PROCEEDINGS of 26th International Conference On ...

26th International Conference on CAD/CAM, Robotics and Factories of the Future 2011 26th-28th July 2011, Kuala Lumpur, Malaysia

PROCEEDINGS of th

26 International Conference On

CAD/CAM, Robotics and Factories of the Future CARs&FOF 2011

KUALA LUMPUR, MALAYSIA th TH 26 – 28 July 2011

EDITORS: Dr. M. Khurshid Khan and Assoc. Professor Dr. Ni Lar Win

1


ACKNOWLEDGEMENTS The Editors wish to sincerely thank the following people/ for their sterling contributions, efforts and input in producing this proceeding: Kwong Chiew Foong Chay Nyit Sun Leow Fui Theng (Proceeding Cover Design) Nee Tani Mariah Mohamad Malathy Batumalay Nor Zahrina Mohd Nor All the Keynote and Plenary Speakers All the Authors All the Referees/Reviewers Special thanks are due to all our Co-sponsors, without whose support the CARs&FOF 2011 Conference would have been a much smaller event: British High Commission (Kuala Lumpur) SIRIM Bhd. MIYAZU Sdn. Bhd. The Institution of Mechanical Engineers (IMechE, UK) The Institution of Engineers, Malaysia (EM) The Institution of Engineering and Technology (IET) Malaysia Network SAE International Malaysian Chapter Malaysia Convention & Exhibition Bureau (MyCEB) The Editors are also grateful for the support of the following four Academic journals (and their Chief Editors), who will publish special editions of selected papers from the CARs&FOF 2011 Conference: International Journal of Intelligent Systems Technologies and Applications (ISSN: 1740-8873 Online ISSN: 1740-8865 - Print) International Journal of Modelling in Operations Management (ISSN: 2042-4108 - Online ISSN: 2042-4094 - Print) International Journal of Customer Relationship Marketing and Management (ISSN: 1947 – 9247) International Journal of Engineering and Technology (ISSN: 2180 – 3633)

2


FOREWARD The International Society for Productivity Enhancement (ISPE) was founded in 1984 in the USA with the goal to accelerate and augment the international exchange of ideas and scientific knowledge in the field of technological applications. The primary aim of ISPE is to unite, on a common platform, policy makers, researchers, people from academia, engineers and users of CAD/CAM, Robotics and Automation, and Advanced Manufacturing Technologies. Twenty four different international institutions have hosted (globally) the CARs&FOF conference since this prolific platform's conception in 1984. The 26th CARs&FOF 2011 Conference has been jointly organised with INTI International University (INTI IU) and the University of Bradford (UoB, UK), with the Host being INTI IU in Kuala Lumpur, Malaysia. th

This volume contains refereed papers accepted for the 26 CARs&FOF 2011 Conference. In arranging the Conference the scope of the meeting was deliberately set wide to attract both research and industrial contributions which reflect current trends and interests in the broad area of CAD/CAM and Advanced Manufacturing. The call for papers stimulated a vigorous response (more that 140 abstracts were received) from which over 110 high quality submissions were reviewed and accepted. We are also happy to report that papers from over 18 countries were received, reflecting the global aspects of manufacturing. INTI IU, UoB and ISPE are pleased to welcome all delegates and hope that future collaborations can be instigated. There was no difficulty in identifying clear themes and current trends for the Conference. The quality of papers reviewed and accepted for presentation achieved very high standards. Three keynote addresses were given by: Professor Olaf Diegel (Auckland University of Technology, New Zealand); Professor Madya Dr. Ishak Bin Aris (University Putra Malaysia); and Dr. Tariq Sattar (London South Bank University, UK). In addition to these, three plenary addresses were arranged. On behalf of the Organising Committee, we would like to sincerely thank the keynote and the plenary speakers for taking the time to prepare and present very relevant keynote addresses. This CARs&FOF 2011 Conference would not have been possible without the efforts of the administration and secretarial support provided by Colleagues and Staff members at INTI IU. A special mention needs to go to Mr. Kwong Chiew Foong, Conference Hon. Secretary, whose sterling efforts made this conference possible, from the initial Call for Papers, the development of the CARs&FOF 2011 website, setting up the EasyChair Conference organising system, to the final organisation of the Conference. Colleagues in the School of Engineering, Design and Technology (EDT) at UoB are also thanked for their assistance and encouragement in preparing for this event. Finally, we are indebted to the Organising Committee and the Programme Committee, who are listed elsewhere in this volume, for selecting the papers, refereeing and assisting in the administrative load and for chairing many of the sessions at CARs&FOF 2011 Conference. Equally, our thanks are extended to the contributing authors and delegates for their splendid, timely and relevant contributions. Dr. M. Khurshid Khan & Associate Professor Dr. Ni Lar Win CARs&FOF 2011 Organising Committee Co-Chairs. Kuala Lumpur, Malaysia, July 2011.

3


CONFERENCE PATRONS Emeritus Professor Walter Wong Vice Chancellor, INTI International University, Malaysia Professor Mark Cleary Vice Chancellor and Principal, University of Bradford, UK

CONFERENCE ADVISORS Professor Dato' Dr. Ibrahim Ahmad Bajunid, Deputy Vice Chancellor INTI International University, Malaysia Professor Raj Gill President of ISPE,(Pro Vice Chancellor of Middlesex University, UK) Professor Phil Coates Pro Vice Chancellor, University of Bradford, UK Professor Alastair S. Wood Dean of Engineering, Design and Technology, University of Bradford, UK Assoc. Professor Dr. Lau Chee Kwong Pro Vice Chancellor, INTI International University, Malaysia

CONFERENCE ORGANISING COMMITTEE Chairs Dr. M. Khurshid Khan and Associate Professor Dr. Ni Lar Win Conference Secretary Kwong Chiew Foong Conference Treasurer Ong See Khin Secretariat Sub-Committee Nee Tani, Mariah Mohamad Finance Sub-Committee Malathy Batumalay, Nor Zahrina Mohd Nor Logistic / Arrangements Sub-Committee Zuraidah Harith (Chair), Audrey Woon Su Fern (Co-Chair), Mazlia Abdul Holit, Ravindran Sayagaran, Mohd Sukri Abdul Malek, Mohd. Iruwan Shah, Go Yun Ii, Ravathy M Rakan Publication Sub-Committee Chay Nyit Sun (Chair), Leow Fui Theng Publicity / Website Kwong Chiew Foong Technical Sub-Committee Dr. Chong Perk Lin (Chair), Dr. Koh Yit Yan, Gerald Victor, Dr. Wan Abdul Rahman (SIRIM Bhd), Dr. Zulkifli Mohamed Udin (Universiti Utara Malaysia), Dr. Mohd Kamal Mohd Nawawi (Universiti Utara Malaysia) 4


INTERNATIONAL PROGRAMME COMMITTEE (2011) Argentina: Eduardo A Destefans, Walter Monsberger; Belgium: Alain Delchambre; Brazil: Marcius Carlvalho, Max Hering de Queiroz, Sergio Eduardo Gouvea da Costa Oscar Salviano Silva Filho; Canada: B.S. Dhillon, Kalyan Ghosh; Chile: Hector Kaschel; Colombia: Hrishi Bera; Cuba: Roberto Rodriguez; France: V. Boschian-Campaner; Germany: Thomas Laengle, Heinz: Westphal Hungary: George. L. Kovacs India P. Radhakrishnan, S.R Dev, Jaimal Singh Khamba Japan: Susumu Sakano, Tohru Kawabe; Libya: Rajab A. Hokoma; Malaysia: Wan Abdul Rahman, Ni Lar Win, Kwong Chiew Foong, Zulkifli Mohamed Udin, Kamaludin Nawawi; Mexico: Arturo Molina Guitierrez; Pakistan: Iftikhar Hussain, Sahar Noor; Peru: Julio Solis Padilla; Poland: Janusz Szpytko; Russia: Vladimir Deviatkov, V.G Gradetsky; Saudi Arabia: Cahill Aslam Awan; South Africa: Glen Bright, Riaan Coetzee; Spain: Emilo Garcia, Julian J Salt; Thailand: H. Paul; Trinidad & Tobago: Chanan Singh Syan, Prakash Persad, Kit Fai Pun; UK: Raj Gill, Mohammed Khurshid Khan, Andrew Day, Derek Godfrey, David K. Harrison, Ken G Swift, A. R. Mileham, Al-Ashaab Ahmed; USA: Jeet Gupta, Yogeshwar Hari, Y. P. Kakad, Biren Prasad; Venezuela Miguel Márquez

5


CARs&FOF 2011 INTERNATIONAL REVIEWERS / REFREES COMMITTEE Ahmed Shaik, Council for Scientific and Industrial Research (CSIR), South Africa Alain Chong, Polytechnic University of Hong Kong, Hong Kong Ali Solati, Shahid Rajaee Teacher Training University, Iran Andrew Day, University of Bradford, UK Angel-Eduardo Gil, Universidad del Táchira, Venezuela Anthony White, Middlesex University, UK Bvijaya Ramnath, Sri Sairam Engineering College, India Chakib Kara-Zaitri, University of Bradford, UK Chanan Syan, University of West Indies, Trinidad and Tobago Chong Perk-Lin, INTI International University, Malaysia Chuah Yea Dat, Universiti Tunku Abdul Rahman, Malaysia Dehong Huo, Middlesex University, UK Dev Anand, Noorul Islam Centre for Higher Education, India Dipnarayan Ray, Central Mechanical Engineering Research Institute, India Ekta Singla, Indian Institute of Technology (IIT) Ropar, India Federico Roy, INTI International University, Malaysia Gerald Victor, INTI International University, Malaysia Glen Bright, University of KwaZulu-Natal, South Africa Iftikhar Hussain, University of Engineering and Technology Peshawar, Pakistan Igor Gorlach, Nelson Mandela Metropolitan University, South Africa Jee Kian Siong, Multimedia University, Malaysia Jeremy Green, Council for Scientific and Industrial Research (CSIR), South Africa John Dickens, Council for Scientific and Industrial Research (CSIR), South Africa John Victory, University of Bradford, UK Karthick Ramananthan, INTI International University, Malaysia Khairul Salleh Mohamed Sahari, Universiti Tenaga Nasional, Malaysia Khin Maung Win, INTI International University, Malaysia Khonzumusa Hlophe, Council for Scientific and Industrial Research, South Africa Koh Yit Yan, INTI International University, Malaysia Kwong Chiew Foong, INTI International University, Malaysia Mauricio Vladimir Peña Giraldo, Universidad Libre, Columbia Mehmet Karamanoglu, Middlesex University, UK Mei Choo Ang, National University of Malaysia, Malaysia Miguel Marquez, Universidad del Táchira, Venezuela Mohammad Khurshid Khan, University of Bradford, UK Mohd Kamal Mohd Nawawi, Universiti Utara Malaysia, Malaysia Mum Wai Yip, Tunku Abdul Rahman College, Malaysia Napsiah Ismail, Universiti Putra Malaysia, Malaysia Ng Poh-Kiat, Multimedia University, Malaysia Ni Lar Win, INTI International University, Malaysia Olaf Diegel, Auckland University of Technology, New Zealand Puramanathan Naidoo, Mangosuthu University of Technology, South Africa Rajay Vedaraj.I.S., VIT University, India Rudolf Reinhard, IMA/ZLW-IfU RWTH Aachen, Germany Sahar Noor, University of Engineering and Technology, Peshawar, Pakistan Shakil Seeraji, Military Institute of Science and Technology, Bangladesh Souilah Zahi, Multimedia University, Malaysia Vladimir Deviatkov, Bauman Moscow State Technical University, Russian Federation Wan Abdul-Rahman, Sirim Bhd, Malaysia Zulkifli Mohamed Udin, Universiti Utara Malaysia, Malaysia

6


INTERNATIONAL SOCIETY FOR PRODUCTIVITY ENHANCEMENT (ISPE) The International Society for Productivity Enhancement (ISPE) was founded in 1984 in the USA with the goal to accelerate and augment the international exchange of ideas and scientific knowledge in the field of technological applications. The primary aim is to integrate technologies, strategies and resources to enhance productivity, competitiveness and thereby improve the quality and standard of life. ISPE has been promoting research and education by organizing conferences, training workshops and seminars and various other professional activities. Among others, the primary aim of ISPE is to unite, on a common platform, policy makers, researchers, people from academia, engineers and users of CAD/CAM, Robotics and Automation, and Advanced Manufacturing Technologies. Twenty four different international institutions have hosted (globally) the CARs&FOF conference since this prolific platform's conception in 1984. The 26th CARs&FOF 2011 Conference will be organised jointly by the INTI International University and the University of Bradford (UK), in Kuala Lumpur, Malaysia from 26th to 28th July 2011.

INTI INTERNATIONAL UNIVERSITY, MALAYSIA Built across an impressive 82 acres of attractive landscaping, INTI International University houses modern amenities and some of the most prestigious career-focused programmes in the country. Located in Putra Nilai, it offers an ideal study environment with state-of-the-art learning facilities and excellent equipment for both study and leisure. And at its heart is a strong academic culture enriched by the expertise and teaching methods of prestigious scholars from around the world. Students of INTI International University get to choose from a remarkable range of established programmes. Graduates will receive top-notch education and leave with certificates that are immediately recognised by the industry. Also, with a strong collaboration with over 330 universities around the world and 43 partner universities that are ranked among the best in the world. INTI International University is a part of the Laureate International Universities network of more than 50 accredited campus-based and online universities with more than 600,000 students around the world, spanning over 24 countries throughout North America, Latin America, Europe, and Asia.

UNIVERSITY OF BRADFORD, UK One of the 12 largest cities in Britain, Bradford is a vibrant modern metropolis located in West Yorkshire. Within Bradford, The University of Bradford has enjoyed a long and distinguished association with engineering and technology. Its forbear, the Mechanics Institute was established during the UK's Industrial Revolution in the 19th Century. The Institute evolved into the Bradford Technical Institute of Technology then, in 1966, a Royal Charter was granted and the University of Bradford was created. The University's School of Engineering, Design and Technology (EDT) is well renowned for its interdisciplinary taught programmes and cutting edge research. Most of EDT's research groups gained international and national rated research quality levels in the recent UK Government Research Assessment Exercise (RAE 2008).

7


TABLE OF CONTENTS KEYNOTE PAPERS A NEW INDUSTRIAL REVOLUTION THROUGH ADDITIVE MANUFACTURING.............................. 15 Diegel, O. ROBOTIC NON DESTRUCTIVE TESTING ................................................................................................. 23 1 2 1 Sattar, T.P. , Leon Rodriguez, H.E. , Salman, H. MOVING TOWARDS HIGH INCOME, SUSTAINABLE AND DEVELOPED NATION BY 2020: CHALLENGES, STRATIGIES AND OPPORTUNITIES .............................................................................. 42 Associate Prof. Dr. Ishak Aris ADVANCED MANUFACTURING PROCESSES DEVELOPING COMPUTER-BASED 2D TOLERANCE ANALYSIS PROCEDURE ................................ 44 1 2 Behnam Moetakef Imani , Seyed Ali Hashemian OPTIMIZATION OF SUPPORT MATERIAL AND BUILD TIME IN FUSED DEPOSITION MODELING (FDM) USING CENTRAL COMPOSITE DESIGN METHODOLOGY (CCD) ............................................ 55 1 2 Pavan Kumar Gurrala , Srinivasa Prakash Regalla CONDUCTIVE FUSED DEPOSITION MODELLING: A STEP FORWARD FOR ROBOTICS?............... 67 1 1 1 2 Diegel, O. , Singamneni, S. , Huang, B. , and Gibson, I. SELFOPTIMIZATION IN ADAPTIVE ASSEMBLY PLANNING ............................................................... 76 Daniel Ewert, Daniel Schilberg, and Sabina Jeschke OPTIMISING SPLIT-RANGE CONTROL ON A BLENDING PROCESS BY MEANS OF RULE-BASED TECHNIQUE ................................................................................................................................................... 83 Puramanathan Naidoo CONCEPTUAL DESIGN OF A RECONFIGURABLE MANUFACTURING SYSTEM FOR THE TOOLING INDUSTRY ................................................................................................................................... 90 1 2 IA Gorlach and BH Roberts IMPROVEMENT OF MILLING TOOL GEOMETRY OPTIMISATION USING A REFINED DIFFERENTIAL EVOLUTION ALGORITHM FOR CHATTER AVOIDANCE ....................................... 100 1 2 3 Ahmad Razlan Yusoff , Nafrizuan Mat Yahya and Mohamed Reza Zalani Mohamed Suffian A DISTINCTIVE MANUFACTURING CELL FOR MASS CUSTOMISATION ....................................... 108 1 2 Nazmier Hassan , Glen Bright VERIFICATION OF FIVE-AXIS TOOL PATH OPTIMIZATION USING VERICUT .............................. 116 1 2 S. Pholpho and M. Munlin STUDY ON TOPOLOGY OF LATERAL SURFACE OF COMPONENTS MANUFACTURED BY DIRECT METAL LASER SINTERING........................................................................................................ 125 1 2 3 C.D.Naiju , M. Adithan , and P. Radhakrishnan A STUDY TO AUTOMATE THE FABRICATION OF BELOW-KNEE (BK) PROSTHESIS SOCKET USING CAD/CAE, OPTICAL DIGITIZING AND RAPID PROTOTYPING TECHNOLOGIES .............. 132 Victor Devadass, Julaiha Adnan, SIMULATION OF 3D FILLING, COOLING AND WARPING OF POLYMERIC BEHAVIOUR IN THE INJECTION MOLDING PROCESS .............................................................................................................. 140 S. Zahi OPTIMIZATION OF WELD BEAD SHAPE IN ND:YAG LASER WELDING USING GREY-BASED TAGUCHI METHOD .................................................................................................................................... 152 1 2 3 Ali Solati , Nasrollah Bani Mostafa Arab , Daavood Mirahmadi Khaki ANGLE SEQUENCING ALGORITHMS FOR FIVE-AXIS MACHINING ................................................ 159 M. Munlin 8


ASSESSMENT OF WCM IMPLEMENTATION IN TRINIDAD AND TOBAGO ..................................... 169 Chanan S. Syan, Krystal Ramoutar SOME ANOMALIES IN THE EXPERIMENTAL RESULTS OF EDM ..................................................... 175 1 2 Bhiksha .Gugulothu, Buschaiah K EMERGING SCENARIOS IN ENGINEERING EDUCATION EFFECTIVE LEARNING OF CONTROL SYSTEM ENGINEERING THROUGH PROJECT-BASED ASSIGNMENTS ............................................................................................................................................ 184 1 2 Federico Roy Jr and Perk Lin Chong WEB APPLICATION AID FOR TEACHING KINEMATICS OF INDUSTRIAL MANIPULATORS ...... 190 1 2 3 4 5 Duarte Franklyn , Márquez Miguel , Gil Ángel , González Freddy and Pérez Wilson E-LEARNING STRATEGY FOR PROJECT-BASED AND COLLABORATIVE LEARNING ENVIRONMENTS USING INTELLIGENT AGENT .................................................................................. 198 Ayisha Qureshi, Rizwana Irfan START-STOP-CONTINUE – WHAT DOES IT TAKE TO CHANGE? ...................................................... 206 Koh Yit Yan AN AUTOMATED ONLINE STUDENT PROJECT ALLOCATION AND MANAGEMENT SYSTEM . 214 M. Ali, P. Pillai and Y. F. Hu THE EFFECT OF STUDENTS‘ LEARNING STYLES ON ACADEMIC PERFORMANCE IN ENGINEERING EDUCATION ..................................................................................................................... 224 1 2 Ni Lar Win and Khin Maung Win HUMAN ASPECTS IN ENGINEERING ACTIVITIES MODELLING AND ANALYSIS OF COMPLEX EM FIELDS PROBLEMS USING HYBRID PROCEDURE ................................................................................................................................................ 232 1,2

2

3

K. N. Ramli, R. A. Abd-Alhameed and P. S. Excell DUAL-BANDS FOUR-ELEMENT ANTENNA ARRAY DESIGN FOR MIMO APPLICATION ............ 239 1, 2 2 2 1, 2 2 Z. Z. Abidin , R. A. Abd-Alhameed , Y. Ma , K. N. Ramli and C.H. See ADVANCED TOOLS IN PRODUCTS/SYSTEMS DEVELOPMENT AND MANAGEMENT OVERALL EQUIPMENT EFFECTIVENESS IMPROVEMENT THROUGH IMPLEMENTATION OF TOTAL PRODUCTIVE MAINTENANCE IN ASSEMBLY CELL OF STEERING GEAR PRODUCTION ........................................................................................................................................................................ 246 N. Ismail, Y. Musa, Z. Leman, and B.T.H.T.Baharudin IMPLEMENTING TQM FOR SUPERIOR ENGINEERING PERFORMANCE: A CASE STUDY........... 255 Poh Kiat Ng and Kian Siong Jee LIFE CYCLE COST ESTIMATION MODELS AND COST ESTIMATION METHODS: FOR THE FACTORIES OF THE FUTURE ................................................................................................................... 266 B.S Dhillon THE IMPLEMENTATION OF RADIO FREQUENCY IDENTIFICATION IN SUPPLY CHAIN MANAGEMENT IN MALAYSIA ................................................................................................................ 277 1 2 3 4 5 Abu Bakar Hamid , Lee Chee Yang , Siti Zaleha Abdul Rasid and Fauziah Sheikh Ahmad Inda Sukati A CONCEPTUAL DESIGN FOR LEAN MANUFACTURING SYSTEM AND ITS IMPLEMENTATION IN SME AND LE ........................................................................................................................................... 289 Amad-Uddin

9


EVALUATION OF PRODUCTION PERFORMANCE AND COST USING OPTIMUM-SEEKING SIMULATION: A CASE STUDY AT AN AIRCRAFT PARTS MANUFACTURING INDUSTRY ......... 296 Mohd Kamal Mohd Nawawi THE MANAGEMENT OF MAINTENANCE IN A LARGE SCALE PRECISION CNC MACHINING MANUFACTURING FACILITY. ................................................................................................................. 302 1 1 1 2 Mr. C. Courtney , Professor A.J. Day , Dr. J.L. Victory , Mr. L. Zeall A HYBRID KNOWLEDGE BASED SYSTEM FOR LOW VOLUME AUTOMOTIVE MANUFACTURING (LVAM): STAGE 2 (DESIGN) .................................................................................. 310 1,2 1 N.M.Z.N. Mohamed , M. K. Khan RELIABILITY ANALYSIS AND REDESIGN OF AN AUTOMOBILE PROPELLER SHAFT ................ 320 G. Riyaz Mohamed1, C. Suresh Kumar2, S. Muthukumaran3, I. Yousuf Imran Ali Baig4, G.Sriram5, B.Vijayaramnath6 EXPLORING THE APPLICATION OF TRIZ IN A GENERATIVE DESIGN SYSTEM........................... 330 1 2 M. C. Ang , and K. W. Ng A FRAMEWORK FOR SIMULATION COUPLING USING SEMANTIC WEB TECHNOLOGIES ....... 338 1 1 1 1 Rudolf Reinhard , Tobias Meisen , Daniel Schilberg and Sabina Jeschke LAYOUT FORMATION IN CELLULAR MANUFACTURING SYSTEMS.............................................. 345 1 2 3 4 5 Sh. Ariafar , N. Ismail , S. H. Tang , M. K. M. A. Ariffin & Z. Firoozi THE DESIGN, DEVEOPMENT AND VALIDATION OF A STRUCTURED REVIEW METHODOLOGY FOR HEALTH AND SAFETY PERFORMANCE MANAGEMENT IN THE CHEMICAL INDUSTRY . 351 Dr. Chakib Kara-Zaïtri and Dr Saad Alquahtani MANUFACTURING ENTERPRISE MANAGEMENT BASED ON THE SERVICE-ORIENTED TECHNOLOGIES .......................................................................................................................................... 361 1 2 3 Konstantinos Kotsopoulos , Yim Fun Hu , and Pouwan Lei INNOVATIVE METHODOLOGY FOR DESIGNING A MODULAR HIGH VOLUME FLOW LINE .... 371 1 2 1 R. Yumbla , S. Lumley , and M. K. Khan THE STRATEGIC ALIGNMENT OF QUALITY FUNCTION DEPLOYMENT (SAQFD) AS A KEY DRIVER FOR THE DESIGN OF A HIGH VOLUME PRODUCTION LINE ............................................. 379 1 2 1 R. Yumbla , S. Lumley , and M. K. Khan LOGISTICS AND MATERIALS HANDLING SYSTEMS AND DEVICES SIMULATION MODEL OF MARITIME INVENTORY ROUTING PROBLEM WITH PARTICULAR APPLICATION TO CEMENT DISTRIBUTION.......................................................................................... 389 1,2 1 1 E. Wirdianto , H. S. Qi , M. K. Khan A STUDY AND IMPROVEMENT OF TEST AND MEASUREMENT INDUSTRY‘S SUPPLY CHAIN SYSTEM ........................................................................................................................................................ 400 1 2 Yea Dat Chuah and Mike Miles CONTROL SYSTEM MANIPULATION OF PROFIT IN SUPPLY CHAINS ............................................ 409 1 2 3 4 A.S.White , M. Karamanoglu , R. Gill , M.Censlive PRODUCT LIFE CYCLE INTEGRATION THE ROLE OF CONCURRENT ENGINEERING IN PROJECTS INVOLVING RADICAL INNOVATION: A CONCEPTUAL STUDY ................................................................................................ 421 Poh Kiat Ng and Kian Siong Jee EFFECT OF AGE OF PRODUCT ON METAL CONTENT AND SOME CHEMICAL AND PHYSICAL PROPERTIES OF CANNED TOMATO PASTE .......................................................................................... 429 Dr. Milad. A. Shalluf A NOVEL INDEX BASED HEURISTIC FOR JOB SHOP SCHEDULING ............................................... 433 1 2 1 1 S. Maqsood , I. Hussain , M. K. Khan , and A. S. Wood 10


DEVELOPMENT OF RECONFIGURABLE MANUFACTURING SYSTEM USING FLOWLINE CONFIGURATION ....................................................................................................................................... 441 1 2 3 A.O. Oke , K. Abou-El-Hossein and N. J. Theron ACTIVITY BASED COSTING APPROACH FOR PRODUCT LIFECYLCE COSTING SYSTEM USING OBJECT ORIENTED PROGRAMMING ..................................................................................................... 450 1 2 Siva Prasad Darla , and S Narayanan AUTOMATION AND ROBOTICS DESIGNING CO-OPERATIVE MANIPULATORS OF HETEROGENEOUS FORM USING WALKING WHEELCHAIR .............................................................................................................................................. 460 1 2 3 4 I.S. Rajay Vedaraj , Tabreez Musheer , Aishwarya and Kirtiraj UNDERGROUND MINE NAVIGATION USING AN INTERGRATED IMU/TOF SYSTEM WITH UNSCENTED KALMAN FILTER................................................................................................................ 470 1 2 Khonzi Hlophe , and Jeremy Green DESIGNING AND MANUFACTURING WITH DEVELOPABLE SURFACES ....................................... 480 1 F. Perez SERPENTINE GAITS FROM COUPLED ORTHOGONAL JOINT ORIENTATION FUNCTION .......... 491 1 2 3 Atanu Maity , Somjyoti Majumder , and Sukamal Ghosh MULTIMODAL ANALYSIS OF UNSTRUCTURED VIDEO STREAMS ................................................. 502 Vladimir Devyatkov, Alexander Alfimtsev HUMAN DETECTION FOR UNDERGROUND AUTONOMOUS MINE VEHICLES USING THERMAL IMAGING ...................................................................................................................................................... 512 1 2 3 J. S. Dickens , J. J. Green and M. A. van Wyk AUTOMATION OF PRE-PROCESSING AND FEATURE EXTRACTION PARAMETER SELECTION FOR A SINGLE-TRIAL P300-BASED BRAIN-COMPUTER INTERFACE USING A GENETIC ALGORITHM ................................................................................................................................................ 522 1 1 2 Randy E. S. Harnarinesingh , Chanan S. Syan and Ramaswamy Palaniappan THE AUTOMATION OF THE ‗MAKING SAFE‘ PROCESS IN SOUTH AFRICAN HARD-ROCK UNDERGROUND MINES ............................................................................................................................ 532 Teleka SR, Green JJ A GEOMETRICAL METHOD FOR SOLVING THE INVERSE KINEMATICS PROBLEM OF SERIAL ROBOTIC MANIPULATORS....................................................................................................................... 541 Firas Subhy Hameed Ahmed, AUTOMATIC CALIBRATION OF A TOOL-CHANGING UNIT FOR RECONFIGURABLE MANUFACTURING SYSTEMS (RMS) USING NINTENDO WII REMOTES ......................................... 550 1 2 J. Collins , G. Bright DYNAMIC LOAD CARRYING CAPACITY COMPUTATION USING NONLINEAR ANALYSIS ....... 558 1 2 3 Jaime Arango , Gustavo Osorio , and Fabiola Angulo VISION BASED OBSTACLE DETECTION MECHANISM OF A FIXED WING UAV.......................... 569 1 2 3 3 Prof. S.N. Omkar , Saurabh Mishra , Sanjay Tripathi , Gaurav Kumar PATH PLANNING OF WHEELED MOBILE ROBOT USING A NEW OBJECTIVE FUNCTION IN GENETIC ALGORITHM .............................................................................................................................. 577 1 2 Varshovi-Jaghargh, Payam , Naderi, Davod A PROPOSED METHODOLOGY FOR BEHAVIOUR-BASED MULTI-AGENT Q-LEARNING FOR AUTONOMOUS EXPLORATION ............................................................................................................... 583 1 2 3 4 Dip N. Ray , Amit K. Mandal , S. Mukhopadhyay and S. Majumder SENSITIVITY ANALYSIS OF GENETIC ALGORITHMS FOR JOB SHOP SCHEDULING PROBLEMS ........................................................................................................................................................................ 594 1 2 1 1 S. Maqsood , S. Noor , M. K. Khan , A. S. Wood 11


DECENTRALISED NAVIGATION OF AGVS IN A COMPUTER INTEGRATED MANUFACTURING ENVIRONMENT ........................................................................................................................................... 603 1

2

Roland Dixon , Prof. G. Bright TRANSMISSION LINE INSPECTION ROBOT FOR MALAYSIA: PROTOTYPE IMPROVEMENT .... 612 1 1 1 1 *Khairul Salleh Mohamed Sahari , Justin Chan , Sarah Navita , Adzly Anuar , Syed Sulaiman Kaja 2 2 Mohideen , and Mohd Zafri Baharuddin A NOVEL VIDEO IMAGE SMOKE DETECTION SYSTEM BASED ON WAVELET & COLOR INFORMATION ANALYSIS ........................................................................................................................ 621 1 2 3 Chai Yoon Yik , Ewe Soo Chait , and Loo Chan Perng NOVEL 6 DOF HYBRID MACHINE DESIGN ........................................................................................... 631 1 2 3 Ahmed Asif Shaik , Prof. Glen Bright , and Dr. Nkgatho Tlale THE ROBOFRIEND RESEARCH PROJECT .............................................................................................. 640 1 2 Sami Salama Hajjaj , et al DESIGN AND DEVELOPMENT OF RECEIVER MODULE FOR WIRELESS VISION BASED RESCUE ROBOT FOR ROUGH TERRAIN USING DAVINCI CODE PROCESSOR .............................................. 651 1 2 3 4 Naveen Prakash , Kailash Vijaykumar , Bharathi V ,and Arun Perumal A DISTRIBUTED VISION SENSORS AIDED INTELLIGENT ENVIRONMENT SYSTEM DESIGN FOR MOBILE ROBOT NAVIGATION ................................................................................................................ 669 1 2 1 Yongqiang Cheng , Ping Jiang , and Yim-Fun Hu INTEGRATING OPENCV BASED MACHINE VISION TO AN ABB INDUSTRIAL ROBOT ............... 678 Srinivas Ganapathyraju, Ph.D AN EYE-TRACKING BASED WIRELESS CONTROL SYSTEM ............................................................. 686 Suraj Verma, Prashant Pillai, and Yim-Fun Hu STRUCTURED LIGHT BASED VISUAL NAVIGATION ON ROCKY TERRAIN FOR A SIX WHEEL LUNAR ROVER ............................................................................................................................................ 697 1 2 3 Vikalp Sachan , K. S. Venkatesh , and Ashish Dutta MINING ROBOTICS SENSORS, PERCEPTION SENSORS ON A MINE SAFETY PLATFORM .......... 706 1 2 3 4 5 Green JJ , Hlophe K , Dickens J , Teleka R , Mathew Price TOWARDS OPTIMUM DESIGN OF MAGNETIC ADHESION WALL CLIMBING WHEELED ROBOTS ........................................................................................................................................................................ 716 Salman, H., Sattar, T.P., Salinas, E. REVERSE ENGINEERING OF CONTROL PROGRAMS INTO RECURRENT NEURAL NETWORKS FOR RECONFIGURABLE MANUFACTURING SYSTEMS ..................................................................... 729 1 2 Vimal Nandhan R.K. and Ramesh Babu N. DESIGN AND EXPERIMENTATION OF A SIX WHEEL LUNAR ROVER FOR MOTION ON UNEVEN TERRAIN ....................................................................................................................................................... 739 Harjinder Singh, Anjali Kulkarni, Biswanath Panda, Anupam Sana and Ashish Dutta PERFORMANCE INDICES FOR SERIAL ROBOTIC MANIPULATORS ................................................ 747 1 2 Ekta Singla and Ashish Singla SEMI-AUTOMATED COLOUR DYEING SYSTEM (YARN AND READY PIECES) FOR SMALL TIME SONGKET MANUFACTURERS ................................................................................................................. 755 1 1 1 Mohd Shahrul Azmi Mohamad Yusoff , Shaifull Azhar Othman , & R Fairuz Indra Didi Indra Tjahya DEVELOPMENT OF A HUMANOID HEAD ROBOT AMIR-III ............................................................... 760 M.F. Alias, A.A. Shafie, and S.H. Hashim A POTENTIAL FIELD METHOD FOR AUTONOMOUS LUNAR ROVER NAVIGATION IN 3D TERRAIN ....................................................................................................................................................... 767 1 2 1 Parth Nanadikar , Rahul Shome and Ashish Dutta VIBRATION CONTROL OF A CART-FLEXIBLE POLE SYSTEM USING A ZVD SHAPER ............... 777 Ashish Singla THE NEW METHOD OF LOCOMOTION IN HEXAPOD ROBOT BY USING EXTRA JOINT ....... 789 1 2 3 4 Sajad Moradian , Karim Mohammadi , Mohssen Heydari Kaiedan , and Mehdi Moradian 12


FUZZY LOGIC APPROACH TO ESTIMATE AVAILABILITY OF AUTOMATED MATERIAL HANDLING DEVICES ................................................................................................................................. 798 1 2 Smoczek J. , Szpytko J. DESIGN OF A FUZZY GAIN SCHEDULING CONTROLLER FOR THE ANTI-SWAY CRANE SYSTEM ........................................................................................................................................................ 809 1 2 Smoczek J. , Szpytko J. OTHERS ADVANCED DUST CONTROL TECHNIQUES IN CEMENT INDUSTRY ELECTROSTATIC PRECIPITATOR –A CASE STUDY ............................................................................................................. 820 1 Engr. Zulfiqar Khattak , Engr: Jamil Ahmad KNOWLEDGE-BASED ARCHITECTURE FOR HIGH SPEED NETWORK INTRUSION DETECTION SYSTEM ........................................................................................................................................................ 833 1 2 3 4 Faeiz Alserhani , Monis Akhlaq , Irfan Awan , Maha Alsarhani GREEN MANUFACTURING MANAGEMENT: INVESTIGATION OF THE PHILOSOPHY PRACTISED IN INDUSTRY ............................................................................................................................................... 841 1 2 3 Adam Shariff Adli Aminuddin , Mohd Kamal Mohd Nawawi , El Mostafa Kalmoun 2D PROFILE RECONSTRUCTION USING CONSTRAINED LOCAL FITTING OF NURBS CURVES 847 1 2 Behnam Moetakef Imani , Seyed Ali Hashemian EFFECT OF ELECTRODE GEOMETRY ON WELD NUGGET SIZE IN RESISTANCE SPOT WELDING ........................................................................................................................................................................ 860 1 2 3 S.M. Hosseini , N.B. Mostafa Arab and V. Panahizadeh PRELIMINARY STUDY OF CRITICAL SUCCESS FACTORS OF KNOWLEDGE MANAGEMENT IMPLEMENTATION IN TOTAL QUALITY MANAGEMENT (TQM) ORGANIZATIONS ................... 867 1 2 3 Pei Pei Hing , Mum Wai Yip , Dominic Lau A COMPACT TUNEABLE PIFLA NOTCHED ULTRA-WIDEBAND ANTENNA FOR WIRELESS APPLICATIONS ............................................................................................................................................ 873 1 1, 1 1 I.T.E. Elfergani , R.A. Abd-Alhameed C.H. See , H.I.Hraga and K.N. Ramli FAST TUNABLE DIRECT DIGITAL SYNTHESIZER FOR HIGH FREQUENCY APPLICATIONS ..... 881 #1 #2 Mrs.R.Sujatha , Mr.M.Marimuthukumar WEB-BASED TOTAL CONDITION MONITORING ................................................................................. 887 H. F. Al-Hajjar, A.Lewalski , K.M. Ebrahimi , TCP PERFORMANCE EVALUATION OVER HETEROGENEOUS WIRELESS NETWORKS USING MIH ................................................................................................................................................................ 893 1 2 M. Ali, P. Pillai and Y. F. Hu MECHANISTIC MODEL OF THE CUTTING PROCESS FOR CONDITION MONITORING ................ 903 1 2 A Nikranjbar , K M Ebrahimi AM ENHANCED SUPPLY CHAIN PERFORMANCE MEASUREMENT SYSTEM WITH INCORPORATION OF R&D AND MARKETING POLICY ...................................................................... 915 a b b a c Felix T.S. Chan, Ashutosh Nayak, Ratan Raj, Alain Yee Loong Chong, and Chiew Foong Kwong

13


KEYNOTE PAPERS

14


A NEW INDUSTRIAL REVOLUTION THROUGH ADDITIVE MANUFACTURING Diegel, O. Auckland University of Technology, Centre for Rapid Product Development Auckland, New Zealand e-mail1: [email protected] ABSTRACT It has been speculated that, in the not too distant future, additive manufacturing will begin to have a marked effect on how we order and manufacture products. When one examines some of the implications of AM more closely, it can be concluded that it could, in fact, have a major Impact not just on products, but on our society, and how we live and do business. This paper briefly describes the state of the art in AM with a particular emphasis on where the technologies are heading, and then examines some of the issues that will need to be grappled with as additive manufacturing comes of age… within the next few decades… Keywords: Additive Manufacturing, Future Focused Manufacturing Technologies.

1

INTRODUCTION In the late 18th century, the invention of the flying shuttle, the steam engine, and other machines that replaced manual labour produced almost unimaginable changes in society. They marked the start of the industrial revolution which influenced almost every aspect of daily life in some way. As the revolution progressed, society changed in every way. For the following two centuries average income and population exhibited unprecedented growth. The world's average per capita income increased over ten-fold, while the world's population grew by a factor of six. Never before had the world seen such unprecedented change. Additive manufacturing may represent the same paradigm shift to society, and will influence every aspect of the way we live. There has been a huge amount of research go into the many technical aspects of additive manufacturing, from materials and processes to applications and management. This paper speculates on some of the sociological and wider economic issues that additive manufacturing may give rise to.

2

ADDITIVE MANUFACTURING Additive manufacturing began in the late 1980s, with one of the most historically important patents being filed by Charles Hull in 1986, and the first commercial SLA machine appearing in 1988. Since then almost every year has seen an exponential rise in available systems, technologies and materials. The Society of Manufacturing Engineers defines Additive Manufacturing (AM) as the process of manufacturing a physical object through the layer-by-layer selective fusion, sintering or polymerization of a material [1]. The additive manufacturing process begins by taking a 3D computer generated file and slicing it into thin slices (commonly ranging from a few microns to 0.25mm per slice depending on the technology used). The additive manufacturing machine then builds the model one slice at a time, with each subsequent slice being built directly on the previous one. As a result of the material deposition and processing operations, the digital model is converted into a physical part or product. Many different additive manufacturing technologies exist, which differ mainly in terms of the materials they use to build the part, which are typically in a powder or liquid raw state, and the process used for creating the model slices. Until recently, many of these technologies, such as stereolithography (SLA), Fused Deposition Modelling (FDM), early Selective Laser Sintering (SLS) systems and 3D printing, just to name a few, were only able to make parts for prototyping purposes, as the processes produced parts that were not strong enough for production [2]. The latest additive manufacturing technologies, however, now allow the production of full-strength polymer and metal parts within hours rather than days [3]. Unlike subtractive manufacturing, where material is removed from a larger block of material until the final product is achieved, most additive manufacturing processes do not yield excessive waste material. It also typically 15


does not require the large amounts of time needed to remove unwanted material, thus reducing time and costs, and producing little waste [3]. It is only over the last few years that additive manufacturing is being used by some companies as a viable production technology. As new polymer and metal materials are developed and the speed and precision of the machines further increase, more additive manufacturing machines are likely to find their way into mainstream production lines [3].

2.1 Freedom from manufacturing constraints Additive manufacturing enables the creation of parts and products with complex features, which could not easily have been produced via subtractive or other traditional manufacturing processes. Injection molded or die-cast parts, for example, must be removable from the die in which they are made and must therefore be designed in such a way that this can be done. The metal part shown in figure 1, for example, could not easily be machined or cast because there is no way of removing the internal part of the die from the component or of machining the interior surfaces [4]. Additive manufacturing, however, does not suffer from these particular restrictions. The complexity of the part does not affect whether it can be made, or even its cost. It allows for components of almost any complexity, freedom in design and increased flexibility in the features and functions of the end product.

Figure 1: Fuel Injection Swirler from Morris Technologies. Tie-down clamp, with moving parts, made as a single integrated moving component on EOS SLS system. With additive manufacturing it is also possible to manufacture complex interlocked moving parts in readymade working assemblies. Though two components may be permanently linked together, they are made as a single component and come out of the machine assembled and ready to work. Figure 1 shows a tie-down clamp made on an EOS laser sintering system out of aluminium filled polyamide material, which is composed of four different components that allow the clamp to operate in the correct way [5]. The entire clamp is, however, manufactured in a single operation with no assembly whatsoever required. If the clamp were to be manufactured using traditional manufacturing methods, it would require, at least, eight components and an assembly procedure to attach all the separate components together. It should be noted that additive manufacturing does not remove all manufacturing restrictions. It, instead, replaces them with a different set of design considerations that designers must take into account if they wish to successfully use the technologies. These new design considerations are, however, much easier for designers to both understand and comply with without them affecting design intent in a major way.

2.2 Mass Customization With additive manufacturing parts can be immediately made as there is no longer a long lead-time to get tooling produced. This has a great impact on new product time to market, and on the ability to easily produce model changes throughout the life of a product. It also has implications in stock control: As components can be made on the spot, companies may no longer need to hold stock of spare parts as they simply manufacture the parts when needed. 16


From a product design perspective, it also means that every component made can be completely different to the others in a production run without significantly affecting the manufacturing cost. This opens the door to mass-customization in which, though mass-manufactured, each product can be customized to each individual customer. For this new way of designing products to be used effectively the industry will need to develop new methods for integrating personalized customer data into their processes. This development has already started, particularly in the hearing aid [6] (figure 2) and the dental industries, in which specialized software exists to automate the processes of patient data acquisition. This now needs to be extended to encompass others, including consumer product industries.

Figure 2: Individually customized mass-produced hearing aid shells.

2.3 Freedom of Design Because of traditional manufacturing technology restrictions a product, which the designer may have originally envisioned as having a certain aesthetic and functionality, may need to be compromised so that it can be cost-effectively made. Most designers are quite accustomed to hearing the response of ―it cannot be made like that‖ from manufacturing engineers. They may then need to compromise their design to the extent that the product loses the essence that truly embodies the designers‘ vision. With additive manufacturing, complexity and geometry no longer affect manufacturability. Almost anything the designer imagines can be made precisely as the designer conceived it (figure 3).

Figure 3: Quintrino lampshade by Bathsheba Grossman [7], Rollercoaster plate by Freedom of Creation (FOC) [8], and Subdivision bracelet by Nervous System [9].

3

HOW ADDITIVE MANUFACTURING IS USED TODAY Beyond the use of AM for making prototype of engineering or design models, AM is beginning to see uses in a range of commercially available products. This section shows just a few of these application areas, without going into any great detail about them, through the use of images. Though each individual application would make an interesting case study, they are beyond the scope of this paper which just attempts to show the range and variety of current applications of additive manufacturing. Note that 17


references are too numerous to include directly in the paper, but most images come a few websites listed in the references section.

Fashion and fashion accessories

Figure 4: Dress from FOC [8], Woman‘s Shoe from FOC and Pauline Van Dongen [8], Jewellery from EOS [6], fashion garment from Joshua DeMonte[10] Medical

Figure 5: Hearing aid by EOS [6], Hip socket by Arcam [11], Dental crowns by EOS [6] Rapid Manufacture and Engineering

Figure 6: Treviso Technologia designer sunglasses made for Crabbi Sun Living[6], Rapid tooling by EOS[6], Robotic arm component for Deka by EOS [6]

4

FUTURE DEVELOPMENTS IN ADDITIVE MANUFACTURING Beyond the huge amount of research going into the many technical aspects of additive manufacturing, such as new materials and processes and faster deposition techniques and the management of AM as in integral part of the design process, there are also some developments in some interesting areas that may have a profound impact on future societies.

18


4.1 NanoPrinting Researchers around the world are trying to create complete components and machines at a nano level. Prof. Satoshi Kawata, in Japan, and Koji Ikuta, also in Japan, are using technologies, such as femtosecond laser manufacturing, to build incredibly small components. This could, ultimately, mean the manufacture of nanobots that could, for example, be injected into the blood stream with the task of cleaning the blood stream of all undesirable molecules.

Figure 7: The goal: nanobots that can clean the undesirable elements from our blood-streams [12]

Figure 8: Nano-bull (10 microns in size), Brandenburg gate, and Buckyball [13]

4.2 Printing houses Research at the University of Southern California, is developing machines to print entire houses. The technology, called contour crafting, extrudes concrete through a nozzle and traces out the wall contours layer-by-layer until the walls are built. The current technology is already capable of building concrete walls in a laboratory situation. Research is on-going to develop the robotic systems to install modular plumbing and electric systems, insulation, window and door lintels, and so on. In parallel, Enrico Dini of D-Shape in Italy is developing a system based on 3D-Printing, in which a layer of stone powder is deposited, and a binder is printed onto the power, solidifying where it prints. The result is a large-scale structure made out of reconstituted stone. Estimates are that, once all the technologies have been developed, it will be possible to build an entire house within a few days.

Figure 8: Printing houses: the University of Southern California‘s Contour Crafting system [14], and Enrico Dini‘s 3D printing system [15] 19


4.3 Bio-printing Tissue engineering, and the ultimate printing of entire organs or body parts, is an active area of research worldwide. Prof. Anthony Atala, of Wake Forrest University, for example, is currently able to bioprint a number of animal organs, and bladders for humans that are engineered out of the patients own stem cells. They have also used inkjet printing technology to print mouse heart valves, and are close to being able to print out a working human kidney.

Figure 9: Bioprinting will give the ability to print entire replacement organs[16]

4.4 Food Printers The Fluid Interfaces Group at MIT Media Lab has proposed concepts for 3D printers capable of printing food. In their concept canisters of raw ingredients, vitamins, colouring and flavouring agents, etc. are used to reconstitute the food requested by the user. Though this concept may seem like pure fantasy, active research is on-going at research institutions around the world to make this a reality.

Figure 10: Food printers able to reconstitute the food you desire when you want it [17]

5

THE FUTURE IMPACTS OF ADDITVE MANUFACTURING. Our current economic models, largely driven by the advent of factories through the 1st industrial revolution, are based around the mass production of products through a factory environment. In these models, products are manufactured in a factory that could be located anywhere, finished goods are then transported to retail stores (often via distribution centres) that hold a number of the products in stock, and the product is, eventually, bought by the customer. In these models, a large part of the cost of goods sold is not in the direct manufacturing of the products, but the indirect costs, such as transport, middle-man infrastructure and margins, etc. Though figures vary hugely depending on the type of product, it is estimated that the manufacturing cost of a product is typically between 10% to 25% of the retail price of the product. The rest goes into marketing, distribution and the various other costs currently required to bring the product to the consumer [18, 19].

20


Figure 11: Examples of the factors that make up the retail cost of a product it In manufacturing, as AM technologies improve, and as material costs eventually get near those of traditional manufacturing methods, additive manufacturing could entirely replace traditional manufacturing, thus heralding true on-demand manufacturing, and a world in which product customisation is the norm rather than the exception. Not only will the manufacturing be on-demand, but it could well become ‗home manufacturing‘ in which each of us has a 3D printing system at home which will supply is with whatever manufactured products we need. Your local garage will no longer need to have a store room full of spare parts, as they will simply print them as and when needed.

Now extend this thought to our economic system as a whole. Imagine the whole new business systems this will create. Business models in which external manufacturing costs, labour, marketing and transport are no longer part of the equation. Instead, the value of products is in their design, and in the knowledge needed to design and use them. What would be the impact on the environment of not having to transport goods around the world? What happens to the traditional workforce? Do they shift into building AM machines, or servicing, or design?

On the medical side, the gradual miniaturisation of AM processes will allow nano-machines to be printed that can patrol our blood-stream to keep diseases and viruses at bay, while advances in tissue engineering will allow us to print tailored replacement organs. Should we engineer ourselves a spare body to keep on ice, just in case? The impact of such advances, and the ethical questions they raise could be tremendous. Beyond the ethical questions, what about the increased leisure time that increased life-span is bound to cause?

Even in food, one can imagine one day walking up to our refrigerator and ordering a cheese burger, medium rare, and the fridge printing the requested food for us out of raw materials, colouring and flavouring agents, and all the vitamins necessary to make it as healthy as it can be. Why not have it print out the cutlery and plate at the same time?

Does this mean that conventional cooking now becomes something that is done purely as a leisure activity? And even within this question, are the ingredients used in leisure cooking grown the old-fashioned way, or are they simply printed?

6

CONCLUSIONS Though the exact timeline is hard to predict, it is foreseeable that additive manufacturing could, one day, replace conventional manufacturing techniques. If this happens, then it is also likely to have an effect on society, as it will change business models, health systems and lifestyles. Though much of this may seem like science fiction, a surprizing amount is already beginning to happen and the rate of development is accelerating. This paper presents some examples of where AM technologies are heading, and speculates on how some of the changes brought about by ever improving AM technologies might influence our futures.

21


7

REFERENCES

[1] SME, (2010), Available from http://www.sme.org/cgibin/communities.pl?/communities/techgroups/ddm/what_is_ddm.htm&&&SME [2] Hopkinson, N., Hague, R.J.M. and Dickens, P.M., Rapid Manufacturing an Industrial Revolution for the digital age, Wiley, 2006 [3] Wohlers, T. (2009). Worldwide progress report on the rapid prototyping, tooling, and manufacturing state of the industry Wohlers Associates. [4] EOS, (2010a), Available from http://www.eos.info/en/applications/aerospace.html, [Accessed February 2010] [5] EOS, (2010b), Available from http://www.eos.info/fileadmin/download/literature/EOSINT_P_760_E.pdf, [Accessed February 2010] [6] EOS, (2010c), Available from http://www.eos.info/en/news-events/pressreleases/pressdetails/select/pressemitteilungen/article/103/neuespatent.html?tx_ttnews%5Byear%5D=2006&tx_ttnews%5Bmonth%5D=11&cHash=dbe63472b8, [Accessed February 2010] [7] Grossman, B., (2010), Available from www.bathsheba.com, [Accessed February 2010] [8] FOC (Freedom of Creation), (2010), Available from http://www.freedomofcreation.com/shop/order.php?cat=31, [Accessed February 2010] [9] Nervous System, 2011, Accessed May 2011 from http://n-e-r-v-o-u-s.com/index.php [10] http://joshuademonte.com/home.html [11] Arcam: www.arcam.se [12] www.e-spaces.com [13] http://www.skawata.com/english [14] http://www.contourcrafting.org/ [15] http://www.blueprintmagazine.co.uk/index.php/architecture/the-worlds-first-printed-building/ [16] http://www.wakehealth.edu/WFIRM/ [17] http://web.media.mit.edu/~marcelo/cornucopia/ [18] http://arstechnica.com/gaming/news/2006/12/8479.ars [19] http://people.hofstra.edu/geotrans/eng/ch5en/appl5en/costs_shoe_China.html Other useful websites: Freedom of Creation (FOC): http://www.freedomofcreation.com/ EOS: http://www.eos.info MGX Design: http://www.mgxbymaterialise.com

22


ROBOTIC NON DESTRUCTIVE TESTING Sattar, T.P.1, Leon Rodriguez, H.E.2, Salman, H.1 1

London South Bank Universit, Faculty of Engineering, Sceince & the Built Environment, London, UK e-mail: [email protected] 2 Nueva Granada Military University, Bogota, Colombia

ABSTRACT This keynote paper aims to highlight the application of mobile robotics to perform inspection and Non Destructive Testing (NDT) in industries such as aerospace, large scale fabrication, pipelines, petro-chemical storage, and power generation. It describes industrial tasks where regular inspection is essential to ensure the integrity of infrastructure such as storage tanks, pressure vessels, pipelines, aircraft, ships, etc, and to provide managers of capital assets with data to plan outages and to make decisions on the life span of their infrastructure. The development of robot prototypes is described for these industrial tasks. These robots deploy NDT systems by first providing access to large vertical structures or to test sites that are inaccessible to humans. They are designed to reduce outage time, or where possible, carry out the NDT on-line thus preventing costly outages. Keywords: Non Destructive Testing, Wall Climbing, Pipe Climbing and Amphibious Robots.

1

INTRODUCTION Robotic Non Destructive Testing (RNDT) is a field that has made some progress over the past two decades 12 [ , ]. The aim is to combine robotics with non destructive testing and evaluation techniques to enable an operator to perform inspection remotely. The 4 essential Ms of RNDT are Monitoring, Mobility, Manipulation and Measurement. Monitoring is the task of obtaining and storing information (data from previous inspection) about safety critical infrastructure to make asset management and outage decisions. Mobility is the task of carrying a payload of NDT sensors to a test site on very large structures that may also be located in hazardous environments. Manipulation is the task of deploying the NDT sensors in the required way e.g. raster scanning, following weld lines, skewing a probe to get higher signal to noise ratio, etc. Finally, the Measurement task is to reliably detect the presence and size of defects such as corrosion, cracks, inclusions, and disbonding in laminate structures. Regulatory bodies require mandatory inspection of safety critical infrastructure both during and after construction. These structures are usually very large and/or located in remote and hazardous environments. The Non Destructive Testing (NDT) system has to be deployed by first providing very expensive access, requiring the erection of scaffolding and lengthy preparation before NDT can start. In many cases e.g. in power plant, pipelines, storage tanks in the petro-chemical and food processing industries, etc., the inspection has to be performed during an outage by shutting down a plant. There is enormous pressure to reduce the outage time by performing the inspections as efficiently and quickly as possible to provide a quick turnaround. Our research has developed a number of mobile wall-climbing, swimming and pipe-climbing robots that greatly reduce the cost of access to a test site by eliminating scaffolding or abseiling and rope deployment 3 of human operators [ ]. This paper describes some of these robots. Probably the World‘s first wall climbing robot was developed in the late 1980‘s by the Institute of Problems in Mechanics, Moscow. The robot uses two platforms that move relative to each other to obtain stepped motion. The platforms attach to a surface using pneumatic suction cups. A rotation of the outer platform (while the inner platform is attached to a surface) enables change of direction. We modified this robot in 1992 by equipping it with a six axis robot arm and an ultrasound flaw detector to perform the NDT of vertical steel plates. The modified robot is shown in figure 1(a).

23

26th International Conference on CAD/CAM, Robotics and Factories of the Future 2011 26th-28th July 2011, Kuala Lumpur, Malaysia 45

Figure 1(b) shows the RRNDT wall climbing robot that we developed in 1995[ , ]. The robot uses pneumatic suction cups to stick to a surface. It carries a six axis PUMA 260 arm to raster scan with wet and dry contact ultrasound probes. The figure shows a C-scan image of corrosion thinning (variable thickness 06 mm measured from the back wall) of a 10mm thick steel plate, adjacent colors corresponding to thickness steps of 0.375 mm. Data obtained with 5 MHz wet contact compression wave probe (8mm diameter). The C-scan image is of the letter ―U‖ which is part of the word ―SBU‖ that is machined into the back of the 10mm thick steel plate.

Other climbing robots for climbing on curved surfaces such as Liquid Petroleum Gas (LPG) spheres and pipes are shown in figure 2. The prototype generic mobile inspection tool called RRNDT shown in figure 2 comprises of a compact robot plus a 7 axis robot arm and a maximum 5 kg payload of NDT sensors to test welds. It is capable of climbing motion over highly curved surfaces of any material- 860 mm diameter pipes and 3m diameter pressure vessels by adapting to the surface curvature. The weld inspection of nozzle joints in 860mm diameter pipes in the primary circuit of nuclear power plant is a hazardous task requiring operators to go into the containment area for short periods to perform manual NDT. The robot is designed to replace the human operator for multi-tasking applications in hazardous environments e.g. nuclear power plants. The main features of the design are: • • • •

Thigh Hinges which tilt leg pairs relative to the rigid vehicle payload platform (chassis) Universal pneumatic actuated ankle joints which can be made alternately free during a walking step, but otherwise locked rigid for vehicle stability during the data acquisition stages Suction feet which can adapt and adhere to curved surfaces whilst remaining sufficiently rigid for vehicle stability Seven degrees of freedom revolute jointed arm equipped with a force sensor in its wrist and a 5MHz ultrasound probe.

The RRNDT robot has been tested on ultrasonic examination of welds on a 350mm nozzle at 45 degrees to 865 mm diameter feeder pipes in the reactor coolant loop of a nuclear power station, Sizewell B, United Kingdom.

24


Figure 1: Nozzle weld inspection in the primary circuit of a nuclear power plant with a climbing robot and 7 DOF scanning arm

2

WELD INSPECTION OF SHIP HULLS AND STRENGTHENING PLATES Figure 3 shows a good example of an application requiring the provision of access to weld lines that run vertically and horizontally on the hull of cargo container ships. The welding together of blocks in the dry dock after they have been constructed in hangars requires the provision of access to the weld lines. Currently this is done by erecting planks on ropes attached to the top deck. After the construction is complete, the welds are inspected with ultrasound NDT. The European funded project ―Climbing Robot Cell for Fast and Flexible Manufacture of Large Scale Structures (CROCELLS)‖ has developed a team of prototype robots that cooperatively perform cleaning, 6 welding and ultrasound NDT of large vertical steel structures [ ].

25


Figure 3: Weld inspection task on new construction cargo container ships showing varying surface curvatures, vertical and cross weld lines, use of scaffolding and internal strengthening plates on the hull The wireless NDT inspection robot, size: 600mm x 375mm x 340mm is shown in figure 4. It has been designed to perform long weld line inspection of new ship hulls and also repaired steel structures (external 78 and internal) of the type shown in figure 3 [ , ]. The robot achieves smooth and continuous movement, as well as excellent manoeuvrability, with a differential drive wheeled robot that uses permanent magnet adhesion. The magnets work over large air gaps of 20mm for the purpose of working on curved surfaces and overcoming small obstacles such as studs and bolts. The payload of the NDT robot is approximately 10kg. The robot has two sections connected by a hinge joint, with two wheels to drive the robot, and two omni-wheels, one in the front and one in the back, to support the robot. The two-section design enables motion through sharp angled corners presented in ship hulls with the back half maintaining strong holding force when the front half of magnet is lifted up. After the front magnets resume strong holding force, then the back magnet is lifted up to complete the transfer. The on-board robot controllers are controlled via an Ethernet network. An interface conversion module converts Ethernet to serial, IO, AD and I2C interfaces, allows the connection of different sensors and equipment and enables their monitoring or control via standard TCP-IP protocol. All the on-board modules are plugged to an Ethernet hub which is carried on the robot. The uplink of the hub is connected to a wireless bridge for Wi-Fi wireless communication with a central task manager. Two infrared distance sensors facing side ways, guide the robot along stiffeners such as those which arise on container ships.

26


Figure 1: The CROCELLS wall climbing robots performing both welding and weld inspection The NDT robot is able to follow the welding robot by using the infrared distance measuring sensors and by sensing the hot welding point with a thermal array sensor of eight thermopiles arranged in a row. It can measure the temperature of 8 adjacent points simultaneously. The sensor reads infra-red in the 2um to 22um range, which is the radiant heat wavelength. The driving wheels are made of aluminium hubs bonded with solid 65 IRHD polyurethane tyres with coefficient of friction of 0.9 on steel walls. The key benefits of the material are resistance to abrasion (non-marking), impact, cuts, and large range of operating temperatures. The polyurethane material has a long working life, good traction and is oxygen and ozone resistant. The NDT robot is required to perform real time inspection of long weld lines with 100% volume coverage, simultaneously with the welding process. Ultrasound phased array NDT with an Omniscan carried by the robot sends data wirelessly to a laptop for analysis. A scan from a weld test piece is shown in figure 5.

Weld defect visible on phased array scan

Figure 2: Weld NDT with phased array ultrasound

27


3

INSPECTION OF AIRCRAFT WINGS AND FUSELAGE Regular NDT of civil aircraft is mandatory. Aircraft wings and fuselage are tested for bond quality, corrosion, impact damage, and cracks around fastener holes. Pressurisation and de-pressurisation during takeoff/landing cycle causes stress fatigue at the rivets that hold the surface skin to its frame. The fatigue results in growth of radial cracks. There are approximately 2,000 rivets in a typical aircraft wing. NDT is mostly manual with limited area coverage. Techniques used are eddy current, ultrasonic and X-ray. The inspection is unreliable due to operator fatigue when performing100% inspection on large structures e.g. aerofoil and wings. Full coverage with more reliable methods such as X-ray is very expensive as the component has to be removed for radiography in shielded bays. Robotic deployment of the NDT techniques offers three operational and economic advantages: thoroughness, correctness, and records of the inspection. Portable C-scanner bridges, fixed by straps or suction cup, are available for semi-automation of ultrasonic field inspection enabled by Microelectronics and PC development. Flexible bridges have been introduced in recent years to deal with the complex geometry of aircraft structures e.g. PANDA (Tektrend, Quebec), MAUS (Boeing, St Louis), and the ISCAN (Fraunhofer Institute, Germany). Automated Non Destructive Inspector (ANDI) [Carnegie Mellon University], The Crown Inspection Mobile Platform (CIMP) [Carnegie Mellon University], The Autocrawler [AutoCrawler LLC] and the Multifunction Automated Crawling System (MACS), NASA Jet Propulsion Laboratory.

3.1 Climbing Robot 9

Our climbing robot called ROBAIR [ ] provides access to the top-side and under-side of aircraft wings and fuselage. The compact robot uses vacuum adhesion to adhere to a surface to climb on the topside and underside of aircraft wings and on all areas of the fuselage. It can carry a 18 kg payload of scanning arm plus NDT Sensors with a safety factor of 4, move over all surfaces with curvatures less than 0.3m and travel with a maximum speed of 1m/min. The mass of the climbing robot is 20kg with outer dimensions 518x518x180mm. Payload including umbilical mass is 18 kg. The umbilical comprises of two 10mm air tubes, 2 twisted pairs RS485, a 2 wire cable for 24VDC. The climbing ability is proven on test frames.

3.2 Non-Destructive Evaluation Sensors and Instrumentation The climbing vehicle carries a Flaw Detector and the scanning arm carries a sensor payload that is changed according to inspection requirements and comprises of a Acoustic Camera, Ultrasonic Phased Array, Eddy Current sensor, Thermo graphic Camera, Ultrasonic Dry Contact Wheel Probes and Defect Visualization Software. Ultrasonic wheel probes constructed of hydrophilic material have been shown to be capable of detecting rivet defects and the Phased array has been shown capable of detecting angled cracks. The Thermo graphic technique is best at detecting loose rivets while Eddy-current detection of angled cracks has been successfully demonstrated.

3.3 Robotic Scanner Mounted on the climbing vehicle is a Cartesian scanning arm with 4 degrees of freedom (X, Y, Z and Roll) that deploys NDT sensors in a work envelope of volume 400 x 400 x 180 mm. Control systems adapt to the changing dynamics of the inspection device as it operates on different structures e.g. on a fuselage or top or bottom of aircraft wing.

28


Figure 7: Inspection of loose rivets with thermo graphy. The climbing robot carrying a thermograph camera and heat source. Bottom: Loose rivet detected

29


Figure 8: Other NDT methods to detect cracks between rivets. On the right: Phased array ultrasound. On the right: Eddy current NDT detects slot between two rivets

4

INSPECTION OF STORAGE TANK FLOORS Storage tanks are normally inspected by opening them after every ten years. Tank outages are very expensive as the tank product has to be emptied by transporting and storing it at other locations. The tank has to be repeatedly cleaned so that it is free of toxic and explosive vapour before the tank is opened for inspection by human operators. A crude oil tank outage can take 8-9 months. Initial NDT is by MFL and more detailed inspection is by ultrasound. Huge savings in cost and inspection times could be obtained by performing in-service inspection of tank floors and walls with robotic devices. Our work has resulted in the development of prototype mobile robots [10] that can enter tanks through minimum manhole openings of 300 mm diameter to deploy a payload of Non Destructive Testing (NDT) sensors for the inspection of top and bottom corrosion on the tank floor. The robots are designed to operate in explosive and hazardous liquids such as crude oil, petroleum products, etc. To inspect the floor of clean storage tanks (size from about 2 to 20 metres in diameter) containing blended oil products and chemicals, for underside corrosion, magnetic flux leakage (MFL) is used for the initial inspection. Suspect areas are further examined using either vacuum box or magnetic particle inspection methods. To inspect larger crude oil storage tanks (20 and 100 metres diameter with construction from carbon steel and floating roofs), either double skin or pontoon type, with many manhole openings (for agitator entry). The preparation periods for entry and internal inspection are lengthy with 6-9 months required for removal of the oil, gas, and sludge banks. Another 3-6 months are required for the process of washing the tank clean of all oil and venting it before men can enter the tank. Tanks are inspected visually followed by MFL techniques to find problem areas. Ultrasound testing is used as a final method to validate the problem areas. Dependent on technique, annular floor plate thickness up to 35mm can be achieved with discrimination between topside and under floor corrosion. A number of robots that gain entry to tanks to perform cleaning 11 12 13 and inspection tasks have been developed [ , , ]. 14

Our prototype wall climbing robot called ROBTANK [ ] can perform ultrasonic non-destructive testing (NDT) while submerged in liquids. The robot is able to rotate through any angle within the full 360 degree maximum and can change surfaces from a floor to a wall and vice versa. It is designed to find application in the in-service inspection of large storage tanks to detect corrosion on tank floors and walls, and in the inspection of floating oil storage platforms that have first been emptied and then filled with water. It can also be applied without modification to inspect the submerged hull of a ship. ROBTANK is designed to be compact and lightweight so that it can be manually handled by one or at most two operators and can be inserted into restricted spaces through manholes of diameter 300 mm or more. It is equipped with an array of four ultrasonic wheel probes, four compression probes and two bulk-wave rotating probes look for corrosion thinning on the floor and walls up to half a metre ahead and under inaccessible floor areas, see figure 10. 30


Figure 9: A Tank farm and manual inspection of a floating-roof tank looking for corrosion and pitting defects on both topside and underside of tank floor

Figure 10: ROBTANK climbing a wall carrying a payload of NDT sensors The dimensions of the mobile robot are 200x200x500 mm. The maximum travel speed is up to 150 mm/sec. The flaw detector is able to measure internal and external corrosion with a thickness resolution of 1 mm on plate thickness ranges from 6 to 25 mm. Two servomotors provide the drive for the wheels of the vehicle while one propeller mounted on top of the vehicle provides the thrust force for adhesion to the wall. The on-board servo systems are controlled from outside the tank via a serial communications link. Trajectory control of the vehicle is by tele-operation via a Windows based software interface.

31


The umbilical comprises of two twisted pairs for serial full duplex communications to a remote station at a distance of 100 metres. The vehicle design incorporates a sealed, purged and pressurised central box where the servo motors, controller cards, NDT instrumentation (24 channel TD Scan Flaw Detector), and navigation sensors are carried on-board the mobile robot to guarantee explosion proof working conditions. The vehicle is able to travel on the tank floor while submerged in liquid (tests have been performed in water), change surfaces from the floor to a wall and vice versa, and climb the walls of a tank. It uses thrust from two propellers to provide vehicle adhesion to a vertical surface and hence is able to climb on all types of surface.

Figure 11:ROBTANK on the roof of a tank next to a manhole through which it is inserted and on the floor of a water tank.

Reflection of drain outlet in tank wall

B-Scan line with origin at centre

Top of tank wall

180°

0° Tank floor weld

Corner of tank floor

Drain outlet Welded stud on tank floor

Tank floor weld

270°

Figure 12: Floor inspection with rotating ROBULK ultrasound probes

32


Two probe arrays, each 30cm long with 15 to 20mm pitch, are mounted to the front and rear of the inspection robot. The minimum detectable area is a 6mm diameter flat-bottomed hole at a range of 3 mm. A surface coverage of 3m2 per minute and surface speed of 10 m per minute are realisable with this arrangement. The inspection system is able to measure plate thickness between 6-25mm with minimum thickness of 3mm. Two sets of 0º (in the front and rear of the vehicle) high efficiency twin wheel probes have been developed to cope with large crude oil tank inspection difficulties and environment conditions. They are designed to European Standard EN10160 (July 1999) for the UT examination of steel planar plates. Tests on single and twin crystal probes for scanning the surface with a fluid gap or direct contact, ability to monitor wall thickness despite changes in probe orientation, size of probe, frequency of element and coverage, and the influence of sludge, sand and other tank constituents resulted in the development of a Wheel probe system consisting of a high efficiency ultrasonic inspection twin wheel probe that in the preliminary laboratory tests showed a promising behaviour working in crude oil tank environment simulations. Two ultrasonic rotating bulk wave probes are mounted on the two sides of the robot to speed up the discovery of potentially corroded areas in the plate with a look forward distance of 50 cm in water. The probe is motorized and encoded to produce a radar type B scan plot, see figure12, to detect the edges of tanks, welds, etc and can therefore be used for navigation. The sound wave dives under unattached obstacles and can therefore inspect under striker plates. Wheel probes that can penetrate debris and sludge on the tank floor provide quantitative data at the required rate. A commercially available TD-Scan 24 channel flaw detector with dimensions of 170x60x104 mm is mounted in the purged box on-board the robot. The TD-Pocket integrates a pulser/receiver, A/D converter, encoder inputs (the requirement is for one bi-directional input to describe forward/backward travel), and 2 unidirectional encoders to control the LORUS probes). Software for data acquisition, display and analysis in all standard NDT formats is provided. The TD Pocket uses TTL signals from one of the robots incremental encoders to position stamp the NDT data. An industrial version of RobTank, certified as intrinsically safe and able to perform reliable NDT, could Save 80% of the average cost of inspecting a storage tank i.e. 56,000 Euro per tank Provide, after a few days of in-service inspection, an initial indication of state of the tank floor and buried tank walls Enable a tank operator to plan an outage for repair and could prevent a mandatory 10 year outage when floor has not suffered from corrosion.

5

INSPECTION OF TANKS FOR FLOATING PRODUCTION STORAGE OF OIL The other robot that we have developed for operation in liquids in storage tanks is the FPSO robot that is designed for inspecting tanks in off-shore oil operations - Floating Production Storage of Oil (FPSO), see figure 13. FPSO provides access to welds on stiffener plates inside oil storage tanks when the tank is either full of oil or emptied to the last few centimetres. It performs non-destructive testing of the welds using a number of NDT techniques. The robot is very compact, inserted through a manhole in the roof, and is able to swim to a test site on the floor of the tank. It is able to follow welds all the way along stiffener plates in a constrained space and find weld cracks and floor corrosion. It is designed to be intrinsically safe in Explosive Environments. Figure 14 shows the FPSO swimming robot in a water tank. Vertical motion is controlled by depth sensor feedback and active buoyancy control. Horizontal motion is with two independently controlled thrusters. The robot descends to the tank floor and moves on the floor using wheels to follow weld lines along stiffener plates and walls. The NDT techniques used are (a) ACFM for weld body inspection and for plate corrosion sizing between stiffeners (b) Ultrasound for weld toe inspection (using creep waves) and for plate corrosion detection (using plate waves).

33


Figure 13: Floating Production Storage of Oil (FPSO) requires NDT of welds on strengthening plates located on the floor of a tank The NDT payload comprises of ACFM probes for weld inspection (5 kHz with 8 sensors in 2 modules) and for corrosion sizing (50 kHz using 2 Bz coils). Two further sensors, a Sonatron S54008 plate wave sensor at 2 MHz and 65⁰ refracted angle and a dual creep wave sensor : RTD Crst4 at 4 MHz, Dual element and 80⁰ refracted angle. To change the direction of the FPSO robot during motion on the floor of a tank, its wheels are rotated by 90 degree to allow the robot to move in an orthogonal direction to its present position.

6

INSPECTION OF NUCLEAR PRESSURE VESSEL WELDS Reactor Pressure Vessels are inspected with an outage on average every 1-5 years. The inspection must not interfere with other maintenance tasks. NDT is performed with Ultrasonic and eddy current techniques. Figure 15 shows the RIMINI wall climbing robot that is designed to inspect shell welds from inside a nuclear reactor pressure vessel (RPV) while submerged in water. It provides access to nozzles to enable another pipe crawling robot (carried by the climbing robot) to enter the nozzle pipe to inspect a circumferential weld located 700mm inside the nozzle. The robot is designed to withstand large doses of radiation. Two DC motors provide the drive actuation, 3 triangular suction cups provide adhesion to the RPV wall (3 air motors provide suction cup actuation). Welds in the RPV are classified as circumferential welds and nozzle welds. Circumferential welds are located at - flange to upper shell, upper shell to middle shell, middle shell to lower shell and lower shell to bottom head. Nozzle welds are located at - nozzle to middle shell, nozzle to nozzle pipe (so called safe end). Flange ligaments are also inspected. Some reactor vessels - vertical welds of reactor shell and safety injection nozzle welds are included in inspection Most RPV have at least six nozzles, the number of nozzle welds to be inspected is 12, and the number of circumferential weld inspections is sixteen. Thus many days are required to inspect one RPV.

Figure 14: The FPSO robot swims to a test location and descends (using active buoyancy control) to the floor. It uses wheeled motion on the floor to move along strengthening plates and perform weld inspection

34


The most difficult inspection is the circumferential weld located 700mm inside each nozzle. The nozzle diameter is 760mm at its opening and tapers to 540mm at weld location.

Figure 15: Nuclear pressure vessel and nozzle weld inspection with the RIMINI wall climbing and pipe crawling robots The system uses a ultrasonic multi-element phased array sensor and systems with 3D focusing capability for fatigue cracks, inclusions and other flaws. inspection of concrete, brick and glass structures A wall climbing wheeled robot called VORTEX that we have designed is able to climb on most types of surfaces by creating a negative pressure by spinning an impeller at 20,000 rpm or higher to generate a vortex. Its dimensions are that of an A4 page and its mass is 1kg with an additional payload of 200g comprising of a camera system. The robot is suitable for visual inspection of non-ferrous surfaces. It climbs reliably on brick, concrete and glass surfaces. Work is progressing to understand the parameters that need to be optimized to increase the payload capability of this robot as it offers the ability to climb on most types of surfaces with wheeled motion. Figures 16 and 17 show the robot climbing on glass, concrete and brick surfaces.

35


Figure 16: The VORTEX robot shown climbing on glass, concrete and brick surfaces

36


1.

INSPECTION OF WIND TURBINE TOWERS AND BLADES There is currently great interest in finding solutions to the in-situ inspection of wind blades. Wind turbine farms for sustainable electric power production are being planned worldwide. The largest wind turbines planned for the future will generate 5MW and involve fibre reinforced composite (FRP) blades up to 100m in length. Wind blades are subject to enormous stresses, especially in storm conditions in offshore locations. At the same time the use of FRP in safety critical structures located in such extreme environments is relatively new and it is likely that structural defects of a previously unknown nature may arise. Effective regular inspection for structural integrity inspection is thus essential. Access to offshore wind turbine blades poses tremendous problems, danger to human operatives and costs in the event of blades having to be taken out of service and transported to shore for scheduled inspections.

Figure 18: RING robot climbs on pipes and towers Robotic in-situ blade inspection of offshore wind turbines is a promising solution. A climbing robot carrying a micro focus X ray source and digital detector could deploy radiography to test a blade. Computed axial X ray tomography has been identified as the optimal if not the only solution for identification of safety critical defects in the thickest blade sections. The weight of such an inspection system is very high, typically 200kg and typical cross sectional scanner dimensions of 1m x 2 m to encircle as blade, clearly involve very high destabilizing moments to be countered by the deployment robot. Our solution [15] is a climbing ring robot completely encircling a turbine tower (typically 3 meter in diameter). Because of the size and thus development costs of such a huge robot, we have designed and prototyped a small scale model. The key design innovation is that the adhesive forces between the robot and climbing surface are provided entirely by mechanical means rather than by using the usual methods of vacuum suction or magnetic force. Figure 18 shows the CONCEPT ‗ring‘ climbing robot, funded by the European Commission [16]developed from several modular frames that decrease in diameter from 4.5 m to 3 m at the top of the tower. The system has a Cartesian scanning arm to scan the blades from the top of the tower to the bottom. The blade can usually be rotated to change its ―pitch‖. This ability is useful to turn the blade in the radiation beam to enable 3D computation of a defect. The robot is then moved to a new position to obtain new results along the blade. The prototype has three modules which are completely identical and can be easily joined together to climb on any circumferential tube. The tower has a tapering radius. The robot is placed around the tower and it uses spring forces to grip it. Active force control could also be used to adapt to changing radius but this method has not been used here. Each module uses two motors, one for the drive motion and the other to turn the angle of the wheel so that the robot climbing trajectory is spiral. The robot has the capability to face the driver wheels in different angles which means that the robot can either climb along the tube, or with a certain pitch angle it can spiral 37


around the tube, or if the wheel is turned through 90 degrees then the robot will not climb but it will rotate around the tube in the same spot. The prototype has been built to a linear scale of 1:10 (for both the robot and test pipe) and tested successfully performing the three types of motion i.e. up/down, spiral, and rotation on the spot. The robot weight is 3kg, the payload capacity is 2kg with a safety factor of 2 and maximum speeds of climbing and circumferential motions are 10m/min. In the full scale model the cross sectional area over which adhesive forces between the wheels and turbine tower could be developed would increase by a factor of 100 (assuming the wheel widths and diameters to be scaled up by a factor of 10 and the payload capacity can thus be potentially increased in the same proportion to about 200kg, the target figure. However, if necessary, adhesion forces can always be augmented in the full scale design by the inclusion of a number of rare earth magnet arrays.

7

INSPECTION OF TIDAL STREAM GENERATORS AND BLADES Generation of energy from tidal streams is a fast developing industry with several commercial systems showing promise. Tidal stream generators are located in regions of fast-moving tidal flow and can be completely submerged, making access hazardous.

Figure 13: The turbine blades of SeaGen (Marine Current Turbines Ltd) Marine turbines, located on a sea bed, generate power from tidal flows with rotating blades similar in appearance to those of a wind turbine. A major advantage of these is that they can be located out of sight and deep enough not to obstruct shipping channels. Because water is about 800 times denser than air, tidal turbines generate more energy than wind turbines. However, they also experience turbulence and axial forces due to the velocity of the flow at a given location varying greatly across the actuator area with significant variations in loading across the actuator and associated fatigue and vibration problems. Inspecting turbine blades with robotics is another challenging task that has yet to be attempted but the need for it will grow in line with growth in their use.

8

INSPECTION OF PLATFORM MOORING CHAINS Offshore oil and gas exploration and production operations are being conducted in increasingly deeper waters from floating platforms which are moored to the seabed by chains.

38


Figure 9: Inspection of mooring chains. Top left: The concept Top right: Example of wear and corrosion on a chain link (from the Sea-bed Touch Down Zone). Bottom: Marine Growth after long term deployment of chain. The projected 23% increase by the end of 2011 compared to 2010 figures of large scale deployment of offshore Renewable Energy systems will rely upon similar mooring systems. 14-17% of Europe‘s 2011 total need for electricity (approximately 40GW) will come from offshore and deepwater platforms. Mooring chains are safety-critical systems which are subject to immense environmental and structural forces such as currents, oceans waves, and hurricanes. Other forces include impact with the seabed, abrasion, increased drag due to accumulation of marine organisms and salt water corrosion. Failure of one or more of these mooring lines can result in disastrous consequences for safety, the environment and production. Periodical inspection of chains systems is mandatory. It is usually done either outside the water that necessitates the decommissioning of production or in-water with the chains in situ. The in situ inspection is extremely dangerous for divers because the chain dynamics generate huge forces. The European MoorInspect project [17] will bring a step change in chain inspection systems through the development and introduction of robotics that provide access to each link of the chain for detection of fatigue cracks in the large chain links used in deepwater offshore facilities.. A vision system on the submersible robot will give early indication of problems. The robot will clean a link before strapping a Medium Range Ultrasonic Transducer collar to a cleaned area. Each link will be tested with medium range ultrasound guided waves.

39


9

CONCLUSION The mobile robots presented here are designed to provide access to inspection sites on very large structures and/or test sites located in hazardous environments. The robots deploy sensors to implement an appropriate technique from the full range of NDT techniques to find defects such as cracks, inclusions, lamination debonding and the extent of corrosion on steel structures. Robotic access both speeds up the inspection and reduces costs by eliminating the expensive and lengthy erection of scaffolding or the preparation of the site before humans can manually perform the inspection. Thus outage turnaround can be reduced or an outage prevented where the robotic inspection can be performed while the plant is in service. Robotic deployment of NDT is the only means of performing testing where the test site is located in hazardous and dangerous environments.

10 RECOMMENDATIONS There are numerous industrial inspection tasks requiring these types of inspection robots. Recent developments in cheap wireless control, mobile communications, improved battery technology and spatial positioning systems now offers the means to build small umbilical-free mobile robots that can be deployed cheaply and quickly to go to a remote test site, gather NDT data, stamp its position and have it analyzed in real-time by an operator sitting safely some distance away. Further research and development is required to develop robots to go inside petro-chemical storage tanks (while full of product) to inspect floors for pitting and corrosion, to climb on the hulls of steel ships to inspect hundreds of kilometres of weld, to inspect the walls of petro-chemical storage tanks for corrosion and weld integrity, to inspect nozzle welds inside nuclear pressure vessels, to inspect structures such as dams and bridges for cracks, to inspect overhead power cables, to internally inspect buried pipelines that are currently not reachable by intelligent pigs, to climb up off-shore wind turbine towers to inspect the blades, and to climb on aircraft wings and fuselage to detect for cracks and loose rivets.

11 ACKNOWLEDGEMENTS The support of the European Commission under projects Moorinspect FP7-SME-2011-1, 286976, COOPCT-2006-032949, and NMP2-CT-2005-017509 is gratefully acknowledged, and Nueva Granada Military University of Colombia for sponsoring Dr Leon Rodriguez.

12 REFERENCES [1] Topics On Nondestructive Evaluation (TONE), Volume 4, Automation, Miniature Robotics and Sensors for Nondestructive Evaluation and Testing, Technical Editor: Yoseph Bar-Cohen, ASNT, ISBN:1-57117-43-X [2] Special issue on NDT Robots, In Industrial Robot: An International Journal, Vol.37 , No. 5, 2010, Emerald Group Publishing Limited, ISSN 0143-991X [3] Sattar, T., Robotic Non Destructive Testing, Viewpoint: Robotic non destructive testing, Special issue on NDT Robots, In Industrial Robot: An International Journal, Vol.37 , No. 5, 2010, Emerald Group Publishing Limited, ISSN 0143-991X [4] Bridge B, Sattar T., Chen S, Khalid A , On the design of multi-task, compact, climbing robotic NDT systems for remote operation on large surfaces and in hazardous environments, Nondestructive Testing and Evaluation, 1997, Vol 3, pp 85-111. [5] Chen S., Sattar T.P., Khalid A., Bridge B. (1996) Design, development and performance Evaluation of a new pneumatically powered versatile wall climbing robotic NDT system suitable for hazardous environments, Proceedings of the 14th World Conference on NDT (14th WCNDT), Vol. 1-5, Ch. 566, pp 1023-1026. [6] European FP6 STREP programme CROCELLS Contract No. NMP2-CT-2005-017509, Climbing Robot Cell for Fast and Flexible Manufacture of large Scale Structures [7] Shang, J., Bridge, B., Sattar, T.P., Mondal, S., Brenner, A., Development of a climbing robot for the NDT of long weld lines, Industrial Robot: An international Journal, Volume 35 Issue 3, 2008, Emerald Group Publishing Ltd., ISSN 0143-991X 40


[7] Mondal, S.C., Bryan, B., Sattar T. P., Patel, T., Remote mobile vehicle and its suitable non destructive testing (NDT) inspection methods for the melt weld inspection, Proceedings of 13th International Conference On Climbing and Walking Robots and the Support Technologies for Mobile Machines, 2010 [8] Shang, J.; Sattar, T.; Chen S.; Bridge, B. Design of a Climnbing Robot for Inspecting Aircraft Wings and Fuselage, Industrial Robot: An International Journal, Vol. 34, No. 6, 2007, Emeral Group Publishing Limited, ISSN 0143-991X [9] Sattar, T.P., Leon-Rodriguez, H.E., Shang, J., Amphibious NDT Robots, Chapter 6 Climbing and Walking Robots, Towards New Applications, International Journal of Advanced Robotics Systems, 2006, ISBN 978-3-902613-16-5, 24 pages [10] Berger,A., Knape, B.,Thompson, B., Development of a Remote Tank Inspection (RTI) Robotic System, Proceedings of 1990 American Nuclear Society Winter Meeting,Washington D.C., November 1990 [11] Schempf H. (1994). Neptune-Above-Ground Storage Inspection Robot System, Proceeding of IEEE International Conference on Robotics and Automation, San Diego, Vols 1-4, Part 2. pg. 1403-1408 [12] Maverick Demonstration ―Submarine that goes in Gasoline‖, Solex Robotics, http://www.solexrobotics.com/Solex6.html [13] Sattar, T.P., Leon-Rodriguez, H.E., Shang, J., Amphibious NDT Robots, Chapter 6 Climbing and Walking Robots, Towards New Applications, International Journal of Advanced Robotics Systems, ISBN 978-3-902613-16-5, 24 pages, 2007 [14] Sattar, T.P., Leon Rodriguez, H.E. and Bridge, B. Climbing Ring Robot for inspection of off-shore wind turbines, Industrial Robot: The international journal of industrial and service robotics, Number 4 Mobile robots + CLAWAR, Vol. 36 No. 4, 2009, pp326- 330Emerald Group Publishing Limited, ISSN 0143-991X [15] European CRAFT project CONCEPT-INSPECT, ―Computerised Open Environment Portable Tomography‖, 6th Framework Programme CRAFT COOP-SME: COOP-CT- 2006-032949. [16] MoorInspect: Development of an advanced medium range ultrasonic technique for mooring chains inspection in water, European FP7 Collaborative projects, Networks of Excellence, Coordination and Support Actions, research for the benefits of specific groups (In particular SMES), Grant agreement no.: 286976

41


MOVING TOWARDS HIGH INCOME, SUSTAINABLE AND DEVELOPED NATION BY 2020: CHALLENGES, STRATIGIES AND OPPORTUNITIES Associate Prof. Dr. Ishak Aris Department of Electrical and Electronic Engineering, Faculty of Engineering, Universiti Putra Malaysia, 43400, Serdang, Selangor, Malaysia Email: [email protected]

ABSTRACT

Vision 2020 was unveiled in 1991 by former Prime Minister Dato‘ Seri (now Tun) Dr. Mahathir Mohamad. The main objective is to make Malaysia to become a fully developed country by 2020 according to its own mould. In order to be a balance developed nation Malaysia should not be developed only in the economic sense but it should also be developed in all aspects: technologically, politically, socially, spiritually, psychologically and culturally. The present government has introduced transformation programmes namely Government Transformation Programme (GTP) and Economic Transformation Programme (ETP)on top of the existing policies to make sure that the main objective of the Vision 2020 is achievable during the time frame. The cooperation between the government, public and private sectors are very importance to realise this vision. This paper will discuss the challenges, strategies and opportunities that may exist during the time frame of achieving the status of high income, sustainable and developed country. It will also discuss how advanced technologies can be used to support industries in Malaysia in order to fullfill the government‘s vision.

Keywords: High Income Nation, Vision 2020, Government Transformation Programme, Economy, Transformation Programme, Sustainable Industry and Green Technology.

42


ADVANCED MANUFACTURING PROCESSES

43


DEVELOPING COMPUTER-BASED 2D TOLERANCE ANALYSIS PROCEDURE Behnam Moetakef Imani1, Seyed Ali Hashemian2 CAD/CAM Lab., Dept. of Mechanical Engineering Ferdowsi University of Mashhad, Mashhad, Iran e-mail1: [email protected] e-mail2: [email protected] ABSTRACT Tolerance analysis methods become of extreme importance to improve product qualities. Considering the role of tolerance analysis in the design phase, this process is significantly taken into consideration by designers in recent decades. In this case, in order to increase the speed and accuracy of the process, it is necessary to apply an automatic approach to the process with minimum user interactions. The objective of the presented article is to develop a computer-based approach to the tolerance analysis procedure and implement a software-package in MATLAB® environment. The developed code is able to read a DXF file containing a 2D assembly and apply the tolerance analysis procedure to the assembly. Dimensional tolerances (manufacturing variables) and also key characteristics (variables of interest) of the assembly must be defined by the user through the process. The output of the process is the variation limits of the key characteristics, number of assembly rejects and the percent contribution of manufacturing variables. The capability and functionality of the presented package are investigated by some theoretical assemblies. Keywords: Computer-based tolerance analysis, Manufacturing variables, Key characteristics, limit. 2.

Variation

INTRODUCTION Tolerance analysis methods have a crucial role to associate product design with manufacture and become of extreme importance to improve product qualities as well as decrease the production costs. By nature, manufacturing processes are not exact and there are always unwanted variations such as tool wear, fixture imperfections, chatter, etc. which lead to deviations from part ideal dimensions. Such variations may cause costly problems during assembly process and result in unacceptable performance of the finished products. Therefore, tolerance analysis procedure is significantly taken into consideration by designers in recent decades. For all assembly products, there is always one (or more) critical feature(s) by which the overall performance of the assembly is measured. This feature is referred to as the assembly specification [1] or key characteristic (KC) [2] and could be the position of a point, a gap or a geometry feature in the assembly. Generally, KCs are the assembly features whose variation from nominal significantly impacts the final performance, cost or safety of the products and special control should be applied to them [3]. A common step in all tolerance analysis procedures is to express the variation of the key characteristics (assembly variables of interest) as a function of part variations (manufacturing variables) and calculate how sensitive those characteristics are to the input tolerances. In this regard, several methods such as Monte Carlo Simulations (MCS), Direct Linearization Method (DLM) and Global Coordinate Method (GCM) have been developed [4-6]. Once the KCs are expressed in terms of manufacturing variables, sensitivities of those specifications to each individual tolerance can be determined by deterministic Worst Case (WC) or statistical Root Sum Squared (RSS) methods [1]. Along the advance of engineering science, computerization techniques find a remarkable role to enhance the speed and accuracy of the engineering processes with minimum user interactions. In the area of tolerance analysis, several software-packages have been developed in order to automate the tolerance analysis procedure such as SolidWorks® TolAnalyst™, Sigmetrix® CETOL™ and VARATECH® SigmundWorks™. The objective of the current work is to develop a computer-based approach to the tolerance analysis procedure which presents the TA2D™ package using the MATLAB® graphical user interface development environment (GUIDE). The developed TA2D™ package is able to read a DXF file containing a 2D assembly and apply the tolerance analysis procedure to the assembly. Once dimensional tolerances (manufacturing variables) and also assembly specifications (KC‘s) are defined by the user, the output of the process is the variation limits of the KC‘s, their variation correlation (for more than one KC),

44


number of assembly rejects and the percent contribution of manufacturing variables. The capability and functionality of the presented package are investigated by some theoretical assemblies. The remainder of this paper is organized as follows. Section 2 reviews the principal theories of tolerance analysis. Section 3 describes the TA2D™ package and discusses all the steps and modules through an example. Section 4 illustrates a bivariate assembly model and finally, Section 5 draws the conclusions. 3.

PRINCIPAL THEORIES OF TOLERANCE ANALYSIS

Liaison Diagram The first step in creating a tolerance model for an assembly is to find the relationship between parts. This relationship is schematically defined by the assembly graph which replaces the parts with dots or balloons and contacts or joints between mating parts with lines. This graph which also known as the Liaison diagram [7] can determines how many loops (dimension chains) will be required to build the tolerance model. As an example for the well-known 2D stacked blocks assembly (Figure 4) the liaison diagram is depicted in Figure 5. This diagram contains one open and two closed loops. It should be noted that for a 2D assembly, there are five types of kinematic joints or contacts between mating parts which are illustrated in Figure 6.

Figure 4: 2D stacked blocks assembly [8].

Figure 5: Liaison diagram for the stacked blocks assembly [8].

Figure 6: Kinematic joints for 2D assemblies [8].

45


Direct Linearization Method (DLM) DLM is one of the most powerful methods in 2D tolerance analysis by which the assembly specifications (KC‘s) can be mathematically expressed in terms of manufacturing variables. This method, which was firstly presented by Marler [5], describes the assembly in terms of vector loops where each vector corresponds to a manufacturing dimension. Small dimensional variations are applied to the vectors, and tolerance accumulation will be estimated using a statistical approach. DLM simplifies complex nonlinear assembly equations to a linear system using first-order Taylor‘s series expansion. Once the relationship between parts in an assembly is found, vector loops can be defined through manufacturing variables. In this regard, a datum reference frame (DRF) is defined for each part to locate features. Vector loops are to be started from DRF of the first involving part and go through the manufacturing variables to reach the second part at the joint, and again go through manufacturing variables of the second part to reach its DRF. The process continues until the loop is closed, i.e. the vector chains reaches the first part‘s DRF. Vector loops for the current example are illustrated in Figure 7 and Figure 8.

Figure 7: Closed loops for stacked blocks assembly (left: loop1 and right: loop2).

Figure 8: Open loop for stacked blocks assembly.

46


Considering Figure 7, in addition to manufacturing variables (a, b …), there are some variables designated by ―u‖ and ―phi‖ which are defined through the assembling process. These dimensions are called kinematic or assembly variables [1] and remain unknown until the vector loop equations are solved. Not that in Figure 8, dimension ―Gap‖ is the assembly specification (KC). Each closed loop is described by summing the vector components in x and y directions, and summing the vector rotations. For the open loop only one equation is needed, since the gap has only a vertical component.

h1x  b  u1 cos  d cos1  a  0  h1 y  c  u1 sin  d sin 1  u2  0  h1    1  90  0 (1)

h2 x  a  r  r cos φ 2  R cos φ3  e cos φ1  u1 cos θ  b  0  h2 y  u3  r sin φ 2  R sin φ3  e sin φ1  u1 sin θ  c  0  h2θ  φ 2  φ1  φ3  θ  270  0 (2)

Gap  f  r  u3

(3)

By solving the above equation system, the nominal (mean) value of the assembly KC (Gap) can be computed. The sensitivity of the Gap to the manufacturing variables will also be calculated by:

S  C  DB1 A

(4)

If manufacturing variables are stored in vector X and assembly variable in vector U, matrices A, B, C and D (matrices of first-order derivatives of vector loop equations) can be determined as:

Aij = ∂hi /∂Xj , Bij = ∂hi /∂Uj , Cj = ∂Gap/∂Xj , Dj = ∂Gap/∂Uj

(5)

Using the RSS method, the standard deviation of the KC will be calculated as follows [9]:

 KC 

 S n

j 1

j

Xj



2

(6)

where σXj is referred to the one-third of tolerance limit of the j-th manufacturing variable. For problems concerning more than one KC, the sensitivity vector S becomes a matrix and standard deviations of Xj‘s are stored in variance matrix [VX]. Therefore, σKC is replaced by covariance matrix [ΣKC] whose diagonal components are KCs‘ variances and off-diagonal terms are their interrelations [10].

47


(7)

[ KC ]  [S ][VX ][S ]T Assembly Rejects

Computing the assembly rejects is one of the most common steps in applying tolerance analysis procedure to the assemblies. Once the number of assembly rejects is calculated, it is possible to determine if the tolerance model is capable of delivering the KC in the designed range. The number of rejects can be computed by integrating of the normal distribution of the KC within the desired limits [11]. The alternative solution is to estimate the rejects from Standard Normal tables by calculating the z-value associated with the normal distribution as expressed in equation (8) where UL and LL are referred to as upper- and lowerlimits.

zUL 

UL   KC

 KC

z LL 

,

LL   KC

 KC

(8)

Percent Contributions The percent contribution factor determines how each manufacturing variable contributes to the final assembly variation. By modifying the tolerance of those variables that have the greatest contributions, the tolerance of the KC can be modified and the number of rejects will decrease. The percent contribution of each manufacturing variable Xj is calculated by [8]:

% Cont X j 

S   j

 S   n

i 1

4.

2

Xj

i

2

(9)

Xi

TA2D™ PACKAGE As stated earlier in Section 1, TA2D™ package is developed using MATLAB ® GUIDE environment to facilitate the tolerance analysis procedure. The package has six major steps which are listed as: Import DXF file, Define dimensions and tolerances, Define assembly specification, Define joints, Create loops, and Results.

Import DXF file In the first step of the developed package as demonstrated in Figure 9, the CAD model of the assembly is imported as a DXF format. The DXF file which has a text format contains the information of all points, line and arcs and their corresponding layers in a standard routine [12]. Part information is extracted in this step and stored in a data structure in order to be used in the next steps.

48


Figure 9: TA2D™ Step 1, importing DXF file. Define dimensions and tolerances Once the CAD model of the assembly is imported, the next step is to define the manufacturing variables. As depicted in Figure 10, part dimensions are classified into 3 types of linear, angular and radius dimensions in accordance with CAD standards. The nominal values of the manufacturing dimensions are automatically extracted from the CAD model; however, their tolerances must be defined by the user in this step. Define assembly specification After defining the manufacturing dimensions and their corresponding tolerances, the next step is to define the assembly specification (KC). As illustrated in Figure 11, three types of assembly specification can be defined for an assembly: Gap, Position, or Orientation [10]. The gap specification is defined as the distance of two distinct features of the assembly like the distance of two points or a point and a line, etc. Position specification is the x-y coordinates of a specified point of an assembly or a mechanism. Finally the orientation specification represents the orientation of a line in the 2D space. In this step, after specifying the KC, its nominal value will be automatically determined from the CAD model and user must define the desired variation limits of the KC. Define joints As demonstrated earlier in Figure 6, there are five joint types for a 2D assembly. In this step, the part DRF‘s are automatically specified by the program, but the type and location of the joints should be defined by the user. The process is accomplished by choosing the appropriate joint type, selecting the involving parts, and clicking on the approximate position of the joint. The specified joints will be highlighted in the graphical window and their information listed in the joint table (Figure 12).

49


Figure 10: TA2D™ Step 2, defining dimensions and tolerances.

Figure 11: TA2D™ Step 3, defining assembly specification. Create loops This step of the TA2D™ package is developed according to the loop creating scheme (Section 2.2). Creating loops in this step is totally automatic and vector loops and their corresponding equations will be depicted in the graphical window (Figure 13).

50


Figure 12: TA2D™ Step 4, defining joints.

Figure 13: TA2D™ Step 5, creating loops. Results The final step in applying tolerance analysis procedure using the TA2D™ package is to calculate the results. This step is also applied automatically and the results contain normal distribution of the KC, assembly rejects and percent contributions will be demonstrated in the graphical window (Figure 14). The user is also allowed to modify the input tolerances to improve the results.

51


Figure 14: TA2D™ Step 6, results.

5.

BIVARIATE ASSEMBLY MODEL In many assemblies, it is important to control the variation of two or more assembly KCs simultaneously. Therefore, the tolerance analysis procedure must account for the correlation between the KCs in order to accurately predict the quality level of the assembly. Correlation arises when one set of manufacturing variables is involved in the delivery of more than one KC. This event is quite likely to occur because most products must deliver many KCs. Figure 15 illustrates a four-bar mechanism with important dimensions and their corresponding tolerances. In this assembly, the position of the coupler point (CP) is the assembly variable of interest (KC) whose variation from nominal must be within the desired limit. Since the position specification has two components in the 2D space, the problem is called a bivariate model. Therefore, the KC three-sigma variation limit, computed using equation (7), becomes an elliptical contour whose radii are the standard deviations of the CP in x and y directions multiplied by three [13]. The required closed and open loops for applying the tolerance analysis procedure are created using TA2D™ package and demonstrated in Figure 16 and Figure 17, respectively. As stated above, since there is only one set of manufacturing variables involved in the delivery of both x and y coordinates of the CP, a correlation will occur between variations in these two directions.

52


Figure 15: A four-bar mechanism with major dimensions and tolerances.

Figure 16: Closed loop model of the four-bar mechanism. Once the closed and open vector loops are created and the KC sensitivities are calculated, the final step in applying TA2D™ to the four-bar mechanism is to computing the results. Figure 18 represents the elliptical contour of the coupler point CP whose mean position is denoted by ‗x‘. The percent contributions of manufacturing dimensions in both x and y directions are also illustrated. The desired variation limit of the assembly KC, which becomes a circular contour in the case of bivariate distributions, is demonstrated in the result‘s graphical window as well as computed assembly rejects.

Figure 17: Open loop model of the four-bar mechanism.

53


Figure 18: TA2D™ results for four-bar mechanism. 6.

CONCLUSIONS The objective of the presented article was to develop a computer-based approach to the 2D tolerance analysis procedure. For this purpose, the TA2D™ package was developed using MATLAB® graphical user interface development environment (GUIDE). The TA2D™ package consists of six main steps which are organized in accordance with the principal theories of tolerance analysis and can be applied for problems concerning one or two assembly specifications. The applicability of the presented package was also investigated by two theoretical assemblies. The future work includes extending the current package to be applied for multivariate problems, developing a package for 3D assemblies and finally accounting for flexibilities in the assembly.

7.

REFERENCES [17] Chase, K. W. and Parkinson, A. R., "A survey of research in the application of tolerance analysis to the design of mechanical assemblies," Research in Engineering Design, 3, pp.23-37 (1991). [18] Whitney, D. E., Mechanical assemblies : their design, manufacture, and role in product development, Oxford University Press, New York (2004). [19] Thornton, A. C., "A mathematical framework for the key characteristic process," Research in Engineering Design, 11, pp.145-157 (1999). [20] DeDoncker, D. and Spencer, A., "Assembly Tolerance Analysis with Simulation and Optimization Techniques," SAE Technical Paper Series, Report No. 870263, (1987). [21] Marler, J. D., "Nonlinear Tolerance Analysis Using the Direct Linearization Method," M.Sc Thesis, Department of Mechanical Engineering, Brigham Young University, Provo, Utah (1988). [22] Gao, J., Chase, K. W. and Magleby, S. P., "Global Coordinate Method for Determining Sensitivity in Assembly Tolerance Analysis," Proceedings of ASME International Mechanical Engineering Conference and Exposition, Anaheim, California, (1998). [23] Bourjault, A., "Contribution a une Approche Methodologique de l'Assemblage Automatise: Elaboration automatique des sequences Operatoires," Thesis to obtain Grade de Docteur des Sciences Physiques, l'Universite de Franche-Comte, Besançon, France (1984). [24] Chase, K. W., "Tolerance Analysis of 2-D and 3-D Assemblies," Report No. 99-4, Department of Mechanical Engineering, Brigham Young University, Provo, Utah (1999). [25] Fortini, E. T., Dimensioning for interchangeable manufacture, Industrial Press, New York, (1967). [26] Law, M. J., "Multivariate Statistical Analysis of Assembly Tolerance Specifications," M.Sc Thesis, Department of Mechanical Engineering, Brigham Young University, Provo, Utah (1996). [27] Montgomery, D. C. and Runger, G. C., Applied Statistics and Probability for Engineers, 3rd edition, John Wiley & Sons Inc, New York (2002). [28] AutoCAD®, DXF Reference, Autodesk, Inc., (2007). [29] Johnson, R. A. and Wichern, D. W., Applied multivariate statistical analysis, 6th edition, Pearson Prentice Hall, Upper Saddle River, N.J. (2007).

54


OPTIMIZATION OF SUPPORT MATERIAL AND BUILD TIME IN FUSED DEPOSITION MODELING (FDM) USING CENTRAL COMPOSITE DESIGN METHODOLOGY (CCD) Pavan Kumar Gurrala1, Srinivasa Prakash Regalla 2 Birla Institute of Technology and Science-Pilani (BITS-Pilani), Hyderabad Campus Jawahar Nagar, Shameerpet Mandal, Ranga Reddy District Hyderabad – 500078, India e-mail1: [email protected] e-mail2: [email protected] ABSTRACT Fused deposition modeling (FDM) has evolved as one of the fastest growing layer manufacturing (LM) technology. The quality of the product depends on process parameters, the most important of them being raster angle, orientation, contour width and part raster width. In the present study, the influence of these parameters on two process quality parameters, namely, build time and the support material volume are studied on a part modeled on a FDM 200mc machine. A CCD methodology was employed and the results for these quality parameters were analyzed using the Design of Experiments (DOE). A CCD full factorial DOE methodology is employed and the results of 25 experiments are analyzed using Design Expert and Analysis of variance (ANOVA). A model equation for the quality parameters in coded and original factors has been developed. Comments on the results obtained and interaction effects are included at the end of the paper. Keywords: FDM; ANOVA; Layer Manufacturing; build time; CCD 8.

INTRODUCTION FDM is one of the most popularly used solid freeform fabrication (SFF) systems for various applications. The early days of industrial manufacturing following the industrial revolution saw manual prototyping, wherein the paper drawings of the products would be sent to shop floor for a skilled machinist to make the prototype by machining and allied manufacturing processes in several weeks‘ time. During the early 1980s, virtual prototyping followed by quicker physical prototyping became possible. More recently, technologies called as rapid prototyping (RP) and later standardized as Layered Manufacturing (LM) were invented and developed. In these techniques, the prototype is built layer-by-layer in a specialized machine directly from the solid model. Several LM techniques have been invented the most popular of which are the stereolithography (STL), layered object manufacturing (LOM), fused deposition modeling (FDM) and selective laser sintering (SLS). Details of these processes can be found in some good books, for example [1] or [2], and hence those process details will not be discussed here. Instead, only a brief account of FDM, on which the work of this paper is concerned, is presented below for completeness. A 3D solid model created in any of the modeling software is exported to the FDM‘s InsightTM software using the stereolithography (STL) format. The software generates the process plan and controls the FDM machine‘s hardware. The hardware for the FDM machine is represented in Fig. 1[3]. An ABS filament is fed through a heating element, which heats it to a semi-molten state. This heated filament is then fed through a nozzle and deposited on to the partially constructed part. As the material is extruded in a semi-molten state, the newly deposited material fuses with adjacent material that has already been deposited. The head then moves around in the x–y plane and deposits material according to the part geometry. The head also carries support material that is used whenever the part needs to support the build material. The platform holding the part then moves vertically downwards in the z-direction to begin depositing a new layer on top of the previous one. After a period of time, the head will have deposited a full physical representation of the original CAD file. The cooled model is then cured by treating with solvents to remove the support material [3].

55


Figure 1: Schematic of FDM process [3] Investigation into the effect of various process parameters of FDM on the part accuracy and other quality parameters of the part has been pursued in the near past and is documented in some research papers. Researchers have attempted to improve the dimensional accuracy by taking the processing parameters and its dependency on the FDM processed part. Different road patterns (or raster pattern) were considered and void density in FDM parts was estimated [4] in the literature. Anitha et al. [5] used Taguchi method to find the optimum strength and quality as affected by various parameters. Lee et al. [6] considered the effect of four parameters, namely, air gap (void space), raster angle, raster width and layer thickness on throwing distance (which is a measure of elasticity) of a flexible ABS object. They ignored the interaction effects of the parameters. Ang et al. [7] considered air gap(void space), raster width, build orientation, hatching direction and the scheme of build layers in a 2level full factorial DOE to study strength, porosity and modulus of elasticity. Zhang & Chou [8] investigated into the part distortion affected by tool path in different directions, namely, long raster, short raster and alternate raster. In their later paper [9], they considered the effect of road width (or raster width), scan speed and layer thickness on residual stresses and distortion of the part. Study using gray Taguchi method was carried out to investigate the effect of layer thickness, raster angle, raster width, air gap (void spacing) and part orientation on distortion of FDM parts [10]. They investigated both full factorial DOE and CCD. In their next paper [11], they investigated the mechanical properties of FDM parts as affected the same set of parameters. They carried out ANOVA to identify the functional relationship. The foregoing literature review indicates that good amount of work has been done to estimate the mechanical properties of FDM parts. Empirical models were derived for different other quality parameters also such as distortion as a function of the processing parameters with or without their interaction effect. Examining literature review, it was felt that there is scope for further insight into the optimization of the quality parameters of FDM parts as a function of as many process parameters as possible. Several quality parameters such as strength, distortion, build time and support material utilization need to be considered, optimizing all of which at the same time necessitates multi-objective optimization. In the literature, not much work is found focusing on multi-objective optimization. Therefore, the authors of this paper have set out to investigate into different quality parameters of FDM parts as functions of various factors in a multiobjective optimization paradigm. In this paper, the work on optimization of two quality parameters, namely, build time and support material volume as affected by five different factors of the process, namely, raster angle, orientation, contour width and part raster width on the product build time and the support material volume has been presented. An attempt was made to derive an empirical model between the above process factors and process parameters of build time and support material volume using 25 full-factorial DOE and response surface methodology. 9.

METHODOLOGY FDM machines, like other layered manufacturing processes, build part on a layer-by-layer basis. In FDM process, a computer generated design data is converted into a physical object via computer controlled robotic extrusion of a small fiber in the form of semi-liquid state. Molecular diffusion bonding process takes place between the interfaces between the extruded and the substrate fibers resulting in definite structural model. FDM uses separate nozzles for part material deposition and the support structure material deposition. Part material used is preferably that melts at a pre-selected temperature and solidifies quickly upon sudden cooling. The output of the part produced preferably depends on various processing parameters

56


viz., raster angle, orientation, contour /part width and part raster width. The above parameters are defined in brief as follows. a. b. c.

d.

Raster angle: The part raster angle as measured from the X-axis on the bottom part layer. All part and support raster angles are relative to this angle. Orientation: Part build orientation refers to the inclination of part in a build platform with respect to X, Y and Z axis. X and Y-axis are considered parallel to build platform and Z axis is along the direction of part build. Contour width: The width of the contour tool path that surrounds each of the part curves. Every part curve is filled by using at least one contour. Part raster width (or road width): The tool path width of the raster pattern used to fill interior regions of the part curves.

A full factorial DOE methodology is employed and the results for build time and support material volume of the 25 experiments are analyzed using Design Expert.The following design summary of FDM 200mc machine has been used. In the present study, the influence of these parameters on two process quality parameters, namely, build time and the support material volume are studied on a rotational part (Fig-2) modeled on the FDM 200mc machine. Increased frequency of changes in the design even during the prototyping phase can also be very well accounted in the RP. The build time shall be one of the major criteria that play a vital role in rapidly generating an initial prototype. In FDM, the build time depends on various fixed and variable parameters. Therefore it is very much important to optimize the build time. FDM uses support material to support the build material while modeling. In order to minimize the cost of production, the consumption of support material has to be minimized. The support material deposition path is dependent on various parameters. In the present paper, the variable parameters that critically affect the prototype build time and the volume of the support material viz., raster angle, orientation, contour width and raster width along with their mutual interdependence have been considered.

Figure 2: The rotational part considered for analysis in this paper The table-1 shows the level settings as applicable in FDM 200mc machine for different factors considered. In the present study, the influence of these factors on two process quality parameters, namely, build time and the support material volume are studied on a rotational part modeled on a FDM 200mc machine.

57


TABLE 1: LEVEL SETTINGS FOR ALL FACTORS Factor

Name

Level -1

0

1

A

Raster Angle (degree)

5

47.5

90

B

Orientation (degree)

5

47.5

90

C

Contour width (in)

0.012

0.021

0.0301

D

Raster width (in)

0.012

0.021

0.0301

A face centered CCD (α=1) full factorial DOE, methodology is employed and the results for build time and support material volume of the 25 experiments (table-2) are analyzed using Design Expert. Analysis of variance (ANOVA) was done and based on the ANOVA results; the model equation for the two quality parameters in both coded and original factors has been developed. The ANOVA details for support material volume (R1) are given in table-3. 10. RESULTS AND DISCUSSION Based on the ANOVA details for a final equation in terms of the coded factors have been obtained for each of the response variables; support material volume and the build time.

Support Material Volume (R1) = 1.06 + 3.72E-03*A + 0.079*B - 9.44E-04*C - 1.67E-04*D + 0.011*A*B + 1.88E-04*A*C + 1.88E-04*A*D - 1.06E-03*B*C + 1.88E-04*B*D + 1.88E-04*C*D - 1.92E-03*A2 0.065*B2 - 1.92E-03*C2 - 1.92E-03*D2

Build Time (R2) = 544.54 + 0.17*A + 40.72*B - 9.33*C - 128.28*D + 1.25*A*B + 0.13*A*C + 0.13*A*D - 0.62*B*C - 2.12*B*D + 3.75*C*D - 1.13*A2 + 29.87*B2 - 0.63*C2 + 54.87*D2

Values of "Prob > F" less than 0.0500 indicate that the model terms are significant. Based on this criterion from the values of tables 3 and 4 for the support material and build time respectively, the coded factors A, B, AB, B2 are significant model terms for the R1 and parameters and their interaction terms, B, C, D, AB, BD, CD, B2, D2 are significant model terms for R2.

TABLE 2: THE 25 EXPERIMENTS OF FACE CENTERED FULL FACTORIAL CCD Support material volume R1 (cm3)

Build Time R2 (min)

Std

Run

A (Degrees)

B (Degrees)

C (Inches)

D (Inches)

16

1

90

90

0.0301

0.0301

1.082

533

19

2

47.5

5

0.02105

0.02105

0.916

534

3

3

5

90

0.012

0.012

1.058

812

20

4

47.5

90

0.02105

0.02105

1.077

615

58


21

5

47.5

47.5

0.012

0.02105

1.059

551

1

6

5

5

0.012

0.012

0.921

727

25

7

47.5

47.5

0.02105

0.02105

1.060

544

18

8

90

47.5

0.02105

0.02105

1.069

545

15

9

5

90

0.0301

0.0301

1.052

530

9

10

5

5

0.012

0.0301

0.918

467

6

11

90

5

0.0301

0.012

0.903

699

23

12

47.5

47.5

0.02105

0.012

1.060

727

4

13

90

90

0.012

0.012

1.085

814

12

14

90

90

0.012

0.0301

1.085

545

24

15

47.5

47.5

0.02105

0.0301

1.060

472

13

16

5

5

0.0301

0.0301

0.920

456

10

17

90

5

0.012

0.0301

0.904

465

22

18

47.5

47.5

0.0301

0.02105

1.061

537

8

19

90

90

0.0301

0.012

1.082

786

5

20

5

5

0.0301

0.012

0.920

702

11

21

5

90

0.012

0.0301

1.058

543

17

22

5

47.5

0.02105

0.02105

1.051

542

14

23

90

5

0.0301

0.0301

0.903

454

2

24

90

5

0.012

0.012

0.904

724

7

25

5

90

0.0301

0.012

1.052

783

As the normal probability plot indicated in Fig 3 and 11 is of the "S-shaped" curve, it indicates that the residuals follow a normal distribution and the transformation of the response R1 and R2 provide a better analysis. From fig 3 to 9 and fig 12 to 17 of the response of support material volume and build time, it is clear that the orientation plays the vital role in minimization of the support material volume and the build time. Any increase of the parameters like thickness, raster angle, contour width and raster width tends to decrease the build time whereas increase in orientation angle increases the build time. It is evident from the fig 9 and fig 10 that the interaction of the contour width and part raster width shall influence the support material volume and is maximum when the values of C and D are respectively 0.02 and 0.021 with the A and B values of 47.5o. These observations are important because the orientation angle plays a dominant role in part strength; dimensional accuracy and surface finish [5].

TABLE-3: THE ANOVA TABLE FOR RESPONSE SURFACE QUADRATIC MODEL FOR SUPPORT MATERIAL VOLUME Source

Sum of Squares

df

Mean Square

F Value

p-value Prob > F

Model

0.14

14

9.90E-03

1138.77

< 001

2.49E-04

1

2.49E-04

28.69

003

A

59


B

0.11

1

0.11

12921.26

< 001

C

1.61E-05

1

1.61E-05

1.85

0.204

D

5E-07

1

5E-07

0.058

0.8153

AB

2E-03

1

2E-03

230.34

< 001

AC

5.63E-07

1

5.63E-07

0.065

0.8044

AD

5.63E-07

1

5.63E-07

0.065

0.8044

BC

1.81E-05

1

1.81E-05

2.08

0.18

BD

5.63E-07

1

5.63E-07

0.065

0.8044

CD

5.63E-07

1

5.63E-07

0.065

0.8044

A2

9.37E-06

1

9.37E-06

1.08

0.3237

B2

0.011

1

0.011

1253.61

< 001

C2

9.37E-06

1

9.37E-06

1.08

0.3237

D2

9.37E-06

1

9.37E-06

1.08

0.3237

Residual

8.69E-05

10

8.69E-06

Cor total

0.14

24

TABLE-4: THE ANOVA TABLE FOR RESPONSE SURFACE QUADRATIC MODEL FOR BUILD TIME Source

Sum of Squares

df

Mean Square

F Value

p-value Prob > F

Model

3.57E+05

14

25476.82

13009.1

< 001

A

0.5

1

0.5

0.26

0.6243

B

29849.4

1

29849.4

15241.9

< 001

C

1568

1

1568

800.66

< 001

D

2.96E+05

1

2.96E+05

1.51E+05

< 001

AB

25

1

25

12.77

051

AC

0.25

1

0.25

0.13

0.7283

AD

0.25

1

0.25

0.13

0.7283

BC

6.25

1

6.25

3.19

0.1043

BD

72.25

1

72.25

36.89

001

60


TABLE-4: THE ANOVA TABLE FOR RESPONSE SURFACE QUADRATIC MODEL FOR BUILD TIME CD

225

1

225

114.89

< 001

A2

3.27

1

3.27

1.67

0.2255

B2

2271.84

1

2271.84

1160.06

< 001

C2

1.02

1

1.02

0.52

0.4871

D2

7666.81

1

7666.81

3914.87

< 001

Residual

19.58

10

1.96

Cor Total

3.57E+05

24

11. FIGURES

Figure 3: Normal Plot of Residuals for R1

Figure 4: Orientation and Raster Angle on R1

61


Figure 5: Contour Width and Raster Angle on R1

Figure 6: Part Raster Width and Raster Angle on R1

Figure 7: Orientation and Contour Width on R1

62


Figure 8: Orientation and Part Raster Width on R1

Figure 9: Part Raster Width and Contour Width on R1

Figure 10: Contour plot of C and D on R1

63


Figure 11: Normal Plot of Residual for R2

Figure 12: Raster Angle and Orientation on R2

Figure 13: Raster Angle and Contour Width on R2

64


Figure 14: Raster Angle and Part Raster Width on R2

Figure 15: Contour Width and Orientation on R2

Figure 16: Part Raster Width and Orientation on R2

65


Figure 17: Part Raster Width and Contour Width on R2

12. REFERENCES [1] D. T. Pham and S. S. Dimov, Rapid Manufacturing: The Technologies and Applications of Rapid Prototyping and Rapid Tooling, Springer, UK, 2001. [2] S. P. Regalla, Computer Aided Analysis and Design, IK International Publishers, New Delhi, 2010. [3] S. Lee, S.G. Kim, H. J. Kim and S. H. Ahn, ―Measurement of anisotropic compressive strength of RP parts‖, Journal of Materials Processing Technology, vol. 187-188 (1), 2007, pp. 627-630. [4] J. F. Rodriguez, J. P. Thomas and J. E. Renaud, ―Characterization of the meso-structure of fused deposition of acrylonitrile butadiene styrene (ABS)‖, Rapid Prototyping Journal, vol. 6 (3), 2000, pp.175-185. [5] R. Anitha, S. Arunachalam, and P. Radhakrishnan, ―Critical parameters influencing the quality of prototypes in fused deposition modeling‖, Journal of Materials Processing Technology, vol. 118 (1), 2001, pp. 385-388. [6] H. Lee, J. Abdullah and Z. A. Khan, ―Optimization of rapid prototyping parameters for production of flexible ABS objects‖, Journal of Materials Processing Technology, vol. 169, 2005, pp.54-69. [7] K. C. Ang, K. F. Leong, C. K. Chua and M. Chandrasekaran, ―Investigation of mechanical properties and porosity relationships in FDM fabricated porous structures‖, Rapid Prototyping Journal, vol. 12, 2006, pp.100-105. [8] Y. Zhang and K. Chou, ―Three-dimensional finite element analysis simulations of FDM‖, Proc. IMechE, J. Engg. Manufacture, vol. 220, 2006, pp. 1663-1671. [9] Y. Zhang and K. Chou, ―A parametric study of part distortion in FDM using three dimensional finite element analysis‖, Proc. IMechE, J. Engg. Manufacture, vol. 222, 2008, pp. 959-967. [10] A. K. Sood, R. K. Ohdar and S. S. Mahapatra, ―Improving dimensional accuracy of FDM processed parts using gray Taguchi method‖, Materials and Design, vol. 30, 2009, pp.4243-4252. [11] A. K. Sood, R. K. Ohdar and S. S. Mahapatra, ―Parametric appraisal of mechanical property of FDM processed parts‖, Materials and Design, Vol. 31(1), 2010, pp.287-295. [12] Douglas C. Montgomery, Design and analysis of experiments. 5th ed. Singapore: John Wiley & Sons Pvt. Ltd., 2003.

66


CONDUCTIVE FUSED DEPOSITION MODELLING: A STEP FORWARD FOR ROBOTICS? Diegel, O.1, Singamneni, S.1, Huang, B. 1, and Gibson, I.2 1

Auckland University of Technology, Centre for Rapid Product Development Auckland, New Zealand e-mail1: [email protected] 2 National University of Singapore, School of Mechanical Engineering Singapore e-mail2: [email protected]

ABSTRACT This paper describes research to develop a novel Fused Deposition Modeling (FDM) process in which the layers of material that make up the part are deposited as curved layers instead of the conventional flat layers. This technology opens up the possibility of building complex curved plastic parts that have conductive electronic tracks and components printed directly as part of the plastic component. It is not possible to do this with existing flat-layer additive manufacturing technologies, particularly on parts that are curved, as the continuity of a circuit would be interrupted between the layers. The elimination of the flat PCBs that are used in most electronic products creates a new type of product in which the housing of the product becomes its electronic circuit. This has implications for mechatronics and robotics in which dealing with large wiring looms becomes an issue of both space and flexibility. Keywords: Additive Manufacturing, Fused Deposition Modeling, Conductive 3D printing. 13. INTRODUCTION Additive Manufacturing (AM), which, in the past, was referred to as Rapid Prototyping (RP) is an additive fabrication technology which creates complex 3dimensional models in short times: a 3D computer model is cut into thin 2D slices; these are transferred to a machine which stacks the thin flat layers sequentially to recreate a physical version of the Computer Aided Design model [1].

The additive manufacturing process begins by taking a 3D computer generated file and slicing it into thin slices (commonly ranging from a few microns to 0.25mm per slice depending on the technology used). The additive manufacturing machine then builds the model one slice at a time, with each subsequent slice being built directly on the previous one. The technologies differ mainly in terms of the materials they use to build the part, and the process used for creating each slice of the model.

Some of the earlier additive manufacturing processes, which were only able to make plastic-like parts, are now producing parts in metals such as titanium, and even stainless steel [2]. Not only is the choice of materials and processes increasing, but the last few years have seen a significant reduction in the cost of these technologies.

Fused Deposition Modeling (FDM), the core AM technology used in this particular project, works by extruding a thin filament of plastic as the nozzle of the machine traces each slice. This is then repeated for each subsequent slice of the model. It, in fact, uses 2 nozzles: One for the part material and another for the support material which is used to support overhanging parts. [2]. The parts currently produced by FDM systems are reasonably strong plastic components that are well suited to basic functional testing and can easily be sanded and painted to reproduce the aesthetics of the production product thus also making them useful for consumer testing.

67


Though each AM technology has advantages and disadvantages over the others, one of the weaknesses common to all current flat-layer RP technologies is a relatively poor surface finishes caused by the ‗staircase‘ effect on curved surfaces, and many of the technologies suffer from a lamination weaknesses in a direction perpendicular to the layer direction (Fig. 1). If smooth surfaces are required for the component the staircase effect can sometimes require substantial post-processing of the part (sanding and polishing) in order to produce smooth surfaces.

Figure 1. The staircase effect and lamination weakness problems caused by conventional flat-layer rapid prototyping. This paper looks at the application of curved-layer FDM [3] (Fig 2) for producing plastic components with integral conductive tracks that allow for the elimination of wiring or printed circuit boards from products.

Figure 2. A curved-layer part. This technology may open up entirely new possibilities of building complex curved plastic parts that have conductive electronic tracks and components printed directly as part of the plastic component. It is not possible to do this with existing flat-layer additive manufacturing technologies, particularly on parts that are curved, as the continuity of a circuit would be interrupted between the layers.

The elimination of the flat printed circuit boards (PCBs) and possibly even some of the electronic components, such as transistors, that are used in most electronic products creates a whole new type of product in which the housing of the product becomes its electronic circuit. This, in turn, could revolutionize the field of product design which would no longer be constrained by having to design around flat PCBs.

The ability to print conductive tracks directly as part of a plastic housing has quite dramatic implications for fields such as mechatronics and robotics in which dealing with large wiring looms becomes an issue of both space and flexibility. If the wiring loom can become part of the plastic shell of the robotic system, then all the extra space can be more efficiently used for additional sensors, or the system can be reduced in size. If one takes a humanoid robot, for example, just the motors from the ankles to the waist will number around ten, which can mean a wiring loom which, at the waist will have at least twenty or more wires depending on the type of motors being used. Such a thick wiring loom can restrict the ability of the robot to bend at the waist. 14. CURVED LAYER FUSED DEPOSITION MODELING This project began with the development of an additive manufactruing machine capable of constructing a part by depositing the layers of material as curved layers instead of the current flat layers. This new process could be named Curved-Layer Fused Deposition Modeling (CLFDM).

68


The concept behind the technology is as follows: A substructure of ‗support material‘ to the curved part is first created through existing flat-layer methods using a soluble, or removable, support material. This support structure forms the base onto which the curved layers of product material can then subsequently be deposited by having the deposition head precisely follow the contour of the part (Fig. 3). The effect of these curved layers would be to both eliminate the staircase effect altogether, as well as removing the inherent lamination weakness in the direction of the layers.

The bulk of the research being carried out in FDM additive manufacturing at different universities has been related to investigating alternative materials for FDM (and other AM methods) and working with a variety of materials including ceramics and metals [4], high performance thermoplastic composites [5] and metal/polymer composites [6]. While special FDM systems have been designed for experimental deposition of different types of materials with different techniques and much work has been done on the analysis of the mechanism of deposition [7, 8], very little research has been done on depositing material as curved layers. The literature on RP reveals one research project in which the LOM (Laminated Object Manufacturing) process was used to create curved layers [9] at the University of Dayton in the USA but the results were limited by the ability to evenly stretch a material over a curved mandrel and the very small range materials that could be used in the process.

Figure 3. Support material deposited as flat layers to form structure onto which build material can then be deposited as curved layers. This CLFDM technology opens up an entirely new possibility of building complex curved plastic parts that have conductive electronic tracks and components printed directly as part of the plastic component. It is not possible to do this with existing flat-layer additive manufacturing technologies, particularly on parts that are curved, as the continuity of a circuit would be interrupted between the layers (Fig. 4). With curved-layer fused deposition modeling (CLFDM) this problem is removed as continuous filaments in 3 dimensions can be produced, allowing for continuous conductive circuits.

Figure 4. Curved layer vs flat layer for conductive polymers. The elimination of the flat printed circuit boards (PCBs) and possibly even some of the electronic components, such as transistors, that are used in most electronic products creates a whole new type of product in which the housing of the product becomes its electronic circuit. This, in turn, could revolutionize the field of product design which would no longer be constrained by having to design around flat PCBs. It opens up new possibilities for miniaturization and could lead to a new paradigm in direct digital manufacturing in which the cost and size of many electronic components no longer affects the product as they are simply printed as part of the products plastic housing.

69


15. CLFDM SYSTEM OVERVIEW Two CLFDM systems were developed in parallel at the National University of Singapore and at Auckland University of technology in New Zealand. On the system developed at NUS in Singapore composite material was extruded using an in-house, screw feed system mounted on a Sony Robokids Cartesian robot (called the Screw Extrusion System (SES), shown in fig. 5. One of the objectives of the NUS project was to investigate whether the addition of fibers to the polymer could be used to achieve enhanced part strength.

Figure 5. The Screw Extrusion System developed at NUS. Since the curved surfaces are normal to the load direction, this makes it feasible to investigate the use of fibre reinforcement within the build material. In conventional FDM, the inclusion of fibres would not make sense since loads may be applied to have the effect of separating the layers. In curved FDM, the fibres would make it possible to spread the load over the surface. This would be particularly true if subsequent layers were built in a different direction to provide a simple weave pattern. Work carried out at the National University of Singapore has focused on the effect of inclusion of short wood fibres into a polypropylene matrix [10]. It was found necessary to include a coupling agent to facilitate a good bond between the two materials. Test samples were found to be approximately 30% stronger under tensile loading. Little difference was noted under compression load. It is hypothesized that further improvements can be made should a higher temperature filler material be used. The wood fibre showed signs of degradation caused by the elevated temperature inside the heated chamber of the extruder. Further research is currently being conducted using short glass fibres. Should these fibres indicate an improvement over regular particulate fillers, the research will continue using biodegradable polymers and calcium phosphate fibres that could be used for tissue engineering applications. The curved layer rapid prototyping machine used at Auckland University of Technology was built by modifying an existing Fab@Home desktop RP machine (Fig. 6). The Fab@Home machine consists of an XY axis gantry type system that moves a dispensing head along a preprogrammed path. This constitutes a relatively low-cost apparatus that is ideal for doing development work on rapid prototyping technologies, or for other research work. The platform provides a Z axis which, on the standard machine, moves a build platform down by a unit of measure after each XY slice is completed. The standard dispensing head that is included with the machine allows for the dispensing of material from a syringe. The motor control system consists of 4 stepper motors, a Xylotec XS3525/8S4 Stepper Motor Driver Board, and an Olimex LPCH2148 Microcontroller Board. On the mechanical side of the design, a new deposition head was designed and built with an extruding unit that allows a filament of molten plastic to be extruded. This was to allow for the eventual production of more durable parts than that allowed by the materials that could be deposited through the syringe system. This head is now being improved to have 3 depositions heads. One for support material, one for ABS, and one for conductive polymer material.

70


Figure 6. Fab@Home desktop RP machine. The standard machine comes with a single syringe deposition head, thus only allowing the deposition of one type of material at a time. To further increase the number of deposition material options, the single syringe deposition head was replaced with a two-syringe head that allowed the system to deposit both the support material and the build material without needing to stop the build operation in order to change from the support material to the build material. The addition of an extra syringe meant the addition of an extra syringe drive motor which, in turn, meant the addition of an extra motor control board. The modular design of the system, however, meant that the addition of extra motors or an extra control board was not a great problem. A number of materials were initially tested for the proof of concept stage of the project. These materials included silicone, icing sugar, RTV sealant, light curing epoxy and standard two pot epoxy all of which were able to produce parts, though of varying quality. The material finally selected was FabEpoxy, a special epoxy formulated for the machine to be thixotropic, so it does not flow after extrusion. The standard Fab@home machine is designed to receive a set of tool-path commands contained in a standard text file. The tool-paths consist of a series of xy coordinates that define how the deposition head moves for each flat slice. At the end of each slice program a z control command is sent which moves the build platform down by one slice height increment. The program then continues with the next slice of the model. The PC software provided with the machine takes an stl file, the defacto standard file format for RP applications, of the 3D part to be produced and slices it into flat slices. From these slices the software derives the xy coordinates for the tool path. The text file containing the comma separated values (CSV) of the xy coordinates are then sent to the Olimex LPCH2148 microcontroller over USB. This allowed for an easy method of creating curved tool-paths simply by including the z coordinate with every set of xy coordinates in a set of CSV data, and allowed the deposition head to be dynamically controlled in any of the 3 axis much like a conventional 3 axis CNC machine. New software was written, in Matlab, that accepted an stl file of a curved part. STL is the accepted de facto international standard for rapid prototyping machines, and it was therefore important that the software for this system be able to directly import files in this format. The software used a simple algorithm to split the part up into the real component and its support material structure by examining the bottom most surface of the part. Any section of that surface that was not at the zero point of the part was considered to have support material below it. The support material component of the part was then put through an algorithm that sliced it up into flat layers spaced, in the case of the Fab@Home machine, at 1mm spacings. This resolution was a variable that could be reduced or increased as needed. The algorithm started at the bottom surface and created a new flat plane above the first surfaces spaced away from the first plane by whatever thickness variable had been set.

71


Figure 7. Matlab Program for flat support material structure. Points were located wherever this plane intersected the bottom surface of the curved part, and these points were used as the extremities of the new flat layer that was to be created. The process was then repeated by adding a new plane above the previous one, until no new intersecting points could be added. A separate algorithm was then written to take the real component part of the stl file and split it up into curved layers, also spaced 1mm apart. By treating the bottom surface of the model as an infinitely thin geometry, x, y and z, coordinates could simply measured from the model slice and used to approximate the 3D geometry for that slice. A variable was introduced into the algorithm that determined how close any x, y and z sets of coordinates were to each other. This, in effect, determined the resolution of the tool-path. In the initial trials, this resolution, like the z resolution, was set to 1mm. From the stl input data, the Cartesian coordinate values of each point were individually recorded in the form of matrices. A matrix extension procedure was used, which made the boundary conditions of the surface lift offset available. The extension direction used was perpendicular to the filament deposition direction, and along with main deposition tool path direction. After the extension procedure, a new extended M by (N+2) matrix was obtained which compared to the original M by N matrix, as shown below.

Original Matrix:

Extended Matrix: Then, A Four Vector Cross Product (FVCP) algorithm was used to process the way in which each subsequent layer was offset form the previous surface. The FVCP algorithm used four different vectors to solve the new locations of every single point in the offset layers. The four vectors were formed by a point (P0) (which is the point being offset), two adjacent points (P1, P2) in the X axis direction and two presumption points (P3, P4) as shown in the diagram below. P1 and P2 were used to calculate the positions of the new lifting points, whereas P3 and P4 were used to define that the new offset point was on the plane which P0, P2 and P3 were on. This, in effect, determined the direction normal to the surface being worked on. The four vectors formed by the equations are shown below.

Then, the direction of the new vector was required. As the new vector is a combination of four different vectors, a further calculating procedure was needed.

Where

and

72

represented the offset direction.


After combining these vectors, the final offset vector was obtained. The offset position of new point was given by the equation below:

Figure 8. Matlab Program for curved build material structure. Using the same procedure, every point on the surface was calculated and the offset surface was generated from these points. A fourth algorithm was then used to combine the results of the second and third algorithm into a single text file containing first the tool-path for the support material, and then that for the build material. The order of the combined file was critical, as the support material needed to be printed in flat layers before the build material was to be deposited as curved layers. The text file was then sent to the machine‘s microcontroller and used to control the appropriate x, y or z stepper motor to build first the support material structure, and then the real component on top of the support structure. The extrusion head, in the initial tests, was kept extruding at a constant rate.

Figure 9. Matlab Program support and build material structures. Through these systems, parts were successfully produced that demonstrated the principle of curved-layer fused deposition modeling.

73


Figure 10. curved layer part produced by modified Fab@home rapid prototyping machine.

Figure 11. curved layer part produced by National University of Singapore machine.

(a)

(b)

(c)

Figure 12. Curved models using the SES made from short fibre reinforced composite material. Figures a and b show samples using a simple foam support. Figure c shows crossover paths with the support removed. 16. FURTHER WORK Work is currently underway to allow the CLFDM system described above to print parts that contain tracks of conductive material within the plastic part itself. The conductive material project is being undertaken along two routes. The first is to extrude a polymer mixed with carbon-black (initial tests have shown this material can be successfully extruded by the system), and/or other conductive elements (such as carbon nanotubes), while the second is investigating the possibility of ink-jetting a conductive ink onto the plastic substrate. The conductive polymer system has the advantage of using the same deposition mechanism for all 3 materials, while the inkjet system has the advantage of printing potentially more complex circuits faster than the polymer system. It would also have the potential of allowing other inks with varying properties (insulation, heat transfer, etc.) to be printed onto the plastic substrate.

74


The development of the triple-material deposition head (for support material, polymer material and conductive material) is expected to be complete by the time this paper is presented, and proof-of-concept parts are expected to be produced by the end of 2011. Work is also underway to develop effective methods of connecting components to such integral conductive tracks. In parallel with the development of the electromechanical systems, work is also under way on the software to convert 2 dimensional PCB designs into 3 dimensional paths. Consideration will also need to be given to double-sided or multi-layer PCBs if it is possible to develop an effective means of connecting conductive tracks on different layers. 17. CONCLUSION This paper describes a novel process referred to as Curved-Layer Fused Deposition Modeling. This new technology has the potential to allow the printing of conductive tracks integral to a product, which is not possible with conventional flat-layer technologies. This opens up a new field of design in which products can ultimately be designed without the encumbrances of having to design around PCBs and wiring looms. This could prove to be of particular benefit to the field of mechatronics and robotics in which large and complex wiring looms can become a design encumbrance. Proof-of-concept machines were built and software algorithms were written that allowed the system to create parts in which support material was first deposited as conventional flat layer structures, and build material was then deposited over the support structure as curved layers. The initial components built by the machine successfully demonstrated the proof-of-concept of Curved-Layer Fused Deposition Modeling. The creation of the research platform now opens the field to further areas of investigation into curved-layer fused deposition modeling, the first of which is the printing of conductive polymer tracks as part of the other normal polymer tracks that make up the part. 18. REFERENCES [1] [2] [3]

[4]

[5] [6] [7] [8] [9]

[10]

Chua, C.K., Leong, K.F., Rapid Prototyping: Principles and Applications. (2nd ed). World Scientific Publishing Co, Singapore, 2003 Wohlers, T. Wohlers Report 2005, Worldwide progress report on the rapid prototyping, tooling, and manufacturing state of the industry, Wohlers Associates, 2005 S. Singamneni, O. Diegel, B. Huang, I. Gibson, R. Chowdhury, Curved Layer Fused Deposition Modeling, Rapid Product Development Association of South Africa (RAPDASA) 2008, 9th Annual International Conference on Rapid Product Development, South Africa, 2008 Agarwal M.K, Van Weeren R, Bandyopadhyay A, Whalen P.J, Safari A, Danforth S.C, 1996, ―Fused deposition of ceramics and metals: An overview,‖ Proceedings of Solid Freeform Fabrication Symposium, The University of Texas, Austin, p 38592. Gray R.W, Baird D.G, Jan Helge Bohn, 1998, ―Effects of processing conditions on short TLCP fiber reinforced FDM parts,‖ Rapid Prototyping Journal, 4(1), p 1425. Masood S.H, Song W.Q, 2004, ―Development of new metal/polymer materials for rapid tooling using fused deposition modelling,‖ Materials and Design, 25, p 587594. Tseng A.A, Tanaka M, 2001, ―Advanced deposition techniques for free form fabrication of metal and ceramic parts,‖ Rapid Prototyping Journal, 7/1, p 617. Anna B, Lauren S and Seluck I.G, 2005, ―New developments in fused deposition modelling of ceramics,‖ Rapid Prototyping Journal 11/4, p 214220. Klosterman D.A, Chartoff R.P, Osborne N.R, Graves G.A, Lightman A, Han G, Bezeredi A, Rodrigues S, 1999, ―Development of a curved layer LOM process for monolithic ceramics and ceramic matrix composites,‖ Rapid Prototyping Journal, 5/2, p 6167. Yuan, L., Gibson, I., ―A Framework for Development of a Fibercomposite, Curved FDM System‖, Proc. Int. Conf. on Manufacturing Automation, ICMA‘07, 2830 May, 2007, Singapore, pp93102.

75


SELFOPTIMIZATION IN ADAPTIVE ASSEMBLY PLANNING Daniel Ewert, Daniel Schilberg, and Sabina Jeschke Institute of Information Management in Mechanical Engineering of the RWTH Aachen University Aachen, Germany e-mail1: [email protected] ABSTRACT To improve production especially in high wage countries, new concepts are needed which allow for more flexible and more robust production processes. One promising approach is the automation of assembly planning. In this paper, we present a hybrid approach for autonomous assembly planning that allows fast adaption to changing products as well as a flexible assembly. Due to the use of cognitive techniques, especially learning, the presented approach can deal with changes in the material flow as well as changes in machine performance. Keywords: Self optimization, Assembly planning, Cognitive production systems 1.

INTRODUCTION In the last years, production in low-wage countries became popular with many companies by reason of low production costs. To slow down the development of shifting production to low-wage countries, new concepts for the production in high-wage countries have to be created [1]. One promising approach is to automate planning processes preceding the actual production. This would first result in reducing the costs for planning, and secondly it would allow for easily switching between products or at least allow for a high number of variants, and hence enabling more adaptive production strategies. Automatic planning processes would also allow to quickly adapt to changes within the production system, e.g. malfunction of machines, lack of materials or similar. This paper presents our approach for a cognitive control unit (CCU) which is capable to autonomously plan and execute a product assembly by relying entirely on a CAD description of the desired product. The CCU is evaluated with the robot cell depicted in Figure 19. This setup simulates a nondeterministic production environment [2]: Only Robot2 is controlled by the CCU. Robot1 independently delivers parts in unpredictable sequence to the circulating conveyor belt. The parts are then transported into the grasp range of robot2 who then can decide to pick them up, to immediately install them or to park them in the buffer area. Incoming Parts

Leaving Parts Photo Sensor V1

Switch Photo Sensor

Robot Photo 1 Sensor Light Sensor

V1

Buffer

Assembly Area V1

V1

V=0

Photo Sensor

Photo Sensor

Robot 2

Figure 19 Schematic of the robot cell

76


The scenario also incorporates human-machine cooperation. In case of failure, or if the robot cannot execute a certain assembly action, the CCU is able to ask a human operator for assistance. To improve the cooperation between the operator and the machine, the operator must be able to understand the behavior and the intentions of the robot [3]. Therefore, machine transparency is a further major aspect in our concept. 2.

RELATED WORK In the field of artificial intelligence planning is of great interest. There exist many different approaches to planning suitable for different applications. Hoffmann et al.[4] developed the FF planner, which is suitable to derive action sequences for given problems in deterministic domains. Other planners are capable to deal with uncertainty [5], [6]. However, all these planners rely on a symbolic representation based on logic. The corresponding representations of geometric relations between objects and their transformations, which are needed for assembly planning, become very complex even for small tasks. As a result, these generic planners fail to compute any solution within acceptable time. Other planners have been designed especially for assembly planning and work directly on geometric data to derive action sequences. A widely used approach is the Archimedes system by Kaufman et al [7] that uses And/Or-Graphs and an ―Assembly by Disassembly‖ strategy to find optimal plans. U. Thomas [8] follows this strategy, too, but where the Archimedes system relies on additional operator-provided data to find feasible subassemblies, Thomas uses only the geometric information about the final product as input. However, both approaches are not capable of dealing with uncertainty.

3.

3.1

AUTONOMOUS ASSEMBLY PLANNING

Hybrid Assembly Planning The overall task of the CCU is to realize the autonomous assembly in a nondeterministic environment. The availability of parts or the sequence of their arrival in the assembly area of the robot cell cannot be predicted as well as if an invoked assembly action is successfully executed. While assembly planning is already hard even for deterministic environments where all parts for the assembly are available or arrive in a given sequence [8], the situation becomes worse for unpredictable environments. One approach to solve the nondeterministic planning problem would be to plan ahead for all situations: Prior to the assembly all plans for all possible arrival sequences are computed. However, this strategy soon becomes unfeasible: A product consisting of n parts allows for n! different arrival sequences, so a product consisting of 10 parts would result in the need to compute more than 3.6 million plans. Another approach would be to replan during the assembly every time an unexpected change occurs in the environment. This strategy, however, leads to unacceptable delays within the production process. To overcome these problems, our approach is based on a hybrid strategy where all computational intensive tasks are executed once prior to the actual assembly by an Offline Planner component. The results of this step serve as basis of decision-making for the Online Planner component, which adapts planning to the actual situation and unforeseen events. Due to this separation, our approach (see Figure 20) allows for detailed planning as well as fast computation during the assembly, therefore enabling appropriate assembly duration even in nondeterministic environments. The Offline Planner contains a CAD Parser which derives the geometric properties. The currently supported format is STEP [9]. This data is then processed by the graph generator. The details of that process are explained in 0. The Online Planner consists of the components Graph Analyzer, Parallelizer and Cognitive Control. These are detailed in 0.

77


Offline Planner CAD

CAD Parser geometric data

geometric data Graph Generator state graph Online Planner Graph Analyzer current system state

assembly seq. Parallelizer

Robot Cell

assembly sets Cognitive Control

command

Figure 20 System overview of the hybrid approach 3.2

Offline: Graph Generation As only input, the Offline Planner receives a CAD description of the desired final product. From this input it derives the relations between the single parts of the product via geometrical analysis as described in [10]. The results are stored in a connection graph. Assembly sequences are now derived using an assembly-bydisassembly strategy: Based on the connection graph, all possible separations of the product into two parts are computed. The feasibility of those separations is then verified using collision detection techniques. Unfeasible separations are discarded. The remaining separations can then be evaluated regarding certain criteria as stability, accordance to assembly strategies of human operators or similar. The result of this evaluation is stored as a score for each separation. This separation is recursively continued until only single parts remain. All found separations are added to an and-/or-graph [11] (see Figure 21). Here, hyper edges connect a node containing the original assembly with the two nodes containing the respective results of the separation. This graph then holds all possible complete separations of the final product. Read in the opposite direction these separations represent all feasible assemblies. 1 1a

2

1 2 3

1 2 3 4

1c

1b

2b

3

3a

2a

2 3 4 3b

4

5

1 2

4a

7

1

6

2 3

5a

8

6a

9

2

3 4

3

10

4

Figure 21 And-/or- graph representation of a disassembly of a four block tower The and-/or-graph is then converted into a state graph as displayed in Figure 22. Here nodes represent subassemblies of the assembly. Edges connecting two such nodes represent the corresponding assembly action which transforms one state into the other. Each action has associated costs, which depend on the type of action, duration, etc. Also, each edge optionally stores information about single additional parts that are needed to transform the outgoing state into the incoming state.

78


Figure 22 State graph representation of the assembly of a four blocks tower The conversion from and-/or-graph is necessary, since the and-/or-graph is a hyper graph. The following process steps however can only work on normal graphs. The conversion process is relatively straightforward: To transform an hyper edge connecting a node containing (partial) assembly to the two nodes containing its separations into a normal edge, one must distinguish two cases as shown in Figure 23: In the first case one of the resulting separations consists only of a single part. The hyper edge can then be replaced by a normal edge leading from the original assembly to the other child node separation. The node containing only one part is deleted, and the related part is added to the edge as additionally needed part for the transformation. In the other case, both child nodes hold partial assemblies. These nodes can be merged into a node that holds both separations together. The resulting node then represents a state where two subassemblies have been assembled in parallel. A)

B)

1

1 2 3 4

1 2 3 4

1a

2

1 2 3

1

req: 4

10

1a

4

1 2 3

4

1 2 3 4

1 2 3 4

1 2

req:

6

3 4

1 2

3 4

Figure 23 Conversion of an and-/or-graph into a state graph. A) Conversion if one child node contains only one single part B) all other cases 3.3

Online: Graph Analysis The state graph builds the basis of decision-making for the Online Planner. During the assembly the Online Planner executes the following process iteratively until the desired product has been assembled: The Graph Analyzer perceives the current situation of the assembly and identifies the corresponding node of the state graph. Now the state graphs edge costs are updated due to the realizability of the respective action which depends on the availability of parts to be mounted. Unrealizable actions receive penalty costs which vary depending on how close in the future they would have to be executed. This cost assignment makes the planning algorithm avoid currently unrealizable assemblies. However, due to the weaker penalties for more

79


distanced edges, the algorithm prefers assembly sequences that rely on unavailable parts in the distant future to assemblies that immediately need those parts. Preferring the latter assembly results in reduced waiting periods during the assembly since missing parts have more time to be delivered until they are ultimatively needed. Using the A*-algorithm the Online Planner now derives the cheapest path connecting the node matching the actual state with a goal node, which presents one variant of the finished product. This path represents the at that time optimal assembly plan for the desired product. Using the same technique Zaeh et al [12] guide workers during assembly. In a further step the derived plan is analysed by the Parallelizer component regarding actions which could be executed in parallel. Where the the original plan is represented by a sequence of assembly actions, the parallelized plan is a sequence of sets of parallelizable assembly actions. Identifying parallelizable actions allows accelerating the assembly in case that more than one operator (robots or human workers) takes part in the assembly. Also the actual decision-making component, the Cognitive Controller, can choose freely between parallelizable actions, allowing it to consider further criteria not covered by the Graph Analyzer. The Cognitive Controller is based on Soar, a framework modelling the cognitive processes of man. It receives the derived plan, which represented as a sequence of sets of parallelizable actions. It can now choose to execute one action, wait for additional material or ask a human operator for support. 4.

INCORPORATING LEARNING INTO THE PLANNING PROCESS The approach described above bases its decision entirely on theoretical data or data based on the input of an experienced operator. This knowledge is stored in the edge costs which are assigned to the state graph during the graph generation. These costs cover average expected values regarding the duration of the related action, the energy consumption, complexity/chances of failure or if the human operator must assist for this action. However, these are only theoretical values. Miscalculations or wrong estimates can cause a considerable harm to the plan quality. Also, these costs are fixed. Changes of machine properties, which change the duration and energy consumption of an action are not considered. To improve this situation, the edge costs have to be regularly updated to better mirror the actual situation. This is realized by a feedback loop: In case that an action is executed, the result of this action is reported back to the system and the costs of the action are updated respectively regarding execution duration or other criteria. However, to be able to deal with momentarily fluctuations and to avoid oscillation between plans, the costs are updated with a certain inertia, so that edge costs vary slowly over time. In the on-going assembly and also for future assemblies the new costs are considered during planning, so that the system can learn to avoid certain actions which are error prone or take too much time.

5.

SUMMARY In this paper we presented means for improving autonomous assembly planning by incorporating learning into our hybrid approach of an offline and an online planner. We described the workflow of the offline planner, which analyses CAD data describing the desired product. The outcome of the offline planner is a state graph which holds alle possible (and feasible) assembly sequences. This graph is generated by following an assembly by disassembly strategy: Recursively all possible separations of the final product are computed until only single parts remain. This results in an and-/or-graph which is then transformed into the desired state graph. During the actual assembly, this state graph is updated to mirror the current situation of the assembly, specially the availability of parts. Using the A* algorithm, the at that time optimal assembly sequence is derived and handed over to the cognitive control unit, which then decides which assembly step gets to be executed. This step is then executed and the outcome of that step is reported back to the planning systems. The effected results are used to update the edge weights of the state graph. This feedback loop continuously adapts the state graph to the real conditions of the robot cell and therefore improves the precision of the CCU‘s planning behaviour. The complete workflow is depicted in Figure 24.

80


Figure 24 Complete workflow of the CCU

6.

OUTLOOK Future work must optimize the described approach regarding strategies for cost assignment, especially for the behaviour over time. Different assignment functions have to be evaluated regarding the resulting plan

81


quality. The same holds for learning: How fast can the system adapt to new situations without losing process stability? Other research must focus on expanding the described planning approach to a complete production line consisting of cooperating and collaborating robots which assemble several products simultaneously. 7.

ACKNOWLEDGEMENTS The authors would like to thank the German Research Foundation DFG for supporting the research on human-robot cooperation within the Cluster of Excellence "Integrative Production Technology for HighWage Countries".

8.

REFERENCES [1] Brecher, C. et al., Excellence in Production, Apprimus Verlag, Aachen, Germany, 2007. [2] Kempf T, Herfs W, Brecher C., Cognitive Control Technology for a Self-Optimizing Robot Based Assembly Cell., In: Proceedings of the ASME 2008 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, America Society of Mechanical Engineers, U.S., 2008. [3] Mayer M, Odenthal B, Faber M, Kabuß W, Kausch B, Schlick C., Simulation of Human Cognition in Self-Optimizing Assembly Systems, In: Proceedings of 17th World Congress on Ergonomics IEA 2009. Beijing, China, 2009 [4] Hoffmann, Jörg, FF: The Fast-Forward Planning System. In: The AI Magazine,2001. [5] Hoffmann, J., Brafman, R., Contingent planning via heuristic forward search with implicit belief states. In: In Proceedings of ICAPS‘05. 71–80, 2005. [6] Castellini, C., Giunchiglia, E., Tacchella, A., Tacchella, O., Improvements to sat-based conformant planning. In: In Proc. of 6th European Conference on Planning, 2001. [7] Kaufman, S.G., Wilson, R.H., Jones, R.E., Calton, T.L., Ames, A.L., LDRD final report: Automated planning and programming of assembly of fully 3d mechanisms. Technical Report SAND96-0433, Sandia National Laboratories , 1996. [8] Thomas, U., Automatisierte Programmierung von Robotern für Montageaufgaben.Volume 13 of Fortschritte in der Robotik. Shaker Verlag, Aachen, 2009. [9] Anderl, R, Tripper, D., STEP Standard for the Exchange of Product Model Data, B. G. Teubner, Stuttgart/Leipzig, 2000. [10] Röhrdanz, F.; Mosemann, H.; Wahl, F. M., HighLAP: a high level system for generating, representing, and evaluating assembly sequences. In: 1996 IEEE International Joint Symposia on Intelligence and Systems, Seiten 134–141., 1996 [11] Homem de Mello, L. S.; Sanderson, A. C. , And/Or Graph Representation of Assembly Plans. Proceedings of 1986 AAAI National Conference on Artificial Intelligence, p. 1113–1119, 1986. [12] Zaeh, M.; Wiesbeck, M., A Model for Adaptively Generating Assembly Instructions Using State-based Graphs. In: Manufacturing Systems and Technologies for the New Frontier, Springer, London, 2008

82


OPTIMISING SPLIT-RANGE CONTROL ON A BLENDING PROCESS BY MEANS OF RULE-BASED TECHNIQUE Puramanathan Naidoo Mangosuthu University of Technology, Faculty of Engineering: Electrical Durban, KwaZulu Natal, South Africa e-mail: [email protected] ABSTRACT Expert systems differ from traditional computer programs in three ways: they perform computational or problem-solving tasks based on incomplete and subjective knowledge, they explain their reasoning to the user by displaying the rules applied toward solving given problems, and they have mechanisms built in for the acquiring of new knowledge. Expert systems based on ―semantic‖ networks can eliminate whole classes of objects without testing each object in turn. In the ―frames‖ paradigm, facts and logical rules are built into the same framework using an "object oriented" approach. On the highest level rule-based technique must receive input from its environment, determine an action or response, and deliver an output to its environment. The interpretation must be embedded in the control algorithm and represented in some form that can be manipulated by the computer system. Split-range control is applied to manufacturing systems for specific process requirements, in multi-final correcting element configurations, to optimise conventional control. Siemens Fuzzy Control++ design tool will be configured to simulate and test a suitable rule-based, split-range control algorithm. Keywords: Rule-based Technique, Design and Analysis, Split-range Control. 1.

INTRODUCTION In order to achieve the goal of optimal control, techniques of knowledge representation are invoked together with advanced methods like split-range control. The artificial intelligence interpretation of this, together with knowledge obtained previously, is manipulated within a typical manufacturing system by means of a rule-based technique. The system thus arrives at an internal representation of the response of the control action. This requires techniques of natural-language generation. The system will show that facts and rules (declarative knowledge) can be represented separately from decision-making algorithms (procedural knowledge) [1,3,6]. By adopting a particular procedural element, called an inference engine, development of the rule-based technique is reduced to obtaining and codifying sufficient rules and facts from the problem domain. This codification process, called knowledge engineering, will be utilised. The objective of the control strategy in this project is to maintain a desired, constant pressure of the supply process variable and blend chest level, by means of rule-based technique combined with split-range control [7,8,10]. Split-range control is implemented to optimise control by regulating two or more final correcting elements from a single controller. This method is subjected to restricted optimal control under certain conditions, due to the mathematically based structure of the control algorithm. The aim of the research is to develop and test the response of an appropriate decision-making control algorithm, to maintain a desired level in the blend chest and pressure of the process supply under varying conditions [5,13,14,18].

2.

THE MANUFACTURING PROCESS The main problem in maintaining a consistent blend chest level and supply pressure is nullifying the effect of outside influences. Both the blend chest level and supply pressure depend primarily on the amount of fluid entering the blend chest and the demand required by the manufacturing plant. The research focuses on the regulation of the correct amount of supply through both control valves, PV1a and PV1b. Both valves are air-to-open valves with PV1a operating from 0% to 50% of the control signal and PV1b operating from 50% to 100% of the control signal. When the supply and demand conditions vary the desired blend chest level and supply pressure can be greatly affected, and it becomes necessary to quickly readjust them to meet process conditions to maintain quality and logistical requirements.

83


SV3

Recirculation

Supply

FUZZY PT1

CONTROL

BLEND CHEST Input 1 Input 2

Delivery LT1

Pump Output 1 Output 2 PV1b

PV1a

BLENDED PRODUCT TO MANUFACTURING PLANT Figure 1: Process plant layout configured in combined split-range and fuzzy control The level in the Blend Chest is measured by LT1 (level-transmitter 1) and the pressure on the delivery side of the pump is measured by PT1 (pressure-transmitter). The supply to the manufacturing plant is regulated at two points via PV1a (pressure-valve a) and PV1b (pressure-valve b). Split-range control benefits this manufacturing process by integrating the two final correcting elements, to maintain a consistent supply to the manufacturing plant as per process requirements [14]. However, there are limitations in any mathematically based control system due to the algorithm relying on mathematical calculations on deviations from the desired value. A 2-input, 2-output fuzzy controller is tested as indicated in Figure 1. 3.

CONFIGURING THE CONTROL ALGORITHM Two inputs, process pressure and blend chest level, and two outputs applied to pressure valves were defined. After naming the inputs and outputs the membership functions had to be defined, for each input and output. The trapezoid form was used for the inputs, in order to increase the number of corner points, for clear distinction of one function from the other. The outputs were inserted as singletons. The rules were then edited in the inference engine, in either the rule table or rule matrix form [2,4,9,11]. In order for the desired process pressure to be maintained, it was dependent on certain plant and control variables. These variables had to be analysed at different values, within a specified band, in order to maintain the process pressure at the desired value. The membership functions (procedural knowledge) for both, the inputs and outputs were derived from the following plant variables, for the specified band:  



Process pressure Blend chest level Position of pressure control valves a and b, configured in split-range

The rules (declarative knowledge) for the rule-based system were derived from the following control variables, for the specified band: 

Data communication signals from the process pressure and blend chest level transmitters

84




Data communication signals to pressure control valves a and b

Figure 2 represents the knowledge engineering process in a generic form [9,16,19].

A

B

X Y Z

RULE 1:

C

PV

IF PT100=…

PT 100

100a

THEN, PV100a =… RULE 25: A

B

C

X Y Z

IF LT100=…

LT 100

PV

THEN,

100b

PV100b =… Figure 2: Generic knowledge engineering process of the control algorithm Figure 3 represents the edited input membership functions in trapezoid form, for only process pressure (PT100) as configured in Siemens FuzzyControl++ design tool. Table 1 represents the actual corner points of each membership functions. This facilitates fuzzification of a crisp value by scaling and mapping the input‘s domain, a linguistic variable, into an internal computer code. The second input, blend chest level (LT100) was configured in the same way as discussed above [9,12]. Figure 4 represents the edited output membership functions in singleton form, for only process pressure valve a (PV100a), as configured in Siemens FuzzyControl++ design tool. Table 2 represents the actual values of each membership function. This facilitates de-fuzzification of the internal computer code to a crisp value by scaling and mapping the output‘s domain. The second output, process pressure valve b (PV100b), was configured in the same way as discussed above. It will be noted that valve a, as indicated, operates between 0.00% to 50.00% and valve b operates between 50% and 100%, due to the split-range principle. In the absence of a firing rule the output is maintained at the last value [9,12]. Table 1: Edited Pressure (Actual values) MEMB. FUNCT.

PT 1

PT 2

PT 3

PT 4

PT_vlo

0.0

0.0

15.0

20.0

PT_lo

15.0

20.0

35.0

40.0

PT_med

35.0

40.0

60.0

65.0

PT_hi

60.0

65.0

80.0

85.0

PT_vhi

80.0

85.0

100.0

100.0

85


Figure 3: Edited Pressure (Graphical)

Figure 4: Edited Pressure Valve a Table 2: Edited Pressure Valve a

MEMB. FUNCT.

VALUE

a_clsd

0.00

a_sclsd

15.00

a_half

25.00

a_sopn

35.00

a_opn

50.00

86


Figure 5 represents the rules that govern pressure valve a and figure 6 represents the rules that govern pressure valve b in the knowledge engineering process of the inference stage. The facts and rules (declarative knowledge) are represented separately from decision-making algorithms (procedural knowledge). From figure 5 the rule numbers are read from left to right in ascending order, e.g. rule 5 states that IF PT100=PT_vhi and LT100=LT_lo, THEN PV100a=a_sclsd. Similarly figure 6 indicates for the same rule that IF PT100=PT_vhi and LT100=LT_lo, THEN PV100b=b_sclsd [9,15].

Figure 5: 25 Rules for Valve a 4.

Figure 6: 25 Rules for Valve b

THREE-DIMENSIONAL ANALYSIS

The numerical value 5 as indicated on Figure 7(a) and Figure 7(b) represents rule number 5 as discussed in the previous section. The common zero point for the x, y and z-axis is also indicated. It can be seen that the process pressure is within the fifth of the five input membership functions, PT_vhi (80.0-85.0-100.0-100.0 are the corner points as discussed in section 3) and the blend chest level is within the second of the five input membership functions, LT_lo (with corner points, 15.0-20.0-35.0-40.0). For rule 5 the output membership functions for both the pressure valves are represented on figures 7a and 7b as a_clsd and b_clsd for valves a and b, respectively. However, the value of a_clsd is 15.0% and b_clsd is 65.0%. As discussed in section 2 both the valves operate over different range values due to combined split-range and fuzzy control. The assigned rules indicate that the process pressure is directly proportional to valve a between 0% to 50% and is directly proportional to valve b between 50% to 100% respectively. This concept can be seen on figure 1 as the pressure transmitter is installed between the delivery side of the pump and both the valves. When the blend chest level increases the head pressure causes the process supply pressure in increase thereby causing the valves to open with increasing data communication signals [9, 20, 21].

5 5

Zero Point

Zero Point

(a)

(b)

Figure 7: Spatial representation of Input 1(PT100) on the x-axis and Input 2 (LT100) on

87


the y-axis versus; (a) : Output 1(PV100a) on the z-axis (b) : Output 2(PV100b) on the z-axis

5.

SIMULATING AND TESTING The designed control algorithm was tested by two-dimensional analysis of simulating the actual input signals by allowing the rule firing sequence to generate the corresponding output signals and concurrently verifying the active rules. In the two-dimensional analysis the process pressure was set at 5 cycles per screen, 40% aspect ratio, 0% phase angle, and the blend chest level was set at 3 cycles per screen, 30% aspect ratio, 0% phase angle. As recorded in figure 8a the process pressure is 9.99%, the blend chest level is 66.25%, pressure valve a is 35% and pressure valve b is 85%. Like in figure 7 both the valves vary by a difference of 50% from each other due to the combined split-range decision-making control algorithm. The recorded readings are captured at the point of the ruler, as indicated in figure 8a. At this point of analysis figure 8b captures the rule firing on another screen, to verify the active rule. Figure 8b indicates that rule 4 is 35% active, on a activity scale of 0.00 to 1.00, for output 1, pressure valve PV100a [9,14,17,21].

PV100b

PV100a

Ruler

PT100

LT100

Figure 8: (a) Simulation of rule base control

6.

(b) Verification of rule firing

CONCLUSION From the response of the rule-based split-range control algorithm, at a point of analysis in figure 8, the stepped response of the control signals indicates that the controller response is based on an intelligent decision-making process and not mathematical reasoning. A typical response of a conventional control algorithm causes the manipulated variable to increase and decrease over a period of time allowing the overshoots and undershoots of the process variable from the desired value to decrease 0.25 of the previous amplitude, to allow the process variable to follow the trend of this oscillation. This is referred to as 0.25 wave damping. This is unavoidable with this method of control. During this settling time there is a negative impact on production objectives, product quality, etc. It can clearly be seen that the time delay that is required for the process variable to reach the desired value in the conventional controller is not present in a rule-based controller. The response of the control algorithm minimises instability, optimising control.

7.

ACKNOWLEDGEMENTS The author expresses his gratitude to Mangosuthu University of Technology for the funding and support provided in the research, the Process Instrumentation students who have contributed in the installation and commissioning of various phases of the project, JVR Consulting engineers for technical support, John Farr

88


of Tiger Brands for sponsoring all fabrication on the project and affording the participating Process Instrumentation students the opportunity of gaining industrial exposure during their academic programme. 8.

RECOMMENDATIONS From the benefits observed in the research under simulated conditions, the designed control algorithm will be tested on-line in real-time, on the next phase of commissioning of the plant to verify the application of the research under dynamic conditions.

9.

REFERENCES [1] Muñoz C., Vargas F., Bustos J., Curilem M., Salvo S., Miranda H., Fuzzy Logic in Genetic Regulatory Network Models, International Journal of Computers, Communications & Control IV (4), 2009 [2] Al-anni M.K, Sundarajan V., Detecting a denial of service using artificial intelligent tools, genetic algorithm, Indian Journal of Science and Technology 2 (2), 2009 [3] Ismail R., Jusoff K., Ahmad T, Ahmad S., Ahmad R.S, Fuzzy State Space Model of Multivariable Control Systems, Computer and Information Science, 2009 [4] Sandhu P.S., Salaria D.S., Singh H., A Comparative Analysis of Fuzzy, Neuro-Fuzzy and Fuzzy-GA Based Approaches for Software Reusability Evaluation, World Academy of Science, Engineering and Technology (39), 2008 [5] Satean T., Santi W., Level Control in Horizontal Tank by Fuzzy-PID Cascade Controller, World Academy of Science, Engineering and Technology, (25), 2007 [6] Yohn E. G. Z., Fuzzy Logic in Process Control: A New Fuzzy Logic Controller and an Improved Fuzzy-Internal Model Controller, University of South Florida, 2006 [7] Cox E., The Fuzzy Systems Handbook, AP Professional, 1999 [8] Langari R., Yen J., Fuzzy Logic, Prentice Hall, 1999 [9] FuzzyControl++ Manual, Siemens, C79000-G8276-C144-01, 1998 [10] Wang L., A Course in Fuzzy Systems and Control, Prentice Hall, 1997 [11] Reznik L., Fuzzy Controllers, Oxford Newnes, 1997 [12] Kosko B., Fuzzy Engineering, Prentice Hall, 1997 [13] Jamshidi M., Parsaei H.R., Design and Implementation of Intelligent Manufacturing Systems, Prentice Hall, 1995 [14] Liptak B., Instrument Engineer‘s Handbook: Process Control, Butterworth-Heinemann, 1995 [15] Kosko B., Fuzzy Thinking, London : Flamingo, 1994 [16] Kruse R., Gebhardt J., Klawonn F., Foundations of Fuzzy Systems, John Wiley, 1994 [17] Wang L., Adaptive Fuzzy Systems And Control, Prentice Hall, 1994 [18] Jamshidi M., Vadiee N., Ross T.J., Fuzzy Logic and Control, Prentice Hall, 1993 [19] Kosko B., Neural Networks and Fuzzy Systems, Prentice Hall, 1992 [20] Winstanley G., Artificial Intelligence in Engineering, John Wiley, 1991 [21] Naidoo P., Automation and control of an industrial process using fuzzy logic, Elektron, Vol. 19, No 4, pp 46 – 47

89


CONCEPTUAL DESIGN OF A RECONFIGURABLE MANUFACTURING SYSTEM FOR THE TOOLING INDUSTRY IA Gorlach1 and BH Roberts2 1

Nelson Mandela Metropolitan University, Faculty of Engineering, the Built Environment and Information Technology, Port Elizabeth, South Africa e-mail1: [email protected] 2 Nelson Mandela Metropolitan University, Faculty of Engineering, the Built Environment and Information Technology, Port Elizabeth, South Africa e-mail2: [email protected] ABSTRACT Mould, tool and die making, or tooling, is an important sector of the modern manufacturing industry, supporting many other sectors of the economy, such as: production of consumer goods, transport, aerospace, biomedical and service. Tooling manufacturing can be characterized by high investment costs due to the need for a large variety of machine tools, high precision equipment and high skill levels required to produce complex tools. A tooling manufacturer is required either to have a large variety of machine tools to facilitate all the required manufacturing processes, which could be economically unviable for Small, Micro, and Medium Enterprises (SMMEs), or to outsource some production processes, which can create logistical and quality issues. In addition, tooling manufacturing is job-shop type production, which makes it difficult to automate without considerable investments. Flexible Manufacturing Systems (FMSs), for example, can be used for tooling manufacturing; however, due to high costs they are more suitable for small and medium production volumes than for job-shop production. One of the solutions in reducing the cost of equipment and facilitating required manufacturing processes is to employ a Reconfigurable Manufacturing System (RMS), which can be designed around a part-family. RMSs allow rapid changeover between products by adjustment and rearrangement of the sub-systems or processing units, thus providing required manufacturing flexibility at lower costs. This paper presents the conceptual design of a Reconfigurable Manufacturing System for rubber injection mould making. The process of development of the RMS is based on a part-family formation philosophy, which leads to selection of common manufacturing processes and the design of reconfigurable equipment to facilitate these processes. The developed Reconfigurable Manufacturing System can be used for manufacturing of rubber and plastic injection moulds and extrusion dies. Keywords: Advanced Manufacturing Systems, Reconfigurable Manufacturing Systems 1.

INTRODUCTION The tooling industry is faced with high frequency changes in products, which makes it difficult to be competitive without updating or developing technological processes. In addition, characteristics specific to the tooling industry must be considered when developing new manufacturing technologies, namely: small lot sizes, large variant diversity, widely differing product groups, high levels of complexity, dimensional precision/accuracy requirements, stringent quality requirements, and the need for minimal cost incursion and manufacturing times [1-6]. These challenges have led to the development first of FMSs and later RMSs intended for utilisation within the tooling industry. RMSs exhibit flexibility and have three capabilities, namely rapid changeover between products, rapid introduction of new products and unattended operation [7]. The capabilities are the result of efficient adjustment and rearrangement of the RMS components; with the effectiveness of the RMS stemming from the identification of part families [8]. The development of reconfigurable machine tools that can be utilised in RMSs has been reported in [9].

This paper presents the conceptual design of a RMS for rubber injection mould making, as well as extrusion die making. The process of development of the RMS is based on the formation of part families, and the employment of group technology (GT), which leads to selection of common manufacturing processes and

90


equipment. Along with this, the design of specialised reconfigurable equipment is carried out to facilitate production of moulds and dies. The developed RMS can however also be used for machining of conventional parts in general manufacturing. 2. 2.1

PART FAMILIES FORMATION Mould Selection In this research, the case study is focused on manufacturing of an injection die for an automotive component. Figure 1(a) shows a rubber component produced by a local component supplier for the VW Polo vehicle, which is manufactured by means of an injection moulding process. The geometric complexity of this particular component demonstrates a level of sophistication of mould design and manufacturing, as well as the intricacy of the rubber injection process. The component geometry includes complex NURBS surfaces, a thin support, a bulky cap with four cavities for weight reduction, and a 2 mm diameter blind hole in the head. The main features of the rubber component are achieved in the following way. The part Cap is moulded in the cavity. The internal part Cavities are formed by the set of sections, which forms the Core as shown in Figure 1(b). The sections can be separated, which is necessary for releasing finished parts. The process of opening or partial disassembly of a mould in the horizontal direction can only be achieved manually as a press is not equipped with such a device. Since the operator needs to accomplish this function every time the cycle is complete, the mould design needs to be ergonomic to simplify the process and yet it should guarantee the required accuracy. The 2 mm blind diameter hole in the Cap is formed by a Pin. Pins are inserted in the Cavity (not shown) from the back and secured with grub screws. The complete mould consists of a number of parts and sub-assemblies, which facilitates the injection moulding process.

Cap

Blind hole

NURBS surfaces Cavities

Support

Step a)

Cavities to form Cap

Cores to form Cavities

b)

Figure 1: CAD model of the rubber component

91


Mould cavity plate

Upper mounting plate

Upper mounting plate

Core mounting plate Core mounting plate

Core sections Core locating plate

Lower mounting plate

Figure 2: Complete Mould CAD Model 2.2

Mould Parts Classification All the parts of the complete mould shown in Figure 2 were classified to form five Part Families. Two Part Families are described in detail. The Upper, Lower, Core mounting and Core locating plates form Part Family 1 having common geometric features such as: plate shape, through holes, tapped holes and straight slots. The geometric and positional accuracy of the dimensions and features of the parts of Part Family 1 is between ISO IT9 and IT14, except for the dowel holes, which are made according to ISO IT7. The faces of the parts need to be parallel within a 0.1 mm positional tolerance in order to provide a reference horizontal plane, therefore surface grinding is necessary. The surface finish quality of the main features of the parts is between Class 8 - 10 (Ra = 3.2 – 25 µm), except for the bottom and top faces and the dowel holes, which have Class 7 (Ra = 1.6 µm). Therefore, Part Family 1 has common geometric features and accuracy tolerances, which can be obtained by means of face/slot milling, surface grinding, drilling, reaming and tapping processes using conventional machine tools of manual or automatic (CNC) operation, general engineering fixtures, such as manual or mechanised vices, and standard cutting tools. Referring to Figure 2, the Core sections form Part Family 2, which are non-rotational parts and have common geometric features such as: a rectangular shape, convex and concave NURBS surfaces, through holes and rectangular steps. These parts can be made of tool steel or hardened high carbon steel as they are subjected to wear and thus need to have higher hardness. The geometric and positional accuracy of the dimensions and features of Part Family 2 is between ISO IT7 and IT10, which include NURBS surfaces and the dowel holes. The faces of the parts need to be parallel within a 0.1 mm positional tolerance in order to provide a reference horizontal plane, therefore surface grinding is necessary. The surface finish quality of the features of Part Family 2 is Class 6 - 10 (Ra = 0.8 – 25 µm), which can be achieved by finish (highspeed) milling and/or polishing. The bottom and top faces are ground. Therefore, Part Family 2 has common geometric features and accuracy tolerances, which can be obtained by means of face/slot milling, surface grinding, 3-5 axis milling, drilling and reaming processes using conventional machine tools for general processes and CNC machines for sculptured surfaces, general engineering fixtures such as manual or mechanised vices, and standard cutting tools. An Electric Discharge Machining (EDM) (or spark erosion) process is used to produce the Cavities of the mould plate. The sinking EDM electrode itself must be manufactured (machined), and falls under Part Family 2. The Cavity plate forms Part Family 3, which has the following features: a rectangular plate shape, concave NURBS surfaces, through and tapped holes. Dowels and guide pins form Part Family 4, which are rotational parts. Although the case study component does not require an extrusion die for manufacture, it is desired that they form part of the range of parts applicable to the system, as extrusion dies are common in

92


the mould making industry. Extrusion dies differ from the above Part Families, as they require Wire EDM to make intricately shaped slots, therefore they form Part Family 5. An overview of the manufacturing processes relating to the Part Families and the individual mould assembly parts is shown in Figure 3.

Cylindrical Grinding

Polishing

PART FAMILY 5

Turning

PART FAMILY 4

Hardening

3 – 5 Axis Milling

High Speed Milling

PART FAMILY 3

Boring/Tapping

Face/Slot Milling

Drilling/Reaming

Surface Grinding

Electric-Discharge Machining

PART FAMILY 2

PART FAMILY 1

Figure 3: Manufacturing processes relating to the selected Mould Part Families

2.3

Part Families Manufacturing Processes Selection The range of part families to be manufactured by the RMS is defined above. For each family (and related features), specific processes, and hence specific production equipment, is required, see Table 1 below.

Table 1. Processes and equipment for manufacture of part families for moulds and dies Family Type and Features Die plates and blocks (Family 1)     

Prismatic Simple geometries Orthogonal or angular surfaces Slots Holes

Processes

    

Equipment 

2-and 3-axis rough and semi-finish milling Drilling Rough and finish reaming Internal thread cutting and/or tapping Surface grinding

93



3-axis CNC milling machine (2-D for milling and 2½-D for thread cutting and/or tapping; but machining centre if automated tool change etc. required) Surface grinder (or possibly incorporate into multi-function CNC machine)


  

Threads High accuracy Good surface finish of original block (before other machining)



Internal cylindrical grinding

Convex and Concave cores and cavities (Families 2 and 3)





  

5-Axis rough milling and/or high speed milling Hardening EDM Contour grinding or polishing

 

Protrusions with contoured surfaces having complex geometry Hard and abrasion resistant materials Textured or smooth surface finishes Rotational guide and dowel pins etc. (Family 4)

   

Rotational (axissymmetrical) Simple geometries (grooves, steps, etc.) Possibly flat surfaces, nonaxial holes, threads, etc. Possibly high accuracy and good surface finish

Extrusion dies (Family 5)   

3. 3.1

Complex geometry orifices with potentially sharp corners (small radii) High accuracy and smooth surface finish Hard and abrasion resistant materials

      

   

Turning (contour turning unlikely) Boring Axial drilling Axial reaming Internal and external thread cutting Possibly milling and drilling, perpendicular to axis (inclined unlikely) External cylindrical grinding 2-and 3-axis rough and semi-finish milling Possibly hardening (for metal extruding) Wire–cut EDM Surface grinding

   





   

5-axis CNC milling machine, equipped with high speed spindle Heat treatment equipment. Die sinking EDM machine, preferably 3-axis CNC. Polishing machine (hand held); or polishing incorporated into multifunction CNC machine.

Horizontal CNC lathe with additional spindle for off-centre longitudinal holes (but mill-turn centre required if flat faces and/or automated tool change etc. required) Cylindrical grinder (or possibly incorporate into multi-function CNC machine)

3-Axis 2-D milling machine Heat treatment equipment Wire EDM Surface grinder (or possibly incorporate into multi-function CNC machine)

CONCEPTUAL DESIGN OF THE RMS Design Methodology A classical systems engineering approach was used for the conceptual design of the RMS [10]. The system design requirements were determined by operational and performance requirements of a system, which are identified through Technical Performance Measures (TPMs). The TPMs also include maintenance and support requirements. Based on the prioritised TPMs, a Functional Analysis of a system is performed, which can be structured in the form of a Quality Function Deployment (QFD) model. The advantage of the QFD model is that it allows the system requirements and the system functions to be matched during the concept design phase, which can be further facilitated on sub-system level in the detailed design phase. QFD models can be presented in a table or graphical format, and include sub-levels as required. As a result, Functional Analyses and QFD models provide a basis for establishing a System Architecture, which defines how the system parts interact and are interrelated on the system level. Following this step is a concepts generation phase in the design process, during which various design alternatives are evolved based on a selection of methods, processes and technologies, which will provide the required functions in order to achieve the operational/design requirements.

94


3.2

Analysis of the System Requirements and Associated Functions The purpose of the proposed reconfigurable manufacturing system is to provide manufacturing capabilities for moulds and dies based on the types of Part Families that were developed previously. Therefore, the system requirements, shown in Figure 4, are identified for these particular five Part Families.

SYSTEM REQUIREMENTS

Product Type

System Type

Organizational Structure

   

 Moulds for plastic and rubber injection moulding  Forming dies  Extrusion dies  Product size is small and medium  High accuracy  High geometric complexity

 Production  Manufacturing & Assembly  Reconfigurable  Semi-automated

Job-shop Highly automated Scalable Expandable

Data

Materials

    

Rotational Non-rotational Heat treated Tools Fixtures

 CNC Programmes  Robot programmes  PLC  SCADA  Networks

Figure 4: Diagram of the manufacturing system requirements The system functional analysis is based on the system requirements analysis and includes the functions that are required for advanced manufacturing systems (Figure 5). The diagram shows only the two main levels of functions, which are important for conceptual design. The other sub-levels would be defined in the detailed design phase. SYSTEM REQUIREMENTS

Manufacturing

SYSTEM

Product

Data

FUNCTIONS

Data

Level 1 Safety & Security

Control

Materials Handling

Processing

Support

Machines

Storage & Retrieval

Technologies

Machine setting

Types

System

Robots

Transport

Assembly

Robot setting

Scheduling

Access Control

Storage

Machine tending

Inspection

Tool presetting Fixture presetting

Transport

Networks

Maintenance

Level 2

Figure 5: Diagram of the Manufacturing system functions

95

Users


3.3

Hierarchical System Architecture In order to develop the hierarchical architecture of a production system, it is important to include the process preparation and planning stage, which can be described as follows. A new product, a mould, is classified according to the classification system and its parts are assigned to a certain Part Family. This is followed by a step where a Computer-aided Process Planning (CPP) system will generate production process plans for the parts of the mould, using either a variant process-planning or generative processplanning system. For the proposed system, a variant process-planning system is more suitable, since the system is designed for Part Families having a limited number of features and commonalities. Once the process plan has been generated by the computer programme, it needs to be edited by engineering staff since some parts might have features which are not explicitly defined in the product data management system, and, therefore, would require interactive planning. Then, manual and machine processes are developed further by a process engineer. Manual processes like polishing, assembly etc. require the generation of standard operating procedure sheets, which are used as instructions for operators. Automated machining operations, such as turning and milling, are programmed using CAD/CAM software programmes depending on the type of operation and the machine tool. This process can be done fully automatically or interactively. As a result, CNC programmes together with tool data are generated and stored in the database. CNC programmes will be transferred to the machine tools when required according to the schedule. The tool data will be sent to the tool setting facility for tool presetting and storage. Cutting tools will be installed in the machining centre‘s tool magazines according to the process plan sequence. Material handling systems, such as industrial robots and AGVs, are programmed using dedicated packages or directly using robot programming utilities. A design engineer is required to design or select grippers and tool holding devices for the robots. This can also be done with the aid of a software programme, which guides a selection process. Specific grippers need to be designed and manufactured or ordered accordingly. The complete process plan with all the required machining and tooling information is transferred to a scheduling software programme, which generates a processing order and a bill of materials, and allocates the machine tools, based on the production process requirements and availability of machines. Control of the manufacturing system is achieved with the aid of a Supervisory Control and Data Acquisition (SCADA) software programme, which controls the automated machinery in real-time through the infrastructure, which provides a combination of central and distributed control using ‗Master-Slave‘ controllers and networks. The system hierarchical architecture is shown in Figure 6. On the sub-system level, the main units are CNC machining centres, robots and AGVs.

96


Central Computer

Software: Supervision, Scheduling, SCADA view, Logistics, Quality

Controller

File transfer Software:

Server

Databases

LAN: Ethernet

Hardware:

Wireless network

Industrial PC

SCADA

LAN or Industrial Ethernet

PLC (Master)

Profibus File transfer

PLC (slave)

PLC (slave)

Industrial Robots/

Profinet/ Profisafe

Profinet/ Profisafe

AGVs

Wireless network

CNC Machining Centre

Machine Loading System

Device

Sensor

Hardware: CNC, PLC, I/O Software: Network protocols

Manufacturing Cell

Material Handling/Inspection Cell

Figure 6: Diagram of the manufacturing system hierarchical architecture 3.4

System Layout Architecture The system layout architecture, depicted in Figure 7, includes: a. b. c. d. e.

Multi-axis Milling/ Turning reconfigurable centre (Machining Centre 1) 5-axis Milling/ High-Speed-Milling/ EDM (sinking) reconfigurable centre (Machining Centre 2) Sinking & Wire EDM reconfigurable centre (Machining Centre 3) Intelligent mobile material handling system Support sections

A common platform, which is used for all three types of reconfigurable machining centres was proposed in [11] . Machining Centre 1 would be used for machining of all Part Families requiring basic milling, drilling and turning. Part Families 2 and 3 will be further machined on the 5-axis Milling/ High-Speed-Milling/ EDM (sinking) reconfigurable centre (Machining Centre 2) for machining of cores and cavities requiring 5axis milling and/or EDM (sinking). The Sinking & Wire EDM reconfigurable centre (Machining Centre 3) will be used for Part Families 2, 3 and 5, depending on the machining requirements and their features.

97


Manual Fixture Setting Station

Station

Material Handling Station


Fenced area

Inspection Station (CMM)

Manual Tool Setting Station Central Control

Intelligent, mobile material handling system

Inspection Station (vision)


Manual Mould Polishing Station

Automated Storage/Retrieval System

Grinding Section

Reconfigurable/ multi-axis/ Milling/Turning Centre

Reconfigurable/ 5-axis Milling/High Speed/EDM Centre

Reconfigurable/ EDM sinking/ wire EDM Centre

Figure 7: Diagram of the manufacturing system layout An example of a reconfigurable machining centre configuration with 3-axis milling/ HSM and EDM (sinking) capabilities is shown in Figure 8. The machining centre consists of a base platform of a gantry type and interchangeable units. Such a centre could form the basis of each of the three Reconfigurable Centres which would provide the majority of the required machining processes; modules could be added or interchanged according to specific needs. The tool depicted could for example be converted into a Mill/Turn Centre by the inclusion of a part holding device and an additional spindle, for turning and boring applications. A controllable cradle and a tilting spindle head could also be added giving the machine 5-axis capabilities. In these and other ways a variety of modular variations could be achieved, giving the centre reconfiguration abilities.

Figure 8: CAD model of a Reconfigurable Machining Centre 4.

CONCLUSION In this research, the conceptual design of a Reconfigurable Manufacturing System for production of moulds and dies is presented. As a case study, an injection die for an automotive component was selected as it has the most typical features. The component geometric complexity demonstrated a level of sophistication of mould design, and related manufacturing processes, which were ultimately reflected in the design of the manufacturing system.

98


The parts of the mould were classified to form Part Families having common features and requiring common manufacturing processes. This classification was used for the selection of manufacturing processes, equipment, and tools, which led to the development of the proposed reconfigurable manufacturing system. The system conceptual design was accomplished based on the requirements of a high level of automation and stringent quality. The heart of the system is the reconfigurable machining centres, based on a common platform, which provides the required flexibility and functionality to produce all the major components of a typical mold and die. 5.

RECOMMENDATIONS The research can be extended to the investigation of the manufacturing of press dies of different sizes, to demonstrate the flexibility and range of capabilities of the conceptual system design. The incorporation of new advanced manufacturing processes into the design of reconfigurable machining centres should improve productivity and quality, and reduce the cost of production. It should be noted that the proposed system would be most cost effective and warranted in facilities producing significant volumes of moulds and dies, but where the range of tooling produced is large. The manufacturing system design variations can be explored to demonstrate the system reconfigureability and agility. This could be achieved through modelling and simulation.

6.

ACKNOWLEDGMENTS The authors would like to acknowledge the financial support from the Technology Innovation Agency (TIA) of South Africa, and General Motors South Africa, through the GMSA Chair of Mechatronics.

7.

REFERENCES

[1]

Aachener Werkzeug-und Formenbau, Tool and die making for the Future. http://www.ipt.fraunhofer.de/en/Businessunits/Toolanddiemaking/index.jsp, 2008. [2] Fallböhmer, P., Rodriguez, C., Özel, T., Altan, T., High-speed machining of cast iron and alloy steels for die and mold manufacturing, Journal of Materials Processing Technology, 2004. 98: 104 – 115 [3] Kelkar, A., Nagi, R., Koc, B., Geometric algorithms for rapidly reconfigurable mold manufacturing of free-form objects. Computer-Aided Design, 2005. 37: 1-16 [4] Sandvik, C., Die and mould production news, 2003. [5] Vivancos, J., Luis, C., Costa, L., Ortiz, J., Optimal machining parameters selection in high speed milling of hardened steels for injection moulds, Journal of Materials Processing Technology, 2004. 155 – 156: 1505 – 1512 [6] Yamamura, H., Yoshikawa, M., Iwata, K., Tomimatsu, K., Tsumura, K., High-speed & high-accuracy precision die milling machine: MVR-FM developed for die & mold manufacturing as base of massproduction, Mitsubishi Heavy Industries, Ltd., Technical Review. 2005. 42(2): 1-4 [7] Xing, B., Bright, G., Tlale, N., Potgieter, J., Reconfigurable manufacturing system for agile mass customization manufacturing, 22nd International Conference on CAD/CAM, Robotics and Factories of the Future, 2006. 473 – 482 [8] Galan, R., Racero, J., Eguia, I., Garcia, J., A systematic approach for the product families formation in reconfigurable manufacturing systems, Robotics and Computer-Integrated Manufacturing, 2007. 23: 489 – 502 [9] Simpson, M., Gorlach, I., Development of a Reconfigurable Machine Tool, International Conference on Competitive Manufacturing, COMA‘10, Stellenbosch, South Africa, 2010. [10] Blamchard, B., Fabrycky, W., Systems Engineering and Analysis, 5 th Ed., Prentice Hall, New Jersey, 2011. [11] Estment, W., Gorlach, I., Wiens, G., Design of sub-systems of a reconfigurable machine tool, CAD/CAM Robotics and Factories of the Future Conference, Pretoria, South Africa, 2010.

99


IMPROVEMENT OF MILLING TOOL GEOMETRY OPTIMISATION USING A REFINED DIFFERENTIAL EVOLUTION ALGORITHM FOR CHATTER AVOIDANCE Ahmad Razlan Yusoff1, Nafrizuan Mat Yahya2 and Mohamed Reza Zalani Mohamed Suffian3 1

Manufacturing Process Focus Group, Faculty of Mechanical Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia e-mail: [email protected] 2 Faculty of Manufacturing Engineering, Universiti Malaysia Pahang, Lebuhraya Tun Razak, 26300 Kuantan, Pahang, Malaysia e-mail: [email protected] 3 Manufacturing Process Focus Group, Faculty of Mechanical Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia e-mail: [email protected]

ABSTRACT Machining with high productivity causes unstable self-excited vibration for a particular spindle speed and depth of cut. This unstable problem can be mitigated by modifying the helix and pitch tool geometry. Optimization method was applied by combining semi discretisation method (SDM) and Differential Evolution as optimization algorithms to search optimal tool geometry. Differential Evolution (DE) algorithm was used in present study to optimize the tool geometry for chatter avoidance. During optimization process, however, the poor performance population happens where the population does not change. The mixed population update and bounce back strategy were proposed and applied to overcome this problem. The results showed that the refined DE approach can significantly avoid chatter with better performance than the original algorithm. Keywords: 1.

Self excited vibration, variable helix and variable pitch and Differential Evolution

INTRODUCTION A greater productivity and an improved quality in aerospace, automotive, mould/die and general manufacturing industries need to be faced to ensure lower cost productivity for making a sustainable machine tool industry development. When machining with a high material removal rate or productivity, the dynamic deflection of tool and workpiece can occur. This causes a problem called chatter as a self-excited type of vibration. Chatter can occur when machining with too large chip width with respect to the dynamic stiffness of the system during metal cutting. This results a low quality of workpiece finish and accelerated tool wear and can even break machine spindle speed. This boundary of stability is a function of depth of cut and spindle speed can be plotted to differentiate the region either stable or unstable in machining. A passive approach is explored in this paper for chatter suppression to interrupt the chatter vibration by variable helix and variable pitch tool geometry. A uniform helical end milling tools is compared to nontraditional variable helix and variable pitch design. Variable pitch tools by Altintas et al. [1] were reconsidered by using an invariant time constant and a non-uniform multiple regeneration time delay to optimize pitch geometry. Meanwhile, Budak [2] optimized and applied a non-constant pitch angle cutter model with an analytical stability model with linear pitch variation. Recently, Olgac and Sipahi [3] optimized the material removal rate and irregular pitch cutter simultaneously using a complex mathematical objective function. Sims et al. [4] modeled variable helix and variable pitch milling tools. However, Sims‘s model only predicted the chatter stability of variable helix and variable pitch cutter, and did not optimize the tool design for minimizing chatter. Yusoff and Sims [5] optimized variable helix and variable pitch milling tool for chatter suppression. However, Differential Evolution (DE) algorithms trapped in premature convergence can be referred as poor performance in minimizing chatter. In optimization algorithm, the population improvement includes several processes. The processes are mutation, crossover, objective function assessment and selection. This takes several generations before the global optimal solution is achieved. During the population process, it is desirable to produce a robust feature and a high convergence rate to create a population with high probability. In modified DE, instead of one array on population update, Babu and Angira [6] applied two population updates during the mutation

100


and crossover process and the original population to ensure each population had equal opportunity. To relocate the violated bound vector to the interior bound, a penalty function was used for avoiding local optimal. Meanwhile Nearchou and Omirou [7] and Zhang and Xu [8] used random keys encoding to handle discrete variables to produce high performance of the modified DE. By adjusting of minimum space distance, a population‘s being located in the same area was prevented, as proposed by Hendershot [9]. To accelerate the mutation process and exploration region, Kaelo and Ali [10] recommended a random uniform mutation factor and localisation around best vectors, respectively. A refined DE algorithm is used to improve poor performance in [5] and the results will be compared to show the effectiveness of refined DE in optimizing variable helix and variable pitch for chatter avoidance. A refined DE consists of mixed population update and bounce back strategy are applied to modify and improve the current DE algorithm. This paper is presented as follows. A theory of DE is introduce to the current problem is briefly described. Chatter minimization approach and application into refined DE are then presented. A refined DE optimization results is then presented and compare with previous optimization or original DE algorithms. 2.

THEORETICAL BACKGROUND Evaluation Algorithms (EA) such as Genetic Algorithm (GA), Evolutionary Programming (EP) and Evolution Strategy (ES) have been researched for several decades. Differential Evolution (DE) was introduced by Price et al. [11], and can be considered to be an improved Genetic Algorithm (GA) version with different strategies for faster optimization. This is similar to other EA in which mutation plays the key role, with real valued parameters that directly search for the global optimum. A basic idea in DE is that of adapting the search during the evolution process. Compared to other algorithms, DE has the advantages of simple structure, ease of use, speed and robustness. In machining applications, Saikumar and Shunmugan [12] applied DE to select the best cutting speed, feedrate and depth of cut to achieve optimum surface finish while Krishna [13] applied DE in a grinding process to search for suitable process grinding in minimizing surface grinding DE can solve objective functions that are non-differentiable, non-linear, noisy, flat, multi dimension, and with multiple local minima. Such functions are difficult to solve analytically, and the variable helix optimization problem fits within this scope. DE begins using initial samples at multiple random chosen initial points. With simple algorithms, DE can search for the optimal condition very quickly with minimal control parameters such as mutation, crossover, selection and population. The concept is evolved from GA‘s with layer population and special evolutionary strategy of self adaptive mutation. Instead of a binary in GA, differential evolution however deals with a real coded population with its own processes of mutation and crossover. Mutation process is created randomly from the selection of three individual vector differences. In the crossover process, any individual population member has equal opportunity to survive in the next generation based on its fitness value. The process of evolving mutation, recombination and selection through generations or new population is repeated until the optimum solution is achieved, as shown in Fig 1. By combining with the analytical method for chatter stability prediction, the process of DE optimization can be used to optimize tool/helix geometry. The sequence of operations is optimization setting process, Semi Discretisation Method (SDM) process, objective function evaluation process and DE optimization process. For the current study, the procedure adopted is as follows: 1. First, DE selects the input parameter value together with optimization parameters such as strategy, population size, number of generation NP crossover factor CR and scaling factor SF, as give in Table 1. 2. Optimization setting process parameters are read by the numerical algorithm process through the DE optimization process as new input parameters. 3. Then analytical chatter stability will predict the output value to evaluate by the objective function evaluation process. The parameters in SDM analytical chatter stability is referred to the Table 2 4. The output generated from this prediction is evaluated and compared with the next output. 5. These steps are repeated until the optimal input values of chatter stability are found. 6. This is an interactive process at the end of which the DE arrives at the optimum set of input parameters, i.e. variable helix i and variable pitch i, which generate optimum output.

101


In the present work, the DE source code by Markus Buehren [14] was initially used based on the DE algorithm of Price et al. [12]. To enhanced the optimization performance in [5], DE is refined by adding with bounce back and mixed population approaches. 3.

METHODOLOGY Chatter minimization For a single objective function that considers minimised chatter as target, where chatter relates to characteristic multiplier or eigen values CM that must be less than one to fully eliminate chatter. In the current case, the worst chatter indicated by eigen value is minimised by optimization algorithms. This occurs within a specific region of spindle speed and axial depth of cut. The final optimization problem can be specified as follows: Objective function:

Subject to constraints: Helical angle Pitch angle Helical height difference

Minimize

f (  i , i )  CM

(1)

25   i  55 i  1,2,3 n

  22.5  i    22.5 i  1,2,3 n h  5mm

During optimization, each point of specific spindle speed and axial depth of cut was evaluated on characteristic multiplier value to produce a matrix of characteristic multiplier. A single value of characteristic multiplier is calculated after two determinations of maximum characteristic multiplier matrix. This seems to be Matlab-specific i.e. max function operates on one dimension of a matrix. The constraint on pitch angle i (55) is to ensure good chip evacuation as suggested by previous work [1,2]. The DE needs to search for the suitable values of variable helix i and variable pitch i that produce minimum chatter across the chosen spindle speed and depth of cut range (along with the additional constraints). Helical angle height difference h is constraint greater than 5 to prevent intersection. Machine speed limits and workpiece thickness are two factors which need to be considered from a practical viewpoint. A constraint spindle speed should be considered for specific range of machine speed limits, while the workpiece thickness is limited by axial depth of cut. A single value of Equation 1 is calculated after two determinations of average value. At the same time, structural dynamics from the variable helix will be changed through characteristic multiplier evaluation either in stable or unstable condition at optimum performance. The iteration will finish when the optimal variable helical angles have achieved the maximum performance.

As stated before, variable helix and pitch were considered for the current study. The helical angle purpose is actually to break chip formation, change line of contact between tool and workpiece and reduce chatter [12]. By adjusting the helix randomly, the optimization process will be more generalized, besides avoiding premature population and preventing local optimal problem in the optimization processes.  i is the helical angle value range from low helix (25) to high helix (55) conditions. For variable pitch, the chip evacuation should be prevented when high chip removal is used [2, 3], particularly when higher value pitch angles are selected for the finishing process. Without chatter frequency and phase angle as constraint, as suggested by Altintas et al. [1] and Budak [2], the selection of the appropriate helical and pitch angles will be randomized. During the optimization process, however, the candidate values of tooth helix i and pitch i may result in milling cutters whose flutes intersect with each other. This is clearly inadmissible from a practical viewpoint. To prevent helical angles intersecting, the helical angle height different h is

102


introduced as a constraint in the DE optimization. h calculations for variable helix and variable helix with variable pitch at its end are as follows:

h  h 

d  d 2m  tan  i

d  d 2m  tan  i

1

d  2d m  (variable helix) tan  i 1 1



1

d  2d m sin2 1



tan  i 1

1

  (variable helix and pitch)

(2) (3)

Here, Equation (2) relates to optimization problems where the free end of the milling cutter has teeth uniformly spaced, but the variable helix of each flute means that the teeth are irregularly spaced along the rest of the cutter‘s axial length. Equation (3) relates to cutters with both variable helix and variable pitch at the free end. The h value is based on the parameters along with the number of cutter teeth m and cutter diameter d and pitch difference . The refined DE consists of mixed population and bounce back boundary are now presented. In every new generation, the next population will consist of 75 percent of the best current population and 25 percent from a randomised population. This randomised population reacts as noise to improve the next population. In every generation, a 25 percent population with additional noise will improve DE performance to overcome premature problem of the population during the optimization process. In constraint optimization, several methods [11] have been proposed to solve the problem, such as penalty function, random initialisation, bounce back method and rejection of the vectors. Previously, the rejection of the vector was applied; however, the point outside bounds may have a better solution but unfortunately not in the feasible region. A bounce back method function can be used to modify an out of bounds trial parameter with one located on the boundary. Besides escaping from the local optimal, especially at the boundary, this replaces out of bounds vectors to have a highly diverse population. 4.

RESULT AND DISCUSSIONS In this study, a single degree of freedom dynamic milling model with conditions of milling tool, modal and cutting parameters the same as in Table 2 is used, but the helical and pitch angles and other constraints need to be set before using DE and SQP as optimizer. Each case is solved for variable helix tools and variable helix and variable pitch tools to investigate the influence on chatter stability limit. Parametric study of DE parameters was made before analysis for chatter minimisation and chatter performance. Before DE is applied, DE parameters, such as crossover (CR), scaling factor (SF), number of population (NP) and number of generation (NG) for each ‗strategy 7‘, need to be determined. For this purpose, the problem to minimise chatter (objective function in Equation 1) of variable helix and pitch was selected. The parameter settings were evaluated based on the effects on DE performance. An initial study was made for different CR (0.2, 0.4, 0.5, 0.9), with other parameters kept constant, as shown in Table 2: 50 NG, 0.7 CR, 0.6 SF and 70 NP (10 multiplied by number of real parameter (7)). The second attempt was for SF of 0.5, 0.6, 0.7, 0.8 and 0.9, with other parameters kept constant. The DE was executed for 50 generations, 0.6 CR and 70 NP. A similar setting of the DE parameter was used to examine the effect of population size or NP. Various population sizes (4, 35, 40, 70 and 105) based on minimum NP, 5, 8, 10 and 15 times, respectively, with other parameters kept constant. The fourth study used a maximum number of generations of 10, 50, 75, 100 and 150, other parameters kept constant. The optimum values of DE parameters summarised in the present work are given in Table 3. Crossover rate CR is 0.9, scaling factor SF is 0.9 and the number of population NP is 10 times the real parameter (10*RP) and 70 generations are employed in DE optimization. A ‗strategy-7‘(DE/rand/1/bin) methodology [11] was implemented in view of its wide application in the literature. This methodology involves random perturbation of a population vector (‗/rand‘), perturbation of a difference vector for the mutation process (‗/1‘) and binomial crossover (‗/bin‘). The performance of the DE and Sequential Quadratic Programing (SQP) algorithms is summarized in Figure 2. Note that DE algorithms showed better performance than SQP. For both cases of variable helix

103


tools and variable helix and variable pitch tool, however, the problem which occurred was illustrated for the performance of variable helix and variable pitch tools [5]. During the optimization, the generation cannot be improved and performed after the second iteration although showed the best result. In order to indicate that the refined DE performs well and overcomes the previous problem, minimization of chatter is reconsidered with the same DE settings and machining parameters. In Figure 3, the original DE and refined DE were compared in their performance in optimising chatter of variable helix and variable pitch tool. It shows that the refined DE is significantly better than the original where generation improved during iterations although the population of refined DE showed improved maximum eigen value at the 63th generation converged at 0.8. Characteristic multiplier CM value is 6-fold greater than the previous optimization. This corresponds to a variable helix (41, 43, 40) and variable pitch (68, 283, 36) that showed stable or unchattered behaviour (Figure 4). By comparing with Figure 5, the refined DE is better than the original DE results, indicated by a larger gap in the stable border line in the CM diagram. Moreover, the magnitude of the absolute eigen value contour for the refined DE is between 0.6 and 0.85 and for the original DE is between 0.8 and 0.9. This not only represents that the CM value of the refined DE has a lower value or better than the original result, but also more damping behaviour from the original DE. This clearly indicates the bounce back and mixed population can improve the DE result. Next, the refined DE algorithm is reconsidered for previous problems. 5.

CONCLUSION The practical implementation of refined DE improves the optimization of variable helix and pitch in this paper. It has been indicated that for poor performance and local optimal problem in original DE optimization performance, the mixed population and bounce back boundary are introduced to prevent the optimization simulation. Using refined DE can improve the optimization performance and can be easily applied for current problem. For the future work, the optimized refined DE of the variable helix and pitch tool should have experimental validation.

6.

ACKNOWLEDGEMENT Authors extend their sincere thanks to the support of the Dr Neil D Sims at Department of Mechanical Engineering, the University of Sheffield.

7. [1] [2]

[3] [4] [5]

[6] [7] [8] [9]

REFERENCES Altintas, Y., Engin, S. and Budak, E. Analytical stability prediction and design of variable pitch cutters. Journal of Manufacturing Science and Engineering, Transactions of the ASME, (1999), 121(2), p. 173. Budak, E. An analytical design method for milling cutters with non-constant pitch to increase stability, Part 1: Theory and Part 2: Application. Journal of Manufacturing Science and Engineering, (2003), 125, p. 29. Olgac, N. and Sipahi, R. Dynamic and stability of variable pitch milling. Journal of Vibration and Control, (2007), 13(7), p. 1031. Sims, N.D., Mann, B. and Huyanan, S. Analytical prediction of chatter stability for variable pitch and variable helix milling tools. Journal of Sound and Vibration, (2008), 317(3-5), p. 664. Yusoff, A.R. and Sims, N.D. Optimization of variable helix end milling tools by minimising selfexcited vibration. Journal of Physics: Conference Series on 7th International Conference on Modern Practice in Stress and Vibration Analysis, (2009), 181, 012026. Babu, B.V. and Angira, R. Modified differential evolution (mde) for optimization of non-linear chemical processes. Computers & Chemical Engineering, (2006), 30(6-7), p. 989. Nearchou, A. Balancing large assembly lines by a new heuristic based on differential evolution method. The International Journal of Advanced Manufacturing Technology, (2007), 34(9), p. 1016. Zhang, J.Z. and Xu, J. A new differential evolution for discontinuous optimization problems. Third International Conference on Natural Computation (ICNC 2007), 0-7695-2875-9/07, (2007). Hendershot, Z.V. A differential evolution algorithm for automatically discovering multiple global optima in multidimensional, discontinuous spaces. Proceedings of the Fifteenth Midwest Artificial Intelligence and Cognitive Science Conference (MAICS 2004), Illinois, (2004).

104


[10] [11]

[12]

[13]

Price, K.V., Storn, R.M. and Lampinen, J.A. Differential evolution a practical approach to global optimization. (Springer, Berlin Heidelberg, 2005). Saikumar, S. and Shunmugan, M.S. Parameter selection based on surface finish in high speed finish in high speed end milling using differential evolution. Materials and Manufacturing Processes, (2008), 21(4), p. 341. Krishna, A. Selection of optimal conditions in the surface grinding process using a differential evolution approach. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, (2007), 221(7), p.1185. Markus, B. Differential evolution. 2008 [cited 2008 15th June ]; Available from:http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?objectId=18593&objectType =file. Table 1 DE parameter settings for machining optimization Parameter Initial value Optimum Strategy 7- DE/rand/1/bin Number of Generation (NG) 50 70 Number of Population (NP) 10*RP 10*RP Crossover factor (CR) 0.7 0.9 Scaling factor (SF) 0.6 0.9 Table 2 Cutting, modal and tool parameters for optimization Tool and cutting parameters Tool diameter d (mm) 19.05 Radial immersion RI (mm) 1.00 Tangential cutting stiffness Kt (MN/m2) 550 Tangential cutting stiffness Kn (MN/m2) 200 Modal property in x-direction mode Natural frequency fn (Hz) Modal effective mass mm (kg) Damping Ratio 

169.3 6.5363 0.0056

Table 3 Parametric study of DE to optimize variable helix for ‗strategy 7‘ Performance of DE DE parameters Crossover rate (CR) 0.2 0.4 0.5 0.7 0.9 Minimum CM

0.7533 0.5

Minimum CM

0.7543

0.7481

0.7489

0.7487

0.7481

Scaling factor (SF) 0.6 0.7 0.8 0.7490

0.7532

0.7487

0.9 0.7486

Number of Population (NP)

Minimum CM

4* RP

5* RP

8* RP

10*RP

15* RP

0.7711

0.7489

0.7491

0.7487

0.7487

Maximum Number of Generation (NG)

Minimum CM

10

50

70

100

150

0.7514

0.7487

0.7470

0.7470

0.7470

105


xi,g

x1,g

x2,g

x3,g

xNp-2,g xNp-1,g

f(xi,g)

f(x1,g)

f(x2,g)

f(x3,g)

f(xNp-2,g) f(xNp-1,g)

xr2,g

xr1,g

+

Current Population

xr3,g

+ +

Mutation (Strategy 1)

F

vi,g

v1,g

v2,g

v3,g

vNp-2,g vNp-1,g

f(vi,g)

f(v1,g)

f(v2,g)

f(v3,g)

f(vNp-2,g) f(vNp-1,g)

Mutant Population

Mutant Vector

crossover Trial Vector

Target Vector

ui,g

u1,g

u2,g

u3,g

uNp-2,g uNp-1,g

f(ui,g)

f(u1,g)

f(u2,g)

f(u3,g)

f(uNp-2,g) f(uNp-1,g)

Trial Population

selection

xi,g+1

x1,g+1 x2,g+1 x3,g+1

f(xi,g+1) f(x1,g+1) f(x2,g+1) f(x3,g+1)

xNp-2,g+1 xNp-1,g+1 f(xNp-2,g+1)f(xNp-1,g+1)

New Population

Figure 25: Process of generating one population to next

Figure 26: Performance of DE and SQP in chatter minimisation. (▬) DE variable helix and variable pitch, (-.-)DE variable helix, (---) SQP variable helix and (…) SQP variable helix and variable pitch

Figure 27: Performance of DE and improved DE on optimising three-flute variable helix and variable pitch. (▬) DE and (---) refined DE variable helix with variable pitch at end

106


Figure 28: Optimised stability prediction for chatter minimisation of three-flute variable helix (52, 52, 41) and variable pitch (107, 163, 90) using DE. ( ) stability contour

Figure 29: Optimised stability prediction for improved chatter minimisation of three-flute variable helix (41, 43, 40) and variable pitch (68, 225, 87) using DE. ( ) stability contour

107


A DISTINCTIVE MANUFACTURING CELL FOR MASS CUSTOMISATION Nazmier Hassan1, Glen Bright2 1

University of KwaZulu-Natal, Mechatronics and Robotics Research Group, Engineering Durban, South Africa e-mail1: [email protected] 2 University of KwaZulu-Natal, Mechatronics and Robotics Research Group, Engineering Durban, South Africa e-mail2: [email protected] ABSTRACT Customer driven demand for mass customised products has elevated the requirements for manufacturers. Manufacturers are constantly in search for feasible techniques towards meeting customer needs and to remain competitive in sustaining their place in the global economy. A technique for developing a Hybrid Reconfigurable Computer Integrated Manufacturing (HRCIM) cell for accommodating mass customisation was therefore researched. The HRCIM cell extracted and utilised specific FMS and RMS functionalities that were neither excessive nor inadequate in its capabilities. The key objective of the research was to analyse manufacturing cell behaviour for discrete cell operations and part flow. This particularly is focused on the production of mass customised parts that resulted from customer driven orders. Keywords: Computer Integrated Manufacturing (CIM); Reconfigurable Manufacturing Systems (RMSs); Flexible Manufacturing Systems (FMSs); Mass Customisation 1.

INTRODUCTION Present-day technology in manufacturing is still not sufficient to feasibly produce mass customised parts. Flexible Manufacturing Systems (FMSs) which are based on Computer Integrated Manufacturing (CIM) technology [1], possesses excessive functionalities that are not utilised in relation to its high cost. In contrast, Reconfigurable Manufacturing Systems (RMSs) are yet to be further enhanced and be acknowledged by manufacturing industries. RMSs are also limited to the manufacture of part families and are incapable of producing these mass customised parts [2]. To generate a distinct solution towards the manufacture of mass customised parts, research has directed a development of a HRCIM cell. The HRCIM cell architecture was developed with integrated hardware and software that encapsulated RMS and FMS characteristics. The hardware contained storage systems, material handling systems and processing stations that were primarily software controlled. Software related control was initiated from vision and algorithm based scheduling processes. These software processes generated seamless, synchronised operations and provided augmented reconfigurable and flexible manufacturing cell characteristics. Based on actual HRCIM cell data, a software based simulation was utilised. It generated in-depth manufacturing results. Results generated provided for improved operational efficiency, buffer status, processing data and cycle times. Actual laboratory and simulation results were used to investigate the HRCIM cell capabilities in relation to minimum product cycle times, reduced lead times and manufacturing cell changeover times. This paper focuses on the HRCIM cell behaviour, when exposed to discrete manufacturing events. Discrete events consisted of software and hardware transformations that responded to frequently changing material handling and part processing requirements.

2.

LITERATURE REVIEW Appropriate concise theory from existing market related approaches and manufacturing strategies were used to encapsulate modern manufacturing knowledge. This was used to direct research in developing a distinctive manufacturing cell for mass customisation. The following literature provides a study and details

108


the requirements for mass customisation. Conventional manufacturing strategy analysis that entails RMS, FMS and CIM is also detailed. 2.1

Mass Customisation Present-day customer demands for customised products are more diverse. Manufacturers are therefore compelled to produce customised products and sustain high production efficiency [3]. This approach is termed mass customisation and is defined as, ―the technologies and systems to deliver goods and services that meet individual customer needs with near mass production efficiency‖. It is also an approach where adequate customisation and variety is provided to meet customer needs [4]. This approach allows manufacturers to produce products after having orders at hand (make-to-order) [5]. Manufacturers are however inhibited to efficiently produce mass customised parts due to the lack of existing technology that can be reconfigured easily, rapidly and cost effectively [6]. Some of the advantages of inducing mass customisation include the reduction in material waste and inventory levels. It also facilitates customer satisfaction and maximises market share [5].

2.2

Reconfigurable Manufacturing Systems RMSs are designed to generate a cost effective response to alterations in manufacturing requirements. Its design facilitates rapid transformation in its structure as well as its hardware and software components. RMS involves system adaptability in response to manufacturing requirements that require reconfiguration [7]. It is designed to rapidly manufacture diverse product families. These product families that are processed by a RMS, is generated at the lowest cost and shortest time, avoiding any decline in quality [2].

2.3

Flexible Manufacturing Systems FMSs are based on CIM technology [1], and is defined as "a computer-controlled production system capable of processing a variety of part types". Due to its capability in producing a variety of quality products, FMSs are able to meet market demands. A typical FMS consists of automated material handling systems, Computer Numerically Controlled (CNC) machines and robots [8].

2.4 Comparing RMS and FMS Table 1 [9-10] tabulates comparisons between RMS and FMS. As tabulated, adjustable machine and system structure facilitates system scalability. This is in reaction to varying market demands and machine or system adaptability when introduced to new products [9]. Scalability refers to the ability to alter production capacity by reconfiguring an existing manufacturing system, and or individually altering the production capacity of reconfigurable stations [11]. Machine structure includes machine hardware and software, and system structure includes the addition of machines [9]. From an RMS perspective, its manufacturing system is designed with customised flexibility to produce part families. FMS in contrast possesses general flexibility as equipment such as CNCs are developed prior to being selected by a manufacturer and before process planning is incorporated for a part. The high cost of FMSs are due to this flexibility [9]. FMSs also possess the capability to produce a high variety of products [7]. Table 1: Comparison between FMS and RMS RMS

FMS

Machine structure

Adjustable

Fixed

System structure

Adjustable

Adjustable

System focus

Part Family

Machine

Flexibility

Customised

General

109


Scalability

Yes

Yes

Simultaneous operating tools

Yes

No

Intermediate

High

Cost

2.5

Computer Integrated Manufacturing CIM is a manufacturing and an essential management strategy. It is used to integrate manufacturing systems and facilities in an enterprise, by utilising computers and its peripherals [12]. CIM technologies include engineering functions such as computer-aided design and computer aided manufacturing. Other relevant technologies include Automatic Storage and Retrieval Systems (ASRSs), CNCs and FMSs [1, 13-15].

3.

HRCIM CELL METHODOLOGY Analysis generated from existing manufacturing strategies, that included FMS and RMS, provided a methodology in conceptualising a HRCIM cell. The HRCIM cell characteristics were extracted from existing RMS and FMS strategies. Some of the HRCIM cell characteristics were intermediate to that of RMS and FMS. Highlighted intermediate characteristics contained the ability to produce a standard library of mass customised parts, as compared to RMS and FMS that respectively processes part families and a diverse variety of parts. The vena diagram presented in figure 1 illustrates this methodology. This approach was facilitated to incorporate sufficient functionalities in producing mass customised parts, as compared to existing manufacturing strategies that lack or contain excessive functionalities in accommodating mass customisation.

RMS

HRCIM

CIM FMS based on CIM technology

Figure 1: Vena diagram for the HRCIM cell implementation

The HRCIM architecture composed of integrated hardware and software that generated synchronised manufacturing cell behaviour. The HRCIM cell at a systematic level was divided into two processing cells. Processing cell 1 and 2 respectively facilitated machining and quality inspection of parts. The simulated version of the HRCIM cell setup is depicted in figure 2. This setup illustrates the various machine level hardware used to construct the HRCIM cell.

110


Figure 2: Simio 3D overview of the HRCIM cell Performance results were based on the manufacture of four parts. Each part possessed different part geometries and processing parameters that required discrete HRCIM cell behaviour. This flow of heterogeneous part types was used to symbolise a mass customisation based problem that provided just enough variety and customisation. The part types also possessed identical directional path flow through the manufacturing cell. For more complex mass customisation based experiments an extended library of customised parts can be created. The hardware and software used to coordinate HRCIM cell functionalities are listed as follows: 3.1

Hardware Machine level hardware composed of storage systems which included an ASRS and an intermediate buffer. The intermediate buffer formed a common point to generate seamless part flow between processing cell 1 and 2. Material handling equipment consisted of an Automated Guidance Vehicle (AGV), conveyors and robotic arms. Two Fanuc six axes robotic arms as seen in figure 2, possessed reconfigurable manipulators and facilitated material handling for discrete part geometry flow. Robot 1 performed material handling in processing cell 1. Robot 1 facilitated material handling from the conveyor to the CNC machine for automatic part loading. It also facilitated unloading of parts from the CNC machine to placing of parts at the intermediate buffer. Robot 2 facilitated part transfer from processing cell 1 to processing cell 2 via the intermediate buffer. The robotic arms were integrated with vision processing techniques that generated part geometry capture along the material handling system and in response, based on the detected part geometry, reconfiguration of the end effector manipulators was generated automatically. Machine level hardware that facilitated part processing included a CNC milling machine and an Automated Modular Inspection Apparatus (AMIA).

3.2

Software Highlighted software based functionalities include an integrated Fanuc robotic vision software (iRvision) and the developed Master Control Program (MCP). The Fanuc incorporated vision software, known as iRvision, facilitated calibration and part geometry teaching. This mapped reconfiguration of the robotic arm end effector for specific part geometry manipulation. The MCP depicted in figure 3 and figure 4 was developed from the Visual basic programming language.

111


Figure 3: MCP This MCP provided a framework to engage the input of customer order parameters. Customer order parameters included the number of parts per order, and the time a customer expects the order to be complete. Each part type required different processing times that was standardised within the developed MCP. These parameters were used to compute a priority based scheduling on four customer orders. The MCP initiated ASRS retrieval of specific parts for processing based on these priority values. As seen in figure 3, part type A and B had similar geometries. This identical geometry is also illustrated for part C and D. This dissimilarity between part geometries, required reconfiguration of the robotic arm end effector manipulators in response to part handling between different detected part geometries. 4.

RESULTS AND DISCUSSION Simio, a manufacturing simulation software, generated detailed HRCIM cell results. Operational parameters from the physical cell were used to model the simulated manufacturing cell. According to the computed coefficient of priority values as detailed in section 3.2 and denoted in figure 4, the processing arrangement followed part A, D, C and B. Coefficient of priority values were dependent on part processing times, number of parts to be processed and delivery time. The order with the lowest coefficient of priority value was processed first and the highest coefficient of priority value was processed last [16]. This arrangement was applied to the simulated manufacturing cell. This arrangement was used to highlight machine changeover and reconfiguration of manipulators. This resulted respectively from the heterogeneous processing requirements and geometry properties between part A and D, and alternatively between part C and B. Due to identical geometry properties between part D and C, reconfiguration of the robotic arm manipulators was not necessitated. Each customer order contained a number of 10, 5, 5 and 15 parts for part type A, B, C, D respectively (see figure 4) .

112


Figure 4: MCP with customer input values The graphs presented in figure 5 and figure 7 respectively maps the initial times and manufacturing cell exit times for each part type. Reducing the time frame for each graph in the Simio framework provides easier user analysis. The graphs presented in this paper are of an hourly time frame that captures the whole simulation. Preferred user time frames that capture the simulation over a minute or second time frame can also be selected. As depicted, figure 5 maps the retrieval of parts based on the priority values. The vertical axis represented the number of parts for each part type. Steps in the graph signified part retrieval. Prolonged time lengths between part retrieval were in reaction to CNC changeover and robot reconfiguration.

Figure 5: Initial retrieval time for all part types

113


Figure 6 provides a Simio generated report that provides user clarification in determining the total number of parts (throughput) retrieved. This total was computed by incrementing part retrieval events that were initiated from the ASRS.

Figure 6: Simio report for the total number of parts retrieved Figure 6 shows the time for which each part exits the manufacturing cell. Extended time lengths for part exit arise from CNC changeover or part buffering at the CNC output buffer. Part buffering time at the CNC output buffer increased until robotic arm manipulation was facilitated. The order in which parts exited the manufacturing cell was based on a first in, first out basis.

Figure 7: Time for which parts exit the HRCIM cell Figure 8 in contrast to figure 6 provides a Simio generated report that denotes the total number of parts (throughput) that exited the manufacturing cell.

Figure 8: Simio report for the total number of finished parts These results were used to analysis HRCIM cell behaviour in response to discrete part type flow that represented a mass customisation approach. It provided means of capturing bottlenecks within the system due to part build. This arouse from machine transformations in order to match the architecture required to produce different part types. The graphs also provided sufficient detail in extracting cycle times for each part. This is achieved by computing the time difference from when a part is retrieved to when it exits the manufacturing cell. Cycle times are more precise when analysed over a minute or second time frame.

114


In contrast to RMS and FMS, the HRCIM cell possessed sufficient flexibility and reconfigurability to manufacture the part types available. Its design did not go beyond the flexibility required, nor was it deficient. This approach therefore reduced cost in producing customised parts at near mass rate. By bounding the number of customised parts that can be produced, less flexible and cost effective manufacturing equipment can be purchased. Standardising the customised parts available for manufacture reduces setup times that arise from the elimination of process planning, computer-aided design and computer-aided manufacturing procedures. 5.

CONCLUSION The HRCIM cell provided an innovative solution to producing mass customised parts. The manufacturing cell transformed to the requirements needed to accommodate discrete manufacturing events. This HRCIM cell provided a feasible approach that possessed sufficient flexibility and reconfigurability in facilitating mass customisation. Further improvement in RMS platforms can be used to perform research and development of more complex HRCIM cell designs. The HRCIM cell in response also adhered to engage reduced lead times and setup times, which results in short delivery times and significantly provides customer satisfaction.

6. [1] [2] [3] [4]. [5] [6] [7] [8] [9] [10] [11] [12] [13]

[14] [15] [16]

REFERENCES H. K. Shivanand, M. M. Benal, and V. Koti, FLEXIBLE MANUFACTURING SYSTEM. 2006, New Delhi: New Age International. V. Malhotra, T. Raj, and A. Arora, Reconfigurable manufacturing system: an overview. International Journal of Machine Intelligence, 2009. 1(2): p. 38-46 E. Sundin, et al., Integrated product and service engineering enabling mass customization, in International Conference on Production Research. 2007. Nambiar, A.N. Mass Customization: Where do we go from here? in Proceedings of the World Congress on Engineering. 2009. D. Pollard, S. Chuo, and B. Lee, Strategies For Mass Customization. Journal of Business & Economics Research, 2008. 6: p. 77-86. M. T. Fralix, From mass production to mass customization. Journal of textile and apparel, technology and management, 2001. 1(2): p. 1-7. V. Malhotra, T. Raj , and A. Arora, Excellent Techniques of Manufacturing Systems: RMS and FMS. International Journal of Engineering Science and Technology, 2010. 2: p. 137-142. R. M. Setchi and N. Lagos, Reconfigurability and Reconfigurable Manufacturing Systems: State-ofthe-art Review. Industrial Informatics, 2004: p. 529-535. Y. Koren, et al., Reconfigurable manufacturing systems, in Annals of the CIRP. 1999. p. 527-540. H. A. ElMaraghy, Flexible and reconfigurable manufacturing systems paradigms. Int J Flex Manuf Syst, 2006. 17: p. 261-276. C. Stoian and G. Frumusanu, Reconfigurable Manufacturing Systems Design Principles. The annals "Dunareae de jos" of galati fascicle v, technologies in mechanical engineering, 2007: p. 62-65. S. V. Nagalingam and G.C.I. Lin, CIM—still the solution for manufacturing industry. Robotics and Computer-Integrated Manufacturing, 2008. 24: p. 332-344. M. Yurdakul, Selection of computer-integrated manufacturing technologies using a combined analytic hierarchy process and goal programming model. Robotics and Computer-Integrated Manufacturing, 2004. 20: p. 329–340. K. D. Kumar, et al., Computers in manufacturing:towards successful implementation of integrated automation system. Technovation, 2005. 25: p. 477–488. Rehg, J.A., Introduction to robotics in CIM systems. Third ed. 1997, New Jersey: Prentice-Hall, Inc. M.K. Lim and Z. Zhang, A multi-agent based control strategy for responsive manufacturing. Journal of Materials Processing Technology 2003. 139: p. 379-384.

115


VERIFICATION OF FIVE-AXIS TOOL PATH OPTIMIZATION USING VERICUT S. Pholpho 1 and M. Munlin 2 Faculty of Information Science and Technology, Mahanakorn University of Technology, 140 Cheum-Sampan Rd., Nong Chok, Bangkok 10530 Thailand Tel: +662-988-3655 ext. 4112 Fax: +662-988-4027 Email: [email protected], Email: [email protected]

ABSTRACT When the tool of a five-axis milling machine travels near a stationary point, the rotation angles may change sharply leading to unexpected deviations from the estimated trajectories. Several algorithms have been proposed to optimize the rotations of the machine drives without increasing the number of tool positions or changing the tool orientations. The main idea is minimization of the distance traveled by the tool in the angular space at the expense of using multiple solutions of the inverse kinematics equations and switching the rotation angles or inserting new points at certain positions. We employ Vericut to verify theses algorithms. A virtual fiveaxis model of MAHO600E is constructed using its inverse kinematics and real machine parameters. Numerical experiments and cutting by a virtual five-axis milling machine built in Vericut validates the results of these optimization algorithms. The efficiency of these algorithms has been verified by a virtual machine as well as by real cutting on five-axis machine MAHO600E at the CIM Lab, Asian Institute of Technology of Thailand. Keywords: Five-axis machines, inverse kinematics, tool path simulation, kinematics error, CAD/CAM, Vericut 1.

INTRODUCTION Milling machines are programmable mechanisms for cutting industrial parts. The machine consists of several moving parts designed to establish the required coordinates and orientations of the tool during the cutting process. The axes of the machine define the number of the degrees of freedom of the cutting device. The movements of the machine parts are guided by a controller which is fed with NC program comprising commands carrying spatial coordinates of the tool-tip and angles needed to rotate the machine parts to establish the orientation of the tool. The main goal of five-axis tool path planning is minimization the difference between the desired and the actual surface while producing the actual surface for a minimum time. However, mathematical formulations presented in the literature vary in terms of the error criteria and the set of optimized variables. The tool path is optimized with regard to the machining time, accuracy, the length of the tool path, the width of the machining strip, the volume of the removed material, the size of the remaining scallops, etc. [1,2,3]. Furthermore, the error analysis and optimization in the areas of large variations of the rotation angles have not been provided by commercial CAD/CAM systems such as Unigraphics [8], EdgeCam, Vericut, etc. Besides, only a few research papers deal with the subject. However, the rotations invoke large kinematics errors. Besides, machines with rotation axes on the table often have to turn around heavy workpieces. As a result, the machines have low capacities for acceleration which significantly increases the machining time. This effect is amplified in high speed machining, when the rotation axes reach greater speeds. In [4] the authors analyze the sequence of rotations to minimize the number of the phase reverse steps at discontinuities of the first derivative of the surface (corners etc). A method of avoiding singularities has been presented in [5]. The method certainly has its merits since it allows inserting additional points without any modifications. However, the computation is complex, computationally expensive and does not preserve the original CC points. Modifying of the machine codes to solve the surface singularities problems have also been presented in [10, 11]. However, the methods keep inserting new CC points and slightly changing tool orientation until the cutting error is less than the tolerance.

116


In [12], the authors present a generic five-axis postprocessor system for various five-axis machine tools and verification by Borland C++ Builder and OpenGL. The wireframe model of the configured five-axis machine tool can be promptly shown and rotated or zoomed dynamically on the screen to assist the user to input relevant parameters correctly and efficiently. Through the implementation of the developed postprocessor and the verification by the solid cutting simulation software as well as the real machining experiment, the effectiveness of the proposed scheme was confirmed. In [13], the authors proposed a new simulation and verification system of belt grinding with industrial robots. The work piece surface is represented by a discrete height field, an array of extended height segments, and a fast collision detection algorithm using k-DOP bounding volumes is adopted to accelerate the localization of the contact area. A local grinding model is incorporated to decide the real material removal. Unlike the usual global linear model, it determines the removed material in the contact area based on the acting force distribution and some other grinding parameters. With this new system, robot programmers can improve the path planning by visualizing the manufacturing process, predicting potential problems and measuring dimensional errors. In [6] the authors proposed an angle switching algorithm to optimize the sequence of the rotation angles without increasing the number of tool positions or changing the tool orientations. The main idea is to minimize the distance traveled by the tool in the angular space at the expense of using multiple solutions of the inverse kinematics equations, which is, switching the rotation angles at certain position. Considering the entire set of angles requires the shortest path techniques to minimize the total angular distance. In [7] the algorithm was extended to the case when the cost function differentiates between damaging undercuts and repairable overcuts. Modern industries manufacture complex parts design using CNC machine to achieve higher accuracy and less machining time [1, 3]. However, using the real machine with the real work piece without computer simulation involves iterative trial and errors and therefore, increases costs and times. This paper presents the 3D simulation model of five-axis milling machine maho600e, constructed by using Vericut program. The simulator resembles the functions of the real maho600e from the given NC program. This makes it possible to analyze an accuracy of machining and well as perform collision detection. The results of the simulation can be used to simulate the milling process, verify the final cut and estimate the errors of the actual tool path before the real work piece is actually cut with the real machine. 2.

FIVE AXIS SIMULATION USING VERICUT Vericut is a CNC based machining simulation process. It simulates the exact depth, width, and angle of each cut. It knows exactly how much material is removed by each cut segment and divides the motion into smaller segments. Where necessary, based on the amount of material removed in each segment, it assigns the best feed rate for each cutting condition encountered. It then outputs a new tool path, identical to the original but with improved feed rates, but does not alter the trajectory. Tool path verification and optimization using Vericut are two of the best ways that can dramatically improve the manufacturing operation and save cost with relatively little work. In this paper, we employ Vericut to construct a virtual five-axis model of MAHO600E from its inverse kinematics and real machine parameters to verify the tool path optimization algorithms.

2.1

Inverse kinematics of a five-axis milling machines A five-axis milling machine normally has three translational and two rotary degrees of freedom. These configurations allow programming in any arbitrary tool axis orientation. Usually both rotary axes are perpendicular to each other and their orientation equals the direction of the linear axes. The primary rotary axes have a constant orientation, while the orientation of a secondary axis changes due to a rotation around the primary axis. One rotary axis has a fixed inclination with respect to the tool axis. The second rotary axis has a fixed orientation with respect to the clamping table. To simplify the formulae in this section the rotary axis which orientation remains fixed with respect to the tool axis will be defined as the principal rotary axis. If the conventional kinematics of a five axis milling machine is considered, the conclusion follows that each orientation of the tool-axis (i,j,k) can be achieved at two different positions of the rotary axes. However a limited range of the rotary axes often eliminates one of these solutions. If the principal B-axis is zero ( i=0, j=0, k=1) then any value for the second axis is satisfactory. This is called a degenerated position. If the tool axis is moved from the first solution space (B > 0) into the other solution space (B < 0) then the tool axis goes through the degenerated position (B = 0 and i=0, j= 0, k=1) [5].

117


Consider a typical configuration of the MAHO600E five axis milling machine with the rotary axis on the table as shown in Figure 1. Recall that the machine is guided by axial commands carrying the three spatial coordinates of the tool tip in the machine coordinate system M and the two rotation angles. The CAM software generates a set of successive coordinates called cutter location points or CL-points (X, Y, Z, I, J, K) in the work piece coordinate (X, Y, Z, A, B). Typically, the CAM software distributes the CL-points along a set of curves, which constitutes the so-called zigzag or spiral pattern. A post processing which includes a transformation to the M-system generates a set of machine axial commands which provide the reference inputs for the servo-controllers of the milling machine [14].

Figure 1. The corresponding reference coordinate systems

Figure 2. Non-linearity of the tool-path in the workpiece coordinates

C1

The kinematics of the machine depend on matrix-functions A(a), B(b) associated with the rotations a and b and around the primary (the rotary table) and the secondary (the tilt table) axes shown in Figure 1. A simple analysis of the inverse kinematics equations reveals that a linear trajectory of the tool tip in the machine coordinates may produce a non-linear trajectory in the work piece coordinates as shown in Figure 2. We shall call the deviation from the non-linear trajectory the kinematics error. Note that a fine cut of a smooth surface employing small spatial and angular steps may not demonstrate the detrimental effects near

118


the singularity points. However, a rough cut characterized by large gradients could produce considerable errors. It is because of the sharp angular jumps that the machine produces the loop-like trajectories of the tool. Moving along such trajectories may destroy the work piece and even lead to a collision with the machine parts. Furthermore, suppose that the tool vector changes the sign from positive to negative or vice versa, the inverse kinematics produces the singularity for which any value of the rotation angle is admissible [9]. Such singularity point on the surface presents a special case the rotation angles may jump considerably leading to unexpected deviations from the prescribed trajectory. Approaching the stationary point (minimum, maximum of saddle) involves sharp variations of the rotation angles even when the spatial steps are small. Moving along such trajectories may destroy the work piece and even lead to a collision with the machine parts. If the tool is aligned along the surface normal, then the rotation angles are evaluated by

abase

 1 j  tan ( i ) , i  0 and j  0,  j     tan 1 ( ) , i  0, i   1 j 2  tan ( i ) , otherwise. 

(1)

1

bbase   sin (k ) , (2) where (i,j,k) is the tool orientation vector. Furthermore, there are four sets of a-angles and b-angles within the range [0, 2 ] that can rotate the tool vector into the required orientation. The set of the a-angles is

defined by

{abase , abase  2 , abase   , abase   } as shown in Figure 3.

i0

+tan-1(j/i)

j

i>0,j>0 Tool vector

tan-1(j/i)

i

+tan-1(j/i)

+tan-1(j/i) i 1) vanishes for odd values of m, while for even values of m it is equal to (2/m) (tan1/2 ß)m, where ß denotes the operating angle

5.2 JOINT SELECTION (TORQUE RATING) The capacity of a universal joint is the torque which the joint can transmit. The torque capacity of a single Cardan joint of standard steel construction is determined as follows: i. From the required speed in RPM, operating angle in degrees, and service condition (intermittent or continuous) are calculated, ii. Multiply the required torque, which is to be transmitted by the input shaft, by the use factor. ii. Refer to the torque capacity column in the SOP catalogue and select a suitable joint having a torque capacity not less than the standard value. If a significant amount of power is to be transmitted and/or the speed is high, it is desirable to keep the shaft operating angle below 15°. For manual operation operating angles up to 30° may be permissible.

5.3 UNIVERSAL JOINT SELCTION FOR CONTINUOUS OPERATION A single universal joint is to transmit a continuously acting torque of 88.96 Nm, while operating at an angle of 15° and at a speed of 600 RPM. The required torque of 6049.58 Nm is obtained. There is no shock load and the dynamic factor.

5.4 UNIVERSAL JOINT SELECTION FOR INTERMITTENT OPERATION WITH SHOCK LOADING A single universal joint is to transmit 1/4 horsepower at 300 RPM at an operating angle of 15°. Select a suitable joint for intermittent operation with shock loading. Here we make use of the equation:

Torque = (horse power* 280349.16)/ rpm Nm Hence, operating torque = ((0.25)*(280349.16))/ 300 = 233.53 Nm (for intermittent running condition) RPM=300 ANGLE=15°, Assume use factor as 16. Due to shock loading there should be an additional dynamic factor of 2. Hence, the rated torque = (233.53) (16) (2) = 7473.01 Nm. Thus the same joints found in the previous example are usable in this case.

5.5 DETERMINING THE MAXIMUM SPEED OF AN INPUT SHAFT A universal joint is rated at 1112.05 Nm, and operates at an angle of 12°. Driving a rotating mass, which can be represented (together with the inertia of the driven shaft) by a steel, circular disc, radius r = 6", thickness t = 1/2", attached to the driven shaft. For ß = 12°, we have αmax/ω² = 0.0442. The weight, W, of the disc is π r² t y. where y = 1.25 Nm/in³ and denotes the density of steel. Thus W = π (6)² (0.5) (1.25) = 71.17 Nm. The polar mass moment of inertia, I, of the disc is given by I = Wr²/ 2g I = (71.17) (36) / (2*9.81)

326


I = 3.318 Nm-sec² The inertia torque = I α (max) = 50% of 250 in Nm = 125 Nm Hence, 0.0442 ω² I = 125 Nm Giving ω = 61.6 rad/sec = 588 RPM Hence, if the inertia torque is not to exceed its limit, the maximum speed of the input shaft is 588 RPM. For joints made with thermoplastic material, consult the catalogue, which contain design charts for the torque rating of such joints.

5.6 SECONDARY COUPLES In designing support bearings for the shafts of a Cardan joint and in determining vibrational characteristics of the driven system, it is useful to keep in mind the so-called secondary couples or rocking torques, which occur in universal joints. These are rocking couples in the planes of the yokes, which tend to bend the two shafts and rock them about their bearings. The bearings are thus cyclically loaded at the rate of two cycles per shaft revolution. The maximum values of the rocking torques are as follows:

Maximum torque on input shaft = T (input) tan β Maximum torque on output shaft = T (input) sin β Where Tin denotes the torque transmitted by the input shaft and ß the operating angle. These couples are always 180° out of phase. The bearing force induced by these couples is equal to the magnitude of the couple divided by the distance between shaft bearings. For example if the input torque, T in, is 1000 in-Nm. And the operating angle is 20°, while the distance between support bearings on each shaft is 6", the maximum secondary couple acting on the input shaft is (1000) (tan 20°) = 364 in-Nm. and on the output shaft it is (1000) (sin 20°) = 342 in-Nm. The radial bearing load on each bearing of the input shaft is 364/6 = 60.7 Nm. and for output shaft 342/6 = 57 Nm. The bearings should be selected accordingly.

5.7 PHASING In order to obtain a constant angular-velocity ratio (1:1) between input and output shafts, proper phasing of the joints is required. This phasing can be described as follows: two cardan joints in series will transmit a constant angular-velocity ratio (1:1) between two intersecting or non-intersecting shafts, provided that the angle between the connected shafts and the intermediate shaft are equal (ß = ß') and that when yoke 1 lies in the plane of the input and intermediate shafts, yoke 2 lies in the plane of the intermediate shaft and the output shaft. If shafts 1 and 3 intersect, yokes 1 and 2 are coplanar. When the above phasing has been realized, torsional and inertial excitation is reduced to a minimum. However, inertia excitation will inevitably remain in the intermediate shaft 2, because this shaft has the angular acceleration of the output shaft of a single universal joint It is for this reason that guidelines exist limiting the maximum angular accelerations of the intermediate shaft. Depending on the application values between 300 rad/sec² and values in excess of 1000 rad/sec² have been advocated. Stress determination is necessary for an accurate determination of allowable speed.

5.8 DETERMING THE MAXIMUM SPEED OF AN INPUT SHAFT IN A SERIES In a drive consisting of two universal joints in series, phased so as to produce a constant (1:1) angular velocity ratio between input and output shafts, the angle between the intermediate shaft and input (and

327


output) shaft is 20°. If the maximum Angular acceleration of the intermediate shaft is not to exceed 1000 rad/sec,

Since α (max) / 0.1250 = (1000)/ 0.125 = 8000 rad/sec² ω = (8000)´ = 854 RPM Hence, the speed of the input shaft should not exceed 854 RPM.

Figure .14. Maximum Angular Acceleration of the output shaft of a single cardan joint as a function of input speed and operating angle

Figure .15. Redesigned Universal joint 6.

CONCLUSION

From the static and the dynamic analysis of the propeller shaft, it has been observed that the maximum displacement occurs at the Universal joint and the following conclusions are made:  Failures are occurred as a result of fatigue process.  The crack beginning location of the joint yoke corresponds to highest stress points. Modification on the design of the joint may be considered for prevention of later failures.  The AISI 5046H is the most agreeable standard steel. AISI 5046H is a material in the low alloy steel group. Typical mechanical properties are 1750 MPa of tensile strength and 1400 MPa of yield strength. For certain strength levels, low-alloy steels have a good combination of strength, toughness and ductility.  The propeller shaft failure seems to be originated from improper heat treatment conditions; however the failed section and the crack beginning locations also coincide with the highly stressed regions. The mild stress concentration also speeds up the failure.

328


 Here, we offset the spider bearing from the center position so that the load transfers along the support member. By this the stress and the displacement value is reduced which will increase the life of the universal joint transmission which in turn increase the life of propeller shaft.

7.

REFERENCES [1] H. Bayrakceken , S. Tasgetiren, I. Yavuz ‖ Two cases of failure in the power transmission System on vehicles: A universal joint yoke and a drive shaft‖ Afyon Kocatepe University, Technical Education Faculty, Afyon 03200, Turkey [2] Gummadi Sanjay, Akula Jagadeesh Kumar ―Optimum Design and Analysis of a Composite drive shaft for an automobile‖ M.A.K. Chowdhuri et al./International Journal of Engineering and Technology Vol.2(2), 2010, [3] T.Rangaswamy, S. Vijayarangan, R.A. Chandrashekar, T.K. Venkatesh and K.Anantharaman ―Optimal Design and Analysis of Automotive Composite drive shaft‖ [4] Optimal design of the press fit joint for a hybrid aluminum/composite drive Shaft Hak Sung Kim, Dai Gil Lee, Korea Advanced Institute of Science and Technology, 373-1, Guseong-dong, Yuseong-gu, Daejeon-shi 305-701, South Korea [5] R. Clough, and J. Penzien, Dynamics of Structures, Second Edition, McGraw-Hill, Inc., ISBN 0-07011394-7, 1993. [6] A. Chopra, Dynamics of Structures, Prentice-Hall, Inc., Englewood Cliffs, New Jersey 07632, ISBN 013-855214-2, 1995. [7] K. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Inc., Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4, 1982.

329


EXPLORING THE APPLICATION OF TRIZ IN A GENERATIVE DESIGN SYSTEM M. C. Ang1, and K. W. Ng2 1

Institute of Visual Informatics and Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, UKM Bangi, Malaysia e-mail: [email protected] 2 Design Engineering Section, Product Design and Engineering Centre, SIRIM Berhad Bukit Jalil, Kuala Lumpur, Malaysia e-mail: [email protected]

ABSTRACT Generative design systems have been used in the support of the generation and exploration of design concepts. A generative design system is a system that generates design concepts and requires interaction with human designers to finalise on the final design concepts. Some generative design systems applied rules and dimensional constraints to avoid generating awkward design concepts whilst exploring design concepts that meet design requirements. However, most generative design systems focus on physical design requirements that define the form of the product. The theory of inventive design or TRIZ has been widely applied by major consumer products companies like Samsung and Motorola to derive innovative design concepts for their products and to solve their design problems. The design concepts are typically derived based on improving and worsening features required by the designers. In this research work, the feasibility of the TRIZ (theory of inventive problem solving) inventive principles with their related improving and worsening features were explored and further investigated to assist a generative design system to improve their process of generating design concepts in meeting design requirements. The outcomes of this feasibility study enable the derivation of a framework for an innovative generative design system with TRIZ support that can assist designers to explore design concepts within the specified design requirements. Keywords: TRIZ technical contradiction matrix, generative design system, product design 1.

INTRODUCTION Designers have been burdened with ever increasing challenges to derive innovative products with better specifications and less time at a lower cost in the current competitive global market. In order to assist designers to derive innovative products with better specifications and shorter time, researchers have come up with design methodologies, various design tools, and ways to stimulate cognitive capabilities of the designers. One of the interesting design tools that are widely researched is the generative design systems. These systems were ultimately researched to automatically derive design solutions on the notion that such automatic derivation of design solutions is able to explore a bigger solution space in a much shorter time. However, the generative design systems have severe short comings which will be explored more in the next section.

2.

GENERATIVE DESIGN SYSTEMS IN DESIGN Generative design systems have been used in the support of the generation and exploration of design concepts. Eckert et al. [1] define generative design systems as methods that generate product concepts based on a set of input specifications. These systems include meta-heuristic computations [2-6], heuristic rule methods [7], shape grammar approaches [8] and a combination of one or more approaches [5, 9]. For a generative design system, the knowledge and information derived from an existing product are crucial as they provide inputs to the system to develop a new improved product. These inputs are also used by the generative system to improve the existing product. Inputs can be ranged from the function of the product to the shape of the product. A schematic diagram representing a generative design system is shown in Figure 1 [1].

330


Generation of product concepts using generative design system

Selection/Setting of constraints by user or system

Evaluation by user or system

Figure 1: Components of a generative design system (adapted from [1]) The inputs for designing new products can be obtained via market research of consumers and users of an existing product in the market. A generative design system supports and allows the re-utilisation of the knowledge, information and data obtained from existing products. Though this method requires careful preparations to gather these inputs, the method can produce good results rapidly and is able to explore the space of possible solution concepts [1]. The ability to explore a large of amount of possible solution concepts rapidly is a significant advantage but there are critical drawbacks with such ability. This is because generative design systems can generate huge amount of design solutions with significant number of the poor design solutions that are unable to meet the design requirements. For generative design systems that are applied to produce new design shape, the design solutions will include a large number of awkward shapes. This will pose the problem of evaluating the suitability of the design solutions among a large amount of possible design solutions generated that are able to meet the design requirements particularly, if the evaluation is done interactively by the designers. Though, the problem can be reduced with appropriate constraints [4], any excessive use of constraints defeat the purpose of exploring design solution space [6, 10]. Graham [10] has applied genetic algorithms to evolve the shape of animal sculptures, seats and other products to assist designers in creating aesthetic shapes while Bentley [11] uses evolutionary algorithms to generate a variety of coffee table shapes. Agarwal [8] used shape grammar to generate various configuration of coffeemaker shapes. All these generative systems required designers to evaluate the generated design solutions. The cognitive abilities of designers are limited and prone to cognitive fatigue [6] when designers evaluate a huge number of possible design solutions generated. Furthermore, all generative design systems require an existing model of the product and the knowledge of key parameters of the existing model to control the generation. The current generative systems focus on modifying the shapes of the product and lacks of consideration for functional parameters of a product. This research work attempts to explore the feasibility of utilising the theory of inventive problem solving (TRIZ) particularly the improving and worsening features as well as the inventive principles in facilitating the process of evaluating the huge amount of design solutions generated by a generative design system particularly from functional aspects. The next section will explore the application of TRIZ particularly the technical contradiction matrix tool in design. 3.

APPLICATION OF TRIZ IN DESIGN TRIZ is widely applied in design among the industries [12] and are used to assist designers to create innovative products and solve design problems. TRIZ is a philosophy consists of a set of tools or methods that can be applied to solve design problems [13]. TRIZ was developed from years of studies on the patent information. One of the well-known TRIZ tools is the technical contradiction matrix and Figure 2 illustrates the schematic representation of how the technical contradiction matrix is developed. In this research work, we will focus only on the technical contradiction matrix of TRIZ.

331


Worsening Features

1

Inventive Principles

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

1: Weight of moving object *

-

15 8 29 34

-

29 17 38 3 4

-

29 2 40 28

-

2 8 15 38

8 10 18 37

10 36 37 4 0

10 14 35 4 0

1 35 19 39

28 27 18 4 0

5 34 31 35

-

6 29 4 38

19 1 32

35 12 34 3 1

-

12 36 18 3 1

6 2 34 19

5 35 3 31

10 24 35

10 35 20 2 8

3 26 18 31

1 3 11 27

28 27 35 2 6

28 35 26 1 8

22 21 18 2 7

22 35 31 3 9

27 28 1 36

35 3 2 24

2 27 28 11

29 5 15 8

26 30 36 3 4

28 29 26 3 2

26 35 18 1 9

35 3 24 37

2: Weight of stationary -

*

-

10 1 29 35

-

35 30 13 2

-

5 35 14 2

-

8 10 19 35

13 29 10 1 8

13 10 29 1 4

26 39 1 40

28 2 10 27

-

2 27 19 6

28 19 32 2 2

19 32 35

-

18 19 28 1

15 19 18 2 2

18 19 28 1 5

5 8 13 30

10 15 35

10 20 35 2 6

19 6 18 26

10 28 8 3

18 26 28

10 1 35 17

2 19 22 37

35 22 1 39

28 1 9

6 13 1 32

2 27 28 11

19 15 29

1 10 26 39

25 28 17 1 5

2 26 35

1 28 15 35

3: Length of moving object 8 15 29 34

-

*

-

15 17 4

-

7 17 4 35

-

13 4 8

17 10 4

1 8 35

1 8 10 29

1 8 15 34

8 35 29 34

-

1 35

7 2 35 39

4 29 23 10

1 24

15 2 29

29 35

10 14 29 4 0

28 32 4

10 28 29 3 7

1 15 17 24

17 15

1 29 17

15 29 35 4

1 28 10

14 15 1 16

1 19 26 24

35 1 26 24

17 24 26 1 6

14 4 28 29

4: Length of stationary -

35 28 40 2 9

-

*

-

17 7 10 40

-

35 8 2 14

-

28 10

1 14 35

13 14 15 7

39 37 35

15 14 28 2 6

-

1 10 35

3 35 38 18

3 25

-

-

12 8

6 28

10 28 24 3 5

24 26

30 29 14

-

15 29 28

32 28 3

2 32 10

1 18

-

15 17 27

2 25

3 15 40 14

63

-

2 15 16

15 32 19 1 3

19 32

-

19 10 32 1 8

15 17 30 2 6

10 35 2 39

30 26

26 4

29 30 6 13

29 9

26 28 32 3

2 32

22 33 28 1

17 2 18 39

13 1 26 24

15 17 13 1 6

15 30

14 1 13

2 36 26 18

2 10 19 30

35 39 38

-

-

-

17 32

17 7 30

10 14 18 3 9

30 16

10 35 4 18

2 18 40 4

32 35 40 4

26 28 32 3

2 29 18 36

27 2 39 35

22 1 40

40 16

16 4

16 15 16

1 18 36

2 35 30 18

29 1 40

15 13 30 1 2

10 15 29

26 1

29 26 4

35 34 16 2 4

10 6 2 34

1 31

2 17 26

-

35 37 10 2

19 -

10 15 19

32 8 35 24

3 1 35

1 26

26 -

30 14 7 26

5: Area of

Patent Information

Improving Features

moving object

2 17 29 4

-

14 15 18 4

-

*

-

7 14 17 4

-

29 30 4 34

19 30 35 2

10 15 36 2 8

5 34 29 4

11 2 13 39

6: Area of stationary

-

30 2 14 18

-

26 7 9 39

-

*

-

-

-

1 18 35 36

10 15 36 3 7

-

2 38

7: Volume of moving object

2 26 29 40

40 -

14 30 28 2 3

10 26 34 2

23 10 15 17 7

-

1 7 4 35

-

1 7 4 17

-

*

-

29 4 38 34

15 35 36 3 7

6 35 36 37

1 15 29 4

28 10 1 39

9 14 15 7

6 35 4

-

34 39 10 1 8

2 13 10

35 6 13 18

7 15 13 16

36 39 34 1 0

2 22

2 6 34 10

29 30 7

14 1 40 11

25 26 28

25 28 2 16

22 21 27 3 5

17 2 40 1

8: Volume of stationary -

35 10 19 1 4

19 14

35 8 2 14

-

-

-

*

-

2 18 37

24 35

7 2 35

34 28 35 4 0

9 14 17 15

-

35 34 38

35 6 4

-

-

-

30 6

-

10 39 35 3 4

-

35 16 32 1 8

35 3

2 35 16

-

35 10 25

34 39 19 2 7

30 18 35 4

9: Speed

2 28 13 38

-

13 14 8

-

29 30 34

-

7 29 34

-

*

13 28 15 1 9

6 18 38 40

35 15 18 3 4

28 33 1 18

8 3 26 14

3 19 35 5

-

28 30 36 2

10 13 19

8 15 35 38

-

19 35 38 2

14 20 19 3 5

10 13 28 3 8

13 26

-

10 19 29 3 8

11 35 27 2 8

28 32 1 24

10 28 32 2 5

1 28 35 23

2 24 35 21

35 13 8 1

32 28 13 1 2

34 2 28 27

15 10 26

10 28 4 34

3 34 27 16

10 18

-

10: Force (Intensity )

8 1 37 18

18 13 1 28

17 19 9 36

28 10

19 10 15

1 18 36 37

15 9 12 37

2 36 18 37

13 28 15 1 2

*

18 21 11

10 35 40 3 4

35 10 21

35 1014 2 7

19 2

-

35 10 21

-

19 17 10

1 16 36 37

19 35 18 3 7

14 15

8 35 40 5

-

10 37 36

14 29 18 3 6

3 35 13 21

35 10 23 2 4

28 29 37 3 6

1 35 40 18

13 3 36 24

15 37 18 1

1 28 3 25

15 1 11

15 17 18 2 0

26 35 10 1 8

36 37 10 1 9

2 35

3 28 35 37

11: Stress or pressure 10 36 37 4 0

13 29 10 1 8

35 10 36

35 1 14 16

10 15 36 2 8

10 15 36 3 7

6 35 10

35 24

6 35 36

36 35 21

*

35 4 15 10

35 33 2 40

9 18 3 40

19 3 27

-

35 39 19 2

-

14 24 10 3 7

-

10 35 14

2 36 25

10 36 3 37

-

37 36 4

10 14 36

10 13 19 3 5

6 28 25

3 35

22 2 37

2 33 27 18

1 35 16

2 36 37

35 24

10 14 35 3 7

12: Shape

8 10 29 40

15 10 26 3

29 34 5 4

13 14 10 7

5 34 4 10

-

14 4 15 22

7 2 35

35 15 34 1 8

35 10 37 4 0

34 15 10 1 4

*

33 1 18 4

30 14 10 4 0

14 26 9 25

-

22 14 19 3 2

13 15 32

2 6 34 14

-

462

-

14 10 34 1 7

36 22

10 40 16

28 32 1

32 30 40

22 1 2 35

35 1

1 32 17 28

32 15 26

2 13 1

1 15 29

16 29 1 28

15 13 39

15 1 32

17 26 34 1 0

13: Stability of the object 21 35 2 39

26 39 1 40

13 15 1 28

34 28 35 4 0

33 15 28 1 8

10 35 21 1 6

2 35 40

22 1 18 4

*

17 9 15

13 27 10 3 5

39 3 35 23

35 1 32

32 3 27 16

13 19

27 4 29 18

-

35 27

15 32 35

-

1 8 35

14: Strength 1 8 40 15

40 26 27 1

1 15 8 35

15 14 28 2 6

3 34 40 29

9 40 28

10 15 14 7

9 14 17 15

8 13 26 14

10 18 3 14

10 3 18 40

10 30 35 4 0

13 17 35

*

27 3 26

-

30 10 40

35 19

19 35 10

-

29 3 28 10

29 10 27

11 3

15: Durability of moving obj.

10 20 10 28 1 8

3 35 10 40

11 2 13

10 28 20 10 1 6

19 5 34 31

37 2 11 13

39 28 10 19 3 9

35 -

15 13 10 1

14 35 29 3 5

32 35 27 31 14 2 39 6

35 10 26 35 2 8

-

2 19 9

-

3 17 19

-

10 2 19 30

-

3 35 5

19 2 16

19 3 27

14 26 28 2 5

13 3 35

27 3 10

*

-

19 35 39

2 19 4 35

28 6 35 18

-

16: Durability of non moving obj. -

6 27 19 16

-

1 40 35

-

-

-

35 34 38

-

-

-

-

39 3 35 23

-

-

*

19 18 36 4 0

-

-

-

17: Temperature 36 22 6 38

22 35 32

15 19 9

15 19 9

3 35 39 18

35 38

34 39 40 1 8

35 6 4

2 28 36 30

35 10 3 21

35 39 19 2

14 22 19 3 2

1 35 32

10 30 22 4 0

19 13 39

19 1836 4 0

*

32 30 21 1 6

19 15 3 17

-

18: Illumination intensity 19 1 32

2 35 32

19 32 16

-

19 32 26

-

2 13 10

-

10 13 19

26 19 6

-

32 30

32 3 27

35 19

2 19 6

-

32 35 19

*

32 1 19

32 35 1 15

19: Use of energy by moving

12 18 28 3 1

-

12 28

-

15 19 25

-

35 13 18

-

8 35

16 26 21 2

23 14 25

12 2 29

19 13 17 2 4

5 19 9 35

28 35 6 18

-

19 24 3 14

2 15 19

*

-

6 19 37 18

12 22 15 2 4

20: Use of energy by stationary

-

19 9 6 27

-

-

-

-

-

-

-

36 37

-

-

27 4 29 18

-

19 2 35 32

-

*

-

21: Power

8 36 38 31

19 26 17 2 7

1 10 35 37

-

19 38

17 32 13 3 8

35 6 38

30 6 25

15 35 2

26 2 36 35

22 10 35

29 14 2 40

35 32 15 3 1

16 6 19

16 6 19 37

-

*

22: Loss of Energy

15 6 19 28

19 6 18 9

7 2 6 13

6 38 7

15 26 17 3 0

17 7 30 18

7 18 23

36 38

-

-

14 2 39 6

23: Loss of substance

35 6 23 40

10 28 24

35 2 10 31

10 18 39 3 1

1 29 30 36

3 39 18 31

10 13 28 3 8

14 15 18 4 0

3 36 37 10

29 35 3 5

2 14 30 40

35 28 31 4 0

35 6 22 32

14 29 10 3 9

24: Loss of Information 10 24 35

10 35 5

1 26

25: Loss of Time

10 20 26 5

15 2 29

26: Quantity of substance/th e 35 6 18 31

27 26 18 3 5

27: Reliability

3 10 8 28

28: Measureme nt accuracy 32 35 26 2 8

29: Manufacturi ng precision 28 32 13 1 8

-

19 35 10 3 8

26 -

28 27 3 18

16 2 14 17 25

-

16 -

2 14 17 25

28 27 3 18

18 35 24 30 1 8

3 27

3 3 27 16 40

3 35 31

34 27 6 40

10 26 24

21 36 29 3 1

-

35 28 21 1 8

3 17 30 39

19 35 3 10

32 19 24

13 1

16

19 1 26 17

1 19

-

11 15 32

3 32

35 24 18 5

-

35 38 19 1 8

34 23 16 1 8

19 21 11 2 7

3 1 32

-

-

28 27 18 3 1

-

-

3 35 31

10 36 23

-

10 35 38

28 27 18 3 8

10 19

35 20 10 6

4 34 19

19 24 26 3 1

32 15 2

21 17 35 3 8

32 13 16 1 6

27 16 18 3 8

13

3 27 16

-

1 -

11

2

35 19 1 35

35 40 27 3 9

35 19

32 35 30

2 35 10 16

35 30 34 2

2 35 22 26

35 22 39 2 3

18 35 37 1

15 35 22 2

11 3 10 32

32 40 25 2

27 11 3

15 3 32

2 13 25 28

27 3 15 40

22 15 33 2 8

21 39 16 2 2

27 1 4

12 27

29 10 27

1 35 13

10 4 29 15

19 29 39 3 5

17 1 40 33

24 22 33 35 2

22 35 10

1

1

2 -

23 35 40 3

15 29 35 10 1 4

6 10

35 17 14 1 9

25 34 6 35

1 20 10 16 3 8

22 35 2 24

26 27

26 27

4 10 16

2 18 27

2 17 16

3 27 35 31

26 2 19 16

15 28 35

15 19

35 19 32 3 9

19 35 28 2 6

28 26 19

15 17 13 1 6

15 1 19

6 32 13

32 15

2 26 10

2 25 16

1 35 6 27

2 35 6

28 26 30

19 35

1 15 17 28

15 17 13 1 6

2 29 27 28

35 38

32 2

12 28 35

-

10 2 22 37

19 22 18

14

-

-

-

-

19 35 16 2 5

-

16

32 2

19 22 31 2

2 35 18

26 10 34

26 35 10

35 2 10 34

19 17 34

20 19 30 3 4

19 35 16

28 2 17

21 22 35 2

21 35 2 22

-

35 32 1

2 19

-

7 23

35 3 15 23

15 34 33

32 28 2 24

2 35 34 27

15 10 2

35 10 28 2 4

35 18 10 1 3

19 38 7

1 13 32 15

-

-

3 38

*

35 27 2 37

19 10

10 18 32 7

7 18 25

11 10 35

27 16 18 3 8

21 36 39 3 1

1 6 13

35 18 24 5

28 27 12 3 1

28 27 18 3 8

35 27 2 31

*

-

15 18 35 1 0

6 3 10 24

10 29 39 3 5

16 34 31 2 8

35 10 24 3 1

33 22 30 4 0

10 1 34 29

10 19

19 10

-

*

24 26 28 3 2

24 28 35

10 28 23

-

-

22 10 1

10 21 22

-

-

-

35 33

10 5 18 32

35 18 10 3 9

24 26 28 3 2

*

35 38 18 1 6

10 30 4

24 34 28 3 2

24 26 28 1 8

35 18 34

35 22 18 3 9

35 28 34 4

4 28 10 34

32 1 10

35 28

6 29

18 28 32 1 0

24 28 35 3 0

-

2 22

26 32

-

-

-

-

-

2 5 34 10

35 16 32 1 8

-

10 37 36 5

37 36 4

4 10 34 17

35 3 22 5

29 3 28 18

20 10 28 1 8

28 20 10 1 6

35 29 21 1 8

1 19 26 17

35 38 19 1 8

29 14 35 18 -

15 14 29

2 18 40 4

15 20 29

-

35 29 34 2 8

35 14 3

10 36 14 3

35 14

15 2 17 40

14 35 34 1 0

3 35 10 40

3 35 31

3 17 39

-

34 29 16 1 8

3 35 31

6 3 10 24

24 28 35

35 38 18 1 6

*

18 3 28 40

13 2 28

33 30

35 33 29 3 1

3 35 40 39

29 1 35 27

35 29 25 1 0

2 32 10 25

15 3 29

3 13 27 10

3 27 29 18

8 35

13 29 3 27

15 9 14 4

15 29 28 1 1

17 10 14 1 6

32 35 40 4

3 10 14 24

2 35 24

21 35 11 2 8

8 28 10 3

10 24 35 1 9

35 1 16 11

-

11 28

2 35 3 25

34 27 6 40

3 35 10

11 32 13

21 11 27 1 9

36 23

21 11 26 3 1

10 11 35

10 35 29 3 9

10 28

10 30 4

21 28 40 3

*

32 3 11 23

11 32 1

27 35 2 40

35 2 40 26

-

27 17 40

1 11

13 35 8 24

13 35 1

27 40 28

11 13 27

1 35 29 38

28 35 25 2 6

28 26 5 16

32 28 3 16

26 28 32 3

26 28 32 3

32 13 6

-

28 13 32 2 4

32 2

6 28 32

6 28 32

32 35 13

28 6 32

28 6 32

10 26 24

6 19 28 24

6 1 32

3 6 32

-

3 6 32

26 32 27

10 16 31 2 8

-

24 34 28 3 2

2 6 32

5 11 1 23

*

-

28 24 22 2 6

3 33 39 10

6 35 25 18

1 13 17 34

1 32 13 11

13 35 2

27 35 10 3 4

26 24 32 2 8

28 2 10 34

10 34 28 3 2

28 35 27 9

10 28 29 3 7

2 32 10

28 33 29 3 2

2 29 18 36

32 23 2

25 10 35

10 28 32

28 19 34 3 6

3 35

32 30 40

30 18

3 27

3 27 40

-

19 26

3 32

32 2

-

32 2

13 32 2

35 31 10 2 4

-

32 26 28 1 8

32 30

11 32 1

-

*

26 28 10 3 6

4 17 34 26

-

1 32 35 23

25 10

-

26 2 18

-

26 28 18 2 3

10 18 32 3 9

30: Objectaffected harmful 22 21 27 3 9

2 22 13 24

17 1 39 4

1 18

22 1 33 28

27 2 39 35

22 23 37 3 5

34 39 19 2 7

21 22 35 2 8

13 35 39 1 8

22 2 37

22 1 3 35

35 24 30 1 8

18 35 37 1

22 15 33 2 8

17 1 40 33

22 33 35 2

1 19 32 13

1 24 6 27

10 2 22 37

19 22 31 2

21 22 35 2

33 22 19 4 0

22 10 2

35 18 34

35 33 29 3 1

27 24 2 40

28 33 23 2 6

26 28 10 1 8

*

-

24 35 2

2 25 28 39

35 10 2

35 11 22 3 1

22 19 29 4 0

22 19 29 4 0

33 3 34

22 35 13 2 4

31: Objectgenerated harmful 19 22 15 3 9

35 22 1 39

17 15 16 2 2

-

17 2 18 39

22 1 40

17 2 40

30 18 35 4

35 28 3 23

35 28 1 40

2 33 27 18

35 1

35 40 27 3 9

15 35 22 2

15 22 33 3 1

21 39 16 2 2

22 35 2 24

19 24 39 3 2

2 35 6

19 22 18

2 35 18

21 35 2 22

10 1 34

10 21 29

1 22

3 24 39 1

24 2 40 39

3 33 26

4 17 34 26

-

*

-

-

-

-

19 1 31

2 21 27 1

32: Ease of manufacture 28 29 15 1 6

1 27 36 13

1 29 13 17

15 17 27

13 1 26 12

16 40

13 29 1 40

35 12

35 19 1 37

1 28 13 27

11 13 1

1 3 10 32

27 1 4

35 16

27 26 18

28 24 27 1

28 26 27 1

14

27 1 12 24

19 35

15 34 33

32 24 18 1 6

35 28 34 4

35 23 1 24

-

1 35 12 18

-

24 2

-

*

2 5 13 16

35 1 11 9

2 13 15

27 26 1

6 28 11 1

8 28 1

35 1 10 28

33: Ease of operation

25 2 13 15

6 13 1 25

1 17 13 12

-

1 17 13 16

18 16 15 3 9

1 16 35 15

28 13 35

2 32 12

15 34 29 2 8

32 35 30

32 40 3 28

29 3 8 25

1 16 25

26 27 13

13 17 1 24

1 13 24

-

35 34 2 10

2 19 13

28 32 2 24

4 10 27 22

4 28 10 34

12 35

17 27 8 40

25 13 2 34

1 32 35 23

2 25 28 39

-

2 5 12

*

12 26 1 32

15 34 1 16

32 26 12 1 7

-

1 34 12 3

15 1 28

34: Ease of repair

2 27 35 11

2 27 35 11

1 28 10 25

3 18 31

15 13 32

16 25

25 2 35 11

2 35

11 1 2 9

11 29 28 2 7

15 1 13

15 1 28 16

-

15 10 32 2

15 1 32 19

2 35 34 27

-

32 1 10 25

2 28 10 25

11 10 1 16

10 2 13

25 10

35 10 2 16

-

1 35 11 10

1 12 26 15

*

7 1 4 16

35 1 13 11

-

34 35 7 13

1 32 10

35: Adaptability or versatility 1 6 15 8

19 15 29 1 6

35 1 29 2

1 35 16

35 30 29 7

15 16

15 35 29

-

35 10 14

15 17 20

35 16

15 37 1 8

35 30 14

35 3 32 6

13 1 35

2 16

27 2 3 35

6 22 26 1

19 35 29 1 3

-

19 1 29

18 15 1

15 10 2 13

-

35 28

3 35 15

35 13 8 24

35 5 1 10

-

35 11 32 3 1

-

1 13 31

15 34 1 16

1 16 7 4

*

15 29 37 2 8

36: Device complexity

2 26 35 39

1 19 26 24

26 14 1 13 16

6 36

34 26 6

1 16

34 10 28

26 16

19 1 35

29 13 28 1 5

2 22 17 19

2 13 28

10 4 28 15

-

2 17 13

24 17 13

27 2 29 28

-

20 19 30 3 4

10 35 13 2

35 10 28 2 9

-

6 29

13 3 27 10

13 35 1

2 26 10 34

26 24 32

22 19 29 4 0

19 1

27 26 1 13

27 9 26 24

1 13

29 15 28 3 7

*

15 10 37 2 8

15 1 24

12 17 28

37: Difficulty of detecting 27 26 28 1 3

6 13 28 1

16 17 26 2 4

26 2 13 18 17

2 39 30 16

29 1 4 16

2 18 26 31

3 4 16 35

30 28 40 1 9

35 36 37 3 2

27 13 1 39

11 22 39 3 0

27 3 15 28

19 29 39 2 5

25 34 6 35

3 27 35 16

2 24 26

35 38

19 35 16

18 1 16 10

35 3 15 19

1 18 10 24

35 33 27 2 2

18 28 32 9

3 27 29 18

27 40 28 8

26 24 32 2 8

-

22 19 29 2 8

2 21

5 28 11 29

25

12 26

1 15

15 10 37 2 8

*

34 21

35 18

38: Extent of automation

28 26 35 1 0

14 13 17 2 8

23 17 14 13

-

35 13 16

-

28 10

2 35

13 35

15 32 1 13

18 1

25 13

69

-

26 2 19

8 32 19

2 32 13

-

28 2 27

23 28

35 10 18 5

35 33

24 28 35 3 0

35 13

11 27 32

28 26 10 3 4

28 26 18 2 3

2 33

1 12 34 3

1 35 13

27 4 1 35

15 24 10

34 27 25

*

5 12 35 26

28 27 15 3

18 4 28 38

10 35 17 7

2 6 34 10

35 37 10 2

-

28 15 10 3 6

10 37 14

14 10 34 4 0

35 3 22 39

29 28 10 1 8

35 10 2 18

20 10 16 3 8

35 21 28 1 0

26 17 19 1

35 10 38 1 9

28 10 29 3 5

28 10 35 2 3

13 15 23

-

35 38

1 35 10 38

1 10 34 28

18 10 32 1

22 35 13 2 4

1 28 7 10

1 32 10 25

1 35 28 37

12 17 28 2 4

35 18 27 2

5 12 35 26

*

26 30 34 3 6

28 26 18 3 5

39: Productivity 35 26 24 3 7

30 7 14 26

10 26 34 3 1

4 18 39 31

18 13 34

1 34 9

1 11 10

13 1 13 2 4

1 4 10

-

1 35 20 10 6

35 7 18 25

1 35 20 10

32 27 22

28 35 10 2 3

-

35 35 13 8 1

19 -

35 10 18

10 35 17 4

3 8 10 40

10 -

2 28 10 29 3 5

30 16

30 24 14 5

10

32 -

28 35 34

-

26 4 5 16

10 20 37 3 5

26 30 26

7 16 35 38

35 -

26 10 28

19 10 35 3 8

2 14 30 40

35 35 28 31 4 0

35 -

2 1 26 13

35 22 18 3 9

35 28 2 24

35 13 23 15

2 22 35 18 3 9

1 27 34 35

35 28 6 37

Figure 2: Schematic representation of the technical contradiction matrix development The classical technical contradiction matrix consists of 39 improving and 39 worsening features in which, given any pair of improving and worsening feature, a corresponding list of inventive principles are recommended to solve the design problems that are related to the pairing of improving and worsening features[14]. However, the technical contradiction matrix has no recommended inventive principles when at the diagonal region of the matrix where an improving feature meets the same worsening feature. In order to utilise the technical contradiction matrix, the designers need to determine the features he is trying to improve and as a consequence of the improving these features, what are the features that will be worsened. Though the designers can select several improving features and worsening features, it would be highly

332


recommended that a design problem is reduced or further broke down into the root level where the improving features and worsening features are not too many [13]. Hence, one of the difficulties of using the technical contradiction matrix is to determine the relevant improving and worsening features for a design problem. This is because in reality, most of the improving and worsening parameters that a design problem need to be solved are not found in the list of 39 features given by the matrix. The designer has to map the actual improving and worsening parameters to the nearest relevant improving and worsening features of 39 in the matrix. Ultimately, the inventive principles recommended by the technical contradiction matrix are very abstract, ambiguous and require significant contributions of knowledge from the designers to translate them into specific solutions as shown in Figure 3. This is the one of the key issues with TRIZ. Designers are given examples of several patents related to each inventive principle in hoping that these examples can simulate the designers to come up with design solutions. Since TRIZ only provide generic design solutions and not specific solutions, novice designers that lack the experience and design knowledge may not be able to solve design problems using TRIZ.

Generic design solution (Improving and worsening features)

(Inventive principles) Designer‘s knowledge Specific design solution

Specific design problem

Figure 3: Components of TRIZ problem solving (adapted from[13]) As shown in Figure 2, improving and worsening features as well as the inventive principles of the matrix are derived from years of study on patent information and the study practically concluded in 1985 [15]. According to Mann [15], a significant amount of patents have been filed since 1985, there are changes in the deployment of the inventive principles in relation to the features though the number of inventive principles remain 40. The technical contradiction matrix was updated in 2003 by Mann [15] to include additional nine features (from 39 to 48) with the position of the features re-arranged to reflect the evolutionary S-Curve, one of the TRIZ tool that depicts the evolution of engineering system from infancy to decline. The re-arrangement of the position of the features is done in a manner that features that are highly active in infancy stage of a product is placed to towards the left for worsening features and towards to top for improving features. Further updates of the matrix were done in 2010 and the features were increased 50 with the addition of positive intangible feature and negative intangible feature [16]. Furthermore, the classical TRIZ technical contradiction matrix have quite a few pairings of improving and worsening feature that do not have any recommended inventive principles to help solve design problems. The new updated version of the contradiction matrix has recommended inventive principles for all those pairings of improving and worsening feature that does not have recommended inventive principles in the classical matrix with exception of the diagonal region where the improving and worsening feature is the same. In the new updated matrix, some of the pairings of improving and worsening feature have recommended inventive principles that differ from the classical matrix [15]. This is because the updating work was performed by re-analysing the patent information. By extracting features and inventive principles from the patent information, the pairing of an improving with the corresponding worsening feature may come up with different recommended inventive principles. Such work requires a huge amount of effort and time [17]. Li [18] combined TRIZ with analytical hierarchy process (AHP) to design automated assembly systems. He used TRIZ to determine the related improving and worsening features to the design parameters of selecting a suitable manufacturing and then determine the corresponding inventive principles. These inventive principles are then translated into relevant assembly processes. The design parameters are productivity,

333


safety and environment, quality, flexibility, and cost. AHP is used to evaluate the assembly systems. However, he also required domain experts to clarify and identify the features as well as to translate the inventive principles into specific solutions. Pham [19] applied TRIZ on top of a descriptive design framework to support designers in designing a construction device to support wood planks for concrete pouring in between beams. The TRIZ tool applied in the descriptive design framework was the technical contradiction matrix but with adaptations to allow trade-offs. The designer was using a descriptive design tool to capture his design activities. This descriptive design tool enabled him to decompose his design activities from requirements to sub-requirements and finally into ideas to solve his design problem. However, if the designer has difficulty in solving the design problem, the designer will apply the adapted contradiction matrix tool to generate inventive principles to assist him in solving the problem. In this scenario, the designers decided on how and where design requirements are decomposed into sub-requirements. However, he could call upon TRIZ to assist him to solve design problems whenever he encountered difficulties to solve any design problems to meet the subrequirements. Further experiments with TRIZ by Pham [19] indicated that the improving and worsening features as well as the inventive principles used by a novice designer can be identified and affiliated with the characteristics of the design model derived by an experienced designer without using TRIZ. The same experiment also found that the novice designer had difficulty in translating the inventive principles into a feasible design solution due to the abstractness of the inventive principles. 4.

APPLICATION OF TRIZ IN A GENERATIVE DESIGN SYSTEM With the research work elaborated in the generative design systems and the application of TRIZ, it can be observed that the issues with the evaluation problem in a generative system can be reduced by TRIZ. However, several key aspects need to be considered within the application of TRIZ itself. One of key aspects of the TRIZ application would be to expand the number of improving and worsening features as carried out by Mann [16]. With the expansion of these features, the functionality of each product can be better represented by these features. In this research work, study on the improving and worsening features as well as the inventive principles found that it is important for designers to apply TRIZ at root level of a design problem i.e. solving design problems for sub-requirements. The study also found that the inventive principles were too abstract and required crucial knowledge of the designers to translate the general solutions to specific solutions. For the generative design, it is discovered that the knowledge of the designers in evaluating the design solutions generated is also critical. In addition to that, the generating of design solution needs vital input provided by the designers on the control parameters determined by the designers themselves. The generative design systems should start with design requirements instead of an existing model of the product and should stress more on the functionality of the product. Investigation on both domains found that the success and failure of both systems are severely dependent on the designer‘s knowledge. In order to apply TRIZ in a generative design system successfully, TRIZ must provide knowledge feedback to the generative system to enable some level of evaluation to perform to reduce the burden of evaluation done by the designers. This research work then looked into the patent information and attempted to decompose each patent into its main functionality and their sub-functionalities that contribute to the main one. Each of these subfunctionalities has been tagged with a list of the possible 50 features from the updated TRIZ related to them. Therefore, instead of representing the sub-requirements with a textual description representation, they are replaced with numbers that represent the improving and the worsening features. In addition to that, similar to each of these sub-functionalities will be also tagged by inventive principles that correspond to it. Figure 4 illustrates the tagging of TRIZ (technical contradiction matrix) parameters to a design patent. Similar tagging works have been performed by Mann [15] to develop an updated technical contradiction matrix. Unlike his work, this research work attempts to derive knowledge templates from patent information. The designer needs to decompose the design requirements into sub-requirements and then into related improving and worsening features as well as the inventive principles. With the features and inventive principles specified, the generative design system can search the patent information tagged with similar features and inventive principles. The outcome of the search is a list of design patents that best match each sub-requirement or main requirements, depending on the preference of the designer. Figure 5 shows the framework of the generative design system with TRIZ support.

334


Patent No. XXXX Item description: xxxxxxxxxxxxxxx xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxx

MainSub-requirement requirement: xxxxxxxxxxx Level 1: xxxxxxxxxxxxx Improving Feature: x, x x, x Improving Feature:

xxxxxxxxxxxxxxxxxxxxxxx

Sub-requirement Level 2: xxxxxx Worsening Feature: x Worsening Feature: x

xxxxxxxxxxxxxxxx

Improving Feature: x, x Inventive Principle: x, x, x, x, x Worsening Feature: x Sub-requirement Levelx,1:x,xxxxxx

xxxxxxxxxxx

Improving Feature: x, x Sub-requirement Level 2: xxxxxx Worsening Feature: x, x, x

xxxxxxxxxxxxxx

Improving Feature: x, x

xxxxxxxxxxxxxx

Worsening Feature: x, x, x

xxxxxxxxxxxxxxxxx

Figure 4: Tagging of design patent information with TRIZ features and inventive principles

xxxxxxxxxxxxxxxxxxxx

xxxxxxxxxxxxxxxxxxxxxx

Design Patent Information

Main design requirements

xxxxxxxxxxxxxxxxxxx

xxxxxxxxxxxxxxxxxxxxxx

Decompose into main and sub-requirements

xxxxxxxxxxxxxxxxxxxxxxxxx Decompose into xxx

subrequirements

xxxxxxxxxxxxxxxxxxxxxxxxxx

Translate into TRIZ features and corresponding inventive principles

Translate into TRIZ features and corresponding inventive principles

Search, compare and match design solutions

Evaluate by designers

Final design solution Figure 5: The framework for a generative design system with TRIZ support

335


5.

CONCLUSION This research has explored both the application of the generative system and TRIZ in design. The exploration found that both TRIZ and the generative design system require significant contribution from the designer to be effective. TRIZ was derived from years of study in patent information to produce abstract and generalised principles to assist designers. However, these abstract and generalised principles should be able to trace back to the patent information. Such trace back would be more effective in helping designers to design better. The generative design systems were found to be helpful in generating design solutions from the perspective of exploring design space. Most of the current generative design systems focus on evolving on the existing product shapes and hardly stress on the functionality of the product. In addition to that, the huge number of design solutions generated also posed cognitive fatigue to the designers. In view of these findings, this research has attempted to trace back the features and inventive principles to the patent information. This is done by tagging patent information with related TRIZ features and inventive principles creating knowledge templates. For the generative design systems, the systems require the designers to translate the design requirements into TRIZ features and inventive principles. The generative design system then performs a search to compare the translated design requirements with knowledge templates created from the patent information. The outcome of the search is a list of best matched patent information that fulfils pre-determined design requirements. The designer still needs to evaluate and decide which design solution is preferred but the task of evaluation would be more focused as those solutions evaluated will have the best conformance to the design requirements.

6.

ACKNOWLEDGEMENTS The authors would like to thank Universiti Kebangsaan Malaysia for sponsoring this research by using the funding of the research project UKM-GGPM-ICT-102-2010.

7. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

REFERENCES Eckert, C., I. Kelly, and M. Stacey, Interactive generative systems for conceptual design: An empirical perspective. Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 1999. 13(4): p. 303-320. Bentley, P.J. Is Evolution Creative? in In Proceedings of the AISB‘99 Symposium on Creative Evolutionary Systems (CES). 1999: The Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB). Bentley, P.J. and J.P. Wakefield. Overview of Generic Evolutionary Design Systems. in In Proceedings of the 2nd On-Line World Conference on Evolutionary Computation (WEC2). 1996. Case, K., I. Graham, and R. Wood, Shape modification using genetic algorithms. Proc. Instn Mech. Engrs Part B: J. Engineering Manufacture, 2004. 218: p. 827-832. Pham, D.T., et al. Generating branded product concepts: Comparing the Bees Algorithm and an evolutionary algorithm. in Proc. of 4th Int. Virtual Conf. on Innovative Production Machines and Systems (I*PROMS 2008). 2008: Whittles Publishing. Krish, S., A practical generative design method. Computer-aided design, 2011. 43(1): p. 88-100. Kolodner, J.L. and L.M. Wills, Case-based creative design. 1993, Association for the Advancement of Artificial Intelligence (AAAI): Menlo Park, California. p. 95-102. Agarwal, M., J. Cagan, and K.G. Constantine, Influencing generative design through continuous evaluation: associating costs with the coffeemaker shape grammar. Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 1999. 13: p. 253-275. Ang, M.C., et al. Combining evolutionary algorithm and shape grammar to generate branded produce design to meet functional requirement. in Second International Conference on Design Computing and Cognition (DCC'06). 2006. Technical University of Eindhoven, Netherlands. Graham, I.J., K. Case, and R.I. Wood, Genetic algorithms in computer-aided design. Journal of Materials Processing Technology, 2001. 117(1-2): p. 216-221. Bentley, P.J., ed. Evolutionary Design by Computers. 1999, Morgan Kaufman: San Francisco. Shirwaiker, R.A. and G.E. Okudan, Triz and axiomatic design: a review of case-studies and a proposed synergistic use. Journal of Intelligent Manufacturing, 2008. 19(1): p. 33-47. Mann, D., Hands-on systematic innovation 2002, Leper, Belgium: CREAX Press.

336


[14] [15] [16] [17] [18] [19]

Altshuller, G., 40 principles: TRIZ keys to innovation, ed. L. Shulyak and S. Rodman. 1997, Worcester, MA: Technical Innovation Center. Mann, D., et al., Matrix 2003; Updating the TRIZ contradiction matrix. 2003, Leper, Belgium: CREAX Press. Mann, D.L., Matrix 2010; Re-updating the TRIZ contradiction matrix. 2009, Devon, UK: IFR Press. Cong, H. and H.T. Loh, Grouping of TRIZ Inventive Principles to facilitate automatic patent classification. Expert Systems with Application, 2008. 34(1): p. 788-795. Li, T., Applying TRIZ and AHP to develop innovative design for automated assembly systems. Int. J. Adv. Manufacturing Technology, 2010. 46(1-4): p. 301-313. Pham, D.T., K.W. Ng, and M.C. Ang. Applying TRIZ to support designers in a descriptive design framework. in Proc. of the ASME 2009 Int. Design Engineering Tech. Conf. & Computers and Information in Engineering Conf. (IDETC/CIE 2009). 2009. San Diego, California, USA.

337


A FRAMEWORK FOR SIMULATION COUPLING USING SEMANTIC WEB TECHNOLOGIES Rudolf Reinhard1, Tobias Meisen1, Daniel Schilberg1 and Sabina Jeschke1 1

RWTH Aachen, Institute of Information Management in Mechanical Engineering Aachen, Germany e-mail: {rudolf.reinhard, tobias.meisen, daniel.schilberg, sabina.jeschke}@ima.rwth-aachen.de

ABSTRACT Using simulation tools with the aim to support industrial planning processes became common over time. An often considered way to realise such an aim is the interconnection of simulations. Though, the interconnection of individual simulation tools of specific sub-processes often lacks optimal settings for several parameters, e.g. boundary parameters.

Furthermore, often there is no advantage taken of the experiences concerning production processes by doing so. In order to improve planning quality using simulations, the individual simulations need to be linked to form a continuous simulation chain. This paper contains a description of a methodology for the interconnection of simulations allowing the flexible extension of the overall system and thus an incorporation of a variety of heterogeneous simulation chains. Beyond the interconnection of distributed simulation tools there also tools provided, which enable members of a group the collaborative setting up of simulation chains. These methods are implemented comprehensively in a framework providing the extensibility and scalability for the interconnection of demanding simulations. Within the framework, the task of ensuring the correct syntactic, structural and semantic transformation of data between the individual heterogeneous simulations is completed by a data integration component. It ensures the integration of all needed simulation data into a common database. In this paper such a framework will be presented. Keywords: Virtual Production, Data Integration, Application Integration, Semantic Web Technologies 1.

INTRODUCTION Complexity in modern production processes increases continuously. Therefore, the virtual planning of these processes simplifies their realisation extensively and decreases their implementation costs. So far, several institutions have implemented their own simulation tools, which differ in the simulated production technique and in the examined problem domain. On the one hand, there are specialized simulation tools simulating a specific production technique with exactness close to the real object. On the other hand there are simulations which comprise production processes as a whole. The latter do not achieve prediction accuracy comparable to the one of specialized tools. However, both types are unexceptionally state-of-theart and commonly applied in university research. Furthermore most of the applied algorithms in these tools are not yet implemented in commercial tools. Hence, the simulation of a whole production process is often not realisable due to insufficient prediction accuracy or the missing support of the asked production techniques. In solving the problem, it is necessary to interconnect different specialized simulation tools and to exchange their resulting data. However, the interconnection is often not achievable because of incompatible file formats, mark-up languages and models used to describe the simulated objects. Therefore, the simulation of a production process as a whole using different simulation tools is hard to realise because of the missing consistency of data and interfaces [1]. Therefore, results received within a simulation can often only be integrated into another one after being checked manually and being adapted to the needs of following simulations, which is both tedious and faultprone. On the one hand, the huge data volumes being characteristic for simulation processes are not

338


supported by current available solutions. On the other hand, the possibilities to adapt a simulation process as a consequence of changes (e.g. integration of a new application, modification of a simulated object) are poorly supported. In this paper, the architecture of a framework for adaptive data integration is presented, which enables the interconnection of simulation tools of a specified domain. The framework provides generic functionality which, if customised to the needs for a specified domain (e.g. by transformation rules or data interfaces), supports the system to integrate any domain specific application in the process by making use of adaptive integration. For this purpose, this paper focus on the integration of data generated during the applications‘ usage, whereas the applications‘ link-up technique, which can be handled with the help of modern middleware techniques, will not be stressed. The framework is getting developed within the project ―Integrated Platform for Distributed Numerical Simulation", which is a part of the Cluster of Excellence ―Integrative Production Technology for High-Wage Countries" at RWTH Aachen University. The paper is structured as follows: In section 2 related works in the area of information integration and its relation to the results presented within this paper is presented. The consecutive section 3 presents the framework‘s architecture. Section 4 describes in a comprehensive manner how to perform the interconnection of simulations by an ontologies-based approach. Section 5 summarizes the results presented in this paper. 2.

STATE OF THE ART In the last three decades, data integration as well as Enterprise Application Integration (EAI) belongs to the most frequented topics across application boundaries [2]. Today, a multitude of data integration products can be found which are used in different fields of application. In general, the functionality of those products can be sub-divided into three categories [3] and illustrated in Figure 50: • Data propagation • Data federation • Data consolidation

339


Figure 50 - Main areas of data integration Regarding the operational section, Data Propagation is applied in order to make use of data on a crossapplication basis, which is often realised via data propagation. As already presented in [3], data propagation mainly focuses on small data volumes like messages and business transactions that are exchanged between different applications. In order to realize EAI, a contemporary architecture concept is used, which was developed in connection with service-based approaches [4] and which will be emphasized within this contribution - the so called Enterprise Service Bus (ESB). The basic idea of ESB, which can be compared to the usage of Integration Brokers, comprises the provision of services within a system [5]. Each service provides a technical or technological functionality with the help of which business processes are supported. The Integration Bus provides the services‘ connection. Transformation services provide general functions in order to transfer data from one format and model into another one. Against that, routing services are used to submit data to other services. Both transformation and routing services are used by adaptors in order to transfer data provided by the Service Bus into the format and the model of an application. Consequently, transformation services support the reuse of implemented data transformations. The advantage of a solution

340


based on the ESB pattern is to be seen in the loose interconnection of several services, whereas the missing physical data interconnection can be regarded as a disadvantage [6]: If recorded data has to be evaluated or to be analysed subsequently (e.g. with the help of data exploration techniques like OLAP or Data Mining), it will have to be read out and to be transformed once again. According to this fact, a historic or at least long-term oriented evaluation of data is inconvertible. In order to realize a unified examination on a cross-data basis, other sections belonging to the field of data integration need to be taken into consideration. Data Federation, which is examined within the field of Enterprise Information Integration (EII), might serve as one possible solution to enable a unified examination. With the aid of EII, data stored in different data sources, can be unified in one single view [3] and [7]. This single view is employed by the user to query this virtual, unified data source. The query itself is processed by the EII system by interrogating the underlying, differing data sources. Because of the fact that most EII do not support advanced data consolidation techniques, the implementation will only be successful if the data of the different data sources can be unified and if access to this data is granted (e.g. via query interfaces). Otherwise, techniques belonging to the field of data consolidation, which comprises the integration of differing data into a common, unified data structure, need to be utilised. Extract-Transform-Load (ETL) - a current process with regard to data integration - can be seen as one example of data consolidation [8]. ETL consists of the following aspects: The extraction of data from one or several - mostly operational - data sources, the transformation of the data format as well as of the data model into a final schema and, finally, the uploading of the final schema to the target data base. The presented sections of data integration (and not just those) have in common that, independent of the type of integration. The aim is to overcome the heterogeneity of data. In literature, different kinds of heterogeneity are distinguished [9] and [10]. In this paper, the types of heterogeneity listed in [11] will be stressed: • Technical heterogeneity • Syntactic heterogeneity • Data model heterogeneity • Structural or schema heterogeneity • Semantic heterogeneity The problem of technical heterogeneity, which addresses the problem of accessing data, can be handled with the help of modern middleware techniques [6]. Syntactic heterogeneity, a problem arising as a result of the representation of data (e.g. number formats, character encoding), is solved by converting the existing representation into the required one; in most cases, the conversion is carried out automatically. The handling of data model heterogeneity is more complex, as this kind of heterogeneity can be traced back to data using different data models (e.g. relational database, XML data model, structured text file). Nevertheless, modern data integration solutions provide readers and writers to access data from popular data models like relational databases or XML. Besides that, the support of other data models can be implemented. The combination of both structural and semantic heterogeneity is the most complex form of heterogeneity. Structural heterogeneity addresses the problem of representing data in one data model in different ways, for instance the usage of element attributes versus nested elements in a XML document. Semantic heterogeneity comprises differences in meaning, interpretation and in the type of usage of schema elements or data. Schema and ontology matching as well as mapping methods can be used to find alignments between data schemas as well as to process these alignments. Thereby, an alignment is a set of correspondences between entities of schemas that have to be matched. In the past years, several matching and mapping algorithms have been published [12]. However, these methods often focus on database schemas, XML schemas and ontologies without taking into account the background domain specific information [13]. This paper will not take a closer look at the last point mentioned.

341


3.

THE FRAMEWORK‟S ARCHITECTURE

The framework‘s architecture is based on the Enterprise Service Bus‘ (ESB) architectural concept. In order to realise a communication process between the integration server and the applications, a middleware is used that encapsulates the functionality of routing services, which are typical of those ones used in ESB concepts. Since a service does not provide necessarily any capability to communicate over messages it needs a further instance to undertake this task. This instance is the Service Activator. Each Service has its own Service Activator, whereas a Service Activator might also handle several Services. The Service Activator listens to the Integration Bus with the intention to identify any messages containing queries, which could be executed by one of the services the Service Activator cares about. Beside the query itself, the message contains also information about the requirements that need to be fulfilled by the Service. In the case there is a query, matching the capability of one of the services entrusted to the Service Activator‘s care, it locks one of its services to process this message and marks it as ―in work", so that there is no other service processing this query. The result of the service is packed into a message and is sent to the specified reply queue. Each process within a simulated production process is managed by the Process Manager. It writes messages containing queries into the Integration Bus‘ Queue, so that processes can be executed by a service. In addition, it cares about the process initiation and eradication. The Integration Bus consists in particular of a queue containing the different queries the Process Manager, beside other plugged Services, writes into. The framework ensures that each message can be read by at least one active Service Activator. The framework is employed with the intention of realising an integration level at which service providers, which are directly linked to the Integration Bus, make different services available. Due to the fact that the integration architecture needs to allow the easy substitution of one application by another one, the choice of a service-oriented architecture was helpful, to obtain an adaptable solution. In the following, there will be a concise explanation of the architecture‘s components. The services considered in this architecture comprise the following tasks: integration, extraction (both of them act as translators), analysis, transformation and planning. The Integration Services care about the processing of data for the further employment by making use of a particular application. That‘s why the Service Integration interface needs an own specialised implementation for each integration purpose. The Analysis Service checks the data that has been inserted into the database concerning their current structure as well as its semantics requested by the next simulation tool within the simulated production process. Thereby it determines how the current data have to be transformed for the next step. To define the transformation steps needed to prepare the data, in a way that they can be processed by the next simulation tool within the simulated production process, the Analysis Service has to parse the message in order to know which processes are necessary to fulfil the requirements written into the message. Each implementation of the Transformation Service cares about exactly one special aspect in the existing data. In most cases, it is not sufficient to make more than one step to modify output data of one application such that they can be processed by the next application. An important constraint is the order in which these transformations have to be executed as a request exists to obtain a fully automated interconnection of applications on the one hand and the determining of the kind of transformations and their execution order on the other hand. At this point, Planning Services come into consideration. They determine the kind of Services needed to perform the required operations and how these services have to interact. After their preparation by the appropriate, transformed data they get extracted by an Extraction Service. The Extraction Service cares about the extraction of data, which got recently processed by an application and is meant to get used by another one. In turn, the simulation results are stored within a file with a particular format. In certain cases it might be necessary to modify the input data. This step is called Enrichment. Since the communication between all components is message driven the question arises, how to activate the adequate service for a certain task. The Process Manager controls the realisation of the current step by an appropriate service instance within a running integration or extraction process. It does not know which functionality can be provided by any service, not about the data a service needs to run. Thus there is the need for an instance having exactly this knowledge. This instance is called Service Registry. It contains

342


information about available services, the functionality provided by each service and the input data that is required to run the service properly. A Gateway always belongs to a single application, which does not possess any capability to communicate with other architecture components over the Message Bus. The Gateway provides access for the architecture components to the application it belongs to and vice versa [14]. The described components, in particular the service oriented architecture allows to implement the concept of data integration in an adaptive way. This point will be considered in the following section. Figure 51 illustrates the concept described in this section.

Figure 51 - The framework's architecture

4.

ADAPTIVE DATA INTEGRATION

The main goal of the adaptive data integration is to overcome the problems of structural and semantic heterogeneity. The adaptive data integration is part of the enrichment process step, which can be assigned to the extended ETL process being used during the extraction of data. The objective of the extraction process consists in the generation of data in a given data format, taking into account the data model and structure as well as the semantics of this format. Therefore, the implemented enrichment allows the discovery and exploitation of background-specific information. The concept is based upon ontologies and planning algorithms that are usually applied in artificial intelligence. In the first instance, the existing data is analysed. The goal of the analysis is the determination of so-called features that are fulfilled by the data. A feature is domain specific, which means that it is expressing a structural or semantic property of the domain. Besides, the analysis step determines features that have to be fulfilled by the data to satisfy the requirements of the specific output format of the extraction process. Subsequent to the analysis, planning algorithms are used to find a data translation that transforms and enriches data in a way that allows for the fulfilment of features needed by the output format. After the planning is finished, the data translation, which is part of the executed step, is processed. The domain-specific data transformation algorithms are stored in transformation services following the ESB architectural concept, whereas the information about existing transformations and features is stored within an ontology. According to [15], ontology is an explicit specification of a conceptualization. In this chapter, the ontology-driven data integration will not be focused due to the limited space, which will not suffice to describe it in a proper way. 5.

CONCLUSION The development of the framework presented in this paper can be regarded as an important step in the establishment of digital production, as the framework allows a holistic, step-by-step simulation of a production process by making use of specialized tools. Both, data losses as well as manual, time-consuming

343


data transmissions from one tool to another are excluded by this approach. The suggested framework facilitates the linking of simulation tools, which were, ―until now―, developed independently from each other and which are specialized for certain production processes or methods, too. Furthermore, the integration of data generated in the course of the simulation is realized in a unified and process-oriented way. Apart from the integration of further simulation tools into an application that was already established, it is essential to extend the domain of simulations reflected upon with additional simulations covering the design domains of machine and factory. In this way, a holistic simulation of production processes is provided. Thereby, a major challenge consists in generating a central data model, which supports the possibility of illustrating data uniformly and in consideration of its significance in the overall context, which, in turn, comprises the levels of process, machines as well as materials. Due to the methodology presented in this article, it is not necessary to adapt applications to the data model aforementioned. On the contrary, this step is realized via the integration application, which is to be developed on the basis of the framework. Because of the unified data view and the particular logging of data at the process level, the framework facilitates a comparison between the results of different simulation processes and simulation tools. Furthermore, conclusions can be drawn much easier from potential sources of error. This is a procedure, which used to be characterized by an immense expenditure of time and costs. The realization of this procedure requires the identification of Performance Indicators, which are provided subsequently within the application. In this context, the development of essential data exploration techniques on the one side and of visualization techniques on the other side turns out to be a further challenge. 6.

ACKNOWLEDGEMENT The approaches presented in this paper are supported by the German Research Association (DFG) within the Cluster of Excellence ―Integrative Production Technology for High-Wage Countries".

7.

REFERENCES [1] G. Schmitz and U. Prahl, "Toward a virtual platform for materials processing" JOM Journal of the Minerals, Metals and Materials Society, vol. 61, pp. 19-23, 2009. [2] A. Halevy, A. Rajaraman, and J. Ordille, "Data integration: the teenage years" in VLDB'2006: Proceedings of the 32nd international conference on Very large data bases, pp. 9-16, VLDB Endowment, 2006. [3] C. White, "Data Integration: Using ETL, EAI, and EII Tools to Create an Integrated Enterprise" tech. rep., The Data Warehousing Institute, 2005. [4] D. Chappell, Enterprise Service Bus: Theory in Practice. O'Reilly Media, 2004. [5] R. W. Schulte, "Predicts 2003: Enterprise service buses emerge" tech. rep., Gartner, 2002. [6] T. Rademakers and J. Dirksen, Open-Source ESBs in Action. Greenwich, CT, USA: Manning Publications Co., 2008. [7] P. A. Bernstein and L. M. Haas, "Information integration in the enterprise" Commun. ACM, vol. 51, no. 9, pp. 72-79, 2008. [8] P. Vassiliadis, A. Simitsis, and S. Skiadopoulos, "Conceptual modeling for etl processes," in DOLAP '02: Proceedings of the 5th ACM international workshop on Data Warehousing and OLAP, (New York, NY, USA), pp. 14-21, ACM, 2002. [9] W. Kim and J. Seo, "Classifying schematic and data heterogeneity in multidatabase systems" Computer, vol. 24, no. 12, pp. 12-18, 1991. [10] C. H. Goh, Representing and reasoning about semantic confliicts in heterogeneous information systems. PhD thesis, Massachusetts Institute of Technology, 1997. Supervisor-Madnick, Stuart E. [11] U. Leser, Informationsintegration : Architekturen und Methoden zur Integration verteilter und heterogener Datenquellen. Heidelberg: Dpunkt-Verl., 1. Auflage, 2007. [12] J. Euzenat and P. Shvaiko, Ontology matching. Berlin/New York: Springer, 2007. [13] F. Giunchiglia, P. Shvaiko, and M. Yatskevich, "Discovering missing background knowledge in ontology matching" in Proceeding of the 2006 conference on ECAI 2006, (Amsterdam, The Netherlands, The Netherlands), pp. 382-386, IOS Press, 2006. [14] G. Hohpe, B. Woolf, Enterprise Integration Patterns, Addisson-Wesley, 2004 [15] T. R. Gruber, "A translation approach to portable ontology specifications" Knowledge Acquisition, vol. 5, pp. 199-220, 1993.

344


LAYOUT FORMATION IN CELLULAR MANUFACTURING SYSTEMS Sh. Ariafar1, N. Ismail2, S. H. Tang3, M. K. M. A. Ariffin4 & Z. Firoozi5 1

Department of Mechanical and Manufacturing Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor, Malaysia Department of Industrial Engineering, College of Engineering, Shahid Bahonar University, Kerman, Iran [email protected] 2 Department of Mechanical and Manufacturing Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor, Malaysia [email protected] 3 Department of Mechanical and Manufacturing Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor, Malaysia [email protected] 3 Department of Mechanical and Manufacturing Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor, Malaysia [email protected] 4 Department of Mechanical and Manufacturing Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor, Malaysia [email protected] ABSTRACT Facility layout is the arrangement of facilities, including aisles, departments, machines, tools and instruments in the shop floor in the most efficient position. Facility layout has an important role on the performance of manufacturing systems and factories. An efficient facility layout aims to decrease the material flow, work in process, lead time, and throughput time, while an inefficient arrangement of facilities, results in the accumulated work in process, and inefficacy of the material handling system. The aim of this study is to develop a mathematical model for layout design in a cellular manufacturing system that minimizes the inter-cell and intra-cell material handling costs. To validate the model several cases have been generated randomly, and solved by two methods; first by Lingo 12.0 optimization software, and then by an enumeration algorithm. The results show the feasibility and validity of the mathematical model. Keywords: Group technology, Facilities planning, Cellular manufacturing, Mathematical model, Material handlings, Lingo, Exact method, Enumeration, Inter-cell, Intra-cell. 1.

INTRODUCTION In today's competitive world economy, manufacturing industries have to keep up with the technological advances in order to be able to compete and to have a fair share of the market. Increasing demand for more customized products, a growing trend toward more automation to produce cheaper products with greater consistency, incompatibility of process layout in an automated environment, high cost of labour, and low utilization of machines are recent challenges of manufacturing systems [1]. To overcome these challenges, flexibility of manufacturing system plats an important role [2]. Group technology, which attempts to exploit similarities in design and processing of products to take advantage of these similarities, is a response to new requirements of manufacturing systems to increase the flexibility of the systems [3]. Cellular manufacturing system (CMS) is an important application of group technology [4]. It involves producing of similar parts on dedicated collections of dissimilar manufacturing processes or machines (machine cells) [5]. A machine cell is a collection of different machines dedicated to produce part families, and a part family is a group of parts that are similar in some factors such as size, geometric, shape or even manufacturing process. A variety of benefits have been reported by the companies which have implemented the cellular manufacturing system, including reduction in the material handling cost, reduced production lead time, setup time, work in process, rework and scrap materials, simplifications of production control procedures and improvement in quality of products [6].

345


The implementation process of a cellular manufacturing system involves the identification of part families, and grouping the machines (cell formation), and the arrangement of machine cells in the shop-floor, and machines in each cell (Layout problem) [7]. Cell formation is an important stage in the design of a CMS [8]. In this step, parts with similar manufacturing requirements are grouped into part-families, and different machines are dedicated to process these parts. Cell formation is a well studied problem in the CMS, and there are a lot of taxonomy and reviews in this field of the study [9], but the layout problem in the cellular manufacturing system rarely has been absorbed the attention of researchers as much as the cell formation [10]. Facility layout problem aims to arrange a group of facilities within a shop-floor in such a way that minimizes one or more qualitative or quantitative objective function[11]. Material handling cost, costs of laying communications wiring, or costs of scheduling product manufacturing are among the non-qualitative objectives, while vibration, noise disturbances, or flow of information or work between areas are some qualitative measures. In layout design of manufacturing systems, material handling cost is the most ever used objective function. However, in cellular manufacturing systems, facility layout problems aim to find the arrangement of facilities in the machine cells, and the layout of cells in the shop-floor in such a way that satisfies the zoning constraints. Chandrasekharan and Rajagopalan (1993) developed a method to minimize inter-cell movement in a CMS. For this purpose, they ranked the distance material movement, and changed the problem to a multidimensional non-metric scaling problem. For validation, they illustrated their method [12]. Grznar et al. (1994) proposed a non-linear mixed integer mathematical model for a layout problem in a CMS to minimize inter-cell material handling cost, when there is some kind of capacity and part requirement constraint. They developed a heuristic algorithm to solve the problem, and for validation, they numerically illustrated their method [13]. Bazargan-Lari and Kaebernick (1996) integrated intra-cell and inter-cell layout problem in a multi-objective model. They formulated their method as a goal programming model and developed a simulated annealing based algorithm to solve the model. They illustrated their model for validation [14]. Elwany et al. (1997) developed an expert system that combines a knowledge base with an improvement algorithm to solve an inter-cell layout problem in a cellular manufacturing system [15]. Wang et al. (1998) proposed a mathematical model for layout problems in a CMS that minimizes the intra-cell and inter-cell material handling cost. They developed an algorithm based on simulated annealing and verified the algorithm by comparing their results with similar results from the literature [16]. Daita et al. (1999) proposed a method for machine part clustering based on a production flow analysis. They demonstrated their approaches for several cases from the literature and also experiments from the industry [17]. Salum (2000) proposed a two-phase methodology for a layout design in a CMS that minimized manufacturing lead time and material handling cost [18]. Bazargan-Lari and Nahavandi (2001) proposed a three-phase procedure for developing a cellular manufacturing system, including the cell formation, intercell layout an intra-cell layout. They investigated the impact of each phase on other phases of procedure. Their model was verified by applying an example from the literature [19]. Vilarinho and Guimaraes (2003) developed a method for layout problem that determines the orientation and position of pick up/ drop off points for machines and manufacturing cells. They illustrated an example for their approach [20]. Hicks (2004) developed a genetic algorithm for design of a layout problem that minimizes the material handling cost for a given planning horizon. His proposed algorithm determines the geometric information on building and resource constraints [21]. Chrysostomos and Vlachos (2005) proposed a linear mathematical model to minimize the backward flow, and find the layout of machines in a manufacturing cell. They developed an ant colony system algorithm to solve their model, and demonstrated their algorithm [22]. Seo et al. (2006) addressed a methodology to determine a global solution to design manufacturing work cells and concurrently find a unidirectional flow path for the layout of facilities. They developed a heuristic for their problem by a decomposition approach [23]. Hu et al. (2007) integrated cell system layout design with a flow path structure. They developed a genetic algorithm for their problem, and compared it with existing methods from the literature [24]. TavakkoliMoghaddam et al. (2007) addressed the issues of stochastic demand on the inter-cell and intra-cell material handling cost in a cellular manufacturing system. They developed a model for their problem, and solved it by Lingo optimization software after linearization [7]. Mahdavi and Mahadevan (2008) developed an algorithm in a cellular manufacturing system that not only determines the cell design, but also identifies the

346


sequence of machines in each machine cell. They verified their algorithm by comparing its results with the available results from the literature [25]. Defersha and Chen (2009) integrated production planning and dynamic system reconfiguration in a cellular manufacturing system with an assumption that production quantities are decision variables. They developed a mathematical model for their problem, and also a heuristic algorithm to solve the model. They demonstrated their algorithm by several numerical examples [26]. Paydar et al. (2010) developed a multiple traveling salesman problem based on a simulated annealing algorithm to solve a cell layout and cell formation problem simultaneously. They illustrated their algorithm with several cases [9]. Jayachitra and Prasad (2011) studied the performance of various manufacturing systems such as the functional layout, cellular manufacturing system, and virtual cellular manufacturing system in an automotive manufacturing industry. They used simulation methods to validate their method [27]. The main objective of this paper is to develop a mathematical model for the layout design in a cellular manufacturing system that minimizes the total material handling cost (both inter-cell and intra-cell). The remainder of the paper is organized as follows. The mathematical model is presented in section 2. The solution methods (an enumeration algorithm and the application of Lingo 12.0 optimization software) have been illustrated in section 3, and finally section 4 concludes the paper. 2.

MATHEMATICAL MODEL The proposed mathematical model minimizes both intra-cell and inter-cell material handling costs. In the following variables and sets have been defined.

2.1. Assumptions The assumptions of the proposed model are as follows:       

The cell formation stage has been done before i.e. the type of facilities for each machine cell has been specified and known as a prior. Parts are moved in batches between the equipments and flow of material is static and deterministic. Besides, the material handling cost between each of the facilities is identified. The layout of facilities in each cell is considered to be U-shape i.e. the material handling system flows through a U-shape layout where the width and length of the layout are known beforehand. The size of all equipments is considered to be equal and the space and shape of the shop floor are not restricted. The distance between each two facility is measured from a centre of one facility to the centre of the other. The unloading point for each machine cell is at the leaving point of the cell and the loading point for the machine cell is at the entry point of that cell. The loading and unloading points of the facilities are at the centre of each facility.

2.2. Variables and sets The symbols and variables of the model are as follows: If machine i is appointed to the location j

1 X ij   0

Otherwise If machine i belongs to machine cell p

1 X ip   0 Otherwise

347


  M 1 , M 2 ,...., M n  : Set of machines

C  C1 , C 2 ,...., C w : Set of machine cells MFij : Material flow between machine i and machine j

Dkl : Distance between location k

and l

NMC (cr ) : Number of machines in machine cell Cr w : Total number of machine cells n : Total number of machine

n

n

n

n

w

w

MinZ   MF jl d ik X ij X kl X ip X kq

(1)

i 1 j 1 k 1 l 1 p 1 q 1

n

X i 1

ij

1

i  1, ..., n

ij

1

j  1, ..., n

(3)

ip

1

i  1, ..., n

(4)

ip

 NMC (C p )

n

X i 1 w

X p 1

n

X i 1

p  1, ..., w

(2)

(5)

The objective function (1) is to minimize the total handling cost (both inter-cell and intra-cell). Constraints (2) ensure that each machine is only assigned to one machine location, and constraint (3) assigns each machine location to one machine. Then, constraint (4) is to make sure that each machine location is assigned as a part of one machine cell, and constraint (5) allocates each cell the exact number of machines existing in that machine cell.

3.

ILLUSTRATION In this study, to solve and validate the mathematical model, several cases have been generated randomly and solved by two solution approaches. At first, the cases have been solved by Lingo 12.0 optimization software (Branch and Bound), and then by an exhaustive enumeration algorithm, which has been developed in C/C++. Both Lingo and the developed enumeration algorithm have been run on a Pentium 4 desktop 2.8 GHz with 1 GB RAM. The results of two methods have been summarized in the following tables.

348


Table 1: Computational results for Lingo 12.0 optimization software Number of Machines and Cells

Inter-cell layout

M=5 & C=2

Cells:

2- 1

M=6 & C=2

Cells:

1- 2

M=7 & C=3

Cells:

2- 3- 1

M=9 & C=3

Cells:

3- 2- 1

Intra-cell layout Cell 1: Cell 2: Cell 1: Cell 2: Cell 2: Cell 3: Cell 1: Cell 3: Cell 2: Cell 1:

5- 4 3- 2- 1 1- 2- 3 6- 4- 5 4- 5 7- 6 3- 1- 2 8- 9- 7 3- 4- 6- 5 2- 1

Cost ($) 1,310 2,120 1,830

2,260

Table 1: Computational results for the Enumeration Algorithm Number of Machines and Cells

Inter-cell layout

M=5 & C=2

Cells:

2- 1

M=6 & C=2

Cells:

1- 2

M=7 & C=3

Cells:

2- 3- 1

M=9 & C=3

Cells:

2- 3- 1

Intra-cell layout Cell 1: Cell 2: Cell 1: Cell 2: Cell 2: Cell 3: Cell 1: Cell 2: Cell 3: Cell 1:

5- 4 3- 2- 1 1- 2- 3 6- 4- 5 4- 5 7- 6 3- 1- 2 5- 6- 4- 3 7- 9- 8 1- 2

Cost ($) 1,310 2,120 1,830

2,140

As the results show in all the cases except the forth one, the Lingo optimization software has found the same solutions that have been found by the enumeration algorithm. It means that Lingo has been terminated with global optimal solutions in the first three cases. Since the mathematical model is a Mixed Integer NonLinear programming model and Lingo can only find the local optimums for this type of model. Finding global optimums for these cases shows the feasibility of the results for the first three cases. On the other hand, although for the last case Lingo has not found the optimum solution, the local solution which has been found, satisfies all the zoning constraints. So for this case, also the solution is feasible. However, feasibility of all the results in these cases is evidence that shows the validity of the mathematical model. 4.

CONCLUSION This study aimed to examine the issues of layout models in cellular manufacturing systems. For this purpose, a mathematical model for layout formulation in a CMS was proposed. Then two solution methodologies for solving the model were applied; (1) Lingo 12.0 optimization system (2) A full enumeration (exhaustive search) algorithm which was developed in C/C++. Illustration of the solution approaches on several random generated cases and comparison of the results demonstrated the applicability of two methods for small size cases. The results showed the feasibility and validity of the mathematical model. For future study, development of a heuristic algorithm to solve the model is suggested.

5.

ACKNOWLEDGEMENT The authors would like to thank sincerely the referees for their valuable comments in an earlier version.

6.

REFERENCES

[1] Z. Car and T. Mikac, "Evolutionary approach for solving cell-formation problem in cell manufacturing," Advanced Engineering Informatics, vol. 20, pp. 227-232, 2006. [2] N. C. Nayak and P. K. Ray, "Flexibility and performance relationships: evidence from Indian bearing manufacturing firm," International Journal of Modelling in Operations Management, vol. 1, pp. 67-83, 2010. [3] S. Ah kioon, A. A. Bulgak, and T. Bektas, "Integrated cellular manufacturing systems design with production planning and dynamic system reconfiguration," European Journal of Operational Research, vol. 192, pp. 414-428, 2009.

349


[4] X. Wu, C. H. Chu, Y. Wang, and W. Yan, "A genetic algorithm for cellular manufacturing design and layout," European Journal of Operational Research, vol. 181, pp. 156-167, 2007. [5] M. Solimanpur, P. Vrat, and R. Shankar, "Ant colony optimization algorithm to the inter-cell layout problem in cellular manufacturing," European Journal of Operational Research, vol. 157, pp. 592-606, 2004. [6] G. A. Süer, A. Subramanian, and J. Huang, "Heuristic procedures and mathematical models for cell loading and scheduling in a shoe manufacturing company," Computers and Industrial Engineering, vol. 56, pp. 462-475, 2009. [7] R. Tavakkoli-Moghaddam, N. Javadian, B. Javadi, and N. Safaei, "Design of a facility layout problem in cellular manufacturing systems with stochastic demands," Applied Mathematics and Computation, vol. 184, pp. 721-728, 2007. [8] L. K. Saxena and P. K. Jain, "Dynamic cellular manufacturing systems design-a comprehensive model," International Journal of Advanced Manufacturing Technology, vol. 53, pp. 11-34, 2011. [9] M. M. Paydar, I. Mahdavi, I. Sharafuddin, and M. Solimanpur, "Applying simulated annealing for designing cellular manufacturing systems using MDmTSP," Computers and Industrial Engineering, vol. 59, pp. 929-936, 2010. [10] S. Ariafar and N. Ismail, "An improved algorithm for layout design in cellular manufacturing systems," Journal of Manufacturing Systems, vol. 28, pp. 132-139, 2009. [11] K. S. N. Ripon, K. Glette, O. Mirmotahari, M. Høvin, and J. Tørresen, "Pareto optimal based evolutionary approach for solving multi-objective facility layout problem," vol. 5864 LNCS, ed, 2009, pp. 159-168. [12] M. P. Chandrasekharan and R. Rajagopalan, "A multidimensional scaling algorithm for group layout in cellular manufacturing," International Journal of Production Economics, vol. 32, pp. 65-76, 1993. [13] J. Grznar, A. Mehrez, and O. Felix Offodile, "Formulation of the machine cell grouping problem with capacity and material movement constraints," Journal of Manufacturing Systems, vol. 13, pp. 241-250, 1994. [14] M. Bazargan-Lari and H. Kaebernick, "Intra-cell and inter-cell layout designs for cellular manufacturing," International Journal of Industrial Engineering : Theory Applications and Practice, vol. 3, pp. 139-150, 1996. [15] M. H. Elwany, A. B. Khairy, M. G. Abou-Ali, and N. A. Harraz, "A combined multicriteria approach for cellular manufacturing layout," CIRP Annals - Manufacturing Technology, vol. 46, pp. 369-371, 1997. [16] T. Y. Wang, H. C. Lin, and K. B. Wu, "An improved simulated annealing for facility layout problems in cellular manufacturing systems," Computers and Industrial Engineering, vol. 34, pp. 309-319, 1998. [17] S. T. S. Daita, S. A. Irani, and S. Kotamraju, "Algorithms for production flow analysis," International Journal of Production Research, vol. 37, pp. 2609-2638, 1999. [18] L. Salum, "The cellular manufacturing layout problem," International Journal of Production Research, vol. 38, pp. 1053-1069, 2000. [19] M. Bazargan-Lari and S. Nahavandi, "The impact of cell formation on layout designs in cellular manufacturing," Intelligent Automation and Soft Computing, vol. 7, pp. 13-22, 2001. [20] P. M. Vilarinho and R. C. Guimaraes, "A procedure for the facility layout problem with fixed geometry resources," International Journal of Industrial Engineering : Theory Applications and Practice, vol. 10, pp. 413-419, 2003. [21] C. Hicks, "A genetic algorithm tool for designing manufacturing facilities in the capital goods industry," International Journal of Production Economics, vol. 90, pp. 199-211, 2004. [22] F. Chrysostomos and A. Vlachos, "Optimal solution of linear machine layout problem using ant colony system," WSEAS Transactions on Information Science and Applications, vol. 2, pp. 652-662, 2005. [23] Y. Seo, D. Sheen, C. Moon, and T. Kim, "Integrated design of workcells and unidirectional flowpath layout," Computers and Industrial Engineering, vol. 51, pp. 142-153, 2006. [24] G. H. Hu, Y. P. Chen, Z. D. Zhou, and H. C. Fang, "A genetic algorithm for the inter-cell layout and material handling system design," International Journal of Advanced Manufacturing Technology, vol. 34, pp. 1153-1163, 2007. [25] I. Mahdavi and B. Mahadevan, "CLASS: An algorithm for cellular manufacturing system and layout design using sequence data," Robotics and Computer-Integrated Manufacturing, vol. 24, pp. 488-497, 2008. [26] F. M. Defersha and M. Chen, "A simulated annealing algorithm for dynamic system reconfiguration and production planning in cellular manufacturing," International Journal of Manufacturing Technology and Management, vol. 17, pp. 103-124, 2009. [27] R. Jayachitra and P. S. S. Prasad, "Performance analysis of Virtual Cellular Manufacturing: A simulation study," International Journal of Services and Operations Management, vol. 8, pp. 92-107, 2011.

350


THE DESIGN, DEVEOPMENT AND VALIDATION OF A STRUCTURED REVIEW METHODOLOGY FOR HEALTH AND SAFETY PERFORMANCE MANAGEMENT IN THE CHEMICAL INDUSTRY Dr. Chakib Kara-Zaïtri and Dr Saad Alquahtani University of Bradford, School of Engineering, Design and Technology Bradford, UK e-mails: [email protected], and [email protected] ABSTRACT Despite the increasing prevalence of H&S standards and tighter regulations, accidents in the chemical industry continue to rise with a significant cost to the industry and economy. Current audit systems have worked well to a certain extent but they remain vague and open to interpretation. The methodology developed uses extensive H&S performance literature reviews and three stages of empirical research. Results have been analysed based on central tendency and variation and yielded six critical but orthogonal aspects for review: Prevention, Surveillance, Response, Achievements, Resources, and H&S Management and Enhancement. The review methodology promotes the rapid rectification of any H&S issues identified through the publication of associated shortcomings using a graded profile (4: Standard Exceeded, 3: Standard met, 2: Standard almost met, and 1: Standard not met) as well as good practice. Indications from a pilot implementation of the methodology in Saudi Arabia are encouraging and have already unveiled critical gaps, which would have hitherto been missed using previous methods. The methodology can readily be applied to other industries. Keywords: H&S, Performance, Review, Empirical Research, Chemical Industry, Saudi Arabia. 1.

LITERATURE REVIEW ON H&S MANAGEMENT The chemical industry currently has a poor public image due to well-publicised major chemical disasters. This, as well as more demands from stakeholders, regulatory bodies, government, media, and the public [1] has forced chemical industries to be even more vigilant with regard to H&S. ‗H&S pays‘ is a commonly used phrase, but many companies are paying lip service to this when it comes to putting the concept into practice [2]. More often than not, the role of management and organisational factors are reported as major causes H&S accidents. All governments stipulate the establishment and maintenance of an H&S management system [3, 4, 5]. The implementation of such maintenance programmes demands a significant amount of time, especially in Small Medium Enterprises [6, 7]. The question remains as to whether or not audit performance measures used reflect actual safety effectiveness. In order to survive in today‘s highly competitive global business environment, chemical companies continue to be challenged to undertake a variety of activities for measuring, managing and controlling H&S risks [8]. There is therefore a clear and urgent need for a new proactive approach for reviewing H&S management performance [8] coupled with a change from measuring loss-type accidents to measuring the potential occurrence of accidents before they occur [9].

1.1 H&S management activities One of the basic requirements in H&S is the establishment of a safety culture [10]. This has been defined as ―a product of individual and group values, attitudes, perceptions, competence, and patterns of behaviour that determine the commitment to, and the style and proficiency of, an organisation‘s safety management‖ [11]. Three cultural factors contributing to the development of a strong safety culture have been identified [12] as Commitment, Competence and Cognisance.

Integration of H&S with other management activities, such as Quality and Environment Management, has been long recognised [13]. For instance, there are many common features between BS-8800 [14], ISO-9001 [15] and ISO-14001 [16] in respect of H&S activities. Weinstein [17] explained how TQM development steps can be applied to the development of a safety management system. Similarly, Total Quality Management and environmental hazards management have many elements similar to safety management

351


[17, 18]. Nowadays, integrated SHEQ (Safety, Health, Environment, Quality) systems have been introduced in many organizations [19, 20], with various models available [14]. Understanding why and how accidents and other unwanted events develop is important when preventive activities are planned [21]. A number of safety management-related standards, directives, and regulations have been published during the 1990s. BS 8800 [14] became the first widely used safety management standard while the ‗Seveso‘ Directive 82/501/EEC in 1982 presented the principles for the management of major accident hazards in the chemical and petrochemical industry. 1.2 Recent Occupational H&S standards These include the International Safety Audit system (ISA 2000) and the Occupational Heath and Safety Assessment System by the British Standards Institute [22] (OHSAS 18001). Interestingly, ISA 2000 represents a departure from the usual approach to managing H&S. Its starting point is not legislation, but good management practice. ISA 2000 contains a set of initiatives, comprising over 200 elements, which are categorised as either mandatory or supplementary. OHSAS 18001 is more focused on H&S policy, planning, implementation and operation, corrective action and management review. 1.3 Evaluation of H&S management systems Assessing the different aspects of the H&S management system can be done with different evaluation methods [23] as discussed in the sections below. 1.3.1 Measurement of performance: H&S performance is a measure of the completeness and adequacy of the H&S management system operating on-site. Only when a number of control loops integrate effectively all relevant system elements at all levels of operation, the system can operate fully. Such control loops include fundamental elements, such as implementation procedures and training; policy setting and standards drawn by regulations and norms; monitoring, control and revision; and operator and equipment reliability [24]. Reactive indicators such as Fatal Accident Rate (FAR), Lost Time Injury (LTI) rate, and other output indicators have been used extensively in assessing process safety. 1.3.2 H&S audits A safety audit is intended to assess the company‘s safety status reviews, safety policy, safety organisation, implementation of the safety activities, and the performance measurement system [14]. Cooper [25] points out that although an audit may be able to identify the most serious problems, it does not always identify every existing problem, no matter what sort of audit is used, and what the focus of the audit (e.g. quality, safety, and environment) might be. Petersen [26] mentioned that the first tools developed for assessing safety management systems were checklists. These were followed by simple yes-no type audit methods combined with the use of complicated audit tools. Today, safety audit tools typically consist of a list of safety activities to be assessed, and associated criteria for evaluation. Diekemper and Spartz [27] developed one of the earliest multidisciplinary audit tools. Since then, several other methods can be found in the literature. For example, the eighth edition of the International Safety Rating System (ISRS) [28] includes updates to reflect changes in international standards including the

international occupational health and safety management system specification OHSAS 18001:2007, the international quality management standard ISO 9001:2008 and the Global Reporting Initiative:2006. Another good example is the computerised health and safety and environment audit tool CHASE [29]. 1.3.3 Discussions on current management safety audit methods While the Diekemper and Spartz (D&S) method gives little weight to policy, organisation and administration, and a lot of emphasis is put on hazard control activities, CHASE and ISRS focus more on management aspects. However, CHASE gives little importance to behavioural safety, whereas both D&S and ISRS methods give more weight to these activities. Thorough scientific validation studies have been carried out only with the ISRS method. In all of these studies, accident records were used as a reference. The outcomes of the validation studies are varied. Some studies showed reduced accident rates among ISRS users, and others showed no link at all.

352


1.3.4 Critique on literature review on H&S Measuring performance is very important since this reflects achievements in H&S performance and management. Measuring is not a single action but should be considered as a continuous process, i.e. as a normal function of management. It is good to have target setting for performances, so that these can be properly coordinated, and people know what they are supposed to achieve. There are clearly several performance measures, which can be used including the ISRS method, which has been used extensively. The safety audit methods described, focus on the management of accident hazards that can lead to an injury. Behavioural safety, that is, the management of personnel through leadership and motivation is not considered to the same extent. 2.

SYNTHESIS OF H&S PROBLEMS IN SAUDI CHEMICAL INDUSTRIES In spite of the Saudi government‘s significant effort to support the chemical industry, the country suffers from significant H&S problems as follows: a.

Worrying increase in major occupational accidents and associated injuries.

b.

Difficulty in measuring H&S performance [30].

c.

Difficulty in integrating current Safety, Health and Environmental Management (SHE) audits [31].

d. A lack of technical and skilled manpower with an over-dependence on a foreign work force which is usually untrained in safe work practices. f. Lack of involvement and co-ordination among the many different government ministries and agencies in establishing, implementing and monitoring H&S regulations. g. 3.

Low number and lack of experience of auditors.

EMPIRICAL RESEARCH The empirical research carried out is based on three instruments: a pilot study, structured interviews and a comprehensive questionnaire.

3.1 Phase 1: Pilot study A comprehensive pilot study was conducted to establish the awareness and commitment to H&S, and to assess the applicability of H&S indicators obtained from previous studies to the Saudi chemical industry. A number of experts from 8 chemical industries, accounting for 15 to 20% of the Saudi workforce, were identified and interviewed based on findings from a similar European survey on H&S and Environment in small process plants called SPASE [7]. The pilot study revealed a number of key H&S factors including complying with regulations, ethical duty to protect workers and organisational learning and communication. 3.2 Phase 2: Structured Interviews This focused on identifying the most commonly used H&S indicators within Saudi chemical companies based on a structured questionnaire incorporating H&S activities, motivators, management system, external information sources, performance measures, and communication and learning. Industries in this research included chemical, petrochemical and other companies manufacturing fertiliser, plastic, and rubber. Two out of six industrial areas, Riyadh and Dammam, have been selected by purposive sampling including a total of 35 companies of different sizes and types. The two selected areas are in effect the largest in Saudi Arabia and account for 45% of the entire population. Interviewees, who were mainly H&S managers, answered the questions in a series of meetings, recorded verbatim for further analysis. They had enough confidence in the interviewer and were reassured about data confidentiality. The data collected from the structured interviews were subjected to statistical analyses including central tendency and variation analyses and yielded a list of 60 most commonly used H&S indicators (modal distribution 50-100%) within Saudi chemical companies.

353


3.3 Phase 3: Questionnaire The objective of phase 3 of the empirical research was to identify more precisely the ―important‖ and ―critical‖ indicators from the 60 identified in Phase 2 that could affect H&S performance in Saudi chemicals companies. A carefully designed questionnaire was developed to measure the perceived criticality of these H&S indicators commensurate with that by Thiagarajan [32], which was used to measure the perceived importance of quality management factors in a company. Respondents were asked to rate H&S indicators as Critical, Important or of Minor importance in their respective companies. The questionnaire was administered to the same sample of 74 interviewees of Phase 2 of the empirical research. The indicators were given in the questionnaire as simple statements without any leading questions. Respondents were assured of confidentiality of information provided. As argued in the literature [33, 34] two measures of spread were used in the analyses and include the Variation Ratio and the Index of Diversity. The Variation Ratio (VR) is the proportion of cases that do not fall into the modal category [33]. A value of zero means unanimity. Values of 0.5 or less mean majority consensus, while values more than 0.5 indicate no majority consensus. However, the VR does not take into account the full distribution of responses. The measure of spread that does this is the Index of Diversity. This is defined as a dispersion measure based on a proportion of cases in each category [34] as:

IoD  1  Pk

2

Where Pk is the proportion of cases in category k, and k is the number of categories.



A low index value means general agreement on the importance of an aspect, while a high index value  means general disagreement. This means that an index value close to zero implies near unanimity. A value close to 0.05 is, when there is equal cluster (concentration) around two large categories. A near uniform distribution in the three rating categories gives a maximal value close to 0.06, which means a high level of disagreement. Results from the 74 respondents are shown in Table 1 including Variation Ratio (VR) and Index of Diversity (IoD) values for each of the 60 indicators. Overall, from the 74 questionnaires, it was found that the respondents had rated 40 indicators as critical and 20 indicators as important. It was interesting to observe that many of the H&S indicators rated as critical are similar to the ones discussed in the literature review especially the SPASE survey. Interesting new results were responses to ‗Personal protective equipment to control H&S risk‘, ‗H&S awareness‘, ‗Pressure from industry organisation as motivator to H&S‘, and ‗Importance of risk assessment to control H&S risk‘ as critical indicators. 4.

DEVELOPMENT A STRUCTURED REVIEW METHODOLOGY FOR H&S PERFORMANCE MANAGEMENT The conceptualisation of the review methodology for H&S management performance has been derived from a combination of two methods by the Quality Assurance Agency for Higher Education (QAA), which describes the method and procedures for carrying out subject reviews in England and Northern Ireland during 2000 and 2001, and the Health and Safety Executive, which describes a H&S Performance Measurement Tool in 2000 [35]. The aspects of review were arrived at from a grouping of the critical H&S indicators, identified in Phase 2, after a detailed brainstorm session with experts.

Table 1. Empirical research results for Phase 2 with Variation Ration (VR) and Index of Diversity (IoD) values No 1 2 3 4 5 6 7 8

H&S Indicator H&S issues to be managed by company H&S policy to manage H&S enhancement H&S manual and procedure to manage H&S enhancement H&S committees to manage H&S enhancement H&S prizes/rewards to manage H&S enhancement Job analysis in terms of H&S to control H&S risk Risk assessment to control H&S risk Emergency planning to control H&S risk

354

Critical 70 55 30 47 50 44 66 61

Important 3 14 36 25 20 21 7 9

Minor 1 5 8 2 4 9 1 4

VR 0.05 0.26 0.51 0.36 0.32 0.41 0.11 0.18

IoD 0.10 0.41 0.59 0.48 0.47 0.55 0.20 0.30


9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Personal protective equipment (PPE) to control H&S risk Ventilation system to control H&S risk Chemical handling and labelling to control H&S risk Accident record and investigation to evaluate H&S performance Inspection& housekeeping to evaluate H&S performance Attitude/ perception surveys to evaluate H&S performance H&S audits/housekeeping audits to evaluate H&S performance. Financial losses to evaluate H&S performance. H&S awareness influence on effectiveness of H&S activity Leadership influence on effectiveness of H&S activity Supervisor/line management influence on H&S activity Employee behaviour/attitude influence on H&S activity Employee participation influence on H&S activity Documentation and data control influence on H&S activity Internal communication influence on H&S activity Employee training influence on H&S activity Financial resource influence on H&S activity Performance evaluation influence on H&S activity Complying with regulations motivator for H&S Expensive cost of accident/ill health motivator for H&S Providing safe work place for employees motivator for H&S Publicity/images motivator for H&S Pressure from authorities motivator for H&S Pressure from employees motivator for H&S Meeting customers/suppliers requirements motivator for H&S Satisfying society motivator for H&S Pressure from industry organisation motivator for H&S Good management procedures reason for implementing H&S Competitive advantage reason for implementing H&S Acquiring customers reason for implementing H&S Facilitating compliance with H&S legislation Improve H&S performance reason for implementing H&S Publications as a source for company on H&S matters Contact with relevant authorities as a source for company on Workshops/seminars/ meetings source for H&S H&S Chambers of Commerce source for H&S Advice from consultants source for H&S Advice from insurance companies as a source for H&S Monitoring and measurement of performance vis-avis H&S Promotion of goods and services for performance measures Performance monitoring based on performance measures Compliance with authorities based on performance measures Communication of performance measures to interested parties Improve H&S motivator for using performance measures Use accident investigation as learning for H&S Use unsafe activities reported as learning for H&S Use H&S meeting learning activity for H&S matters Use team observation learning activity for H&S matters Use enforcing safe job procedures as learning for H&S Employees involved in learning exercise at the company Supervisors line management involved in learning H&S committee involved in learning exercise in company

72 54 27 63 26 35 40 64 70 30 50 53 55 26 45 49 33 34 65 52 55 30 60 47 27 25 66 55 30 25 40 50 31 40 56 61 47 23 51 44 48 42 27 45 57 29 43 23 30 50 46 29

1 17 37 9 45 38 25 10 3 40 22 13 15 40 26 21 40 37 7 17 15 40 10 18 40 29 7 11 33 40 30 22 38 31 12 11 22 41 15 26 23 24 40 25 15 41 23 45 39 20 16 38

1 3 10 2 3 1 9 0 1 4 2 8 4 8 3 4 1 3 2 5 4 4 4 9 7 20 1 8 11 9 4 2 5 3 6 2 5 10 8 4 3 8 7 4 2 4 8 6 5 4 12 7

0.03 0.27 0.5 0.15 0.39 0.49 0.46 0.14 0.05 0.46 0.32 0.28 0.26 0.46 0.39 0.34 0.46 0.5 0.12 0.3 0.26 0.46 0.19 0.36 0.46 0.61 0.11 0.26 0.55 0.46 0.46 0.32 0.49 0.46 0.24 0.18 0.36 0.45 0.31 0.41 0.35 0.43 0.46 0.39 0.23 0.45 0.42 0.39 0.47 0.32 0.38 0.49

0.05 0.41 0.60 0.26 0.51 0.51 0.58 0.23 0.10 0.54 0.45 0.44 0.40 0.57 0.51 0.48 0.51 0.54 0.22 0.45 0.40 0.54 0.32 0.52 0.57 0.66 0.20 0.41 0.61 0.58 0.54 0.45 0.56 0.53 0.39 0.30 0.50 0.58 0.47 0.52 0.48 0.56 0.57 0.51 0.36 0.54 0.55 0.53 0.55 0.47 0.54 0.57

The six aspects of review are Prevention, Surveillance, Response, Achievements, Resources and Management and Enhancement. A mapping of the H&S indicators from the empirical research and the six aspects are shown in Figure 1.

355


Empirical research

H&S aspects

findings Literature review  Regulation, inspection and enforcement  Workplace culture  Safety policy  Management responsibility  Communications and learning  H&S training  Employees‘ participation in safety promotions  Industrial environment  Supply of PPE 

Prevention:  Risk assessment  Emergency planning  H&S motivators  Relevant authorities source on H&S

Surveillance:    

Phase 1: Pilot study  Complying with regulations  Ethical duty  Protect workers  Management responsibility  Learning and communications  Regulations and awareness  Government encouragement

Response:    

Phase 2: Structured interviews          

    

H&S meeting Team observations Knowledge exercises Learning new lessons

Achievements:

Compliance with H&S regulations Workplace accidents Prevention and awareness practices Surveillance for H&S H&S monitoring Learning and communications H&S performance achievement H&S information Sources of H&S information Promote H&S activities

    

Manage H&S performance Evaluate H&S performance Workplace analysis Accident investigation Adverse event

Resources:  H&S manuals  Equipment and information technology  Environmental  Training

Phase 3: Questionnaire survey     

Identify adverse trends Effectiveness H&S activities Monitor exposure Confidentiality of information

H&S policy Hazards prevention & control Management commitment Training & instruction Communicate with employer and employees Confidentiality of information H&S awareness Management commitment Employee training Enhance safety in workplace

H&S management and enhancement:  Implement management standards  Enhance safety in workplace  Effective manage net arrangements  Promotion of H&S activities  Action to update

Figure 1. Mapping of empirical research findings to the six Health and Safety Aspects The six aspects are sufficiently broad and inclusive to enable companies and reviewers to give full weight to the particular features of the companies in different contexts.

356


4.1 Prevention This aspect covers all work carried out to prevent H&S incidents and accidents from happening in the first place. It covers the following four areas: a. Hazard identification: How structured are the methodologies used to identify all aspects of H&S, including a consistent risk assessment model with clearly defined risk attributes, and an organised methodology clearly showing what might go wrong, how it might go wrong, what might cause it to go wrong, and what corrective options are available. b. Hazard mitigation: How appropriate are the control measures and emergency plans in relation to the working environment and the underlying industry, considering the staff profile, the hazards identified, and the internal and external reports on the effectiveness of the control measures. c. H&S awareness: How motivated are staff to recognising and observing H&S rules, taking into consideration the working environment, the culture of the company, and the pressure from the stakeholders and regulatory bodies. d. Currency and innovation: Is there evidence that hazard management is informed by recent developments in the field, industrial and professional advice where relevant, and developments in physical resources. 4.2 Surveillance This aspect evaluates the extent to which companies are continuously monitoring early signals and covers the following three areas: a. Health monitoring: What precautions are there for monitoring the health of employees and what analyses and corrective actions are carried out as a result, including the monitoring of sickness records, health incidents, and regular check-ups. b. Safety monitoring: What systems are in place to monitor any safety risk including exposure to hazardous substances and what analyses and corrective actions are carried out as a result, including the monitoring of exposure, safety incidents, and safety reports. c. Continuous improvement: What systems are in place to continuously improve the process of early detection using a variety of means including recent developments in the field. 4.3 Response This aspect focuses on how well companies have responded to H&S incidents and covers four key areas as follows: a. H&S strategy: Does the strategy articulate clearly the requirements of everybody in the organisation? Is it appropriate in terms of the overall protection vis-à-vis H&S, the planned incident response, and the resources available, including all staff. b. Incident Response: What does the evidence (derived from scrutiny of materials, observations, staff questionnaires and meetings with staff) reveal about the strengths and weaknesses identified in relation to the quality of response, communication, command and control during the incident. c. Incident investigation: What systems are in place to properly investigate the response after the incident and identify areas for improvement including structured methodologies used to review the entire incident and the options for quick recovery. d.

Lessons learned: To what extent are lessons learned captured, disseminated and implemented.

357


4.4 Achievements This aspect focuses on hard evidence in terms of key achievements that reflect how well the H&S standard is applied at the company and covers the following two main areas: a. Key Performance Goals (KPG): What targets have been set by the company to monitor H&S Management including accident rates, incident rates, lost Work Days, insurance Costs, staff complaints, non-compliances with given standards, and the number of high priority corrective actions. b. Key Performance Indicators (KPI): How close are the KPIs to the stated KPGs in terms of corrective action plans. 4.5 Resources Reviewers gather evidence through direct examination of all H&S resources available at the company. This aspect covers the following main six areas: a. Resources strategy: Is there an appropriate overall strategy for resources? Are there effective arrangements for maintaining, replacing and updating resources? Is the resources strategy consistent with support for the industry, incident response strategy, and the needs implied by the staff profile and the working environment? b. H&S Manuals: Are the H&S Manuals available, accessible and appropriate in terms of the requirements of incident response, the standard operating procedures and the arrangements for staff induction. c. Equipment and Hazardous Substances: Are all such resources protected adequately, accessible, and appropriate in respect of the industry under consideration. d. The working environment: Is available and appropriate in terms of the range and layout of the plant, space for work, and social, dining and recreational facilities. e. Training of Staff: Are all staff adequately trained with regards to H&S taking into consideration training needs, materials, programmes and reviews. f. Technical support: Is appropriate technical support available for specialist areas, maintenance for safety critical equipment and devices and support review. 4.6 H&S management and enhancement This aspect focuses on how the H&S process is managed ―now‖, and how it is continuously enhanced, and covers the two areas as follows: a. H&S Management: How effective are the internal arrangements for monitoring and evaluating the current H&S system? What H&S policy exists and to what extent does it cover all aspects relating to H&S. Do these arrangements involve appropriate consideration of appropriate management structures, budgets and plans, internal monitoring data, external reports, views of staff, contractors, visitors, and professional bodies, where appropriate, other internal or external reviews and staff development needs. b. H&S Enhancement: This includes the significant outcomes of the H&S management process in terms of revision of the H&S strategy, identification and implementation of action points required to meet the strategy more fully, identification of internal indicators/measures of effectiveness, plans for future enhancement, dissemination of good practice, processes for opportunities for enhancement identified and considered, induction arrangements for new staff appraisal of staff in terms of H&S issues, and take-up and application of staff development activities related to H&S. This area also covers the quality of the selfassessment and the associated consultative process with employees. 5.

PROPOSED H&S REVIEW METHODOLOGY The review methodology consists of the following steps:

358


a.

Appointment of a Review Chair and Specialist Peer H&S reviewers avoiding all conflict of interest.

b. Preparation of a self-assessment document by the company including an assessment of each of the six aspects supported by evidence. c. Preparation for Review Visit by the review team including advance planning so that both parties can plan and agree the timing of individual reviews and the range of H&S work available for scrutiny. d. Review Visit for enabling the reviewers to assess, consider and test the evidence of the effectiveness of H&S management performance. e. Grading and overall summative judgment which includes the assignment of grades to each of the six aspects drawing on evidence from both the self-assessment and the review visit as shown in Table 2. Table 2. H&S aspects grading

Grade 4

Qualifier Standard exceeded

Description Commendable: Above the required levels of performance

3 2

Standard met Standard almost met

1

Standard not met

No shortfalls: achieving the required levels of performance Minor shortfalls: no major deficiencies and required levels of performance seems achievable without extensive extra activity Major shortfalls: significant action is needed to achieve the required levels of performance

f. Making an overall judgment, which is informed by reviewers‘ collective judgements on the effectiveness of H&S management performance in relation to each aspect of H&S. g. Managing the outcome of the review: a 1 grade in any aspect results in an unsatisfactory outcome and a further full review will be conducted after six months. A profile with all aspects graded 2 or better will be reported as approved. However: where the graded profile includes two or more 2 grades, the company will be requested to write an improvement plan. 6.

VALIDATION OF THE METHODOLOGY Five companies were selected for a pilot implementation and included mixed-sized companies. Fifteen H&S experts were recruited for conducting the review. Several feedback questions were posed to the expert team after the pilot implementation and included the following aspects: Contents of the self-assessment document, validity of the six aspects of review, the review methodology and the overall judgment. Analysis of feedback data resulted in a small number of enhancements including a minor change in the naming of two aspects (Prevention and Control, and Achievements and Control), and most interestingly, a recommendation to create a new middle grade (3) in the overall judgment – apparently to address a partial implementation approach to H&S aspects, in that they have a systematic approach in some areas, but not in others.

7.

CONCLUSIONS The proposed H&S review methodology presented in this paper has been designed based on extensive literature review and comprehensive empirical research, and validated in a pilot implementation programme. Early results are very encouraging. The proposed review methodology is expected to be useful to a wide range of chemical companies in Saudi Arabia and elsewhere.

8.

REFERENCES [1] Al-Dhalaan K. A. (2003). Safety, health and environment (SHE) performance in the Saudi chemical industry. PhD Thesis. University of Bradford.

359


[2] Ahmad R. (2000). Developing a Proactive Safety Performance Measurement Tool (SPMT) for Construction Sites. [3] Gombault M. (1999). "Cleaner production in SMEs through a partnership with (local) authorities: successes from the Netherlands." Journal of Cleaner Production 7, 4,: pp. 249-261. [4] Health and Safety Executive (1998). "HSG65, Successful H&S Management." ISBN 7176: 1276. [5] Porter S. and Wettig J. (1999). "Policy issues on the control of major accident hazards and the new Seveso II directive." Journal of Hazardous Materials 65(1): 1-14. [6] Fletcher I. London. (1995). Health & safety in small firms. Association of British Chambers of Commerce. [7] Harms-Ringdahl, L., Jansson T., et al. (2000). "Safety, Health and Environment in Small Process Plants- Results from a European Survey." Journal of Safety Research 31(2): 71-80. [8] Quazi H., Khoo Y. et al. (2001). "Motivation for ISO 14000 certification: development of a predictive model." Omega 29(6): 525-542. [9] Tarrants W. (1980). The Measurement of Safety Performance, Garland STPM Press. [10] Murley T. (1990). "Development in Nuclear Safety." Journal of Nuclear Safety 31: No.1. [11] Lee T. (1994). "Employee Attitudes: The Quintessence of Safety Culture: Paper presented at the 3rd European seminar." Human Factors in Offshore Safety: Their Importance in Safety Case Implementation, Aberdeen. [12] Reason J. (1993). "The Identification of Latent Organizational Failures in Complex Systems." Verification and Validation of Complex Systems: Human Factors Issues, Berlin, Germany, SpringerVerlag: 223-237. [13] Karapetrovic S. (1998). "Integration of quality and environmental management systems Stanislav Karapetrovic and Willborn in The TQM Magazine 10(3): 204 213. [14] BS8800. (1996). "A guide to occupational H&S management systems." British Standards Institution. [15] International Standard Organisation. ISO-9001 [16] International Standard Organisation. ISO-14001 [17] Weinstein M. (1996). "Total quality approach to safety management." Professional Safety 41(7): 18-22. [18] Krause T. and Hidley J. (1989). "Behaviourally based safety management: parallels with the quality improvement process." Professional Safety 34(10): 20±25. [19] Spedding L., D. Jondes, et al. (1993). Eco-Management and Eco-Auditing: Environmental Issues in Business, Chancery Law Publ. [20] Shillito D. (1995). "Grand unification theory - Should Safety, Health, Environment and Quality be Managed Together or separately?" Process Safety and Environmental Protection 73(b): 194-202. [21] Svenson O. (2001). "Accident and Incident Analysis Based on the Accident Evolution and Barrier Function (AEB) Model." Cognition, Technology & Work 3(1): 42-52. [22] BS-OHSAS 18001: Occupational Health and Safety. [23] Hasle P. and Jensen P. (2006). "Changing the internal H&S organization through organizational learning and change management: Research Articles." Human Factors in Ergonomics & Manufacturing 16(3): 269-284. [24] Flin R., Burns C., et al. (2006). Measuring safety climate in health care, Quality and Safety in Health Care. [25] Cooper D. (1998). Improving Safety Culture: A Practical Guide, Wiley. [26] Petersen D. (1998). Techniques of Safety Management: A Systems Approach, American Society of Safety Engineers. [27] Diekemper R. and Spartz D. (1970). "A quantitative and qualitative measurement of industrial safety activities." ASSE Journal, Dec: 12-19. [28] International Safety Rating System. Publisher: Det Norske Veritas 1996. [29] Glendon A., Boyle A., et al. (1992). "Computerized H&S Audit Systems." Computer Applications in Ergonomics, Occupational Safety and Health. Elsevier, Amsterdam: 241-248. [30] Ballal S. G., Ahmed H. O. and Sebiany, A. M. (2002). "Occupational health in Saudi Arabia." Occupational Medicine: (Philadelphia, Pa.), 17, 3, pp. 491-507. [31] Alalfi F. (1989). Early planning to prevent disaster in industrial area. Riyadh, High institute for security science, Arabic centre for security and training studies. [32] Thiagarajan T. (1996). An empirical study of total quality management (TQM) in Malaysia: a proposed framework of generic application: the development of a generic framework to assist in the implementation of TQM in Malaysian-based organisations through empirical study of critical quality factors and best practice. Bradford,: 1 vol. [33] Weisberg H. F. (1992). Central tendency and variability. Newbury Park; London, Sage. [34] Levin R. (1994). Statistics for management, Prentice-Hall Englewood Cliffs, NJ. [35] HSE (2000). Development of an H&S Performance Measurement Tool. HSE Books

360


MANUFACTURING ENTERPRISE MANAGEMENT BASED ON THE SERVICE-ORIENTED TECHNOLOGIES Konstantinos Kotsopoulos1, Yim Fun Hu2, and Pouwan Lei3 1, 2, 3

University of Bradford, School of Engineering, Design and Technology Bradford, United Kingdom e-mail1: [email protected] e-mail2: [email protected] e-mail3: [email protected]

ABSTRACT Manufacturing plants contain hundreds of different applications, programmable equipment, sensor and measuring devices, with some running on legacy systems, others in standard hardware devices, or in embedded manufacturing equipment. These applications need to be interconnected and managed accordingly. The IT infrastructure landscape in these plants is a complex heterogeneous environment that is difficult to quickly adapt to ever changing business needs. Managing the heterogeneous nature of the IT infrastructure of a manufacturing plant is a difficult, complex and expensive process. Very often, interoperability and functional reuse among individual systems are difficult, if not impossible. Architectures following the Service-oriented paradigm allow the creation of composite business applications from independent, self-describing, and interchangeable software modules called services. Serviceoriented architectures use standardized technologies such as XML, WSDL, SOAP and JMS. Realizing the huge benefits of establishing a standard-based framework for applications and their integration through utilizing a common canonical schema and transaction set, this paper proposes a Service-based management framework that integrates heterogeneous management systems in a loose coupling manner. The key benefit of the proposed management architecture is the reduction of the complexity through service and data integration. Keywords: Management Framework, Service Oriented Architecture, Middleware, Web Services, SOA-based Manufacturing Plant. 1.

INTRODUCTION Manufacturing plant management has to deal with multiple vendors, multiple applications, multiple physical devices, multiple databases, and multiple service layers (infrastructure plane, control plane, service plane). Any management solution for a manufacturing plant must be architected in a way that it can scale to manage the current and future infrastructure. This scalability challenge is a requirement for flexibility so that the solution can be rapidly adapted to support new services and technologies in the future without the need for long term and complex upgrades. The management plane of a manufacturing plant should be flexible and scalable enough to accommodate heterogeneous legacy management systems as well as new generation management systems that span across different layers of the plant‘s infrastructure and need to operate as one agile entity. Moreover, the management architecture should be able to reduce the complexity of the involved management systems, increase the potential for reuse of management functionality and increase the speed of development and deployment of these systems. In addition, the level of automation of the management system needs to be high in order to provide greater capability and to manage higher levels of complexity in networks and systems. The management architecture needs to adopt mainstream information technologies and development techniques rather than maintaining a reliance on stovepipe specific technologies. Service Oriented Architecture (SOA) constitutes a very promising approach to integrate enterprise applications, including management systems. SOA can provide a high level of scalability and flexibility that is required in heterogeneous environments. Moreover, organizations require data and service sharing but due to the heterogeneity of the technology, data and services may reside in different platforms and each platform communicates according to its specific communication protocol. This gives rise to many challenges in interoperability and integration. Before the emergence of the SOA, middleware technologies such as Common Object Request Broker (CORBA) from the Object Management Group (OMG), Distributed Component Object Model (DCOM) and Remote Method Invocation (RMI) were commonly

361


adopted for data and services integration. The drawback of these technologies is that the interoperability among different system components residing on different platforms remains weak and difficult to achieve. By adopting the SOA philosophy, vital management operations can be applied as services (for example, retrieving the status of a device, controlling it, changing its configuration settings and provisioning). In the SOA paradigm, services are software components with formally defined, message-based, request-response interfaces and the logic behind those interfaces is hidden from the users. 2.

ADOPTING THE SOA FOR DEVELOPING MANAGEMENT SYSTEMS SOA has gained popularity due to the wide use of Web Services [1]. Web Service technology enables Service-Orientation that makes use of autonomous, self-described services which are loosely-coupled by using technologies such as Simple Object Access Protocol (SOAP) [2] as a communication protocol, Web Service Description Language (WSDL) [3] used for service description and Universal Description, Discovery and Integration (UDDI) [4] used as a service registry. Beyond the basic framework of Web Services (publishing, searching and invoking) SOA defines service composition - the next step in developing and extending Web Services. Through service composition, it is possible to compose new and complex services from other simple atomic services.

2.1

The Web Service Framework The most prevalent technology that enables Service-Oriented implementations is the Web Services technology [5]. Web Services define interfaces to perform a specific task or a collection of operations through standardized XML messages. Web Services use a standard, formal XML notion (its service description) which covers all the details needed to interact with the service, including transport protocols, message formats and location. Services can be independent from the software or hardware platform on which they are implemented and they are independent from the programming language in which they are written. Users need only to know the interfaces whereas the implementation details of the services are hidden from the users. Hiding the implementation details allows Web Services to be loosely coupled with cross-technology implementations and to be used independently or with other Web Services to complete a business transaction or a complex aggregation [6]. Web Services provide a way of communication among applications running on different operating systems, written in different programming languages and using different technologies whilst using the internet as their common transport [7].

2.2

The emergence of the Enterprise Service Bus Enterprise Service Bus (ESB) is commonly used to provide technological solutions to intercept messages between services. ESB incorporates the concept of mediation and allows interoperability between clients and data sources in Information Systems. An ESB is a middleware that provides service integration and composition by building services upon industrial standards such as XML, SOAP, WSDL, WS-Addressing, and WS-Security. Moreover, ESB provides a communication channel that is mostly asynchronous by applying Message-Oriented Middleware (MOM) and Publish/Subscribe methods. In addition, ESB provides transformation functions, dynamic routing and security (for example authorization, cryptography, etc.) and other QoS management functions such as quality measurement, tracing and data management.

2.3

Service Orchestration and Service Composition Two main models are used to perform Web Service composition: the orchestration model and the choreography model [5], [8]. The orchestration model is the main process that coordinates interactions between services. It provides a language for the formal specification of business processes and business interaction protocols, thus, providing more support to business transactions. The Business Process Execution Language (BPEL) [9] is commonly adopted to specify the Web Service behaviors through an XML grammar to describe the logic needed to coordinate services that participate in a flow. The choreography model is more collaborative than the orchestration model. In choreography, each part involved in the process describes its role in the interaction. Service choreography is associated with messages exchanged among many Web Services. Web Services Choreography Interface (WSCI) is a specification created by Sun Microsystems, BEA and SAP that defines an XML-based language for such scheme of Web Service collaboration [10].

362


These languages define an interoperable integration model that should facilitate the integration of intra/inter-organizational processes between businesses and organizations. 3.

MESSAGE EXCHANGE PATTERNS BASED ON THE MOM CONCEPT In the MOM concept, message applications employ a message client API to communicate with each other through a proposed middleware infrastructure called Core NMS Service Bus. The Core NMS Service Bus is based on the ESB concept that allows service composition and service integration. Furthermore, it uses the MOM communication paradigm for exchanging information among the involved management components. In the MOM communication paradigm, an application can act as a message producer that produces the message or a message consumer that consumes the message. An application may have dual functionalities of being a producer and a consumer at the same time. In the proposed management architecture the message producer is a management system (i.e. NMS, ERP, Device Management System, etc.) that sends management information to a destination. The destination is a message consumer that could be either another management system at the shop-floor of the manufacturing plant acting as a local consumer or an external management system residing at another manufacturing plant and acting as a global consumer. Figure 1 shows the relationship between the producers and the consumers. The shop-floor systems produce the management information and the Core NMS Service Bus publishes the information into virtual channels, the local and global users act as consumers, requesting management information from the Core NMS Service Bus.

I need Performance measurements from all Shop Floors

I need Shop Floor A Faults I need the configuration Profile of all devices at Shop Floor B

I need a fault overview from all Shop Floors

I need Performance from Shop Floor C

I need to process Faults from Shop Floor A Consumer C

Consumer D

Consumer B

Consumer E

Consumer F

Consumer A

Core NMS Service Bus

Management Systems at Plant’s Shop Floor A

Management Systems at Plant’s Shop Floor C Management Systems at Plant’s Shop Floor B

Figure 1: Communication Scenario between Core NMS Service Bus and consumers Management Information is stored in the Core NMS Service Bus into queues or topics, and is categorized into different groups according to the requirements of each management service. Consumers can subscribe to the group of their interest and receive all messages sent to the groups. This categorization can help filtering messages accordingly.

363


PROPOSED MANAGEMENT ARCHITECTURE FOR THE MANUFACTURING PLANT Figure 2 illustrates the management architecture model of the manufacturing plant based on the SOA layering. The Enterprise-Grade Management Services (EGMS) layer consists of global management services residing at the top-floor of the manufacturing plant in order to support generic management functions such as configuration of the shop-floor network, fault management, device management, performance management functions etc. Devices of the manufacturing and automation processes are represented in the shop-floor network. The smart networked devices and IT systems of the manufacturing plant are managed by several applications at the foundation IT and Management System Layer and exchange messages via standardized management interfaces such as SNMP [11]. The IT and Management Systems are traditionally strongly connected to the infrastructure layer operating locally within the shop-floor of the manufacturing plant and are being interconnected in order to exchange information with each other. Furthermore, these Systems expose their functionality as web services to allow easy integration and to expose their functionality to the higher layer of the EGMS services. Enterprise-grade Management Services Performance & Activity Monitoring Service

Trouble Ticketing Service

Failure Management Service

External Partner Data Sharing Service

Plant floor Operations

Alert/Event Management Service

Configuration Management Service

Inventory Management Service

Management Information Broadcast

Device/Service Discovery

Core NMS Service Bus Data Collection

Event Action Mgt.

Message Validation Service

Message Transform ation Service

Routing Service

Persistent Store

Message Archiving Service

(

(

Service Repository

(

Device/Service Availability

(

Messaging Service

Service Orchestration

(

Service Composition

(

4.

Request/Reply and Publish/Subscribe Technology

ERP

Network Management System

CRM

Controller RFID

Device Management System

PLC

File Server

Message Abstraction Layer

Communication Layer

Plant Floor Management Applications

Production Line Process

Database Database

PLC

Alarm/Failure Generator

Middleware layer

Network Router

Foundation IT and Management Systems

Application Server

Shop Floor (Devices, Network and Systems)

Figure 2: Integration Architecture The Core NMS Service Bus as depicted in figure 2 consists of two layers named Middleware Layer and Message Abstraction Layer, which are explained in the sections that follow.

364


4.1

Middleware Layer The Middleware Layer of the Core NMS Service Bus consists of several middleware services that are available to give support to the enterprise applications. An example of a middleware service is the Device/Service Availability service that contains a history of all the available devices and services at the production line of the shop-floor. A list of services has been defined in the Middleware layer and is listed below:       

4.2

Data Collection Service Event Action Management Service Management Information Broadcast Service Device/Service Discovery Service Composition and Orchestration Device/Service Availability Service Repository

Message Abstraction Layer The Message Abstraction Layer consists of the following services:      

Messaging Service Message Validation Service Message Transformation Service Message Routing Service Persistent Store Service Message Archiving Service

4.2.1

Messaging Service

The messaging service is responsible for the communication and data transfer from one management system to another. This service may use three message exchange patterns (MEPs): Synchronous Request/Reply messaging communication, asynchronous point-to-point messaging or Publish/Subscribe messaging. These MEPs are based on technologies such as Java Messaging Service (JMS) [12] and Web Service. The messaging operations that are performed in the messaging service are explained below. 4.2.2

Point-to-Point

The point-to-point messaging model allows message clients to send and receive messages asynchronously via virtual channels known as queues. Messages from the message producer are routed to the message consumer via a message queue. While there is no restriction on the number of message producers who can publish to a queue, a message in the queue can only be received by a single message consumer. This property enables load balancing to be supported in the system. In this model, messages are always delivered and are stored in the queue until a consumer is ready to retrieve them. 4.2.3

Publish/Subscribe

In the publish/subscribe model, management messages are published to a virtual channel called topic. Unlike the point-to-point model which only supports one-to-one message distribution, the pub/sub model supports one-to-many and many-to-many distribution mechanism, allowing a single producer (i.e. Shop Floor Management System) to broadcast a message to hundreds of consumers. There are two types of subscription within the publish/subscribe paradigm; the durable subscription and the non-durable subscription. The non-durable subscription allows temporary subscriptions to receive messages only when they are actively listening to the specific topic. Topics cannot hold messages except if the consumer uses durable subscription. In duration subscription, when a subscribing consumer is disconnected from the messaging server, the message server stores the message and holds the data until the consumer reconnects. Thus, durable subscription can survive the failure of the subscribing consumer.

365


4.2.4

Request/Reply

The request/reply model is used for synchronous communication and it is supported by the Web Service framework. In the Core NMS Service Bus, the Messaging Service uses the request/reply approach in order to allow synchronous communication between the Enterprise-grade Management Services and the shopfloor management Systems. The publish/subscribe and point-to-point message models are primarily for asynchronous communications between message producers and consumers. However, synchronous interactions between these two parties are sometimes required. A request/reply message pattern can be built on top of the two MOM message models to perform both asynchronous and synchronous request/reply. MOM message channels (topics and queues) are not bidirectional. To perform a request/reply operation, a requester must use two channels: one for the request and one for the response (reply). A correlation ID can be used to correlate the request message with the reply message. 4.2.5


Generally, the management information extracted from the shop-floor network could contain errors regarding the content of the information that they share or could share messages that cannot be understood by other management applications. A solution for this problem is to subjecting their information to reference validation. For this reason, a validation mechanism is proposed in order to eliminate the creation of unnecessary faults and errors in invalid messages which store information regarding specific managed nodes (i.e. network rooter, file server, etc.) that cannot be later processed by the Plant Floor Management Applications. 4.2.6

Message Validation Service Architecture

Management Services, i.e. the Failure Management Service, Trouble Ticketing Service etc., must be able to interpret the messages published by other management systems in the shop-floor and understand their meaning. This is not always possible, because a message that is based on XML could be invalid. For example, the message body may cause parsing errors or lexical errors, or there are missing information in the message header, or the properties values in the message itself are wrong. In other cases, when virtual channels are categorized into different groups for different management information type, if a message is put in the wrong category, the Message Validation Service should be able to detect such error. In Figure 3, the messaging service creates an incoming messaging channel (validation.in) and two outgoing message channels. The incoming message channel receives management messages transmitted by a Local Management System (LMS) at the shop-floor. As for the two outgoing channels, one named Validation.out messaging channel is responsible for connecting the message validation service with the EGMS and the other (Validation.error) with the error message handler.

Core NMS Service Bus LMS1

.. ..

x

x

Outgoing Messaging channel EGMS


x

Incoming Messaging channel

LMSn Schema

Invalid Messaging channel

Figure 3: Message validation Service

366

Error Message Handler


A message sent by a LMS will be validated before reaching its destined message consumer, the EGMS. The message may contain management information regarding a failure in a network node at the shop-floor. The message is passed to the virtual channel and is processed through the Message Validation Service, where it will be compared against a validated XML schema. If the messages satisfy the requirements of the XSD schema, then they can successfully proceed to the destination, which is the EGMS. If they fail, the Message Validation Service initiates an invalid fault alert and sends the invalid messages to the error message handler. 4.2.7

Message Transformation Service

Legacy systems only understand their own proprietary protocols and messages and rarely agree on a common data format. This makes system and data integration virtually impossible. One solution for integrating heterogeneous systems is to modify the systems through data transformation, where data of one system is transformed into the data format of the other. However, this is not the most efficient way to integrate systems due to the fact that it requires a lot of changes in the system‘s logic and data format changes are not economically feasible [13]. Furthermore, adjusting the data format of one system to match that of another system makes the overall architecture more tightly-coupled. Another approach is to use XML-based messages to enable service interoperability. Transformation is performed using a stylesheet language called XSLT (eXtensible Stylesheet Language Transformation) to restructure XML documents from one format to another and to transform and/or enhance the content of the XML message. The stylesheet specifies how the XML data will be displayed. XSLT uses the formatting instructions in the stylesheet to perform the transformation. These instructions inform the transformation processor of how to process a source document in order to produce a target document that is understood by all systems. MM1


LMS1


MM2

LMS2

Incoming Messaging channel

x

CM CM EGMS Outgoing Messaging channel

Metadata Repository

Figure 4: Message Transformation Service created in the Network Management Platform Figure 4 demonstrates the Message Transformation Service that transforms messages from one format into a common format. In this scenario, messages are sent to the EGMS by two different LMSs. The LMS1 could be a management system at the shop-floor that needs to provide management information such as PLC device failure notifications to an EGMS service (i.e. Failure Management Service) at the top-floor of the management plant. The LMS2 could be another management application at another department of the shop-floor that manages another set of equipment such as production line IT systems that it also needs to send failures to the same Failure Management Service at the top-floor of the manufacturing plant. The messages they transmit need to be transformed into a common information model to be understood by the EGMS. Messages from LMS1 (MM1) and messages from LMS2 (MM2), each having its own proprietary data formats, are passed to a common message incoming channel created by the Message Service in order to be delivered to and processed by the Message Transformation Service. The Messaging Service also creates an outgoing messaging channel (transformation.out) responsible for connecting the EGMS to the Message Transformation Service. The Message Transformation Service has a central repository for storing metadata defining the appropriate message format understood by the EGMS. The metadata can be stored in a number of formats. A common format for XML messages is defined in the XSLT. The Message Transformation Service makes an external

367


call to the metadata repository for a lookup (searching the data structure of the XSLT). The messages (MM1 and MM2) are compared against the XSLT schema and the content is being transformed according to the XSLT schema. Finally, the Messaging Transformation Service will place the transformed messages to the outgoing messaging channel (transformation.out) for delivery to the EGMS. In the case the Transformation Service component is required to transform information based on different information models, each XML namespace (xmlns) included in the messages, has to be mapped to a particular XSLT stylesheet. 4.2.8

Message Routing Service

Routing functions target messages that have been sent by the LMSs and need to be distributed to different EGMSs. By implementing the Routing Service at the Core NMS Service Bus, neither the EGMSs nor individual LMSs need to be concerned with routing functions (i.e. the destination of the message, message priority etc.). As a result, systems and services become more loosely-coupled and more reusable because they do not have to specify the number of consumers that will be attached to or how to prioritize the message exchange. The Routing Service is responsible for performing routing functions and this is achieved by applying routing rules based on the content of each message. Moreover, the Routing Service provides intelligent routing rules for routing messages to the appropriate destination. Motivated by the Enterprise Application Integration (EAI) Patterns that introduce solutions for integrating applications, the Routing Service implements three functions based on EAI [14]: 1. 2.

Content-based routing functions, Content-enrichment functions, 3. Content splitting functions.

Figure 5 demonstrates the Routing Service performing content-enrichment functions, splitting functions and content-based routing.

MS1

MM1

EGMS1

LMS1

.. . LMSn

Routing Service Messaging channels

MMn

Contentenriching Function

Splitter Function

MS2 Routing Function

Messaging channels

EGMS2

MSn


.. .

EGMSn

Figure 5: Routing Service performing routing functions in the Middleware Layer When a message is sent from a LMS to the Core NMS Service Bus via the incoming message channel, the Routing Service is activated. The content-enrichment function will inject additional information on each message indicating its origin. Each message is a large XML-based message, thus the splitting function is applied in order to split the message into smaller messages, where each message will contain one event of an individual network node. These event messages are processed by the routing function. Routing function routes the event messages based on the actual content of the message, rather than by the destination specified in the message header. The Routing Service parses the message and applies a set of rules to its content to determine the message‘s destination. As a result, the Routing Service provides a high degree of flexibility and adaptability to change.

368


4.2.9

Persistent Storage Service

One of the middleware services offered by the Core NMS Service Bus is the provision of a persistence store that is designed for message persistence. The persistent storage is a database that stores all incoming and outgoing messages to and from the Core NMS Service Bus and to examine messages by querying the database. It is used in order to recover the data in case of a Middleware failure or failures of the LMSs or the EGMSs in order to increase reliability. The use of persistence store makes the management architecture more reliable and fault tolerant. 4.2.10

Message Archiving Service

The messages that are passed through the Core NMS Service Bus are stored into folders for inventory purposes. The Message Archive Service accommodates management information in XML-based form messages. This service creates folders according to the message destination and stores all the messages that have been transmitted from different LMSs. This function allows external access from service providers to request information for inventory purposes. 5.

MESSAGE TRANSACTION AMONG THE INVOLVED ENTITIES Figure 6 illustrates a typical message transaction among a LMS, the Core NMS Service Bus and an EGMS. In this case the EGMS is a Trouble Ticketing Service that allocates failures identified in the shop-floor to particular system administrators.

Core NMS Service Bus LMS

Trouble Ticketing Service


Messaging Service


Routing Service

Create session with 143.53.36.62:61616/Validation.in Session accepted Subscribe to 143.53.36.62:61616/Topic1 Subscription ack Send Management message to Validation.in queue Read message from Validation.in queue Store to Validation.out queue

Read messages from Validation.out Store to Transformation.out queue Read messages from Transformation.out queue Store event message to Topic1 Store event message to Topic2 Store event message to Topic3 Store event message to Topic4

Send event message to 143.53.36.44 ack

Figure 6: Interactions between remote services and the Core NMS Service Bus In figure 6, the LMS creates a session with the Core NMS Service Bus. The Core NMS Service Bus sends an acknowledgement back to the LMS. The Messaging Service contains all the queues and topics as explained in the previous section. The LMS sends its management information in the form of management messages to the validation.in queue. As messages are stored into the queue, the Message Validation Service is initialized and reads the messages from the queue. It processes each management message and sends the

369


validated management message back to the Messaging Service to store it in the validation.out queue, if it complies with the XSD schema, or else it sends an error message and stores it in the validation.error queue. The Message Transformation Service reads the messages from the validation.out queue and transforms them. After the transformation process the messages are sent back to the Messaging Service and they are stored into the transformation.out queue. Next, the Routing Service reads the messages from the transformation.out queue and processes them. Each management message completes a series of processes as explained in the previous section by the Routing Service and the output of each management message is stored to the different topics. The final step is the transmission of the event messages to the subscribed Trouble Ticketing Service. Each message that is sent to the subscriber is acknowledged. 6.

CONCLUSION Architectures using the Service-Orientation principles could deliver agility, scalability, reusability, flexibility and automation in distributed heterogeneous environments. SOA uses open standard principles and facilitates loose coupling and ―plugability‖ of new interfaces. In order the manufacturing plants to move from stovepipe systems towards the enterprise-grade reusable systems that can quickly adapt to the fast changes of the business needs, the shop-floor IT and Management Systems have to be easily integrated in a Service-Oriented way with modern top-floor enterprise services. This paper has presented the design of a management integration architecture based on messaging and asynchronous communication that removes the complexity from the management systems and integrates the low level management information residing at the shop-floor with enterprise-grade services residing at the office-floor of the manufacturing plant.

7. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

REFERENCES T. Erl, Service-Oriented Architecture: A Field Guide to Integrating XML and Web Services, New York: Prentice Hall, 2004. W3 Consortium, ―SOAP Version 1.2 Part1: Messaging Framework,‖ W3Consortium Recommendation, April 2007. W3 Consortium, ―Web Services Description Language (WSDL) 1.1,‖ W3 Consortium Recommendation, March 2001. OASIS, ―UDDI Version 3.0.2,‖ July 2008 T. Erl, C. Utschig, B. Maier, H. Normann, B. Trops, T. Winterberg and P. Cheliah, Next Generation SOA: A Real-World Guide to Modern Service-Oriented Computing, New York: Prentice-Hall, 2010. H. Kreger, ―Web Services Conceptual Architecture (WSCA 1.0),‖ IBM Software Group, May 2001. K. Gottschalk, S. Graham, H. Kreger and J. Snell, ―Introduction to Web Services Architecture,‖ IBM Systems Journal, Vol. 41, Iss. 2, 2002. A. Peltz, ―Web Services Orchestration and Choreography,‖ Computer, Vol. 36, Iss. 10, pp 46-52, Oct 2003. T. Andrews, F. Curbera, H. Dholakia, Y. Goland, J. Klein, F. Leymann, K.Liu, D. Roller, D. Smith, S. Thatte, I. Trickovic, S. Weerawarana, ―Business Process Execution Language for Web Services version 1.1,‖ May 2003. W3C, ―Web service choreography interface 1.0,‖ August 2002. www.w3.org/TR/wsci/. D. Mauro and K. J. Schmidt, Essential SNMP, O‘Reilly Media, 2001. Sun Microsystems, ―Java Message Service,‖ version 1.1, April 2002. P. Carey, New Perspectives on XML, Course Technology, 2nd edition, 2002. G. Hohpe and B. Woolf, Enterprise Integration Patterns, Pearson education, 2004.

370


INNOVATIVE METHODOLOGY FOR DESIGNING A MODULAR HIGH VOLUME FLOW LINE R. Yumbla1, S. Lumley2, and M. K. Khan1 1

School of Engineering, Design and Technology/ University of Bradford Bradford, UK 1 e-mail1: [email protected] 1 e-mail1: [email protected] 2 Flexitallic, Bradford, UK Bradford, UK 2 e-mail2:[email protected]

ABSTRACT This paper proposes an innovative factory planning methodology to achieve the objectives defined by Flexitallic for the future expansion of the Thermiculite 866 production line. The concepts under investigation extend to the analysis of flow benefits and restrictions considering product features and demand addressed in a proposed Batch / Flow Comparative Matrix. Furthermore, this study describes a Conceptual Factory Planning Methodology to incorporate new tendencies for project planning, process manufacturing design and layout design. The definition of the future value stream is based on the analysis of the process, the definition of the manufacturing capacity and future line expansion strategy. The Thermiculite 866 case study exemplified the utilization of the proposed methodologies and demonstrates its importance during the design of a high volume production line. The paper concludes that the implementation of a line-flow equipment-paced will support the future production and quality requirements for Thermiculite 866. Keywords: Plant Design, Lean Philosophy, Flow Production Line, Value Stream Mapping. 1.

MARKET AND PROCESS DEFINITION Manufacturing has been radically changed over the past decade because the relatively ‗static‘ nature of market now has been replaced by high changeable market requirements. The new demand is hardly satisfied by the mass production, so new terms and requirements to manufacturing are fundamental. In fact, Kidd emphasizes in Chute (1) that the need for every organization is to be able to switch frequently from one market-driven objective to another. Crowson (2) argues that most businesses get flexible primarily to reduce costs and thereby improve their competitive position in the market. However, the real objective of this investment is to make money by high volumes with reduced unit costs. The SOFC industry, because of its current low production quantities and low-rate production, has not been strongly influenced by high volume production concepts and automation requirements. Although, automation can be justified for this industry due to precision and accuracy are required.

1.1

Business and Productivity Strategy Considerations The productivity on shop floor can be identified by six interrelated factors: Planned Production Times, Physical Working Conditions, Economic Working Conditions, Degree of Centralization for Decisions, Acceptance of responsibility, Attitude Towards Time (3). In fact, all these condition are key drivers to warranty a production line linked to organizational priorities that according to The Department for Business Innovation depends on seven pillars of activities for success. These are: Macroeconomic Stability, Investment, Science and Innovation, Best Practice, Skills and Education, Infrastructure and Policies to ensure the right market framework. In addition, it has been stressed that infrastructure and policies should ensure the right market framework (4). In other words, every intention to implement new technology in any manufacturing process must fit the real market requirement. This concept can be extended to other manufacturing strategies such as flexibility, reliability and capacity. Flexibility, reliability and capacity are the most sought-after properties in modern manufacturing systems, but they are poorly understood in theory and poorly utilized in practice. One reason for this is the lack of general agreement on how those terms should be applied to Manufacturing. Even though, Boyle (5) argues

371


they are not only an operational aspect but also an attribute of decision making, an economic indicator, and a strategic tool. It must be considered that increasing the potential flexibility to a point much greater than the required flexibility may result in an over-investment in manufacturing equipment (5). For this reason, the proposed factory planning methodology define the level of flexibility, reliability and capacity during the Conceptual Design and Detailed Design stages described in the Proposed Conceptual Factory Planning Methodology. 2.

FLOW PRODUCTION AND LEAN MANUFACTURING SOLUTIONS After the World War I, Henry Ford and General Motors moved the world into mass production and flow process. The next Manufacturing milestone was after World War II when Toyota pioneered the concept of Lean production (6). Today, lots of companies have implemented their own ―Lean System‖ and flow concepts that are based on the first Toyota Production System (TPS). Womack (6) mentioned that the success of Lean philosophy is based in the process stability which permits to combine to produce 100% quality products when they are needed to satisfy customer demand. Standardized and leveled production work which helps to stabilize the process, and Just in Time (JIT) managed by a pull system will reduce the inventory. In addition, one of the most aggressive mechanisms to implement continuous improvement is through production stop policy implementation to visualize the location of the problem.

2.1

Continuous Flow: the solution for high production volumes In 1913, Ford implemented the mass production concept to assembly a vehicle that minimized the time that elapsed between beginning and completing production. The solution for Ford was to standardize huge volumes of products in a continuous flow and low in house inventory facility. The second important development in the manufacturing field was the link of this concept to Lean philosophy. Minoru M. president of Toyota Motor Manufacturing implies it when he mentioned: ―If some problem occurs in one-piece flow manufacturing then the whole production line stops. In this sense it is a very bad system of manufacturing. But when production stops everyone is forced to solve the problem immediately. So team members have to think, and though thinking, team members grow and become better team members and people‖. In addition, not only all people will be involved in the problem solution but the continuous process flow brings problems to the surface (7). In addition, Stewart (8) claims that companies generally try to settle into more stable ways of working such as more fixed working patterns. For example, static product lines like the current Thermiculite 835 line has been a reference for product development at Flexitallic. Stewart (8) suggests that the best companies make easy to figure out the organization‘s structure as well as its process/flow of work through the production systems which gives inertia to the operation. This argument suggests that these organizations are always clear, visible to employees for quicker track of difficulties. Finally, visual business processes allow everyone in the organization at all levels to understand specific roles in the company and in what form their contribution has helped to the company‘s revenue.

Miltenburg (9), claims that “a line flow system is appropriate when the product design is stable and the volume is high enough to make efficient use of a dedicated line.” To conclude, flow is a concept developed along to the process standardization in order to reduce total lead time and promote mass production success. After flow concepts were implemented in most of the industries, the system needed to gain competitiveness through Lean practices. The Lean philosophy and TQM provides waste elimination methods and continuous improvement techniques to the system. For this reason, Lean flow system provides total customer satisfaction and employee engagement. 2.2

Lean Manufacturing and Production Lines The intention of change in an organization is the most motivating challenge and also a scaring issue for most of the people. Change disrupts organizations and it is not an overnight process. The first step to change in a company should be promoted from top executives and transmitted to shop floor through approaches designed by Human Resources. Change cause non-conformance in the first phase which is the least comfortable for the people and a lack of stability for the operation (8).

372


Today, a number of companies have implemented their own ―Lean System‖ and flow concepts that are based on the first Toyota Production System. Although, many of these companies has not been able to demonstrate the financial benefits at the overall enterprise level (10). Probably one of the reasons is that Lean has been considered as a process whereas as a philosophy Baker (11) and O‘Corrbui and Corboy (12). Unfortunately, Lean interpretation usually becomes so misapplied and the popular mentality is to do everything with less: half of the factory space, half of human effort, half of investment. Actually, most of the activities were applied in discrete activities that transformed the industry ―one step away from anorexic‖ making them ―fragile‖ and inflexible instead of Lean (13). On the other hand, Alavi (14) argues that there is no roadmap for achieving a kaizen or Lean culture, and suggest to left each organization to their own devices and methods. 3.

INTRODUCTION TO THE THERMICULITE 866 PROJECT Flexitallic Ltd designed Thermiculite 866 as a compression seal for Solid Oxide Fuel Cells (SOFC), and other range of applications. Thermiculite 866 is based upon the use of extremely thin, flexible plates of the natural mineral vermiculite. Vermiculite is the short mineroligical name for hydrated laminar magnesiumaluminum-ironsilicate which is similar to mica in appearance. This material is known for high temperature capability, high chemical resistance and as an electrical insulator. In addition, Thermiculite 866 has a second material which is steatite, another silicate with a plate-like crystal structure, which is perhaps better known as talc or soapstone. The combination of steatite, a very soft mineral, with the Chemically Exfoliated Vermiculite (CEV) results in a soft sheet material that compresses under modest loads and this means that on assembly it conforms easily to the surfaces thus forming a seal (15). The aim of the Thermiculite 866 project is to design and commission the implementation of a high volume production line or batch process to provide gaskets for the SOFC industry that may generate £7m per annum income to produce a good financial return. The Thermiculite 866 production process requires accurate controls to warranty the product‘s quality. For this reason, a Strategic Alignment of Quality Function Deployment (SAQFD) was developed to define the major requirements in the future production line (16). This Neo-QFD approach delivered strategic alignment in addition to engineering and customer requirements. The required high volume production line must support a production quick increase for 2015, and warranty supply of Thermiculite 866 sheets for probable UK SOFCs market that is projected to be 1,600,000 per year (17) . This study proposes a methodology for designing a modular high volume line supported by the strategic alignment defined during the Thermiculite 866 SAQFD.

4.

BATCH / FLOW COMPARATIVE MATRIX In section 2.1, several strong reasons were exposed to promote implementation of flow line, although, further analysis needs to be conducted to design a suitable production solution for Thermiculite 866. SOFC technology has been developing for the last eighty years and its high technical requirement and expensive materials have not permitted to start high volume production regimens. For this reason, the SOFC industry has had reduce contact with high volume production line challenges. Flexitallic will face several challenges such as required operators‘ skills and high level of operations management, etc. At Flexitallic, the Thermiculite 866 demand has been forecasted to increase three hundred times compared to the current demand during the next three years. Although, the next couple of years the demand is expected to slightly increase. This means that the Thermiculite 866 implementation needs to be highly reactive to considerable demand acceleration by 2015 to 2020. For this reason, the new Thermiculite 866 production line requires a flexible concept to justify budget expenditure according the forecasted demand. The definition of a particular production line can be based on the evaluation of product/volume – layout/flow matrix proposed by Miltenburg (9) that provides a general description of different layouts and types of operations. This paper suggests a novel comparative approach between batch and flow to justify a strategy and future expenditure in the implementation of any high volume production line. The layout can be easily defined by using the link to the material flow and process flow. In fact, the material flow depends on the layout, but a particular layout can vary according to the operation. Consequently, this study proposes a structured evaluation matrix to define the product and operation according to different production parameters such as product, process and material supply.

373


In Figure 1, the first column displays three concepts required to achieve Flow, and the difference with Job Shop and Batch operations. On the top of the matrix are displayed different levels regarding Lean Manufacturing Controls that promote reduction of variation in the process. The sloped arrow represents the additional controls required to achieve higher Lean Manufacturing levels to support Flow process. Finally, the Thermiculite 866 project is represented by the orange box and some specific improvements of the current experimental production line such as layout redesign and Strategic Aligned Quality Function Deployment (SAQFD). The matrix proposed to evaluate three production features: product, process and production control in order to define the required action to pass the barrier between batch and flow. The Batch/Flow Comparative Matrix (Figure 1) displays an analysis of the current Thermiculite 866 process based on process, product and layout concepts previously explained. This analysis intends to summarize the product characteristics and required operational improvements to implement and achieve the reliability of a high volume manufacturing production line. The red dots represent the current possible limitation for the flow implementation at the Thermiculite 866 production line. Two of those requirements: pull control and control boards will be achieved with the implementation of the Thermiculite 866 project. Although, there is a product volume demand that is a fundamental requirement to get the advantages of the flow concept.

Figure 1: Batch / Flow Comparative Matrix On the other hand, the product complexity and process flexibility provide great conditions for implementation of flow concept. Finally, there are two required improvements (yellow balls) in order to transpose from batch to flow. This novel analysis intent to envisage the best option to be considered during the concept design of the production line.

374


4.1

The Thermiculite 866 Production Line After the conceptual definition the future Thermiculite 866 investment process is defined based in the requirements exposed in Figure 1. In addition, the following bullet points are important observations to define the guidelines in the definition, detailed design and implementation stages: 



5.

The DEMAND analysis (X thousand parts per year in 2015 in the UK) has defined that the process requires and improved drying process (manufacturing bottleneck) and automatization of the packaging process THE PRODUCT and PROCESS are favourable to promote flow concepts and achieve the required high volume production levels. The Thermiculite 866 manufacturing can be summarized as the accurate spreading of wet material and posterior quick dry of this material. By adding more intermediate processes, only more waste would be obtained

INNOVATIVE DESIGN METHODOLOGY AND PROJECT PLANNING FOR THE HIGH VOLUME PRODUCTION LINE DESIGN In fact there are two main reasons to use a specific manufacturing system. First, consider the mix and volume required to accomplish the demand. Second, recognize the output required in terms of cost, quality, performance, delivery, flexibility and innovativeness. This paper intends to combine analysis of demand, facility design methods and project management techniques to develop the conceptual design of a high volume production line. Heaton (18) suggest that the design must consider the following strategies:    



Take into account customer requirements Take into account competitors Take into account manufacturing capabilities Consider all options available to manufacturing List of the required outputs that manufacturing will provide

A more specific approach is suggested by Lee in Tompkins (19), who defines the following steps as part of the information, strategy and layout definition:   

Strategy: Develop operational strategy and business architecture, define space planning units Layout: Analyse material flow, calculate space, identify constrains Product Analysis and Current Process: Display the current state value stream and definition of product family

In addition, Halen (20) suggest a plant layout analysis that includes a study of the production line flow charts, material flow diagrams, product routings, processing times, relationship diagrams between different departments in the facility and the cost of material movement. The reader must understand that the development of a new production line or layout must not include only a process design but also the management during implementation. For this reason, The Proposed Conceptual Factory Planning Methodology needs a first stage of planning and a second stage of commercial awareness proposed by Heaton (18). In addition, it must include material flow and layout analysis as suggested by Tompkins (19) and Halen (20). In addition, the characteristics and relationships defined in the SAQFD highlighted the necessity to incorporate a parallel Lean & Quality development process (16). Figure 2 displays the proposed model for high volume production line design and the required support from process development and equipment trials (grey flow line). Each section of the project (blue boxes in Figure 2) is defined by subtasks grouped according to different phases described by bullet points in page 6. The implementation of this structured process warranties the success of the project and the best Quality Assurance (QA) system and Lean implementation. In fact, the partition between conceptual design and detailed design represents the required Lean evaluation before continue with the future implementation.

375


Figure 2: Proposed Conceptual Factory Planning Methodology In the following section are listed the different outcomes from the Thermiculite 866 project. The author has included several specific outcomes and the date those were achieved or when it is expected to be finished. All the design and selection of equipment have gone through the process proposed in Figure 2. For example, the drying and packaging systems are currently in the detailed design stage and it is expected to be fully designed by the end of 2011. In addition, the mixer was already designed and implemented in Dec/2010. The reader must consider the following manufacturing processes as reference: Weight, Mix, Cast, Dry, Peel, Cut, Rolling and Packaging. Plan Project: The project plan structure follows a logical implementation sequence which is based on the following organizational and managerial chapters: Team Organization, Product Trials and Planning, Training, Production Line Design, Plant Preparation, In-Plant Build Activities and Continuous Improvement. The Thermiculite 866 Project Master Plan has achieved 18% (50 tasks) of the Gantt Chart tasks. The 300 tasks defined with the project Team aim to prepare Flexitallic for the Start of Production event in year 2015. Analyse Product and Volumes: The Batch / Flow Comparative Matrix (Figure 1) provides the first approach to define the impact of the product features in the process. In addition, the Analysis of Volumes was performed during Demand & Manufacturing Strategies reviews that produced figures of manning, investment and line capacities. It was concluded that the batch process would require three times the manning required by the proposed line-flow equipment-paced. Analyse Current Process: Due to the relative simple manufacturing process: Weight, Mix, Cast, Dry, Peel, Cut, Rolling and Packaging. It was not required to utilize all the extend of the Value Stream Mapping technique to recognize production bottle necks and improvement requirements. By 2015, the projected utilization of the process capabilities will be: 8% of Mix and Cast, 88% of Dry, 2% of Peel, Cut and Packaging. This means that all efforts should be directed to the drying process to achieve high volume capacities. In addition, the criticality of QA requires more attention for the control of variables such as

376


mixing and drying times. For this reason, the team developed a mixing cycle control based on the ―mixing energy‖ to register mix parameter and Weigh per Unit Area (WPUA) checks. Conceptual Design: A SAQFD was developed to support the implementation of the entire Thermiculite 866 production line, although, its outcomes are focused in the Conceptual Design. In addition, the definition of the future target cycle time and value stream has been defined by the projected demand in the next five years. As explained above, 88% of the drying capacity would be used by 2015. The design of the modular drying solution and the ―four conveyor‖ casting strategy will cope with the expected 200% acceleration slope. The four conveyor casting strategy intends to start extracting moisture by spreading the wet material just after mixing. Furthermore, considering the relative low drying temperature of 90 ᴼC, a drying time of less than 150 minutes can be achieved for Thermiculite 866 dough of 29.0 % solids. In Thermiculite 866 the aligned exfoliated vermiculite acts as an inorganic binder to bind all ingredients. As a result any microscopic paths that exist perpendicularly through the formed sheet have a very low degree of permeability and thus it is difficult for water vapour molecules to travel along them. This means that drying time cannot be reduced by increasing temperature because it causes blistering on the material surface. The ―four conveyor‖ strategy intends to increase the drying capacity by increasing the drying area with a 90 ᴼC continuous drying solution. On the other hand, it has been defined the Plant Distribution and Layout that defined the material flow and equipment location to use the 600 square meter area for future increase of the drying capacity. Finally, it has been defined the strategy and general budget expenditure plan for the implementation of the Thermiculite 866 Equipment-Paced Line Flow during the next four years. Detailed Design: Four gates have been defined before final approval of the equipment specifications and CAD drawings. These gates are: Concept Development and Trials, System Level Design and Detail Design. The intention is to engage the Technical Group, Operations and Compliance in the equipment selection and to promote communication, cooperation and ―risk sharing‖. The success of the equipment design Current detailed design of drying solution will consider predetermined water content on conveyor band to extract water and reach 6% moisture content. The conceptual design and trials suggest modular design to facilitate future capacity expansion and the application of drying zone concept. In fact, the future Thermiculite 866 flow line will reach high volumes by using four parallel conveyors aligned to a pair of 90 ᴼC continuous zoned ovens. Finally, the end line will cut and package cut rectangles of shaped gaskets. Implementation: The implementation has been highly supported by trials and continuous communication with the equipment suppliers. In addition, Flexitallic defined payment terms and internal approbation gates to trial the technical specifications included in the supplier‘s proposal. As result, the vacuum mixer was designed, trailed and commissioned at Flexitallic. Continuous Improvement: As explained in the Batch / Flow Comparative Matrix (Figure 1), Flexitallic needs to fill a gap to achieve the required production control to manage with the future high volume requirements. For this reason, the Flexitallic team requires to design, develop and implement the Thermiculite Business Plan according to the Flexitallic vision/mission and values: Quality, Production, H&S and PDCA Deming Cycle. Experimentation and Trials: Most of the trials are based on testing one variable at a time, although, other multiple variation Taguchi methods may be explored in the future. There are number of trials that supported the detailed design and experimentation in 2010. For example, it was tested different concepts such as vacuum oven, zoned and continuous oven concept, and different types of mixers. 6.

CONCLUSION AND RECOMMENDATIONS This paper has defined an innovative methodology to design a high volume production line based on literature review and conceptualization of the flow line advantages. The Thermiculite 866 has already successfully implemented the conceptual design stage after the definition of the business requirements that started with the SAQFD developed by the authors. The Batch / Flow Comparative Matrix (Figure 1) has successfully defined key parameters to analyze and define the requirement of product, process and supply to adopt flow concept to an specific production line. In addition, the matrix proposes clear Lean Manufacturing requirements to reach the required production control and obtain the real benefits of continuous flow. The Proposed Conceptual Factory Planning Methodology (Figure 2) is the simplification of three different methodologies that consider the company strategy, material flow and layout design. The aim of this

377


methodology is to promote cooperation and communication within different departments at the company and support every stage with experimentation. The Thermiculite 866 case study provided a view of the application of the Batch / Flow Comparative Matrix and the methodology to define the gap between Batch and Flow. Finally, the Proposed Conceptual Factory Planning Methodology is the core of the Thermiculite 866 line design and the outcomes has been seen in the implementation of the project during 2010. The success of the proposed Methodology has been demonstrated with the current implementation of the Thermiculite 866 line. The structured project management approach proposed in this study needs to be supported by training and communication strategies within the organization. The development of the Thermiculite 866 project required of continuous communication and training of people on shopfloor. In addition, it is important to mention that a robust organizational structure is fundamental to tackle every challenge with team synergy. 7.

REFERENCES [1] Crute, V., Ward, Y., Brown, S., & Implementing Lean in aerospace—challenging the assumptions and understanding the challenges. Crute, V., Ward, Y., Brown, S., & Graves, A. 2003, Technovation , pp. 917–928. [2] Crowson, R. Assembly Processes. New York : Taylor & Francis Group, 2006. [3] International shopfloor level. Bruijn, E J and Steenhuis, H J. 2006, Journal of Manufacturing Technology Management, pp. 42-55. [4] Department for Business Innovation and Skills. The Government‘s Manufacturing Strategy. UK : Department for Business Innovation and Skills, 2006. [5] Towards best management practices for implementing manufacturing flexibility . Boyle, T. A. 2006, Boyle, T. A. (Vol. 17 No. 1, 2006). Towards best manaJournal of Manufacturing Technology Management, pp. 6-21. [6] Womack, J, Tones, D and Roos, D. The Machine that Changed the World. New York : Macmillan Publishing Company., 1990. [7] Liker, J. The Toyota Way 14 Management principles from the worlds gratest manufacturer. New York : McGraw-Hill., 2004. [8] What great companies do well. Stewart, J. 2004, IEE Manufacturing Engineer, pp. 14-15. [9] Miltenburg, J. How to formulate and implement a winning plan. New York : Productivity Press., 2005. [10] Ward, Yvonne, et al. Cost management and accounting methods to support lean aerospace enterprises. Bath, UK : University of Bath, 2003. [11] Why is lean so far off? Backer, P. s.l. : Works Management, 2002, Vol. October. [12] The seven deadly sins of strategy. O'Corrbui, D and Corboy, M. s.l. : Management Accounting, 1999, Vol. No 10. [13] Amos, J. Transformation to agility. New York & London : Garland Publishing Inc., 1998. [14] Leaning the right way. Alavi, S. Jun-Jul, s.l. : Manufacturing Engineer, 2003. [15] Flexitallic. Thermiculite 866 ---- A Service Proven , High Temperature, Compression Gasket for SOFC Applications. UK : Flexitallic, 2010. [16] The Strategic Alignment Of Quality Funtion Deployment (SAQFD) as a Key Driver For The Design of a High Volume Production Line. Yumbla, R, Lumley, S and Khan, M. 2011. [17] Delta Enargy & Enviroment. Exploring the Market Opportunity for SOFC. Edinburgh : European Fuel Cell Forum 2010, 2010. [18] Heaton, P. Product definition module - Setting the Scene -. Derby : Rolls-Royce EEPDS, 2003. [19] Tompkins, J. Facilities Planning. s.l. : John Wiley & Sons, Inc, 2003. [20] Hales, H. Computer aided facilities planning. New York : M. Dekker, 1984.

378


THE STRATEGIC ALIGNMENT OF QUALITY FUNCTION DEPLOYMENT (SAQFD) AS A KEY DRIVER FOR THE DESIGN OF A HIGH VOLUME PRODUCTION LINE R. Yumbla1, S. Lumley2, and M. K. Khan1 1

School of Engineering, Design and Technology/ University of Bradford Bradford, UK 1 e-mail1: [email protected] 1 e-mail1: [email protected] 2 Flexitallic, Bradford, UK Bradford, UK 2 e-mail2:[email protected]

ABSTRACT This paper covers the introduction of a novel Quality Function Deployment (QFD) to support the manufacturing line design using a mechanism of incorporation commercial awareness in any stage of the product deployment. The original QFD ensures process planning by bringing parts deployment into parts characteristics through the House of Quality. This study renews the original QFD by developing the Strategic Alignment of Quality Function Deployment (SAQFD) to achieve proactive management of Houses III and House IV. The design of SAQFD is based on Neo-QFD that has been proposed during the last ten years by different authors. Finally, the SAQFD applicability is demonstrated during the implementation of the Thermiculite 866 high volume production line. Keywords: Quality Function Deployment, Quality Assurance System, Quality Management. 1.

INTRODUCTION Quality Function Deployment (QFD) has continued to spread since Mizuno and Akao published their first book on the topic in 1978. This paper intends to identify the best utilization of the QFD as an important section of Total Quality Management (TQM). The modified QFD approach integrates product and process development (IPPD) as an extension of concurrent engineering in the design of a high volume production lines. The main QFD method was not modified, and it has been proposed the use of a systematic way to conduct integrated product and process development. The proposed QFD approach is driven by business priorities and adapts the differences in applications resulted from changes in product development. This approach is presented in the Thermiculite 866 case study, although, this paper does not include any technical specification that would compromise any intellectual properties of the Thermiculite 866 or manufacturing process at Flexitallic Ltd. The term ―Quality Function Deployment‖ (QFD) refers to the concept and methodology of New Product Development (NPD) under the umbrella of TQM. The QFD is a methodology for transforming the customer‘s requirements into product characteristics and further more into process and production characteristics. QFD uses four ―houses‖ to integrate the information and requirements of marketing, engineering, R&D and manufacturing. According to the first traditional scheme published by Mizuno and Akao in their first book in 1978; there are four houses: House of Quality, Parts Deployment, Process Planning and Process and Quality Control (1). Despite the popularity of this concept, Shiu (2)claims that there have been several misperceptions about the ―QFD essence‖. The common misunderstandings include the utilization of the Quality Deployment equivalent to ―quality chart;‖ and QFD being equivalent to quality deployment (2). Pinto and Kharbanda (3) identified other major causes of NPD failures such as ignoring the project environment, stakeholders‘ requirements and project objectives. The broad concept of quality and its philosophy provides different possibilities to alter the basic QFD concept and redefine a Neo-QFD to achieve an optimal product development. The structural change of the ―four houses‖ has produced different approaches and new variables into the practice of QFD for product development. For example, the Matrix of Matrices Model deals with quality, technology, reliability and cost

379


considerations in addition to the popular QFD Four Matrices. This modified QFD approach aims to extract any bottleneck-technology, to prevent potential failures and to achieve target cost. The development of QFD has demonstrated that modern product development requires more accurate market analysis and business integration, although, this continuous sophistication of the QFD has brought other issues in the information management. 2.

FURTHER DEVELOPMENT IN QUALITY FUNCTION DEPLOYMENT Several problems can be encountered during the implementation of sophisticated QFD. For example, erroneous conclusions can be magnified by serial processing. Therefore errors introduced at one stage will propagate to the entire analysis. In addition, Han (4) argues that the QFD complexity may be a timeconsuming process requiring a lot of detail. In fact, 30 customer requirements and 50 design requirements lead to 1,500 different links to be discussed. This is a real issue for QFD practitioners considering that a typical application can have 30 to 200 requirements (5). The QFD literature has demonstrated that there is significant complexity in the use of rating scales to prioritize the QFD final outcome (6). Furthermore most QFD researches have focused on the scoring mechanics (2). Ten percent of the QFD practitioners use the four-matrix model; another 10% use the Matrix of Matrices approach exclusively. Finally, the remaining 80% use an integrated approach combining the best features of both models. This clear message defines the importance of ―customized‖ QFD in order to achieve practicality in the industry. For example, Marsh (7) integrated a model based on Deming‘s Plan, Do, Check and Act (PDCA) cycle in order to link QFD with Lean Manufacturing philosophy. In fact, this seems to be the start of more customized approaches. For example, ten years after the first QFD publication, some practitioners intended to link QFD to technology, cost and reliability in NPD cycle (2). Furthermore, Zultner, a student of Akao, designed a streamlined approach called ―Blitz QFD‖ that intended to select and deploy only the top most influential customer needs (8).

3.

ALTERNATIVE QUALITY FUNCTION DEPLOYMENT Sullivan (9), and Clausing (10) were important figures in the initial development of QFD in the industrialized Western countries. Afterwards, QFD was gradually introduced to researchers and practitioners in different of manufacturing fields. Then, QFD was combined with various design methodologies and numerical analysis methods that promoted more research afterward. Jiang (11) defined three main aspects that have been explored:  

 3.1

QFD combined with TRIZ, Taguchi methods in order to improve its effectiveness QFD as part of product design and process design QFD combined with numerical methods in order to strength the analysis accuracy

Commercial awareness as a mechanism of knowledge incorporation in QFD The level of TQM and ISO9000 development defines the major quality methodologies in organizations. Historically most of the organizations in Europe were more exposed to ISO9000. Although, there are several important reasons to promote TQM in this companies. Planning and QFD intends to improve new product development and reduce delays in projects. This paper proposes a renewed approach to use the original QFD and integrated organization objectives during product development. The Flexitallic Thermiculite 866 project is an illustrative example of the methodology proposed to deploy the quality system using a ―forced QFD‖. As mentioned in the introduction of this study, ―Neo-QFD‖ provides several reasons to promote flexibility and customization to the original QFD. In fact, there are several aspects to be covered in QFD. For example, cost and product life cycle are important inputs to be incorporated in the analysis. Several authors have assertively implemented more sophisticated QFD scoring mechanisms. Although, these approaches constrain the QFD due to the excessive time required for its implementation. On the other hand, the proposed QFD must promote communication to improve product development. Successful companies optimize product life-cycle designs by employing well organized design reviews and utilizing the culturally inherent communication between designers and engineers responsible for production and maintenance.

380


The aim to envisages the process prior QFD has been a common point of view among QFD practitioners. Han (4) proposes a hierarchical framework to improve the effectiveness of decision making. This Hierarchical framework suggest six stages before starting the house of quality: voice of the customer, competitive analysis, and voice of the organization, design targets, relationship matrix and correlation matrix. In fact, the initial forced priorities often depend on product positioning in the market and life-cycle. For example, QFD initial priorities should be affected by the role of product features, cost, and time to market. Chao (12) proposes a matrix that begins by identifying constrained factor (e.g. hard limit on timeto-market, hard budget/cost target, a new level of features/functions). Second, the priority to be optimized needs to be defined (e.g. quicker time-to-market, minimize cost, maximize features). Furthermore, Sarbacker (13) identified the importance of risk assessment in three main aspects: Envisioning risk, Design risk. And Execution risk (12). The quality function deployment (QFD) is one of the most important part of the total quality control concept. It focuses on customer needs determination and on organization-wide commitment to satisfy these needs in the long term. Lu (14) suggests that the application of QFD in the strategic planning process could be applied for corporate departments such as marketing, finance, accounting, research and development, etc. (14).This view incorporates corporate strategy into the process to decide about products, processes and operations, and suggest continuous reviews of customer strategy. In addition, Killen (15) suggest that QFDbased methods should start with customer and stakeholder outcomes. Finally, the inherent culture of the company is an important driver for the application of QFD. For example, in companies that emphasize new product development, the ―forced product features‖ involve the setting of product specifications. Early stages of the QFD begin with demanded-quality deployment, and determine critical quality characteristics and design quality. On the other hand, in companies that emphasize manufacturing, the QFD activities begin with acceptance of product specifications. Demanded-quality deployment and quality planning before the setting of product specifications. 4.

THE STRATEGIC ALIGNMENT OF QUALITY FUNCTION DEPLOYMENT (SAQFD) As explained in the previous sections the original QFD supports modification. This study proposes a NeoQFD defined by forcing some of the product features and process requirements prior to starting the first House of Quality. The application of ―inherent‖ commercial strategies leads to define the optimal process specifications for specific components. This paper intends to prioritize stakeholders‘ requirements and commercial awareness as a key factor to define QFD requirements. Company situation defines restrictions in the development of NPD. For example, economic stability, investment in the NPD project, staff skills and education are key factors to define the scope of the level Quality Assurance (QA) implementation. In addition, infrastructure and inherent policies have an important impact significally in the approach to the NPD. The SAQFD propose two fundamental conjunctions to link marketing plan and corporate strategy to QFD. This key grouping are Workforce and Strategic Alignment displayed in Figure 1.

381


MISION & VISION

Marketing Plan & Corporate Strategy

House of Quality

Parts Deployment

I

II

Workforce Alignment

Process Planning

III

Production Planning

IV

1

Strategic Alignment

Cost: DOC Responsiveness: DFM, DFA, DFSS Quality: FMEA

Figure 1: The Strategic Alignment of Quality Function Deployment (SAQFD) SAQFD can be represented by two main streams of influence and information: Workforce Alignment and Strategic Alignment. Both of these sources of forced requirements will define a Neo-QFD shaped by organizational mission & vision through corporate strategies. In Figure 1, the QFD structure is represented by the four Houses of Quality, and the introduction of specific requirement is represented by green and blue arrows. For example, the number one in the circumference represents an early forced requirement in the process for one specific teams involved in the QFD. The specific requirement may be the use of Design for Manufacturing (DFM) to modify product features instead of filling the gap, for example, of poor skilled manpower or technological restrictions. This demonstrates how a specific solution is forced to product redesign instead of influencing the process redefinition to accomplish customer requirements. This downstream feedback might be considered an atypical modification of the QFD, although this is fully compatible with new Concurrent Engineering practices in product development. In fact, SAQFD provides a structured development process for companies that need to include their limitations during NPD. The major difference with the original QFD developed by Akao is the importance that SAQFD gives to business awareness and the relative flexibility required during the HoQ. This means that Operations needs to interact closely with Designers to define some parameters that would be reflected in the product. SAQFD requires specialized support analysis in three major fields: cost, responsiveness and quality to support any forced requirement in house I, II, III or IV. The brownish base in the bottom of the graph represents the set of product development techniques to facilitate the introduction of any feedback or loop to HoQ. The Workforce Alignment or Organizational Alignment, which is part of a greater Human Resources study, is partially covered in this paper. Each department in the organization need to understand their specific role and responsibility in each specific stages of the project. The communication between department and the ability to identify some of the requirement in the different Houses will define the success of the NPD. Ideally, the SAQFD should start by involving people from different levels in implementing and thinking about company practices. Provide feedback to management and engineers on cost-effectiveness and project specific aims. It is important to use factual information about what is happening based on an unbiased objectives with informed staff. After defining teams and their roles, try to answer what they are trying to accomplish and define some specific roles. Are the company vision and mission aligned? Major requirements should sound like solutions but avoiding the high-level project goals and challenges, like ―make cheaply‖, ―produce better quality‖, and ―meet new demand‖. The flow of information and involvement of the team is a key factor for defining successfully the Strategic Alignment.

382


The Strategic Alignment adapts the original QFD to the business perspective requirements in order to effectively implement process planning and production planning. A well informed team promotes ideas from different hierarchical levels which produces valid forced relationships. The business and company philosophy defines the major project drivers. For example, a company that prioritize manufacturing upon other departments might promote production capacity and capability as part of the Strategic Alignment. The Workforce would be aligned under a Manufacturing leader that would manage the project to fulfill operations requirements. A successful Strategic Alignment produces a detailed project plan with tasks within an specific timescale. The SAQFD will define the structure of this plan based on the four QFD Houses: Engineering Characteristics (House of Quality), Parts Characteristics, Key Operations Process and Production Requirements. This plan structure is based on Fundamental Principle of System Engineering which suggests to start with a conceptual design and continue with detailed design. The mechanism to define each section will be based on the Plan, Do, Check, Act Deming Cycle, and supported by Facility Planning Project and Design. In this stage, the SAQFD would produce a detailed project plan and all of its advantages. For example, it would be possible to identifying a ―critical path‖ and the necessary backup plans can be addressed to accomplish a success NPD. To conclude, the SAQFD provide an improved utilization of the original QFD approach by developing a strategic plan that consider the project environment such as company vision. 4.1

Introduction to the Solid Oxide Fuel Cells and Definition of Basic Customer Needs Solid Oxide Fuel Cells (SOFCs) are highly efficient energy conversion devices that produce electricity by the electrochemical reaction between fuel and an oxidant. Fuel-cell devices are composed of an anode electrode (exposed to the fuel), an electrolyte and a cathode electrode (exposed to the oxidant). These components are stacked to produce enough electricity and heat to be recovered. This sealing solution for this technology has several key factors that need to be accomplished: The materials selected as sealants must be thermo-mechanically and thermo-chemically stable in both oxidizing and wet-reducing environments at 800 ◦C for long-term exposures (500–1000 h) (16). Smeacetto claims that five main approaches are being studied for sealing SOFCs (17) & (18) : brazing, compressive seals, glass, glass–ceramic and glass-composite seals. Flexitallic designed Thermiculite 866 as a compression seal for Solid Oxide Fuel Cells (SOFC), and other range of applications. Furthermore, the SOFC market needs an effective solution with reduced prices considering that other technologies, such as gas boilers, have defined a low reference price.

5.

SAQFD CASE STUDY – THE THERMICULITE 866 PROJECT Thermiculite 866 is based upon the use of extremely thin, flexible plates of the natural mineral vermiculite. Vermiculite is the short mineroligical name for hydrated laminar magnesium-aluminum-ironsilicate which is similar to mica in appearance. This material is known for high temperature capability, high chemical resistance and as an electrical insulator. In addition, Thermiculite 866 has a second material which is steatite, another silicate with a plate-like crystal structure, which is perhaps better known as talc or soapstone. The combination of steatite, a very soft mineral, with the Chemically Exfoliated Vermiculite (CEV) results in a soft sheet material that compresses under modest loads and this means that on assembly of a connection it conforms easily to the surfaces thus forming a seal (19) Thermiculite Critical Service materials are rated for temperatures up to 1000°C and have passed the API 607 fire test. (20) SAQFD for Thermiculite 866 starts with the Organizational Alignment or Workforce Alignment as mentioned in the previous section four. In order to understand their impact and involvement in the project, the stakeholders and team project members were grouped into three different hierarchical groups according to the power to force a variable or ownership of the project: Priority 1 - Project Owners - They have high interest in the commercial and economical success of the project. Their main goal is to accomplish the levels of production, quality and flexibility required for the future Thermiculite 866 demand. In addition, they are looking for revenue and investment return periods. Priority 2 - Project Developers - They are the experts in the product, materials and manufacturing process. In addition, in this group, I included the academic supervisors who transfer knowledge from the University of Bradford. Priority 3 - Project Coordinators and Executors - They are the people that perform the required tasks to support the design, commissioning and implementation of the Thermiculite 866 production line. They are

383


considered ―partial-stakeholders‖ because their role change from internal customers to providers along the project. The first outcome from the project owner was a general Marketing Plan & Corporate Strategy for the Thermiculite 866. The plan for the Thermiculite 866 production line implementation is defined by the SOFC technology development during following years. It has been predicted that the Thermiculite 866 demand would increase 300 times for one of the SOFC major customers during the next four years. This is a major concern that defined two forced inputs for the process and production requirements in the SAQFD: a.

Expected high volume production line needs to be considered in the QFD Process Planning

b. Production volume ramp-up in 2015 year must be considered in the QFD Process Planning & Production Planning Due to the novel nature of the SOFC market, it is important to get to the market and support customers with their technical challenges by aligning strategies according to their demand requirements. The original target was to design and to implement a high volume production line to introduce the Thermiculite 866 range by 2015, but the SOFC development and experimental phases were delayed ten months. For this reason, the Production Line implementation would be accomplished by implementing modular equipments concept during the following years. This is and example of Workforce and Strategic Alignment, and first important milestone of the SAQFD, in the design of the Thermiculite 866 production line. Figure 2 displays in a green arrow the organizational prioritization of R&D and their definition of the process as a key driver to develop the product. On the other hand, Manufacturing and Engineering act as project developers and support R&D objectives. In addition, the impact of the expected sudden demand ramp up is represented by the forced requirement of modular high volume capacity line. The SAQFD suggest to force those requirements during the first House of Quality (HoQ) and identify them through the entire QFD. The mechanism to introduce these requirements is by using a team solution technique to tackle the proposed forced variables. For example, workshops and brainstorming sessions, before the HoQ session, would communicate and fix the strategies in the team‘s perception. After that, the HoQ can be deployed with the team and some of the previous conclusions can be included in the Design Requirements row. The HoQ for Thermiculite 866 defined, after brainstorming with the Technical Team, the following main conclusions: a.

b.

Operations mechanisms and technology requirements to reach high volume production of the Thermiculite 866 by 2015 Quality Assurance needs to be linked to the Manufacturing Process

The major outcomes from the brainstorming were used as forced design and customer requirements. For example, the drying time was defined as process bottle neck and its forced design requirement to maximize throughput by probable increase of the Thermiculite 866 solid contents in the wet material. This is a good example of how SAQFD promotes loops from the House III to HoQ to change product specifications when required. This early feedback is valuable to gain time and other resources during novel products development. On the other hand, the reader must understand that the Thermiculite 866 project intends to use some of the experimental equipment to construct the future high volume production line. For this reason, most of the Engineering Characteristics were defined from the third house (Process Planning), and the proposed Strategic Alignment is applicable to the Thermiculite 866 project. The manufacturing of the Thermiculite 866 has four main processes: mix, spread, dry and rolling. As stated in the introduction of this paper, no technical data will be presented, and the main relationships that where defined in the HoQ are:    

Density of the Thermiculite 866 dough as a key driver of other product features such as Weight per Unit Area (WPUA) and settings of the spreading head settings The mixing settings would have a moderate impact on the WPUA and probably a weak relationship with the specks presence The cost of the product and production capacity may be strongly affected by Drying capacity and it has a close relationship with the supply of raw material from alternative resources The importance of the material formulation was defined as priority 5 above WPUA and thickness

384


The HoQ produced sixteen design requirements and eight customer requirements (product technical requirements) and twenty eight relationships. Considering the outcomes of the House of Quality (HoQ) and Strategic Alignment, the diagram of the Strategic Alignment of Quality Function Deployment (SAQFD):

Figure 2: The Strategic Alignment of Quality Function Deployment for the Thermiculite 866 The SAQFD is especially useful for new product development, as the Thermiculite 866 project, that was not originally conceived from the HoQ. As explained in the introduction of this paper different authors designed more specialized approaches to support the first HoQ. Four methodologies to improve the Thermiculite 866 quality, cost and responsiveness are explained bellow.

FMEA: However, the original QFD did not emphasize the implementation of the process design FMEA, it is necessary to add this analytic mechanism to the proposed SAQFD. The Flexitallic team developed a FMEA to envisage any problem in the Thermiculite 866 manufacturing process. The team envisaged potential minor variable during spreading and potential significant variation during ingredients weighing.

Tolerances Design Methods: Weight Per Unit Area (WPUA) is a key parameter that need to be monitored and analyzed with statistical methodologies. The analysis of standard deviations would envisage important variables relationships in the processes and provide valuable information for Quality Assurance (QA) and automatization. 5.1

Case Study Analysis and Outcomes Automatization implemented at any cost: Considering the circled number one in Figure 2, which is a forced requirement of automatization. It has been defined that automation need to be considered for production capacity improvement and QA. The automatization of the cutting and packaging system increases the capacity of the line. In addition, the control of critical parameters such as raw material

385


weighing, mixing cycle parameters, spreading head settings, WPUA and thickness checks are important for QA. The automated system requires a centralized information system to control QA in the line. Quality Assurance defined from Process Planning: This forced requirement is represented by the circled number two in Figure 2. The impact of this requirement on the Thermiculite 866 high volume production line is that the QA is defined according to the process limitations rather that the product features. For this reason, the Thermiculite 866 QA is based on process control rather than product checks. In fact, the process control affects the product specifications through a simple lineal relationship. The density/mass/volume equation defines the required control system loop for controlling the dough spreading process according to QA requirements. The Themiculite 866 SAQFD in detail: The SAQFD process required about four months from the team definition to define most of the process planning (House III). The Thermiculite 866 project team provided the expected organizational alignment and engagement with the project. Although, its definition and official presentation was delayed because it was the first time the QFD methodology was performed at Flexitallic. Regular meeting reviews permitted to define major requirements for the project and requirements for the SAQFD. The Strategic Alignment analysis provide a better understand of the project requirements with stakeholders and it produces the Thermiculite 866 Implementation Plan with four hundred specific task to be delivered in the next two year period. In addition, the brainstorming prior the formal HoQ delivered enough documentation to define technical requirements, and raised forced design requirements such as the change of product characteristics by increasing the solid contents to improve the line throughput. Finally, it was difficult to fully perform the Process Planning (House II) due to the unexpected outcomes from the mixing process after installing a new mixer. In fact, this issue conducted our effort to performing statistical analysis and apply tolerance design methods. In conclusion, the SAQFD reduced time in process analysis and provided the flexibility required to design the future Thermiculite 866 high volume production line. 6.

CONCLUSION AND RECOMENDATIONS Strategic Alignment of Quality Function Deployment (SAQFD) is an alternative approach that does not change the QFD structure, and force variables in the HoQ according to stakeholders requirements, Process Planning and Production Planning. First, a strong and well defined team is required for the project success. Second, the team problem resolution techniques provide a good forum to discuss, document, and share process definition before the HoQ. SAQFD simplify the QFD methodology and integrate narrowly the different requirements such as technology, reliability, and implementation cost prior HoQ. The proposed Neo-QFD saves time and provides a flexible approach, compared to the original QFD, that aligns the organization and define major features in any stage of the product life cycle and utilize resources currently engaged in the process. For example, the Thermiculite 866 is already introduced in the market and it has been produced in a laboratory scale, so SAQFD provides the best approach to scale up the process and design the future high volume production line. The Thermiculite 866 study case demonstrated that SAQFD supports product development from experimental processes of laboratory scale products. In fact, the case study demonstrates the application of the SAQFD concept by considering the current process limitations in the laboratory scale production line and stakeholders‘ needs. The future development of the Thermiculite 866 production line will require implementing QA in the supply chain. In fact, one of the major customers requires vendor quality assurance system (VQA) to ensure that unqualified product, in terms of feasibility and materials are not chosen or approved at Flexitallic. In conclusion, SAQFD provides a structured and flexible approach to design or extend a high volume production line to introduce new product in any stage of its product life cycle.

386


7.

REFERENCES

[1] Akao, Y. Integrating Customer Requirements into Product Design. Cambridge : Productivity Press, 1990. [2] Reconstruct QFD for integrated product and process development management. Shiu, M, Jiang, J and Tu, M. 2007, The TQM Magazine, pp. 403-418. [3] How to fail at project management (without really trying). Pinto, J and Kharbanda, O. Pinto, J.K. and O.P. Kharbanda, (1996), "How to fail at project management (without really trying)," Business Horizons, 39 5th Edition, s.l. : Pinto, J.K. and O.P. Kharbanda, (1996), "How to fail at project management (without really trying)," Business Horizons,IEEE Computer Society Press., 1998. [4] A conceptual QFD planning model. Han Bruce, S, et al. 2001, International Journal of Quality & Reliability Management, pp. 796-812. [5] The house of quality . Hauser, J R and Clausing, D. 1988, Harvard Business Review, pp. 63-73. [6] Rating scales and prioritization in QFD. Franceschini, F and Rupil, A. 1999, International Journal of Quality & Reliability Management, pp. 85-97. [7] Facilitating and training in quality function deployment. Marsh, S. s.l. : Methuen, 1991, Vol. GOAL/QPC. [8] ReVelle, Jack B, Moran, John M and Cox, A Charles. The QFD Handbook. New York : John Wiley & Sons, 1998. [9] Quality function deployment. Sullivan, L. 6, 1986, Vol. 19. [10] Taguchi methods to improve the development process. Clausing, D. Vol. 2. [11] Quality function deployment (QFD) technology designed for contract manufacturing. Jiang, Jui-Chin, Shiu, Ming-Li and Tu, Mao-Hsiung. 2007, The TQM Magazine, pp. 291-307. [12] Project quality function deployment. Chao, Lawrence and Ishii, Kosuke. 2004, International Journal of Quality & Reliability Management, pp. 938-958. [13] Sarbacker, S. The value feasibility evaluation method: improving product, innovation through the management of risk arising from ambiguity and uncertainty. s.l. : PHD dissertation, Stanford University, Stanford, CA. [14] Strategic marketing planning:a quality function deployment approach. Lu, Min Hua and Kuei, ChuHua. 1995, International Journal of Quality & Reliability Management, pp. 85-96. [15] Strategic planning using QFD. Killen, Catherine P, Walker, Mike and Hunt, Robert. 2005, International Journal of Quality & Reliability Management, pp. 17-29. [16] Characterization and performance of glass–ceramic sealant to join metallic interconnects to YSZ and anode-supported-electrolyte in planar SOFCs. Smeacetto, F, et al. 2008, Journal of the European Ceramic Society, pp. 2521–2527. [17] Sealants for solid oxide fuel cells. Fergus, J W. 2005, Power Sources, pp. 46–57. [18] The state-of-the-art in sealing technology for solid oxide fuel cells. Weil, K S. 2006, JOM, pp. 36–44. [19] Flexitallic. Thermiculite 866 ---- A Service Proven , High Temperature, Compression Gasket for SOFC Applications. UK : Flexitallic, 2010. [20] —. Thermiculite®Thermiculite®866 Sealing Material Solid Oxide Fuel Cells. www.flexitallicsofc.com. [Online] [Cited: 15 April 2011.] http://www.flexitallicsofc.com/files/Flexitallic_SOFC_thermiculite_866.pdf.

387


LOGISTICS AND MATERIALS HANDLING SYSTEMS AND DEVICES

388


SIMULATION MODEL OF MARITIME INVENTORY ROUTING PROBLEM WITH PARTICULAR APPLICATION TO CEMENT DISTRIBUTION E. Wirdianto1,2, H. S. Qi1, M. K. Khan1 1

School of Engineering, Design and Technology, University of Bradford, Bradford, UK e-mail:[email protected] e-mail:[email protected] e-mail:[email protected] 2 Department of Industrial Engineering, Andalas University, Padang, Indonesia

ABSTRACT Simulation is undoubtedly a very useful tool for modelling a system specifically in the presence of stochastic elements and complex interactions between the system entities. In this paper, a simulation model to support decision making in ship scheduling for Maritime Inventory Routing Problem (MIRP) with particular application to cement distribution is presented. The system under study is a combined discrete and continuous system, where a heterogeneous fleet of ships with various sizes and types of contracts transport bulk cement products from production facility (Central Supply, CS) of a cement company to its packing plants (Distribution Centres, DCs). The simulation model in this study has been designed and developed thoroughly to emulate the complexity of the real system of the MIRP. The simulation model has demonstrated the capability to provide support for decision making in ship scheduling of the heterogeneous shipping fleet in the following forms: (a) real time states of inventory levels at CS and DCs and (b) ships‘ routing. In addition, one of the main strength of this simulation model is its flexibility. It can be easily expanded or adjusted to different size of system entities for example number of CSs, DCs, berths, vessels, and products. Keywords: Simulation, Routing, Maritime Transportation, Inventory Management, Bulk Cement 8.

INTRODUCTION Cement industry is not only disadvantaged by its plant location constraint, but also by the characteristics of its product. Cement is bulky and heavy product. Transportation costs of this product contribute a high percentage in its cost of goods sold. As cement is also a make to stock product, the immensity of its volume will create high inventory costs. Contrary with these two facts, price of cement is very cheap. Profit margin from this industry is very low. For these reasons distribution plays a very important role in this high volume business.

Mostly, a cement company uses Packing Plants (PPs) for the extension of the market. The input of these PPs, that is bulk cement, is supplied by production facilities as CS. The alternatives of transportation modes used are: truck, train, and vessel. The choice depends on the location of the PPs, support of transportation systems, and economies of scale. However, vessel is the most preferable transportation mode as the volume of shipments is concerned; especially in the condition of poorly developed land transportation system or the PPs are located in different islands. Distribution problem in a cement industry can be considered as an Inventory Routing Problem (IRP). The products are shipped to its own DCs or manufacturing facilities, on a recurrent basis. Shipments are usually not initiated by orders from the receiving locations, but rather the shipper (central supply) must schedule the shipments to assure that the receiving locations do not run out of stock. According to Ronen [1], this type of problems is known as the Inventory Routing Problems (IRPs). Campbell and Savelsbergh [2] also add that IRP is a variation of the Vehicle-Routing Problems (VRPs) that arises in situations where a vendor has the ability to make decisions about the timing and sizing of deliveries, as well as the routing, with the restriction that customers are not allowed to run out of product. Since the system under study uses vessel as

389


transportation mode, this distribution problem is more precisely called as Maritime Inventory Routing Problem (MIRP).

This paper is organized as follows: Next section introduces an overview of MIRP and simulation. The third section explains the design and development of the MIRP simulation model which includes description of the system, conceptual model, and computerized (Arena® simulation) model. Section four shows initial output of the model. Conclusions and next phase for the study are presented in the last section. 9.

OVERVIEW OF MIRP AND SIMULATION This section presents a brief review about the main topics considered in this study: Maritime Inventory Routing Problems (MIRPs) and simulation in MIRPs.

2.1

Maritime Inventory Routing Problems Proper scheduling of ocean transportation presents great potential of improving company‘s profit and economic performance of shipping. Vessels operating costs may easily amount to thousands of dollars daily or tens of thousands of dollars for the larger vessels and consume fuel while under way, at a similar rate [35]. A significant amount of attention has been directed towards IRPs where trucks are used to deliver the products [1, 6]. Even though approximately 90% of the volume and 70% of the value of all goods transported worldwide are carried by sea, until recently, relatively little work has been done on shipment planning in maritime inventory routing [7]. Christiansen et al. [8] also state that the number of research literatures within maritime routing and scheduling problems has been left far behind to those of within land and air transport. Their survey presents almost 60 references on ship routing and scheduling subject published during that last decade, whereas Ronen‘s [4] survey includes only about 30 references for the earlier decade. There are several works that have been carried out in the area of MIRPs. Shih [9] uses a Mixed Integer Programming (MIP) for planning of fuel coal import problem including multiple suppliers and multiple power plants for the Taiwan Power Company to minimize the total inventory costs of the fuel coal. As an extension of Shih‘s [9] work, Liu and Sherali [10] study the optimal shipping and blending decisions of coal fuel from each overseas contract to each power plant using a mixed-integer zero-one programming model to minimize the total costs of purchasing, shipping and inland delivery. Vukadinović et al. [11] develop a decision support system that can decrease the work load for the dispatcher and improve the quality of decisions for vessel dispatching problem. They propose a neural network approach as solution methodology. Ronen [1] also uses MIP for determination of the slate of shipments in maritime inventory routing. Some other works in this area also can be found in Ronen [4] and Christiansen et al. [8]. They also call these MIRPs as tactical and operational problems in industrial shipping, where the cargo owner or shipper also controls the vessels. From their surveys, in total, there are less than twenty literatures in this subject for the time span of two decades.

2.2

Simulation in MIRPs Simulation is one of the most widely used operation research and management sciences techniques [12-13]. Fu [14] also notes that one of the most successful interfaces between operations research and computer science probably has been the development of Discrete-Event Simulation (DES) software. For several decades, simulation has been used as a descriptive tool by the operations research community in the modelling and analysis of a wide variety of complex real systems [15]. A MIRP or marine shipping involves many stochastic variables and complex interactions between the system entities in its practice. The presence of stochastic elements and complex interactions between the system entities often preclude the possibility of obtaining an analytical solution [16]. Ronen [1] models shipment planning in maritime inventory routing using MIP and finds that the smaller size problems are solved to optimality. However, he finds that the modelled MIP problems are very hard to solve and the optimality of the solution for the larger ones is not verified. Vukadinović et al. [11] in their research on vessel dispatching problems find certain disadvantages incorporated in the mathematical programming

390


approach. They assert that mathematical programming models become invalid in the situation where unexpected delays existed. Paolucci et al. [17] state that defining an optimization model based on a mathematical formulation of the problem to support a decision-making process is neither simple nor suitable. Instead of using mathematical formulation, they use a simulation-based decision support system to the allocation problem of crude oil supply to port and refinery tanks. Cheng and Duran [18] study logistics for world-wide crude oil transportation using discrete-event simulation and optimal control. In general, their model is typified by one supply point with multiple discharging ports, one discharging port per vessel voyage, multiple vessels with similar capacity, one type of product, identical daily costs for each type of tankers, no port entry constraints on vessels, and one-day discrete time slice. This set of system characteristics may work well for a worldwide crude oil logistics system; however some of the system‘s characteristics tend to make the model being not valid for other MIRPs. In Ronen‘s [4] review paper, there is only one published work that uses simulation as the technique to solve the problem. This work is a design of an interactive decision support system to simulate ship voyage alternatives for only one vessel [19]. While in Christiansen et al. [8], there is no simulation model found to be utilised as the method for solving the problems in commercial vessels routing and scheduling in industrial shipping. Nevertheless, there are a small number of works in the area of strategic ship routing and scheduling system design that use simulation approaches with the major decisions on design of transport system and fleet size, and also there is one work on ship routing and scheduling in supply chain problem with major decision on logistic system design. 10. SIMULATION MODEL OF MIRP 3.1

System Description The MIRP system under study can be classified as combined discrete and continuous system. Mainly, the system‘s events occur at isolated points in time or in other word according to discrete jumps. The sequence of detailed activities of vessel in MIRP system can be modelled as discrete model. However, the MIRP system also contains inventory system where the state of the system might change continuously over time. In this situation, the inventory part of the MIRP system should be modelled as continuous event model. Fortunately, Arena® can handle both situations. This MIRP system consists of one CS (also named as PP1) and six DCs (DC A to F and also named as PP2 to PP7). The CS has two berths for loading the cargo into the vessels. DC A to DC E only has one berth each for unloading the cargo, while DC F has two berths. All DCs have land equipment for unloading process except DC B; hence only vessels with self unloading equipment can be assigned to DC B. Following the classification of ship routing and scheduling problems and models from Ronen [20], the problem entities of the MIRP system under study can be described as in Table 1. Table 1: Problem Entities of MIRP System under Study

a. b. c. d. e. f. g. h. i. j. k. l. m. n.

Problem Entity Mode of operation Loading and discharge times Number of origins Number of discharging ports Number of loading ports per vessel voyage Number of discharging ports per vessel voyage Number of commodities Fleet size Types of vessel Demands (shipment sizes) Cruising speed as a decision variable Fleet size and composition Port entry constraints on vessels Sea route constraints on vessels 391

Preference Industrial Open One Multiple One One Multiple Multiple vessel Multiple Stochastic (full shipload) No Constant over scheduling period Exist Exist


o. p. q. r. 3.2

Ports precedence requirements Costs Cargo transhipment Time between events

None Fixed costs and variable costs Excluded Stochastic

Conceptual Model of the MIRP System The sequence of activities for a vessel in the MIRP system can be represented as in the Figure 1. Every arriving vessel at CS (Packing Plant 1, PP1) Pilot Station EOSV will be assigned its DC destination and cargo (i.e. its cement type and amount). If all servers (berths) are busy, the vessel will wait at anchorage position with First In First Out (FIFO) queuing discipline, otherwise it will seize one of the available berths. After the vessel successfully berthing, it will then get Initial Draft Survey (IDS), loading the cargo, and End Draft Survey (EDS). The vessel then releases the berth (unberthing) and routes to CS Pilot Station BOSV. From this station the vessel is sailing to its assigned DC destination. In real system the vessel will send a notice (e.g. 1 hour) before it arrives at DC Pilot Station EOSV. This also applies when vessel arrives at CS. From DC Pilot Station EOSV the vessel sets the route to the available berth. If all berths are busy, the vessel will wait at anchorage position according to FIFO queue. As in CS, the vessel at DC will follow the IDS, unloading the cargo, and EDS processes sequence. Having completed the processes, the vessel then releases the berth and routes to DC Pilot Station BOSV. From this station the vessel is sailing back to CS in empty condition. This sequence of activities will be repeated as a cycle.

Berth

Unberthing

Depart from Pilot Station (BOSV)

Sailing

Arrive at Pilot Station (EOSV)

Waiting for Berthing

EDS

Berth IDS

Loading Process

Unloading Process

EOSV: End of Sailing Voyage BOSV: Begin of Sailing Voyage

Port B - DC

Port A - CS

EDS

IDS

Berth

Waiting for Berthing

Arrive at Pilot Station (EOSV)

Sailing

Depart from Pilot Station (BOSV)

Unberthing

Berth

Figure 1: Vessel Round Voyage (VRV) in MIRP Entities are the dynamic objects in the simulation. While entities usually are created, move around for a while, and then are disposed as they leave, the vessels entities in this MIRP simulation model are created at time 0, move around during the simulation, and never leave the system. It is possible to have entities that never leave but just keep circulating in the system [21]. Vessels entities and other components of simulation model for this MIRP system are explained in Table 2. Table 2: Components of MIRP Simulation Model

Component

Name

Detail

Description

Entities

Vessel

Quantity Arrival pattern

Attributes Variables

Vessel Index

Number of vessel All vessels are created at time 0 Labelling the vessel: Vessel 01, Vessel 02, ... Various variables are used to keep information those reflect some characteristics of the MIRP system. These variables such as current stock and maximum capacity of each cement type stored at CS and DC

392


Resources

Berths

Quantity Capacity Where required

2 at CS, 1 each at DC A to E, and 2 at DC F Only 1 vessel can berth at one time The berth is required for IDS, loading or unloading, and EDS activities Shifts 24 hours a day and 7 days a week Queues Berths Queues Capacity No queue is allowed at particular berth (0) Anchorage Capacity Instead of queuing at berth, vessel will queue at Positions anchorage position at particular CS or DC with infinite capacity Dwell time Until at least one berth available Rule FIFO From the detailed sequence of activities of a vessel in the MIRP system (see Figure 1), the stations are defined, for example: station number 21 (i.e. PP1-Pilot Station EOSV), station number 22 (i.e. PP1Anchorage Position), and station number 87 (i.e. PP7-Pilot Station BSOV). These stations will be used in the design and development of the MIRP simulation model. The network representation of these stations and their relationship to activities duration in VRV can be seen in Figure 2. As examples, IDSPTaCS means IDS Process Time at CS and STfDCtCS means Sailing Time from DC to CS.

PP1 Berth 2 IDS Position

PP1 Berth 2 – Loading Process Position

PP1 Berth 2 PP1 Berth 2 EDS Position Unberthing Position LPTaCS EDSPTaCS ELT UITdLP = LPTaCS - ELT

IDSPTaCS BERTH 2 PP1 Berth 1 IDS Position BPTaCS

PP1 Berth 1 – Loading Process Position

PP1 Berth 1 PP1 Berth 1 EDS Position Unberthing Position LPTaCS EDSPTaCS ELT UITdLP = LPTaCS - ELT

IDSPTaCS BERTH 1

Or

UPTaCS

CS (PP1)

WTfBAaCSAP

PP1 – Pilot Station BOSV

To DC A/B/C/D/E/F

PP1 – Anchorage Position

e.g. DC F (PP7)

STfCSEOSVtAP

STfCStDC

PP1 – Pilot Station EOSV

PP7 – Pilot Station EOSV

STfDCtCS

STfDCEOSVtAP PP7 – Anchorage Position

PP7 – Pilot Station BOSV

WTfBAaDCAP

DC F (PP7)

Or UPTaDC

PP7 Berth 1 Unberthing Position

PP7 Berth 1 EDS Position EDSPTaDC

BERTH 1 PP7 Berth 2 Unberthing Position

UPTaDC EUT UITdUP = UPTaDC - EUT


BPTaDC

IDSPTaDC

PP7 Berth 2 EDS Position EDSPTaDC

BERTH 2

PP7 Berth 1 – Unloading Process Position

PP7 Berth 2 – Unloading Process Position UPTaDC IDSPTaDC EUT UITdUP = UPTaDC - EUT


Figure 2: Design of Stations within VRV in MIRP Simulation Model Several assumptions are made to the MIRP simulation model as the following:

a. Time required for sailing from CS to a DC or from a DC to CS is expressed as a statistical

b.

distribution. At the current time of the MIRP system, a vessel may be on its way to a DC or CS (sailing). The remaining time required to complete its journey is assumed to be the proportion of the distance left. If the silos of associated product at particular DC are full during the unloading process, the process will be interrupted for one hour to bring the level of inventory down.

393


c. Similarly, if the silos of associated product at CS are empty during the loading process, the process will be interrupted for one hour to wait the stock being replenished.

d. If the silos of associated product at CS are full, replenishment process will be interrupted for one hour to wait the level of inventory down. 3.3

Simulation Model of the MIRP System The structure of computer or simulation model of the MIRP system under study can be seen in Figure 3. The logics begin with reading external data, which were created in a Microsoft ® Excel file (i.e. MIRP Initial Values.xls). This file contains parameters and initial conditions of the MIRP system at the current time. Create Vessel Logic then creates unique vessels and assigns initial conditions and attributes values to each vessel. These vessels are then located at their current position in the system, such as sail-go, sail-back, CS, or a DC. Each vessel will then follow the logic at its current position, which is representing the detailed activities in VRV. Sail-Go and Sail-Back Logic is only used once at the beginning time of simulation run if there is any vessel at that position. Each vessel will route inventory from CS to an assigned DC in full load and back in empty. This cycle will continue until the simulation is terminated.

READ EXTERNAL DATA LOGIC  Parameters  Initial Conditions

CREATE VESSELS LOGIC  Create unique vessels at t=0  Assign initial conditions and attributes values

WRITE TO EXTERNAL FILE  Simulation Output  System States

SUPPLY AND DEMAND LOGIC  Current inventory status  Inventory cost calculation

ROUTING AND SCHEDULING LOGIC

Send the vessels into their current positions

SAIL-GO AND SAIL-BACK LOGIC (These are only temporary (transient) station at initial time (t=0) of the simulation

CENTRAL SUPPLY LOGIC  Stations and logic for the processes at Central Supply

DISTRIBUTION CENTRE A LOGIC  Stations and logic for the processes at DC A

...

DISTRIBUTION CENTRE B LOGIC  Stations and logic for the processes at DC B

DISTRIBUTION CENTRE i LOGIC  Stations and logic for the processes at DC i

Vessel movement

Data and Information

Figure 3: Structure of Module‘s Logic in the MIRP Simulation Model As previously defined in the conceptual model, the CS (PP1) area is divided into several sections (stations) according to detailed sequence of vessel activities in this area. This detailed division not only gives flexibility in developing and animating the MIRP simulation model, but it will also promote the validity of the model by eliminating the possibility of miscalculating activities durations of the vessels during the early stage of the simulation run. This miscalculation is due to placing the vessel not at its exact position and consequently not at its exact activity start time, whereas time precision is a very crucial factor in this MIRP simulation model. In term of inventory calculation part of this simulation model, every movement (time advance) of the vessel will affect the inventory status. Time will also be used for calculating other performance measures in the model, such as costs of each stage of a time charter vessel, despatch, and

394


demurrage. Similarly, DCs areas are also divided into several stations according to detailed sequence of vessel activities at DCs. The fundamental reason for this is also similar. Duration for each activity within the VRV is recorded in a global variable every time a vessel has moved from one station to the next station. The detailed logic in Figure 4 reveals the challenge in modelling this MIRP simulation. Due to the size of the model, only the CS part is presented.

Write S im ula tio n Re s ult Ve s s e l Ro und Vo y a g e (Write to M s E x c e l F ile )

Wri te Ve s s e l Ro u n d Vo y a g e

As s i g n m e n t a t Sta ti o n 2 1 0 1

Sta ti o n


Ro u te

21

PP1 Ber t hing Posit ion

0 Sta ti o n

Tr ue

2 2 Is th e F i rs t Sta ti o n Vi s i te d b y th e Ve s s e l ?


Se i z e Be rth 1 a t PP1 Be fo re Be rth i n g Sta te


22


Ro u te PP1 Ber t h1 I nit ial Dr af t Sur vey Posit ion

0

Fals e

Ho l d a t PP1 fo r Be rth Av a i l a b l e Si g n a l

Bra n c h NR( PP1 Ber t h2) == 1 NR( PP1 Ber t h1) == 1 ( NR( PP1 Ber t h1) == 0 && NR( PP1 Ber t h2) == 0) && ( ( Am ount Car r ied( Vessel I ndex) / PP1 Loading Capacit y( 1) ) ( Am ount Car r ied( Vessel I ndex) / PP1 Loading Capacit y( 2) ) )

If If If If

Se i z e Be rth 2 a t PP1 Be fo re Be rth i n g Sta te


Ro u te PP1 Ber t h2 I nit ial Dr af t Sur vey Posit ion

0 Sta ti o n

2 3 1 Is No t th e F i rs t Sta ti o n Vi s i te d b y th e Ve s s e l ?

Tr ue

As s i g n m e n t a t Sta ti o n 2 3 1 0 2

Sto re Be fo re IDS a t PP1 Be rth 1

De l a y fo r In i ti a l Dra ft Su rv e y a t PP1 Be rth 1

Un s to re Afte r IDS a t PP1 Be rth 1


Sto re Be fo re IDS a t PP1 Be rth 2

De l a y fo r In i ti a l Dra ft Su rv e y a t PP1 Be rth 2

Un s to re Afte r IDS a t PP1 Be rth 2

231

0

Fals e


Se i z e Be rth 1 a t PP1 Be fo re IDS Sta te

0 Sta ti o n

2 4 1 Is No t th e F i rs t Sta ti o n Vi s i te d b y th e Ve s s e l ?

Tr ue

241

0

Sta ti o n

Fals e


232


Se i z e Be rth 1 a t PP1 Be fo re L o a d i n g Sta te

Se i z e Be rth 2 a t PP1 Be fo re IDS Sta te


Bra n c h

Sta ti o n


242

Se i z e Be rth 2 a t PP1 Be fo re L o a d i n g Sta te


Bra n c h If Els e

Sta ti o n

Un s to re Afte r Wa i t u n ti l Oth e r Sto re fo r L o a d i n g Ve s s e l De l a y a t Loading Po s tp o n e m e n t a t L o a d i n g a t PP1 Po s tp o n e m e n t a t PP1 Be rth 1 PP1 Be rth 1 Be rth 2 Fi n i s h

Ho l d a t PP1 Be rth 1 fo r Loading


Wa i t u n ti l Oth e r Un s to re Afte r Sto re fo r L o a d i n g Ve s s e l De l a y a t Loading Po s tp o n e m e n t a t L o a d i n g a t PP1 Po s tp o n e m e n t a t PP1 Be rth 2 Be rth 1 Fi n i s h PP1 Be rth 2


Ho l d a t PP1 Be rth 2 fo r Loading


W ait Unt il O t her Vessel Delay at PP1 Ber t h1 Finis h == 1 && Cem ent Type I ndex Loaded at PP1 Ber t h1 == Cem ent Type I ndex


Se i z e Be rth 1 a t PP1 Be fo re EDS Sta te


Sto re Be fo re EDS a t PP1 Be rth 1

De l a y fo r En d Dra ft Su rv e y a t PP1 Be rth 1


Se i z e Be rth 2 a t PP1 Be fo re EDS Sta te


Sto re Be fo re EDS a t PP1 Be rth 2

De l a y fo r En d Dra ft Su rv e y a t PP1 Be rth 2

233


W ait Unt il O t her Vessel Delay at PP1 Ber t h2 Finis h == 1 && Cem ent Type I ndex Loaded at PP1 Ber t h2 == Cem ent Type I ndex

If Els e

Sta ti o n

Un s to re Afte r EDS a t PP1 Be rth 1

Re l e a s e Be rth 1 a t PP1

As s i g n m e n t a t Sta ti o n 2 3 4

Ro u te

234

Sta ti o n 243

Ro u te Sta ti o n

As s i g n m e n t a t Sta ti o n 2 9

Sta ti o n

Ro u te

31

Bra n c h

244 PP3 Pilot St at ion EO SV

29 If If If If If If

PP PP PP PP PP PP

Dest inat ion Dest inat ion Dest inat ion Dest inat ion Dest inat ion Dest inat ion

of of of of of of

Vessel( Vessel Vessel( Vessel Vessel( Vessel Vessel( Vessel Vessel( Vessel Vessel( Vessel

I ndex) I ndex) I ndex) I ndex) I ndex) I ndex)

== == == == == ==

2 3 4 5 6 7

As s i g n m e n t a t Sta ti o n 2 4 4

Re l e a s e Be rth 2 a t PP1

Ro u te PP1 Pilot St at ion BO SV

Ro u te PP4 Pilot St at ion EO SV

Ro u te Ro u te

Ro u te

PP1 Pilot St at ion BO SV

Un s to re Afte r EDS a t PP1 Be rth 2

PP5 Pilot St at ion EO SV



Figure 4: Discrete Part of the MIRP Simulation Model – Logic for CS Continuously during the simulation run, Supply and Demand Logic will collect data from CS, DCs, and vessels. This data will be used to control the process in the MIRP system and for routing and scheduling purposes. The structure of this logic consists of the following: a. b. c. d. e. f.

Defining continuous approach in inventory model and initialisation of PPs inventory levels and cumulative rates for each cement type Observing inventory level to interrupt the unloading process Observing inventory level to interrupt the loading process Observing inventory level to end the unloading process Observing inventory level to end the loading process Observing inventory level to interrupt stock replenishment at CS

The detail logic can be seen in Figure 5 below. Again, due to the size of the model, only some of them are presented. Periodically during the simulation run, the simulation outputs are stored in a Microsoft ® Excel file (i.e. MIRP Simulation Results.xls).

395

SUPPLY AND DEMAND LOGIC Continuous Approach in Inventory Model Continuous

th

Rates

Levels P P1 S i l o s S t o c k f o r Ce m e n t T y p e P P2 S i l o s S t o c k f o r Ce m e n t T y p e P P3 S i l o s S t o c k f o r Ce m e n t T y p e P P4 S i l o s S t o c k f o r Ce m e n t T y p e P P5 S i l o s S t o c k f o r Ce m e n t T y p e P P6 S i l o s S t o c k f o r Ce m e n t T y p e P P7 S i l o s S t o c k f o r Ce m e n t T y p e A m o u n t o f Ca rg o On b o a rd i n Ve s s e l

P P1 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P2 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P3 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P4 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P5 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P6 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P7 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e Ca rg o Cu rre n t L o a d i n g o r Un l o a d i n g Ra t e f o r V e s s e l

SUPPLY AND DEMAND LOGIC

26 International Conference on CAD/CAM, Robotics and Factories of the Future 2011 th Approach in Inventory Model Malaysia 26th-28Continuous July 2011, Kuala Lumpur, SUPPLY AND DEMAND LOGIC Continuous

and Rates

P P1 S i l o s S to c k fo r Ce m e n t T y p e S i l o s S to c k fo r Ce m e n t T y p e P P3 S i l o s S to c k fo r Ce m e n t T y p e P P4 S i l o s S to c k fo r Ce m e n t T y p e P P5 S i l o s S to c k fo r Ce m e n t T y p e P P6 S i l o s S to c k fo r Ce m e n t T y p e P P7 S i l o s S to c k fo r Ce m e n t T y p e A m o u n t o f Ca rg o On b o a rd i n Ve s s e l

P P1 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P2 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P3 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P4 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P5 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P6 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e P P7 Cu m u l a ti v e Ra te f o r Ce m e n t T y p e Ca rg o Cu rre n t L o a d i n g o r Un l o a d i n g Ra te fo r V e s s e l

P P1 P P2 P P3 P P4 P P5 P P6

P P1 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P2 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e Ra t e f o r Ce m e n t T y p e Ra t e f o r Ce m e n t T y p e Ra t e f o r Ce m e n t Tto ype Assign

0 P P2

Detect

0

If If If If If

Continuous Approach in Inventory Model

Detect

S i l o s S t o c k f o r Ce m e n t T y p e S i l o s S t o c k f o r Ce m e n t T y p e Ty pe Ty pe Ty pe Ty pe

S i l o s S t o c k f o r Ce m e n t Assign Initial S i l o s S t o c k f o r Ce m e n t Inventory S i l o s S t o c Levels k f o r Ce m e n t Assign to S i l oand s S tRates o c k f o r Ce m e n t Interrupt

P P3 Cu mu Dispose the 2l a t i v e P P4 Cu m u l a t i v e Entities for P P5 Cu m u l a t i v e

Cement P P6 Types Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e Delay Unloading Continue P P7 Cu m u l a t i v e Ra t e f o r Ce m e nt Ty p e 0 Ca for rg o CT1 Cu rre n t L o a d i n g Unloading o r Un l o a d i nCT1 g Ra t e at PP2

S i l o s S t o c k f o r Ce m e n t T y p e 0 PA P7 m o uUnloading n t o f Ca rg o CT1 On b o a rd i n Ve s s e l

Branch

P P1 S i l o s S t o c k f o r Ce m e n t T y p e (1 )

Rates Continuous Inventory LevelsPPs Detect Level to Interrupt ProcessRates at a Distribution CentreType Initialisation of Inventory LevelsUnloading and Cumulative for Each Cement Create 2 Entities for Cement Types

Dispose Detect Unloading CT1 at PP2

PP2 Initialisation of PPs Inventory Levels and atCumulative Rates for0Each Cement Type P P2 S i l o s S t o c k f o r Ce m e n t

at PP2 T y p e (1 )

Assign to

PP2

Assign to Continue

Dispose the 2 Delay Unloading

Assign Initial Interrupt

fo r Ve s s e l

Dispose Detect

Detect2 Entities Create Unloading CT2 at Entities for Unloading CT2 Inventory Levels Unloading CT2Process at at PP2 for CT2 Detect Inventory Level to Interrupt Unloading for Cement Types PP2 a Distribution Centre at PP2 Cement Types and Rates at PP2

Initialisation of continuous PPs Inventory Levelsapproach and Cumulativein Rates for Each Cement Type Defining inventory model and initialisation ofUnloading PPs inventory levels and Detect Inventory Level to Interrupt Process at a Distribution Centre PP2 PP3 cumulative rates for each cement type P P2 S i l o s S t o c k f o r Ce m e n t T y p e (2 )

0

Delay Unloading Dispose the 2 Delay Unloading at PP2 for CT1 Entities for at PP3 for CT1 Cement Types

0

Detect Detect

0

0

Assign to Assign Interruptto Assign Initial Interrupt Unloading CT1 Inventory Levels Unloading for Cement Types at PP2CT1 P P2 S i l o s S t o c k f o r Ce m e n t T y pand e (1 Rates at) PP3 P P3 S i l o s S t o c k f o r Ce m e n t T y p e (1 )

Detect Create Detect2 Entities

P P2 S i l o s S t o c k P P3 S i l o s S t o c k

Assign to Assign to Interrupt Interrupt Unloading CT2 CT2 Unloading Assign at) PP3 PP2to f o r Ce m e n t T y p e (2 at f o r Ce m e n t T y p e Interrupt (2 ) Unloading CT1 at PP2 fo r Ce m e n t T y p e (1 )

0

Assign to Assign to Continue Continue Unloading CT1 Unloading CT1 at PP2 at PP3

Dispose Detect Dispose Detect Unloading CT1 at Unloading PP2CT1 at PP3

Assign to Assign to Continue Continue Unloading CT2 CT2 Unloading Assign at PP3 PP2to at

Delay Unloading Unloading Delay at PP3 PP2 for for CT2 CT2 at

Delay Unloading

Dispose Detect Dispose Detect Unloading CT2 CT2 at at Unloading PP2 PP3 Dispose Detect

Continue

0 0

Detect

Detect Detect Detect

Assign Interruptto Assign Interruptto Unloading CT1 Interrupt Unloading at PP3CT2 Unloading CT1 at PP2 f o r Ce m e n t T y p e (1 ) Assign to Interrupt Interrupt Assign to Unloading CT2 Unloading CT2 at PP3 PP4to Interrupt Assign at ff o o rr Ce Ce m me en n tt T T yy p pe e (2 (2 )) Unloading InterruptCT2 at) PP2CT1 f o r Ce m e n t T Unloading y p e (2 at PP3 fo r Ce m e n t T y p e (1 )

at) PP2 P P2 S i l o s S to c k fo r Ce m e n t T y p e (2 P P2 S i l o s S t o c k

Detect Detect

Detect Detect

P P P4 P3 S S ii ll o o ss S S tt o o cc kk

P P2 S i l o s S t o c k

P P3 S i l o s S to c k

Detect Detect Detect Detect

at PP2 for CT1

Delay Unloading Unloading Delay at PP3 PP4 for for CT2 CT2 at Delay Unloading Delay Unloading at PP2 for CT2 at PP3 for CT1

Assign to Assign Interruptto Interrupt Assign to Unloading CT1 Assign Unloading CT1 Interrupt at PP5to Interrupt at PP4CT2 Unloading Unloading CT1 at)PP3 fo r Ce m e n t T y p e (2 at PP3to f o r Ce m e n t T y p Assign e (1 ) Assign Interruptto Interrupt Unloading CT2 Assign to Unloading CT2 f o r Ce m e n t T y p eat (2 )PP5 Interrupt at PP4to f o r Ce m e n t T y p eAssign (2 ) Unloading InterruptCT2 at PP3CT1 f o r Ce m e n t T Unloading y p e (2 )

Detect Detect P P5 S i l o s S t o c k P P4 S i l o s S t o c k

Detect Detect


Detect Detect

Detect

at PP6 for CT1 Delay Unloading at PP5 for CT1 Delay Delay Unloading Unloading at PP4 for CT2 CT1

P P5 S i l o s S t o c k f o r Ce m e n t T y p e (1 ) P P4 S i l o s S tto oc k

Detect

Detect


Detect


Dispose Detect Dispose Detect Unloading CT1 at Unloading CT1 at Dispose Detect PP5 Dispose Detect PP4 Unloading CT2 at Unloading PP3CT1 at PP3 Dispose Detect Dispose Detect Unloading CT2 at Unloading PP5CT2 at Dispose Detect PP4 Unloading CT2 at Dispose Detect PP3CT1 at Unloading

PP4

Assign at PP4to Continue Assign to Unloading CT1 Continue at PP6to Assign Unloading CT1 Assign to Continue at PP5 Continue Unloading Unloading CT2 CT1 Assign at PP4to Continue Assign to Unloading CT2 Continue at PP6 Assign to Unloading CT2 Continue at PP5 Assign to Unloading CT2 Continue at PP4

Delay Unloading

Interruptto Assign Unloading InterruptCT1 at PP6to Assign Unloading CT1 Assign to Interrupt Interrupt at PP5 Unloading Unloading CT2 CT1 Assign at PP4to (2 ) ffo o r Ce m e n t T y p eInterrupt (1 Assign to Unloading CT2 Interrupt at f o r Ce m e n t T y p e (2 )PP6to Assign Unloading CT2 Interrupt at PP5 f o r Ce m e n t T y p Assign e (2 ) Unloading to CT2 Interrupt at PP4 f o r Ce m e n t T y p e (2 )


Dispose Detect Detect Dispose Unloading CT2 CT2 at at Unloading Dispose Detect PP4 PP3 Unloading CT2 at Dispose Detect PP2CT1 at Unloading PP3

…

Delay Unloading Delay Unloading at PP5 for CT2 at PP4 for CT2 Delay Unloading at PP3 for CT2 Delay Unloading at PP4 for CT1

at) PP4to P P4 S i l o s S to c k fo r Ce m e n t T y p eAssign (1

Dispose Detect Unloading CT1 at Dispose Detect Unloading CT2 at PP4 PP3 Unloading PP2CT1 at PP2

Assign to Assign to Continue Continue Assign to Unloading Assign CT1 to Unloading Continue at PP5CT1 Continue at PP4 Unloading UnloadingCT2 CT1 at atPP3 PP3to Assign Assign to Continue Continue Unloading CT2 Assign Unloading CT2 at PP5to Continue at PP4to Assign Unloading CT2 Continue at PP3CT1 Unloading

Delay Unloading

Delay Unloading at PP5 for CT1 at PP4 for CT1 Delay Unloading Delay Unloading at PP3 for CT2 at PP3 for CT1

P P5 S i l o s S t o c k f o r Ce m e n t T y p e (1 ) P P4 S i l o s S t o c k f o r Ce m e n t T y p e (1 )

P P3 S i l o s S to c k P P3 S i l o s S t o c k

Assign to Continue Assign to Continue Unloading CT1 Continue Unloading PP3CT2 at PP4 Unloading at PP2CT1 at PP2 Assign to Continue Continue Assign to Unloading CT2 Unloading CT2 at PP3 PP4to Continue Assign at Unloading CT2 Continue at PP2CT1 Unloading at PP3

Delay Unloading Delay Unloading PP4 at PP3 for CT1 Delay Unloading at PP2 for CT2

at PP4 P P P4 P3 S S ii ll o o ss S S tt o o cc kk ff o o rr Ce Ce m me en n tt T T yy p pe e (1 (1 ))

If If If If If

0

Delay Unloading at PP6 for CT2 Delay Unloading at PP5 for CT2 Delay Unloading at PP4 for CT2 Delay Unloading

Dispose Detect Unloading CT1 at Dispose Detect PP6CT1 at Unloading Dispose Detect PP5 Dispose Detect Unloading Unloading CT2 CT1 at at PP4 PP4 Dispose Detect Unloading CT2 at Dispose Detect PP6 Unloading CT2 at Dispose Detect PP5 Unloading CT2 at Dispose Detect PP4

0 0 0

PP4 PP3 PP2

0 0

0 0

0 0

0 0

PP3 PP5 PP4 PP3

0 0

0 0 0 0

0 0

Branch


Detect 0 Unloading CT1 at Unloading CT1 at PP2 for CT1 Unloading Detect Inventory Level to Interrupt Process at PP2 a Distribution Centre at PP2 0 PP2 Assign to Assign to Dispose Detect P P2 S i l o s S to c k

at PP7 Situation 12

Delay Unloading at PP7 for CT2 Situation 20

Assign to Continue Unloading CT2 at PP7 Situation 20

Assign to Interrupt Unloading CT2 at PP7 Situation 21






Detect Inventory Level to Interrupt Vessel Loading Process at Central Supply

SUPPLY AND DEMAND LOGIC Rates Entities for Cement Types

Levels Inventory Levels

for Cement Types

Situation 12


Rates

Levels

P P1 S i l o s S t o c k f o r Ce m e n t T y p e P P1 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P2 S i l o s S t o c k f o r Ce m e n t T y p e P P2 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P3 S i l o s S t o c k f o r Ce m e n t T y p e P P3 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P4 S i l o s S t o c k f o r Ce m e n t T y p e P P4 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P5 S i l o s S t o c k f o r Ce m e n t T y p e P P5 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P6 S i l o s S t o c k f o r Ce m e n t T y p e P P6 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e P P7 S i l o s S t o c k f o r Ce m e n t T y p e P P7 Cu m u l a t i v e Ra t e f o r Ce m e n t T y p e A m o u Assign n t o f Ca rg o On b o a rd i n Ve s s eDispose l 2 n t L o a d i n g o r Un l o a d i n g Ra t e f o r V e s s e l Initial Ca rg othe Cu rre

Initialisation of PPs Inventory Levels and Cumulative Rates for Each Cement Type Continuous Approach in Inventory Model Create 2 Entities Continuous

at PP7 Situation 12

Assign to Interrupt Loading CT1 at PP1 Situation

Delay Loading at PP1 for CT1 Situation 01

Assign to Continue Loading CT1 at PP1 Situation

01 Ce m e n t T y p e I n d e x L o a d e d a t P P1 B e rt h 1 = = 0 & & Ce m e n t T y p e I n d e x L o a d e d a01 t PP 1 Be rt h 2 = = 1 Ce m e n t T y p e I n d e x L o a d e d a t P P1 B e rt h 1 = = 1 & & Ce m e n t T y p e I n d e x L o a d e d a t PP 1 Be rt h 2 = = 0 Ce m e n t T y p e I n d e xAssign L o a d e dtoa t P P1 B e rt h 1 = = 1 & & Ce m e n t T y p e I n d e x L o a dAssign e d a t PPto 1 Be rt h 2 = = 1 Ce m e n t T y p e I n d e x Interrupt L o a d e d a t P P1 B e rt h 1 = = 1 & & Ce m e n t T y p e I n d e x L o a d e d a t PP 1 Be rt h 2 = = 2 Continue Ce m e n t T y p e I n d e x L o a d e d a t P P1 B e rt h 1 = = 2 & & Ce m e n t T y p e I n d e x L o a d e d a t PP 1 Be rt h 2 = = 1 Loading CT1 at PP1 Situation 10


Dispose Detect Loading CT1 at PP1

0

Loading CT1 at PP1 Situation 10

Assign to Interrupt Loading CT1 at PP1 Situation 11


Assign to Continue Loading CT1 at PP1 Situation 11







Assign to Interrupt Loading CT2 at PP1 Situation In d e x L o a02 d e d a t PP1



PP1 for CT2 Situation 12










Assign to Continue Loading CT2 at PP1 Situation L o a d e d a02 t PP1 Be rth 2

PP1 Dispose Detect Loading CT2 at PP1

Ce m e n t T y p e Be rth 1 = = 0 && Ce m e n t T y p e In d e x == 2 Ce m e n t T y p e In d e x L o a d e d a t PP1 Be rth 1 = = 1 && Ce m e n t T y p e In d e x L o a d e d a t PP1 Be rth 2 = = 2 Ce m e n t T y p e In d e xAssign L o a d e dtoa t PP1 Be rth 1 = = 2 && Ce m e n t T y p e In d e x L o a dAssign e d a t PP1 to Be rth 2 = = 0 Ce m e n t T y p e In d e x L o a d e d a t PP1 Be rth 1 =Delay = 2 &&Loading Ce m e n t at T y p e In d e x L o a d e d a t PP1 Be rth 2 = = 1 Continue Ce m e n t T y p e In d e x Interrupt L o a d e d a t PP1 Be rth 1 = = 2 && Ce m e n t T y p e In d e x L o a d e d a t PP1 Be rth 2 = = 2

0


Detect inventory level to interrupt loading process at CS

PP4 PP6 PP5 PP4

0

Detect inventory level to interrupt unloading PP5 process at DCs with one berthPP6

Detect


Unloading CT1 at PP5 to Interrupt f o r Ce m e n t TUnloading y p e (1 ) CT1 Assign to to Assign at f )PP6Ce m e n t Interrupt f o r Ce m e n t T y p eIInterrupt (1 Ce m e n t If Unloading CT1 Ce m e n t Unloading CT2 If PP5Ce m e n t I at f(1 )PP5 ffo o rr Ce e Ce m me en n tt T T yy p pAssign eIat toCe m e n t f(2 ) Interrupt Unloading CT2 Assign to at f o r Ce m e n t T y p eInterrupt (2 )PP6 Assign to Unloading CT2 Interrupt f o r Ce m e n t T y p eat (2 )PP5 Unloading CT1 at PP6 fo r Ce m e n t T y p e (1 )

at PP5 for to CT1 Assign

Interrupt Unloading CT1 Delay Unloading at atPP7 PP6Situation for CT1 I n d e x Un l01 o a d e d a t P P7

P P5 S i l o s S to c k fo r Ce m e n t T yBranch p Assign e (1 ) Detect

Detect P P7 S i l o s

Sto c k

Detect


P o cc kk P P5 P5 S S ii ll o o ss S S tto

Detect Detect


Detect



Detect Detect Detect

P P7 S i l o s S t o c k P P6 S i l o s S t o c k P P6 S i l o s S to c k

Detect

Unloading CT1 PP5 DelayatUnloading Assign to at PP7 for CT1 Continue Situation 01 Unloading CT1 Assign to to B e rt h 1 = = Assign 0at&& Ce m en t Ty pe PP6 B e rt h 1 = = Continue 1 && Ce m e n t T y p e Continue B e rt h 1 Unloading = = 1 && Ce m en t Ty pe CT1 Unloading CT2 B e rt h 1Delay = = 1 && Ce m e n t T y p e Unloading at&& PP5 PP5 B e rt h 1 = = 2at Ce m e n t T y p e Assign to at PP7 for CT1 Continue Situation 10

Unloading CT1 at PP5to Assign

0

Continue 0 Dispose Detect Unloading CT1 Unloading CT1 at at PP7 Situation T y p e Delay Unloading I n d e x Dispose Un l o a PP6 d01 e d Detect a t P P 7 B e rt h 2 = = 1 Dispose Detect T y p e I nDelay d e x UnUnloading l o a d e d a t P P7 I n d e x Unloading Un l o a d e d aCT1 t P P 7atB e rt h 0 2 == 0 T y p e I n at d e PP5 xAssign Un l ofor a d to e d a t P P7 I n d e x Unloading Un lAssign o a d e d aCT2 t P P 7atB e rt h 2 = = 1 CT1 to at PP5 for CT2 T y p e I n d e x Un l o a d e d a t P P7 I n d e x Un l o a PP5 dPP5 e d a t P P 7 B e rt h 2 = = 2 Interrupt T y p e I n d e x Un l o a d e d a t P P7 I n d e x Un lContinue o a d e d a t P P 7 B e rt h 0 2 == 1 0 Unloading CT1 Unloading CT1 Dispose Detect Delay at PP7Unloading Situation at PP7 Situation Unloading CT2 at at PP610for CT2 Unloading CT2 Assign to 10 Detect PP6 Dispose Delay Unloading at PP6 Continue 0 Unloading CT2 at Assign to Assign to Assign to at PP5 for CT2 Unloading CT2 Dispose Detect Delay Unloading Delay Unloading PP5 Interrupt Continue Continue at PP5 Unloading CT1 at Unloading at PP7 for CT1 0 Unloading CT1 at PP6 for CT1 Unloading CT1 PP6 at PP7 Situation at PP7 Situation at PP611 Situation 0 Assign 11 to Assign 11 to Delay Unloading Interrupt Continue Branch Assign to Assign to Assign to AssignDetect to Unloading CT1 at PP7 for CT1 Unloading CT1 Dispose Assign to Assign to Delay Unloading Delay Unloading Interrupt Interrupt Continue Dispose Detect Continue at PP7 Situation f o r Ce m e n t T y p eInterrupt (1 ) at PP7 Situation Delay Unloading Unloading CT1 at Situation 01 Continue Unloading CT1 Unloading CT1 at PP7 for CT1 at PP6 for Unloading CT1 Unloading CT1 at Unloading CT2 Ce m e n t T y p e I n d xPP6 Un l01 ofor ade d a t P P7 B e rt h 1 Unloading = = 0 && Ce m e n t T y p e I n d e x Un l o a PP6 d01 e d a t P P 7 B e rt h 2 = = 1 Unloading CT2 CT2 Iat f PP6CT2 atatePP7 Situation PP7 Situation at PP6 Situation 12 Ce m e n t T y p e I n d e xatUn l o a dPP6 e d a t P P 7 B e rt h 2 = = 0 f o r Ce m e n t T y p eI (1 f )PP6 Ce m e n t T y p e I n d e x Un l o a d e d a t P P7 B e rt h 1 = = 1at&& 0 PP6 fo r Ce m e n t T y p eat (2 ) Ce m e n t T y p e I n d e xAssign Un 12 l o a d to e d a t P P7 B e rt h 1 = = 1 && Ce m e n t T y p e I n d e x Un lAssign o a d12 e d ato t P P 7 B e rt h 2 = = 1 If 0 Ce m e n t T y p e I n d e x Un l o a d e d a t P P7 B e rt h 1Delay = = 1 && Ce m e n t T y p e I n d e x Un l o a d e d a t P P 7 B e rt h 2 = = 2 Unloading If Interrupt Continue e xAssign Un l o a d to e d a t P P7 B e rt h 1 = = 2 && Ce m e n t T y p e I n d e x Un lAssign o a d e d ato t P P 7 B e rt h 2 = = 1 If Assign toCe m e n t T y p e I n dUnloading CT1 Assign to at PP7 for CT1 Unloading CT1 Delay Unloading Dispose Detect Interrupt Continue Delay Interrupt atUnloading PP7Unloading Situation Continue at PP7 Situation Situation 10 CT1 Unloading CT2 at PP7 for CT1 Unloading CT1 at Unloading CT2 PP6 for to CT2 10 Unloading CT2 10 Assign atatPP7 Situation PP6 Assign to at PP7 Situation Situation 21 at PP6 at PP6 f o r Ce m e n t T y p e (2 ) Delay Unloading 21 to Interrupt 21 to Continue Branch 0 Assign Assign Unloading at PP7 for CT1 Delay Unloading Unloading CT1 InterruptCT1 Continue atUnloading PP7 Situation fo r Ce m e n t T y p e (1 ) atUnloading PP7 Situation CT1 01 atSituation PP7 for CT1 CT1 Assign to Assign to Ce m e n t T y p e Inat d ePP7 x Interrupt Un l01 o a d e d a t P P7 B e rth 1Delay = = 0 && Ce m e n t T y p e In d e xatUn l o a d01 e d a t P P 7 B e rth 2 = = 1 Situation Unloading PP7 Situation If Situation 11 Continue Branch Ce m e n t T y p e In d e xAssign Un l11 o a d to e d a t P P7 B e rth 1 = = 1 && Ce m e n t T y p e In d e x Un lAssign o a d e d ato t P P 7 B e rth 2 = = 0 If Unloading CT2 11 CT2 at PP7 for CT2 Ce m e n t T y p e In d e xAssign Un l o a d to e d a t P P7 B e rth 1Delay = = 1 && Ce m e n t T y p e In d e x Unloading Un lAssign o a d e d ato t P P 7 B e rth 2 = = 1 Unloading I f Interrupt f o r Ce m e n t T yBranch p e (2 ) Ce m e n t T y p e Inat d ePP7 x Un l oSituation a d e d a t P P7 B e rth 1Delay = Situation = 1 && Ce m lContinue o a d eSituation d a t P P 7 B e rth 2 = = 2 02e n t T y p e In d e xatUnPP7 Unloading If Interrupt Assign to Continue Unloading CT1 PP7 Assign at PP7 for CT1 CT1 Ce P7 B e rth Un eedd aato tt P P7 B e rth Ce m m ee nn tt TTyy pp ee In In dd ee xx Un Un l02 loo aa ddee dd aatt P Be rth 1Delay 1 ==== 20 && && Ce Ce m m eenn tt TTyy ppee In In dd eexx Unloading Unllooaa dd02 PP7 Be rth 22 ==== 12 IIff Unloading Unloading CT1 Unloading f o r Ce m e n t T y p e I(1 ) Ce m e n t T y p e Inat d ePP7 x Interrupt Un lSituation o a d e d a t PP7 Be rth 1at =Situation =PP7 1 &&for CeCT1 m e n t T y p e In d e xat Un lContinue o a d Situation e d aCT1 t PP7 Be rth 2 = = 2 PP7 01 f Ce m e n t T y p e Inat d ePP7 x Un 01 lSituation oade d a t PP7 Be rth 1 =Situation = 2 &&for Ce 10 m e n t T y p e In d e x Unloading Un l o a d Situation e d aCT1 t PP7 Be rth 2 = = 0 Unloading CT1 PP7 If Ce m e n t T y p e I n d e xAssign Un l o a d to e d a t P P7 B e rt h 1 at = =PP7 0 && Ce CT1 m e n t T y p e I n d e xat Un lAssign o a d01 e d ato t P P 7 B e rt h 2 = = 1 If Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 1Delay = = 2 && Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 2 = = 1 Unloading 10 Situation PP7 If Ce nat dd eePP7 xx Interrupt Un P7 B e rt h 11 ==Situation == 12 && n dd eexxatUn llContinue ooaa dd10 eedd aa tt P P7 B e rt h 22 ==== 02 PP7 Situation 12 If Ce m m ee nn tt TTyy pp ee IIn Un lloo aa ddee dd aatt P Be rth && Ce Ce m m eenn tt TTyy ppee IIn Un PP7 Be rth If Ce m e n t T y p e I n d e xAssign Un 12 l o a d to e d a t P P7 B e rt h 1 at = =PP7 1 && for Ce m e n t T y p e I n d e x Un lAssign o a d e d ato t P P 7 B e rt h 2 = = 1 Unloading CT2 CT2 If Unloading 12 CT2 Ce m e n t T y p e I n d e xAssign Un l o a d to e d a t P P7 B e rt h 1Delay = = 1 && Ce m e n t T y p e I n d e x Un lAssign o a d e d ato t P P 7 B e rt h 2 = = 2 Unloading If Interrupt at PP7 Situation 12e n t T y p e I n d e xatUnPP7 Ce m e n t T y p e I n d e x Un l o a d e d a t P P7 B e rt h 1Delay = Situation = 2 && Ce m lContinue o a d eSituation d a t P P 7 B e rt h 2 = = 1 Unloading If Interrupt Assign Unloading CT1 Continue Assign 12 to at PP7 for CT1 Unloading CT1 12 to Delay Unloading Interrupt CT1 atSituation PP7 for CT1 atUnloading PP7 Situation Continue CT1 atUnloading PP7 Situation 10 Assign to Assign to CT1 atUnloading PP7 Situation atSituation PP7 for CT1 10 CT1 atUnloading PP710 Situation 11 Delay Unloading Interrupt Continue at PP7 Situation 11 at PP7 Situation 11 Situation 21 Assign at PP7 for CT2 Unloading CT2 Unloading CT2 Assign 21 to 21 to Delay Unloading at PP7 Situation Interrupt Assign to at PP7 Situation Continue Assign to Situation 20 Delay Unloading 20 CT1 Unloading Interrupt 20 CT1 at PP7 for CT1 Unloading Continue Assign to Assign to atUnloading PP7 Situation CT1 atSituation PP7 for CT1 atUnloading PP7 Situation CT1 11 Assign to Delay Unloading Assign to Interrupt Continue Branch at PP711 Situation 11 at PP7 Situation Situation 12 Delay Unloading InterruptCT2 Continue Unloading at PP7 for CT2 Unloading CT2 12 12 at PP7 for CT2 Unloading CT2 CT2 Assign to at PP7 Situation Assign to f o r Ce m e n t T y p e (2 ) atUnloading PP7 Situation Situation 02 at PP7 Situation Delay Unloading at PP7 Situation Interrupt Situation 21 Ce m e n t T y p e In d e xAssign Un 02 l o a dto e d a t PP7 Be rth 1 = = 0 && Ce m e n t T y p e In d e x Un Assign lContinue o a d02 e d ato t PP7 Be rth 2 = = 2 If CT1 Ce m e n t T y p e In Unloading d e x Interrupt Un21 lo a de d a t PP7 Be rth Delay 1at = =PP7 1 Unloading &&for CeCT1 m e n t T y p e In d e x Unloading Un lContinue o a d21 e d aCT1 t PP7 Be rth 2 = = 2 If Ce m e n t T y p e Inat d ePP7 xAssign Un lSituation o a dto e d a t PP7 Be rth 1 = = 2 && Ce m e n t T y p e In d e x Un lAssign o a d e d ato t PP7 Be rth 2 = = 0 If atUnPP7 Situation CT1 12 at=Situation for CT1 Ce m e n t T y p e In Unloading d e xAssign Un l o a dto e d a t PP7 Be rth 1Delay =PP7 2 && CeCT1 m e n t T y p e In d e x Unloading l o a d e d ato t PP7 Be rth 2 = = 1 Assign Unloading If Interrupt Continue Ce m e n t T y p e Inat d ePP7 x Interrupt Un 12 lSituation o a d e d a t PP7 Be rth 1Delay = = 2 && Ce m e n t T y p e In d e xatUn lContinue o a d12 e d a t PP7 Be rth 2 = = 2 Unloading PP7 Situation 21 If Unloading CT2 atSituation PP7 for for CT2 CT2 Unloading CT2 21 to Unloading CT2 at PP7 21 to Unloading CT2 Assign Assign at Situation at PP7 PP7 Situation Situation 12 at PP7 PP7 Situation at Situation Situation 22 Delay Unloading Interrupt Continue 12 12 22 22 Unloading CT1 at PP7 for CT1 Unloading CT1 Assign to Assign to Assign to at PP7 Situation Assign to Delay Unloading at PP7 Situation Situation 21 Interrupt Continue Branch Delay Unloading Interrupt 21 CT2 Continue 21 CT2 Unloading at PP7 for CT2 Unloading at PP7 for CT2 Unloading CT2 CT2 at PP7 Situation fo r Ce m e n t T y p e (2 ) atUnloading PP7 Situation Situation 02 at PP7 02 Situation at PP7 Situation Situation 20 Ce m e n t T y p e In d e xAssign Un l o a d to Ce m e n t T y p e In d e x Un lAssign o a d02 e d ato t PP7 Be rth 2 = = 2 If 20 e d a t PP7 Be rth 1 == == 01 && Ce m e n t T y p e In d e x Interrupt Un l o a d e d a t PP7 Be rth 1Delay && Ce m e n t T y p e In d e x Un l o a d20 e d a t PP7 Be rth 2 = = 2 Unloading If Continue Branch Ce m e n t T y p e In d e xAssign Un l o a dto e d a t PP7 Be rth 1 = = 2 && Ce m e n t T y p e In d e x Un lAssign o a d e d ato t PP7 Be rth 2 = = 0 If at= =PP7 for CT2 Ce m e n t T y p e In dUnloading e xAssign Un l o a dto eCT2 d a t PP7 Be rth 1Delay 2 && Ce CT2 m e n t T y p e In d e x Unloading Un Assign l o a d e d ato t PP7 Be rth 2 = = 1 Unloading If Interrupt Continue Delay Unloading Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 1 = = 2 && Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 2 = = 2 at PP7 Situation Interrupt f o r Ce m e n t T y p eI (2 Continue at PP7 Situation 02 f ) Assign to Unloading CT2 atSituation PP7 for for CT2 CT2 Assign to Unloading CT2 02 at PP7 Unloading CT2 Unloading CT2 02 Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 1 = = 0 && Ce m e n t T y p e In d e x Un l o a d e d a t PP7 Be rth 2 = = 2 Delay Loading at If Interrupt Branch atUnPP7 PP7 Situation 12e n t T y p e In d e xat Ce m e n t T y p e Inat x Un lSituation o a d e d a t PP7 Be rth 1 =Situation = 1 && Ce m lContinue o a d Situation e d a t PP7 Be rth 2 = = 2 atd ePP7 PP7 Situation If Situation 21 for CT1 Ce m e n t T y p e In Loading d e xAssign Un12 l o CT1 a dto e d at a t PP7 Be rth 1 =PP1 = 2 && Ce m e n t T y p e In d e x Loading Un lAssign o a d12 ed ato t PP7 CT1 at Be rth 2 = = 0 If 21 Ce m e n t T y p e In dPP1 e x UnSituation l o a d e d a t PP7 Be rth 1Delay = = 2 && Ce m e n t T y p e In d e x PP1 Un l o a Situation d21 e d a t PP7 Be rth 2 = = 1 f o r Ce m e n t T y p e I(1 Unloading f ) Situation 01 Ce m e n t T y p e In d e x Interrupt Un l o a d e d a t PP7 Be rth 1 = = 2 && Ce m e n t T y p e In d e x Un lContinue o a d e d a t PP7 Be rth 2 = = 2 01 If Assign to Assign to Ce m e n t T y p e I n d e xAssign Loa d e dto a t P P1 B e rt h 1 = = & & Cefor m eCT2 n t T y p e I n d e x L oUnloading a dAssign e d a01 t PPto 1 Be rt h 2 = = 1 Unloading CT2 at0PP7 If CT2 Delay Unloading Ce m e n t T y p e I n d e x L o a d e d a t P P1 B e rt h 1 = = 1 & & Ce m e n t T y p e I n d e x L o a d e d a t PP 1 Be rt h 2 = = 0 Interrupt Delay Unloading Continue If Interrupt at PP7 Situation Continue Ce m e n t T y p e I n d e xAssign L o a d e dtoa t P P1 B e rt h 1 = = Situation & & Ce m e12 n t T y p e I n d e x Lat o aPP7 dAssign e d aSituation t PPto 1 Be rt h 2 = = 1 If at11PP7 PP7 for CT2 Unloading CT2 Unloading CT2 Unloading CT2 12 at for CT2 Ce m e n t T y p e I n d e x L o a d e d a t P P1 B e rt h 1 = = & & Ce m e n t T y p e I n d e x L o a d e d a t PP 1 Be rt h 2 = = 2 Unloading CT2 12 Delay Loading at If Interrupt Continue Ce m e n t T y p e I nat d ePP7 x L o a d e d a t P P1 B e rt h 1 = = Situation 2 & & Ce m e20 n t T y p e I n d e x Lat o aPP7 d e d aSituation t PP 1 Be rt h 2 = = 1 If at PP7 Situation Situation 22 LoadingSituation CT1 at PP1 for CT1 Loading Assign 20 to Assign to at 20CT1 22 22 PP1 Situation PP1 Situation Delay Unloading Situation 10 Interrupt Continue 10 to 10 to Assign Assign at PP7 for CT2 Unloading CT2 Unloading CT2


Detect

Detect

Detect


0

PP5

PP6

Figure 5: Continuous Part of the MIRP Simulation Model – Supply and Demand Logic

11. SIMULATION OUTPUTS P P7 S i l o s S to c k



0

PP6 Dispose Detect Unloading CT1 at PP7 Dispose Detect Unloading CT2 at Dispose Detect PP7 Unloading CT1 at PP7

PP7 0

This section will present the designed performance measures for simulation model of the MIRP system under study and typical results using hypothetical data. The results of the simulation model will be given in three forms: graphical animation, real-time statePP7report and summary report. P P7 S i l o s S t o c k

Detect P P7 S i l o s S t o c k


0

0

Animation provides visualization of the entities flow throughout the system. Current state of a vessel can easily be observed. The movement of inventoryPP7levels in silos at CS and DCs and also onboard the vessels PP7 are real time. TheProcess typical animation of the MIRP simulation model is given in Figure 6. Detectpresented Inventory Level in to Interrupt Vessel Loading at Central Supply Detect




Delay Unloading

Interrupt at PP7 Situation Assign to Unloading 20 CT2 Interrupt

Dispose Detect Unloading CT2 at PP7 Dispose Detect Unloading CT2 at PP7 Dispose Detect Loading CT1 at PP1

0

0

0 0

Continue at PP7 Situation Assign to Unloading 20 CT2 Continue

Situation 20 at PP7 for CT2at Delay Loading

atLoading PP7 Situation atLoading PP7 Situation CT1 at Situation 21 PP1 forLoading CT1 CT1 at at Central Supply Detect Inventory Level to Interrupt Vessel Process Assign Assign 21 to 21 to PP1 Situation PP1 Situation

Detect

Branch

P P1 S i l o s S t o c k f o r Ce m e n t T y p e (1 ) If If If If If

Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e

Interrupt 11 Assign to Unloading CT2 Interrupt Assign to at PP7 Situation Assign to Interrupt Unloading 21 CT2 Interrupt CT1 at atLoading PP7 Situation Loading CT1 at Assign PP1 Situation 22 to PP1 Situation Interrupt 12 01 Unloading In d ex Loa d e dCT2 a t P P1 I nat d ePP7 xAssign L o aSituation d e dtoa t P P1 I n d e xAssign L o a d e dtoa t P P1 Interrupt I n d e x Interrupt L o 22 a d e d a t P P1 I n dLoading e x L o a dCT1 e d a tat P P1 Loading CT1 at PP1 Situation PP1 Situation 21 10

B e rt h 1 B e rt h 1 B e rt h 1 B e rt h 1 B e rt h 1

Situation 11 Delay Unloading at PP7 for CT2 Delay Unloading Situation 21 Delay Loading at at PP7 for CT2 Delay Loading at PP1 for CT1 Situation 22 PP1 for CT1 Situation 12 Delay Unloading Situation 01 == & & Cefor me n t Ty pe at0PP7 CT2 = = 1 & & Ce m e n t T y p e =Delay = Situation 1 & &Loading Ce m e22 n t at Ty pe =Delay = 1 & &Loading Ce m e n t at Ty pe CT1 = = PP1 2 & & for Ce m en t Ty pe PP1 for CT1 Situation 21 Situation 10

In d e x In d e x In d e x In d e x In d e x

Continue 11 Unloading CT2 Assign to at PP7 Situation Assign to Continue Assign to 21 CT2 Continue Unloading Continue CT1 at atLoading PP7 Situation Loading CT1 at Assign PP1 Situation 22 to PP1 Situation Continue 12 L oUnloading a d e d a01 t PP CT2 1 Be rt h 2 L o a dAssign e d a t PPto 1 Be rt h 2 at PP7 Situation L o a dAssign e d a t PPto 1 Be rt h 2 L o a d Continue e d a22 t PP 1 Be rt h 2 Continue at rt h 2 L oLoading a d e d a t CT1 PP 1 Be Loading CT1 at PP1 Situation PP1 Situation 21 10

Dispose Detect Loading CT1 at PP1 == 1 == 0 == 1 == 2 == 1

0

Detect Inventory Level to Interrupt Vessel Loading Process at Central Supply Detect Inventory Level to Interrupt Vessel Loading Process at Central Supply Assign Assign Assign to to Assign to to Detect

Detect

Branch Branch


P P1 S i l o s S to c k fo r Ce m e n t T y p e (1 ) If II ff II ff I ff P P1 S i l o s S t o c k f o r Ce m e n t T y p eI (1 ) If II ff If If If If

Detect

Detect

Branch

Ce m e n t Ce m me en n tt Ce Ce m e n t Ce m e n t Ce m e n t Ce Ce m me en n tt Ce m e n t Ce m e n t Ce m e n t Ce m e n t

Ty p e T yy p pe e T Ty p e Ty p e Ty p e T Ty yp pe e Ty p e Ty p e Ty p e Ty p e

Branch

P P1 S i l o s S t o c k f o r Ce m e n t T y p e (2 ) If If If If If

Detect

Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e Ce m e n t T y p e

Branch

P P1 S i l o s S to c k fo r Ce m e n t T y p e (2 )

Detect

Branch

If f ) P P1 S i l o s S t o c k f o r Ce m e n t T y p eI (2 If If If If If If

Ce m e n t Ce m e n t Ce m e n t Ce m me en n tt Ce Ce m me en n tt Ce Ce m e n t Ce m e n t Ce m e n t

Ty p e Ty p e Ty p e T yy p pe e T T yy p pe e T Ty p e Ty p e Ty p e

Interrupt Interrupt CT2 at Loading Loading CT1 CT1 at PP1 Situation PP1 Situation Assign to 02 11 In d e x L o a d e d a t PP1 01 In d de e xx Interrupt Lo oa a de ed d a a tt PP1 P P1 In L d P P1 In dLoading e xAssign L o a dCT1 e dtoa tat PP1 P P1 In d e xAssign L o a d e dto a t PP1 Interrupt In dPP1 e x Interrupt L oSituation a d e d a t PP1 P P1 Interrupt Loading CT2 at 01 In d e x L o a d e d a t P P1 I n dLoading e x L o a dCT1 e d a tat Loading CT1 atP P1 I n dPP1 e x L oSituation a d e d a t P P1 PP1 Situation I n dPP1 e xAssign L oSituation a d e dtoa t P P1 12 12 I n d e x Interrupt Loa d e d a t P P1 10 I n d e x L o a d e d a t P P1 Assign to Loading CT1 Assign to at Assign to Interrupt PP1 Situation Interrupt Interrupt Loading 10CT2 Loading CT1 at at PP1 Loading CT1 at PP1 Situation Situation 20 Assign PP1 Situation 21 to Interrupt 11 Assign to at Loading CT1 Interrupt Assign to PP1 Situation Assign to Loading CT2 at Interrupt 11 PP1 Situation Loading CT1 CT2 at 21 to Assign PP1 Situation Interrupt 12 In d e xAssign L o a02 d e dtoa t PP1 In dLoading e x Interrupt L o a dCT1 e d a tat PP1 In dPP1 e xAssign L oSituation a d e dtoa t PP1 Loading CT2 at In d e x L o a d 12e d a t PP1 In dPP1 e x Interrupt L oSituation a d e d a t PP1 CT1 at Loading 22CT2 Assign to PP1 Situation Interrupt 12 21 Loading CT1 at PP1 Situation Assign to 21 to Assign Interrupt Interrupt Loading CT2 at Loading CT2 at PP1 Situation Assign to 20 PP1 Situation Interrupt In dLoading e x L o a02 d CT2 e d a t at PP1 In d e xAssign L o a d e dtoa t PP1 Interrupt In dPP1 e xAssign L oSituation a d e dtoa t PP1 In d dLoading e xx L Lo oa a02 dCT2 ed d a a ttat PP1 In e d e PP1 In d de e xx Interrupt Lo oa ad de ed d a a tt PP1 PP1 In L PP1 Situation In dLoading e xAssign L o a dCT2 e dtoa tatPP1 21 In dPP1 e x L oSituation a d e d a t PP1 In d e x Interrupt L o a d e d a t PP1 12CT2 Loading Assign to at PP1 Situation Interrupt Assign to 12CT2 Loading at Interrupt PP1 Situation Loading CT2 at Assign 22 to PP1 Situation Interrupt 20CT2 at Loading PP1 Situation Assign 20 to Interrupt Loading CT2 Assign to at Interrupt PP1 Situation Loading 21CT2 at PP1 Situation Assign 21 to Interrupt Assign to at Loading CT2 Interrupt PP1 Situation Loading 22CT2 at PP1 Situation 22

Be rth 1 B e rth rth 1 1 Be B e rth 1 Be B e rth 1 Be Be B e rth 1 B Be e rth rt h 1 1 B e rt h 1 B e rt h 1 B e rt h 1 B e rt h 1

Delay Loading Delay Loading at at PP1 for for CT1 CT2 CT1 PP1 02 Situation Situation 11 01 = = 0 && Ce m e n t T y p e =Delay = 1 0 && & &Loading Ce m me en n tt at T yy p pe e = = Ce T 1 && & & Ce m e n t T y p e = = PP1 2 for CT1 1 && & & Ce m e n t T y p e == 2 Loading =Delay = Situation 2 Ce m e01 n t at Ty pe 1 && & &Loading Delay at PP1 form CT2 = 2 Ce e CT1 == = PP1 0 & && & for Ce m en n tt T Ty yp pe e CT1 = = PP1 1 & & for Ce m e12 n t Ty pe Situation Situation 12 = = Situation 1 & & Ce m e10 n t Ty pe =Delay = 1 & &Loading Ce m e n t at Ty pe = = 2 & & Ce m e n t T y p e PP1Loading for CT1 at Delay Delay Loading Situation 10 at Delay Loading at PP1 for CT2 PP1 for CT1 PP1 for CT1 Situation 20 Situation 21 Situation 11 Delay Loading at PP1Loading for CT1 at Delay Situation 11 at PP1 for CT2 Delay Loading Situation 21 PP1 for CT1 CT2 Situation 12 02 at Delay Loading

In d e x Ind de e xx In Ind e x In Ind e x In In Ind e x II n nd de ex x In d e x In d e x In d e x In d e x

Be rth 1 = = 0 && Ce m e n t T y p e In d e x for CT1 Be rth 1 =Delay = PP1 1 &&Loading Ce m e n t at T y p e In d e x Be rth 1 = = Situation 2 && Ce m e12 n t T y p e In d e x formCT2 Be rth 1 =Delay = PP1 2 &&Loading Ce e n t at T y p e In d e x Be rth 1 = = Situation 2 && Ce m e22 n t T y p e In d e x

Delay Loading at PP1 for CT1 CT2 12 at Situation 21 Delay Loading PP1 for CT1 Situation 21 Delay Loading at Delay PP1Loading for CT2at PP1 for CT2 Situation 20 Situation 02 at Delay Loading = = PP1 0 && Ce en t Ty pe formCT2 = = 1 && Ce m e n t T y p e =Delay = Situation 2 &&Loading Ce m e02 n t at Ty pe =Delay = PP1 2 && &&Loading Ce mCT2 en n tt at T yy p pe e = = 0 Ce m e T for == = 1 2 && && Ce Ce m me en n tt T T yy p pe e = Situation formCT2 = = PP1 2 && Ce e21 n t Ty pe =Delay = 2 &&Loading Ce m e n t at Ty pe = = Situation 2 && Ce m e12 n t Ty pe PP1 for CT2 Delay Loading Situation 12 at PP1 for CT2 Delay Loading at Situation 22 PP1 for CT2 Delay Loading Situation 20 at PP1 for CT2 Situation 20 Delay Loading at PP1 for CT2 Delay Loading Situation 21 at PP1 for CT2 Situation 21 Delay Loading at PP1 for CT2 Delay Loading at Situation 22 PP1 for CT2 Situation 22

Continue Continue CT2 at Loading CT1 Loading CT1 at PP1 PP1 Situation Situation to Be rth 2 02 L o a dAssign e d a11 t PP1 01 Continue

Lo oa ad de ed d a a tt PP1 PP 1 Be Be rth rth 2 2 L PP 1 Be rth 2 L o a d e d a t PP1 Loading CT1 at PPto 1 Be rth 2 L o a dAssign e d a t PP1 Assign to Continue L oPP1 a d Continue e d Situation a t PP1 PP 1 Be rth 2 Continue Loading CT2 atrth L a tt CT1 PP Lo oLoading ad de ed d a a01 PP 1 1 Be Be rt h 2 2 at at rt h 2 L oLoading a d e d Situation a t CT1 PP 1 Be PP1 Situation PP1 L oPP1 a dAssign e d Situation a12 t PPto 1 Be rt h 2 12 Loaded a t PP 1 Be rt h 2 10 Continue L o a d e d a t PP 1 Be rt h 2

Assign to at Loading CT1 Assign to Assign to Continue PP1 Situation Continue Loading CT2 at Continue 10CT1 Loading at PP1 Situation Loading CT1 at PP1 Situation 20 Assign PP1 Situation 21 to Continue 11 Assign to Loading CT1 at Continue Assign to PP1 Situation Assign to at Loading CT2 Continue 11 PP1 Situation Loading CT1 at 21CT2 Assign to PP1 Situation Continue 02 to Be rth 2 L o a dAssign e d a12 t PP1 at rth 2 L oLoading a d Continue e d a t CT1 PP1 Be L oPP1 a dAssign e d Situation a t PP1 rth 2 to Be atrth 2 L oLoading a d e d a t CT2 PP1 Be L oPP1 a d Continue e d Situation a12 t PP1 Be rth 2 Loading CT1 at CT2 22 Assign to PP1 Situation Continue 12 21 Loading CT1 at PP1 Situation Assign to 21 to Continue Assign Loading CT2 at Continue PP1 Situation Loading CT2 at Assign 20 to PP1 Situation Continue L o a d e d a02 t PP1 Be rth 2 Loading CT2 atrth 2 to Be L o a dAssign e d a t PP1 L oPP1 a dAssign e d Situation a t PP1 Be rth 2 Continue to Be Lo oa ad de ed d a a02 PP1 Be rth 2 2 L tt CT2 PP1 Loading atrth Continue Lo oa ad de e d a a tt PP1 PP1 Be Be rth rth 2 2 L d PP1 atrth 2 L oLoading a dAssign e d Situation a t CT2 PP1 to Be L oPP1 a d e d Situation a21 t PP1 Be rth 2 Continue Loade d a t PP1 Be rth 2 12CT2 Loading Assign to at PP1 Situation Continue Assign to 12CT2 at Loading Continue PP1 Situation Loading CT2 Assign to at 22 PP1 Situation Continue

== == = = == == == = == = == == == ==

2 1 2 0 1 2 1 1 0 1 2 1

Dispose Detect Detect Dispose Loading CT1 CT2 at at Loading PP1 PP1 Dispose Detect Loading CT1 at PP1

PP1

0 0

0

PP1 Dispose Detect Loading CT2 at PP1 == 2 == 2 == 0 == 1 == 2

0

PP1

Figure 6: AnimationPP1of MIRP Simulation Model

Be rth 1 Be rth 1 Be rth 1 Be rth rth 1 1 Be Be rth rth 1 1 Be Be rth 1 Be rth 1 Be rth 1

In d e x In d e x In d e x In d de e xx In In d de e xx In In d e x In d e x In d e x

== == == == = = == = = == == ==

2 2 0 1 2 2 2 0 1 2

Dispose Detect Loading CT2 at PP1 Dispose Detect Loading CT2 at PP1

0 0

At current stage of this research, the designed performance measures are the real time states of inventory levels at CS and DCs and VRVs which contain DC destination, cargo, berthing position at CS and DC, and duration of each activity. Table 3 shows the element of VRV which will be recorded during the simulation run. A Microsoft® Excel file (i.e. MIRP Simulation Results.xls) has been defined to record the performance measures of the MIRP simulation model as the simulation runs progress. The examples of inventory status and VRVs recorded from the MIRP simulation result can be seen in Figure 7. 20CT2 at Loading PP1 Situation Assign 20 to Continue Loading CT2 Assign to at Continue PP1 Situation Loading 21CT2 at PP1 Situation Assign 21 to Continue Assign to at Loading CT2 Continue PP1 Situation Loading 22CT2 at PP1 Situation 22

396


Table 3: Vessel Round Voyage Performance Measures

Element

Unit

Element

Unit

Vessel

-

EDS Time at DC

Hours

PP (DC) Assigned

-

Unberthing Position – DC BOSV

Hours

Cement Type

-

Sail Back

Hours

Cargo Delivered

Tons

CS EOSV – Anchorage Position

Hours

Sail Go

Hours

Waiting for Berthing at CS

Hours

DC EOSV – Anchorage Position

Hours

Berth Seized at CS

Waiting for Berthing at DC

Hours

Anchorage Position – Berth at CS

Hours

IDS Time at CS

Hours

Berth Seized at DC

-

-

Anchorage Position – Berth at DC

Hours

Loading Time

Hours

IDS Time at DC

Hours

Expected Loading Time

Hours

Unloading Time

Hours

Unexpected Idle Time during Loading

Hours

Expected Unloading Time

Hours

EDS Time at CS

Hours

Unexpected Idle Time during Unloading

Hours

Unberthing Position – CS BOSV

Hours

Figure 7: Microsoft® Excel Sheets for Simulation Results – Real Time States

Arena® has capability to create a summary report of simulation results. This summary report provides useful statistics regarding to the performance measures of the MIRP simulation model. Some of these statistics can be seen in Figure 8. As examples, from this summary report utilization of each resource and its idle time can be identified.

397


Figure 8: Arena® Summary Report of MIRP Simulation Model Regarding to further development, this MIRP simulation model has already had the capability to provide support for decision making in ship scheduling and to some extent to support integration with the optimization model. This simulation model also has capability to provide information regarding to calculation of inventory and transportation costs and also calculation of unexpected delays due to waiting time for berthing, DCs silos full, and CS silos empty. This information will be useful to determine other performance measures for the MIRP simulation model. Beforehand, this MIRP simulation model is needed to be validated using the data collected from the real system. 12. CONCLUSION A MIRP or marine shipping is characterized by the present of many stochastic variables and complex interactions between the system entities in its practice, which often preclude the possibility of obtaining an analytical solution. This paper presented a simulation approach as solution methodology for a MIRP with particular application to cement industry. The simulation approach used is a combined discrete and continuous model. The simulation model in this study has been designed and developed thoroughly to emulate the complexity of the real system of MIRP. This MIRP simulation model is taking into account the presence of stochastic elements and complex interactions between the system entities. The simulation model has shown an encouraging result as it performs as expected. This MIRP simulation model has been designed and developed systematically, so that the model can be easily expanded or adjusted to different size of system entities for example number of CSs, DCs, berths, vessels, and products. The simulation model developed has already had the capability to provide support for decision making in ship scheduling. Even though the effectiveness of the model in term of optimization has not been established yet, current achievement promises further development of the MIRP simulation model to the next stage as has been planned, which is the integration with the optimization model. 13. ACKNOWLEDGEMENTS Support for this research is provided by the Directorate of Higher Education, Ministry of National Education, Republic of Indonesia and the University of Bradford, the UK. The authors gratefully acknowledge Mr. Afrizal and Mr. Wieky Gusta who provided data for case study and helped in validation of the MIRP simulation model.

398


14. REFERENCES [1] Ronen, D., Marine inventory routing: shipments planning. Journal of the Operational Research Society, 2002. 53(1): p. 108-114. [2] Campbell, A.M. and M.W.P. Savelsbergh, A decomposition approach for the inventory-routing problem. Transportation Science, 2004. 38(4): p. 488-502. [3] Brown, G.G., G.W. Graves, and D. Ronen, Scheduling ocean transportation of crude oil. Management Science, 1987. 33(3): p. 335-346. [4] Ronen, D., Ship scheduling: the last decade. European Journal of Operational Research, 1993. 71(3): p. 325-333. [5] Christiansen, M., et al., Maritime transportation, in Handbooks in Operations Research and Management Science, Vol. 14: Transportation, C. Barnhart and G. Laporte, Editors. 2007, Elsevier B.V.: Amsterdam. p. 189-284. [6] Baita, F., et al., Dynamic routing-and-inventory problems: a review. Transportation Research Part A: Policy and Practice, 1998. 32(8): p. 585-598. [7] Al-Khayyal, F. and S.-J. Hwang, Inventory constrained maritime routing and scheduling for multicommodity liquid bulk, part i: applications and model. European Journal of Operational Research, 2007. 176(1): p. 106-130. [8] Christiansen, M., K. Fagerholt, and D. Ronen, Ship routing and scheduling: status and perspectives. Transportation Science, 2004. 38(1): p. 1-18. [9] Shih, L.-H., Planning of fuel coal imports using a mixed integer programming method. International Journal of Production Economics, 1997. 51(3): p. 243-249. [10] Liu, C.-M. and H.D. Sherali, A coal shipping and blending problem for an electric utility company. Omega, 2000. 28(4): p. 433-444. [11] Vukadinović, K., D. Teodorović, and G. Pavković, A neural network approach to the vessel dispatching problem. European Journal of Operational Research, 1997. 102(3): p. 473-487. [12] Law, A.M. and W.D. Kelton, Simulation Modeling and Analysis. 3rd ed. 2000, Boston: McGraw-Hill. [13] Henderson, S.G. and B.L. Nelson, Stochastic computer simulation, in Simulation, S.G. Henderson and B.L. Nelson, Editors. 2006, Elsevier: North Holland. p. 1-18. [14] Fu, M.C., Optimization for simulation: theory vs. practice. Informs Journal on Computing, 2002. 14(3): p. 192-215. [15] Tekin, E. and I. Sabuncuoglu, Simulation optimization: a comprehensive review on theory and applications. IIE Transactions, 2004. 36(11): p. 1067-1081. [16] Medaglia, A.L., S.-C. Fang, and H.L.W. Nuttle, Fuzzy controlled simulation optimization. Fuzzy Sets and Systems, 2002. 127(1): p. 65-84. [17] Paolucci, M., R. Sacile, and A. Boccalatte, Allocating crude oil supply to port and refinery tanks: a simulation-based decision support system. Decision Support Systems, 2002. 33(1): p. 39-54. [18] Cheng, L. and M.A. Duran, Logistics for world-wide crude oil transportation using discrete event simulation and optimal control. Computers & Chemical Engineering, 2004. 28(6-7): p. 897-911. [19] Boykin, R.F. and R.R. Levary, An interactive decision support system for analyzing ship voyage alternatives. Interfaces, 1985. 15(2): p. 81-84. [20] Ronen, D., Cargo ships routing and scheduling: survey of models and problems. European Journal of Operational Research, 1983. 12(2): p. 119-126. [21] Kelton, W.D., R.P. Sadowski, and D.T. Sturrock, Simulation with Arena. 4th ed. 2007, New York: McGraw-Hill.

399


A STUDY AND IMPROVEMENT OF TEST AND MEASUREMENT INDUSTRY‟S SUPPLY CHAIN SYSTEM Yea Dat Chuah1 and Mike Miles2 1

Universiti Tunku Abdul Rahman, Faculty of Engineering and Science, Kuala Lumpur, Malaysia e-mail: [email protected] 2 University of Plymouth, School of Marine Science and Engineering, Plymouth, UK e-mail: [email protected]

ABSTRACT The objectives of the study are to understand the constraints and the areas that have to be improved in selected test and measurement manufacturers‘ supply chains system and developed a better supply chain model that suitable for test and measurement product manufacturer. This paper includes the analysis outcome of environmental influences using PEST (Political, Economic, Social and Technology) and company position assessment using SWOT (Strengths, Weaknesses, Opportunities and Threats) of existing supply chain management system for selected test and measurement manufacturing companies and proposed solutions to overcome the problems. 1.

INTRODUCTION The test and measurement manufacturing industries use high mix, low volume manufacturing methods to supply measuring instruments. Most of the parts used to build measuring instruments are unique and have low demand. Due to this reason, test and measurement manufacturers always have difficulty sourcing suitable parts and finding suppliers. Also, the test and measurement manufacturers face difficulties finding alternative sources if the existing supplier supplies parts that have high reject rate. Substantial amounts of research on supply chain have been carried out over the last twenty years. Researchers focus on different aspects of the system or sub-system. For example, Kumar et al identified four sources of uncertainty in supply chain systems [1]. Suarez et al. and Gerwin proposed frameworks that can be used for implementing and managing supply chain flexibility [2, 3]. Stewart performed a supply chain performance benchmarking study and proposed the importance of integration between functions, suppliers and customers [4]. Beach et al. presented a consolidation model for manufacturing flexibility integrating environment uncertainty [5]. Narain et al. outlined the link between manufacturing, marketing and organisation strategies [6]. Zhang et al. created a model that applies competence and capability theory to value supply chain flexibility [7]. Chan et al. proposed a performance measurement method for supply chain management [8]. Noonan et al. created value-focused strategies to build responsive contract manufactures in a supply chain system [9]. Kapuscinski et al. have performed a study on Dell supply chain and developed an inventory model for Dell‘s supply chain [10]. Foster performed a comparison of 18 companies and lists the reasons why their supply chains worked well and concluded that excellent supply chains will be found in companies that have a clear business strategy enabled by a complementary operating model aimed at achieving a balanced set of operating objectives [11]. Jaafari presented a project health check methodology by using a systems approach, which is useful to provide a graphical picture of the health level of the system at the time of assessment [12]. Most of the research outcomes mentioned above have an academic and industrial focus on overall supply chain systems but none of the researchers performed detailed studies on a high mix, low volume industry particularly in the test and measurement industry. The purpose of this work is to fill the knowledge gap in supply chain system research by providing a supply chain model for the test and measurement industry.

2.

METHODOLOGY Primary data about the supply chain system in test and measurement industry has obtained from Managers and engineers who work in test and measurement company and its contract manufacturer in Malaysia. Secondary data related to supply chain systems research outcome, government policies, companies businesses performance, social economy situations has obtained from various articles, journals, government

400


official website and books. The data collected has been analysed by using PEST (Political, Economic, Social and Technology) to identify environmental influences on the test and measurement industry, particularly in Malaysia. This followed by company position assessment using SWOT (Strengths, Weaknesses, Opportunities and Threats) to identify the company position compare to other test and measurement companies in the world. The finding of PEST and SWOT analysis were used to proposed supply chain system for test and measurement company particularly in Malaysia and probably can also be used as reference by other test and measurement companies in other countries. 3. 3.1

PEST ANALYSIS Political Factors The studied company‘s manufacturing centre is located in Penang, Malaysia. The Malaysian government provides a good business environment with opportunities for growth and profits that have made it an attractive manufacturing and export base in the region. According to the Malaysian Industrial Development Authority, the Malaysian government offers a wide range of tax incentives for manufacturing projects under the Promotion of Investments Act 1986 and the Income Tax Act 1967. Among the main incentives are the Pioneer Status, Investment Tax Allowance, Reinvestment Allowance, Incentives for High Technology Industries and Incentives for Strategic Projects and Incentives for the Setting-up of International/ Regional Service-based Operations [13]. Employer-employee relationships in Malaysia are harmonious and government also strongly discourages strikes. Thus, strikes and industrial actions are extremely rare [14].

3.2

Economic factors Based on Table 4-2 from the official website of the Ministry of Finance Malaysia, the Gross Domestic Product (GDP) growth of Malaysia was 7% in 2010. The GDP of Malaysia is expected to grow at 5% to 6% in the subsequent years [15]. The continuing growth of the Malaysian economy will increase its domestic buying power and hence it is expecting to enjoy better social and economic stability.

Table 4-2: Malaysia Gross Domestic Product by Kind of economy activity at constant 2000 prices (RM million) (Source: GDP by sector, official website of Ministry of Finance, Malaysia [15])

Services Manufacturing Agriculture Mining and quarrying Construction

2010

2011

6.5 10.8 3.4 1.0 4.9

5.3 6.7 4.5 2.9 4.4

The International Monetary Fund forecasts the global recovery is continuing but its strength is not yet assured [16]. Asia is leading the global recovery whereas North America recovery is moderating in the face of debt and continued uncertainty. In Europe, the recovery is gradual and uneven due to unsustainable policies in some member countries, the sovereign debt crisis in the spring 2010 erupted before the euro area‘s recovery could gain traction. This indicates that buying power will low in US, Japan and European region but it will rise in China, India, other countries in Asia and South East Asia (Figure 1). The test and measurement companies can find new market opportunities in these fast growing countries by performing marketing surveys to identify potential market demand and consumer behaviour in the regions.

401


Figure 1: Average Projected Real GDP Growth during 2010–11 (Source: International Monetary Fund staff calculation [16]) 3.3

Social Factors Test and measurement products are used mainly in the telecommunication industry, life-sciences research, electronic components performance measurement, education and the defence industry. Religions, culture, languages used and gender roles have little influence on demand or use of test and measurement products. Customer perception of product pricing is also not the major factor but the social influences such as consumers‘ perception of product quality and reliability will have a direct effect on its business. In view of this factor, test and measurement companies such as Rohde-Schwarz from Germany and Agilent Technology from US are putting great efforts on their product quality, precision and reliability in order to gain customers‘ loyalty and attract new customers [17,18]. Malaysia is multiracial and multi-ethnic, but each ethnic group has been able to retain its own fundamental beliefs and traditions. Rashid has performed a study to obtain a better understanding of the relationship between organizational culture and attitudes towards organizational change in Malaysia, the result indicated that organizational culture was associated with attitudes toward organizational change. Overall, this study showed that the respondents have a positive or strongly positive attitude toward change [19]. Malaysian can get along with each other because they have common values of politeness, gentleness, loving, obedience are accommodating and humble. Malaysians, regardless of ethnic group, generally prefer to work with people who were easy to relate to and understand their culture, traditions and sensitivities. The increased social health awareness and medical care will increase the demand for life science test instruments. For example, Agilent Technologies has strong Forth quarter earning results gained from the high demand of bioanalytical instruments with robust growth in both the life sciences and chemical analysis markets [18]. It is believed that the increase of living standards and health consciousness of the people in this region will encourage life sciences research activities which in turns need more test and measurement products.

3.4

Technological Factors Global technology advancement has provided ample business opportunities for test and measurement companies. This is reflected in most of the test and measurement companies‘ annual financial reports. Among them are Agilent Technologies and Tektronix. There is a continuous improvement of global technology in telecommunication and networking, health sciences and various types of scientific and engineering research all needing test and measurement equipment. Test and measurement companies such as Agilent Technologies have to maintain the product quality and measurement accuracy of existing products. At the same time it should focus on research and development activities to produce instruments that have a higher measuring capability to fulfil future requirements. Malaysia is able to provide the investor with a diligent, disciplined, educated and trainable labour force. The Malaysian government has taken measures to increase the number of engineers, technicians and other skilled personnel graduating each

402


year from local and foreign universities, technical training institutions to cater for the demand for technically trained workers. [13]

4.

SWOT ANALYSES After performing PEST analysis, SWOT (Strength, Weaknesses, Opportunities and Threats) analysis will be applied to selected US based test and measurement company in Penang, Malaysia in order to evaluate the organisation‘s strengths and weaknesses in relation to environmental opportunities and threats.

4.1

Strengths Data obtained from selected test and measurement company in Malaysia indicated that the company is producing quality products that have less than 2 percent return due to quality and reliability problems. Also, the company has good track record for its after sales service. This has helped the company to retain existing customers and promote quality products. The relocation of the main manufacturing plant from the US to Malaysia enables a reduction in product unit cost and increases its global market competitiveness. The company‘s performance report indicates that the business performance is improving. Top management has implemented an effective corporate strategy that will increase market share; especially in the Life Sciences industry. The company is giving rewards to employees who are able to provide new ideas that will result in improved product quality, reduced cost and improved business performance. Promoting an innovation culture among the employees is one of the key factors that has helped the company to become the market leader in the test and measurement industry.

4.2

Weaknesses Producing high mix, low volume products causes the company difficulties in obtaining alternative suppliers and it cannot fully automate the product assembly line. Therefore, it needs to recruit and retain highly skilled workers who are able to work on a wide product range. Due to the high frequency performance of the products‘ nature, the products require complex and long testing times and most of the raw material reliability can only be tested after its have been assembled in the microcircuit.

4.3

Opportunities The recent developments of the life sciences and research industry have provided a huge market opportunity to the company. The company could also develop low cost instruments to increase its share of the low Radio Frequency (RF) test and measurement market. The International Monetary Fund forecasts China and India to have a large positive GDP growth [16] and this will increase the demand of the test and measurement products in telecommunication and health care industries. Also, outsourcing activities will further reduce the manufacturing cost and increase profits.

4.4

Threats With the increase of global competition in the test and measurement market customers become more demanding of products‘ performance, quality and price. The unpredictable new Asia Pacific economic region; the increase of material cost of about 10 % per annum and inconsistent material quality also become threats for the company. Also, the lost of experienced workers would affect its manufacturing operations.

4.5

Summary of the finding The summary of the analysis finding are listed below: a. External factors that are located outside the supply chain system boundary such as political factors, social economic factors and technological factors in Malaysia provide a positive effect on the test

403


b.

c. d. e. f. g.

5.

and measurement company supply chain system. Any proposed supply chain system must be flexible enough to face unpredictable global market demand. The selected test and measurement company in Malaysia is able to produce high quality test and measurement products but cost reduction projects such as outsourcing activities may affect its product quality level. It is necessary to set up an effective suppliers‘ and contract manufacturers‘ product quality monitoring system. Manufacturers face difficulty sourcing parts and finding suppliers because most of the raw materials are unique and the numbers required are low. Also, the manufacturers face difficulties finding an alternative source if the existing supplier supplies parts that have high reject rate. Material shortages or quality problems have impacted on production output. It is necessary to establish effective suppliers and contract manufacturers‘ management and procurement system to overcome material shortages and quality problems. Customers become more demanding of the performance, quality and price of the products. An effective supply chain system that fully integrates sub-systems has to be created to cater for this challenging business environment. A flexible, lean, agile and continuous improvement culture is essential in high mix low volume supply chains in order to accommodate large product ranges and rapid changes in global market demand.

PROPOSED SUPPLY CHAIN MODEL The proposed high mix, low volume supply chain model for test and measurement should have following characteristics: agility, flexibility, team work culture, waste reduction or Lean, integration between subsystems and continuous improvement of the system (Figure 2). These characteristics are important because high mix, low volume manufacturers require a flexible materials sourcing system to obtain its unique raw materials from suppliers that may be located in different geographical areas while maintaining low costs and consistent quality. This industry also needs an effective manufacturing and logistics system to produce and supply products to customers from all parts of the world rapidly and with the minimum of transportation cost and without product damage in transportation. An agile supply chain is able to respond to the unexpected global demand and sudden change of consumers‘ behaviour in a short time frame. A team work culture in the supply chain can promote supply chain sub-system integration. Lean and continuous improvement is important in order to maintain it competitiveness and reduce operating costs. A Supply chain is a complex open system that can consist of various sub-systems. Every sub-system must have its own departmental strategy that is aligned with company goals to fulfil customers‘ requirements and fully integrate with other sub-systems to maximise its performance (Figure 3). It is important to ensure that the sub-systems have these characteristics and are aligned with the company goal.

Figure 2: Supply chain sub-systems, systems‘ characteristic and systems‘ goal.

404


Figure 3: Proposed Supply Chain Model for Agilent Technologies 5.1

New material sourcing

One of the problems found in the existing supply chain is the difficulty of sourcing parts and finding suppliers because most of the raw materials are unique and the numbers required are low. Therefore it is necessary to develop an effective sourcing strategy to eliminate or reduce the risk. Figure 4 shows a proposed sourcing system. The material supply engineering team consists of a small group of people who examine the sourcing options from material engineers for different product categories such as metal stamping parts, plastic injection moulding parts, casting parts, solid metal housing, epoxies, solder paste, PCB, IC, capacitor/diode, gold wire and gold mesh. The material engineers obtain material specifications from new, local or overseas suppliers and work with product mechanical/ electrical engineers who compare those specifications with those required. If the recommended suppliers‘ specifications are acceptable, the material supply engineers will work with quality engineers and process mechanical engineers to study these suppliers‘ process capability, reliability, quality control systems and technology capability to meet the company‘s existing and future raw product requirements. Suppliers that meet all the process and engineering requirements must also be willing to build a close relationship with the company to work together in developing new specifications that meet customer requirements. The list of potential suppliers will then be submitted to a sourcing team to further evaluate their abilities to meet basic company rules in terms of prices, rebates, breadth of offering, delivery frequency and options, and order entry systems before awarding the business to the supplier.

Figure 4: Proposed sourcing system. 5.2

Contract manufacturer and suppliers In this highly competitive market, test and measurement manufacturers who used to be original equipment manufacturers can no longer compete as autonomous entities. As the industrial giants trim down, most of the products which need low or moderate assembly technology capability are manufactured or assembled

405


by contract manufactures. Contract manufacturers (CM) are different from the traditional manufacturing model in that they build a variety of products for various clients, whereas original equipment manufacturers (OEM) build only proprietary products [20, 21,22]. To achieve optimum supply, the company which practices product assembly outsourcing to the contract manufacturers must ensure the contract manufacturers is fully included in the development of new products. According to Noonan et al., understanding the new complexities of value criteria (For example, high quality and low cost) will help the contract manufacturer choose the correct strategy configuration for achieving continuous competitive advantage through creating value and contributing to the supply chain in a manner that is perceived as indispensable by its customer‘s supply chain managers [9]. The company has to realize that the contract manufacturer is a critical supply chain player and the contract manufacturer has to fully understand the company‘s strategy for achieving value criteria. Strategies such as expansion, cutting back, or moving towards a virtual enterprise greatly influence the value criteria of the company especially in its purchasing department [9]. The contract manufacturer that does not understand its customer‘s plans and how they influence the overall supply chain will adversely affects the entire supply chain system. To achieve maximum value, the company must ensure the contract manufacturers fully understand their product requirements and that they work closely with the company to maximize supply chain efficiency and produce quality products. Co-ordination of supply chain performance can only be achieved when contract manufacturers and suppliers fully understand the customer value criteria [23, 24] .It is also important that the company understands its contract manufacturers‘ value criteria which may be different from its own. Also, the suppliers who stay on the list of contract manufactures have to create, exploit and sustain competitive advantage. Matching resource commitment with changing value opportunities has become the priority. Promoting trust between the contract manufacturer, suppliers, procurement team, materials engineers and design engineers who are involved in new material sourcing is important in order to promote the information and knowledge flow in the whole supply chain system, [25]. This can be done by sharing knowledge with suppliers about the ultimate customers‘ behavioural patterns and their requirements. The proposed steps to further improve Agilent Technologies and contract manufacturers/suppliers‘ relationship are listed below: a. b. c. d. e.

Develop mutual agreement on product specifications. Build partnership relationship with suppliers to achieve win-win situation. Constantly provide information on the customers‘ feedback and new requirements. Provide monthly raw material forecasts to ease suppliers‘ production scheduling. Share knowledge between suppliers and the company‘s Process and Product Engineering team for continuous product quality and process improvement. f. Work together to develop a cost saving strategy and develop new specifications in order to meet customer requirements. g. Reduce and eliminate exploitation policies which can affect suppliers and company relationship.

5.3

Material quality solving system Unpredictable material defects and variations in the percentage of the defects will affect the whole supply chain. Therefore, an effective quality monitoring system is important. Figure 5 shows the suggested process flow for solving a product quality problem. First of all, a particular product that has a quality issue is identified. Assembly process mapping has to be prepared for the selected product. A detailed analysis should be carried out for every stage of the assembly process to make sure that the product quality problem is not due to the assembly process. After that the product‘s Bill of Material (BOM) list should be studied to identify the possible components that could cause the quality problem. These suspected components will then be assembled on the trouble shooting circuit to identify the particular component that causes the problem. After this stage, the original material manufacturer has to be identified and a meeting should be held to identify the root cause. The supplier has to provide a complete material quality issue report and propose countermeasures. Company Quality Assurance (QA) engineers have to monitor closely the quality of the new incoming lot for a fixed period of time. If the problem persists, a QA engineer should work closely with the procurement team to find an alternative supplier or work closely with both the process and product engineering teams of both organisations to find the solution. Finally, higher management of both

406


organisations have to get involved in this issue to discuss, negotiate and revise the material cost because it has probably impacted on reputation and on product quality; also engineering resources will have been used to solve the problem.

Figure 5: Process flow of solving a product quality problem.

6.

CONCLUSIONS AND RECOMMENDATION The main objective of this research is to create a supply chain model that is suitable for a high mix, low volume industry particularly the test and measurement industry. The model that has been created has been developed and verified by using comments from experienced managers in the test and measurement industry and its contract manufacturer. The proposed supply chain which has characteristics such as agility, flexibility, a team work and continuous improvement culture, a focus on inter-departmental integration and lean should be able to address the existing global test and measurement market‘s requirements. An agile and flexible supply chain is able to respond to changes of the customer expectations without excessive costs, time, organizational disruptions or performance losses. Although the proposed model is based on the analysis results from the selected test and measurement company and its contract manufacturer in Malaysia. The proposed supply chain model can also be applied in other high mix, low volume product manufacturers. The application of cell manufacturing systems, total quality management, world class manufacturing thinking such as 5S, continuous improvement, TPS (Toyota Production System), Lean and JIT (Just in time) can be integrated into the proposed supply chain system to suit various manufacturing functions and requirements to maximise the system flexibility, efficiency and effectiveness. An application of world class manufacturing approaches to a supply chain can help the company to adapt to the highly competitive and rapidly changing global market.

7.

REFERENCES [1] Kumar, V. and Kumar, U. (1988), ―Five years into measuring manufacturing flexibility‖, paper presented at APOR‘88 Meeting, Seoul [2] Suarez, F. F., Cusumano, M. A. and Fine, C.H. (1991), ―Flexibility and performance: a literature critique and strategic framework‖, working paper, Sloan of Management, MIT, Cambridge, MA: 50-59 [3] Gerwin, D. (1993), ―Manufacturing flexibility: a strategy perspective‖, Management Science, 39(4): 395-410 [4] Stewart, G. (1995), ―Supply chain performance benchmarking study reveals keys to supply chain excellence‖, Logistics Information Management, 8(2): 38-44 [5] Beach, R., Muhlemann, A. B., Price, D. H. and Sharop, J. A. (2000), ―A review of manufacturing flexibility‖, European Journal of Operation Research, 122:41-57 [6] Narain, R., Yadav, R., Sarkis, J. and Cordeiro, J. (2000), ―Strategic implementations of flexibility on manufacturing systems‖, International Journal of Agile Management Systems, 2(3):202-13 [7] Zhang, Q., Vonderembse, M.A. and Lim, J. (2002), ―Value chain flexibility: a dichotomy of capability‖, International Journal of Production, 40(3):561-83 [8] Chan, Felix T. S. and Qi, H.J. (2003), ―An innovative performance measurement method for supply chain management‖, Supply Chain Management: An International Journal, 8(3): 209-223

407


[9] Noonan, J. and Wallace, M. (2004), ―Building responsive contract manufacturers through valuefocused strategies‖, Supply Chain Management: An International Journal, 9(4): 295-302 [10] Kapuscinski, R., Zhang, R.Q., Carbonneau, P. and Bill Reeves, R. M. (2004) ―Inventory Decisions in Dell‘s Supply Chain‖, Interfaces, 34(3): 191-205 [11] Foster, T. A. (2005), ―In Pursuit of Supply-chain Excellence‖, Global Logistics & Supply Chain Strategies January, 2005 http://www.glscs.com/archives/01.05.mit.htm?adcode=10 [Date Accessed 1st August 2008] [12] Jaafari, A. (2007), ―Project and program diagnostics: A systemic approach‖, International Journal of Project Management Systems, 25:781-790 [13] Malaysian Industrial Development Authority (MIDA) http://www.mida.gov.my/ [Date Accessed 24th February 2011] [14] World Bank: 2010 Investment Climate Statement Malaysia http://www.state.gov/e/eeb/rls/othr/ics/2010/138774.htm [Date Accessed 24th December 2011] [15] GDP (Gross Domestic Product) By Sector, Ministry of Finance Malaysia http://www.treasury.gov.my/pdf/ekonomi/dataekonomi/2010/forecast/GDPbysector2010_2011.pdf [Date Accessed 24th February 2011] [16] International Monetary Fund: Country and Regional Perspectives (2011) http://www.accessbankplc.com/Library/Documents/2010%20Bankers%20Committee%20Conference %20Resources/Economic%20Development/IMF%20Outlook%20Report%202010.pdf [Date Accessed 26th February 2011] [17] Rohde & Schwarz http://www.rohdeschwarz.com/www/dev_center.nsf/frameset?OpenAgent&website=com&navig=/www/dev_center.nsf/ html/nav,10,12&content=/www/prod_center.nsf/startpage/_10_12.html [Date Accessed 25th February 2011] [18] Agilent Technologies Reports Forth Quarter 2010 Results http://www.agilent.com/about/newsroom/presrel/2010/12nov-gp10024.html [Date Accessed 26th February 2011] [19] Rashid, M. Z. A., Sambasivan, M. and Rahman, A. A. (2003) ―The influence of organizational culture on attitudes toward organizational change‖, The Leadership & Organization Development Journal, 25(2): 161-179 [20] Baker, H (1999), ―Choosing your contract manufacturer‖, Global Cosmetic Industry, May: 48-9 [21] Engardio, P. (2000), ―The barons of understanding‖, Business Week, 28 August: 177-8 [22] Hannon, D. (2001), ―The CM needs to offer more‖, Purchasing, 25 January: 76-8 [23] Holden, N.J. (2002) Cross-cultural management: A knowledge management perspective. Pearson: Harlow [24] Guaspari, J. (1988), The customer Connection: Quality for the Rest of Us, AMACOM: New York [25] Bhote, K.R. (1991), Next Operation as Customer, AMACOM, New York, NY [26] Kwon, Ik-Whan G. (2004) ―Factors Affecting the Level of Trust and Commitment in Supply Chain Relationships‖, The Journal of Supply Chain Management: A Global Review of Purchasing and Supply Supply, Spring: 4-14

408


CONTROL SYSTEM MANIPULATION OF PROFIT IN SUPPLY CHAINS A.S.White1, M. Karamanoglu2, R. Gill3, M.Censlive4 1

School of Engineering and Information Systems, Middlesex University The Burroughs, Hendon, London NW4 4BT, UK Email1: [email protected] 2 School of Engineering and Information Systems, Middlesex University The Burroughs, Hendon, London NW4 4BT, UK Email2: [email protected] 3 School of Engineering and Information Systems, Middlesex University The Burroughs, Hendon, London NW4 4BT, UK Email3: [email protected] 4 School of Engineering and Information Systems, Middlesex University The Burroughs, Hendon, London NW4 4BT, UK Email4: [email protected] ABSTRACT This paper describes the investigation of production profit for an APVIOBPCS, automatic pipeline, variable inventory and order based production control system using electronic RAM production. Results show that a PID controlled system will allow payback sooner that a proportionally controlled one. Changing the desired inventory affects the point at which cash inflow starts, hardly changing the costs. A higher value of desired inventory level also allows earlier payback. PID control allows profit to be generated earlier and at a larger rate than for proportional control. Variable desired inventory also allows earlier payback. Production constraints cause later payback and the effect of sales data smoothing also has a deleterious effect. The principle conclusion is that by using PID control of the inventory system profits can be maximized and can be as effective as substantial manufacturing cost reductions. Keywords: Profit, Inventory system, PID control, APVIOBPCS, RAM 1.

INTRODUCTION Business organisations primary goal is to achieve maximum profit using normal marketing strategies and conventional economic models. Revenue is a function of price and demand in most of the time and ignores the impact of the time and service level on the product price on customer demand. In supply chains the cost of materials and manufacture and distribution inventory counts of between 70 and 80% of the value of the product. In Supply chain designs for suitable for efficiency gains and often they use optimised solutions to maximise the profit of the Supply chain revenue. Currently Eastern examine how everyday low pricing or high low pricing strategies give clear advantage in the Supply chain environment and maximising profit. In the current environment businesses must improve their efficiency in particular their supply chains in autumn rain can maintain competitive advantage and stop profitability is their primary objective and simple cost-cutting or revenue enhancement will not guarantee overall best supply chain performance (Economist intelligence unit). The principle of lean manufacturing just-in-time inventory management helped Toyota leapfrog General Motors as the world's largest beer, infection and they also enabled Wal-Mart to come to the forefront of global retailers. Not simply enough to maintain and improve or streamline your global supply chain to eliminate excess costs companies must apply sophisticated analysis to make their supply chains more responsive to customer demand rather than letting availability of the Supply drive chain product-liability and profitability linked to marketing keeps inventions and the boost profit margins. Dell has slipped from the leading position as the leading PC maker Dell's demand shaping has been named primary given inventory levels are low and its manufacturing the rather than maximising profit by building and promoting its highest margin products (Economist intelligence unit 2009). Benita Beamon [3] in her benchmark review of current industry practice outlined a set of criteria used by researchers to categorise supply chain design methods. Many of these include cost as pivotal criteria. She categorises supply chain performance measures involving economics as: Cost minimisation, Sales

409


Maximisation, Profit maximisation (Shen 2006, Kathmandu 2009, Satyaveer & Jean-Marie 2003, Kahn & Easton), Inventory investment minimisation and Return on investment maximisation (Qian 2009) These have been investigated by a number of authors with no definitive solution. It is only in the last few years that analysts have paid any attention to marketing trends in the sales problem [5]. In order to make any realistic offers to customers they would need to know the implications of their choices on their inventory strategy. The purpose of this paper is to explore the development of a model that can be used in real-time to compute profit and costs of various inventory ordering policies. Disney and Grubbström [6] have examined the AVIOBPCS model used here to obtain analytic expression for the economic performance of a generalised order-up-to policy for random demands. They do show that an optimal solution is no longer the same as that for inventory alone. Several authors have used Net Present Value (NPV) Analysis to investigate e-commerce [7], disassembly processes [8], MRP systems [9]; EOQ models [10] and make –to-order and make-to-stock systems [11]. Most business operators do not use NPV and it may not be as relevant in the present almost zero interest rate market conditions. 2.

AVIOBPCS MODEL The problems of production and inventory have posed significant problems throughout the history of manufactured goods. Many different analytical techniques have been used with no one technique being wholly adequate in providing all the necessary solutions. However the consideration of the information flows, the effects of batch sizes and scheduling has now reached a position where the economic penalties of excess stock and the western poor practices have been replaced by the Toyota lean production regime. New e-manufacturing environments are concerned with rapid response and are generally concerned with Factory-to-Business (F2B) initiatives with the use of Lean and Agile manufacturing processes driving manufacturing engineers to look for greater efficiency in the Supply Chain. Simon [17] originally proposed the use of transform techniques to inventory and order based production systems. Tustin [18] and Vassian [19] also applied transform techniques to economic and inventory systems in the 1950's. Later in the 1960's Forrester [20] applied the methods of Industrial Dynamics simulation to the problem. Forrester devised the method of System Dynamics to describe problems that were not amenable by other means, including human decisions. The principle arguments were that ALL such processes could be described in simple forms by the feedback loops inherent in the information flows and the delays represented in those flows. He used exponential transfer functions to represent the delays, as he observed that they represented many real situations more closely. In the late 1960's Adelson [21] used z transforms in an inventory and order-based system, producing an analytical expression for bullwhip. More recent work using analytical methods by Towill's group at Cardiff University is the origin of the work here. The whole family of IOBPCS models is described in Ferris and Towill [22]. These problems can also be represented as state-space equations, either as continuous or discrete models. A review of dynamic modeling in supply chains is given by Sarimveis et al. (2008).Disney & Towill [12] at Cardiff University and White and Censlive [13] have analyzed Supply Chains with Automatic Pipeline, Inventory and Order Based Production Control Systems APIOBPCS) extensively, to determine its‘ stability and optimization [14]. The actual inventory at the distributor is compared with the agreed re-order point. This decision is arrived at by consideration of Customer Service Levels (CSL) whilst not building up excessive stocks. The agreed procedure embedded in the APIOBPCS model was found by Riddalls and Bennett [15] to follow human behavior. In this APVIOBPCS model of a factory and sales system we see that the distributor produces virtual sales orders assuming a typical pattern of behavior. These orders are further modified by the factory using experience of the learning curve over time (An effective exponential delay). This is added to a fraction of the inventory error, plus a fraction of the Work In Progress (WIP) error. This comprises the order rate, which then will after a delay lead to production being completed. From this completion rate the virtual sales rate is subtracted and the accumulation of these products leads to the inventory.

3.

MODEL In the early descriptions of System Dynamics, delays were assumed to be exponential in form, in control terms-an exponential lag. Hence the delays due to production for example are described by a simple time constant Tp. This is typically the modeling process used in Stella or Vensim. In the transform models used here the differential equations can be modeled using Laplace transforms assuming that a large number items are being handled or for a smaller set of objects, that are better described by difference equations, which can

410


then be modeled using z transforms. Both these techniques are limited to linear models but have great power in allowing general expressions to be obtained. The main structural system that has been investigated here is that of the Automatic Pipeline, Variable Inventory and Order Based Production Control System (APVIOBPCS). The inventory error (EINV) is found as the difference between a fixed desired level of inventory (TINV) and the actual inventory (AINV). A smoothing function is used to obtain the average sales consumption (AVCON) as a function of the virtual consumption rate VCON. In his earlier work, Towill [23] showed that the averaging techniques commonly used in industry could be modeled by an exponential lag in a continuous model. This is combined to obtain the order rate (ORATE) given to the production facility. There is almost always a production delay inherent in the manufacturing process. The Order rate (ORATE) is obtained from the sum of a fraction of the exponentially smoothed virtual sales plus a fraction of the error in inventory plus a fraction of the perceived error in Work In Progress. The error in inventory is supplied by a variable demand. Payment is often requested at this point. This model, implemented in SIMULINK, is shown in figure 1 with the cost functions included. The basic structure of these models as implemented by Towill‘s group is to use a proportional controller in the loops from the inventory and WIP error. Other authors [24] and [25] have proposed PID controllers. These were shown by Censlive [26] to give lower inventory stock outs than the best proportional control. A second model was therefore implemented with PID controllers to see whether they were more cost effective. The model we use here is a standard representation of APVIOBPCS but with non-linear features to compute the cash flows. These equations can be modified to include different formulations of costing or additional terms. The actual values can be changed to do what if calculations very easily. It is clear that the performance measures described by Beamon can be evaluated relatively easily in this model. It is also clear that the exact level of inventory does not have the penalty effect commonly assumed.

Delivery

1 Td.s+1 Transfer Fcn2

Switch1

Scope

Cash inflow -KScope 12 1

-K-

orate -KGain1 -K-

payment delay

0

salerate

Ta.s+1 Transfer Fcn1

Gain2

Gain7

Scope 10

targetinv

Constant1

Out1

Scope 6

Scope 7

Rate Transition Subsystem

comrate 1

1/s

Tp.s+1 Transfer Fcn3

Integrator

Saturation

Gain3 Gain8 -K-

Profit

Scope 2

wip 1/s Integrator1

Scope 8

Scope 1

inventory holding cost -K-

1/s

Gain4

Integrator2

Scope 9

Scope 4

-CConstant Scope 11

backlog cost

-K-

Switch

Scope 3

Gain5

costs -KGain6

production costs

Scope 5

Figure 1. Schematic of the APVIOBPCS with cost functions implemented in SIMULINK. 4.

TYPICAL EXAMPLE The example used here is of an automated PCB assembly plant described in Deif & Elmaraghy [27]. The product is a RAM module with data from a notational case. This example has cost functions to allow for: Holding cost CH, calculated from the quantity of unsold RAM/week, Q H, and then converted to holding cost using the following equations:

𝑄𝐻 = 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 − 𝐷𝑒𝑚𝑎𝑛𝑑

(1)

411


𝐶𝐻 = 𝑄𝐻 𝑖𝑃𝑟

(2)

The backlog cost, CB, is calculated from backlog quantity QB and then multiplying by the backlog penalty PB and the cost of loss of goodwill CLGW as shown below: 𝑄𝐵 = 𝑑𝑒𝑚𝑎𝑛𝑑 − 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 + 𝑄𝐻

(3)

𝐶𝐵 = 𝑄𝐵 𝑃𝐵 + 𝐶𝐿𝐺𝑊 𝑃𝑆 The values quoted by Deif & Elmaraghy [27] are: Pr=$30, Ps=$100, PB=0.01% of the selling price, CLGW=0.01% of the selling price, i=0.2% Since there is a high volume flow rate of product we are justified in using a continuous model in this case. RESULTS

The simulation is run over 52 weeks as no data beyond that time was available, and the results for the inventory, order rate, completion rate and demand are shown in figure 2 for the proportional control using Disney and Towill‘s‘ gains. In the paper by Deif the sales varies from month to month and this effect is shown in figure 5. The inventory follows the sales rate after the initial stock out at the start of the simulation. Notice that the variable desired inventory level is a much smoothed version of the sales. After the initial peak both the order rate and the completion rate follow the sales demand. Figure 2 shows the cumulative profit with these proportional gains. These are plotted with the results obtained by replacing the proportional gains in the inventory and WIP The results in figure 5 and 6 illustrate a maximum value of profit is gained, for the base values of PID gains, with Tv=1.3 feedforward with PID controllers using optimum PID gains [23]. The cumulative profit shown in figure 2 illustrates the economic superiority of the PID controlled system. This shows a smaller period when a loss is made and this is responsible for ensuring a greater final profit. INVENTORY PERFORMANCE

5

2

x 10

1.5

RAM

1

0.5

AINV targetinv ORATE COMRATE SALES

0

-0.5

-1

0

10

20

30

40

50

60

Time/weeks

Figure 2. Inventory Performance of the APVIOBPCS with proportional control using sales data from Deif (2007).

412


EFFECT of CONTROL FORM on PROFIT

8

x 10

3

Cumulative Profit/$

2.5

2

1.5

1

0.5

Proportional PID

0

-0.5

0

10

20

30

40

50

60

Time/weeks

Figure. 3. Cumulative Profit for the APVIOBPCS comparing Proportional and PID control for a step sales input. The PID system has a lower value of deficit than the Proportional system and is paid back in a shorter time. For this sales input the cash inflow and costs are plotted in figure 4. No cash inflow occurs before week 10 and the costs peak at $4 million/week, while the cash in from sales peak at around $11 million/week. Changing the desired inventory constant only affects the point at which cash inflow starts, hardly changing the costs. This is confirmed in the weekly profit figures. The deficit is nearly the same but the date at which profit starts coming in is earlier for the higher DINV constant. The cumulative profit is greater when the desired inventory constant is higher as shown in figure 5 with a payback period reduced by 2 weeks. However the variation is close to the maximum value of profit for a wide range of Tv values. INFLOW and OUTFLOW OF MONEY

6

Cash/$

15

x 10

10 5 0

0

10

20

30

40

50

60

Time/weeks 6

Costs/$

4

x 10

3 2

Tv=0.5 Tv=0.8

1 0

0

10

20

30

40

50

60

Time/weeks

Figure 4. Flow of Money in the APVIOBPCS with proportional control using sales data from Deif (2007) for two values of desired inventory gain.

413


TOTAL PROFIT for DIFFERENT DESIRED INVENTORY

8

x 10

2.5

Cumulative Profit/$

2

1.5

1

0.5

Tv=0.8 Tv=0.5

0

-0.5

0

10

20

30

40

50

60

Time/weeks

Figure 5. Cumulative Profit for the APVIOBPCS with proportional control using sales data from Deif (2007) for two values of desired inventory gain. If we compare the changes in maximum cumulative profit in figure 7 with the changes caused by changing the Pi gain by 100 we can see that the change in cumulative profit is of the same order as changing the production cost by $5. Figures 8 & 9 illustrate that the inventory changes are small for the change of Pi gain of 100. This means that the use of Pi changes to increase profit instead of reducing the cost price is possible.

8

2.84

Variation of Cumulative Profit with variable desired inventory

x 10

X: 14 Y: 2.83e+008

2.82

2.8

Cumulative Profit/$

Interest rate =2% 2.78

Interest rate=4%

Peak value Tv=1.3

2.76

2.74

2.72

2.7

2.68

0

5

10

15

20

25

Index

Figure 6 Variation of Cumulative profit with variable desired inventory for different interest rates.

8

4

Variation of Profit with production price

x 10

3.5

Pr=$10

3 Pr=$30 2.5

Profitp/$

2

1.5

1

Pr=$55

0.5

0

-0.5

0

10

20

30 Time/weeks

40

50

60

Figure 7 Effect of production cost on the cumulative profit

414


8

Variation of Profit with PID gain Pi

x 10

3

Pi=160

2.5

2 Pi=60

Profitp/$

1.5

All other parameters at base values

1

0.5

0

-0.5 0

10

20

30 Time/weeks

40

50

60

Figure 8 Effect of Inventory PID proportional gains Pi on the cumulative profit We can consider the case of optimising the inventory error a common suggestion in the literature. The case investigated here is of using the time integral of absolute inventory error (ITAE) as a weighting function. 4

Inventory for two PID gains

x 10

7

6

5

4 Pi=60

AINV/RAM

3

2

Pi=160

1

0

-1

-2

-3

0

10

20

30 Time/weeks

40

50

60

Figure 9 Inventory for two PID gains with Tv=0.5 The gains obtained by this are: Pi=34, Pw=12.4, Ii=0.333, Iw=0.25, Di=0.4, Dw=0.4. The maximum cumulative profit for this situation is $2.86e8 after 52 weeks. This compares with a value of profit of $2.83e8 when no weighting of the inventory takes place. 4

16

Profit with twp Pi gains for Tv=1.3

x 10

14

12

10

Pi=60

Profitp/$

8

Pi=160

6

4

2

0

-2

-4

0

10

20

30 Time/weeks

40

50

60

Figure 10 Inventory for two PID gains with Tv=1.3, optimum value

415


Variation of Maximum Profit with PID gains 8

x 10 3

Max profit

2.9 2.8 2.7 2.6 2.5 60 40 20 0

Pw

100

50

200

150

250

Pi

Figure 11 Most profit at 52 weeks for variation of P terms in PID control Maximum Profit for variation of Integral PID values 8

x 10 2.9128 2.9127

Pi=200,Pw=1

Max Profit/$

2.9127 2.9126 2.9126 2.9125 2.9125 2.9124 0.8 0.6

6 0.4

4 0.2

Iw

2 0

0

Ii

Figure 12 Most profit at 52 weeks for variation of I terms in PID control A second set of experiments were performed to maximise the cumulative profit after 52 weeks by varying the gains separately. This is shown in figures 11, 1 2 & 13. In contrast to optimising the inventory we are hear optimising the profit and the cumulative profit is higher than the ITAE case by around $10M. Most of this is due to the higher proportional gains, but variation of the integral and derivative terms also contribute by at least $100000! However the shape of the surface in figure 11 illustrates that nearly the same result is given by 70 200

Zone Number

0

1

2

3

4

Each individual circle surrounding the robot (circles around it) is again divided in 8 sectors (A to H). Total nos. of states for obstacles-avoiding agent can be found out in the similar manner as done in the case of light-seeking agent. Suppose an obstacle is detected inside the first circle (no. 0) and in the second sector (no. B) as shown by a red square in fig. 3 (b). So, for this single obstacle, there may be 5 possibilities (as there are only 5 circles) of an obstacle to be present in the next sectors i.e. first, second, third and fourth circle. This is to be noted that it is immaterial whether more objects are detected behind the first obstacle in the same sector. In such cases only the first obstacle (nearest to the centre of the circles) is considered. So together, 1 × 5 combinations are possible. Again, if the third sector is considered, total 1×5×5 (or 5 2) combinations can be obtained. For all the sectors of the different circles, in total 5 8 i.e. 390625 combinations (i.e. states) are available. But as in the earlier case, it is not possible to handle such large amount of data which require a lot of computational power, memory space and processing time. For the simplicity, rather only those sectors of the circles are considered, where the brightness is maximum, second maximum and third maximum. Thus the total nos. of states reduces to 5 3 i.e. 125 nos. instead of 58 combinations. The robot should also move continuously to avoid trap. So, to avoid traps repeated readings of the robots should be avoided. This is denoted by another state: ‗Repeated reading?‘- Yes or No which is referred by 1 and 0. So, here 2 combinations are possible and in total 125 × 2 (=250) nos. states can be obtained. For this agent the state table will also consist of 4 digits as given in table 5. The state IDs along with their description is given in the table 5 below. TABLE 9.

THE STATE TABLE FOR THE AGENT OBSTACLE-AVOIDANCE

Obstacle in the Maximum Intense Zone

Obstacle in the Second Maximum Intense Zone

Obstacle in the Third Maximum Intense Zone

Repetition

Percept ID 1

Percept ID 2

Percept ID 3

Percept ID 4

0

1

2

0

TABLE 10.

ACTION TABLE FOR OBSTACLE-AVOIDANCE AGENT

Between 1st and 2nd circle

Between 2nd and 3rd circle

Between 3rd and 4th circle

Outside 4th circle

34

67

84

100

117

0

1

2

3

4

Position of obstacles

Within 1 circle

Speed (RPM)

Action ID

st

The Avoid-obstacle agent locates the obstacles and finds out their position (Table 6). This agent works independently and separately without any intervention by other agents. It also shares the information of three maximum brightness zones obtained by the light-seeking agent.

589


6.4

Co-ordinator and the modifier All the sub-states and the recommended actions for different agents are forwarded to co-ordinator for selecting the refined action. The modifier chooses the refined actions against the above simplified states from the initial or updated Q-table based on the Q-values. All the three agents are sending data (sub-states and action) simultaneously to the Co-ordinator. As the agent Hunger, is of highest priority, the co-ordinator sorts out this agent first and forward its recommended action and sub-state to the modifier. If the recommended action is ‗shut-down system‘, the same is passed for action. No other command by other agents will be considered. But if the action is ‗do not shut down‘, the modifier looks for the next priority agent. The next priority agent is the Goal-seeking agent. The agent gathers the data (brightness) from the surroundings and arranges only three of them (numbers of sectors with different brightness) in descending order starting with the maximum. Suppose the sub-states for such a case is ‗2 – 3 – 1‘. The action related to the maximum brightness is rotation through an angle (= number of the sector × 45°) in the clockwise direction (as the reading has been taken in a step of 45° in the clockwise direction). Here it will be rotation through 90° (2 × 45°). The sub-states and the action are passed to modifier for further action. Obstacle-avoidance agent is the agent with last priority. This agent gathers the data about the presence of obstacles in the sectors and circles (1st, 2nd, 3rd and 4th) and compares them with the sectors of first, second and third maximum brightness. Then the agent passes the sub-states and action to the Co-ordinator for refinement. Here for 5 different zones (within 1st , 2nd , 3rd and 4th circle and outside 4th circle) and 3 different brightness regions, total 215 combinations are possible for presence (‗1‘) or absence (‗0‘) of obstacles. The modifier chooses the action according to following rules. Rule - I: If there is an obstacle within the first circle of the brightest zone, go to second bright zone. If there is an obstacle also within the first circle of the second bright zone, go to the third bright zone. If there is an obstacle with in the first circle of the third bright zone, the system may stand still. Rule - II: If the obstacle is not within the first circle, but within second or third or fourth circle of the considered bright zone, the system will use modified speed as given in table 6. In this regard a point may be noted that if there is an obstacle within the first circle of the brightest zone; another obstacle within the second circle (more specifically between first and second circle) of the second bright zone and no obstacle in the third bright zone, the modifier will make the system move along the second bright zone with modified speed. The reward obtained due to the past action is updated in the Q-table for respective agents. Here two cases may happen. Case-I: When the agent hunger detects that the voltage level is below the predefined threshold limit, the reward will be totally updated to the Q-table of the agent hunger, as this agent solely defines the action. Case-II: When the agent hunger detects that the voltage level is above the predefined threshold limit, the reward is distributed with reducing weightage as mentioned in equation 2. Here as only three agents are active, goal-seeking, obstacle-avoidance and direction-correction, the rewards are distributed with the weightages 0.6, 0.3 and 0.1 respectively. The reward for goal-seeking is calculated as:

[(Lmax )i  ( Lmax ) ] i 1

(2a)

= maximum light value of the previous state (i  1) in the direction of motion of the ) max i  1 robot and ( L ) = maximum light value in the (i) current state for that same direction. Similar policy max i Where ( L

is also adapted for obstacle-avoidance agent as mentioned in eqn. 2b.

[( D ) ] max i

(2b)

Where ( D

) = maximum distance of the obstacle (if present or ‗255‘ if not present. 255 is the max i

maximum range of the ultrasonic sensor) in the same direction in which the robot has moved to reach the (i) current state For no repetition in any of the above mentioned agents, a positive scalar reward of magnitude ‗10‘ or ‗100‘ and for repetition a negative reward (i.e. punishment) of ‗-20‘ or ‗-200‘ can be

590


used. The punishment should be always higher than the reward so as to stop any policy leading to punishment. The high value of punishment will completely abandon the associated action. These rewards multiplied with the weightage factors are updated in the respective Q-tables of different agents. 7.

EXPERIMENT Different experiments in the field of Behavior-based robotics, Reinforcement learning has been carried out in different environments using different robotic systems. Initially it was a challenge to identify the right Behavior-based architecture for implementation. For this purpose two most popular architectures (Subsumption architecture and Motor-schema theory) have been tested in simulated environment. Using the identified architecture, the next series of experiments have been carried out in simulated environment (fig 4(a)), indoor environment (fig 4(b)) and different outdoor terrains (fig 4(c)-(f)). Most of these experiments include only three agents: hunger, goal-seeking and the obstacle-avoidance using the battery-level detector, light sensor and ultrasonic sensor respectively. The data of the experiments have been saved on the robot itself and later downloaded for further processing.

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 4: Experimental environments (a) Simulated environment (b) Indoor environment; Outdoor terrains - (c) Plain grassland (d) Loose sand (e) Laterite tableland (f) Roof of the laboratory during night The experiments with HuMAQ are carried out in five trial runs on different outdoor terrains each with five sets. The first set of each trial run uses a randomly initialized Q-table and from second run onwards, final updated Q-table of the previous set is used as the initial Q-table of the next set. Initially the Q-table is filled up with random values, so as to make sure that for a particular state any of the actions may be selected (which cannot be possible if the cells of the Q-table are filled up with equal values), not a specific one. 7.1

Experimental Systems In all the experiments four different robots have been used. Initially the experiments have been carried out with indigenously developed three different robots: ARBIB (Autonomous Robot Based on Intelligent Behaviors) – I, II and III. Testing for HuMAQ has been carried out on LEGO Mindstorm NXT due to its simple analog sensors.

7.2

Experimental Results The goal of the experiments for the validation of HuMAQ was to reach the brightest zone avoiding the obstacles. The time to reach the goal for all the five sets of the six different trials in both indoor and outdoor environments are plotted in 5 (a) –(c). All these curves are showing a reducing nature for time to reach the goals and this conforms to the real learning methodology. Another important issue for measuring the performance of learning, is convergence of the policy [17][22]. The convergence of the learning can be proven by the reducing nature of individual time required for each sets of the different trial runs as shown in fig. 6(a). Also the changes in initial and final Q-values for different agents prove the refinement of the data as well as the convergence of the learning. Fig 6 (b) and (c) show the initial and final data respectively for light sensor plotted in 3D during testing on plain grassland.

(b) (a) (c) Fig 5: Experimental data plotting (a) Curves for learning time vs. no. of iterations to reach the goal for different sets of experiments using human-like gradual single agent Q-learning. The ideal curve (marked as

591


‗Ideal‘) of learning has also been indicated. (b) The nos. of updates to reach the goal Vs. nos. of trial runs shows a gradually reducing nature which in fact supports the actual learning using HuMAQ in indoor environment. (c) The time to reach the goal Vs. run nos. shows a good acceptability of HuMAQ for autonomous exploration in outdoor terrains.

(b) (a) (c) Fig 6(a): The individual change of Q-values of the goal – seeking agent for different sets of the different trial runs. (b) Initial Q-table for light sensor during testing on plain grass land (c) Final Q-table for light sensor during testing on plain grass land 8.

CONCLUSION The behavior-based robotics has opened up a new field of robotics which uses the Sense →Act paradigm for achieving the low-level intelligence easily. Four different behavior-based architectures are popular, out of which only two (Subsumption architecture and the Motor-schema Theory) have been tested to implement here. Q-learning, a sub-issue of reinforcement learning, uses delayed reward/punishment for the previous action. It uses a state-action mapping table based on the different conditions of the sensors (states) and commands to the actuators (action). Agent is a software entity which performs an assigned task independently. In most cases Q-learning uses single agent to perform all the tasks for learning. Use of multiagent is preferred to handle with large numbers of sensors and therefore large numbers of state-action mapping. The concept of cumulative and gradual learning of humans can be incorporated with the multiagent Qlearning to get a better performance in outdoor terrain explorations by mobile robots. The testing of HuMAQ in different terrains, different static environments reveal the acceptance of the algorithm to be used for autonomous explorations by mobile robots. Also the measurement of the performance of the learning algorithm has been done from the proof of convergence.

9.

RECOMMENDATIONS Further work is recommended for experimentation with dynamic obstacles and using multisystem as multiagents for establishing HuMAQ as a good tool for field explorations.

10. REFERENCES [1] R. S. Sutton & A. Barto, Reinforcement Learning: An Introduction, MIT Press, 1998. [2] C. J. C. H. Watkins, ―Learning from Delayed Rewards‖, PhD thesis, Cambridge, 1989. [3] R. S. Sutton, ―Integrated architectures for learning, planning, and reacting based on approximating dynamic programming,‖ Proc. 7th International Conference on Machine Learning, Morgan Kaufmann, 1990, pp. 216 –224. [4] S. Mahadevan and J. Connell, ―Automatic programming of behavior-based robots using reinforcement learning‖, Proc. National Conference on Artificial Intelligence, New York, 1991, pp. 768-773. [5] C. Gaskett, L. Fletcher and A. Zelinsky, ―Reinforcement learning for a vision based mobile robot,‖ Proc. IEEE International Conference on Intelligent Robots and Systems, Takamatsu, Japan, 30 Octr – 5 Nov 2000, pp. 403 – 409. [6] S. Hagen and B. Krose, ―Neural Q-learning‖, Neural Computing and Applications, Vol. 12 (2), 2003, pp. 81 – 88. [7] A. K. Agogino and K. Tumer , Quicker Q-learning in Multi-Agent Systems, http://ti.arc.nasa.gov/m/pub/ 940h/0940%20(Agogino).pdf. [8] T. Fujii, Y. Arai , H. Asama, I. Endo, ―Multilayered Rienforcement Learning For Complicated collision Avoidance problems,‖ Proc. of ICRA, pp. 2186-2191, 1998. [9] J. Kim and P.Vadakkepat, ―Multiagent Systems: A survey from the Robot Soccer Perspective‖, J. of Intelligent Automation and Soft computing, Vol. 6 (1), pp. 3-17, TSI Press, USA, 2000.

592


[10] K. Park, Y. Kim and J. Kim, ―Modular Q-learning based multi-agent cooperation for robot soccer‖, Journal of Robotics and Autonomous Systems, Vol.35, 2001, pp.109-122. [11] William D. Smart and Leslie Pack Kaelbling, "Effective Reinforcement Learning for Mobile Robots," Proc. of the IEEE International Conference on Robotics and Automation, Washington DC, May 2002, Vol. 4, pp. 3404 – 3410. [12] Ronald C. Arkin, Behavior-Based Robotics, The MIT Press, 1998. [13] R. Brooks, ―A Robust Layered Control System for a Mobile Robot‖, IEEE Journal of Robotics & Automation, Vol. 2, No.1, Page 14-23, 1986. [14] P. Maes, ―The Dynamics of Action Selection‖, Proceedings of the International Joint Conference on Artificial Intelligence, pp. 991-997, 1989. [15] Ronald C. Arkin, ―Motor Schema — Based Mobile Robot Navigation‖, The International Journal of Robotics Research, Vol. 8, No. 4, 92-112, 1989. [16] L. Steels, ―A Case study in the behavior-oriented design of autonomous agents‖, Proc. of International Conference on Simulation of Adaptive Behavior, The MIT Press, 1994. [17] P. Kaelbling, M. L. Littman and A. W. Moore, ―Reinforcement learning: a survey,‖ J. of Artificial Intelligence Research, Vol. 4, 1996, pp. 237 – 285. [18] C. J. C. H. Watkins and P. Dayan, ―Technical note: Q-learning,‖ Machine Learning: Special Issue on Reinforcement Learning, Vol. 8 (3-4), 1992, pp.279 – 292. [19] Y. Dahmani and A. Benyettou, ―Seek of an Optimal Way by Q-learning‖, Journal of Computer Science, Vol. 1 (1), 2005, pp. 28 – 30. [20] S. Vosniadou, ―How Children Learn,‖ International Academy of Education, http://www.ibe.unesco.org/publications/EducationalPracticesSeriesPdf/prac07e.pdf. [21] D. N. Ray, A. Mondal, S. Majumder, S. Mukhopadhyay, ―Human-like Gradual Learning of a Qlearning based Light Exploring Robot‖, Proc. of 2010 IEEE Conference on Robotics & Biomimetics, Tianjin, China, Page 1411 – 1416, 2010 [22] E. Martinson, A. Stoytch and R. Arkin, ―Robot behavioral selection using Q-learning‖, Proc. of the IEEE IROS 2002, Switzerland, 2002.

593


SENSITIVITY ANALYSIS OF GENETIC ALGORITHMS FOR JOB SHOP SCHEDULING PROBLEMS S. Maqsood1, S. Noor2, M. K. Khan1, A. S. Wood1 email1: [email protected] email1:[email protected] email1:[email protected] email2: [email protected] 1

School of Engineering, Design and Technology, University of Bradford, Bradford, UK

2

Department of Industrial Engineering, NWFP University of Engineering and Technology, Peshawar, Pakistan

ABSTRACT In literature, Genetic Algorithms (GAs) are widely used in manufacturing scheduling problems. The sublime GA technique, mostly dependant on few critical parameters such as number of generation, population size, crossover and mutation rates. No technique is yet known to find the best combination of the parameters set for optimum output of the GA. In this paper, a GA-based algorithm for comprehensive GA parameter analysis (also called sensitivity analysis) is proposed for popular Job Shop Scheduling Problems (JSSP). Several JSSP benchmark problems are solved with the help of proposed algorithm for sensitivity analysis. The effect of each parameter is explained with the help of detailed charts. In future, this paper will provide a reference for selection of GA parameters in optimisation of JSSP. Keywords: Genetic Algorithm (GA), Sensitivity Analysis, Parameters, Optimisation, Job shop Scheduling Problem (JSSP)

1.

INTRODUCTION In operation‘s research, Evolutionary Computation (EC) is a subfield of AI (more particularly computational intelligence) that involves combinatorial optimisation problems. EC deals with Genetic Algorithms (GA), Evolution Strategies and Genetic Programming (GP). This approach is based on the computational models of natural selection and genetics. The GA is a class of stochastic search algorithms based on biological evolution [1-3]. It is one of the most popular optimisation tool and capable of being applied to an extremely wide range of large problems. However, some researcher used GA from an experimental perspective [2]. The application of the GA to scheduling problems has interested many researchers due to the fact that offers the ability to cope with the huge search spaces involved in optimisation and constraints satisfaction problems [4-6]. Coley, [7] has listed a range of practical optimisation problems to which GA has been successfully applied. A typical GA might consist of the following: a)

A population, of estimates of solution to the problem. Rather than starting from a single point (or guess), GAs are initialized with a population of estimates. The population is normally random and spread throughout the search space. The initial estimates (or chromosomes) are held as binary encodings (or strings) of the true variables, although an increasing number of GAs use ―real valued‖ (i.e. base-10) encodings [8].

b) A procedure to calculate the goodness or badness of individual solutions within the population. This procedure is known as selection procedure. For the selection of chromosome many methods are used such as roulette wheel selection, tournament selection, rank selection and steady state selection [1].

594


c)

A way of mixing fragments of the better solutions to form a better new solution.

d) A mutation operator to avoid permanent loss of diversity within the solutions. As discussed earlier, the GA use a stochastic search method, the fitness of a population may remain stable for a number of generations (or iteration) before a superior chromosome appears.

In such cases, the use of conventional terminating criteria is problematic [1]. Therefore, researcher‘s common practice is to terminate a GA after a specified number of iteration. After termination the chromosome are examined for best chromosome in the population. GA is restarted if no satisfactory solution is found.

2.

THE SENSITIVITY ANALYSIS The operational performance of GA is mostly dependant on number of generation, population size, crossover, and mutation rates [9]. In GA, larger number of generation and population size give better results because both parameters tends to explore more solution space; however, it has proved to be computationally expensive. To achieve a cost effective GA the best possible combinations of Crossover Rate (XR) and Mutation Rate (MR) are very important.

In manufacturing scheduling problems, there is no such data found in the literature which can be used as benchmark for using the combination of these two parameters. Therefore, a GA-based algorithm for comprehensive GA parameter analysis (also called sensitivity analysis) is proposed for popular Job Shop Scheduling Problems (JSSP). JSSP is a classical operations search problem and a recurrent problem in many industrial settings, ranging from the planning of a small workshop to the allocation of computing resource [10]. The proposed simple GA is applied to a test bed. The test bed consists of a number of benchmark problems (FT06, FT10, LA01, LA06, LA11, LA12, LA26 and LA36)[11, 12] with the given standard processing time and constraints. The primary aim of these experiments is to study the effect of the various combinations of XR and MR for JSSP. The secondary objective is to present the results in such a form that can be used for future references. 3.

DESCRIPTION OF THE GA BASED ALGORITHM In this paper, a simple GA base algorithm has been used for sensitivity analysis as shown in Figure 1. The proposed GA flowchart is summarized in the following steps:

595


Figure 77: Simple proposed GA based algorithm Step 1: The population has been initialized randomly which spreads throughout the search space. For initial guesses (or chromosomes) is encoded using the permutation technique. Two chromosomes are shown in Figure 2. Each chromosome is a potential solution. Chromosome contains sets of information to solve a particular problem [8].

Chromosome A: 1 5 3 2 6 4 7 9 8 Chromosome B : 8 5 6 7 2 3 1 4 9 Figure 78: Chromosomes used in GA based algorithm [9] The gene‘s arrangements in a chromosome differentiate it from other chromosomes and thus giving it a unique identity. Changing the position of genes in the chromosomes drives us toward the solution. Table 1 Shows chromosomes with different identity or job sequences with a population size of 20 used in this paper.

Table 16: A Sample of Chromosomes of Population Size 20 Chromosome No 1

1

2

2

3

4

1

3

3

2

4

1

Job Sequence 2 3

596

4


4

3

4

1

2

5

4

3

2

1

6

3

1

2

4

7

3

1

4

2

8

2

1

3

4

9

2

4

1

2

10

3

1

2

3

11

2

1

3

2

12

4

2

3

4

13

3

1

4

3

14

1

4

3

1

15

1

2

3

1

16

1

4

2

1

17

2

1

3

4

18

4

1

3

2

19

3

1

2

4

20

1

4

2

2

Step 2: A novel hybrid evaluation algorithm or heuristic (HybH) is used to evolve the schedule and obtain the fitness values for each of the chromosome [6]. The HybH is based on the combination of Index Based Values (IBH) and Finished Job Based Heuristic (FJBH). The minimum the Makespan value of the chromosome, the fittest it is. The fitness values have been used for selection of chromosome for next generation.

Step 3: The SUS (Stochastic Universal Sampling) technique has been used to for selection of parents for crossover and mutation as shown in Figure 3[9].

Figure 79: Stochastic Universal Sampling Chromosomes are mapped over the line with their fitness values. Equally spaced pointers are placed for the selection of N chromosomes. The distance between pointers is given by 1/N and the position of the first pointer is given by a randomly generated number in the range [0, 1/N] [8]. Step 4: For crossover standard job based crossover technique has been used as shown in Figure 4. In crossover process, two randomly selected parents (P1 and P2) are initially selected. Then genes 3 5 7 have to be preserved and will be copied to offspring chromosome C1 (child1) in the same absolute positions as its parent P1. The remaining vacant position will be copied from P2 shown as underline.

597


P1

P2

1 2 3 4 5 6 7 8

5 6 781234

C1

C2

6 8 3 1 52 7 4

51724638

Figure 80: Job Based Crossover Scheme Step 5: For mutation approach a randomly selected gene but not the last one is exchanged with next adjacent gene as shown in Figure 5. 6 8 3 5 1 2 7 4

6 8 3 1 5 2 7 4

Figure 81: Mutation Process Step 6: For each chromosome the process will be repeated until the termination condition is satisfied. 4.

IMPLEMENTATION The proposed GA based algorithm for sensitivity analyses has been implemented in MATLAB 7 and on an Intel(R) Core 2 Duo processor (2.00GHz). The input data required for the algorithms are in the form of processing times and process constraints in spread sheets.

5.

BENCHMARK PROBLEMS To determine the strength of various methods and algorithm used to solve JSSP, they need to be compared with the benchmark problems. The benchmark problems provide a common standard on which all JSSP algorithms can be tested and compared. As the benchmark problems are of different dimensions and grades of difficulty, it is simple to determine the capabilities and limitations of a given method by testing it on these instances. These test findings may help to determine if improvements are required and where they should be made [13]. These benchmark problems are created by various authors (Fisher and Thompson [11] - FT; Lawrence [14] - LA; Adams et al. [15] - ABZ; Applegate and Cook [16] - ORB; Storer et al. [17]SWV; Yamada and Nakano [18] - YN and Taillard [19] - TD) and are freely available on http://people.brunel.ac.uk/~mastjjb/jeb/orlib /files/jobshop1.txt. In This paper GA based algorithm is applied to FT and LA benchmark problems (FT06, FT10, LA01, LA06, LA11, LA12, LA26 and LA36) with the given standard processing time and constraints.

6.

COMPUTATIONAL EXPERIMENTS AND RESULTS In literature, it has been reported that large number of population and generation number tends to provide the optimal solutions at computational cost [9]. Therefore, with large number of these mentioned two characteristics it will be very hard to find the best combination for crossover and mutation probability. Therefore, for experiments the generation number and population size for all benchmark problems are taken as 20. These lower values have not only reduced the computational cost, but have also helped in finding the best possible combination of XR and MR. The XR and MR best combination are very important parameters. Their best possible combination not only resulted in decreased computational time, but will also enable schedulers to save production time. For each experiment a wide range of the Crossover Rates (XR) 0.1, 0.5 to 1.0 and Mutation Rate from 0.01 to 0.10 are taken. The experiment results are shown in bar charts from Figure 6 to Figure 13. Each chart represents the MR on x-axis and Makespan on y-axis. The bars represent the XR. The GA based algorithm has achieved global Makespan value for most of the cases. Interestingly, it has been observed that there is no effect of XR-MR combination on the solution for cases like LA06, LA11, and LA 12. It may be due to two reasons. Firstly, the problems such as LA06 and LA12 are easy to solve computationally. Therefore, the proposed evaluation algorithm used for initial evaluation of the schedule found the global Makespan and results in termination of the algorithm without the use of GA. Secondly, the type of operator used for crossover, mutation, selection suite the problems such as LA11 and LA12, and the solution converges quickly (See Results in Figure 7 and Figure 8). For hard problems

598


like FT10, LA01, the combination plays their role. The XR-MR combinations (0.7, 0.8 – 0.04, 0.06, 0.07, 0.8) achieve new lower bound or minimum Makespan value as shown in Figure 7 and Figure 8. For larger problem such as LA 26 the same XR-MR combinations (0.8-0.04, 0.05 and 0.08) proved to have provided best results as in Figure 12. In another large problem LA 36, the XR 0.07 and 0.08 provided best result with almost all MR values except MR 0.1. In summary, for test bed the best results have been achieved for XR 0.7 and 0.8 with MR 0.04, 0.06, 0.07 and 0.08.

61 60 59 58 57 56 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

MR 0.09

MR 0.10

XR-0.1 XR-0.5 XR=0.6 XR-0.7 XR-0.8 XR-0.9 XR-1.0

Figure 82: FT 06 benchmark problem 1350 1300 1250 1200 1150 1100 1050 1000 950

XO-0.1 XO-0.5 XO=0.6 XO-0.7 XO-0.8 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

MR 0.09

MR 0.10

XO-0.9 XO-1.0

Figure 83: FT 10 Benchmark Problem

710 705

XR-0.1

700

XR-0.5

695

XR=0.6

690

XR-0.7

685

XR-0.8

680 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

Figure 84: LA 01 Benchmark Problem

599

MR 0.09

MR 0.10

XR-0.9 XR-1.0


930

XR-0.1 XR-0.5

920

XR=0.6

910

XR-0.7 900

XR-0.8

890

XR-0.9 MR 0.01MR 0.02MR 0.03MR 0.04MR 0.05MR 0.06MR 0.07MR 0.08MR 0.09MR 0.10

XR-1.0


1230

XR-0.1

1220

XR-0.5

1210

XR=0.6

1200

XR-0.7 XR-0.8

1190 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

MR 0.09

MR 0.10

XR-0.9 XR-1.0

Figure 86: LA 11 Benchmark Problem 1050 1030 1010 990 970 950

XR-0.1 XR-0.5 XR=0.6 XR-0.7 XR-0.8 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

MR 0.09

MR 0.10

XR-0.9 XR-1.0

Figure 87: LA12 Benchmark Problem 1400 1380 1360 1340 1320 1300 1280 1260

XR-0.1 XR-0.5 XR=0.6 XR-0.7 XR-0.8 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

Figure 88: LA26 Benchmark Problem

600

MR 0.09

MR 0.10

XR-0.9 XR-1.0


1520 1500 1480 1460 1440 1420

XR-0.1 XR-0.5 XR=0.6 XR-0.7 XR-0.8 MR 0.01

MR 0.02

MR 0.03

MR 0.04

MR 0.05

MR 0.06

MR 0.07

MR 0.08

MR 0.09

MR 0.10

XR-0.9 XR-1.0


7.

CONCLUSION AND FUTURE RECOMMANDATION In this paper, the operational performance of GA is studied with the help of a GA based algorithm. The proposed GA is applied to benchmark problems in combination with a novel hybrid heuristic for evaluation of the schedule. The HybH is initially used for evaluation of job shop schedule from which the fitness value (Makespan) is determined, followed by GA standard processes. The novel combination of HybH and GA is a contribution in this area of research. The results %GAP between the calculated Makespan values and Global Makespan Values is as low as zero in most cases. The second contribution of the research is the study of XR-MR combinations for JSSP in order to achieve optimum combination of these two parameters. In GA, the other two parameters, the generation‘s number and population size their values is have been kept 20 respectively, because larger number results in optimum or near optimum result with showing less variation in XR-MR effect. The results (optimum XRMR values) achieved from the study can be used as platform for selection of the XR-MR in JSSP optimisation problems. For future it is recommended that for each generation the optimum results and best possible combination should be studied in real scheduling environments in order to find more cost effective combination for genetic algorithms.

8.

REFERENCES [1] Negevitsky, M., Artificial Intelligence: A Guide to Intelligent Systems 2002, Essex: Addison Wesley. [2] Maqsood, S., Khan, M.K., and Wood, A.S., A Review of AI techniques for Manufacturing Scheduling, in The 25th International conference of CAD/CAM, Robotics & Factories of the Future. 2010: Pretoria, South Africa. [3] Pohlheim, H. Genetic and Evolutionary Algorithm Toolbox for use with MATLAB. 2009 [cited 2009 12 March]. [4] Varela, R., Vela, C., Puente, J., Serrano, D., and Suárez, A., A New Chromosome Codification for Scheduling Problems, in Information Processing with Evolutionary Algorithms, X. Wu, L. Jain, M. Graña, R.J. Duro, A. d‘Anjou, and P.P. Wang, Editors. 2005, Springer London. p. 74-82. [5] Gao, J., Gen, M., Sun, L., and Zhao, X., A hybrid of genetic algorithm and bottleneck shifting for multiobjective flexible job shop scheduling problems. Computers & Industrial Engineering, 2007. 53(1): p. 149-162. [6] Maqsood, S., Khan, M.K., and Wood, A.S., A Novel Heuristic for Low Batch Manufacturing Process Scheduling Optimization in The Computer Aided Process Engineering (CAPE). 2011: School of Engineering, Design & Technology. University of Bradford, UK. [7] Coley, D.A., An introduction to GA for Scientist and Engineer. 1999, London: World Scientific. [8] Noor, S. and Khan, M.K. Hybrid scheduling system for flow shop and job shop scheduling problem. in 23rd International Conference of CAD/CAM, Robotics & Factories of the Future Conference. 2007. Bogota, Colombia. [9] Noor, S., Operational scheduling of traditional and flexible manufacturing system using genetic algorithms, artificial neural networks and simulation., in School of Engineering Design and Technology. 2007, University of Bradford: Bradford. [10] Grana, M., Duro, R., d'Anjou, A., and P.Wang, P., eds. Information Processing with Evolutionary Algorithms. ed. X. Wu and L. Jain. 2005, Springer.

601


[11] Fisher, H. and Thompson, G.L., Probabilistic learning combinations of local job shop scheduling rules., in Industrial Scheduling, J.F. Muth and G.L. Thompson Eds., Editors. 1963, Prentice Hall: Englewood Cliffs, New Jersey. [12] Lawrence, S., Resource constrained project scheduling: An experimental investigation of heuristic scheduling techniques (Supplement). Graduate School of Industrial Administration, Carnegie Mellon University, Pittsburg., 1984. [13] Morshed, M.S., A hybrid model for job shop scheduling, in Department of Mechanical and Manufacturing. 2006, University of Birmingham, UK. [14] Lawrence, S., Resource Constrained Project Scheduling: An Experimental Investigation of Heuristic Scheduling Techniques (Supplement), . Graduate School of Industrial Administration, Carnegie Mellon University, Pittsburg, 1984, 1984. [15] Adams, J., Balas, E., and Zawack, D., The shifting bottleneck procedure for job shop scheduling. Management Science, 1988. 34(3): p. 391-401. [16] Applegate, D. and Cook, W., A computational study of the job-shop scheduling problem. Informs Journal of Computing, 1991. 3(2): p. 149-156. [17] Storer, R.H., Wu, S.D., and Vaccari, R., New search spaces for sequencing problems with application to job shop scheduling. Management Science, 1992. 38(10): p. 1495-1509. [18] Yamada, T. and Nakano, R., A genetic algorithm application to large scale job shop problems. Parallel Problem Solving from Nature, 2, 1992: p. 281-290. [19] Taillard, E., Benchmarks for basic job shop scheduling problems. European Journal of Operational Research, 1993. 64(2): p. 278-285.

602


DECENTRALISED NAVIGATION OF AGVS IN A COMPUTER INTEGRATED MANUFACTURING ENVIRONMENT 1

Roland Dixon , Prof. G. Bright 1

2

2

University of KwaZulu-Natal Department of Mechanical Engineering Durban, South Africa E-mail: [email protected] University of KwaZulu-Natal Department of Mechanical Engineering Durban, South Africa E-mail: [email protected]

ABSTRACT Material handling systems play a large role in optimised manufacturing techniques. The out dated automated guided vehicles (AGVs) used for material handling in today‘s manufacturing environments are no longer suitable as manufacturing technology strives towards ―mass customisation‖ in computer integrated environments. This paper serves the purpose of discussing current AGV systems and proposing a material handling AGV system design that integrates flexibility, robustness, advanced manoeuvrability and the extended Kalman filter into a localisation and coordination technique suitable for ―mass customisation‖ manufacturing environments. By doing this, the bottlenecks associated with centralised control, out dated navigational techniques and high noise environments can be eradicated, providing a more efficient method of material handling. Key words: Material Handling, Robustness, Flexibility, Decentralised Control, Kalman Filter. 1.

INTRODUCTION ―Mass customisation‖ manufacturing environments require material handling systems to be highly flexible. The navigational techniques used in today‘s AGVs can cause considerable bottlenecks in the path flow of part around a computer integrated environment as well as being inflexible and sensitive to disturbance. Localisation through the extended Kalman filter has become a well-known solution to navigational problems in the mobile robot domain [1]. The prediction-correction technique of Kalman filter allows for sensor fusion between relatively low noise sensors allowing multiple agents to move freely in a noisy, nonlinear, dynamic environment. Navigation techniques of AGVs that are used in manufacturing environments will be discussed in this paper with the purpose of determining their short fallings. A better technique for navigation will be derived based on the extended Kalman filter that makes use of the advanced manoeuvrability and easy control of the omnidirectional vehicle design.

2.

CURRENT AUTOMATED GUIDED VEHICLE SYSTEMS Current AGV navigation techniques, their advantages and disadvantages will be discussed in this section drawing attention to the need for flexible and robust systems.

2.1

Line Following Techniques The first of the line following technique is the embedded guide wire method. Wires are placed in the ground according to the flow of various parts around the factory floor. A magnetic field is induced along the wires which can be sensed and followed by the AGV. The AGV is able to follow certain paths by sensing different frequencies or the user will have to turn off all wires except the path that is needed to be followed. The second technique is the paint strip method. AGVs are able to follow predefined paint strips on the factory floor. The paint strip method may not be as flexible as the embedded guide wire method as far as paths are concerned but is a lot more reliable in a manufacturing environment with a lot of electrical noise [1].

603


The two line following techniques have major disadvantages [2]: 1.

Both methods are highly inflexible as the robots are only capable of following the specific paths that are predefined. The paint strip method being even more inflexible than the guide wire method.

2.

Should the robots overshoot their navigational lines then navigation will cease altogether. Therefore the methods may be described as sensitive and non-robust.

3. Should a process layout change, considerable time and cost would be spent on upgrading the required paths.

2.2

Inertial And Odometric Techniques Inertial navigation methods keep track of the robot positioning by sensing acceleration forces acting on the platform of the robot whilst the odometric method keeps track of the position by counting wheel rotations (dead reckoning). Control of the robot is done through kinematic models and internal error determination through PID controllers. Unfortunately both these systems suffer from unbounded accumulative errors over time due to inaccuracies in sensor feedback measurements and system noise [2].

2.3

Beacon Navigation In beacon navigation, the robot is able to detect beacons that are place in the environment and extract position information from them. Ultrasonic transmitters are an example of an active beacon and can be placed in the environment. The robot is able to navigate through an environment by constantly receiving the ultrasonic pulses and determine its position by triangulation. Some robots are equipped with laser range finders that detect reflective pads that have been strategically place in the environment. The robot can then determine the range to each reflective strip and use trilateration techniques to determine its position. The reflective strip is an example of a passive beacon. Due to beacon navigation been based on external measurements of position and not internal measurements of motion it does not suffer from accumulative ―integral errors‖ [2]. The short fallings of the beacon navigation methods are that the robot needs to be in constant view of a specific amount of beacons in order to get a position measurement and the measurements may be inaccurate due to noisy environments.

2.4

Self-Guided Vehicles Self-guided vehicles (SGVs) [1] are more a more modern type of AGV. They use a combination of wheel distance/speed control (dead reckoning) or inertial measurements and an update on their absolute position by taking external measurements from beacons in the manufacturing environment to ensure the errors in the wheel control do not grow too large. The proposed AGV system is an adaption of the SGV system and attempts to add greater flexibility and efficiency by exploiting advances in navigational, computational, hardware and manoeuvrability design technology.

3.

MASS CUSTOMISATION AND THE NEED FOR EFFICIENT MATERIAL HANDLING SYSTEMS ―Mass customisation‖ can be considered as a hybrid between mass production manufacturing systems and flexible manufacturing systems (FMS). It proposes a system whereby a factory would be able to produce a variety of different, customised products at mass production rates. In order to achieve maximum customisation at high production rates a material handling system that is, in itself, highly flexible and efficient needs to be installed. The material handling system would need to be able to identify, transport and store each work part based on its specific part flow.

604


4.

KINEMATICS The design of the robot platform resembles that of the RoboCup soccer playing robots [3] which make use of omnidirectional (holonomic) wheels. Conventional AGV designs make use of platforms with nonholonomic constraints which restrict a vehicles motion. Although the vehicle will be able to reach every point in a work space, the sequence of movements to get the will complex. An example of a non-holonomic constraint would be parallel parking a car. The car cannot simply go sideways into a parking space; it often has to go through a sequence of movements before it gets there. The omnidirectional wheels allow for movement in the longitudinal and lateral directions of the wheel and thereby permit the vehicle to move in any direction to any point without any constraint. The mathematical kinematic models of a vehicle with omnidirectional wheels are simpler to derive than non-holonomic designs and drastically decrease the complexity of the control architecture. Another advantage of omnidirectional wheels is the ability to move easily in confined space, this will allow the robot to reach places in a manufacturing environment with ease that would be close to impossible for a conventional platform design [4].

Figure 1: (a) A diagrammatic representation of non-holonomic vehicle design moving from point A to point B. (b) Shows the omnidirectional wheel design. (c) A diagrammatic representation of an omnidirectional vehicle moving from point A to point B.

4.1

Modelling The Extended Kalman filter which is briefly described in section 5 requires a state transition model and a measurement model. The state transition model is used to acquire a prior estimate of the vehicle state, which is in turn used to form a prediction of the a measurement that the vehicle would observe by using the measurement model. Kinematic Models The kinematic models are used to convert optical encoder readings into global vehicle movements and vice versa for the low level platform control purposes. They also form the basis of the state transition modelling.

605


Figure 2: A simple diagram of the robot platform and the omnidirectional wheels which are capable of movement in the lateral and longitudinal directions of the wheel. In figure 2 above 𝐶 𝐿 = (𝑥𝐿 , 𝑦𝐿 ) is the vehicle‘s local coordinate system and 𝐶 𝑔 = (𝑥𝑔 , 𝑦𝑔 ) is the global coordinate system. ∝𝑖 is the angle between wheel 𝑖 and the vehicles 𝑥-axis or 𝑥𝐿 . The inverse kinematic equation determines the individual wheel velocities given the global platform velocities and is in the form: (1)

∅𝑖 = 𝐷𝑋

Where 𝐷 maps the global velocities of the platform 𝑋 to the velocities of each wheel ∅𝑖 . The full inverse kinematic equation is shown below. ∅1 −𝑠𝑖𝑛(𝜃 +∝1 ) 1 −𝑠𝑖𝑛(𝜃 +∝2 ) ∅2 = 𝑟 −𝑠𝑖𝑛(𝜃 +∝3 ) ∅3 −𝑠𝑖𝑛(𝜃 +∝4 ) ∅4

𝑐𝑜𝑠(𝜃 +∝1 ) 𝑐𝑜𝑠(𝜃 +∝2 ) 𝑐𝑜𝑠(𝜃 +∝3 ) 𝑐𝑜𝑠(𝜃 +∝4 )

𝑅 𝑅 𝑅 𝑅

𝑥𝑔 𝑦𝑔 𝜃𝑔

(2)

𝜃 is the difference between the vehicles x-axis, 𝑥𝐿 , and the x-axis of the global coordinate system 𝑥𝑔 . 𝑅 is the distance from the centre of the vehicle platform to the each wheel, 𝑟 is the radius of the wheels. For simplicity reasons when finding the inverse of 𝐷, the angle between each wheel was chosen to be 90° with ∝1 =45°. 𝐷 is a non-square matrix and therefore the Moore-Penrose pseudoinverse [5] 𝐷 + needs to be obtained to form a forward kinematic equation in the form of: 𝑋 = 𝐷 + ∅𝑖

(3)

Where 𝐷 + = 𝐷𝑇 × 𝐷

−1

× 𝐷𝑇

(4)

The complete forward kinematic equation is shown below and obtains the global velocities of the platform given the optical encoder velocity output readings.

606


−sin⁡(𝜃 + 45°) 𝑥𝑔 𝑟 cos⁡ (𝜃 + 45°) 𝑦𝑔 = 0.5 2 𝜃𝑔 𝑅

−sin⁡ (𝜃 + 135°) cos⁡ (𝜃 + 135°) 0.5 𝑅

−sin⁡ (𝜃 + 225°) cos⁡ (𝜃 + 225°) 0.5 𝑅

−sin⁡ (𝜃 + 315°) cos⁡ (𝜃 + 315°) 0.5 𝑅

∅1 ∅2 ∅3 ∅4

(5)

The technique for deriving the kinematic equations for platforms utilising omnidirectional wheels can be found in [6]. State Transition Models To obtain a prior state estimate for the prediction phase of the EKF a state transition model is require in form of: + (6) 𝑋𝑘− = 𝑓 𝑋𝑘−1 , 𝑈𝑘−1 + 𝑤𝑘−1 = 𝑓 + 𝑤𝑘 −1 + Where 𝑢𝑘−1 is the control input applied to the posterior state estimate 𝑥𝑘−1 at time step 𝑘 − 1 to get to the prior state estimate 𝑥𝑘− at time step 𝑘. 𝑤𝑘 −1 is the zero mean, Gaussian distributed system noise modelled for movement at time step 𝑘 − 1.

By combining Equation 5 and the equation of motion shown below: 1 𝑋𝑘 = 𝑋𝑘−1 + 𝑋𝑘−1 𝑇 + 𝑋𝑘 −1 𝑇 2 2 where T is the period between time steps in a discrete system we obtain:

(7)

+ 𝑥𝑘− 𝑥𝑘−1 − + 𝑦𝑘 = 𝑦𝑘−1 + 𝑈𝑘 −1 + 𝑤𝑘−1 − + 𝜃𝑘 𝜃𝑘−1

(8)

The acceleration term of the motion equation is considered to be zero due to the discrete manner of the + system. Equation 8 is the state transition model 𝑓 𝑋𝑘−1 , 𝑈𝑘−1 + 𝑤𝑘−1 for the omnidirectional vehicle with relative displacement control inputs. 𝑈𝑘−1 = 𝑇

𝑟 2

𝑥𝑔 𝑦𝑔 + 𝑞𝑘 −1 𝜃𝑔

(9)

Where 𝑞𝑘−1 is the independent Gaussian distributed, zero mean noise between relative displacements and is taken into consideration in the system noise 𝑤𝑘 −1 . 𝑤𝑘−1 ~𝑁(𝑤𝑘−1 , 𝑄𝑘−1 )

(10)

Where the mean 0 𝑤𝑘−1 = 0 0 And the state transition error covariance 𝑄𝑘−1 = 𝐸 (𝑤𝑘−1 − 𝑤𝑘−1 ) 𝑤𝑘 −1 − 𝑤𝑘−1 2 𝜎𝑘−1,𝑥 0 = 0

2 𝜎𝑘−1,𝑦

0

0 0

0

2 𝜎𝑘−1,𝜃

607

𝑇

(11)


The diagonal of the covariance matrix represents the variance between the 𝑥 , 𝑦 and 𝜃 measurements respectively. The off diagonal elements of the matrix are zero due to assumption that the noise between location estimates 𝑥 , 𝑦 and 𝜃 are independent of each other [7].

Landmark Measurement Model To obtain the posterior state estimate the EKF will compare actual range measurements from an ultrasonic rangefinder mounted on the front of the vehicle with an estimated measurement obtained from the prior state estimate and the observation model. 𝑧𝑘 = 𝑕 𝑋𝑘− , 𝐿𝑘 + 𝑣𝑘 = 𝑕 + 𝑣𝑘

(12)

Where 𝐿𝑘 is a known land mark location and 𝑣𝑘 is the zero mean, independent Gaussian distributed noise in the sensor readings.

Figure 3: Diagram showing measurements of a landmark 𝑖 relative to the vehicles priori state estimate. From figure 3 above, the estimated distance 𝑟𝑖 and orientation 𝜃𝑖 between the vehicle and the landmark 𝑟𝑖 can be determined. 𝑟𝑖 =

𝑥𝑘− − 𝑥𝐿,𝑖

𝜃𝑖 = tan−1

2

+ 𝑦𝑘− − 𝑦𝐿,𝑖

𝑦𝑘− − 𝑦𝐿,𝑖 𝑥𝑘− − 𝑥𝐿,𝑖

2

+ 𝑣𝑟,𝑘

+ 𝑣𝜃 ,𝑘

(13)

(14)

Therefore 𝑟𝑖 𝑧𝑘 = 𝜃 + 𝑣𝑘 𝑖

(15)

𝑣𝑘 ~𝑁 𝑣𝑘 , 𝑅𝑘

(16)

Where

With mean

608


𝑣𝑘 =

0 0

And the measurement error covariance 𝑅𝑘 = 𝐸 𝑣𝑘 − 𝑣𝑘 𝑣𝑘 − 𝑣𝑘 =

𝜎𝑟2𝑖 0

𝑇

0 𝜎𝜃2𝑖

(17)

As with the system noise, the noise of the distance estimate and the noise of the orientation estimate are assumed to bare no relation to each other and the off diagonal elements of the measurement error covariance 𝑅𝑘 are taken to be zero [7]. 5.

THE EXTENDED KALMAN FILTER The extended Kalman filter is a recursive data processing algorithm which is able to accurately estimate the state of a noisy, non-linear, dynamic system [7], [8]. It is able to do this by using Taylor series to linearize the estimation about the current estimate by using partial derivatives of the process and then minimise the mean squared error of the state estimations [9].

Figure 4: Diagram showing the recursive manner of the EKF with the Time update and Measurement update. Figure 4 above shows the recursive manner of the EKF as well as the algorithms that are needed to obtain accurate state estimations from time step to time step. Where 𝐴𝑥,𝑘 =

𝜕𝑓 𝜕𝑥

+ 𝑥=𝑥 𝑘−1 ,𝑢=𝑢 𝑘−1

(18)

(19) And

609


𝐻𝑘 =

𝜕𝑕 𝜕𝑥

𝑥=𝑥 𝑘−

(20)

The innovation or residual is the difference between the estimated measurements and the actual measurements made by the vehicle 𝑧𝑘 − 𝑧𝑘 . The Kalman gain 𝐾𝑘 has a weighted average effect on the innovation. Should the error covariance 𝑅𝑘 of the estimated measurement be large then more of the actual measurements will be taken into consideration and vice versa [7]. 6.

LOCALISATION AND CONTROL

Figure 6: Flow representation of the three localisation and control process loops. The control and localisation will take place in three loops. The inner loops for wheel angular velocity control, the outer loop for vehicle position control and the middle loop for error correction using the EKF. The EKF requires a large amount of computational resources when implemented on a robotic platform [10]. The computational complexity can be reduced somewhat by allowing multiple prior state estimations to be computed by using the outer and inner loops before applying EKF corrections or the middle loop.

7.

CONCLUSION The material handling systems that are used in manufacturing environments are becoming fast out dated as industry moves towards highly efficient manufacturing techniques. These manufacturing techniques call for robust material handling systems with high flexibility that takes advantage of modern computational techniques and hardware availability.

610


Localisation by the extended Kalman filter is a modern computation that would be suited for high noise manufacturing environments. With high powered microprocessors becoming more and more affordable comes the ability to use the extended Kalman filter in a decentralised, on board manner. A mechanical system was proposed allowing AGVs more manoeuvrability in a manufacturing environment and all the parameters required for the vehicle to utilise the EKF were derived. This allows for a system with high flexibility as far as movement paths are concerned and the robustness and reliability of the popular EKF to be implemented in a manufacturing environment. 8.

REFERENCES

[1] [Mikell P. Groover, Automation, Production Systems and Computer-Integrated Manufacturing, 3rd ed. Upper Saddle River, New Jersey, United States of America: Pearson, 2008. [2] Michael Csorba, "Simultaneous Localisation and Map Building," Department of Engineering Science, University of Oxford, Oxford, Thesis 1997. [3] The RoboCup Federation. (2011) RoboCup. [Online]. http://www.robocup.org/ [4] T.A Baede, "Motion Control of an Omnidirecional Mobile Robot," Department of Mechanical Engineering, Eindhoven University of Technology, Eindhoven, Traineeship Report 2006. [5] R. Penrose, "A Generalized Inverse for Matrices," Cambridge University, Cambridge, Mathematical Proceedings of the Cambridge 1955. [6] R.P.A van Haendel, "Design of an Omnidirectional Universal Mobile Platform," Eindhoven University of Technology, Eindhoven, DCT traineeship report 2005. [7] Rudy Negenborn, "Robot Localization and Falman Filters," Institute of Information and Computing Science, Utrecht University, Copenhagen, Thesis Number: INF/SCR-03-09, 2003. [8] Rudolf Kalman, "A New Approach to Linear Filtering and Prediction Problems," Journal of Basic Engineering, March 1960. [9] Greg Welch and Gary Bishop, "An Introduction to the Kalman Filter," University of North Carolina, Chapel Hill, 2006. [10] Hugh Durrant-Whyte and Tim Bailey, "Simultaneous Localisation and Mapping (SLAM): Part 1 The Essential Algorithm," IEEE Robotics and Automation Magazine, vol. 2, 2006, Available at: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.128.4195.

611


TRANSMISSION LINE INSPECTION ROBOT FOR MALAYSIA: PROTOTYPE IMPROVEMENT *Khairul Salleh Mohamed Sahari1, Justin Chan1, Sarah Navita1, Adzly Anuar1, Syed Sulaiman Kaja Mohideen2, and Mohd Zafri Baharuddin2 1 Department of Mechanical Engineering, Universiti Tenaga Nasional Kajang, Malaysia *e-mail: [email protected] 2 Department of Electronics & Communication Engineering, Universiti Tenaga Nasional Kajang, Malaysia

ABSTRACT Transmission line inspection has always been related to high risk and high cost. Normally, inspection is done online, making it even more challenging. Hence, to solve the problem, a transmission line inspection robot is proposed. The inspection robot must be able to avoid obstacles on the line as it traverses along the power transmission line. Since the transmission system may differ from country to country, it is important to develop an inspection robot that suits the Malaysian transmission system environment. An initial prototype was developed in 2009. This paper presents a new traversing mechanism for the transmission line inspection robot. The new design reduces the number of actuators while providing the same obstacle avoidance capabilities to the robot. The development includes mechanical analysis to ensure that the final product is workable. Tests are conducted to check whether the proposed mechanism works or not. Keywords: transmission line inspection, robot, obstacle avoidance 1.

INTRODUCTION Electrical power transmission is the process to transfer electrical power to consumers. Electricity is transmitted in high voltages; 110 kV and above using the transmission line [1]. After a certain period, the line has to be checked and inspected to ensure that the lines do not suffer from break, abrasion, and corrosion mainly due to the aging of the material over time. Other than that, the transmission line has to be inspected for hot spots to prevent further power lost during the transmission. Dangerous it might seem, the inspection work is carried out manually by a group of trained workers. These workers have to wear insulated suit and climb and crawl along the transmission lines. Thus, there is a high chance of workers falling from high places or getting electrocuted.

Hence, to solve the problem, inspection robots are designed and developed around the world. There are a few researches doing research on development of transmission line inspection robots [2]-[6]. These robots are developed based on the transmission system in the researchers‘ countries and might not function properly in the transmission system in Malaysia. Due to this, a transmission line inspection robot capable of avoiding obstructions encountered along the transmission lines (500kV, 275kV, 132kV) in Malaysia was developed [7]. The robot developed has 9 actuators for traversing and obstacle avoidance and weighs 12.5kg. In order to further improve the robot, a new mechanism for the arm is proposed. The mechanism reduces the number of actuators required from 9 to 6. Mechanical analyses are conducted prior to fabrication to check and optimize the parameters of the design. Lab tests are conducted to check the traversing and obstacle avoidance capabilities of the new developed arms. 2. 2.1

MECHANICAL DESIGN Objectives and requirements The main objective of the development of the new arm is to improve the existing prototype of the inspection robot in terms of weight and complexity of control.

612


Before the design can be started, it is important to understand the problems and engineering specifications. The followings are the requirements when designing the inspection robot:-

a)

Semi/autonomous travelling

b) The weight of the robot is within 10kg c)

Able to bypass obstructions along the transmission line as shown in Figure 1 [5]

Figure 1: Obstructions found along the transmission line [5] 2.2

Study of the existing prototype In order to develop an improved and better design of the transmission line inspection robot, a detailed study has to be conducted on the existing prototype. The study includes a psychical examination of the robot and a thorough stress analysis on the critical parts of the prototype. The purpose of conducting the test is to identify the weaknesses of the previous prototype. From the data obtained, the design phase will proceed and new concepts can be developed.

The previously developed transmission line inspection robot possesses 3 arms which act as grippers to move along the transmission line and also to avoid obstacles. The previous robot has been tested in the lab on actual 132kV transmission line and the results showed that the prototype was able to move along the transmission line and avoid obstacles (Figure 2). However, after some research and careful observation, there are a few weaknesses on the design:a)

The mass of the existing prototype has exceeded the allowance limit which is 10kg

b) When the arm rotates more than 20 degrees, it will just fall down and unable to rotate back into the original position c)

There are too many motors used in the prototype (9 motors)

Computational analysis of critical parts likely to fail first is conducted to determine their safety factors. The followings are the parts which are selected:a)

Shaft

b) Pin holes and pins c)

Arm of the prototype at various positions

d) Worm gears and worm shafts

613


Figure 2: Previously developed prototype [7] The complete calculation can be found in the thesis entitled ―Design and Development of UNITEN Transmission Line Inspection Robot‖ [8]. From the analysis and calculation, it is found that all the parts of the previous prototype have safety factors (SF) well beyond the allowable safety factor of 1.0. For the worm gear calculation, an alternative method is used to determine whether it is safe or not. In the end, the results of the analysis show that the worm gears are also safe in the design. Based on the analysis, it is concluded that the size of several parts can be reduces to optimize the material usage while maintaining the SF at more than 2.0. Other than that, it is also important to reduce the weight by reducing the number of components and fix any other weaknesses on the prototype. 2.3

Conceptual Design The design phase of the new prototype is based on a series of steps conducted accordingly in order to ensure high quality end product that meets the objectives and requirements. Two concepts were generated using Working Model. Both concepts use linkages to create movement at the arm and the bottom part of the gripper by using only 1 motor. Concept 1 is the earliest concept created with the objective to decrease the usage of the motors and to reduce the weight of the arm. Basically, the mechanism uses a power screw to translate the rotation motion to linear motion. Thus, the force will push the scissors mechanism up. The scissors links are connected to the arm of the robot. Therefore, as the scissors links rise up, the arm will then move at the same time following the path of the slot. This creates a movement that will release the gripper from the transmission line. Concept 2 is almost similar to the first concept but with more refined design on the linkages. With this mechanism the number of linkages are reduced (the slider and the extra link connecting the slider and the arm are removed from the mechanism). The pin is the slot is design with a curve shape to ensure a smoother movement of the arm compared to the first concept.

2.4

Concept evaluation and selection The comparison of both the concepts is shown in Table 1.

614


Table 1: Comparison between the concepts generated

Concept 2 has been chosen based on its simplicity and efficiency compared to concept 1. Further simulation is conducted using Working Model to ensure the followings: i)

The mechanism is able to release the gripper from the transmission line when there is an obstacle.

ii) There is no collision with other components when the robot arm is moving and re-orientating itself. iii) The trajectory of the mechanism during obstacle avoidance is minimized to avoid collision with other lines The new mechanism is supposed to be better as the amount of motors will be reduced from 9 to 6. Besides that, the lower gripper is designed much simpler without the ball screw. These will reduce the weight of the robot. The usage of power screw is also an advantage as the movement of the robot will be more precise. The linear motion of the mechanism depends on the rotation of the screw which will then affect the lead of the thread. There will be a groove slot for the arm to move along with and to support the arm. Therefore, the arm can rotate till the maximum limit while having enough torque to rotate back to its original position. The movement of the mechanism is shown in Figure 3.

615


Figure 3: Simulation of the proposed arm mechanism Based on the results of the simulation, the design is found to be too complicated and involves a lot of links. Hence, the cost of fabrication of the prototype will be expensive. Besides that, with each additional link, the mechanical efficiency of the mechanism will decrease due to friction. Therefore, after consideration, the design of the robot is simplified by replacing the scissors linkage with ball screw. The rotation of the ball screw will cause the nut to move upwards. As the arm moves up, it follows the curve of the slot. The bottom part of the arm will have a pin joint with a support to enable it to rotate as the arm follows the curve of the slot. The detailed drawings are prepared using Solidworks and the 3D assembled drawing is shown in Figure 4. The movement of the arm mechanism done in Solidworks is shown in Figure 5.

Figure 4: 3D drawing of the robot prototype

616


Figure 5: Simulation of the improved arm mechanism 3.

FABRICATION OF PROTOTYPE Based on the drawings, the new prototype is then fabricated and assembled. The motors used for the previous prototype are reused Motor model SPG30-20K (Cytron Technologies) with rated torque of 78.4mNm and speed of 185rpm is used for the rollers used for traversing on the line while the motor model SPG30-30K (Cytron Technologies) with rated torque of 3.43Nm and speed of 15rpm is attached to the ball screw. Figure 6 shows the assembled prototype. The current prototype weighs 10.5kg, a little over the desired weight of 10kg.

4.

CONTROL SYSTEM Since the robot is designed for semi-autonomous or autonomous control, the control system has to be able to cater for both requirements. Currently, for initial testing purposes, the control system for semiautonomous control is designed and developed with a wireless Play Station 2 (PS2) controller used as the remote controller. Figure 7 shows the fundamental arrangement of the preliminary control system used. The hardware consists of an H-bridge, an AT-Mega 8535 controller board (Figure 8), a PS2 controller adapter and a PS2 wireless controller.

PS2 wireless controller PS2 controller adapter AT Mega 8535 H-bridge DC motor Figure 7: Flow chart of basic control system

617


Figure 8: AT Mega 8535 controller board 5.

TESTING OF PROTOTYPE For evaluation purposes, the robot is tested whether or not it will interfere with the transmission line when avoiding the obstacles. The followings are the concerns regarding the movements of the robot during the testing:a)

The arm might not be able to avoid the obstacles.

b) During the process of loosening and tightening of the gripper, the arm might interfere with the transmission, causing possible damage to the line. The testing results are shown in Figure 9. From testing, it can be concluded that the developed mechanism can successfully bypass obstacles on the transmission line with radius of not more than 9cm.

618


Figure 9: Lab testing of newly developed arm mechanism 6.

CONCLUSION A new improved mechanism for arm of semi-autonomous transmission line inspection robot is proposed and developed. The proposed mechanism is successfully tested and will enable obstacle avoidance along the transmission lines. Although the development of the robot is currently still at its early stages, it serves as solid platform for further development. Future work includes installation of cameras and sensors onto the robot and where obtained images and data are transmitted back to the ground workstation via wireless communication system. Study on EMF effects is also taking place. This is vital to ensure that the robot is not damaged during online inspection.

7.

ACKNOWLEDGEMENTS Authors would like to thank UNITEN for the funding of the project under research grant UNITENJ5100525.

8.

REFERENCES [1] Wikipedia, ―Electrical Power Transmission,‖ July 2009. [Online]. Available: http://en.wikipedia.org/wiki/Electric_power_transmission [Accessed: 20th July 2009]. [2] F.Y. Zhou, J.D. Wang, Y.B. Li, J. Wang and H .R. Xiao. ―Control of an Inspection Robot for 110KV Power Transmission Line Based on Expert System Design Method‖. Proceedings of the 2005 IEEE Conference on Control Application, Torronto, Canada, August 28-31, 2005. pp. 1563-1568. [3] S. Fu, Z. Liang and M. Tan. ―Visual based navigation for Power Transmission Line Inspection Robot‖. Proc. 7th IEEE Int. Conf. on Cognitive Informatics (ICCI‘08), 2008. [4] P. Debenest, M. Guarnieri, K. Takita, E.F. Fukushima, S. Hirose, K. Tamura, A. Kimura, H. Kubokawa, N. Iwama and F. Shiga. ―Expliner – Robot for Inspection of Transmission Lines‖, 2008 IEEE International Conference on Robotics and Automation, Pasadena, CA, USA, May 19-23, 2008, pp. 3978-3983. [5] S.Y. Jiang, Y.J. Hu, Y. Wang, H.Z. Jiao, and L.M. Ren. ―Development of Hanging-Arm Inspection Robot for High-Voltage Transmission Line‖, ICIRA 2008, Part I, LNAI 5314, pp. 1089–1098. [6] H. Harano, K. Syutou, T. Ikesue and S. Kawabe, ―The development of the Manipulator method for 20kV class overhead distribution system‖, Transmission and Distribution Conference and Exhibition, 6, 2002, pp. 2112-2116.

619


[7] S.Y. Koh, Khadijah H., Khairul Salleh M.S, Syed Sulaiman K.M., Adzly A. ―Development of Transmission Line Inspection Robot‖, Proceedings of Second International Conference on Control, Instrumental and Mechatronics Engineering (CIM09) Malacca, Malaysia, June 2-3, 2009. [8] J.C.T. Leong. ―Design and Development of UNITEN Transmission Line Inspection Robot‖, Universiti Tenaga Nasional Bachelor of Mechanical Engineering Final Year Project Thesis, 2010.

620


A NOVEL VIDEO IMAGE SMOKE DETECTION SYSTEM BASED ON WAVELET & COLOR INFORMATION ANALYSIS Chai Yoon Yik 1, Ewe Soo Chait 2, and Loo Chan Perng 3 1

INTI International College Penang School Of Engineering And Technology Malaysia e-mail: [email protected] 2

INTI International College Penang School Of Engineering And Technology Malaysia e-mail: [email protected] 3

INTI International College Penang School Of Engineering And Technology e-mail: [email protected] ABSTRACT A novel method to detect smoke in video is proposed. Video image regions containing moving objects are detected and segmented. These segmented regions are analyzed spatially using different type of wavelets processing for edge energy variation. Evaluation results show that ‗db1‘ wavelet has the highest ability to differentiate smoke and non-smoke region. The segmented regions also tested for RGB value variation to confirm any moving objects in those regions. A special flickering model based on temporal variation in edge energy content of the video frames is developed to extract any smoke clue. All the clues extracted from different algorithm are ―AND‖ together to produce the final decision. Overall performance showed 98% success smoke detection rate and zero false alarm. The worst case detection time delay is 9-10 seconds. Keywords: wavelet decomposition, edge energy content, flickering and RGB. 1.

INTRODUCTION A smoke detector is a device that detects smoke and issues an alarm to people. It may alert people within hearing range, and also interface with a security system. Conventional smoke detector works either by optical detection (photoelectric) or by physical process (ionization) or both. These methods make the conventional smoke detector have distance limitation and high probability of failure in an open or large area. The main purpose of this project is to overcome the weakness of conventional smoke detector. It has the ability to detect smoke at a greater distance and also a wider area. Furthermore, it should improve the probability of early smoke detection.

2.

METHODOLOGY The characteristics of the smoke are used to determine the presence of smoke in a particular region. Smoke which is naturally in random moving nature, regardless fast or slow movement, a smoke moving region detection must be implemented to extract those moving regions because the non-moving regions would not contribute to smoke detection, yet incurred more processing time. Therefore, moving regions detection process will be able to greatly reduce the amount of data for subsequent data processing and analysis. Smoke always obstructs the texture and edges of the original background. Edges contribute by the high frequency content of a video image. As smoke is normally semi-transparent, the original edges at the background still visible. This leads to a decrease in the edge energy content [5]. Generally, low density smoke is greyish and semi-transparent initially. This means that the smoke would preserve the colour information of the original background with little discrepancies [2]. This allows RGB colour value from all moving regions data to be used to determine the presence of the smoke.

621


Smoke does flicker due to its movement. As smoke flicker, the smoke at the same location would end up to be having different thickness at different times. Different amount of smoke could cause different amount of edges to be blocked, for example, a thin smoke obstruct a few edges only, but a thick smoke could obstruct almost all edges. This would cause variation to the edge content energy in different time. The statistic of energy content can be analyzed to determine the presence of the smoke. To minimize the possibilities of false alarm, an intelligent set up is required. Since each video image frame was divided into certain number of blocks at the very beginning of the process, if the number of detected smoke region blocks in a video frame is less than 2 blocks, the frame is ignored. The reason is smoke constantly moving with time. If it is a real smoke, eventually the blocks detected with smoke would exceed 2. This intelligence is combined with the precedence processing outcomes to generate an accurate final decision. The proposed smoke detection algorithm consists of five steps: (i) moving regions detection (ii) edge energy content reduction check (iii) RGB data discrepancy check (iv)flickering phenomenon check (v) intelligent decision making routine to minimize the false alarm.

Start

Input Video image

Initialize Background

Input Video image

Moving Region Detection

Identify moving region blocks

Edge Content Detection

Colour Information Analysis

Flickering Analysis

No

Two or more blocks pass smoke detection criteria?

Yes Trigger an alarm

Figure 1 Smoke Detection Algorithm

622


3.

RESULT AND DISCUSSION

3.1 Moving region detection Moving objects in video can be detected using the background estimation method [6]. The camera must be stationary so that every frame is compared with the correct background frame. Moving regions pixels are determined by subtracting the current image from the background image and subsequently gone through a thresholding process [1]. An adaptive thresholding based on the Otsu‘s method [3] is used. The result is simple yet accurate in moving region detection. To improve the efficiency of the subsequent process, the frame is divided into blocks. Any blocks contain the moving image would be subjected to further analysis. 3.2 Edge Content Detection To distinguish the smoke from ordinary moving object, edge content analysis is carried out. Smoke generally obstructs the texture and edges of the background. By performing a wavelet-decomposition on the image blocks [9], we can separate the image signal into high frequency components and low frequency components [4][7]. The energy content of the edge can be calculated based on the high frequency components. From Figure 2, it shows that edge energy content drop in higher level of smoke density which due to the edges of the image become more blurred. However, the edge energy content will not become zero even with very heavy smoke density. This is means even though in heavy smoke condition, the smoke also forms edges and contributes very small amount of energy content.

Figure 2 Edge energy content analysis at different level of smoke density

To determine which type of wavelet suitable for energy content analysis, various experiments in various conditions such as smoke density, light intensity and the presence of moving object were carried out. Based on the evaluation data, db1 wavelet was selected because it is able to produce the largest edge content energy based on the decomposition coefficient. It also gives lower edge content energy variation for control data when compared to other wavelets output. The larger edge content energy of db1 is due to the smallest wavelet support length [4]. The db1 wavelet also able to reduce the camera inherent variation effect on energy content variation base on non-overlapping of both control and test frame (with smoke) data. The edge energy content of different frames calculated based on db1 and other wavelets are shown in Figure 3.

623


624


Figure 3 Edge energy content analysis using different wavelets

The effect of brightness variation and moving object data analysis based on the graphs shown in Figure 4 (a) and (b) indicates:

(i) excessive increase or decrease in the light intensity would cause the visibility of the image‘s edges decrease and as a result, the edge energy content decrease.

(ii) moving object generally will cause extreme changes in the energy content. If a moving object

has more edges as compared to the background frame, it will cause an increase in the energy content. On the other hand, a moving object with less edges as compared to the background frame will cause a decrease in the energy content. However, the energy content for frames with smoke will never reach zero even if the smokes is very thick for two reasons. First, smokes is semi-transparent in nature and second, some of the thick smoke begins to possess some edges of their own.

625


Figure 4(a) Edge energy content variation for static (reference) background, low and brightness level

Figure 4(b) Edge energy content variation for static (reference) background and moving object.

Thus, the adaptive threshold level for edge content detection is calculated based on

Threshold = average energy content for background frames – 3*standard deviation of energy content Furthermore, this threshold is also adaptive to different background image. When edge energy content of the current frame less than the threshold level, then it is considered smoke detected. The same data set are qualified for the following processing to confirm smoke existence validity. Figure 5 clearly shown that the existing of the smoke in the image will be detected as its edge energy content less than the calculated threshold level.

626


Figure 5 Smoke is detected if the edge energy content lower than the threshold level. 3.3 Color Information Analysis The blocks qualified by the edge energy content are subjected to colour information analysis. When the smoke starts to expand, its semi-transparent nature will preserve the RGB data of the background image with little discrepancies. However, when a moving object passes through, the original background image is blocked. The RGB data of the background image undergoes a significant change compared to the reference RGB data from the background. In this project, only the red pixels are analyzed. Based on the evaluation data, the difference in red pixel‘s R value caused by the smoke normally falls in the range of 10 to 30. In addition, the inherent camera noise will produce red pixel value variation of about 10. Therefore, minimum R value changes to classify a data block has significant or valid R value changes, classification threshold is set to 10. To determine the upper limit of R value changes, based on moving object R value changes in Figure 6, the moving object could have easily caused an average of changes in red value of about 50. When evaluating moving object under a huge variation of light intensity, the change in R value showed an average of around 35. When smoke exist and without any moving object or light intensity change, the data shows a change of less 25 for R value. Therefore, by consolidating all these observation, an upper limit of 30 is appropriate to detect smoke. The colour information analysis will generate positive signal when any block average R value change within 10 to 30, otherwise, considered no smoke detected.

627


Moving object alters the RGB value significantly

Figure 6 Moving object could have easily cause an average of discrepancies in red pixel of about 50+.

3.4 Flickering Analysis Flickering analysis is based on the statistic of the edge energy content variation. In the presence of smoke, the edge energy content variation magnitude measured by its standard deviation may be either very much higher or lower when compared to the variation caused by the camera inherent noise. In order to make the classification more accurate, outlier due to camera noise must be filtered using an adaptive filtering. The adaptive filtering spec is tied to a statistical value i.e. the average and standard deviation of the dataset itself. Since this behaviour also related to the smoke density and image background, to distinguish a variation that is caused by the smoke or non-smoke object, the standard deviation of a moving 20 frames is divided by reference background frames standard deviation (based on 20 frames) to produce a Ratio A. Unfortunately, the Ratio A alone is not capable to detect all the frames with smoke as shown in Figure 7 and 9, but it will not produce any false alarm. This problem partially solved by using both the Ratio A and its inverse together with a threshold 3.5 based on evaluation data. The presence of smoke which could not be detected using ratio A can be detected using the inverse of the ratio A as shown in Figure 10.

628


Figure 7 Ratio A

Figure 8 Inverse of the ratio

Figure 9 Ratio A

Figure 10 Inverse of the ratio

629


4.

OVERALL EXPERIMENT RESULTS The proposed method is implemented in a PC with an Intel® Core™ 2 Duo CPU T6400 @ 2.00GHz processor and tested for a large variety of conditions for real time and off-line videos containing various conditions such as video with smokes, shadows, illumination changes and moving objects. The overall computation time is about 450msec for frame size of 576 by 720. This contributes to worst case delay of 9 seconds. Testing results presented in Table 1 shows that smoke density in different stage (stage 1 refer to minor level to stage 4 which is thick) was successfully detected in all of the video clips containing smoke. No false alarms were issued in the live video and recorded off-line videos recorded. Even though the smoke in stage 1 has 20% chances detected, this is not a serious problem because the rate of the video frame at 25 frames/sec and subsequent stage detection is almost 100%.

Table 1 Testing results

5.

CONCLUSION An algorithm combining moving region detection, edges energy content checking, colour information analysis and flickering analysis together with adaptive thresholding and intelligent decision making routine capable to detect the smoke in a region based on video images without false alarm. The average processing speed of 450msec seconds/frame is considered reasonable. Nevertheless, the 20 frames flickering analysis contribute to worst case about 9 seconds delay before triggered an alarm can be further improved.

6.

REFERENCES [1] Yi˘githan Dedeo˘glu (August 2004) ―Moving Object Detection, Tracking and Classification for Smart Video Surveillance‖, Master of science Thesis. Department of computer engineering, Institute of engineering and science Bilkent university, 06800, Bilkent, Ankara, Turkey [2] Arturo Donate and Eraldo Ribeiro, ―Viewing Scene Occluded By Smoke‖ , ISCV 2006, LNCS 4292 pp.1666-1675 [3] Otsu, N., "A Threshold Selection Method from Gray-Level Histograms," IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, No. 1, 1979, pp. 62-66. [4] James S.Walker , ―A Primer on Wavelets and Their Scientific Applications(second edition)‖ [5] Ugur Toreyin, Yigithan Dedeoglu, A. Enis Cetin, "Contour Based Smoke Detection in Video Using Wavelets", 14th European Signal Processing Conference EUSIPCO 2006, Florance, Italy. [6] R.T.Collins, A.J. Lipton , and T.Kanade, ―A system for video surveillance and monitoring,‖ in 8 th Int. Topical Meeting on Robotics and Remote Systems. 1999, American Nuclear Society [7] Jun Li, ―A Wavelet Approach to Edge Detection.‖ Master of Science (Mathematics) Thesis, August 2003, Sam Houston State University. Huntsville, Texas [8] Osslan Oriris Vergara et al ―Digital Image Processing in Wavelet Domain‖ IEEE Looking Forward, vol. 13, pp. 13 – 16, Summer 2006

630


NOVEL 6 DOF HYBRID MACHINE DESIGN Ahmed Asif Shaik1, Prof. Glen Bright2, and Dr. Nkgatho Tlale3 1

Council for Scientific and Industrial Research, Pretoria, South Africa e-mail1: [email protected] 2 University of Kwa-Zulu Natal, Durban, South Africa e-mail2: [email protected] 3 Council for Scientific and Industrial Research, Pretoria, South Africa e-mail3: [email protected] ABSTRACT Robotic manipulators are a central part of high technology industries. They are used for operations such as component assembly, welding, cutting, spray painting, etc. Most robots used for industrial manufacturing have articulated arms equipped with a serial robotic architecture. This is not efficient as the arm itself contains significant inertia due to the location of the motors and gearboxes, which also contributes to inaccuracy and dynamic vibration problems. This research paper will focus on the design of a novel hybrid robotic arm. It is hybrid as its architecture does not fit the definition of either Serial Kinematics Machines (SKMs) or Parallel Kinematics Machines (PKMs) explicitly. It will aim to combine the best aspects of both architectures having an optimized workspace to footprint ratio comparable to a serial robot, with the inertia and speed of a parallel robot. These requirements are conflicting and do not exist concurrently in serial or parallel robot architectures. The design contains a few unique mechanisms which enables a full complement of 6 DOF (degrees of freedom). This hybrid manipulator presents a superior alternative to current industrial robotic arms. Keywords: Parallel kinematics, serial kinematics, hybrid machine, 6 DOF. 1.

INTRODUCTION Before continuing it is necessary to define what qualifies a machine as either PKM or SKM. In a serial topology each actuator axis is in line relative to the preceding one in an open kinematics chain, in other words each motor and gearbox is positioned close to the joint it controls. In a purely parallel topology, the actuator axes, (one for each DOF), have a fixed arrangement and position in space. From the fixed base, a number of arms and links are coupled in parallel to the end effector, forming closed kinematics chains (examples of which are the 3 DOF Delta and the 6 DOF Hexapod). The use of SKMs in industry is nearly ubiquitous, from automotive to nuclear power production environments. The reasons for this are due to the fact that the technology is mature and refined in its current state, as well as the versatility of the SKM to accomplish virtually any task. A few shortcomings of the serial robot lead to the development of specialized PKMs and their adoption in a few industries, most notably the packaging of items, where speed (throughput) and accuracy is paramount. PKMs have severe disadvantages themselves, and their attributes are generally complementary to SKMs. PKMs typically are faster, stiffer and more accurate whereas SKMs have a large useful workspace, and are more dexterous and versatile. One route to improve manufacturing processes is to create robotic manipulators that possess the advantages of both types of architectures, and that is the focus of this paper. The objective is to create a robotic mechanism, that would have the same appeal to industry as the SKM (versatile, large workspace and minimal machine footprint) but coupled to that it would also possess some of the advantages of a PKM (lower inertia, higher speed and greater accuracy). These combined advantages can only be achieved through a hybrid machine design architecture. As automation and flexible production in today‘s industry increase, new applications require higher performance from industrial robots. Current pure serial and pure parallel robot technology are limited in their respective rights, however hybrid structures can provide a greater potential for improvement.

631


1.1

Comparison of Parallel and Serial Robot Architectures Industrial robots with parallel architecture have the advantages of increased stability and arm rigidity. Additionally there is high repeatability (due to reduced arm flexing) and the high stiffness of a closed-loop kinematics structures allows it to exert strong forces in its workspace. [1] As the motors which are responsible for most of the manipulators inertia are positioned in a fixed arrangement on a fixed base, the speed of displacement of the end effector is greater. [2] The load of the end effector is also distributed among its many legs, and may be purely axial but is dependent on the machine configuration. [3] For serial robots each link in the kinematics chain is required to support not only the masses of its link and motor, but also the masses of all the links and drive units preceding it. The inertia is therefore considerable when compared to PKMs, and this limits the capability of the robot in terms of its dynamic performance and acceleration. [4] Additionally each link in a serial machine has flexing errors which are additive, which results in a higher total end-of-arm flexing error as compared to PKMs. Generally all errors, such as manufacturing errors, gear backlash, hysteresis, etc. resulting from a serial structure are cumulative and are amplified. PKM structures by contrast have the effect of averaging all errors. Moreover using large displacement compliant joints can improve the error averaging effect in PKMs which can lead to sub micron positioning accuracy. [5] PKMs are less sensitive to temperature, have lower energy consumption, a lower manufacturing cost and are more reliable. They offer good design variation allowing designers to stretch their creativity and conceptualize machines with varying architectures, far more than they could do with serial topologies. PKMs tend to have a larger footprint to workspace ratio which is the most significant disadvantage. This is due to the positioning of the motors and the resulting configuration. There are some exceptions, however most designs take up a large work area. Also the performance of PKMs is greatly dependent on their geometry and as such optimal design has therefore become a necessity in their development. [6] Payload variations on the end effector drastically affect the machine behaviour. This is due to the fact that the ratio between machine moving mass and payload is significantly lower than in SKMs. Their complex kinematics and dynamic models also make control far more difficult than in serial machines. [1] Most PKM research have been done on machines with six DOFs and they have a small useful workspace, are riddled with design difficulties and their forward kinematics is an extremely difficult problem. On the other hand the kinematics of parallel mechanisms with two and three DOFs can be described as closed forms. Moreover not all the singularities of a six DOF parallel mechanism can be found readily, but these are rapidly identified for PKMs with two and three DOFs. It is for such reasons that PKMs with less than six DOFs have been increasingly attracting more attention for industrial applications. [7]

1.2

Machine Novelty The concept architecture for the machine uses multiple 3 bar linkages, a unique concentric gearbox and dexterous wrist. The machine will mimic the 6 DOF (degree of freedom) motion of a typical serial robot and will have an equivalent footprint and workspace. Additionally all six motors and their associated gearboxes have a fixed spatial location at the base, and through those linkages and gearboxes they transfer their torque to the intended axis, this significantly reduces the inertia of the moving arm. The inertia can be further reduced with the use of high strength, light weight linkages using composite materials. There is still substantial research that has to be done to investigate the feasibility of this solution, however this does represent a first iteration of a workable solution. Hybrid machine designs are typically composed of both serial and parallel machine sections connected serially, and one can distinctly recognize the comprising sections of the machine as a whole. These hybrid designs aim to improve robotic manipulators by localizing a particular architecture at the point where its advantages are most necessary, thereby extracting the best features of each architecture. Typically two and three DOF PKMs are used as component building blocks. SKM components are used where a large range of

632


motion is needed or where less complicated or a non-existing PKM solution exists. The machine design described here has no clear distinction between a parallel or serial nature, and hence it describes a truly unique hybrid structure. 2.

HYBRID MACHINE DESIGN The machine aims to achieve actuation transfer from the motor and gearbox located in one position (the fixed base which makes it similar to parallel architecture) to the target axis located elsewhere, through a series of gears and light weight connecting linkages. The 2 ways in which this actuation transfer could be accomplished is via a rigid link and non rigid link actuation transfer mechanism. The non rigid link option would require the use of toothed belts or chains. However this option would restrict the load carrying and force exerting capability of the machine, however it offers increased manipulator speed. For applications that would require the placement of light objects in a large workspace this option would be suitable. This paper will focus on a design using rigid link mechanisms. The drawings illustrated here are from a rapid prototype of the platform. All links and gears were laser cut from Perspex, and as such to work around the problem of not having bevelled gears the gear teeth were made large enough so that they can mesh at 90 degrees (with their pitch circles tangent at 90 degrees). This worked fairly well, even though the gears now have point contact instead of line contact, and for a working model this was sufficient. Most of the drawings that will be illustrated show this, however it must be remembered that they represent bevelled gears, which will work exceedingly better. The design will be described from the bottom up. The motor units and their associated gear boxes (shown in figure 1a) have a fixed position in space (6 sets in all, shown in figure 1b and 1c). Their arrangement allows them to occupy dedicated space for themselves, and allows them to mesh with the gears of the next part of the design, the concentric gear drive. Those gears that mesh with the concentric drive have to be bevelled as they mesh at an angle, preferably at 90 degrees.

Figure 1 Motor units illustrating their position and arrangement on a fixed base a. Motors and their gearboxes b. Top view of motors and their arrangement c. 3D view of motors and their arrangement

633


The concentric gear drive consists of 7 concentric sections. The outermost section mounts on the fixed base and does not move relative to the 6 inner sections.

Figure 2: Cross Sectional Views of Concentric Gearbox The 6 inner sections are all capable of rotating independently of each other (just one degree of freedom, i.e. rotation), while remaining concentric. Each section holds its nearest inner section in place (the innermost section does not hold anything), via a double ball race bearing (illustrated in figures 2 and 3). The inner bearings do not carry a vertical load, they simply facilitate the transfer of rotation and torque from the base motors to the designated driver gears/links. The outermost bearing is the only one that carries a vertical load, which is the complete mass of the moving machine and the payload it carries. The 5 innermost sections have bevelled gears mounted on both the top and bottom halves of each section (see figure 2). The 6th section (counted from the inside moving outward) has a bevelled gear on its bottom half. On its top half it has a physical mounting for 5 bevelled gears that have all their axes concentric and which mesh with the top half gears of the concentric gear drive at 90 degrees (figure 4). The 6th section (outermost movable section, see figure 2) is responsible for moving the mobile parts of the machine arm about the vertical axis.

634


Figure 3: Exploded View of 2 Consecutive Sections

Figure 4: Bevelled gears mounting on top half of Concentric Gear Drive a. All vertical bevelled gears, with their structural mountings b. Bevelled gear meshing, with structural mounting hidden To transfer actuation away from the base, 3 bar slider-pivot linkages were used, these are illustrated in figure 5. The orbit of the follower need not be a 1 to 1 (the output link would trace a circle of the same radius as the driving link) or 1 to -1 (the output link would trace a circle of the same radius as the input link but in the opposite direction) match with the driving link. The follower links must however match the angular rotation of its driver (no longer positional magnitude); that is the orbit does not have to be a perfect circle but it has to circumnavigate the axis, i.e. the follower must have one complete orbit for every 360 degree rotation of its driver. This also implies that the torque (and rotational speed) delivered to the end of the linkage (which will be transferred to the wrist) will be varying.

The slider has a pivot at the midpoint (allowing the slider to rotate) of the supporting link (this minimises warping of the follower orbit and maintains a somewhat circular profile). Furthermore the follower on the end of the primary slider bar linkage becomes the driver to secondary stage. The orbit of the follower on the secondary stage is further warped but still circumnavigates the main wrist axis, and matches each degree of rotation of the driver on the primary stage.

635


Figure 5: Slider bar linkage a. Primary slider bar linkage b. Primary slider bar linkage connected to secondary slider bar linkage Our initial choice for this torque transfer linkage was a parallelogram but there was no simple mechanical solution to prevent the singularity position (when the parallelogram collapses, or adjacent sides become collinear, and the exit configuration in which it could either be the parallelogram or a crossed quadrilateral – crossed configuration parallelogram). We experimented with designs in which we used extra links to create double parallelograms, with a phase offset so that when one collapses the other prevents the crossed configuration. Another solution was to maintain a crossed configuration, and for this we used a moving slider-pivot joint between the longer sides of the quadrilateral (parallelogram in crossed configuration). The 3 link slider-pivot linkage in figure 5a above was the simplest solution to achieve the required objective. Three of the inner vertical bevelled gears, being driven by the concentric gear drive then serve as the driver links for the primary slider bar linkages. These then drive the secondary slider bar linkages that eventually control the orientation of the wrist through the wrist concentric drive (figure 6 a and b). The 4 th vertical bevelled gear controls the proximal arm (lower arm, as it is closer to the fixed base) whose midpoint holds the pivot axis that connects to the sliders on each of the 3 primary slider-pivot linkages (figure 6). It controls the elevation of the lower arm (proximal arm) with regard to the horizontal plane. The 5 th gear controls the driver of a 3 bar slider-pivot linkage whose follower controls the angle between the upper arm (distal arm) and the lower arm (proximal arm), see figure 6. The distal arm holds the axis that connects to the sliders of the 3 secondary slider bar linkages (figure 7). The slider pivot of the upper arm is located at the mid-point between the end rotational joints. This reduces the warping of the secondary stage follower orbit, much like with the primary slider bar linkages.

Figure 6: Lower Arm a. Vertical gear connections to links b. Simple slider pivot joint

636


The follower end points on the three inner slider-pivot linkages of the secondary stage then connect to three vertical bevelled gears respectively (which are mounted on the upper arm). Since the follower does not have a perfect circle orbit around the main wrist axis slots are cut into the gears and links allowing the follower to move in and out of a perfect circle trajectory/orbit (figure 8 d), these are slider-pivot joints. Those vertical concentric bevelled gears then mesh with the wrist concentric drive gearbox having four concentric sections shown in figure 8 a. This concentric drive again makes use of the double ball race bearing illustrated in figures 2 and 3, which allows each section to move independently of each other. The outer sections hold the inner sections in place. The outermost movable section (or the third section in the concentric gear drive for the wrist) rotates the wrist (this is the first axis) and has mountings for the inner 2 axes, of the 3 DOF wrist. With some additional gearing those remaining 2 axes are set at 90 degrees to each other and the 1st wrist axis, thus allowing a full 3 DOF orientation of the end effector.

Figure 7: Upper Arm a. Upper arm showing secondary slider bar followers connecting to wrist gear drivers b. Upper arm showing slider pivot joint

3.

CONCLUSION The design above has been implemented on a cheap Perspex model to prove the concept. Thus far we were able to prove 4 working degrees of freedom, 3 DOFs to position the wrist and 1 to orient the wrist . The makeshift bearings for the concentric gearing systems then became problematic for the last 2 DOFs. However the physical model of the 3 DOF wrist does work and orients the end effector as desired. The next design iteration will use proper bevelled gears and bearings that will be rapid prototyped via 3D printing. Our end goal is to build a full scale prototype which we can then test and compare to existing serial and parallel robots, and prove our claims of an improved design. We are currently in the process of acquiring sufficient funding to proceed with the research and development, and future research papers will focus on force transfer through the system, the effects of vibrations, the kinematics and dynamics of the machine, and other important metrics that will be used to then effectively compare the design to existing industrial robots.

637


Figure 8: Wrist a. Gear meshing at bottom of wrist concentric gearbox b. Top view of wrist concentric gearbox c. Complete Wrist d. Secondary follower gear with slider slot

Figure 10: Perspex Model

638


4.

REFERENCES [1] ―Mechatronic design of a parallel robot for high-speed, impedance-controlled manipulation‖ by L. E. Bruzzone, R. M. Molfino, M. Zoppi; Proceedings of the 11th Mediterranean Conference on Control and Automation, Rhodes, Greece, June 2003. http://www.dimec.unige.it/PMAR/ [2] ―Design of the ‗Granit‘ Parallel Kinematic Manipulator‖ by Alessandro Tasora1, Paolo Righettini2, Steven Chatterton2. 1 – Università degli Studi di Parma, Dipartimento di Ingegneria Industriale, Parma, Italy; 2 – Politecnico di Milano, Italy. Proceedings of RAAD‘05 – 14th International Workshop on Robotics, Bucharest, May 2005. [3] ―A parallel robot for the Strain Imager (SALSA)‖ by S. Rowe (ILL), Millennium Programme and Technical Developments. http://www.ill.fr/AR-02/site/areport/fset_96.htm [4] ―Modelling And Model Based Performance Prediction For Parallel Kinematic Manipulators‖ by JanGunnar Persson, Kjell Anderson; Engineering Design, Department of Machine Design; KTH – Royal Institute of Technology, Stockholm, Sweden. Presented at Mechatronics Meeting, Gothenburg, August 2003. [5] ―Design of Compliant Parallel Kinematics Machines‖ by Yong-Mo Moon, Prof. Sridhar Kota, Mechanical Engineering, University of Michigan. Proceedings of DETC‘2002 – Biannual Mechanisms and Robotics Conference, DETC‘2002/MECH-34204, Montreal, Canada, September-October 2002. [6] ―Multi-Criteria Optimal Design of Parallel Manipulators Based on Interval Analysis‖ by F. Hao, J.P. Merlet; INRIA Sophia-Antipolis, France, 6 July 2004. Journal of Mechanism and Machine Theory, Vol. 40, No. 2, p157-171, February 2005. [7] ―Two Novel Parallel Mechanisms with Less than Six DOFs and the Applications‖ by Xin-Jun Liu1, Jongwon Kim1, Jinsong Wang2; 1 – Robust Design Engineering Lab, Seoul National University, Seoul, Republic of Korea; 2 – Manufacturing Engineering Institute, Tsinghua University, Beijing, China. Proceedings of the workshop on Fundamental Issues and Future Research Directions for Parallel Mechanisms and Manipulators, Vol. 1, No. 1, p172-177, Quebec, Canada, October 2002.

639


THE ROBOFRIEND RESEARCH PROJECT Sami Salama Hajjaj1, et al2 1

FOEIT, INTI University College - Laureate International Universities, Malaysia e-mail1: [email protected] 2 Refer to team members list (at the end)

ABSTRACT The RoboFriend project is a student-based research project created to utilize elements of Human-Robot Interaction (HRI) in tackling the problem of motion planning in dynamic environment for robots with higher DOFs. By developing a humanoid robotic arm (9 DOFs), to be controlled using visual and/or audio input signals from users, and use these queues to learn to avoid similar obstacles in the future. This project was divided into 4 major sub-projects; development of the gripper (wrist to fingers), development of manipulator (shoulder to wrist), development of the electronics and controls, and finally development of the HRI interface. Each sub-project was given to an engineering student, to be completed as his/her final year engineering project. Being a student-based project, this work had a secondary objective; to investigate the effectiveness of incorporating robotics research in undergraduate engineering education. Challenges included the design/development of the humanoid gripper, due to its conflicting design parameters of flexibility and structural stability. At the end, the humanoid arm was made to response to human voice. Data also showed that research enhanced students‘ learning experience as students were able to demonstrate their recently acquired knowledge in contributing to this work. Therefore, it can be concluded that incorporating research in undergraduate engineering studies is effective. Keywords: Robotics, Human-Robot interaction, HRI, robotics research in undergraduate engineering education, motion planning in dynamic environment. 1.

INTRODUCTION Human Robot Interaction - or HRI - is the study of interactions between humans and robots. HRI is a multidisciplinary field with contributions from Human-Computer Interaction (HCI), Artificial Intelligence (AI), Robotics, natural language understanding, and Social Sciences. [1] This work attempts to utilize elements of HRI in attempting to tackle the problem of motion planning for robots with high degrees of freedom in a dynamic environment. The robot used for this work fits this criteria; it is a humanoid manipulator that mimics a simplified version of the human hand (9 DOFs). This work is student-based, i.e. a significant amount of research, development and construction is performed by final year undergraduate engineering students with the aid and supervision of the project leader. This work was divided into 4 sub-projects; development of the gripper (wrist to fingers), development of manipulator (shoulder to wrist), development of the electronics and controls, and finally development of the HRI interface.

2.

LITERATURE REVIEW Motion planning refers to successfully moving the end-effecter of a robot from point A to point B while avoiding obstacles. Other aspects may be considered in the motion planning analysis; these include obtaining the required speed and/or acceleration for a specific motion. For example, welding robots must be programmed with the right welding velocity, and it must travel on the welding line only, otherwise welding errors will form and cause weak welds. This ultimately results in product failure and production losses. If the environment surrounding the robot is fixed, i.e. does not contain any moving obstacles, then motion planning becomes straightforward. Any Robotics engineering student can perform trajectory analysis and develop complete settings for optimum positioning, speed, and acceleration of the given robot.

640


The challenge begins when objects and obstacles start to move around the robot. This makes the environment a dynamic environment. This gets ever complicated when these motions do follow a predictable pattern, i.e. they become random in nature, and this makes motion planning a challenge [1]. Druy et al showed that the source of the challenge can be traced to a lack of HRI awareness. Considering that the biggest sources of random motions around robots are in fact humans, they argued that by incorporating HRI principles in the design phase of robots, this problem could be solved. [2] Although much progress has been made in robotics research in the last decade, relatively little progress has been made in optimizing the partnership between people and robots through improved techniques for HRI [2]. This is amazing considering that humans work with robots all the time in the industry. In fact, the closer the human and the robot get, the higher the risk of injury However, in the industry solved this problem by enforcing the use of zones, whereby human and robot do not share the workspace at any time. Unfortunately, as robots get more and more complex and more interactive and automated, this separation became impractical and therefore a need to incorporate HRI in the design of robots has raised, as such, guidelines for incorporating HRI in robots design and programming. The recommendations were as follows: [2]  

Enhance awareness, provide a map of where the robot has been. Provide more spatial information about the robot in the environment to make operators more aware of their robots‘ immediate surroundings. Lower cognitive load, provide fused sensor information to avoid making the user fuse the data mentally.



Increase efficiency, provide user interfaces that support multiple robots in a single window, if possible. In general, minimize the use of multiple windows.



Provide help in choosing robot modality. Provide the operator assistance in determining the most appropriate level of robotic autonomy at any given time.

If HRI was linked to HCI, or Human Computer Interaction, as well as to Ergonomics, then this linkage offered a rich resource for research and design in human-robot interaction. This was argued by Kiesler et al. Much less has been learned in the last three decades about how people perceive and think about computerbased technologies. If this could be linked to robotics, then better development of robotics design and programming could be achieved. [3] Adams, J further showed that HRI development for a distributed multiple robot system with a single or small number of human operators requires the HRI to support the human‘s situational awareness (SA). SA is a human's understanding of what is happening around the human operator at the current point in time and in the near future. For example, when driving, a human is aware that: 1) he is in a moving vehicle, 2) if he presses the gas peddle he will accelerate, and if he presses the break peddle he will decelerate, etc. [4,5] Augmenting human capabilities with automated, cooperative robotic devices is important for a wide variety of tasks, including, construction, assembly, repair, search and rescue, and general assistance with every-day tasks for the elderly and handicapped. Effective human-robot cooperation requires robotic devices that understand human goals and intentions. [6,7] These devices must have the capability to track and predict human intention and motion within the context of overall plan task sequences, based on a variety of sensor inputs. Such a capability will enable a fundamentally new kind of collaboration between humans and machines; one where the robots‘ actions are based, primarily, on implicit rather than explicit commands from humans. [4,7] 3.

PROBLEM FORMULATION: THE ROBOFRIEND PROJECT The objective of the RoboFriend Research project is to build a robotic assembly that would respond to human visual and/or audio. In essence allowing the robot to adjust its trajectory upon capturing a specific and a pre-programmed voice or visual signal.

641


These signals, once given by the user (human), not only trigger a change in path, but also define a learning point, that the robot should do the same maneuver in the event that the same obstacle is encountered. As a secondary goal, this work aims at incorporating robotics research in undergraduate engineering education. Incorporating robotics research in undergraduate studies is not new, many universities around the world have already experimented with it [8]. Good progress was made in all educational outcomes for the students. Students developed hands-on experience, both scientifically and personally; they had to apply their knowledge in solving the robotic problems while meeting strict deadlines. [9] Moreover, studies showed that students read one or more papers from the robotics/computer vision literature and implemented portions of these concepts in their work. Also, robotics research projects fit well with the interdisciplinary nature of engineering studies. students from various engineering disciplines worked together as a team. [10] 4.

THE PROCESS The objective of this work is to develop and build a robotic assembly that resembles a simplified version of the human arm, from the shoulders to the fingers. That is controllable with only human voice or visual inputs. Being a student based project, the project was divided into four specific areas, or sub-projects, as follows: 

Development of the Gripper (wrist to fingers)



Development of the Manipulator(shoulder to wrist)



Development of the electronics controls needed to control the actuators



Design and development of the HRI interface.

Development of the Gripper Figure P.1 shows a schematic comparison between a human hand and the humanoid gripper developed for this work. In essence, the Gripper mimics the human fist, but with lesser number of independent motions, or degrees of freedom, DOFs.

Figure 1: Comparison between the human hand and the gripper developed for this work The reduction of DOFs revolves around the motions of each finger. In the human fingers, each joint of each finger – or the knuckles – can move independently. In this gripper, however, they move together, i.e. one actuator drives the motion for the three knuckles of each finger.

Secondly, fingers of the human hand turn away from each other. For example, notice in figure P.1 how the human fingers can create an angle between themselves, while fingers in the gripper are parallel. The result

642


of this simplification is the reduction of the total number of DOFs of the whole gripper from a possible 23 to only 6 DOFs (including the wrist). The reason for this simplification is to reduce both the complexity of the design and the number of motors or actuators needed to create the motions needed for this robotic arm. This simplification is important, because this project – after all – is a student based project, and the students could use this simplification to direct their focus and attention towards other aspects of the design. Detailed analysis of the design of the gripper are discussed here, graphical analysis of the wrist joint will be included in the analysis of the manipulator, which is discussed later. The working envelope of the gripper is defined by angle range of each of its joint. Since the knuckles of each finger are dependent in motion, their working envelope is also a combination of each knuckle‘s range. The range for joint is shown below  Wrist joint:

00 - 340

 Finger joint: (individual knuckle) :

0 - 90

 Finger joint: (3 knuckles):

0 - 180

The driving mechanism: for motion of fingers Figures P.2 and P.3 show the mechanism of the gripping action. Fig P.2 shows the extended configuration of the gripper, while figure P.3 shows the closed configuration. The finger assembly is attached to a loop of rubber wiring (used in fishing lines), which is attached to a servo motor.

Figure P2: Extended Configuration

Figure P3 Closed Configuration As the motor rotates clockwise, the loop of cable to rotate clockwise, and because the cable is fixed to the tip of the finger, the finger follows the motion of the cable and ―extends‖. If the motor rotates counter clockwise, the whole process is reversed, and the finger is ―closes‖. From then on, it became a matter of figuring out the minimum and maximum angle of rotation for each motor. With careful testing, these values were identified for each finger, and a set of ranges of motion for each finger was developed.

643


In order to avoid adding extra weight on the manipulator, the servo motors powering the gripper were not mounted on the robot itself. Instead, cable loops were extended long enough to allow the motors to be mounted on the base of the robot, or the ―control‖ zone. This will be discussed in the controls section. Development of the Manipulator: Again, the manipulator mimics a simplified version of the human hand (shoulder to wrist), in the sense that it has lesser DOFs. Figure P.4 shows this similarity. As it can be seen in the figure, the manipulator resembles the human arm. However, it has fewer DOFs. This was achieved by reducing the number of motions at the shoulder from 3 to 2 and at the wrist from 3 to just 1 (the roll). In essence reducing the number of DOFs from a possible 7 to only 4 as shown in figure P.4.

Figure P4: Development of Manipulator Once again, the reason for this simplification is similar to those discussed in the manipulator section. Also, this simplification does not interfere with the objectives of this project.

Detailed analysis of the design of the manipulator are discussed here, the wrist joint will connect the manipulator to the gripper. As such, it will be included in these analyses. This is because the wrist joint was not discussed in the gripper design analysis.

The working envelop of the Manipulator is defined by angle range of each of its joint. The range for the base, shoulder, elbow and wrist are shown below:

Base joint

: 00 - 350

(10 degrees blocked to avoid collision with controller unit)

Shoulder joint : 10 - 120 Elbow joint : 10 - 135 By combining principles of the Denavit-Hartenberg representation, as well as motion transformation, the T matrix for this robot can be found, this analysis is shown below.

644


Figure P5: Forward Kinematics and development of the T matrix of the manipulator Transformations : Rot( x,  ) Rot(z,) Rot (y,), or 1 0  0

0 cos  sin 

0   sin   cos  

cos sin   0

 sin  cos 0

0 0  1 

cos  0   sin 

0 1 0



therefore Tmanipulator is cos cos

  sin  cos cos  sin  sin  cos sin  sin   sin  cos  

sin   0  cos    sin  cos cos sin  cos

   cos sin  sin   sin  cos  sin  sin  sin   cos cos    cos sin 

Development of the controls For this work, electrical power is used to power the motion of the various joints of the manipulator and gripper. For each degree of freedom, a servo motor is assigned to provide power for that joint. Due to their affordability, reliability, and simplicity of control by microprocessors, RC servos are often used in smallscale robotics applications. Many decisions were to be made by the students regarding the type of servos to be used, how they to be controlled, and the method of programming are, and these challenges were addressed and solved solely by the students. This was intentional, as the student were supposed to use the knowledge they gained from their engineering education to tackle and solve these problems. Several controllers were considered for this project, in the beginning, Programmable Logic Controller was used. However, after some experimenting and research, it was soon realised that PLC is not a suitable option as it cannot – easily – provide instantaneous control for several motors. The kind of motion needed when controlling a robotic arm. For PLC controllers to achieve this it would be too complex to build and construct. As such, micro controllers we selected as an alternative. Servo controllers are cheaper build and construct, do not require space and volume. In fact they are very compact in size and weight. Their performance is also superior to that of the PLC; a standard micro controller can control the simultaneous motions of to 16 motors. For this work, a parallax servo controller was used, Figure P.6 a picture of this controller. This controller proved to very convenient. It could control up to 16 motors at a time, powered only by 4 AAA batteries. But power to the motors was provided separately.

645


Figure P6: A 16 channel Parallax Servo Controller used in this work One of the fundamental challenges with servo motors is the fact that their range is generally fixed; from 0 to 220. When comparing this range to the ranges for some of the joints of the manipulator, it can be seen that some joints will be limited in range. Normally, this would not be a problem, a range of 0 to 220 is big enough and suitable for a reliable robotic motion. However, some of the joints on the manipulator had further gear reductions as part of the structural design. This lead to the range of motion dropping to just 0 to 15 to some joints. Since these reduction gears could not be removed. The servo motors themselves needed to be modified to add more range. Again, this was left to the students to solve, the students studies the making of a typical servo motor and identified the solution. Refer to figure P.7

Figure P7: Components of a typical servo motor As seen from P.7, the servo motor contains its own set of reduction gears, this gives the servo motor the very nature of controlled motion. In essence, a servo motor is a DC motor, a set of reduction gears, and a feedback control circuit. The solution to the above problem was to notice that on the major reduction – the one near number 3 – had an extended piece of plastic. This blocked the gear from rotating a complete 360. Instead, the gear would lock on that position, not able to rotate further, it will stop, in short, it is this extra piece of plastic that defines the 0 to 220 range of the servo motor. Realizing this, the student simply used a soldering gun to melt and remove that extra plastic, and so the motor was allowed to rotate 0 to 360 i.e. we could chose any angle we wanted. Even better, we were able to select angles greater than 360, which was sometimes necessary due to the structural gear reduction. Development of the controls The main objective of the HRI interface to create that connection between the user‘s visual and audio signal and allow it to control the robot‘s motion. In order to achieve this, the HRI interface used in this work must incorporate software and hardware elements. In this interface, visual/audio signals from users (humans) are captured in microphones and cameras and transmitted to the Computer. Next, they get converted into usable digital signals and processed in speech and image processing programs. Based on which, they trigger a specific and pre-defined motion programs from the Parallax GUI, which in turn sends a sequence of commands that correspond to that these motion. These signals are then fed in the PIC 16F877A controller and the parallax micro controller, which cause the robot to move.

646


In this work, three software applications were used in the conversion of input audio into motion command. The first application used is MicroC, which is used in conjunction with the PIC controller. Second is Visual Basic, which is used for speech recognition. Finally, the Parallax GUI is used to convert these signals into motion commands. MicroC is a high level programming language, which is similar to C and C++. This makes MicroC much easier to read and write than Microchip‘s assembly language, which is written in MPlab, a lower level language. The MicroC compiler is a BASIC-STAMP like and has most of the libraries and functions of both BASIC, STAMP I, and STAMP II. All of this makes programming in this language a bit easier for the student researchers, which already studied BASIC and C++. Therefore, writing the code for the speech recognition was not a challenge. Visual Basic 6.0 is used for this work for simple speech recognition and the parallel port coding. For speech recognition, VB is used in conjunction with Microsoft‘s Direct Speech Recognition Engine, a readily available application installed with Microsoft‘s Windows XP installation package. For the Parallel port coding, a dynamic link library file ―inpout.dll‖ is used. 5.

5

FINDINGS AND DISCUSSION

Figures P.2 and P.3 showed the actual design sketches of the gripper used in this work. The dimensions for the palm, fingers, and knuckles were those of the student who developed this gripper. He simply modeled his own hand. Based on the design discussed above, and using light material, such as plastics and aluminum. The gripper assembly was constructed and the gripping action was successfully demonstrated. Figure P.9 below show few configurations of the gripper developed for this work.

Figure 9: Actual gripper developed for this work, in multiple configurations One of the challenges facing the students was the issue of friction of cables powering the fingers. As cables travelled, they rubbed on the rubbing tubing they were house. This friction increased the loading on the servos, which then affected the flexibility of the robot‘s motion. Motors with higher torque solved this problem. Development of the Manipulator The design and construction of the manipulator proved to be a challenge to the students. The manipulator needed to perform two different and distinct tasks. The first is to manipulate the position and orientation of the gripper, hence the name manipulator. The second task is being structurally sound in order to carry its own weight as well as the weight of the gripper and its power cables, etc.

647


Students assigned to this problem were able to conduct torque analysis and were able to estimate the torques needed to power each joint of the manipulator, this allowed the student to select the proper servo motors for these joints. Unlike the gripper, however, these motors will be mounted on the assembly. Table P.2 and figure P.11 show the torque values for each motor, and actual manipulator developed for this work, respectively.

Table P.1: Torque values for each servo motor #

Joint

Torque needed

1

Wrist

500 N.m

2

Elbow

950 N.m

3

Shoulder

1.6 KN.m

4

Base

22.2 KN.m

2.

Figure P10: The actual manipulator developed for this project One can notice the high torque values needed. Starting from the wrist, torque values nearly doubled from joint to joint. This increase was contributed to the heavy weight of the gripper, its power cables and tubing. Secondly, the weights of the motors themselves. This is the part were cost became an issue for this project. Remembering that this is a student based project, the cost of these high torque motors proved to be a challenge for the students. Response of the HRI interface In order to isolate the HRI system aspect of this work from the rest of the mechanical challenge, the control interface designed for this work was connected to LED circuit in order to visualize the outcome. Each data pin from the parallel port (pins 2 through 9) to a LED anode terminal, connect one ground terminal (18 through 25) to connect to the cathode terminal from all LED. Using this technique, students were able to test the software aspects of the HRI interface without any impact from the mechanical side. Students showed that when a signal is captured from user‘s input, the LED associated with that input is switch on, this meant the system was able to capture, process, and deliver a digital signal to the PIC controller, exactly as planned. After mechanical and electronics setup of the manipulator was completed. Students used the teaching method in the parallax GUI to train the robot on some sample programs, or Jobs as they called them. Each of these jobs was created and assigned names, in total 5 programs of motion were developed, each to perform a series of motions of all major links in the manipulator as well as the gripper. Each of these jobs was stored and associated with a pre-determined verbal code, an expression that the user must say to microphone to trigger that job.

To minimize problems, programs were designed to create gestures with arm and fingers. For example, Job 1 required RoboFriend to move its fingers to gesture the sign of number 1, or extend the index finger while close all other fingers.

648


During the final year presentation of the project, students successfully demonstrated this feature. Though not perfect, but some programs were executed, triggered only via voices of students, lecturers, and visitors. Problems persisted though with voice recognition problems, noise from the crowd interfered with the HRI interface Impact of incorporating Robotics research in undergraduate studies When it came to research, a fundamental problem facing any academic is time constrains. Lecturers are required to be active in research, but they are bugged down with academic and administrative responsibilities in their faculties. One of the first benefits of the student-based research project is time. Lecturers can share the tasks of research with their students, perhaps the laborious ones, such as data analysis, construction and fabrication, and repeated cycles of testing. This is exactly what I – the project leader – was able to do, instead my focus shifted on more pressing aspects of research, such as learning, acquiring knowledge, comparing results with other researchers. In essence, students based project do provide an excellent solution to lecturers facing the problem of time. Secondly, this project was a collaborative effort between students of multiple fields of study. This allowed them to work together as a team, augment their knowledge towards a common goal. It also helped them develop the right communication and interpersonal skills needed to conduct their sub-tasks of this work. The impact on students is divided into areas, students‘ perception towards research based project, and the impact of research on their academic performance in final year subjects. The Majority of the feedback from the students was positive. Initially there was some difficulty, but as the semester went on, many started to enjoy the atmosphere. Secondly, students‘ academic performance before and after the project was recorded and compared. As table P.2 show, students‘ performance showed some improvement. Whether this improvement is related directly to the RoboFriend Project is debatable, but the data is welcome, and at least there was no drop in performance. 6.

CONCLOUSIONS Students were able to successfully design, develop and build a humanoid Gripper that mimics a simplified version the human fist, from wrist to fingers, with 5 DOFs. Students were able to successfully design, develop and build a humanoid Manipulator that mimics a simplified version the human hand, from shoulder to wrist, with 4 DOFs. Students were able to connect the gripper and manipulator to develop the RoboFriend, a humanoid robotic assembly that resembles a simplified version of the human Arm, from should to fingers. Students were able to successfully design, develop and build a Human-Robot Interaction (HRI) interface that allowed users to control the motion and behavior of RoboFriend using only their voice as input. Robotics research was successfully incorporated in undergraduate engineering studies. Research tasks could be divided into sub-tasks and given to engineering student as final year projects. This incorporation had a positive impact on research as it freed time for lecturers to pursue research, created an atmosphere of collaboration and team work amongst the students and their supervisors. Furthermore, this incorporation had a positive impact on the students themselves. In general, students‘ response was positive and supportive of the initiative, and their academic performance showed an improvement after the completion of the project.

649


7.

TEAM MEMBERS OF THE ROBOFRIEND PROJECT The names of the students who are members of the RoboFriend research project are shown in table P.3. I would like to take this opportunity to acknowledge their contribution to this work and recognize them as coauthors of this paper. Table P.3: Team members of this project

8.

Name

Contribution

Ong Sue Looi

Parallax Controller interface

Ng Ching Ruey

Gripper

Sathes Thangavelu, Sum Kam Wah, and Chan Weng Seng

Manipulator (part of a subgroup)

Chong Zheng Hao

HRI Interface, MicroC, etc

REFERENCES [1] Casper, J. Murphy, R. Human-robot interactions during the robot-assisted urban search and rescue response at the World Trade Center, IEEE Transactions on Systems, Man, and Cybernetics, vol 33 (3), pp 367–385, June 2003. [2] Druy. Jill, Hestand. Dan, et al. Design guidelines for improved human-robot interaction, Conference on Human Factors in Computing Systems archive. CHI '04 Vienna, Austria, pp 1540 – 1540, 2004, ISBN:1-58113-703-6 [3] Kiesler. S, Hinds, P. Introduction to Human-Robot Interaction. Special Issue of Human-Computer Interaction, vol 19, no 1 & 2. 2004. [4] Adams, J. A. Human-Robot Interaction Design: Understanding User Needs and Requirements, Proceedings of the 2005 Human Factors and Ergonomics Society 49th Annual Meeting, September 2005. [5] A. Hofmann, B. Williams, Intent Recognition for Human-Robot Interaction, Proceedings of 2007 AAAI Spring Symposium - Interaction Challenges for Intelligent Assistants, pp. 60-62, Stanford University, USA, 2007 [6] T. Asfour, P. Azad, N. Vahrenkamp, K. Regenstein, A. Bierbaum, K. Welke, J. Schröder, R. Dillmann, Toward humanoid manipulation in human-centered environments, Robotics and Autonomous Systems, vol. 56, No. 1, pp. 54-65, 2008. [7] Scassellati, B. How robotics and developmental psychology complement each other. Proceedings of the NSF/DARPA Workshop on Development and Learning. Lansing, MI: Michigan State University. 2000 [8] Maxwell, B. A. and Meeden, L. A, Integrating Robotics Research with Undergraduate Education, IEEE Intelligent Systems, November/December, 2000. [9] Mehrl, D. J, Parten E, and Vines. D. L, Robots Enhance Engineering Education, Proceedings of the Frontiers in Education Conference. 1997. [10] Rawat, K.S., and Massiha. H. H, A Hands-On Laboratory Based Approach to Undergraduate Robotics Education, Proceedings of the 2004 IEEE International Conference on Robotics and Automation, New Orleans, LA. April 2004. [11] Salisbury K., Mason M. T., Robot Hands and the Mechanics of Manipulation, M.I.T. Press. Cambridge, MA,1985. [12] Jacobsen. S. et al., Design of the Utah/M.I.T. Dexterous Hand, Proceedings of the IEEE International Conference on Robotics and Automation, San Francisco CA. 1520-1532. 1986.

650


DESIGN AND DEVELOPMENT OF RECEIVER MODULE FOR WIRELESS VISION BASED RESCUE ROBOT FOR ROUGH TERRAIN USING DAVINCI CODE PROCESSOR Naveen Prakash1, Kailash Vijaykumar2, Bharathi V3 ,and Arun Perumal4 1

Velammal Engineering College, India [email protected]

2


3


4


ABSTRACT In this model, the design and implementation of a receiver module for wireless vision based semi autonomous rescue robots that can be utilized in rough terrain using Davinci processor DVM6437, wireless camera receiver, Zigbee Transceiver and Global Positioning System (GPS) kit has been presented. Rescue robots play a pivotal role in aiding the rescue workers by searching for victims and survivors of tragedies that includes mining accidents, explosions and urban disasters. The semi-autonomous robots are generally programmed to perform their task autonomously during temporary loss connections to the control base. The zigbee transceiver module in the receiver side enables the control station to receive the images transmitted from the moving part of the robot and also helps in controlling the moving parts of the robot by sending control signals wirelessly, thereby eliminating the drawbacks of tele-operating rescue robots. The Global positioning system aids the controller by providing the latitudinal and longitudinal location of the moving part. Davinci processor DVM6437 is a digital media fixed point DSP processor based on Very Long Instruction Word (VLIW) architecture supports both half and full duplex communication. It has an expanded instruction set that is best suited for real time rescue operations. The DVM processor can be coded using code composer studio embedded development platform or by using Matlab simulink. Matlab codes can be used with the help of Embedded IDE link. Using these rescue robots, we can reduce human labour and achieve increased access to unreachable areas during natural disasters and manmade catastrophes. 1. 1.1

INTRODUCTION Catastrophes & Disasters A disaster is a natural or man-made hazard that has come to fruition, resulting in an event of substantial extent causing significant physical damage or destruction, loss of life, or drastic change to the natural environment. A disaster can be ostensively defined as any tragic event with great loss stemming from events such as earthquakes, floods, catastrophic accidents, fires, or explosions. Disasters (both natural hazards and man-made catastrophes) have brought loss, grief, and starvation to survivors. An example of a deadly natural calamity was the Great Chilean Earthquake, which took place in Spain in May 1960; roughly 5,000 people died from both earthquake and resulting Tsunamis. In January 2001, massive Earthquake in Gujarat cost the lives of 20,000 people and injuring over a lakh people. In May 2008 more than 50,000 Chinese citizens lost their lives, and over 20,000 people were missing as a result of the massive devastative earthquake. The Indian ocean Tsunami in 2004 caused an irreparable damage in the Indian Ocean region claiming over 2, 30,000 lives. Man-made disasters are also an important motivation, such as the grievous incident happening in USA on 11 September 2001(9/11), terrorist attack in Mumbai, India on 26 th November 2008 etc.

651


1.2

Need for Rescue Robots: Most victims in 9/11 and other disasters have died due to the delay in the availability of assistance. In such conditions, the victims‘ locations and conditions were difficult to determine by rescue crews. Several researchers and academic staff, consequently, have paid more attention in conducting research to develop rough terrain robots, especially for rescue missions. The robots are able to supply images of the environment and specify victims‘ locations to the robot operators at the control base outside the wrecked area. Common situations that employ rescue robots are mining accidents, urban disasters, hostage situations, and explosions. One notable use of rescue robots was in the search for victims and survivors in the remnants of the World Trade Centre. The benefits of rescue robots to these operations include reduced personnel requirements, reduced fatigue, and access to otherwise unreachable areas

1.3

Benefits of Semi- autonomous rescue robots: Rescue robots are generally of three types:  Fully autonomous rescue robots  Manually controlled rescue robots  Semi autonomous rescue robots Fully Autonomous rescue robots are robots that can perform desired rescue operations in unstructured environments without continuous human guidance. But it has limitations that includes complex design and inability to perform the human desired actions at all instants Manually controlled rescue robots are robots that can perform the desired rescue operations with continuous human guidance. It can be made to adapt to the changes in surroundings and can perform the human desired actions at all instants but has a practical drawback that it ceases to function whenever control signal from the user is not received Semi autonomous rescue robots are designed by combining the features of both fully autonomous and manually controlled robots. It can be controlled by the user until it can receive the control signal and can be made to perform the desired tasks. When the control signal fails to reach the robot, the robot starts performing its rescue mission autonomously based upon predefined criteria. Hence the drawbacks of both fully autonomous and manually controlled robots are eliminated here

1.4

Benefits of Wireless rescue robots: Rescue robots can be controlled either by wired medium or wireless medium. Tele-stimulating wired robots are easier in design and are cost effective, but not suited for long distance applications and its functionality will be often interrupted in case of severe twists or bends in the debris or remains of the damaged building. Hence to overcome these drawbacks, we use wireless medium for control. Further, autonomous activity of the robot will be limited when wired medium is used. Hence wireless robots are often preferred over tele stimulating robots especially in the rescue missions

1.5

Receiver module using Digital media processor DVM6437: The receiver module for the semiautonomous rescue robot acts as a control base for the moving robot when it is operating under manual control mode. Davinci processor DVM6437, a digital media processor is widely preferred for the construction of received module. DVM6437 is a fixed point processor based on Very Long Instruction Word (VLIW) architecture. The DVM processor can be coded using code composer studio embedded development platform or by using Matlab simulink. Matlab codes generated can be used with the help of Embedded IDE link. DVM6437 is used here for real time speech processing, video and image processing applications. The receiver module receives the video output from the bot through wireless camera receiver, image through zigbee protocol and sends the control signal using zigbee transmitter.

2.

LITERATURE REVIEW 1.

Roger A. Chadwick, New Mexico state university had worked on the concept of controlling the unmanned ground vehicle (UGV) robots and examined it in a simulation involving a search task.

652


Semiautonomous robotic UGV was simulated, detected and monitored continuously and made to respond to the control signals during a search mission for a radioactive target. He also designed specific interface features including a quick video playback (QVP) function to assist vehicle operations. 2.

Kazunori Ohno, Shouich Morimura, Satoshi Tadokoro, Eiji Koyanagi and Tomoaki Yoshida have collectively worked on the development of a control system for semi autonomous robot. The authors proposed a sensor reflexive method that controls the flipper arm of the moving robot autonomously for getting over unknown and changeable environment

3.

Michael Baker, Robert Casey, Brenden Keyes, and Holly A. Yanc, University of Massachusetts, USA have proposed an interface for human-Robot Interaction in Urban Search and Rescue operations. Human robot interaction has always centred around operators relying heavily on video stream. They have created a new interface that fuses information on and around the video window to exploit this fact. They have increased the functionality of the graphical user interface (GUI) by utilizing the unused information and have modified the INEEL navigation system to perform command remapping task that allows an operator to spontaneously reverse the direction. I. Firmansyah, B. Hermanto and L.T. Handoko, Indonesian Institute of science have proposed a control and monitoring system for modular wireless robot consisting of three main modules: main unit, data acquisition and data processing modules. They have developed a generic prototype with an integrated control and monitoring system to enhance its flexibility, and to enable simple operation through a web-based interface accessible wirelessly

4.

5.

P. Ayyalasomayajula, S. Grassi, N. Deurin, P.-A. Farine and T. Guégue Electronics and Signal Processing Laboratory, Switzerland implemented an image recognition algorithm on the Davinci processor. They developed an Alternative and Augmentative Communication (AAC) portable device called PictoBar which is used in speech rehabilitation therapy. PictoBar recognizes barcodes and images, such as pictograms and pictures. Then it plays a sound message associated with the recognized barcode or image. This paper describes the development of the image recognition algorithm and its implementation using Codec Engine framework on a Davinci processor.

6.

Jae-Min Choi, Byeong-Kyu Ahn, You-Sung Cha,Tae-Yong Kuc, Dept of Electrical and electronics engineering, Sungkyunkwan University have jointly developed remote controlled home robot server with zigbee sensor network. wireless PAN (personal area network) has attracted strong attentions as short-distance networking solution. Convenience of wireless PAN technology has attracted more attentions over traditional wired home network devices such as Ethernet, PLC and Home PNA. They have proposed a home server for an efficient control of internal information and conditions of house from remote location and virtual home robot server to be implemented with Zigbee sensor network

653


3. 3.1

SYSTEM DESCRIPTIONS HARDWARE REQUIRED:

Block Diagram

The various Hardware components required to implement receiver module of wireless vision based rescue robots for rough terrain are:  Davinci code processor TMS320DM6437  Zigbee transceiver CC2520  Wireless camera receiver  GPS Trainer Kit (for tracking) 3.1.1 Davinci processor DVM6437: Davinci processor DVM6437, a digital media processor is widely preferred for the construction of received module. Features: 



High performance digital media processor  2.5, 2, 1.67, 1.51, 1.43 ns instruction time  400, 500, 600, 660, 700 MHz clock rate  Eight 32 bit 64+ instructions/cycle  3200, 4000, 4800, 5280, 5600 MIPS  Fully Software-Compatible With C64x  Low-Power Device Advanced VLIW DSP Core  64 32-Bit General-Purpose Registers  Eight Highly Independent Functional Units with six ALU‘s and two multipliers  Load-Store Architecture With Non-Aligned support  Instruction Packing Reduces Code Size



C64x+ Instruction Set Features  Byte-Addressable (8-/16-/32-/64-Bit Data)  8-Bit Overflow Protection  Bit-Field Extract, Set and Clear  Normalization, Saturation, Bit-Counting and increased orthogonality  C64x+ Extensions  Compact 16-bit Instructions  Additional Instructions to Support with complex multiples



C64x+ L1/L2 Memory Architecture  256K-Bit (32K-Byte) L1P Program RAM/Cache [Flexible Allocation]  640K-Bit (80K-Byte) L1D Data RAM/Cache [Flexible Allocation]  1M-Bit (128K-Byte) L2 Unified Mapped RAM/Cache [Flexible Allocation]

654




Supports Little Endian Mode Only



Video Processing Subsystem (VPSS)  Front End Provides:  CCD and CMOS Imager Interface  BT.601/BT.656 Digital YCbCr 4:2:2 (8-/16-Bit) Interface  Preview Engine for Real-Time Image processing  Glueless Interface to Common Video decoders  Histogram Module to-Exposure, Auto-White Balance and  Auto-Focus Module  Resize Engine  Resize Images From 1/4x to 4x  Separate Horizontal/Vertical Control 

Back End Provides:  Hardware On-Screen Display (OSD)  Four 54-MHz DACs for a Combination of  Composite NTSC/PAL Video  Luma/Chroma Separate Video (S-video)  Component (YPbPr or RGB) Video



External Memory Interfaces (EMIFs)  32-Bit DDR2 SDRAM Memory Controller  Asynchronous 8-Bit Wide EMIF (EMIFA)  Flash Memory Interfaces – NOR (8-Bit-Wide Data) – NAND (8-Bit-Wide Data)

   

Enhanced Direct memory access (EDMA) controller (64 Independent channels) Two 64-Bit General-Purpose Timers One 64-Bit Watch Dog Timer Two UARTs (One with RTS and CTS Flow Control) Master/Slave Inter-Integrated Circuit (I2C Bus)



Functional Block Diagram: The TMS320C64x+™ DSPs (including the TMS320DM6437 device) are the highest-performance fixedpoint DSP generation in the TMS320C6000™ DSP platform. The DM6437 device is based on the thirdgeneration high-performance, advanced VelociTI™ very-long-instruction-word (VLIW) architecture developed by Texas Instruments (TI), making these DSPs an excellent choice for digital media applications. The C64x+™ devices are upward code-compatible from previous devices that are part of the C6000™ DSP platform. The C64x™ DSPs support added functionality and have an expanded instruction set from previous devices. With performance of up to 5600 million instructions per second (MIPS) at a clock rate of 700 MHz, the C64x+ core offers solutions to high-performance DSP programming challenges. The DSP core possesses the operational flexibility of high-speed controllers and the numerical capability of array processors. The C64x+ DSP core processor has 64 general-purpose registers of 32-bit word length and eight highly independent functional units—two multipliers for a 32-bit result and six arithmetic logic units (ALUs). The eight functional units include instructions to accelerate the performance in video and imaging applications. The DSP core can produce four 16-bit multiply-accumulates (MACs) per cycle for a total of 2800 million MACs per second (MMACS), or eight 8-bit MACs per cycle for a total of 5600 MMACS.

655


The DM6437 also has application-specific hardware logic, on-chip memory, and additional on-chip peripherals similar to the other C6000 DSP platform devices. The DM6437 core uses a two-level cache-based architecture. The Level 1 program memory/cache (L1P) consists of a 256K-bit memory space that can be configured as mapped memory or direct mapped cache, and the Level 1 data (L1D) consists of a 640K-bit memory space —384K-bit of which is mapped memory and 256K-bit of which can be configured as mapped memory or 2-way set-associative cache. The Level 2 memory/cache (L2) consists of a 1M-bit memory space that is shared between program and data space. L2 memory can be configured as mapped memory, cache, or combinations of the two. The peripheral set includes: 2 configurable video ports; a 10/100 Mb/s Ethernet MAC (EMAC) with a management data input/output (MDIO) module; a 4-bit transmit, 4-bit receive VLYNQ interface; an inter-integrated circuit (I2C) Bus interface; two multichannel buffered serial ports (McBSPs); a multichannel audio serial port (McASP0) with 4 serializers; 2 64-bit general-purpose timers each configurable as 2 independent 32-bit timers; 1 64-bit watchdog timer; a user-configurable 16-bit host-port interface (HPI); up to 111-pins of general-purpose input/output (GPIO) with programmable interrupt/event generation modes, multiplexed with other peripherals; 2 UARTs with hardware handshaking support on 1 UART; 3 pulse width modulator (PWM) peripherals; 1 high-end controller area network (CAN) controller [HECC]; 1 peripheral component interconnect (PCI) [33 MHz]; and 2 glueless external memory interfaces: an asynchronous external memory interface (EMIFA) for slower memories/peripherals, and a higher speed synchronous memory interface for DDR2.

The DM6437 device includes a Video Processing Subsystem (VPSS) with two configurable video/imaging peripherals: 1 Video Processing Front-End (VPFE) input used for video capture, 1 Video Processing Back-End (VPBE) output.

656


The Video Processing Front-End (VPFE) is comprised of a CCD Controller (CCDC), a Preview Engine (Previewer), Histogram Module, Auto-Exposure/White Balance/Focus Module (H3A), and Resizer. The CCDC is capable of interfacing to common video decoders, CMOS sensors, and Charge Coupled Devices (CCDs). The Previewer is a real-time image processing engine that takes raw imager data from a CMOS sensor or CCD and converts from an RGB Bayer Pattern to YUV422. The Histogram and H3A modules provide statistical information on the raw colour data for use by the DM6437. The Resizer accepts image data for separate horizontal and vertical resizing from 1/4x to 4x in increments of 256/N, where N is between 64 and 1024. The Video Processing Back-End (VPBE) is comprised of an On-Screen Display Engine (OSD) and a Video Encoder (VENC). The OSD engine is capable of handling 2 separate video windows and 2 separate OSD windows. Other configurations include 2 video windows, 1 OSD window, and 1 attribute window allowing up to 8 levels of alpha blending. The VENC provides four analog DACs that run at 54 MHz, providing a means for composite NTSC/PAL video, S-Video, and/or Component video output. The VENC also provides up to 24 bits of digital output to interface to RGB888 devices. The digital output is capable of8/16-bit BT.656 output and/or CCIR.601 with separate horizontal and vertical syncs. The Ethernet Media Access Controller (EMAC) provides an efficient interface between the DM6437 and the network. The DM6437 EMAC support 10Base-T and 100Base-TX, or 10 Mbits/second (Mbps) and 100 Mbps in either half- or full-duplex mode, with hardware flow control and quality of service (QOS) support. The Management Data Input/output (MDIO) module continuously polls all 32 MDIO addresses in order to enumerate all PHY devices in the system. The I2C and VLYNQ ports allow DM6437 to easily control peripheral devices and/or communicate with host processors. The high-end controller area network (CAN) controller [HECC] module provides a network protocol in a harsh environment to communicate serially with other controllers, typically in automotive applications.

3.1.2 Zigbee transceiver: Zigbee is a specification for a suite of high level communication protocols using small, low-power digital radios based on the IEEE 802.15.4-2003 standard for Low-Rate Wireless Personal Area Networks (LRWPANs) for a short range. The technology defined by the Zigbee specification is intended to be simpler and less expensive than other WPANs, such as Bluetooth. Zigbee is targeted at radio-frequency (RF) applications that require a low data rate, long battery life, and secure networking.

The radios use direct-sequencer spread spectrum coding, which is managed by the digital stream into the modulator. BPSK is used in the 868 and 915 MHz bands, and OQPSK that transmits four bits per symbol is used in the 2.4 GHz band. The raw, over-the-air data rate is 250 Kbit/s per channel in the 2.4 GHz band, 40 Kbit/s per channel in the 915 MHz band, and 20 Kbit/s in the 868 MHz band. Transmission range is between 10 and 75 meters (33 and 246 feet) and up to 1500 meters for zigbee pro, although it is heavily dependent on the particular environment. The output power of the radios is generally 0 dBm (1 mW). The basic channel access mode is "carrier sense, multiple access/collision avoidance" (CSMA/CA). That is, the nodes talk in the same way that people converse; they briefly check to see that no one is talking before they start. There are three notable exceptions to the use of CSMA. Beacons are sent on a fixed timing schedule, and do not use CSMA. Message acknowledgments also do not use CSMA. Finally, devices in Beacon Oriented networks that have low latency real-time requirements may also use Guaranteed Time Slots (GTS), which by definition do not use CSMA.

657


In robotic applications, Zigbee transceivers provides wireless program downloading from PC to controller board, plus 2-way wireless control and monitoring of robot from PC. The transceivers connect to a host PC having no RS232 comm.-ports via an off-the-shelf USB-to-RS232 adapter (e.g., IOGear) A simple command interface allows the user to send simple movement commands via the Zigbee downlink, and receive back status info via the Zigbee uplink. On a terminal emulator, we can press single keyboard keys to command the robot. These include: "F"=forward, "B"=back-up, "A"=accelerate, "D"=decelerate, "L"=turn-left, "R"=turn-right and "S"=stop. Commands are sent over the RF, and then the robot sends back status info, such as left-right motor speed and direction, as well as battery voltage plus various sensor readings. The user can watch the robot directly, or monitor its behaviour remotely via the wireless video uplink.

3.1.3 Global Positioning System (GPS) trainer kit: A VI Microsystems GPS/GSM trainer kit is used which is used for tracking the position of the robot at all instants. It is used for calculating the exact latitudinal and longitudinal presence of the robot on the earth. The Global Positioning System (GPS) is a space-based global navigation satellite system (GNSS) that provides reliable location and time information in all weather and at all times and anywhere on or near the Earth when and where there is an unobstructed line of sight to four or more GPS satellites The satellites of the Global Positioning System (GPS) broadcast radio signals to enable GPS receivers on or near the Earth's surface to determine location and synchronized time. GPS signals include ranging signals, used to measure the distance to the satellite, and navigation messages. The navigation messages include ephemeris data, used to calculate the position of each satellite in orbit, and information about the time and status of the entire satellite constellation, called the almanac. The original GPS design contains two ranging codes: the Coarse/Acquisition (C/A) code, which is freely available to the public, and the restricted Precision (P) code, usually reserved for military applications. The software Trackmaker is used in order to control the GPS trainer kit and used to estimate the location of the robot in motion.

3.1.4 Wireless Camera Receiver: The wireless camera receiver is used in the receiver module for wirelessly receiving the video and audio information captured by the camera. The output of this receiver is given to the Davinci processor for real time digital media applications

658


3.2

SOFTWARE REQUIRED:

Block Diagram

The various Software components required to implement receiver module of wireless vision based rescue robots for rough terrain are: 



Matlab Simulink video and image processing block set Code Composer studio (CCS) integrated development environment

3.2.1 Matlab Simulink video and image processing block set: Matlab (Matrix laboratory) is a fourth generation programming language for real time digital media applications. It is a numerical computing environment and fourth-generation programming language. Developed by Math Works, MATLAB allows matrix manipulations, plotting of functions and data, implementation of algorithms, creation of user interfaces, and interfacing with programs written in other languages, including C, C++, and Fortran. Simulink® software models, simulates, and analyzes dynamic systems. It enables you to pose a question about a system, model the system, and see what happens. With Simulink, you can easily build models from scratch, or modify existing models to meet your needs. Simulink supports linear and nonlinear systems, modelled in continuous time, sampled time, or a hybrid of the two. Systems can also be multirate — having different parts that are sampled or updated at different rates. The Video and Image Processing Blockset software is a tool used for the rapid design, prototyping, graphical simulation, and efficient code generation of video processing algorithms. Video and Image Processing Blockset blocks can process images or video data. These blocks can import streaming video into the Simulink environment and perform two-dimensional filtering, geometric and frequency transforms, block processing, motion estimation, edge detection and other signal processing algorithms. You can also use the blockset in conjunction with Real-Time Workshop® to automatically generate embeddable C code for real-time execution. Video and Image Processing Blockset blocks support floating-point, integer, and fixed-point data types.

3.2.2 Code Composer studio (CCS) integrated development environment: Code Composer Studio is an integrated development environment for developing DSP and/or ARM code for the TMS320 DSP processor family, and other processors such as the MSP430, from Texas Instruments. Code Composer Studio includes a real time operating system called DSP/BIOS. It includes accurate instruction set simulators, and supports JTAG based debugging. Code composer Studio provides a single user interface taking users through each step of the application development flow. Familiar tools and interfaces allow users to get started faster than ever before and add functionality to their application thanks to sophisticated productivity tools.

659


Code composer Studio‘s integrated debugger has DSP-specific capabilities and advanced breakpoints to simplify development. Conditional or hardware breakpoints are based on full C expressions, local variables or registers. The advanced memory window allows you to inspect each level of memory so that you can debug complex cache coherency issues. Code composer Studio supports the development of complex systems with multiple processors or cores. Global breakpoints and synchronous operations provide control over multiple processors and cores. CCS has many image analysis and graphic visualization. CCS includes the ability to graphically view variables and data on displays which can be automatically refreshed. CCS can also look at video data in the native format. TI compilers also perform program level optimizations that evaluate code performance at the application level. With the program level view, the compiler is able to generate code similar to an assembly program developer who has the full system view. This application level view is leveraged by the compiler to make trade-offs that significantly increase DSP performance. The TI ARM and Microcontroller C/C++ compilers are specifically tuned for code size and control code efficiency. They offer industry leading performance and compatibility. TI has developed C/C++ compilers specifically tuned to maximize DSP usage and performance. TI compilers use a wide range of DSP-oriented, and sophisticated device-specific optimizations that are tuned to DSP architectures. 4. 4.1

SAMPLE SIMULATIONS AND RESULTS Edge Detection in a video signal:

660


The Edge Detection block outputs a binary image with the edges shown in white. This output is displayed in the Edges window. The Compositing block accepts the original video frames, shown in the Original window, and the output of the Edge Detection block as inputs at its Image1 and Mask ports, respectively. The input to the Mask port tells the Compositing block which pixels to highlight. As a result, the model displays a composite image in the Overlay window, where the original pixel values are overwritten by the white edge values 4.2

People tracking in a video signal:

This block detects and tracks people in a video sequence with a stationary background using the following process: 1) Use the first few frames of the video to estimate the background image. 2) Separate the pixels that represent the people from the pixels that represent the background. 3) Group pixels that represent individual people together and calculate the appropriate bounding box for each person. 4) Match the people in the current frame with those in the previous frame by comparing the bounding boxes between frames.

661


4.3

Histogram Display of a video signal:

The block displays the histograms of R, G, and B values in the RGB Histogram window and the original RGB video in the viewer window. An image histogram is a type of histogram that acts as a graphical representation of the tonal distribution in a digital image. It plots the number of pixels for each tonal value. By looking at the histogram for a specific image a viewer will be able to judge the entire tonal distribution at a glance. 4.4

Motion detection in a video signal: The sum of absolute differences (SAD) method is a popular technique for motion detection in video processing. This block applies the SAD method independently to four quadrants of a video sequence. If motion is detected in a quadrant, the demo highlights the quadrant in red.

662


Double-clicking the Switch block so that the signal is connected to the SAD side, the Video Viewer block displays the SAD values, which represent the absolute value of the difference between the current and previous image. When these SAD values exceed a threshold value, the demo highlights the quadrant in red.

663


4.5

Surveillance recording system:

Security concerns mandate continuous monitoring of important locations using video cameras. To efficiently record, review, and archive this massive amount of data, you can either reduce the video frame size or reduce the total number of video frames you record. This block uses the Video and Image Processing Blockset to demonstrate the latter approach. In it, motion in the camera's field of view triggers the capture of "interesting" video frames The Motion Threshold window displays the threshold value in magenta, and plots the SAD values for each frame in yellow. Any time the SAD value exceeds the threshold, the model records the video frame

664


4.6

Lane Detection:

Input Block:

Lane Detection block:

The block implements this algorithm using the following steps: 1) Detect lane markers in the current video frame. 2) Match the current lane markers with those detected in the previous video frame. 3) Find the left and right lane markers. 4) Issue a warning message if the vehicle moves across either of the lane markers.

665


Departure warning block:

This subsystem uses the Hough Lines block to convert the Polar coordinates of a line to Cartesian coordinates. The subsystem uses these Cartesian coordinates to calculate the distance between the lane markers and the center of the video bottom boundary. If this distance is less than the threshold value, the demo issues a warning. This subsystem also determines if the line is yellow or white and whether it is solid or broken.

666


The Results window shows the left and right lane markers and a warning message. The warning message indicates that the vehicle is moving across the right lane marker. The type and colour of the lane markers are also shown in this window.

667


5.

REFERENCES [1] C.H. Lee, S. H. Kim, S. C Kang, M.S.Kim, Y.K. Kwak (2003). ‖Double –track mobile robot for hazardous environment applications‖, Advanced Robotics, Vol. 17, No. 5, pp 447-495, 2003 [2] K. Osuka, H. Kitajima (2003). "Development of Mobile Inspection Robot for Rescue Activities:MOIRA", Proceedings of the 2003 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems, pp3373-3377, 2003 [3] Mohammed G.F.Uler (1997). "A Hybrid Technique for the Optimal Design of Electromagnetic Devices Usign Direct Search and Genetic Algorithms"IEEE Trans. on Magnetics, 33-2, pp1931-1937, 1997 [4] R. Smith, "Open Dynamics Engine", http://ode.org/ [5] S. Hirose (2000). "Mechanical Designe of Mobile Robot for External Environments", Journal of Robotics Society of Japan, Robotics Society of Japan, vol.18, No.7, pp904-908, 2000 (in Japanese) [6] S. Kawaji et al, (2001). "Optimal Trajectory Planning for Biped Robots"The Transactions of the Institute of Electrical Engineers of Japan. C, vol.121, No.1, pp282-289, 2001 (in Japanese) [7] S. Kobayashi et al, (1995). "Serarch and Learning by Genetic Algorithms"Journal of Robotics Society of Japan, vol.13, No.1, pp57-62, 1995 (in Japanese) T. Inoh et al (2005). "Mobility of the irregular terrain for resucue robots"10th Robotics symposia pp 39-44, 2005 (in Japanese) [8] T. Takayama, et al (2004). Name of paper. "Development of Connected Crawler Vehicle "Souryu-III" for Rescue Application "Proc. of 22nd conference of Robotics Society of Japan CD-ROM, 3A16, 2004 (in Japanese) [9] Y. Yokose et al (2004). "Minimization of Dissipated Energy of a Manipulator with Coulomb Friction using GA Increasing the Calculated Genetic Information Dynamically" Transaction of JSCES, Paper No.20040024, 2004 (in Japanese) [10] Y.Yokose V.Cingosaki, H.Yamashita (2000). "Genetic Algorithms with Assistant Chromosomes for Inverse Shape Optimization of Electromagnetic devices" IEEE Trans. on Magnetics, 36-4, pp10521056, 2000 [11] C. Carreras, I. D. Walker (2001). ‖Interval Methods for Fault-Tree Analysis in Robotics‖, Transaction on Reliability, Vol. 1, pp. 1-11, 2001

668


A DISTRIBUTED VISION SENSORS AIDED INTELLIGENT ENVIRONMENT SYSTEM DESIGN FOR MOBILE ROBOT NAVIGATION Yongqiang Cheng1, Ping Jiang2, and Yim-Fun Hu1 1

Engineering Design and Technology/University of Bradford Bradford, UK e-mail1: [email protected], [email protected] 2 Informatics/University of Bradford Bradford, UK e-mail2: [email protected]

ABSTRACT In the classical-architecture of autonomous mobile robots with on-board centralised intelligence, localisation has always been one of the difficult tasks. As a premise for robot navigation, a robot has to identify its location in an unstructured and dynamic environment, using more than one type of sensors to localise itself, such as ultrasonic sound sensors, laser scanner, vision sensors along with compasses. Wheel slip, parameter shift, and collisions etc. are making localisation extremely uncertain and complicated. Therefore, robot must possess immense computation power to carry out the localisation tasks by means of sensor fusion and inference. Control of a mobile robot in such an environment also encounters similar implementation difficulties. Even if the subsumption approach is adopted to decompose complicated intelligent behaviour into many simple behaviour modules and organised layers, computational complexity does not decrease significantly. Mutual interference between active sensors and signals from different robots create another layer of difficulties for robot control. Besides, if sensors are directly connected to a centralised computer unit, the incrementally learned spatial knowledge can be outdated due to the dynamic changes in the environment. To overcome the shortcomings of the centralized approach, this paper proposes an intelligent environment system with distributed wireless vision sensors for robot navigation control. An architecture design of the wireless vision sensors mimicking the mosaic eye infrastructure and the technique used to support robot navigation, which tries to solve the heavy computation bottle neck faced by the traditional centralised method, are described. The wireless vision sensor network connectivity, the descriptions of the system functions and modules, organisation and connection of the system components as well as the general idea of the proposed algorithms are discussed in detail. Keywords: Autonomous mobile robots, Robot navigation, Distributed wireless vision sensors, Mosaic Eye, Intelligent environment. 1.

INTRODUCTION Over a long period of time, research in intelligent robotics has largely focused on equipping robots with different onboard sensors and a super computer [1, 2] acting as a smart ―brain‖. However, this centralised approach has to resolve many problems in association with essential tasks of maintaining robot autonomy such as localisation, map building and path planning. These will become even more challenging in terms of the computational power and memory as the environment space and complexity grow. Steering away from this smart brain approach, research exploiting low level intelligence such as insect eyes shows a potential to produce highly intelligent functions for autonomous robotic control. Insect eyes with smaller nervous systems, such as mosaic eye, are extremely creative and diverse[3]. The mosaic eye transmits information from the retina to the insect's brain where they are integrated to form a usable picture of the insect's environment in order to co-ordinate their activities in response to any changes in the environment. In bio-mimetic engineering, current works mostly focus on front end sensing, for example, artificial compound eyes [4, 5] and neuromorphic vision systems [6, 7]. Can the highly efficient neuron organization be applied to a large scale and complex environment? In addition, the current bio-mimetic navigation research is mostly concentrated on low-level reflex behaviour control. Can goal-driven and planning behaviour be implemented? These two questions inspire us to develop an intelligent environment with distributed vision to assist in the navigation of un-intelligent mobile robots in an indoor environment. It is expected that this can provide unintelligent mobile devices with free and reliable mobility in any randomly structured environment distributed with wireless sensors connected by wireless communications.

669


Navigation techniques assisted by an environment with distributed information intelligence are very different from conventional ones that rely on centralized intelligence implemented in the robot itself. They need to be carefully reconsidered and developed. By utilising pervasive intelligence [8] distributed in the environment, robots can still maintain a high degree of mobility while utilizing little on-board computational functions and power. A static camera configuration mounted in the environment rather than on the mobile robot will provide the intelligent environment with distributed processing ability to facilitate both localisation and control. This architecture not only reduces the sensing complexity, but also distributes the massive computation to numerous low processing speed and small memory vision sensory devices [9]. Each vision sensor covers one area of the whole target space. Vision sensors are connected wirelessly forming a Wireless Sensor Network (WSN). As each vision sensor is power sensitive [10], the low power IEEE 802.15.4 protocol is selected to exchange information between vision sensors to share their observations in the environment. This paper will provide a detailed module level design of the intelligent environment architecture. The paper is organised as follows: section two presents some background and proposes the overall architecture of the intelligent environment architecture; section three contains details of the definitions and functionalities of the system components; communication interfaces are defined in section four; section five concludes the paper. 2.

MOSAIC EYES AND INTELLIGENT ENVIRONMENT ARCHITECTURE The nervous systems of low level insects and invertebrates are hardwired from birth. Each neuron has its own special predetermined links and functions. This might be one reason why lower insect with smaller nervous systems are so creative and diverse. Inspired by the biological insect nervous systems, the intelligent infrastructure proposed in this section is a large-scale camera network with wireless vision sensors deployed in the environment, resembling that of a insect compound eye made up of thousands of mosaic eyes (lens) connected by the nervous system, which observes the environment and controls the motion of robots directly, as shown in Figure 90.

Snake control points Eye cover area Mosaic eyes Robot

Figure 90 Mosaic eyes inspired intelligent environment In order to guide a robot, a snake algorithm utilising the information captured by the distributed wireless vision sensors is proposed for path planning. The advantage of adopting a snake scheme [11] is that once the start and target positions of the robot are specified, a path can be initialised globally with multiple path segments connected one by one in a serial manner through the vision sensors. A path segment, which is made up of a sequence of connected Control Points (CPs), forms a collision free path within the coverage of a vision sensor node. By connecting the different path segments, a reference collision free optimum path, hereafter referred to as the Reference-snake (R-snake) path is created and maintained by the networked vision sensors. When obstacles appear or disappear or the positions of the obstacles change in the view of vision sensors, the R-snake will deform its shape accordingly and contract to its minimum length in order to keep all the control points away from the obstacles. Due to sensor sensing errors, position calculation errors and robot mechanic movement errors, the robot‘s position may diverge from the R-snake path, causing a

670


deviation of the robot navigation path from the optimum R-snake path. To compensate for such position deviation errors, robot trajectory tracking is performed using an Accompanied-snake (A-snake) [12] that starts from the current robot position and follows as closely as possible the R-snake path for local trajectory tracking and mobile robot motion control. Since overlapping areas between vision sensors exist, a negotiation procedure has to be activated to elect a dominant vision sensor and that sensor will send its planned motion commands to the robot for actuating. The exchange of information among eyes is carried out between neighbouring eyes through wireless communications. At any specific time, there is one and only one dominant eye to collect and fuse the information from its neighbouring eyes. The fused information will be used to deform the segment of the Rsnake in this mosaic eye view and generate an A-snake to accompany the robot and to send motion instructions to control the robot movement. The mosaic eyes supported robot navigation architecture involves three physical system components: 

the remote console responsible for all offline configurations, system status monitoring, mosaic eyes controlling and files transferring service;



the networked wireless vision sensors (or mosaic eye) are the main bodies to implement algorithms and functions including image processing, localisation and robot navigation;



the mobile robot itself which can receive motion control signals from the intelligent environment via its IEEE 802.15.4 communication peripheral, drive and steer in response to these commands.

Figure 91 shows the interactions between these three physical components, their functional modules, processing flows and connections. No direct communication between the mobile robot and the remote console is involved. The next section will describe each and every module in detail. Vision Sensor ……

Goal input module

Vision Sensor 3

Global Topological Map Communication Adaption

Data Download Vision Sensors Control

Executive files

IEEE 802.154

IEEE 802.15.4

Communication Adaption

Mobile Robot

Vision Sensor 1 Path Initialization

Path Re-Initialization

Vision Sensors Status Remote Monitoring

...

Remote Console

System Profile Topological map

...

Camera Calibrating

Camera Params: Homograph Matrix Exposure Time Gain ……

Camera Params

Motion Commands Analyse & Execution

Motor Servo

Meta data parameters or files

Function modules

Communication Adaption

Normal files Executive files

Vision Sensor 2

IEEE 802.154

System Profile: Sensor IDs Parameters Settings

Trajectory Generating

Predictive Controller

Path Adjustment

Position Estimation

Image Acquisition & Processing

Visual Content Configuration

Control Robot

Legend: Physcial component

Data

Wireless links

Processing flow

Figure 91 System components and functional modules An area of consideration of this approach is its complexity in comparison with the traditional onboard navigation. This is analogous to comparing the biological phenomena of insects with very small brains but having many eyes with that of human with binocular vision but a big brain for processing information. The information processing functions of an insect compound eye could be highly efficient subject to proper role allocation and proper information routing mechanism. Therefore, the success of such a distributed vision sensor architecture relies on providing each sensor with unambiguous semantics, predetermined roles and

671


coordinating links with other sensors. Vehicles with little intelligence are expected to perform superior intelligent mobility under the control of numerous mosaic eyes. 3. 3.1

SYSTEM COMPONENTS The Remote Console The Remote Console can be either a Windows based or Linux based PC plus an IEEE 802.15.4 hardware module which is attached with the PC to enable its wireless communication ability. The remote console has the functions of system configuration, global topological map configuration, firmware upgrade, mosaic eyes calibration and human-machine interaction. The console software features as a configuration tool but is also responsible for mapping the global information onto local data via calculated routing tables and acts as a repository centre for storing and distributing profile files. Once initial calibration and configuration functions are carried out, profile files that contain all necessary parameters, routing tables and data will be generated for and launched onto individual vision sensors. Specifically, the files include the global topological map consisting of hashed binary tables representing routes connecting different vision sensors; the system profile describing all necessary settings and parameters; and the executive firmware files and system loadable modules. The function modules are shown in Figure 91 and discussed in the following. 

Visual Content Configuration Module The visual content configuration module provides the tools for creating a topological map with wireless sensor geographic locations. It links a geometric map with a sensor topological map. Each wireless sensor is provided with a node ID and clear semantics, such as ―it is John‘s office connecting to the D wing corridor‖. Optimal routes in the geographic space are obtained by offline searching in the constructed topological map. It is a complete set of all possible connections from one vision sensor to others with semantics. The routes with semantics representations are then converted into hash tables by a set of hash functions in order to compress the geographic routes. The routing tables are split into smaller portions based on vision sensors such that each vision sensor has all the routing tables containing all reachable destinations start from it.



Data Download Module The Data Download module is responsible for reading local files and data in the remote console disk and transmitting them to respective vision sensors by the request of operators. The communications between remote console and vision sensors are IEEE 802.15.4 unreliable connections. Therefore, packet serial numbers will be used for retransmission when packet loss is not tolerant, such as firmware upgrade.



Vision Sensors Control Module This module is designed to send control signals to vision sensors in case the operator needs to interfere with the system during running time. This module enables the operator to re-sample the background image, to fetch the planned path, to inspect the foreground image and to restart the vision sensor.



Vision Sensors Status Remote Monitoring Module As a response to control signals sent by Vision Sensors Control Module, vision sensors will report related information back to the remote console. The data are received and categorized into individual display area by the Control Adaptation module which is a proposed application layer protocol on top of the IEEE 802.15.4 stacks. When a robot enters one mosaic eye‘s coverage, the real time robot location should be displayed by default. Other information such as real time R-snake coordinate, background image or foreground image can be received and displayed on demand.

672




Camera Calibrating Module Due to the manufacturing process whereby the Complementary Metal Oxide Semiconductor (CMOS) sensors in the camera experience random asymmetries during the fabrication process and maybe in the packaging process, the pixel colours of image captured by different cameras on the same object may be interpreted differently even the exposure time and gains are identical. One function of this module is to sample targets and filter the desired colour thresholds for each individual camera by controlling the vision sensors. Secondly, all vision sensors are mounted in the building in a manner that they have overlapping areas with their neighbouring sensors. The multiple geometry homography [13] projection matrix can be obtained by choosing correspondent points in the overlapping area selected by this module. The last function of this module is to calculate the ―meter per pixel‖ ratio. Due to the distance and height of the camera, the projection of the movement of the robot onto the vision sensors are different, this module is used to send a movement patterns commands to the robot to calculate the ratio.



Goal Input module This module accepts navigation goals/destinations from either the remote console itself or IEEE 802.15.4 compatible terminals.

3.2

Vision Sensors Vision sensors are the brains of the intelligent environment. Each sensor covers a certain area of the building and provides service within that region. They are fixed and ―hardwired‖ by communicating with each other thus forming an infrastructure mimicking mosaic eyes and nerves. As the main component in the proposed architecture, vision sensors have the largest range of functional modules, shown in Figure 91, from single view image acquisition and processing to multiple views projections and mapping; from path initialization, adjustment, trajectory generating to eventually robot motion control; from wireless packets receiving, sending and analysing to all kinds of interested data reporting. The process requirement and time consumed also vary. To meet the real time performance as well as real time response requirements, multiple threads are recommended for communication, image processing and path planning separately. Functional modules and their input & output relationships are discussed below: 

Image Acquisition and Processing module The Image Acquisition and Processing module senses the environment, looks after the real time information such as robot location and obstacles coordinates. Then all these data are shared with the path planning modules such as the Path Initialisation module, the Path Adjustment modules and the Position Estimation modules. There are two classes of algorithms involved in the overall image processing: online and offline algorithms. The offline algorithm will sample the colour thresholds of the targets and generate colour lookup tables. The Visual Content Configuration module in remote console is taking care of this. On the other hand, the entire image processing functions including capturing real time images, sampling background frame, subtracting foreground image to background image to get dynamic information, segmentation of interested colours, identifying and tracking objects are performed online during run time. The data obtained in this module are real time obstacles coordinates and robot location and direction. They will be stored in the shared memory for other modules to use, such as the Path Adjustment module.



Path Initialization module

673


When the program starts with the default navigation goal set in the system profile file or a goal of the navigation is passed to vision sensor via the remote console, the R-snake algorithm is initialized. The first vision sensor will check its own global topological routing table, and send route queries to the nodes connected to it to confirm or find the destination, then a search within its coverage for an optimal R-snake segment is carried out to plan a set of control points to be connected in series. Real time obstacles information is taken into account during the search algorithm. 

Trajectory Generating Module This module plans the space position of A-snake based on the R-snake which comes from the Path Initialization module or the Path Adjustment module or the Path Re-initialization module. The start point of the R-snake does not reflect the actual position of the robot; hence an A-snake is generated whenever a new robot position is available to deal with this situation. Once the planned result is ready, it is stored in the shared buffer for the Predictive Controller module to use.



Predictive Controller Module As the name suggested, a predictive technology will be employed in this module to achieve a high performance tracking speed up to robot driving limit. The Communication Adaptation Layer always keeps an eye on the information in the air. This includes the control points from neighbouring vision sensors. By combing the information from neighbouring nodes, vision sensor is enabled with the knowledge out of its view and hence a possible better prediction. A rolling window along with time optimisation algorithm is recommended to generate series of robot control commands including driving force and steering toque with specific associated time. This module processes both information from the Trajectory Generating module and Position Estimation. The former has a larger time interval than the latter since Trajectory Generating only processes the observed robot position while Position Estimation forecasts locations for robot. It is a balance of control accuracy and density.



Control Robot module This module involves a token negotiation procedure to enable a vision sensor to take dominant control of the robot within its coverage. If an area where the robot is in is only covered by one eye, the sensor has the dominant control right straight away; if the area is covered by more than one sensors, they all compete for a token to become the dominant one to ensure that there is one and only one vision sensor controls the robot. Competition is initiated in random, and once the token is with a sensor, it will send out signals regularly to indicate its ownership of the token. When a robot approaches the edge of its coverage, a token handover request message will be sent out. A new owner of the token will be yielded as a result. Once the ownership of the token is resolved, the vision sensor will send the planned series of commands wirelessly to the robot.



Position Estimation Module Due to the image capturing and processing delay, the current positions of obstacles and that of the robot are not real time but estimated based on the history positions and the calculated moving speeds in addition to the newly captured information. Therefore, this module is responsible for estimating the positions even during the interval without input from the observations. In the estimation -> predictive -> control -> estimation processing loop, as shown in Figure 91, the space positions of the R-snake and A-snake are assumed to be unchanged and the robot will follow the desired trajectory blindly. During this time the module predicts the movement of the robot without feedback from the visual observations. By skipping the image processing, this processing loop can generate control commands rapidly. However, the robot may deviate from the desired direction by executing these blind control commands. Thus a very short duration should be set for this loop.

674




Path Adjustment Module Compared with the Position Estimation Module, the processing interval of this module is much longer. It reads the shared memory for all available real time information, abstracts obstacles and the distance between control points into forces exerting on the control points and check whether any violation happens, e.g. the snake collides with obstacles. As a result, the resultant force will push or pull the control points to a new balance position thus forming a new R-snake. If violation happens, all planned paths/trajectory will be discarded and the module will try to re-initiate a snake in the Path Re-initialization module. Once the planned result is ready, it is written to the shared buffer for the Trajectory Generating module and may be sent to the Communication Adaptation Layer for the remote console display if requested by the operator.



Path Re-initialization Module In case the Path Adjustment processing fails, a re-initialization procedure is triggered to perform the process.

3.3

Wireless Controlled Robots In contrast to highly intelligent robots, the mobile robot considered in the intelligent environment can be, for example an off-the-shelf model car or a wheel chair which have very limited on-board computational power but can access the wireless sensor network by integrating an IEEE 802.15.4 module. As shown in Figure 92, some kind of marks or colour patterns should be used to identify the robot from obstacles as well as its moving directions.

Figure 92: Robot on field Since most of the tasks are performed by the intelligent environment, the robot itself only has a very small set of functions. 

Motion Commands Analyse and Execution Module This module will analyse the series of commands sent by the dominant vision sensor. Due to the transmission delay or image processing delay, time tags are used in the commands to eliminate the outdated commands and execute the correct ones.



Motor Servo Module This module converts the commands from the vision sensor to different kinds of driving and steering motors understandable ones. E.g. Pulse-Width Modulator (PWM) control signals.

675


4.

COMMUNICATION INTERFACES Figure 93 shows the protocol stacks of the Communication Adaptation (shown in Figure 91) module between vision sensor and vision sensor/mobile robot. The module is built on top of the physical lay and MAC layer of the 802.15.4 protocol stack to enable vision sensors to communicate with each other.

Vision sensor/ Mobile Robot

Vision sensor

Control protocol layer

Control protocol layer

Information processing and navigation algorithm

Information processing and navigation algorithm MAC layer

MAC layer

PHY layer

PHY layer

2.4GHz wireless Figure 93: Communication Adaptation Layer protocol stacks It is a fundamental module for the information processing and robot navigation control algorithms. Apart from centralised intelligent control system, this distributed architecture requires a harmonic collaboration between all distributed vision sensors to achieve a desired goal. Thus, a uniform packet format, a collection of signal flow scenarios and a set of physical independent function calls are designed to serve robot navigation such as path planning and motion control. The module provides a data exchange mechanism suitable for distributed control purpose. Figure 94 is an outline of all the communication parties and the links between them. As the key intelligent component, vision sensors are involved in all communications, either as a sender or as a receiver. Remote console

Other sensors

Vision sensor

Vision sensor

Other sensors

Mobile robot

Figure 94: Communications interfaces By incorporating an error tolerance mechanism, even with some data loss without re-transmission will not affect the operation of the system. For example, a procedure for re-competing for the robot control token is invoked in case no one has the control token due to a handover process failure caused by a handover related message loss. But message loss is not allowed during a file download. However the IEEE 802.15.4 standard intends to offer the fundamental lower layers of the protocol stack of a type of Wireless Personal Area Network (WPAN) which focuses on low-cost, lower power, low-speed ubiquitous communications between devices, packets loss or network nodes disappearances are inevitable. Therefore, some mechanisms are proposed in this layer to deal with the package loss and package delay. Specifically,

676


5.



Kalman filter[14] is used to filter the received information such as coordinates of the snake control points and predict their new positions;



Data are tagged with serial numbers to make sure that they can be reassembled in the right sequence if they arrive in an unexpected sequence due to delay. The sequence number also provides a mechanism for retransmission if packets are missing;



Error checksum, existence periodical checking for communication nodes and etc. can also be considered.

CONCLUSION AND FUTURE WORK This paper proposes a design of the distributed intelligent environment architecture based on the wireless vision sensor network. The system components including Remote Console, Vision sensors and Mobile Robot are discussed in detail. The functional modules, involved technologies and the possible algorithms are described. The proposed distributed intelligent environment with vision sensors mounted in the building releases the massive computation into smaller sensors, simplifies the localisation problem, promises a scalable environment variation and makes different mobile robots sharing the same software possible. Future research will be focused on the image based distributed visual servoing and self-calibration of the wireless vision sensor network.

6.

ACKNOWLEDGEMENTS This research is partially sponsored by the Royal Society and the NSFC joint project WiME.

7.

REFERENCES [1] [2] [3] [4]

Arkin, R.C., Behavior-based Robotics. 2000, Cambridge: MIT Press. Murphy, R.R., Introduction to AI robotics. 2000, Cambridge: MIT Press. Land, M.F. and D.-E. Nilsson, Animal Eyes. 2002: Oxford University Press. Boussaid, F., C. Shoushun, and A. Bermak. A scalable low power imager architecture for compoundeye vision sensors. in Fifth International Workshop on System-on-Chip for Real-Time Applications (IWSOC'05). 2005. [5] Tanida, J., et al., Thin observation module by bound optics (tombo): concept and experimental verification. Applied Optics-IP, 2001. 40(11). [6] Serrano-Gotarredona, R. and e. al, AER building blocks for multi-layer multi-chip neuromorphic vision systems. NIPS, 2005. [7] Boahen, K., Neuromorphic Microchips. Scientific American, May 2005. [8] Snoonian, D., Smart buildings. IEEE Spectrum, 2003. 40(8). [9] http://www.xbow.com/Products/productdetails.aspx?sid=253. [10] Crocel, S., F. Marcellon, and M. Vecchio, Reducing power consumption in wireless sensor networks using a novel approach to data aggregation. The Computer Journal, 2008. 51(2): p. 227-239. [11] Cheng, Y., P. Jiang, and Y.F. Hu. A snake based approach for robot path planning in an intelligent environment with distributed vision. in Proc. Int. Conf. Automation and Computing. 2007. Stafford, UK. [12] Cheng, Y., P. Jiang, and Y.F. Hu. A-Snake: Integration of Path Planning with Control for Mobile Robots with Dynamic Constraints. in ICCAE. 2010. [13] Hartley, R. and A. Zisserman, Multiple View Geometry in Computer Vision, ed. 2nd. 2003, Cambridge. [14] Welch, G. and G. Bishop, An introduction to the Kalman Filter. UNC-Chapel Hill, TR 95-041, 2004.

677


INTEGRATING OPENCV BASED MACHINE VISION TO AN ABB INDUSTRIAL ROBOT Srinivas Ganapathyraju, Ph.D Sheridan Institute of Technology and Advanced Learning Faculty of Applied Science and Technology Brampton, Canada. Email: [email protected] ABSTRACT This paper presents the practical implementation of OpenCV, an open source machine vision framework from Intel, to an ABB 6 axis industrial robot. An image frame is acquired using a low cost web cam system. An algorithm is presented in OpenCV used for corner detection and for generating the coordinates of the detected corners. The computed coordinates are then communicated to the ABB robot controller through an RS232 serial interface. The calibration of the vision system and the robot to ensure accuracy of the system are presented. The use of RAPID programming language routines for serial communication between the controller and the PC as well as the routine for manipulating the part is presented. Keywords: Harris Corner Detection, Shi-Tomasi, OpenCV, Machine Vision, Robotics, Calibration. 1.

INTRODUCTION Robotic systems require sensors that provide the robot controller valuable information about the environment and objects that need to be manipulated. Probable the most important of these sensors is the vision system. Today vision systems are used widely in automated manufacturing to determine the position, orientation and to identify the parts. With advances in both hardware and software technology, vision systems are expected to expand the capabilities of robotic manipulation to allow complex inspections for close dimensional tolerances with improved recognition and better part location capabilities. The cost of vision systems is still fairly on the high side, which motivated us to look at alternative solutions to introducing vision systems into our existing robotics lab curriculum at Sheridan Institute. There were two main criteria that needed to be addressed. The first being a software package that will allow us to implement a full array of vision based algorithms and the second criteria being that it was desired to have a vision camera system that was low cost. The first criteria was met by using OpenCV [1], which is an open source computer vision library developed by Intel corporation, and the second criteria met by allowing us to use cheaper and more capable cameras such as web cams to be integrated into vision based projects. OpenCV is written in C++, which allows for fast image processing. Several third party wrappers have also been developed that allow OpenCV to run in JAVA, Visual C++.Net and Visual C#.Net. OpenCV has provided the necessary tools for both students and professionals to implement machine vision research projects that were previously available to a few exclusive labs. In this project the objective was to manipulate a square wooden block with no particular orientation or location with a six axis ABB industrial robot using a web cam for part location and recognition. This was achieved by using the Shi-Tomasi corner detection algorithm [5] in OpenCV [1], which is a modified Harris corner detection algorithm. Once the corners are detected the next step is to determine the corner coordinates which would allow us to determine part location and orientation of the wooden block. Corner coordinates were determined in OpenCV using the sub pixel corner sub routine [1]. Also presented in this paper is ABB robot‘s RAPID programming sub routines that are used for both serial communication and also for manipulation of the parts.

2.

THEORY: HARRIS CORNER DETECTOR Corner detection in computer vision works by extracting certain defined features to deduce the contents of an image. Corner detection is used for a number of applications in machine vision such as object recognition and motion tracking to name a few.

678


By definition a corner is an intersection of two edges or a point that has two commanding and differing edge directions within a defined area of the point. Several algorithms [2] detect interest points rather than corners, which will require further analysis of the interest point to determine the real corners. Corner detection using correlation can be computationally expensive and suboptimal [2]. To overcome this problem Harris and Stephens [3] proposed a different approach, and an improvement on the method proposed by Moravec [4]. Moravec‘s algorithm [4], defines a corner to be a point with low self coincidence that tests each pixel in the image for the presence of a corner. The algorithm achieves this by comparing similar areas centered on a pixel to the nearby coinciding area. The sum of squared differences (SSD), are used to test for similarity between two similar areas with a lower number indicating similarity. Three cases need to be considered as shown in Figure 1. Case (a) occurs when the image window patch is flat with no change in intensity along different directions.

Figure 1: Three possible scenarios that results from the Harris Corner detector [6] Case (b) occurs when the image window patch is in on an edge, no significant change occurs along the edge direction, but a significant change occurs when the shift is perpendicular to the edge. Case (c) occurs when the image window patch is a corner or an isolated point, resulting in significant change in intensity along all directions. A general mathematical representation based on the above cases can be derived [3]. The image intensity denoted by I, change E that is produced by a shift (x,y), and image window W can be expressed as:

E(x, y)   Wu ,v [I ( x  u , y  v )  I u ,v ]2

(1)

u ,v

Applying the first order approximation:

f (u  x, v  y )  f (u, v)  xfu (u, v)  yf v (u, v)

(2)

Consequently,

[I 



( x u , y  v )

 I u , v ]2

[ I (u, v)  xIu  yI v  I (u, v)]2

(3)

1st order approximation

  x 2 I u2  2 xyIu I v  y 2 I v2

(4) (5)

679


I u 2   xy   I u I v I u 2  xy (   I u I v

Iu Iv   2 I v 

x   y  

Rewritten in matrix form

(6)

Iu Iv  x  )   2 I v   y 

(7)

For small shifts (x,y), the bilinear approximation of equation (7) can be applied to equation (1) as follows:

x  E(x, y)  [x, y] M    y

(8)

Where M is a 2x2 symmetric matrix

I u 2 M   W (u , v)  u ,v  I u I v

Iu Iv   2 I v 

(9)

Harris [3] points out that if image points are classified using eigenvalues of M, in this case 𝛼 and 𝛽 are to be proportional to the principal curvatures of the autocorrelation function as shown in Figure 2, three different cases will arise: 1. 2.

When the curvatures 𝛼 and 𝛽 are small, it results in a flat region. When one curvature is large and the other is small, it results in an edge region.

3. When both curvatures are large, it results in a corner region.

Figure 2: Auto-correlation principal curvature space [3] Once the edges and corners have been classified, the next step is to measure their quality or response. The size of this response will be used in determining the selection of corner pixels and the thinning of edge pixels. The corner response can be measured using: R = det M –k(trace M)2

(10)

Where det M = 𝛼𝛽 trace M = 𝛼 + 𝛽

680


k (0.04-0.06) is an empirically determined constant [6]. The corner response R is shown in Figure 2, as fine contours. R is positive in the corner region, negative in the edge region, and small in flat region. Shi and Tomasi [5], made an improvement on the selection process for the corners that was proposed by Harris [3]. In order to make the selection better, Shi and Tomasi [5], suggested that only eigenvalues be used to check if a pixel was a corner or not. They modified the response R to be: R = min (𝛼, 𝛽) 2.1

Corner detection in OpenCV: As already discussed, OpenCV is a real time computer vision library developed by Intel corporation. One of the major advantages of OpenCV is that a low cost web cam system can be used for various applications. The routine that is used in OpenCV to detect corners is the cvGoodFetauresToTrack()[1], which is based on the Shi and Tomasi definition [5]. This function was developed for object tracking. The function computes the second derivative using Sobel operators [1] to compute the needed eigenvalues. A list is then returned for different points that meet the criteria for our definition. The function in OpenCV is written as follows [1]:

void cvGoodFeaturestoTrack( const CvArr* image, CvArr* eigImage, CvArr* tempImage, CvPoint2D32f* corners, int* corner_count, double quality_level, double min_distance, const CvArr* mask = NULL, int block_size = 3, int use_harris = 0, double k = 0.4, ); The input images should be either an 8 bit or 32 bit single channel image. cvGoodFeaturestoTrack function has been called, the output is an array of pixel locations. 2.2

Once the

Subpixel Corners in OpenCV: For the current application, detecting the presence of the corners of an object is the first step. Once the corners have been detected, the second step is to determine the coordinates of the corner pixels that have been determined by the previous function. OpenCV has a function called cvFindCornerSubPix() [1], which was developed for camera calibration. As shown in Figure 3, if the vectors are examined starting at point q and ending at p, if p falls on a flat area, the gradient will be zero, however if the vector falls on an edge, the gradient at p on the edge will be orthogonal to the vector q-p. In both cases the dot product between the gradient at p and the vector q-p will be zero. By solving multiple pairs of the gradient at a nearby point p and the associated gradient q-p, it results in a more accurate sub pixel location for q or an exact location of the corner.

681


Figure 3: Finding corners to subpixel accuracy [1] The function in OpenCV to determine the subpixel corner is as follows [1]: void cvFindCornerSubPix( const CvArr* CvPoint2D32f* corners, int CvSize CvSize CvTermCriteria criteria );

image, count, win, zero_zone,

The Figure 4 shows four corners detected in a test sample. Figure 5 shows the output for the pixel coordinates for the test sample shown in Figure 4.

Figure 4: Four corners detected in a test sample.

Figure 5: Pixel coordinates generated using cvFindCornerSubPix function.

682


3.

SYSTEM ARCHITECTURE AND SETUP The system that was used for the experimentation consisted of three basic modules, which are the vision system using a webcam, the corner detection and processing system, and the control of the robotic manipulator. Figure 6 shows the layout of the system that was used.

Object

Vision System

Corner Detection

Robot Controller

6 Axis Robot

Figure 6: Overall architecture of the system The vision system uses a Logitech QuickCam Orbit webcam with a built in motorized tracking system that allows for 189o tracking in horizontal and 102o tracking in the vertical directions. The webcam is connected to the computer by means of a USB cable. The camera is mounted on a stand facing the robot. As mentioned above, the system uses OpenCV which is an open source visual processing library developed by Intel Corporation. The vision processing and corner detection is carried out on a PC with an Intel core (TM) i7 CPU processor running at 2.67GHz. The processed information in this case the coordinates of the part object to be manipulated are then transmitted to the robot controller via serial communication. The robot manipulator is an ABB IRB120 6 axis robot. Figure 7 shows the actual setup in the lab.

Figure 7: Actual setup in the lab 3.1

System Calibration: The system needed to be calibrated, since the pixel dimensions did not represent the actual part dimension. To accomplish this, metrology gauge blocks of known dimension were used to compare it to those of the actual pixel dimensions. The resulting compensation factor would be used in calculating the actual location of the wooden block to allow the robot to grasp it.

3.2

ABB rapid programming language: ABB robots use a control software called RobotWare, whose primary function is to control the motion of the robots six axis. The programming language used to program the different routines is called RAPID. For this application two important RAPID routines that were used were the serial channel handling and the work object manipulation.

683


For serial communication to take place, a serial communication channel needs to be open between the robot controller and the PC. The function reads the serial communication port to get the x and y coordinates and the angle offset. The data then needs to be converted from string to numerical value. The basic structure for the serial function in RAPID to open the com port is as follows: FUNC Bool OpenCommPort() Open "com1:" , channel \Bin; clearIOBuff channel; Return True; ERROR Return FALSE; ENDFUNC To read data from the channel, a set of variables need to be declared in the PROC: PROC PartData() VAR string XCoord:= ""; VAR string YCoord:= ""; VAR string ZAngle:= ""; VAR bool ok; NewXpos:=0; ...... The data can now be read as shown here for the XCoord and similarly applied for the y and z coordinates: XCoord:= ReadStrbin (channel); ....... The string has to be converted into a numerical value as shown here for the x coordinate and similarly applied for the y and z coordinates: Ok:=StrToVal(XCoord, NewXpos); ......... Once the coordinate data has been received by the robot controller from the PC, the data will need to be translated into the manipulator motion. In order to understand the robot joint motions, one will need to understand the different coordinate systems that are used to describe the motion of the robots manipulator joints, and tool. The main coordinate systems used in describing the work space for the robotic manipulator are shown in the Figure 8. The tool center point (TCP) is point that can be located anywhere on the tool; in our case it is located at the edge point of the gripper. The positions recorded during robot motion are the position of the TCP, which is also the point that moves along a given path at a given velocity.

In this application, to move the gripper to the required position to manipulate the wooden block, the frame on the work object coordinates in RAPID were manipulated. When work objects are defined based on the positioning instructions in RAPID, the positional coordinates will be based on the work object coordinates, allowing the X and Y coordinates and also the orientation of the part object detected by the vision system to be manipulated relative to the work object. The instruction in RAPID to manipulate the work object frame is as follows: wobj1.oframe.trans.x := XCoord; wobj1.oframe.trans.y := YCoord; wobj1.oframe.rot:= OrientZYX(ZAngle,0,0);

684


Figure 8: Industrial robot coordinate system [7]

4.

CONCLUSIONS In this paper a solution to detecting the corners of a wooden block and also finding the coordinates of the corners was presented. The algorithm is based on the Harris and Stephens corner detection algorithm. An open source vision library (OpenCV), to run the corner detection algorithm is presented. The use of ABB‘s RAPID programming routines for serial communication and relative tool movement based on the acquired part coordinates are also presented.

5.

RECOMMENDATIONS The system that has been proposed here, applied corner detection for a simple square block. Corner detection and the detection of the coordinates of complex shapes that are not symmetrical needs to be evaluated further, and how best they can be manipulated by the robot. The OpenCV language is also limited in the fact that user interfaces cannot be integrated, which can be overcome by using third party wrappers such as Visual Studio C#, and JAVA, which will allow the creation of a more user friendly user interface and it needs to be investigated further.

6.

ACKNOWLEDGEMENTS I would like to thank Mr.Andrew Orton, CAMDT Lab Manager, for his help in setting up the equipment and guidance with the RAPID programming language. I would like to say a special thanks to the Dean of the Faculty of Applied Science & Technology, Jocelyn Piercy and Sheridan Institute for making the attendance to the conference possible.

7.

REFERENCES [1] G. Bradski and A. Kaehler, Learning OpenCV Computer Vision with the OpenCV Library, O‘Reilly Publications, 2008. [2] http://en.wikipedia.org/wiki/Corner_detection [3] C. Harris and M.J. Stephens, A combined corner and edge detector, Alvery Vision Conference, pages 147-152, 1988. [4] H. Moravec, Obstacle avoidance and navigation in the real world by seeing robot rover, Technical Report CMU-RI-TR-3, Carnegie Mellon University, Robotics Institute, 1980. [5] J. Shi and C. Tomasi, Good Features to Track, 9th IEEE Conference on Computer

Vision and Pattern Recognition, June 1994. [6] http://www.cse.psu.edu/~rcollins/CSE486/lecture06_6pp.pdf [7] ABB, Rapid programming Manual.

685


AN EYE-TRACKING BASED WIRELESS CONTROL SYSTEM Suraj Verma, Prashant Pillai, and Yim-Fun Hu School of Engineering, Design and Technology University of Bradford, Bradford, United Kingdom e-mail1: {s.verma5, p.pillai, y.f.hu}@bradford.ac.uk ABSTRACT The power of vision is taken one step ahead with the introduction of sophisticated eye-tracking and gazetracking techniques which track the movement of the eye and the gaze location to control various applications. This paper describes in detail the low-cost hardware development and the software architecture for a real-time eye-tracking based wireless control system using the open-source image processing framework of AForge.NET. The system developed has been tested in the field of remote robotic navigation, using the Lego NXT Mindstorm robot, and wireless home automation, using the X10 transmission protocol. The system designed uses a USB camera to detect the eye movements and transmit control signals over a wireless channel. This provides a highly useful alternative control system for a person suffering from quadriplegia (i.e. full body paralysis). Test results on system response time and error percentage show the performance of the developed eye tracking systems. Keywords: Eye tracking, wireless control, low-cost headgear, AForge.NET framework, assistive technology. 1.

INTRODUCTION Several people around the world suffer from several physical disabilities that prohibit them from leading a normal life. Several conditions like Amyotrophic Lateral Sclerosis (ALS), cerebral palsy, traumatic brain injury, or stroke may result in loss of muscle movements in the body [1], thus rendering the person paralyzed. However, in several cases not all parts of the body are paralyzed and the person may have limited voluntary movements of the head, eyes or tongue and may even be able to blink, wink or smile [2]. Assistive technology systems, that make use of these available movements, can be developed to aid people suffering from these conditions not only communicate with other people, but also control various electronic devices [3]. For people suffering from quadriplegia, one of the only reliable and convenient sources of voluntary movement that may be used by an assistive technology system is the movement of the eye [4]. Hence, the person‘s eye movements may be used to control computers for browsing the internet, simple word processing or even paint like application. In this paper we present a real-time low-cost eye-tracking based wireless control system called IRIS – Intelligent Recognition for Interactive Systems. The IRIS system can be used to remotely control a wireless robot and can also be used for basic home automation by wirelessly controlling the lights. In the future, the IRIS system aims to provide a single control system to help people suffering from disabilities to continue with their day-to-day activity, helping them to move around without being dependant on somebody and also control various household appliances. The current commercial eye tracking system is the Tobii X60 & X120 eye tracking system which can record eye movements irrespective of any surface or environments such as TVs, projection screens, physical objects such as brochures, magazines, products and even shopping shelves [5]. The eye tracking system is placed near the environment under study and the user, without wearing a head-mounted gear, can look anywhere within that environment and the system analyzes the user‘s eye movements and determines where or what the user is looking at. This system proves to be highly accurate and allows a large degree of head movement and distance from the eye tracking system. However, they are very expensive and hence there is a need for a low-cost eye tracking device. A light-weight eye tracking system called the ‗View Pointer‘ detects eye movements towards ubiquitous computers embedded in real world objects [6]. The View Pointer uses IR tags that are embedded onto each object. The IR tags emit infrared light set at a unique frequency whose reflection on the cornea of the user‘s eye determines what object the user is looking at. The View Pointer uses a light-weight USB camera, as a video fed to the system, to detect the frequency of the infrared light reflected on the cornea of the eye and determines which object the user is looking at depending on the frequency recorded by the system. However, this form of detection cannot be used in robotic navigation as the distance between the user and the robot, embedded with the IR tag programmed to transmit the desired frequency, varies. It can be used for home automation, but the only drawback is that the user cannot remotely control home appliances from a large distance as the user has to be close to the object embedded with the IR tag.

686


The rest of the paper is organized as follows. Section 2 introduces some of the key eye tracking mechanisms and systems. The hardware architecture and the head-mounted gear are described in Section 3. Section 4 describes in detail the software architecture and how the Aforge.NET framework was adopted to develop the eye tracking software. Two applications developed using the IRIS control system is discussed in Section 5. Some performance results of the IRIS system are presented in Section 6. Finally, the conclusion is discussed in Section 7. 2.

EYE TRACKING TECHNOLOGY The process adopted to track the movement of the eye is termed as eye tracking [7]. Eye tracking uses devices, called as ‗Eye Trackers‘, that are responsible for tracking the movement of the eye and act as the front end device which feeds the input (i.e. video frames of eye movement) to the system for further processing. There are various metrics used to track the movement of the eye such as fixations, gaze duration, point of gaze/area of interest, scan path and hot gaze plots [7, 8, 9]. An early technique used to detect the movement of the eye was with search coils [10] which involved attaching a plaster of paris ring directly to the cornea and through mechanical linkages to a recording pen for measurements in a varied electromagnetic field. One commonly used eye tracking mechanism is the Electro-OculoGraphy (EOG) [11, 12] method which is a process of tracking the movement of the eye based on recording the Corneal-Retinal Potential (CRP), the potential difference existing between the cornea and the retina during eye-movements. The EOG signal ranges from 0.05 to 3.5 mV in humans and is linearly proportional to eye displacement which generates a change of the order of around 20 μV in readings for each degree of eye movement [9]. Another commonly used eye tracking mechanism is Video-OculoGraphy (VOG). This method has been proven to be a highly effective non-invasive technology for evaluating eye movements [13]. It is a method in which the movement of the eye is recorded with the help of digital cameras, in particular infra-red cameras [14]. Image processing filters are applied to the images collected from the video source and the movement of the eye is determined from these images. The IRIS system described in this paper also uses a form of VOG eye tracking mechanism. However, even though EOG based systems are more reliable than VOG based systems, they require additional hardware like A/D converter, amplifier, filters and samplers to accurately process the analogue signals before feeding it to the control system. It is here that Video-Oculography plays an important role since it uses only a light-weight portable USB video camera to provide images for the detection process. The complex process required to determine the direction of the eye movement is controlled by the eye tracking software, thereby reducing the size and cost of the eye tracking system. Some of the applications that incorporate eye-tracking and eye-gaze technology are in the following areas: 

Field of media and marketing research: to create a comprehensive study on how people understand, use and respond to web pages [15]  Cognitive psychology: to study mental processes including how people think, perceive, remember and learn [15]  Human Computer Interaction (HCI): to understand human perception of virtual environments and computer games [15]  Indoor/outdoor navigation: to produce eye controlled wheelchairs that aid in navigating the wheelchair through different environments [11, 12, 16] The Attentive Response Technology (ART) [17] has employed eye tracking techniques in home automation using the X10 protocol [18]. The system provides an interface for the physically challenged people to control their household appliances like televisions, lighting, heating and cooling. The ART system uses two cameras, one to monitor the eye movements and the second to monitor the environment. Depending on what device the user looks at, the system maps the real-time image of that device with the pre-known devices, which are saved in an image database, to search for a match. The main disadvantage of this system is that the image database has to be updated if there are any changes around the house. However, the IRIS control system enables the user to control home automation wirelessly in real-time by just looking at the options on the user interface and is independent of the position of the object or surrounding.

687


3.

IRIS CONTROL SYSTEM ARCHITECTURE The IRIS control system uses Video-Oculography eye tracking mechanism to detect the movement of the eye with the help of a low-cost video camera and passes navigation commands over a Bluetooth channel to an NXT Lego Mindstorm robot and home automation unit via the X10 wireless transmitter. The top-level system architecture for the two applications developed, robotic navigation and home automation is shown in Figure 1.

Figure 1: IRIS Control System Architecture 3.1

Eye Tracker Hardware Development The IRIS control system consists of a USB camera and infrared LEDs. Micro IR wireless cameras which are available in the market are expensive and have a low frame rate compared to the inexpensive, high frame rate (60-100 fps) PS3 eye camera which can be easily modified into an IR camera [19]. The eye tracker developed consists of the PS3 eye-camera that captures real time video frames of the eye movements and sends these video frames to the IRIS control system. The eye tracker also consists of an array of Infrared Light Emitting Diodes (IR LEDs) which are housed close to the eye. The main function of the IR LED is to illuminate infrared light around the eye which helps in pupil detection. This process of detecting the pupil is known as corneal-reflection/pupil-centre method [20]. In doing so, the image received consists of two points of interest, the bright pupil reflection and the corneal reflection, as shown in Figure 2. In this paper we consider the bright pupil reflection as the object of interest and track the movement of the pupil which in turns corresponds to different eye movements.

Figure 2: Detection using the bright pupil-detection method Thus, the IR LEDs help the PS3 eye camera to capture a clear image of the bright pupil reflection which gives a more accurate result and enhances the image processing technique [21]. The difference between the images of the eye with and without the IR LEDs being activated is shown in Figure 3. It can be seen clearly from Figure 3, that the IR illumination and bright pupil reflection are key factors for accurate pupil detection and in negating interference from ambient light.

Figure 3: Importance of IR LEDs (Screenshot from the Software Developed) The IRIS eye tracker consists of adjustable aluminium rods and blocks which house the IR LED array and the PS3 eye camera. All these components are housed on an industrial safety glass housing unit along with a

688


button-controlled power source for the IR LED array as shown in Figure 4. Figure 5 shows the developed IRIS eye tracker.

Figure 4: Assembly of the Eye Tracker Unit 3.2

Figure 5: Developed Eye Tracker

NXT Robot & X10 Transmitter Once the eye movements have been detected by the IRIS control system, it analyses the movements and determines the corresponding control command to be transmitted over a wireless channel to the respective application device (i.e. NXT robot or the X10 lamp device). The control system in this case is a HP Pavilion Laptop running on Windows Vista, Home Edition, which acts as an interface between the eye tracker and the application devices. The IRIS control system uses two forms of wireless communication protocols, i.e. Bluetooth 2.0 for robotic navigation using the HP Integrated Bluetooth Module and X10 for home automation using the X10 transmitter unit. A robot that was designed and assembled using the LEGO Mindstorm NXT programmable robotics kit is shown in Figure 6. It consists of the NXT Intelligent Brick [22], which is a 32-bit programmable microprocessor that enables communication with the IRIS control system over a Bluetooth channel. Thus, depending on the command received from the IRIS control system the NXT-brick instructs the respective LEGO servo motors to rotate in the desired direction, speed and turn ratio, thereby navigating the robot.

Figure 6: Assembled NXT Lego Mindstorm Robot The X10 transmitter (CM17A) uses the X10 transmission protocol which is a general purpose network technology used in home automation. It is an open standard for the wireless control of household devices and appliances [18]. The transmission involves short bursts of radio frequency signals that carry control signals in the digital form, such as turn ON and turn OFF signals. The X10 transmitter, which is connected to the IRIS control system, sends wireless control signals to the X10 lamp device which acts as a relay switch and depending on the signal received, it turns ON or turns OFF the lamp connected to it. 4.

IRIS CONTROL SYSTEM SOFTWARE The IRIS system is a sophisticated system that requires the integration of both hardware and appropriate software. Visual C# programming language under the .NET framework of Microsoft Visual Studio, 2008, was used to develop the IRIS eye tracking control software. It uses libraries from AForge.NET, which is an open source C# framework designed for developers and researchers [23]. The framework is used in the field of computer vision for image processing and robotics that comprises of a set of libraries of which the following are used in the software development of the IRIS control system:  

AForge.Imaging AForge.Imaging.Filters

689


 AForge.Video.DirectShow  AForge.Robotics.Lego The software for the IRIS control system can be broken down into several software modules which support the respective applications such as robotic navigation and home automation. The video capturing module retrieves the real-time video frame from the PS3 eye camera and provides the frame to the image processing module. The image processing module is responsible for applying different image filters on the received video frame to facilitate the pupil detection stage. Three different image processing filters are used in sequence. First the greyscale filter is used to convert the colour image into a greyscale image. Second the invert filter is used to invert the image since the colour filter in the final step does not detect the colour black. Hence, the pupil that appears black in colour, prior to this stage will be converted to white and the remaining part of the image is converted to black, thus removing all unwanted objects from the video frame. The final filter used is the colour filter which filters the pixels from the image that are not within the specified RGB range (i.e. set the colour white or fine tune by varying the RGB values until you receive a clear image of only the pupil). The general IRIS software architecture is shown in Figure 7.

Figure 7: Software Architecture The pupil detection module is responsible for detecting the bright pupil and its movement from the processed video frames. The Blob Counter class is used here to filter the objects of interest which in this case is the pupil (white object). The original and the processed video frames are both shown on the computer screen. A rectangle is drawn on the original video frame displayed on the screen around the object of interest (i.e. the blob) to show the detected objects (i.e. the pupil), shown in Figure 3. The calibration module calculates the four co-ordinates of the rectangle blob drawn around the object of interest (pupil). These co-ordinates are used in the program to further map the image co-ordinates on to the main screen and use them as reference points in various applications. The calibration process is performed in order to get the boundary co-ordinates of the screen with respect to the image co-ordinates and then map the image co-ordinates onto the corresponding points of the main screen. Figure 8 shows the calibration points on the screen which the user has to look at in order to complete the calibration process.

Figure 8: Calibration Points on the Developed Software Each calibration point is viewed for a period of 100 recordings of the x & y coordinates and the average value of the coordinates (x1, y1) and (x2, y2) of the blob are considered as the final calibrated image coordinates for that particular point on the main screen as shown in Figure 8. Hence the values for the (x1, y1) and (x2, y2) co-ordinates are:

690


x11, y11, x21, y21 = previous x & y coordinate; x1, y1, x2, y2 = current x & y coordinate On acquiring the corner coordinates (x1, y1) and (x2, y2) we can plot the remaining two coordinates (x2, y1) and (x1, y2) which correspond to the main screen coordinates (a2, b1) and (a1, b2) respectively as shown in figure 8.

Main Screen width = a2 – a1;

Main Screen length = b2 – b1;

Image Screen width = x2 – x1;

Image Screen length = y2 – y1;

For every increment/decrement in the image x-coordinate/y-coordinate value the calibrated point on the main screen will shift by

Thus, for any image coordinate (x, y) the system will map it to the corresponding main screen coordinate (a, b). A high value of (x2 – x1) and (y2 – y1) signifies more system accuracy and higher degree of calibration. A summary of all the classes, methods and fields used from AForge.NET framework for the software development is as shown in Table 1. Table 1: AForge.NET Framework for Image Processing

5.

IRIS CONTROL SYSTEM APPLICATIONS The IRIS eye tracking control system developed was used to test the performance of the system in remote robotic navigation and wireless home automation applications.

691


5.1

Robotic Navigation The first application developed to test the IRIS eye tracking based control system is the Robotic Navigation Application. In this application the IRIS control system remotely controls a robot by passing commands to the robot over a wireless channel. According to the point on the screen the user is looking at the corresponding navigation commands are determined. The user looks at the left side of the screen to make the robot move left, right side of the screen to make the robot move right and looks at the centre of the screen to keep the robot moving forward. The bottom-centre of the screen provides an emergency stop signal that completely halts the robot. The GUI developed using Visual C# for this application is shown in Figure 9.

Figure 9: GUI for Robotic Navigation The AForge.Robotics.Lego Namespace provides the necessary fields required to control the different parameters of the Lego Servo Motors. A list of all the parameters used in the software development is shown in Table 2. Motor C is activated to navigate the robot left, motor B is activated to navigate the robot right and motor B & C are activated to move the robot forward and deactivated to stop the robot. The code snippets for robotic navigation using the fields from Table 2 are shown from Figure 10-13. Table 2: AForge.NET Framework for LEGO Robotics

Figure 10: Code Snippet - Left Navigate

Figure 11: Code Snippet - Right Navigate

692


Figure 12: Code Snippet - Forward Navigate 5.2

Figure 13: Code Snippet – Stop

Home Automation The second application developed to test the working of the IRIS control system is the Wireless Home Automation Application. The main aim of this application is to provide control over household lighting for the physically challenged people by tracking the movement of their eye and providing control signals to the lighting unit. This concept is achieved with the help of the X10 transmitter and the X10 lamp device which acts as a wireless relay switch. The user looks left to turn OFF the light and looks right to turn ON the light with the help of the GUI developed using C# shown in Figure 14. The code snippet for home automation written in C# is shown in Figure 15 where the threshold point corresponding to ‗newtrackx‘ is calculated during the calibration process. If the user looks at a point on the screen that corresponds to a value trackx > (newtrackx + offset), the light turns OFF and if the user looks at a point on the screen that corresponds to a trackx < (newtrackx – offset), the light turns ON. Using the X10-CM17A class, the command X10HouseCode.A selects the X10 lamp device which is set to ‗A‘ on channel ‗1‘ and the command X10Command.TurnOff/TurnOn performs the necessary turn OFF/ON operation.

Figure 14: User Interface for Home Automation

Figure 15: Code Snippet for Home Automation 6.

EXPERIMENTAL RESULTS The two applications developed were used for evaluating the performance of the IRIS control system. The system response time and the system error percentage were measured for each application. The response time of the system is the time elapsed once the control signal has been activated and the corresponding operation has been executed. Figure 16 shows the system response time for 16 different commands. The average system response time of the 16 commands executed was measured to be 2.95 seconds.

693


Figure 16: System Response Time Plot However, the main reason for the delay was the transmission time involved in transmitting the navigation command, from the control system, to the NXT robot via a Bluetooth channel. The delay clocked at 1.6 seconds was because every time a navigation command was sent from the control system, the motors are reprogrammed by the NXT processor since the program is being transmitted from an external device. Hence the average system response of only the eye tracking system was 1.65 seconds. The remaining delay was mainly due to the numerous filters being applied to the same video frame. This is the main disadvantage of AForge.NET framework as the libraries provided were too slow for a fast real-time eye tracking application such as providing mobility for the physically challenged people. The main objective of calculating the system error percentage is to determine whether the command transmitted is the command executed. The experiment transmits a series of 20 consecutive ON/OFF commands using the X10 transmitter which are recorded as the transmitted commands and after the response time of 1.65 seconds elapses, the command transmitted is cross checked with the command executed. If the command transmitted and the command executed differs then an error is registered. The red dots in Figure 17 are commands showing an error which has a response time of over 1.65 seconds. Hence the control system does not execute this command and the next transmitted command gets executed. The error percentage of the IRIS control system is calculated as 15%, shown in Table 3. Note that the X10 transmitter/receiver device has no delay in transmission unlike the NXT robot.

Figure 17: System Error Plot Table 3: Error Percentage of the IRIS control system Commands Transmitted Commands Executed ON commands – 10

ON commands – 9

OFF commands – 10

OFF commands – 8

694

Error Percentage 15%


7.

CONCLUSION The AForge.NET framework used to develop the IRIS control system is a useful tool in software development since it is an open source framework with the available library files for image processing and the integration of the NXT robot and the X10 transmitter. However, the sequence of filters chosen using the AForge.NET framework proved to produce a processing delay since the image is being processed individually by the three filters, thus, making the IRIS control system prone to errors. Hence, through this paper, the performance of a low-cost real-time eye tracking system has been tested using the AForge.NET framework in the field of wireless robotic remote navigation and wireless home automation. The test results yielded an average response time of 1.65 seconds along with a system error percentage of 15%. The IRIS control system can be further extended to applications in robotic wheelchair control, computer mouse control and gaming, mobile phone applications such as making phone calls and sending/receiving text messages. With the rapid accretion and extensive research in the field of Human Computer Interaction and Assistive Technology, eye tracking will soon be as common and easily available as mobile phones or computers.

8.

REFERENCES [8] M. Betke, J. Gips, and P. Fleming, ―The camera mouse: Visual tracking of body features to provide computer access for people with severe disabilities‖, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 10:1, pp 1–10, March, 2002. [9] J.J. Magee, M.R. Scott, B.N. Waber, and M. Betke, ―Eyekeys: A real-time vision interface based on gaze detection from a low-grade video camera‖, Proceedings of the IEEE Workshop on Real-Time Vision for Human-Computer Interaction, Washington D.C., July 2004. [10] N. Garay, I. Cearreta, J.M. Lopez and I. Fajardo, ―Assistive Technology and Effective Mediation‖, An Interdisciplinary Journal on Humans in ICT Environments, vol. 2:1, pp 55-83, April 2006. [11] S. Azam, A. Khan and M.S.H. Khiyal, ―Design and Implementation of a Human Computer Interface Tracking System based on Multiple Eye Features‖, Journal of Theoretical and Applied Information Technology, vol. 9:2, November, 2009. [12] Tobii X60 & X120 Eye Trackers. Internet: [13] http://www.tobii.com/scientific_research/products_services/eye_tracking_hardware/tobii_x120_eye_tr acker.aspx [Accessed: 10 May 2011] [14] J.D. Smith, R. Vertegaal and C. Sohn, ―View Pointer: lightweight calibration-free eye tracking for ubiquitous hands-free deixis‖, Proceedings of the 18th annual ACM symposium on User interface software and technology, Seattle, WA, USA, October 23-26, 2005. [15] A.T. Duchowski, Eye Tracking Methodology – Theory and Practice, 1st edition. Great Britain: Springer-Verlag London Limited, 2003. [16] M. Schiessl, S. Duda, A. Tholke and R. Fischer, ―Eye tracking and its application in usability and media research‖, Eye Square, Berlin. [17] Jacob, R. J. K. and K. S. Karn, ―Eye Tracking in Human-Computer Interaction and Usability Research: Ready to Deliver the Promises‖, The Mind's eye: Cognitive and Applied Aspects of Eye Movement Research, J. Hyona, R. Radach, and H. Deubel, Editors. Elsevier Science: Amsterdam. pp. 573--605, 2003. [18] P.J. Murphy, A.L. Duncan, A.J. Glennie, P.C. Knox, ―The Effect of Scleral Search Coil Lens wear on the Eye‖, Br J Ophthalmol, vol. 85, pp 332–335, 2001. [19] R. Barea, L. Boquete, M. Mazo, and E. López, ―System for Assisted Mobility using Eye Movements based on Electrooculography,‖ IEEE Transactions in Neural System and Rehabilitation Engineering, vol. 10, pp. 209–218, 2002. [20] R. Barea, L.Boquete, L.M Bergasa, E. Lopez & M. Mazo, ―Electro-Oculography Guidance of a wheelchair using eye movement‘s codification‖, International journal of Robotic Research, 2003. [21] E. Vitte, A. Sémont, ―Assessment of Vestibular Function by Videonystagmoscopy‖, J Vestib Res, vol. 5(4), pp 1-7, 1995. [22] Gans, E. Richard, ―Video-Oculography: A new diagnostic technology for vestibular patients‖, Hearing Journal, Vol. 54(5), pp 40, 42, May 2001. [23] C. Ehmke and S. Wilson, (2007) ―Identifying web usability problems from eye-tracking data‖, Proceedings of the 21st British HCI Group Annual Conference on People and Computers, vol. 1, pp 119–128, 2007. [24] H. Yanco, ―Wheelesley: A robotic wheelchair system: Indoor Navigation and User Interface,‖ in Assistive Technology and Artificial Intelligence—Application in Robotics, User Interfaces and Natural Language Processing, Eds. Heidelberg, Germany: Springer, pp. 256–286, 1998.

695


[25] D. Bonino and F. Corno. Dogont, ―An ontology modeling for intelligent domotic environments‖. In 7th International Semantic Web Conference, 2008. [26] ―X10 Transmission Theory‖, Internet: http://www.x10.com/support/technology1.htm , [April 28, 2011] [27] ―The-EyeWriter.pdf‖, Internet: http://fffff.at/eyewriter/The-EyeWriter.pdf [April 28, 2011]. [28] Goldberg, H. J., & Wichansky, A. M. (2003). Eye tracking in usability evaluation: A practitioner‘s guide. In J. Hyönä, R. Radach, & H. Deubel (Eds.), The mind's eye: Cognitive and applied aspects of eye movement research (pp. 493-516). Amsterdam: Elsevier. [29] J.S. Babcock and J.B. Pelz, ―Building a lightweight eyetracking headgear‖, Rochester Institute of Technology. [30] Lego Mindstorm. Internet: http://shop.lego.com/Product/?p=9841 [Accessed: 2nd August 2010] [31] ―AForge.Net: Framework‘s Documentation‖, Internet: [32] http://www.aforgenet.com/framework/documentation.html [April 28, 2011].

696


STRUCTURED LIGHT BASED VISUAL NAVIGATION ON ROCKY TERRAIN FOR A SIX WHEEL LUNAR ROVER Vikalp Sachan1, K. S. Venkatesh2, and Ashish Dutta3 1

2

Dept. of Electrical Engineering, IIT Kanpur, Kanpur 208016, India. e-mail1: [email protected]

Dept. of Mechanical Engineering, IIT Kanpur, Kanpur 208016, India. e-mail2: [email protected]

ABSTRACT In the last few years there has been a renewed interest in the exploration of the moon. Lunar rovers need accurate map building and navigation algorithms for successful navigation on highly rough terrain, containing boulders, craters, dust etc. Past research in the area of map building and navigation use non-contact sensors e.g. time of flight approach using lasers, stereo cameras and other methods. Such methods are often slow and are not very accurate. In this paper we propose map generation on a 3D terrain using structured light for a six wheel rover navigation problem. A line laser is mounted on a rover and rotated at intervals of 1.8 o. At each instant a camera captures the laser profile and at the end of the scan combines all the laser profiles to generate a 3D map. A grid gradient map is generated for the profile from which a navigation index is computed to give the best forward path for navigation. Experimental evaluation of the rover with obstacles and inclinations verify the effectiveness of the proposed method. Keyworks: structured light, lunar rover, laser, camera calibration, patch, navigability. 1.

INTRODUCTION Structured light is a well known and efficient technique to recover the shape of 3D objects by projecting a specially designed light source on in [5]. A camera is used to visualise from an angle the deformation in the reflection for the known light source. In other methods such as, in the time of flight approach a laser is used to emit a pulse of light and the detector measures the amount of time before the reflected laser pulse is seen. The laser rangefinder finds the distance of a surface by timing the round-trip time of a pulse of light. Timeof-flight devices are also available in a 2D configuration as in a Time-of-flight camera. Stereo cameras and stereo vision based depth measurement are prone to sensor noise, and the ever-daunting problem of dense point correspondence errors [6], and heavy computational cost, eg, for a system such as [1], compared to structured light depth measurement, where computational costs are minimal, making it attractive for real-time operation. The triangulation 3D laser scanner is also an active scanner that uses laser light to probe the environment, It can be considered the most primitive format of structured light scanner. In this approach, a laser is projected on an object and a camera placed at slightly offset position looks for the location of the laser dot. Depending on how far away the laser strikes a surface, the laser dot appears at different places in the camera‘s field of view. Such methods are excruciatingly slow, and suffer from occasion failures due to dot occlusion. In the more general structured light based approach, a pattern of light is projected on the surface to be mapped and we look at the deformation of the pattern due to change in surface gradient. The pattern may be one dimensional or two dimensional. The line is projected onto the subject using either an LCD projector or a sweeping laser. The advantage of structured-light 3D scanner is high speed and accuracy of the distanced measured. Instead of scanning one point at a time, structured light scanners scan multiple points or the entire field of view at once. This reduces the problem of distortion from motion. The primary considerations underlying the design of a structured light system include structuring the light so as to allow depth measurements at multiple points simultaneously, to increase scanning speed, while ensuring that the geometry of the structured light obviates ambiguity in the measurements for a

697


predetermined or pre-estimated range of depth values. Where no such estimate is available a-priori, the safe choice of the light structure is a line [8]. In this paper we propose the map generation and navigation on rocky terrain using structured light, where one cannot make assumptions of untextured piecewise-planar surfaces as in [2]. Unlike in [3], where the challenge is an indoor environment that is more structured, and prone to far less illumination variations, our problem is the lunar surface, with ambient illumination capable of varying through 8-10 orders of magnitude sometimes within the same scene. Severe power availability constraints on the craft dictate the choice of a laser against halogen sources. Calibration considerations sometimes dominate the design [4, 7] of structured light systems, but in an application such as lunar terrain navigation with stringent robustness requirements and remote operation, we prefer to rely upon a one-time pre-launch calibration of the system using a known flat graduated surface such as a checkerboard in two orthogonal planes [8]. The expected range of operation is covered by this fixed calibration of the system. In our experiments a diode laser has been used to generate a line pattern, which is projected in the surrounding. The diode laser beam is passed through plano cylindrical lens, which diverges the point laser into a line laser pattern. The wavelength of the diode laser is 650nm and output power of laser is 100mW. 2.

STRUCTURED LIGHT BASED SYSTEM A line laser and a camera are mounted on a rotating platform on the rover as shown in Figure 1. The laser platform is rotated and the terrain is scanned at an interval of 1.8 degrees. At each instant the camera captures the profile of the line laser and at the end of the total scan combines all the frames to develop the 3D map of the surface. The laser profile images are sampled to get surface points (Figure 2). The cameras are calibrated using a chess board pattern before use. The (x, y) and (u, v) coordinates of the sampled points on the reference laser line are calculated using Hxy and Huv respectively. The calculation is shown as : X= Hxy X‘ U= Huv X‘ where X‘ gives the image plane coordinates. X and U give the xy-plane and uv-plane coordinates respectively.

Fig. 1. Two view of the rover showing the camera and the laser source on the rotating platform.

698


(a)

(b)

Figure 2. (a) Laser beam on flat surface (b) Laser beam deformed due to change in gradient.

Terrain data is presented in terms of 3-d coordinates. The x and y coordinates give us the location of points on the horizontal plane. Z coordinates gives us height information of terrain. So terrain can seen as height function which depends on the locations(X;Y). Height at each point(X;Y), f (X;Y) = Z; but the terrain is registered as a set of points each having (X;Y;Z)coordinates the Z, height information is obtained from the shift that is seen in the laser profile. The (X;Y) coordinates are obtained from the straight line profile. The camera is calibrated with respect to two planes. One is the laser plane and other is horizontal plane that is the surface touching the base of robot wheels. Camera calibration is done using checker board pattern. In laser profile, we project each point on the reference line. We calibrate the reference line with respect to the centre of rotary platform. The projected point gives the radial distance of point from the centre, and the pixel shift from projected point to reference point on straight line, gives height information of point ‘P‘.

Mathematically let equation of reference plane Ax + By + Cz + D = 0, where A,B,C,D are known value. for any given laser point (xl,yl,zl), projected point(xp,yp,zp) are obtained through following calculation.

Equation of straight line that passes through laser point and it‘s projected point,

X –xl A

= Y – yl =

Z – zl

B

C

(xp,yp,zp) lies on straight line and on plane as well. Axp + Byp + Czp + D = 0

&

xp –xl

= yp – yl = zp– zl

A

B

699

C


Both equation gives (xp,yp,zp) coordinates. The map contains the average gradient information of the surface in terms of small grids of size 15x15mm for a total radial length of 2m. Several parameters have been identified to determine a navigability index that gives the best path for a six wheel rover. The rover is allowed to move over obstacles less than half its wheel diameter. Also the gradient of a plane the rovers tries to climb should be less than 30 degrees. 2.1

VISUAL MAP GENERATION The complete 360 degrees profile has been divided into grids and then regions are given different colours based on their relative gradients as shown in Figure 3.White color Zone: This is the reference plane of the visual map. This is the plane created by bottom of the wheels in robot. All height or depth of points will be measured with respect to this plane. Red color Zone:This is detectable obstacle zone in the visual map. Any form of boulder will belongs to this zone. Purple color Zone: This is self obstacle zone in the map. This is created by robot itself and it also includes clearance or offset distance which is not scanned by laser. Radius of this zone is called as inner radius of visual map. Black color Zone: In general this zone represents the portion or area which is not scanned by setup. This is possible due to various reasons like discontinuity in laser due to boulder or crater. Blue color Zone: This is crater zone in the map. It belongs to the set of points with negative height in the map.

Figure 3. Different colours signify different gradients in the grid map.

2.2

DETECTION OF CONVEXCITIES When a laser line is projected on terrain it may take multiple fold like concave or convex. In case of convex surface front curvature of terrain is captured properly as it. Curvature of other side remains unknown to us because reflection of laser line is not captured due to hindrance provided by surface itself Figure 4. In situations like moving over to the boulder the front surface and unknown surface both are equally important to us. In our experiment we are scanning the surrounding about 1.2m height to the ground. Proportion of unknown surface will be lesser if height of the boulder is significantly smaller than 1.2m or boulder is closer to rover.

700


Figure 4. Scan of a convex object in which one side is not visible to the laser. A rock on the surface is analysed as a collection of straight lines with gradients as shown in Figure 5. and correlated to the kinematic parameters of the rover. If the combined parameters are acceptable the rover is allowed to travel over the rock. As with all visual navigation methods the rear of objects cannot be seen and hence their gradients cannot be inferred. The developed algorithm outputs a map in this the safe navigation parts are in white colour and red indicates obstacles. A useful feature of this approach is its ability to also detect and profile ‗negative‘ obstacles such as craters, which cannot be seen like a stone, but which still need to be avoided in a navigation task.

Figure 5. A rock approximated with multiple gradients.

2.3

DETECTION OF CONCAVITIES USING VISION Crater are captured with partial curvature missing in visual map. Large portion of crater can be obtained through increasing the height of visual platform. As compare to SONAR based navigation system, VISION based navigation system can detect the presence of crater in region with significant information as in Figure 6.

Figure 6. Scan of a concave object in which nearer side is not visible to the laser.

701


Figure 7. A generic curvature profile of crater with initial discontinuity.

The crater is characterized with two parameter known depth D,unknown length L to verify the navigation feasibility (Figure 7). Where

D = Z(p2) - Z(p3) L = R(p2) - R(p3)

where Z is height function, R is radial function and p2 is projected point of p3 on reference plane. If either of D or L is higher, crater is sensitive to cross. Special techniques of rover kinematics will be required for getting out of the crater.

2.4

PATCH When rover moves in certain direction its own hindrance affects the navigation. Patch is area that is covered by rover itself in the map (Figure 8).

Figure 8. A patch for certain direction.

This patch undergoes for point operation with calculation of local slopes,discontinuity measurement ,etc. Gradient at ith point,

gi = (zi+1 - zi)/ (ri+1 - ri) .

local slope = ∑i= L P ( gi ) /(P-L).

For all

gi < tan(850) & gi > - tan(850)

tan(850) is cutoff value for gradient and tan(300) is cutoff value for local slope (Figure 9).

702


Figure 9. Local slope for any point P.

3.

EXPERIMENTAL EVALUATION First rocks were randomly placed on a flat floor (Figure 10) and scanned by the structured light system. The results are as shown in Figure 11 that shows one big rock and a brick placed on a flat surface. Next inclined floors were made of known inclination and tested using the structured light system, as shown in Figure 12. The errors in regenerating the height of the obstacles and the inclination are as given in Table 1.

Figure 10. Rock on a flat floor.

703


Figure 11. Visual map of previous scene with scan of 1800

Figure 12. Inclined floor.

Table 1. Errors in regenerated 3D map surface. 4.

NAVIGATION INDEX FOR ROVER MOTION After the map is generated the best path that the robot should follow is decided based on the navigation index. The rover is as shown on Figure 13 and there are three wheels at each side. The navigation index consists of the gradients of the wheels on either side as well as the gradient of the central region during a motion forward. If the gradient for any track is more than 60 degrees combines (30 degrees on either side) the rover will not try to climb such profiles. Also if the wheels are not more than 60 degrees combined gradient but the central region between the wheels is more than 60 mm than also the rover will not try to follow the path. In Figure no. 14 navigability is zero for some initial and last directions. It is because of complete patch formation does not take place most likely in left and right sides. Navigability is set to 0. Navigability is lower at the middle, when it encounters the rock and it is higher in case of brick.

704


Figure 13. Kinematic diagram of the rover with frames assigned to joints and its prototype.

Figure 14. Plot of navigability vs. directional index of Fig no. 11 visual map 5.

CONCLUSION A structured light navigation system for a lunar rover system was presented in this paper. Unlike other methods like stereo vision, this method can give very accurate maps. The developed navigation index can also generate the best paths for the rover.

6.

REFERENCES [1] A. Joseph Nsasi Bakambu, Pierre Allard, and Erick Dupuis. 3D terrain modeling for rover localization and navigation. Proceedings of the IEEE Computer Society 3rd Canadian conference on computer and robot vision, pp. 61-67, 2006. [2] T. Heitzmann, C.Doignon, C.Albitar,P.Graebling,‖Position based visual servoing using a coded structured light‖,IEEE Internation workshop on robotics and sensors environments Ottawa-canada,1718,October 2008. [3] Nosan Kwak.Gon-Woo Kim - Sang-Hoon Ji- Beom -Hee Lee ―A mobile Robot Exploration strategy with low cost sonar and tungsten-Halogen Structured Light, J Intell Robot Syst (2008) [4] B.Zhang, Y.F.Li, Y.H. Wu, Self-recalibration of a structured light system via plane-based homography,Pattern recognition., 2006. [5] David Fofi, Joaquim Salvi,El Mustapha Mouaddib,Uncalibrated reconstruction: an adaptation to structured light vision, pattern recognition September 2002 [6] Jens-Steffen Gutmann, Masaki Fukuchi and Masahiro fujita,3D Perception and Environment Map Generation for Humanoid Robot Navigation, The International Journal of Robotics Research, 2007. [7] J. Apolinar Muñoz Rodríguez, Online self-calibration for mobile vision based on laser imaging and computer algorithms, Optics and Lasers in Engineering Volume 49, Issue 6, June 2011, Pages 680-692 [8] Sidharth R Varier, Amey Vaidya, K S Venkatesh, Novel representations, techniques and error evaluation for 3D reconstruction, AMDO'10, Proceedings of the 6th international conference on Articulated motion and deformable objects, Pages: 148-161, 2010.

705


MINING ROBOTICS SENSORS, PERCEPTION SENSORS ON A MINE SAFETY PLATFORM Green JJ1, Hlophe K2, Dickens J3, Teleka R4, Mathew Price5 (1,2,3,4)

CMI, CSIR P.O.Box 91230, Auckland Park, 2006, Johannesburg, South Africa [email protected], [email protected] ,[email protected] [email protected] 5 Cogency 4 Swift Street, Salt River, 7925, Cape Town, South Africa [email protected] Underground mining robotics has not enjoyed the same technology advances as above ground mining. This paper examines sensing technologies that could enable the development of underground autonomous vehicles. Specifically, we explore a combination of three-dimensional cameras (SR 4000 and XBOX Kinect) and a thermal imaging sensor (FLIR A300) in order to create 3d thermal models of narrow mining stopes. This information can be used in determining the risk of rockfall in an underground mine, which is a major causes of fatalities in underground narrow reef mining. Data are gathered and processed from multiple underground mine sources, and techniques such as surfel modeling and synthetic view generation are explored towards creating visualisations of the data that could be used by miners to monitor areas of risk in the stope. Further work will determine this potential. Keywords: Underground Mining Robotics, Perception sensors, Sensor Fusion, Infrared Camera, 1.

INTRODUCTION To date, robotics in the mining industry has seen a lot of advancement in automation for above ground applications where positioning can be achieved with GPS, and often enhanced by a combination of DGPS and machine vision techniques. Underground mining however, has not seen the same advancement due in part to the harsh conditions and inherent difficulties in navigating a robot through a rough, lightless environment. The lack of infrastructure and communication channels across the mine has also hampered the development of autonomous stope systems. Some progress has been made in tramming activities in tunnels with the application of repetitive wall following from load to unload points [1] but no level of autonomy has yet been achieved. This paper looks at attempts to address the significant challenge areas involved in developing a mine safety robot that will enhance mine safety in the stopes of South African Hard Rock mines [2]. A number of critical developments are required for enabling a robot to sense and navigate in the harsh underground stope environment [3]. We discuss some of the issues and show the progress made to date. The current objective of the work is to identify the technical risk areas that are barriers to the implementation of underground robots. We focus on a localization solution based on differential time-offlight (dTOF) beacons, a combined multiple sensor system for visualization in confined, lightless environments, and thermography for assessing the safety and stability of hanging walls. Over the last decade approximately 200 miners have lost their lives per year in South Africa in underground mining activities (128 in 2010) [4]- 50% of those deaths occurred in the stopes, and of those, 70% in rock fall related incidents. It is envisaged that a robot could be used to enhance mine safety by gathering data in the stope (e.g. after a blast) and warn miners of potentially unsafe areas, that must be avoided, prior to their re-entry. Furthermore, the collected data can be used to focus making safe activities on unstable hanging walls. Data will be gathered by the robot using, inter alia, a long-wave infrared (thermal) camera, a short-wave infrared time-of-flight camera, a 3D imaging device, and a sounding device for assessing hanging wall (roof) stability. It will then be combined and processed before being presented to the miner prior to entering the stope. As the eventual deployment of this technology will be on a mobile platform, the goal of realising real-time processing for 3D stope mapping must be kept in mind. While our current implementations process data off-line, the real-time requirement was considered and continues to guide the direction of development.

706


Section 2 covers the localisation techniques for determining where the data is collected. Sections 3 and 4 then cover the thermal and 3D sensors used before section 5 discusses the combination of data sets and potential visualisation techniques that could make the information about risk available to the miners. 2.

LOCALISATION For localizing the robot underground, which is a major challenge, CSIR are pursuing the implementation of a difference in time-of-flight beacon system [5]. Beacons are surveyed into the underground environment and simultaneously transmit ultrasonic and RF signals. The differential time-of-flight between the ultrasonic signal and the RF signal is used to estimate the distance between the beacon (Figure 95) and the receiver [6].

Figure 95: Multi directional dTOF Beacon. Resulting positions will be further refined with on-board inertial sensors using an unscented Kalman Filter (UKF). This will allow the positional uncertainty of the robot to be bounded and will improve registration of sensor data for real-time navigation and mapping with large data sets and multiple sensors. 3.

THERMAL IMAGING Electronic perception underground is hampered by adverse conditions, especially in the deep-level hard rock mines of South Africa where humidity reaches saturation, temperatures are in excess of 40°C, and the threat of explosion requires intrinsically safe construction of equipment. There is also no ambient light available, so any illumination must be carried on-board. To complicate matters, the vast distances underground require tetherless operation and all power must be supplied by battery. Finally, the presence of dust hampers the use of optical techniques, and the abundant use of water to combat the dust creates extreme conditions that are not conducive to the operation of most machinery. Infrared sensing poses a potential solution to some of these problems. Infrared light can penetrate dust very well due to its long wavelength. If the wavelength of an electromagnetic wave is larger than the diameter of an obscuring particle then the wave will tend to pass through it instead of being reflected or scattered. The long wavelength (7 – 14 μm) of thermal infrared allows it to penetrate airborne dust having smaller particle sizes.[7]. All objects radiate infrared at a wavelength dependant on their temperature, As the rock is already heated, the thermal emissions can be used to view the surroundings which negates the need for a standalone illumination system. Past Research [8] and [9-11] have explored the use of thermal imaging cameras in mining environments for the analysis of hanging wall stability. This is made possible by the use of ventilation air which is a vital component in most active South African hard rock mines. Cooled ventilation air is used to keep the stope environment within acceptable limits as the rock can reach temperatures in excess of 65°C. This cools the rock surface, creating a thermal gradient through the host rock. If a crack exists, the heat flow is interrupted and the surface rock will cool preferentially. Analysis of the thermal gradient can therefore be used to identify potentially loose rocks that pose a danger to mine workers. The same technology is applicable for mapping stope areas in order to create a virtual mine that could be used by a small machine to make operational decisions. Three experimental data sets of increasing size and complexity are discussed below. Initial data set.

707


A Wuhan Guide MobIR M2 hand-held thermal imaging camera was used for initial work. The field of view was limited for the application of inspection of a hanging wall in a 1m stope. Since the resolution of the camera is low (120x120 pixels) the resulting images do not provide sufficient coverage for robust detection of thermal features (associated with loose rock). Therefore, multiple overlapping images of an area of interest were collected and an automatic image stitching method was used. Figure 96 shows the first data set collected at Driefontein mine at a depth of approximately 3500m below surface. Stitching is achieved by extracting invariant features [12] for each image and finding a series of 2D homographies (a 3x3 mapping between 3D planes) for pairs of images. Even though the camera is translated during capture, a 2D homography was found to be sufficient for stitching due to the approximately planar nature of the rock surface. Unlike visible-light images in indoor scenes, invariant features are very sparse in the thermal images. This motivated us to use a simplified camera model based on only 4 parameters: 2D translation, scale, and rotation about the optical axis. The approximation worked well, because the images were captured by moving the camera in a parallel manner over the surface.

Figure 96. Stitched thermal images from first dataset (Driefontein Mine). The experiment clearly showed how the ability to stitch thermal images together to create a larger data set could be used to aid analysis. Specifically, it allowed for the determination of a temperature gradient across areas of rock much larger than the field of view of the sensor. It was noted with this first data set that the sharp temperature gradient visible coincided with a step in the hanging wall profile, and not necessarily with an unstable area of rock mass. A larger data set was subsequently captured. Data set 2. A second data set was captured with more overlap between images and more consistent camera motion (in accordance with our simplified model). Although the sequence was captured in an ordered manner, this was not used to determine the pairing relationships during stitching. Rather, robust feature matching is used to find matches between all possible image pairs. The stitched image is then generated by iteratively choosing the image pair with the highest number of matched features that has a common connection to the current merged set. Finally, the images are merged into a common reference frame using the pair-wise transformations and weighted blending function is used to remove the seams. It is apparent in Figure 97 that a metallic roof support (arrow shown) is visible in the image and is at a different temperature to the hanging wall rock. Since different materials have different thermal emissivity constants, this could be due to either a significantly different emissivity and similar temperature, or a real difference in temperature. As the support is metallic and the rock quartzite, and the ventilation air would preferentially cool the support as it protrudes into the airflow, it is likely that both reasons are valid. On the other hand, the blue (cooler) area at the bottom left of the image represents a cooler rock mass, and therefore indicates a potential threat where the rock mass is at risk of sudden separation. (Naturally, this was a concern to the author during data collection as he lay on his back taking images vertically upwards.) However, during the data collection it was noted that the cooler area represented an area that was protruding out of the hanging wall, and could be preferentially cooled having additional surface area exposed to the

708


cooled ventilation air. Therefore, the blue area was not conclusively a high-risk region. Additional data collection and verification is necessary to test this hypothesis in subsequent work.

Figure 97.Stitched thermal image blue=cold and red=warm (Driefontein Mine)

of

mine

hanging

wall

using

17

images,

As a result of this data set it was concluded that three-dimensional profile information is required in order to robustly assess risk from the thermographic data. This led to the development of a multi-sensor setup geared for both thermal and 3D acquisition. This is discussed further in Section 5. Subsequent data set (3rd) The third data set was collected at Bafikeng Rasimone Platimum Mine (BRPM) at a shallow depth of 250m below surface approximately 8 hours after blasting, and after cleaning had been completed. Drilling had commenced. The data was collected with the multi-sensor setup comprising a FLIR A300 thermal imaging camera and two 3D sensors. The following sections discuss the 3D sensors and their use in producing combined thermal and 3D data respectively. 4.

3D MAPPING To complement the thermal data with 3D rock structure, a logical step is to gather 3D data of the mine environment. Creating 3D maps of mines is not new [13], [14] used a laser scanner to map a mine portion and [15] successfully mapped a tunnel section from an abandoned mine. The approach taken with this work was to use a high-end sensor (Reigl laser scanner) to generate a groundtruth data set. This demonstrated the potential of 3D data for our purpose, and resulted in an investigation into less expensive, more portable options for our multi-sensor setup. Ultimately, this led to the selection of a SwissRanger SR4000 TOF (time-of-flight) camera, and later this was compared with the Microsoft XBOX Kinect sensor. The following sections describe each sensor. Riegl data set A ground truth data set was collected using the Riegl LMS-Z390i 3D laser scanner. It is large, heavy, cumbersome, and scans slowly. While an impressive amount of data was collected, the process of transporting the sensor to the mine, and moving it to new locations in the narrow stope environment made this a once-off activity. Newer alternatives that are more compact, such as the Faro focus 3D 20 and 120 scanners are available but they are still prohibitively expensive, large and slow for mobile applications.

709


Figure 98: large scale Riegl laser scanner data set representations. Tunnel and stope data was collected, and the visualisations in Figure 98 show the best level of detail that can be achieved with current technology. Images are textured by depth – blue for close and red for far away. From the data it is seen that there is indeed sufficient resolution for determining whether an area of wall is protruding into the ventilation air. Therefore, further sensors were evaluated for a mobile sensor head. TOF data Set A 5 m range MESA Imaging SwissRanger SR4000 was used for the next data set collection at a shallow (600m) disused gold stope. The distance ambiguity problem (aliasing) [16] typically experienced with this type of sensor proved to be a significant issue for processing. In a tunnel environment this is a persistent problem since extreme points in the direction of the tunnel often exceed the range of the camera, resulting in aliased measurements. Similarly, in a stope environment the hanging wall (ceiling) and foot wall (floor) stretch continuously beyond the sensor range causing the same problem. Although this can be addressed by amplitude filtering, it adds unwanted post-processing and is not completely robust. As a consequence, another 3D sensor option was sought towards achieving a faster mapping solution for thermal data fusion. Figure 99 shows a 3D model generated from the TOF data. A median filter is used to reduce noise and ICP (iterative closest point) is used to register the 3D view of each range image. Following this, ray-carving is used to generate a triangular mesh model using Marching Tetrahedra polygonisation [17]. As the surface is traversed, points are projected into each range image and those that project to points in front of the closest range image are removed. This effectively carves away parts of the volume that are inconsistent with the set of range images. The model is post-processed with an iterative smoothing technique [18].

Figure 99. SR4000 data visualisation using a ray carving technique.(Gold Reef City Mine) XBOX Kinect data set During the evolution of the project a new sensor (the Kinect) was released, and has quickly become a hit in the robot community due to its low cost. The Microsoft XBOX Kinect has been successfully used in indoor mapping [19] using [20] and thus it was a natural progression to test the system underground. In contrast to

710


TOF, Kinect uses a near-IR projector and camera to scan the environment. A colour camera is also built into the unit, but for 3D scanning, using just the IR camera means that no external light source is required.

Figure 100. Two perspectives of the Kinect stope data set. (BRPM Mine) The sensor was tested in an office environment and then mounted together with the other sensors during acquisition at BRPM (details given below). Figure 100 shows two screen shots of different perspectives from the BRPM Kinect data set. As with the TOF data, ICP is used to register the sequence and obtain a global reference frame for each range image. However, instead of using ray-carving, which can be slow and memory intensive, a synthetic combined range image was generated allowing a triangular mesh to be quickly generated using the image connectivity. This method was found to be preferable as it is less processor intensive and lends itself well to our thermal image stitching problem. Occluded areas where the surface cannot be seen from the various viewpoints is clearly seen by the empty sections. Merging more scans from different viewpoints can be used to create a full model, and will be undertaken in subsequent work. 5.

COMBINATION OF THERMAL AND 3D DATA FOR VISUALISATION Preliminary results showed that although the uniformity of the rock temperature produces limited features, it is possible to stitch images together to form a single large image. The combined image extends the fieldof-view of the thermal imaging camera and facilitates analysis of potentially unstable areas.

The sensor head. Our multi-sensor setup thus consisted of the three aforementioned sensors, namely the FLIR A300, the SR4000, and the Kinect, which were all rigidly mounted on a tripod. Relative calibration between the sensors was achieved by collecting correspondences through the use of a calibration object. One of the difficulties of calibrating 3D and thermal cameras is that there is no common viewing spectrum. In one scenario, matching points were manually selected using the bottom of a beer crate. This worked well since the holes in the grid pattern create a depth difference as well as a thermal difference due to the varying background. However, manual calibration is slow so a more automated approach was sought in the form of a tennis ball mounted on a stick. A piece of backing card was mounted further back on the stick so that the ball could be clearly seen from its background. Thus, the sphere-like structure of the ball could be found in the range image, and by flexing the ball several times its glowing counterpart could be detected in the thermal image. Finally, by waving the ball around the scene multiple correspondences were automatically detected and used to calibrate the sensors. Figure 101 below shows images of the sensor head and calibration object.

711


Figure 101: Multi-sensor setup with SR4000 TOF camera, FLIR A300 IR camera and XBOX Kinect rigidly mounted on a tripod (left), and tennis ball calibration object(right). During data capture, scans were taken from a single viewing position with the sensor head rotated about a pivot on a tripod. This provided a stable scanning path that facilitated ICP registration since the motion was less irregular. Data fusion Given the relative calibration between the thermal and 3D sensors, the correspondence between 3D points and their temperatures can be established by projecting into the thermal image. This enables texturing the model with the thermal data. One caveat: since the 3D camera streams approximately 10 times faster than the thermal imaging camera, it is important to ensure good synchronisation so that the thermal texture is correctly aligned. Here, we used a simple time stamping approach to obtain a global reference during capture. During texturing, the 3D and thermal images with the smallest time difference are selected (or rejected if a maximum threshold is reached). Figure 102 shows an example where a single Kinect range image is textured with its closest (temporally) thermal image.

Figure 102. Single frame alignment fusing XBOX kinect and FLIR A300 IR camera. Extending the above case to multiple scans with multiple thermal images requires that the 3D scans be registered in a common reference frame. Using ICP, we obtained an alignment for the BRPM data set and used this to generate a combined model as described in the previous section (3D mapping). Since each thermal image can be linked to its closest range image, the set of thermal images can be addressed in the common reference frame. Therefore, by projecting points from the combined model into the set of thermal images the combined model can be textured from multiple thermal images. We use a greedy approach where points are assigned a temperature from the most recent thermal image. Once assigned, the point is removed from the processing queue and only the remaining points are textured in the following iteration. Once all points have been processed or all thermal images have been exhausted the process terminates. A number of other merging strategies could be applied, such as weighted average. However, the FLIR camera produces extremely repeatable measurements and further filtering was not required.

712


Figure 103. Merged 3D model textured with multiple thermal images (BRPM Mine). Combining thermal and 3D data in this manner therefore provides sufficient information for a machine intelligence algorithm to determine potentially hazardous areas of the mine wall. The next phase of work will involve developing such an algorithm through the collection of multiple data sets that can be labelled by an expert. This will enable us to use the 3D thermal model to estimate the likelihood of a cooler area of rock being loose as opposed to protruding, and therefore having the potential to separate from the host rock. An annotated visualisation can then be generated that can be used by miners to identify rocks that pose a danger. Visualisation Several visualisation techniques were explored so that humans can interact with data. Rendering a 3D textured model as shown in the previous figure is one method of visualisation. However, it may be better to present thermal and 3D structure as separate 2D images so that the distinction between 3D structure and temperatures can be readily seen. Figure 104 below shows an example where the textured model is reprojected into a virtual camera creating a stitched thermal image similar to those that were generated for the initial data set. However, here we are not constrained by our original simplified assumptions of near planar scenes because actual 3D points are known.

Figure 104: Wide angle visualisation - planar projection into a synthetic view (BRPM Mine). As the model becomes larger it becomes more difficult to manage connectivity of the mesh. This is especially problematic when iteratively updating the model with new scans of a previously modelled region. One option is to simply plot the points. However, this creates a sparse structure that is difficult for visualisation. Another alternative is to assign a small patch (surface element or surfel) to each point and allow the patch to adapt to new measurements. Essentially, this provides the flexibility of a point-cloud but with surface characteristics of a mesh. This is demonstrated in Figure 105 below where an iterative model was generated using surfels [21]. Surfels are initialised when points occupying empty regions are added and updated us each range image is processed. Orientations are determined by estimating surface normals from the local pixel neighbourhood in the range images. A scale is assigned depending on the distance to the sensor. This allows model resolution to adapt appropriately when the camera is moved closer to a surface. Each surfel is modelled by a hexagon rendered with four triangles. In the figure, the surfels are coloured by distance from the sensor.

713


Figure 105. 3D model of BRPM data set constructed using surfels (BRPM Mine). 6.

CONCLUSIONS We propose that functional technologies exist for enabling a robot to navigate in the stope environment (30m X 3m X 1m) of South African Hard Rock mines using a dTOF beacon system for sensor localisation, and an array of sensors to map the environment. Using thermal imagery and 3D structure we have shown that a reasonable approximation of the mine can be created as a 3D model. It remains to be shown whether this is sufficiently accurate or consistent for machine intelligence algorithms to reliably generate hanging wall risk criteria. The initial work with only an IR camera to evaluate hanging wall risk was insufficient for complete analysis as rock mass protruding into the stope will also be cooled by the passing ventilation air. This indicated that 3D topographical information would be needed to assess the rock mass stability. Additionally, the lack of features in the thermal data made stitching the data together unreliable. By rigidly mounting the thermal and 3D sensors, it was possible to register the 3D and thermal so that the resulting 3D surface could be thermally textured. This in turn could be used to determine rock mass stability. We have shown that once a risk criterion is achieved, the visualisation of that risk can be achieved in a number of ways. Synthetic view and surfel representation are two likely candidates for use in the mining application.

7.

RECOMMENDATIONS The success of the fusing of thermal data and 3 dimensional structure provides a positive foundation for a mine wall stability assessment tool. The development of robust machine intelligence algorithms to determine a suitable approximation of stability from the thermal and structural information is the next step in development. The data discussed above was captured from a single point-of-view (POV) using only rotation and elevation to vary the sensor viewpoint. This produced stable measurements and consistent features that aided registration. The next step is to combine multiple POV data sets to build a more complete stope model, and then progress to moving sensors using loop closure to limit the accumulation of incremental errors. Visualisation of this information must be presented in such a way so as to be useful to a miner, and this should be explored further taking into account potential use in the unfavourable mining environment.

8.

ACKNOWLEDGEMENTS The author would like to thank colleagues at the CSIR for support of this work as well as the mines that allowed access for the collection of the data sets: GoldFields‘ Driefontein Mine, Gold Reef City mine tours and BRPM. This work is currently solely funded by CSIR and the author would like to express his gratitude for the ongoing support.

714


9.

REFERENCES [1] Sandvik. Automine. 2011(May 2), Available: http://www.miningandconstruction.sandvik.com/. [2] J. J. Green, P. Bosscha, L. Candy, K. Hlophe, S. Coetzee and S. Brink. Can a robot improve mine safety? Presented at CAD/CAM, Robotics and Factories of the Future. [3] J. Green and D. Vogt. A robot miner for low grade narrow tabular ore bodies: The potential and the challenge. Presented at 3rd Robotics and Mechatronics Symposium (ROBMECH 2009). [4] Loni Prinsloo. (2011, 6 JAnuary 2011). South african mine deaths down 24% in 2010. 2011(May 3), pp. 2. Available: http://www.miningweekly.com/article/south-african-mine-deaths-down-24-in-20102011-01-06. [5] K. Hlophe. GPS-deprived localisation for underground mines. Presented at 3rd Biennial CSIR Conference. Science Real and Relevant. Available: http://researchspace.csir.co.za/dspace/bitstream/10204/4225/1/Hlophe_2010_P.pdf. [6] K. Hlophe, G. Ferreira and J. J. Green. A posture estimation system for underground mine vehicles. Presented at CAD/CAM, Robotics and Factories of the Future 2010. [7] FLIR, "FLIR commercial vision systems, 'avoiding accidents with mining vehicles', Application Stories. 2008," vol. 2010, pp. 2, 2008. [8] V. A. Kononov. (2000, September 2000). Pre-feasibility investigation of infrared thermography for the identification of loose hanging wall and impending falls of ground. Safety in Mines Research Advisory Committee. South Africa. Available: http://researchspace.csir.co.za/dspace/handle/10204/1811. [9] D. Vogt, V. Z. Brink, S. Donovan, G. Ferreira, J. Haarhoff, G. Harper, R. Stewart and M. Van Schoor. Mining research for enhanced competitiveness. Presented at Science Real and Relevant: 2nd CSIR Biennial Conference. [10] D. Vogt, V. Z. Brink and S. Schutte. New technology for real-time in-stope safety management. Presented at Hard Rock Safe Safety Conference 2009. Available: http://researchspace.csir.co.za/dspace/handle/10204/3680. [11] D. Vogt, V. Z. Brink, S. Brink, M. Price and B. Kagezi. New technology for improving entry examination, thereby managing the rockfall risk in south african gold and platinum mines. Presented at CSIR 3rd Biennial Conference 2010. Science Real and Relevant. Available: http://researchspace.csir.co.za/dspace/bitstream/10204/4255/1/Vogt_2010.pdf. [12] H. Bay, T. Tuytelaars and L. Van Gool. SURF: Speeded-up robust features. Presented at 9th European Conference on Computer Vision. Available: http://www.mendeley.com/research/surf-speededuprobust-features/. [13] C. Baker, A. Morris, D. Ferguson, S. Thayer, W. Whittaker, Z. Omohundro, C. Reverte, W. Whittaker, D. Haehnel and S. Thrun. A campaign in autonomous mine mapping. Presented at IEEE International Conference on Robotics and Automation. [14] D. Huber and N. Vandapel. Automatic 3D underground mine mapping. Presented at Field and Service Robotics 2003. [15] A. Nüchter, H. Surmann, K. Lingemann, J. Hertzberg and S. Thrun. 6D SLAM with an application in autonomous mine mapping. Presented at IEEE International Conference on Robotics and Automation. [16] D. Droeschel, D. Holz and S. Behnke. Multi-frequency phase unwrapping for time-of-flight cameras. Presented at Intelligent Robots and Systems (IROS), 2010 IEEE/RSJ International Conference on. Available: http://www.ais.uni-bonn.de/papers/IROS-2010-Droeschel.pdf. [17] J. Bloomenthal. (1994, An implicit surface polygonizer. Graphics Gems IV 1pp. 324–349. [18] G. Taubin. Curve and surface smoothing without shrinkage. Presented at Iccv. [19] MIT, " Visual Odometry For GPS-Denied Flight And Mapping Using A Kinect," vol. 2011, March 2011, 2011. [20] P. Henry, M. Krainin, E. Herbst, X. Ren and D. Fox. RGB-D mapping: Using depth cameras for dense 3D modeling of indoor environments. Presented at The 12th International Symposium on Experimental Robotics (ISER). [21] H. Pfister, M. Zwicker, J. Van Baar and M. Gross. Surfels: Surface elements as rendering primitives. Presented at Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques 2000.

715


TOWARDS OPTIMUM DESIGN OF MAGNETIC ADHESION WALL CLIMBING WHEELED ROBOTS Salman, H., Sattar, T.P., Salinas, E. London South Bank University, Faculty of Engineering, Science & the Built Environment, London, UK e-mail: [email protected] ABSTRACT Climbing and walking robots perform tasks that are too difficult, dangerous or time consuming for the human worker. The main design consideration in the climbing robot is its method of adhesion. The aim of this paper is to lay down the foundation for developing a design framework for magnetically adhering wheeled robots having magnets attached to the base of the robot. The different design parameters affecting the magnetic adhesion include the geometry of the flux concentration plate, effect of the variation of the air gap on adhesion and climbing performance, different materials for magnetic flux concentration and different magnetic arrangements. These different parameters affecting adhesion are simulated and optimized using Magnetostatic analysis in ANSYS. Keywords: Wall climbing robot, magnetic adhesion optimization, robot design optimization 1.

INTRODUCTION Climbing and walking robots perform tasks that are too difficult, dangerous or time consuming for the human worker [1, 2] . The main design consideration in the climbing robot is its method of adhesion. Methods of adhesion include Magnetic, Pneumatic, Gecko and Vortex [3]. This paper focuses on magnetic adhesion for Wall climbing robots. There are three different magnet deployment configurations to achieve adhesion in wheeled wall climbing robots. . They achieve adhesion by using magnetic wheels [4-10], magnetic tracks [11-13]and magnets attached to the base of the body [1]. A few researchers have tried to optimize adhesion based on magnetic wheel [4, 13]and magnetic track adhesion [11], but these have been limited to specific robots. Also, this work has focussed on only one or two variables affecting the optimization. To our knowledge, no work has been done to address the entire possible optimization variables and especially the optimization of robot design with magnets attached to the body of the robot. This paper addresses all the possible optimization variables for magnetic adhesion when the magnets are attached to the body of the robot. The comparison of different optimization parameters gives a useful insight into stability of the robot and will help in deciding how to select different design parameters.

2.

BACKGROUND The aim of this paper is to lay down the foundation for developing a design framework for magnetically adhering wheeled robots. The design cycle comprises of a two step process. The steps are to optimize design and to optimize adhesion. In design optimization, a static and dynamic force analysis is carried out. This analysis serves to optimize design parameters by considering the adhesion force requirement, dimensions of the robot, material properties to allow selection of materials, the robot configuration and its center of gravity. The parameters obtained from this step provide a gateway to the second step of the design cycle, i.e. adhesion optimization. The parameters for adhesion optimization depend on methods to strengthen the magnetic field using concentration plates, the geometry of the concentration plates, material of the concentration plate, effect of the variation of the air gap on adhesion and climbing performance, different magnetic arrangements and the effect of wall thickness variations. Their effects are studied using simulations based on Finite Element methods using Magnetostatic analysis in ANSYS. Some of the results are validated by experimentation.

716


3.

DESIGN OPTIMIZATION Static and dynamic force analysis is necessary to analyze and design the optimum design parameters. In this section, Static and dynamic force analysis for the wall climbing robot is carried out.

Static Analysis The stability of the wall climbing robot depends mainly on sliding failure, fall over failure or turn over failure as shown in Figure 106. Static analysis helps to find the design parameters to address these stability concerns.

Figure 106: Stability factors a) turn-over failure; b) sliding failure, c) roll over failure Sliding avoidance The ideal wall climbing robot should do climbing surface transitions and climb on surfaces with different slopes. To understand the forces acting on a robot, consider the forces acting on a robot resting on an inclined plane as shown inFigure 107. The slope of the inclined plane is ―ζ‖.

W= weight of the robot

 = angle of inclination

Fm = magnetic adhesion force

 = coefficient of friction of wheels d = distance of centre of gravity from the climbing surface L = distance between front and rear wheels

717


μN N R/4 θ Wsinθ

L

d

Wcosθ

A

W

Wsinθ

θ W Figure 107: free body diagram of robot moving on an inclined plane

 Fy  W cos  Fm  N  0 N  W cos  Fm  Fx  W sin  N or N

W sin



W cos  Fm  Fm 

W sin 



W sin 



 W cos

For the robot to avoid slipping

Fm 

W sin



 W cos

For the special case of wall climbing robot moving on a vertical surface

  90 Fm 

W



(1)

718


In order to avoid sliding/slipping of the robot, the magnetic adhesion force should be greater than W



.

The stability of the robot can be increased by either increasing the coefficient of friction of wheel tyre or decreasing the robot weight.

Turn over avoidance From Figure 107, Taking moment about point A,

M  W  d 

R L0 2

2W  d L Fm  R R

Fm  

W d 2L

To avoid Turn over, the adhesion force should satisfy equation 2.

Fm 

W d 2L

(2)

For a given adhesion force, equation 2 can be satisfied by minimising the ratio

d . L

This means that the

centre of gravity should be as close to the surface as possible and the distance between the wheels should be large. Equation (2) shows that in order to avoid turnover, the robot centre of gravity should be kept as low as possible.

The stability criteria to avoid sliding and turn over:

W W  d  Fm  max  ,    2L 

(3)

Roll over avoidance For simplicity, we assume that the shear force on the robot always acts perpendicular to the wheel. Consider Figure 3(a), the roll over moment

Mr

will then be

M r  (W  k )  2( Fs  L1 )  0

(W  k )  2( Fs  L1 ) 2 Fs 

(W  k ) L1

719


Figure 108: Roll over forces when a robot is at different orientations on the wall k=distance between center of gravity and point A

Fs =shearing force on each wheel L1 =moment arm, distance between wheel 4 and 1 If order to avoid roll over,

Fs 

(W  k ) 2 L1

(4)

The roll over force varies with the angle of the robot on the wall (Figure 108 (b) and (c)). This is due to the variation of the moment arm with different orientations of the robot. The robot length (distance between the wheels) and the location of its center of gravity are important design considerations. These parameters play an important role in determining the motion and stability of the wall climbing robot. Dynamic Analysis The design for the motor torque requirement will be based on torque analysis when the robot is moving upward. This is due to the fact that the torque required for a climbing robot to move will be maximum when the robot is moving upward. Consider a robot on a wall moving upward as shown in Figure 109.

r a

F

.

r

MO W Figure 109: Moment force diagram for robot moving upward 720


a = distance between center of gravity of robot and wheel center M = total torque required for the robot to move upward

M w = torque required by each wheel O = center of the wheel r = radius of the wheel

Fr

= rolling force required

The moment about point O is

M  (W  a)  Fr  r  0 M  (W  a)  ( Fr  r ) Fr    Fm

M  (W  a)  (   Fm  r )

(5)

If there are ―w‖ numbers of driver wheels, torque required by each wheel

Mw  4.

M w will be

M w

MAGNETIC ADHESION OPTIMIZATION The magnetic adhesion properties can be studied by using finite element software. We use ANSYS Magnetostatic analysis. The magnets were first modelled in ANSYS design modeller. The magnet was then imported into ANSYS Magnetostatic analysis. Like all the Finite element methods procedure, the meshing of the magnets was carried out and the boundary conditions were defined and simulated. The results of the first set of simulations were verified by experimental results to validate the simulation setup and boundary conditions.

(a)

721


(b) Figure 110: Magnetostatic analysis showing magnetic flux lines, (a): modelled block, (b): magnetic flux lines inside modelled figure ANSYS Magnetostatic Analysis The ANSYS Magnetostatic Analysis enables us to analyze different magnetic properties of the designed system. It includes flux density, field intensity, force summation, torque, energy and magnetic flux. In Figure 110(a), schematic of magnetic circuit is shown. The magnetic circuit have flux concentrator, magnets and climbing surface (surface). Flux concentrator has limbs to direct he magnetic lines. It is desirable to design magnetic circuit with and air gap between climbing surface (usually steel) and magnets. In Figure 110(b), the north poles of all three magnets are facing the flux concentrator. The magnetic lines of force travel from North pole into the flux concentrator. To complete the magnetic circuit, these flux lines enter into the South pole after passing through the wall as shown by black arrow. Validation of ANSYS Magnetostatic Analysis Two blocks of permanent magnets, one with an array of 3×3 magnets and other with array of 3×2 magnets were simulated as shown in Figure 111. The arrays were attached to a mild steel plate. This steel plate served to concentrate the magnetic flux lines and thus increases the magnetic adhesion force. N42magnets were used each having dimensions 50×50×25mm. The mild steel plate had a thickness of 10 mm and dimension of 260×200mm and 260×160mm for 3×3 array and 3×2 array respectively.

Figure 111 Blocks of magnetic array 3×2 and 3×3 with a steel plate serving as magnetic concentration plate

722

Attraction force N


5000 4500 4000 3500 3000 2500 2000 1500 1000 500 0

Magnetic Force VS Air Gap (Validation of Simulations Results) 3x3 Simulations 3X3 Experimental 3x2 Simulations 3X2 Experimental

0

10

20 Air Gap (mm)

30

40

Figure 112: Magnetic force vs. air gap (Validation of simulation results), Experimental results taken from the development phase of CROCELLS robot [1].

In Figure 7,experimental results from our previous research work [1] were use to validate the simulations results. The maximum error was found to be 8% at very high adhesion forces. This is due to the experimental apparatus capacity at high loading conditions. The overall result shows a good agreement with the experimental results. Magnetic Arrangements Adhesion force due to different magnetic arrangements is shown in Figure 113. These magnetic arrangements include use of a concentration plate, air gap variation from the wall surface and distance between the magnets. The adhesion force is maximum when the magnets are 5mm apart. As the distance between magnets is increased, the adhesion force starts decreasing. Thus the closer the magnets on the robot base (concentrating plate), the higher will be the adhesion force.

723


Comparison of attraction force at different magnetic arrangements

5mm apart with back plate

2000

5mm apart without back plate

1800

Attraction Force N

1600 1400


1200


1000


800


600 400 200 0 0

5

10

15

20 Air Gap

25

30

35

40

Figure 113: Different magnet arrangement to achieve optimum adhesion

Effect of Wall thicknes on adhesion force (wall 5 mm apart) 2220

Attraction force N

2200 2180 2160 2140 2120 2100 2080 2060 0

2

4 6 8 Thickness of wall mm

10

12

Figure 114: Effect of wall thickness on magnetic adhesion Use of the concentrating back plate shows a significant increase in magnetic adhesion as the value jumps from 1000N to 1835N, thus almost doubling the magnetic adhesion. The air gap refers to the distance between the face of the magnet and the wall. As this gap is increased, the magnetic adhesion decreases. This air gap is necessary to avoid obstacles in some cases and to avoid friction in all the cases. The friction at the wheel is desirable, but the friction at adhesion surfaces is not desirable.

724


Effect of wall thickness For a specifc magnet, the wall thickness determines the ahdesion force. Simulations were carried out with a N52 magnet. The wall material used was structural steel. The adhesion force is minimum at a wall thickness of 0.1mm. When the thickenss of the wall is increased from 0.1mm to 1mm, the adhesion force increases gradually. At 1 mm, the magnetic flux is almost maximum. Any further increase in wall thickness does not have considerable effect on adhesion force as shown in Figure 114.

Effect of different magnet type on adhesion at 10 mm distance fro mthe wall

Attraction force N

1200 1000 800 600 400 200 0 n50

n42

n35

n28uh

Magnet type

Figure 115: Effect of air gap on different magnets having different strength

Attraction froce N

Effect of using differnt materials as concentration plate 888

890

892

899

901

903

906 908.53 910

911 780 599

Figure 116 Effect of using different material for concentration plate When using an air gap in the magnetic circuit, the stronger the magnet, the more adhesion force it will produce as shown Figure 115. N28UH is the weakest material in the analysis and thus producing the lowest adhesion force. N50, being the strongest magnetic force is more desirable when circuit have air gaps.

725


Design of the concentration plates One of the major design considerations is the design of the concentration plate. The geometry and construction material of the concentration plate plays an important role in optimizing the adhesion force. Figure 116 shows the variation of adhesion force, when different materials are used for the concentration plate. The adhesion force is minimum when mu metal is used and maximum when structural steel is used. The use of a concentration plate also serves to provide strength to the chassis of the robot. So the use of structural steel is desirable though the design will perform a trade-off to reduce the weight of the climbing robot. Effect of different shape of the concentration plate The concentration plate shape affects the magnetic flux leakage as shown in Figure 117. When the limb of the concentration plate is skewed inward, most of the magnetic flux leaks into the south pole without passing through the wall.

Climbing surface

Climbing surface

(a)surface Climbing

(b) surface Climbing

Climbing surface

Climbing surface

Climbing surface

Climbing surface (d)

(c)

Figure 117: Magnetic flux leakage due to different concentration plate shapes, (a): limbs inward, (b): straight limbs, (c): No limbs, (d): limbs outwards.

When the limbs are straightened the magnetic flux leakage is improved but is optimum when the limbs of the concentration plate are skewed outward. When there is no limb the magnetic flux also leaks considerably more as compared to the straight or outward limb. Figure 118 shows that the adhesion force is maximum when the concentration plate has limb skewed outwards. The adhesion force by this skewed outward shape is 50% more than the adhesion force produced by either without limb or inward limb.

726


Adhesion force N

Effect of concentration plate shape 100 90 80 70 60 50 40 30 20 10 0 without limb

inward limb

stright limb

Outward limb

Concentration plate limb shape

Figure 118: Effect of concentration plate shape Effect of length of the concentration plate The magnetic adhesion force increases with concentration plate size. This is due to reduction in flux leakage when length of concentration plate is increased. Figure 119 shows that the magnetic adhesion is proportional to the concentration plate length.

Effect of length of the concentration palte on adhesion force

Adhesion force N

100 80 60 40 20 0 80

90

110

130

150

length of the concentration palte mm

Figure 119 Effect of length of the concentration plate on magnetic adhesion 5.

CONCLUSION Different design parameters responsible for the stability of wall climbing robots were analyzed. These parameters help in laying down geometric properties, material properties and configuration of the robot. Finite Element methods were used to study the optimization of the magnetic adhesion. The study reveals the effectiveness of this approach in predicting the magnetic force for optimization purposes. The factors affecting magnetic adhesion are the type of magnet, air gap, configuration of magnetic array, concentration

727


plate material, shape and length, the effect of wall thickness, Optimization of these factors were carried out using FEM analysis. Some of the results of the simulations were applied and validated by comparing it with the results of our robot CORCELLS[1]. These results provide a foundation to construct design rules to develop a wheeled robot using magnetic adhesion with magnets attached to the base of the body. Prototypes will be built based on these optimization results. Also optimization of robot with magnetic wheel will be carried out and will be compared. 6.

REFERENCES [1] Shang, J., et al., Development of a climbing robot for inspection of long weld lines. Industrial Robot: An International Journal, 2008. 35(3): p. 217-223. [2] Sattar, T.P., Wall Climbing Crawlers for Nondestructive Testing,Topics On Nondestructive Evaluation (TONE),. Automation, Miniature Robotics and Sensors for Nondestructive Evaluation and Testing. Vol. 4. 2000. [3] Chu, B., et al., A survey of climbing robots: Locomotion and adhesion. International Journal of Precision Engineering and Manufacturing, 2010. 11(4): p. 633-647. [4] Yukawa, T., et al. Design of Magnetic Wheels in Pipe Inspection Robot. in Systems, Man and Cybernetics, 2006. SMC '06. IEEE International Conference on. 2006. [5] Jin, Y., J. Chen, and Z. Li, A Magnetic Wheel Structure for an Omni-directional Microrobot to Limit Slip Effect. International Journal of Advanced Robotic Systems. 6. [6] Kawaguchi, Y., et al. Internal pipe inspection robot. in Robotics and Automation, 1995. Proceedings., 1995 IEEE International Conference on. 1995. [7] Slocum, A., et al. Magnetically preloaded wheels. 2004. [8] Yukawa, T., H. Okano, and S. Komatsubara. Mechanisms for the movement of piping inspection robot with magnetic elements. 2005. [9] Fischer, W., F. Tâche, and R. Siegwart, Magnetic wall climbing robot for thin surfaces with specific obstacles. 2007. [10] Tache, F., et al. Compact magnetic wheeled robot with high mobility for inspecting complex shaped pipe structures. in Intelligent Robots and Systems, 2007. IROS 2007. IEEE/RSJ International Conference on. 2007. [11] Shen, W., J. Gu, and Y. Shen, Permanent magnetic system design for the wall-climbing robot. Applied Bionics and Biomechanics, 2006. 3(3): p. 151-159. [12] Rochat, F., et al., TRIPILLAR: Miniature magnetic caterpillar climbing robot with plane transition ability. CLAWAR09, 2009. [13] Yuanming, Z., et al. Design and optimization of magnetic wheel for wall and ceiling climbing robot. in Mechatronics and Automation (ICMA), 2010 International Conference on Mechatronics and Automation. 2010.

728


REVERSE ENGINEERING OF CONTROL PROGRAMS INTO RECURRENT NEURAL NETWORKS FOR RECONFIGURABLE MANUFACTURING SYSTEMS Vimal Nandhan R.K. 1 and Ramesh Babu N. 2 1

PhD Scholar, Manufacturing Engineering Section, Indian Institute of Technology Madras Chennai – 600 036, India e-mail: [email protected] 2 Professor, Manufacturing Engineering Section, Indian Institute of Technology Madras Chennai – 600 036, India e-mail: [email protected]

ABSTRACT Ladder Logic (LD) programs are popular for discrete control of events in manufacturing systems. But, their modification, to realize control for reconfigurable manufacturing systems, is quite difficult due to their structure being incomprehensible. Recent use of artificial intelligence concepts for modelling and control of discrete, asynchronous manufacturing systems, have enabled researchers to develop Recurrent Neural Network (RNN) based sequential controllers. Reconfiguration of RNN based control is achievable only when the control specifications are available in a detailed manner so as to consider them for training of RNN. As it is quite tedious and time consuming, the present work attempts to transform the fully developed LD programs into state transition table and is then used for training of RNN. The development of state transition table from LD programs simultaneously considers the configuration of physical system along with the change in the states of the system. The feasibility of the proposed approach is demonstrated with a case study. Keywords: Sequential control, Ladder logic programs, Recurrent Neural Network, Manufacturing control system 1.

INTRODUCTION An essential aspect of modern manufacturing systems is to reconfigure them in order to fulfill changing requirements of manufacturing. Reconfiguration involves the change of both hardware and software components of manufacturing system [1]. Among them, control software reconfiguration is an important task. Ladder diagrams (LD) are employed as control programs for sequential control of manufacturing systems. Reconfiguration of control programs require their modification and is thus possible only with proper understanding of LD program. Though LD programs mimic relay wiring diagrams, the switching logic represented in graphical form is quite incomprehensible [2]. Further, the lack of documentation of LD programs makes it more difficult to understand even by original program developers [3, 4]. Thus, the interpretation of logic in LD programs is possible only by careful analysis of every single path of program. Another method of developing control programs involves the conversion of a set of control specifications into an intermediate model such as Petri net, automata model and is then transformed into control programs with the help of certain heuristic methods [5-7]. Control program development with these approaches is possible only with personnel experienced in these modeling techniques. Therefore, such approaches become quite ineffective for quick reconfiguration of control programs. Recent use of artificial intelligence concepts such as Artificial Neural Networks (ANN) for modeling and control of asynchronous manufacturing systems, eliminates the need for an expert to develop control programs. At the same time, they also capture the non-linear nature of temporal logic among inputs and outputs easily. It is possible to incorporate such logic in ANN with the help of suitable data for training the ANN. Among various types of ANNs, Recurrent Neural Networks (RNN) belong to a special class and have feedback loops from hidden or output layer neurons to input layer that consider the activation from previous propagation and present input in order to generate the output. This sort of closed-loop configuration of RNNs makes them more suitable for training of sequential logic among input-output devices present in sequential control systems [8, 9]. An attempt was made to develop RNN from control specifications for discrete control of manufacturing system [10]. In this method, the detailed control specifications were converted into a state transition table,

729


which in turn was used to train the network. The state transition table essentially shows the logic among the input and output devices, which is quite important to develop RNN based control. An attempt to reduce the efforts in developing state transition table includes the conversion of control specifications into an initial state table, which eventually got transformed into a state transition table with the insertion of transition states [11]. Thus, the generation of state transition table for any existing manufacturing system is a challenging task since it requires detailed control specifications, which are generally formulated by expert programmers and are not normally available with general description of control problem. In view of the difficulties associated with this approach, the present work proposes an approach that aims to generate state transition table from LD program built to control any physical system. As LD programs perform logic control for a particular set of detailed specifications, the simulation of LD program can provide the status of outputs for any particular set of inputs. Along with the physical system configuration, it is possible to identify the sequence of inputs to be activated in order to observe changes in the status of outputs. By means of suitable simulation of inputs/outputs, the state transition table can be generated, which in turn can be used for training of RNN. Such training of RNN with the states in state transition table involves constant updation of its weights and biases using Back Propagation (BP) algorithm. Since, BP algorithm takes considerable time to converge owing to constant learning rate [12], an adaptive learning rate and momentum factor were employed to form RNN for sequential control of manufacturing systems. Trained RNN is then tested for its ability to control the sequential events in any chosen manufacturing system. 2.

PROPOSED APPROACH FOR THE DEVELOPMENT OF RNN BASED CONTROL FROM EXISTING LADDER LOGIC PROGRAMS Figure 1 shows the overall approach proposed to develop RNN controller from any chosen LD program. To achieve the sequential control of manufacturing systems with RNN, two important tasks need to be performed. (1) Development of binary relations among input-output devices i.e. state transition table for any chosen control task and (2) Selection of suitable RNN structure and training algorithm. The proposed approach considers fully tested LD program along with physical configuration of system in order to generate the state transition table, showing the sequential logic among various inputs and outputs in the system. First, it considers the home position of system and identifies the status of inputs. This particular status of inputs will be utilized to identify the status of outputs by simulating the LD program and simultaneously observing the status of outputs in physical system. The status of inputs and outputs will become the initial status of the system and is also the first data set in the state transition table. As these changes in outputs result in a change in the status of certain inputs, these changes in the status of inputs are then considered to simulate the LD program and to identify changes in the status of various outputs. This new set of inputs and outputs will form the next sequence in the state transition table. This procedure is continued until the simulation of LD program results in the status of inputs and outputs same as the initial state in state transition table. In order to explain the process of generation of state transition table from a fully developed LD program, a typical manufacturing system consisting of different units with 5 inputs and 2 outputs is chosen. In this system, the five inputs are labeled as A, B, C, D and E and two outputs are labeled with letters X and Y. The change in status of these inputs from ‗0‘ to ‗1‘ is denoted by the symbols A, B, C, D and E and the change in status of inputs from ‗1‘ to ‗0‘ is denoted by the symbols: !A, !B, !C, !D and !E. Similarly, the change in the status of output from ‗0‘ to ‗1‘ is denoted by the symbols: X, Y and Z and from ‗1‘ to‘0‘ is denoted by the symbols !X, !Y and !Z. Typically, the logic in the rungs of LD program relates the inputs to the outputs of the system. By knowing the status of inputs, the output can be determined. Each rung can be represented as a boolean equation and the set of rungs represents the set of boolean equations. Boolean expressions denoted by (1) show logic in the entire set of rungs in LD programs. IF [(A OR !B) AND C] THEN ‗X‘ ELSE ‗!X‘; 1

IF [D AND E] THEN ‗Y‘ ELSE ‗!Y‘; IF [(A AND !C) OR B] THEN ‗Z‘ ELSE ‗!Z‘;

The physical system is observed for changes in inputs for every change in the output. This can be represented as (X = 1) => A, !B

730


Physical System

LD program

Observe the status of inputs at the home position in physical system

Set the initial status of all inputs as per the home position in physical system

Observe the change in the status of inputs

Analyse LD program to observe change in the status of outputs

IF NO

(Status of inputs and outputs at present condition) = (Status of inputs and outputs at home condition) YES

State transition table

Generate training data for RNN Select the structure of the RNN and training algorithm Train the NN with the training data RNN for control of manufacturing system Figure 1 Overall approach to transform LD program into RNN for control of manufacturing systems

731


The above expression means, when the status of X is ‗1‘, sensor A will go to the status of ‗1‘ and B will reach the status of ‗0‘. The changes in the status of inputs for each of the outputs can be listed as in expressions denoted by (2) i.e. (X = 1) => A, !B; (Y = 1) => A, B, D, !E;

2

(Z = 1) => A, C, !D, E; As mentioned earlier, initial status of inputs is chosen by observing the status of inputs in physical system at its home position or start condition. This particular status of inputs will be used to identify the status of outputs in the system from LD programs by analyzing the boolean expressions shown in (1). The status of various inputs and outputs are then considered as the initial state of state transition table. The second sequence is generated by updating the output status using the expressions denoted by (2) and identifying the status of various inputs. Then, these inputs are updated in the set of expressions shown in (1) to obtain the status of outputs. This process is repeated until the status of inputs and outputs becomes the same as the status of inputs and outputs in the initial state, shown in state transition table. The next step in the development of Recurrent Neural Network based control is the selection of RNN structure, which influences the robustness of RNN for control of manufacturing system. The choice of RNN depends on its capability to learn the input-output relation and the sequence of input-output data. Hence, RNN with an internal memory, to process arbitrary sequences of inputs and an internal state, is chosen. Elman‘s RNNs are capable of learning the sequence of data along with input-output mapping and hence it is preferred for sequential control of discrete manufacturing systems. Elman‘s RNN has three layers, i.e. input, hidden and output layers, where the feedback is given from the hidden layer to the input layer through units called ‗context‘ units, with a fixed weight of one. During each time step, the input is propagated in a feed-forward manner, whereas the previous state is stored in the context units i.e. feedback from hidden layer to train the network about the sequence information. A back propagation training algorithm with momentum and adaptive learning rate is used in this approach for training the RNN. The goal of this training algorithm is to reduce the error between the network output and the desired output by constantly updating the weights and biases in the network connections, so that the network will learn the data and control the system as desired. In this type of network, the number of input and output nodes of RNN will be determined by the number of input and output elements in the system. If the number of inputs and outputs in a system are N1 and N2, then the number of nodes in these layers will be N1 and N2 respectively. Thus, the input data to RNN-based sequential controller is nothing but the status of a set of N1 inputs, while the output data is nothing but the status of a set of N2 outputs. Each set of input –output data is known as a sequence. The number of such set of input-output data i.e. sequences varies for every system. The number of hidden nodes in RNN was chosen by trial and error method with the objective of keeping the time for training to minimum. The maximum number of epochs is chosen arbitrarily to ensure that the training of RNN will attain the desired outputs. Then, testing of the RNN is performed for entire set of input sequences, partial input sequences and individual input sets. RNN thus developed can be adopted for control of the system after successful testing of the network. In case of control of reconfigured manufacturing system, the state transition table is modified and the RNN is trained with the modified training data to achieve its control. Modification of the state transition table is performed in accordance with the new requirements by addition, deletion or in some cases inclusion of status of input and output elements of the system.

3.

A CASE STUDY To demonstrate the feasibility of the proposed approach, a system with a clamp, stamp and push-eject events is considered. Figure 2 shows the system consisting of a table over which the part is placed against the stopper guides. It also consists of four pneumatic cylinders - cylinder A, cylinder B, cylinder C and Cylinder D that control the sequence of events. For illustrating reconfigurability of the proposed approach, another event was introduced and is shown in dotted line. This event is realized by a cylinder D. With the action of cylinder A and cylinder B, clamping and stamping events will take place. Cylinder C is used for

732


pushing the stamped part on to the conveyor. The retracted position of the piston in cylinder A and cylinder B are indicated by the activation of limit switches a0 and b0 respectively and the extended position of cylinder A and cylinder B are indicated by the activation of limit switches a1 and b1 respectively. Similarly, the extended and retracted position of cylinder C is indicated by the activation of limit switches c0 and c1. Apart from the above elements, a push button ‗SRT‘ is used to start the process and a sensor ‗ACT‘ activates the actuator to move the part onto the conveyor after the stamping operation. In the control panel, start button ‗SRT‘, indicator lamps ‗ACT‘ and ‗RJT‘ are shown. The specifications for controlling of the system are as follows (1) Start the system by pushing ‗SRT‘ push button (2) Extend the piston rod of cylinder A until it reaches the limit switch ‗a1‘ to hold the part firmly against the stopper. The movement of cylinder A is accomplished by actuating the solenoid Af (A_forward) (3) Extend the piston rod of stamping cylinder B until it reaches the limit switch ‗b1‘, where it performs the stamping operation on the part. This is achieved by actuating the solenoid B f (B_forward) (4) Retract the piston rod until it reaches the limit switch ‗b0‘ which is actuated by B r (B_reverse) solenoid (5) Retract the piston rod to release the part by actuating Ar (A_reverse) solenoid until the piston rod reaches the limit switch ‗a0‘ (5) When the sensor ‗ACT‘ gets activated to indicate the completion of stamping, the part is moved onto the conveyor with the piston rod of ejector cylinder C. To accomplish this task, the piston in cylinder C is moved forward until it reaches ‗c1‘ by actuating solenoid C f (C_forward) (6) Retract the piston rod to its initial position ‗c0‘ which is actuated by solenoid Cr (C_reverse). Each specification is considered in order to develop LD program for achieving the desired control. Figure 3 shows the LD program for accomplishing the sequential control of events in the manufacturing system.

Figure 2 Stamping System

733


In a typical manufacturing system, an additional requirement may arise such as removal of defective parts from the conveyor. In order to accomplish this particular requirement, the existing system will be modified with the addition of certain elements. In Figure 2, this particular change in the system layout is shown within dotted lines. In this, cylinder D is added to the manufacturing system for the purpose of ejecting the defective part into a bin. Further, a sensor ‗RJT‘ is added in the system in order to detect the defective parts passing through the conveyor. The defective part moving on the conveyor is pushed into the bin with the extension of piston rod in cylinder D, actuated by solenoid Df (D_forward). Then the piston in cylinder D is retracted by actuating solenoid D r (D_reverse) making way for other parts to move onto the conveyor.

Figure 3 Ladder Logic Program for controlling the stamping system Such changes often occur in manufacturing systems and thus necessitate the modification of LD program. Due to the difficulty in understanding the flow of control in LD program and the extent of effort needed to

734


make such changes in LD program by programmers, the LD program can be converted into RNN with which it is easier to incorporate the changes made in the manufacturing system layout. The ease of modifying RNN for affecting changes in control requirements is explained with a case study. LD program and physical configuration of the system, shown in Figure 3 and Figure 2 are considered for developing the state transition table, which is then transformed into RNN for the purpose of controlling the system. Initially, the pistons in the cylinders A, B and C are positioned in their retracted position. Hence, the status of a0, b0 and c0 is ‗1‘. In table 1, the column 0 represents the condition of various sensors and actuators when the machine is powered up. To develop the state transition table, the first step is to observe the initial status of inputs at the home condition of physical system. 



Column 1 shows the initial condition of the system wherein the machine is started by pressing the push button ‗SRT‘. When the status of inputs a0, b0, c0 and SRT is ‗1‘, the evaluation of LD program results in energizing Actuator Af ( A_forward) and is represented in column 1 opposite to label Af ( A_forward). Then, the physical system is observed in order to visualize the change in the status of inputs. A list of observation of all output actuators in the system is given below.

Af ( A_forward) => a0=0; a1=1 Ar (A_reverse) => a0=1; a1=0 Bf (B_forward) => b0=0; b1=1 Br (B_reverse) => b0=1; b1=0 Cf (C_forward) => c0=0; c1=1 Cr (C_reverse) => c0=1; c1=0 From these observations, the change in the status of inputs for any change in the output can be known. When Af ( A_forward) is energized, the status of a0 becomes ‗0‘. Hence, in column 2, it is indicated that a0 is 0. The LD program is analyzed with this set of status of inputs, and the status outputs are found to be unchanged and are the same as in column 1. Then, the observations made from the physical system are again analyzed to obtain the changes in the status of input variables, where the status of a1 changes from 0 to 1 and is denoted in column 3. Again, the LD program is analyzed with this set of inputs to identify the outputs. This procedure is continued until the initial status, shown in column 1, is repeated to complete the development of state transition table. Inputs

0

1

2

3

4

5

6

7

8

9

10

11

12

13

SRT

0

1

1

1

1

1

1

1

1

1

1

1

1

1

a0

1

1

0

0

0

0

0

0

0

1

1

1

1

1

a1

0

0

0

1

1

1

1

1

0

0

0

0

0

0

b0

1

1

1

1

0

0

0

1

1

1

1

1

1

1

b1

0

0

0

0

0

1

0

0

0

0

0

0

0

0

c0

1

1

1

1

1

1

1

1

1

1

0

0

0

1

c1

0

0

0

0

0

0

0

0

0

0

0

1

0

0

735


ACT

0

0

0

0

0

0

0

0

0

1

1

0

0

0

A_forward

0

1

1

0

0

0

0

0

0

0

0

0

0

1

B_forward

0

0

0

1

1

0

0

0

0

0

0

0

0

0

B_reverse

0

0

0

0

0

1

1

0

0

0

0

0

0

0

A_reverse

0

0

0

0

0

0

0

1

1

0

0

0

0

0

C_forward

0

0

0

0

0

0

0

0

0

1

1

0

0

0

C_reverse

0

0

0

0

0

0

0

0

0

0

0

1

1

0

Outputs

Table 1 State transition table By training the RNN with the state transition table, RNN based control for reconfigurable system can be realized. A MATLAB program is developed for the purpose of reading the state transition file and training the RNN. Training of RNN is accomplished by choosing the number of input nodes as 8, which is equal to the number of input control elements and the number of output nodes as 6, which is equal to the number of output control elements and the number of sequences as 14. Then, the state transition table, in .txt format is read and the input and target values for the RNN are fixed to be the same as the one for inputs and outputs in the state transition table. The training parameters of the RNN, such as maximum number of epochs and learning performance index i.e. sum squared error of the network are arbitrarily chosen to be 2000000 and 10-4 respectively. The number of nodes in the hidden layer is chosen to be close to the sum of input (N1) and output (N2) elements i.e. (N1+N2) nodes. Then, it is varied to minimize the time taken for training. Table 3 shows the time taken and the number of epochs for different number of neurons in the hidden layer for this system configuration. The optimum number of neurons in the hidden layer for this initial system configuration is found to be 16. Then, the RNN network is tested in various ways as mentioned below (1) The entire sequences of inputs i.e. the 14 sequences of inputs are provided to the network to identify the sequence of outputs. These outputs are compared with the output sequences in the state transition table to ensure for matching. (2) Partial sequence of inputs, i.e. from sequence 1 to 7 or from sequence 5 to 12 is provided to the RNN in order to match the sequence of outputs with the outputs in the state transition table. (3) Any input set out of the 14 sequence in the state transition table is provided to the RNN to find out the output set and they are checked for its matching with the outputs in the state transition table. Thus, the consistency of the RNN for generating the output sequence for any given set of input sequences is checked before its implementation for the control of manufacturing system. When the stamped part is of acceptable quality, the part moves on the conveyor and reaches the next cell for further operation or to the packaging section. To remove the defective part moving on the conveyor, Cylinder D pushes the rejected part into the bin. Inorder to accommodate the changes in the system configuration, the developed state transition table will be modified. The modifications to be made in the system include  

Add RJT push button and add d0 and d1 limit switches to the inputs in the state transition table. Add Df (D_forward) and Dr (D_reverse) solenoids to outputs in the state transition table.

736


In table 2, the columns represented with bold letter in the state transition table are the additions that are made to the earlier state transition table. Column 13 and 14 indicate the extension of cylinder D by the actuation of solenoid Df (D_forward), where the sensor ‗RJT‘ is activated indicating the presence of rejected part. Column 15 and 16 indicate the retraction of cylinder D by the actuation of solenoid D r (D_reverse). With the addition of these control elements and indicating their status in the state transition table, it is possible to reconfigure the RNN easily. By training the RNN with the modified state transition table, RNN based control for reconfigurable system can be realized. Training of the RNN is carried out by following the procedure outlined for training the initial configuration. It was observed from table 3, that for 20 nodes in the hidden layer the time taken for training minimum. The simulation of RNN with different input combinations was performed to build RNN for suitably controlling the manufacturing systems. From this case study, it was observed that the time taken for modifying the transition table quite less and minimum time for training the RNN is less than 1 minute. Experiences in this case study shows that minimum effort is required for adapting to the control requirement arising out of reconfiguration in 0 1 Inputs Start 0 1 a0 1 1 a1 0 0 b0 1 1 b1 0 0 c0 1 1 c1 0 0 1 1 d0 0 0 d1 ACT 0 0 0 0 RJT Outputs A_forward 0 1 B_forward 0 0 B_reverse 0 0 A_reverse 0 0 C_forward 0 0 C_reverse 0 0 0 0 D_forward 0 0 D_reverse manufacturing system.

2 1 0 0 1 0 1 0 1 0 0 0

3 1 0 1 1 0 1 0 1 0 0 0

4 1 0 1 0 0 1 0 1 0 0 0

5 1 0 1 0 1 1 0 1 0 0 0

6 1 0 1 0 0 1 0 1 0 0 0

7 1 0 1 1 0 1 0 1 0 0 0

8 1 0 0 1 0 1 0 1 0 0 0

9 1 1 0 1 0 1 0 1 0 1 0

10 1 1 0 1 0 0 0 1 0 1 0

11 1 1 0 1 0 0 1 1 0 0 0

12 1 1 0 1 0 0 0 1 0 0 0

13 1 1 0 1 0 1 0 1 0 0 1

14 1 1 0 1 0 1 0 0 0 0 0

15 1 1 0 1 0 1 0 0 1 0 0

16 1 1 0 1 0 1 0 0 0 0 0

17 1 1 0 1 0 1 0 1 0 0 0

1 0 0 0 0 0 0 0

0 1 0 0 0 0 0 0

0 1 0 0 0 0 0 0

0 0 1 0 0 0 0 0

0 0 1 0 0 0 0 0

0 0 0 1 0 0 0 0

0 0 0 1 0 0 0 0

0 0 0 0 1 0 0 0

0 0 0 0 1 0 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1

0 0 0 0 0 0 0 1

1 0 0 0 0 0 0 0

Table 2 Condition table for the problem specification with provision for rejecting parts

Number of Hidden nodes 12 14 16 18 18 20 22 24

Number of Epochs

Time Taken (Hour:Min:sec)

For Initial system configuration 15458 00:03:00 13452 00:02:32 2139 00:00:24 3279 00:00:38 For modified system configuration 36974 00:08:40 4403 00:00:54 8710 00:01:49 6937 00:01:24

Table 3 Effect of number of hidden layer nodes on training

737


4.

CONCLUSIONS This paper presents an attempt to employ RNN for control of existing manufacturing systems. This attempt used LD program along with physical system configuration to develop a RNN based control for any manufacturing system. From the studies reported in this paper, the following observations can be made (1) By considering the status of physical system from time to time, fully developed LD program could be converted into state transition table easily. (2) Sequential control of reconfigured systems with the addition of certain physical units was realised with modification of state transition table and utilizing it for developing RNN based control. (3) The structure of Elman‘s RNN and the parameters chosen for training RNN is found to be effective for developing RNN based sequential control easily. (4)

The time taken for training of RNN for original and modified configurations was about 24 and 54 seconds respectively, which is far less than the time generally needed for developing LD programs for reconfigured manufacturing systems.

All the above points clearly indicate that the proposed approach is effective for developing sequential controllers with RNNs instead of modifying LD programs with extensive effort. Future efforts are directed towards investigating the suitability of RNN based approach for the control of complex manufacturing systems. 5.

REFERENCES [1] Setchi, R. M. and Lagos, N., Reconfigurability and Reconfigurable Manufacturing Systems - State-ofthe-art Review, IEEE International conference on Industrial Informatics, 2004, 529-535. [2] Falcione, A. and Krough, B. H., Design recovery for relay ladder logic, IEEE Transactions on Control Systems Magazine, 1993, 90-98. [3] Lucas, M. R. and Tilbury, D. M., A study of current logic design practices in the automotive manufacturing industry. Int. J. Human- Computer Studies, 2003, 59, 725–753. [4] Hajarnavis, V. and Young. K., An investigation into programmable logic controller software design techniques in the automotive industry, Assembly Automation, 28 (1), 2008, 43-54. [5] Chirn, J. L. and McFarlane, D. C., Petri Nets based design of Ladder logic diagrams, In the Proceedings of the UKACC International Conference on Control, Cambridge, UK, 2000. [6] Venkatesh, K., Zhou. M. C. and Caudill, R., Automatic generation of Petri net models from logic control specifications, Proceedings of theFourth International Conference on Computer Integrated Manufacturing and Automation Technology, 1994, 242-247. [7] Peng. S. S. and Zhou. M. C., Ladder diagram and Petri-net-based discrete-event control design methods, IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews, 2004, 34(4), 523-531. [8] Rodriguez, P., Wiles, J. and Elman, J.L., A Recurrent Neural Network that Learns to Count, Connection Science, 1999, 11(1), 5- 40. [9] Ku C. C. and Lee K. Y., Diagonal recurrent neural networks for dynamic systems control, IEEE Trans. Neural Networks, 1995, 6, 144–156. [10] Abdelhameed, M. M. and Tolbah, F. A., Design and Implementation of a Flexible Manufacturing Control System Using Neural Network, Intl. J. of Flexible Manufacturing system, 2002. [11] Vimal Nandhan R.K. and Ramesh Babu N., Effective Reconfiguration of Control Logic in Manufacturing Systems with Recurrent Neural Network, Proceedings of 3rd International & 24th AIMTDR Conference conducted during December 13-15, 2010 at Andhra University, Vishakhapatnam, India, vol.(1), 145-151. [12] Yu C. C. and Liu B. D., A backpropagation algorithm with adaptive learning rate and momentum coefficient, Proc. of IEEE Int. Joint Conf. Neural Networks (IJCNN'02), 2002, vol. 2, 1218 - 1223.

738


DESIGN AND EXPERIMENTATION OF A SIX WHEEL LUNAR ROVER FOR MOTION ON UNEVEN TERRAIN Harjinder Singh, Anjali Kulkarni, Biswanath Panda, Anupam Sana and Ashish Dutta Dept. of Mechanical Engineering, Indian Institute of Technology Kanpur, Kanpur 208016, India. (E-mail: anjalik/bpanda/sanupam/[email protected]) ABSTRACT The last decade has witnessed a renewed interest in lunar exploration missions. In this paper we describe the design of a rocker-bogie type six wheel rover for lunar exploration. The rover has been designed for minimum mass by analysing the stress and deflection on its links, using finite element method. Kinematic models have been developed for simulating the rover motion on uneven terrain. The ratios of wheel velocities are optimized for control of the ten motors during motion. Experimental studies have been conducted to compare the simulation results with the real motion of the rover for different terrains. Keywords: rocker-bogie mechanism, finite element analysis, trajectory tracking. 1.

INTRODUCTION In this paper we describe the design, development and experimentation of a six-wheel mobile robot (rover) for motion on uneven lunar like terrain. The lunar terrain is expected to contain craters, dust, ash and rocks that make the design of lunar rovers a challenging task. During motion the rover is expected to slip and sink in the soft terrain that would make it deviate from its intended path. The basic suspension mechanism of the six wheel rover was designed using a rocker-bogie mechanism, that has been used in may space rover. This mechanism imparts the rover with a maximum obstacle climbing ability and also ensures the stability of the main body. In order to design the rover with a minimum mass a detailed comparative study was done to compare the stresses and deflections on different linkages of the rover, for a basic rover weight of 20Kg. Kinematic relationships were derived between all the main points of the rover using Denavit Hartenberg (DH) parameters. These relationships were used to simulate the motion of the rover on different terrains. The rover has a total of 10 Degrees of freedom with independent DC brushless motors actuating each wheel. The 3D map of the terrain is obtained from a structured light system based system mounted on the rover. Once the best path is obtained the best wheel velocity ratios are calculated and the motors are actuated to move the rover along that path to the desired goal point. Ishigami et al. [1] proposed a method of steering planetary rovers in very loose soil. Their main focus was on developing soil-wheel interaction models for motion. Reina et al. [2] studied methods for detection of wheel slippage and sinkage on soft terrain. Wettergreen et al. [3] discussed the design of a rover concept for crater exploration. Harrington et al. [4] focus on the analysis and design of the rocker-bogie suspension for mars rovers. Linderman et al. [5] studied the mobility sub-systems for the mars exploration rovers. Chen et al. [6] used evolutionary techniques to design high performance systems for lunar rovers. Patel et al. [7] compared different locomotion systems for mars micro and macro rovers. Parakh et al. [8] derived the kinematic models of a six wheel rover for motion on uneven terrain. The design of the rover using FEM is discussed in Section 2, and the kinematic models are derived in section 3. Section 4 details the experiments carried out on the rover on different types of terrain and the conclusions are drawn in section 5.

2.

ROVER DESIGN Based on the available space the overall dimensions of the rover were set at a length of 550 mm, a width of 500 mm and a height of 350 mm (as shown in Figure 1). The maximum weight of the rover including onboard equipment was set at 20 Kg. The under-body clearance was set at 190 mm from the ground. The wheel-base has been fixed at 450 mm and height of the CG is kept at 285 mm from ground. The track of the rover is 400 mm wide.

739


Figure 1. Basic dimensions of the rover.

Once the basic size of the rover was fixed the links were analyzed using FEM to obtain the least total mass of rover. In this section we analyse four types of links for the rocker and the bogie. The four link cross sections considered are solid link, box type link, solid link with cut out and solid link with circular holes. The best link design for the rover was chosen so that different parameters like stress, displacement are within the limit. As the speed of rover is very slow (10 cm/sec), only static forces will act on the rover structure and the acceleration forces are neglected. The basic modelling of the links is done in Solid works and then the models are imported in Abaqus software for FEM analysis. Rover design parameters were considered for a flat terrain, and total weight of the rover was shared by six wheels equally. A 35 N load was applied on the wheels and corresponding constraints of displacement and rotation are applied in the rocker-body pivot. Al6061-T6 is selected as material for the rocker links. The analysis of the links comprises of 20-node quadratic brick (C3D20R) elements. Figure 2 shows the stresses and deflections for different types of cross sections. The link cross section that gave the least weight and deflection was chosen for the design. Following this procedure all the links were designed.

Figure 2. Stress analysis of links for best design.

740


Figure 3. Spherical wheels of the rover with grouser blades for slip preventions. The rover has six independent wheels and the front and rear wheels have two degrees of freedom each. Each degree of freedom is driven by a Maxon brushless DC servo motor controlled by a Maxon motion controller, via the central controller. The wheels of the rover are spherical in shape in order to ensure that the reaction force from the ground passes through the centre of the wheel for all types of surface profiles. If this condition is not satisfied than a moment will result on wheel that will increase the required torque of the motor. The CAD model of a single wheel is shown in Figure 3. 24 grouser blades have been provided on the wheels for better grip.

Figure 4. The averaging link for keeping the central body at an average position. The four corner wheels are explicitly steered whereas the middle wheels are driven suitably (backward or forward) so that the vehicle turns about the centre of the axis connecting the middle wheels. The body of the rover is pivoted to the rocker at the centre. The body is attached to the suspension at a third point through the pitch averaging mechanism. The pitch averaging mechanism consists of an averaging link which is pivoted on the body using a revolute joint and two connecting rods which are attached to the rocker and the averaging link using revolute joints. At the centre of the averaging link (pivot on the body), the rocking motion of the two rockers is averaged and imparted to the body, which in turn executes pitching motion about the rocker and body pivot axis. Figure 4 shows detail view of averaging link and connecting rods assembly. The CAD model of the rover is as shown in Figure 5. This mechanism ensures that the body is at half the difference in gradients of the two wheels on either side.

741


Figure 5. CAD Model of rover.

3.

KINEMATICS OF ROVER MOTION In order to obtain slip less traction control, kinematic analysis was performed to determine the coordinates of all the main points on the rover as it moves on rocky terrain. Axes were assigned at each joint (as shown in Figure 6) of the rover and the relationship between each frame was determined using homogenous transformation matrices using Denavit Hartenberg (DH) parameters of the rover.

Figure 6. Axes assigned to the rover as per DH conventions.

The transformation from one frame to another is given by the generalised transforamtion matrix:

742


(1)

Where ‗Ө, α, a and d‘ are the DH parameters of th elinks with the reference coordinate system (0,0,0) chosen at the centre of gravity of the rover (pivot point on the body).

TC1,R0 = TC1,A1 ×TA1,K2 ×TK2,K3 ×TK3,B0 ×TB0,R0

(2)

TC1,RO is the transformation from the body centre frame to the wheel1 and ground contact frame as given by equation 2. The wheel frames for the other two wheels are given by:

TR0,C2 = TR0,B0 ×TB0,K6 ×TK6,A2 ×TA2,C2

(3)

TR0,C3 = TR0,K4 ×TK4,A3 ×TA3,C3

(4)

However this formulation does not facilitates motion of the rover. For the simulaiton of rover motion we take a frame which we call as the motion frame and apply this motion frame transformation to the wheel and terrain contact point frames. Depending on the wheel velocity (rӨ) and the turing angle (δ) as given in Figure 7, we determine the motion of the rover as given by equation (5):

(5)

Figure 7. Wheel velocity and truning angle transformation. Rover motion has been experimentally evaluated on a rough terrain in order to prove the effectiveness of our proposed control methods. The complete 3D terrain generated by a structured light system was modelled by using small straight line segments as a grid patterns with an average gradient. Rocks are modelled by straight lines with gradients as shown in Figure 8. In such problems as the wheel moves forward using the kinematic relations given in equations 1-5, there are discontinuities (sharp angles) between line segments as shown in Figure 9. In such cases the wheel contact model switches from one contact to the next as shown in the figure and equation 6.

743


Figure 8. Surfaces are modeled by small straight line segments with a gradient.

Figure 9. Two point conatct when wheel meets a gradient. Depending on the angle of conatct the switching algorithm finds the next contact as given by: XC = XA + r Sin(α) YC = YA + 2 r Sin2(α/2)

(6)

Where ‗α‘ is the difference in the slopes between the two straight segments and ‗r‘ is the radius of the wheel. 4.

EXPERIMENTAL EVALUATION The rover has 10 brushless servo motors for motion with the front and rear wheels having both steering and drive, while the middle wheel has only drive. Each of these are controlled by a Maxon EPOS motor controller via a Laptop based central controller. After the 3D map is built and the best path calculated, the central controller calculates the best velocity ratios to reach a desired goal point. Experiments were carried out in which the rover was moved over various test terrains and the importnat points (wheel centre, CG, etc. ) of the rover tracked using a vision system. Figure 10 shows the motion of the rover over an inclined terrain and the corresponding motion on a real inclined surface as given in Figure 11. Figure 12 shows the motion of the rover on a rectangular obstacle ( rectangular brick). The difference between the simulated and experiemntal trajectories was found to depend on the terrain travesed. In the case of flat ground the two trajectories was the same while for an inclined surface there was diffrences at the two point contact. In the case of sudden change in gradients the motion of the rover was a radius about the surface intersection points. We need to incorporate slip into our control algorithms for more accurate rover motion control.

744


Figure 10. Simulation of the rover on inclined terrain.

Figure 11. Rover moving on inclined terrain.

Figure 12. Experimental trajectory of rover moving over a rectangular obstacle and the corresponding motion of the wheel centre points and body centre.

5.

CONCLUSION In this paper we presented the design and control of a six wheel lunar rover developed for motion on uneven terrain. The kinematic simulaiton shows good agreement with the real trajectories obtained during real rover motion. However in case of very large change in gradients the simualtion and experiemntal resuts show significant differences. In future we are incorporating other effects like wheel slip etc. into our simualtions.

745


6.

REFERENCES [1] G. Ishigami, A. Miwa, K. Nagatani, K. Yoshida, Terramechanics-Based Model for Steering Manoeuvre of Planetary Exploration Rovers on Loose Soil, Journal of field Robotics 24(3), 233-250, 2007. [2] G. Reina, L. Ojeda, A. Milella, J. Borenstein, Wheel Slippage and Sinkade Detection for Planetary Rovers, IEEE/ASME Transactions on Mechatronics, Vol. 11(2), 185-195, April 2006. [3] D. Wettergreen, D. Jonak, D. Kohanbash, S. Moreland, S. Spiker, J. Teza, W. Whittaker, Design and Experimentation of a Rover Concept for Lunar Crater Resource Survey, 47th AIAA Aerospace Sciences Meeting Including The New Horizons Forum and Aerospace Exposition, January 2009. [4] B. Harrington, C. Voorhees, The challenges of designing the Rocker-Bogie suspension for the Mars exploration rover, in: Proceedings of the 37th Aerospace Mechanisms Symposium Galveston, TX, NASA/CP-2004-212073, 185-196, May 2004. [5] R. A. Lindemann, L. Reid, C. Voorhees, Mobility Sub-System for the Exploration Technology Rover, 33rd Aerospace Mechanisms Symposium, 125-140, May 1999. [6] B. Chen, R. Wang, Y. Jia, L. Guo, L. Yang, Design of a high performance suspension for lunar rover based on evolution, Astronautica 64, 925-934, 2009. [7] N. Patel, A. Ellery, C. Welch, A. Curley, Comparative locomotion study for Mars micro-rovers and mini-rovers, in: Proceedings of 55th International Astronautically Congress, Vancouver, Canada, 1-11, 2004. [8] Shrikant Parakh, Pankaj Wahi and Ashish Dutta. ―Velocity kinematics of a rocker-bogie type planetary rover‖. Proceedings of the IEEE TENCON, Fukuoka, Japan, 2010, pp. 939-944.

746


PERFORMANCE INDICES FOR SERIAL ROBOTIC MANIPULATORS Ekta Singla1 and Ashish Singla2 1

School of Mechanical, Materials and Energy Engineering Indian Institute of Technology Ropar, Punjab, India e-mail1: [email protected] 2 Mechanical Engineering Department Thapar University, Patiala, India e-mail2: [email protected]

ABSTRACT To serve the expanding variety of workspaces which need robotic assistance, it is beneficial and at times necessary, to design customized manipulators. The design of such task-based special manipulators is majorly concerned with two decisions --- which are the design parameters? And what are the optimality criteria? This paper is focussed at the later aspect. Various performance measures which are usually used in robot design and motion planning problems are presented in this paper. Besides, a novel path-based global index is discussed in the paper which signifies the ease of a robotic arm to pass through the cluttered environments. Keywords: Global Performance Index, robotic manipulators, optimal design, motion planning objectives 1.

INTRODUCTION To analyze the efficiency of a manipulator, some quantification of its performance is required. The definition of such a performance measure is concerned with the kind of application the robot is intended for. Be it the problem of posture planning, path planning or design of a manipulator, the performance indices play their role in taking several decisions. For non-redundant manipulators, the focus is more on singularity avoidance and determination of the postures with good kinematic conditions. However, for redundant manipulators, with infinite solutions available for each location of the end-effector, the performance indices are usually designed for dexterity improvement and simultaneous fulfilment of desired objectives like safety from collisions, minimum energy usage etc. This section covers the several performance measures proposed and utilized in past. The past work is discussed in two categories, the local and the global formulations.

Local performance indices A performance index is considered local if its formation is based upon the present configuration (local information) only. In general, their definition involves the properties of the manipulator‘s Jacobian matrix, signifying the kinematic conditions of the manipulator at the particular posture. Salisbury and Craig [1] proposed the condition number of the Jacobian matrix for quantifying the dexterity of articulated hands. A manipulability measure, proposed by Yoshikawa [2] for optimal posture planning, defined as the squareroot of the determinant of the product matrix of Jacobian by its transpose. The scalar quantity signifies the ease of changing the position and orientation of the end-effector in any direction. A number of Jacobian based dexterity measures were proposed and analyzed by Klein and Blaho [3]. The work focussed at the comparison of condition number, minimum singular value and joint range availability, on the basis of their suitability to determine optimal work points in task space and to find optimal configurations for specified task locations. The measures were also incorporated for designing optimal link lengths for a planar manipulator. A few modifications were suggested by Gosselin [4] to make the formulation frame invariant, for both planar and spatial manipulators. Mayorga [5, 6] proposed a dexterity measure, based on the rate of change of Jacobian matrix, providing some bounds on the condition number of the Jacobian and on the joint velocities and accelerations. Using these constraints, an optimization problem is formulated to determine link lengths of the manipulators. Global performance indices The definition of global indices is based upon the information of the desired criterion in the entire workspace. These are considered more beneficial, particularly for the design purposes and also for many

747


path planning applications. The attempts made for developing global approaches for redundancy utilization, discussed in previous sections, are related to global manipulability, global minimization of kinetic energy or torque. Stocco [7] proposed an algorithm to compute global isotropy index for design purposes. Recently, Liu et al [8] discussed a number of global indices and utilized these measures for design of a 5-R manipulator. The indices considered were global conditioning index, global velocity index, global payload index and global stiffness index. Wenger [9] presented a few kinematic properties, including cuspability, genericity and solvability, and suggested their use for novel designs of non-redundant manipulators, rather than using conventional structures available. The work is suitable for non-redundant manipulators only. A few recent works discussing the utilization of local and global performance indices are presented in [10,11,12]. Some global performance indices play an important role in motion planning, particularly in optimizing a path based on some criteria or in selecting a planner for a particular application. Lumelsky [13] chose the criteria as minimizing the path lengths for his study on path length performance of several robot motion planning algorithms. In general, the aspects usually considered for a path planning problem are shortest path in joint space, minimum velocity or acceleration over the entire trajectory, minimizing the energy and/or maximizing the safety from collision. The following section highlights the performance metrics used for fundamental manipulators and redundant robotic arms. 2.

DESIGN CRITERIA: FUNDAMENTAL ROBOTIC ARMS The kinematic synthesis of a manipulator involves determination of its D-H parameters, satisfying certain kinematic requirements, e.g. reachability, obstacle avoidance and/or dexterity of the manipulator. Various performance measures utilized for design purposes are discussed in previous sections. This section summarizes the attempts made in past for the kinematic design of serial manipulators. Before analyzing the extent of work done for the designs of redundant manipulators, the discussion first covers the approaches utilized for non-redundant serial manipulators. The design strategies presented for non-redundant serial manipulators are, in general, concerned with the requirement of singularity-free workspaces and/or maximal work volume. A lot of research [14,15,16] in this area was focused on isotropic designs of articulated hands, grippers, closed kinematic chains or robotic arms. Paden and Sastry [17] worked for the design of a 6-R manipulator for well-connected (singularityfree) workspace. Gonzalez et al [18] presented a systematic approach for the kinematic synthesis of serial manipulators for a prescribed Jacobian. Paredis and Khosla [19] introduced another aspect of task-based robotic design. They presented an algorithm to design manipulators for prescribed end-effector locations, for given constrained environments. Basically, in these approaches, the Jacobian condition number is expressed analytically as a function of link lengths and joint angles, based on the requirements of isotropy. These non-linear constraints are solved for the solution of these design parameters. Kircanski [20] presented a comprehensive review and analysis of several works on isotropic manipulator designs. The study reported the complete design problem (with all the manipulator parameters as design variables) as highly complicated and therefore, the designs were confined to the link lengths only. Recently, many researchers worked for the optimal design of fundamental robot manipulators for different objective criteria. Snyman [21] proposed the use of optimization tools for the mechanism design and presented the synthesis of a 3-dof manipulator for minimizing torque over the entire trajectory. Another work presenting the optimal design of 3-R manipulator is by Ceccarelli and Lanni [22], with the objectives of maximal work volume and minimum link lengths within prescribed workspace limits. Kucuk and Bingul [10] presented a comparative study of many local and global performance indices used for the design of fundamental robotic manipulators. He formulated the problems as multi-objective optimization problems and used evolutionary algorithms and sequential quadratic programming (SQP) for the solutions. The comparisons in different designs is presented in a systematic tabular format. A similar study is presented by Liu et al [8] for a 5-R mechanism. The aim of such studies is to provide a reference for the working of industrial manipulators, which, in general, possess a small number of degrees of freedom. Very recently, Carbone et al [23] presented an optimal design algorithm for multiple objectives, suitable for both serial and parallel manipulators. The work emphasized on the use of efficient optimization tools for the fulfilment of many criteria simultaneously.

748


3.

DESIGN CRITERIA: REDUNDANT MANIPULATORS Angeles [15] and Ranjbaran et al [16] attempted design of redundant manipulators for kinematic isotropy. The approaches involved the formation of many non-linear constraint equations to be satisfied by the D-H (Denavit-Hartenbeg) parameters of the manipulator. A step ahead in this direction, Tsai and Wang [25] developed a strategy to deal with larger number of degrees of freedom for isotropic designs.The strategy is based upon the utilization of 6-dof generators (isotropic designs) as modules for the manipulator design of any desired number of degrees of freedom. Another strategy presented by Mayorga [5] includes a performance index based on the rate of change of the manipulator Jacobian. The approach is implemented for planar redundant manipulators with 3-links only and is not recommendable for spatial cases having larger number of design variables, due to the cumbersome analytical formulations involved. A very recent work on design of serial manipulators is reported by Oetomo et al [27]. The strategy works on the collection of all the achievable workspace locations, satisfying the given constraints. The constraints include reachability within the joint limits, singularity-free configurations, ability to exert a prescribed amount of force at the end-effector. The work is presented for a 6-dof manipulator without any obstacles.

4.

PATH-BASED PERFORMANCE INDEX “ROBOGIN” To develop the task-based manipulators, the focus is on the best performance of the robot while executing the prescribed task. In this work, the problem involves some prescribed locations in a given work cell and the desired manipulator should be able to work at all these points. Therefore, the focus is on the capability of the manipulator to operate between these specified locations in the best sense. This is incorporated through a performance index which reflects the uncertainties involved at different stages of design and development of a manipulator, as enumerated below. The proposed measure, named as RoboGin, derived from ―robot margin‖, is defined as ―the minimum margin between the robot and the obstacles or among the robot links‖. For a particular joint configuration, this measure is represented as RoboGinC, which signifies the closest distance of the robot links to the obstacles (or among themselves) at the specific posture (refer Fig. 1 for a pictorial representation).

Figure 1: RoboGinC

The definition of the measure for a complete path of the manipulator is an extension of RoboGinC itself and is represented as RoboGinP. It is described as ‗the minimum RoboGinC of the manipulator, while traversing the path‘. Fig. 2 provides a schematic representation for RoboGinP. The figure shows a few configurations of a robotic arm while performing a task and the minimum distance over this path is marked as RoboGinP. The algorithmic steps to compute this performance index are as follows.

749


1. Determine the joint configurations for all the Task Space Locations (TSL's), using an inverse kinematics procedure. The objective of the optimal inverse kinematics problem is taken as maximization of RoboGinC and the kinematic equations (for reachability of the end-effector at the Cartesian positions) are incorporated as equality constraints. The procedure used for optimal inverse kinematics is not a part of this paper. 2. Plan the optimal path between each pair of joint configurations obtained in the previous step. For N TSL's, (N-1) paths would be developed. 3. Compute the critical distance ($RoboGin_P$) for the total path planned between the TSL's.

Figure 2: RoboGinP To illustrate the significance of the performance measure through a minimum distance plot, Fig.3 presents the approach distance of the manipulator to the obstacles (RoboGinC) corresponding to each grid point of the planned paths connecting four working locations. Ordinates are drawn to indicate the grid points corresponding to each task space location. The critical distance, RoboGinP , is marked as the minimum value for the complete path.

Figure 3: Illustration of RoboGin for four task space locations.

750


5.

IMPLEMENTATION AND ANALYSIS OF ROBOGIN The complete design optimization problem can be defined as the determination of an optimal set of robotic parameters, within the prescribed limits, which can provide maximum RoboGin while performing in the given workcell through the specified task space locations. This section details the formulation of an optimization problem. The manipulator design problem formulated as the optimization problem is supposed to handle the following constraints: 1. the bounds over the design variables, 2. the constraints due to obstacle avoidance, and 3. the reachability of the end-effector at all the desired TSL‘s, with the kinematic condition of the manipulator within the prescribed limits. In each iteration of the optimization technique, a new variable vector is generated through a perturbation within prescribed bounds. With the updated values of the robotic parameters, the inverse kinematics is performed. This segment ensures the reachability of the end-effector at the required locations with good kinematic conditions.

Case study 1 This case presents the design of an 8-link manipulator to be worked between two locations. The path planned between two TSL‘s in a room environment is shown in Fig.4. The case is selected for the illustration purpose, as the globally optimal RoboGinP value for these two TSL‘s is known a priori, which signifies the minimum distance between one of the locations and an obstacle (table). Therefore the result of the RoboGinP, for the optimal design is verified.

Figure 4: Path planned by a manipulator for case study 1 The existing results present the optimal path possible for the D-H parameters obtained through the initial synthesis, and do not show possibility for much improvement in the critical margin because of the location of a TSL being very close to the obstacle. The present result shows the improvement in the critical distance through the iterative process of design optimization.

751


It is observed that the prescribed TSL (120,100,30) is located very close to an obstacle and the resulting objective function value is corresponding to the margin of this working location from the table. With this constrained location, no perturbation in the converged solution can increase the margin.

Figure 5: RoboGinP plots for initial and optimal design for case study 1 Fig. 5 presents the comparison of the margin plots for the previous (initial) and the present (optimal) design parameters. The figure shows the improvement in RoboGinP which has increased from 1.77 to 4.01. This minimum margin for optimal design corresponds to the start configuration. In this case, this is the point which decides the value of the final solution. Case study 2 This case is presented to show the significant improvement in design of another manipulator, reporting increment in RoboGinP from a crucial value of 0.24 to a considerable margin of 1.02. The workspace consists of some shelves and the manipulator is required to work between two TSL‘s. The discussion on the collision avoidance in spatial environment and other design details are not the parts of this paper. RoboGinP plots for the initial and the final robotic parameters are shown in Fig. 6. Each plot shows the RoboGinC values for all the milestones of the paths between each task space locations. The plots corresponding to the initial and the final (optimal) design are shown through hashed and solid lines.

752


Figure 6: RoboGinP plots for initial and optimal design for case study 2 6.

CONCLUSION Thorough discussion on the various performance indices of serial manipulators, used by researchers in design and motion planning problems, is shown in this paper. Besides, a performance index (RoboGin) is proposed in this paper which is the minimum margin between the robot links and the obstacles or among the robot links, while traversing the planned path. The manipulator design can be optimized with respect to the maximization of this proposed measure. The index signifies the ease of a robotic arm to maneuver through the given environement.

7.

REFERENCES [1] J.K. Salisbury and J.J. Craig. Articulated hands: Force control and kinematic issues. The International Journal of Robotics Research, 1(1):4–17, 1982. [2] T. Yoshikawa. Manipulability of robotics mechanisms. Journal of Robotics Research, 4(2):3–9, 1985. [3] C.A. Klein and B.E. Blaho. Dexterity Measures for the Design and Control of Kinematically Redundant Manipulators. The International Journal of Robotics Research, 6(2):72–83, 1987. [4] C.M. Gosselin. Dexterity indices for planar and spatial robotic manipulators. In IEEE International Conference on Robotics and Automation, pages 650–655, 1990. [5] R.V. Mayorga, B. Ressa, and A.K.C. Wong. A dexterity measure for robot manipulators. In IEEE International Conference on Robotics and Automation, pages 656–661, 1990. [6] R.V. Mayorga, B. Ressa, and A.K.C. Wong. A kinematic design optimization of robot manipulators. In IEEE International Conference on Robotics and Automation, pages 396–401, 1992. [7] L. Stocco, SE Salcudean, and F. Sassani. Fast constrained global minimax optimization of robot parameters. Robotica, 16(06):595–605, 2001. [8] X.J. Liu, J. Wang, and G. Pritschow. Performance atlases and optimum design of planar 5R symmetrical parallel mechanisms. Mechanism and Machine Theory, 41(2):119–144, 2006. [9] P. Wenger. Some guidelines for the kinematic design of new manipulators. Mechanism and Machine Theory, 35(3):437–449, 2000. [10] S. Kucuk and Z. Bingul. Comparative study of performance indices for fundamental robot manipulators. Robotics and Autonomous Systems, 54(7):567–573, 2006. [11] M. Badescu and C. Mavroidis. Workspace optimization of 3-legged UPU and UPS parallel platforms with joint constraints. Journal of mechanical design, 126(2):291–300, 2004. [12] K. van den Doel and D.K. Pai. Performance Measures for Robot Manipulators: A Unified Approach. The International Journal of Robotics Research, 15(1):92–109, 1996. [13] V.J. Lumelsky. A comparative study on the path length performance ofmaze-searching and robot motion planning algorithms. IEEE Transactions on Robotics and Automation, 7(1):57–66, 1991.

753


[14] O. Ma and J. Angeles. Optimum design of manipulator under dynamic isotropy condition. in: Proc. IEEE Int. Conf. Robot. Autumn., 1:470 – 475, May 1993. [15] J. Angeles. The Design of Isotropic Manipulator Architectures in the Presence of Redundancies. The International Journal of Robotics Research, 11(3):196–201, 1992. [16] F. Ranjbaran, J. Angeles, M.A. Gonz ́lez-Palacios, and R.V. Patel. The mechanical design of a seven-axes manipulator with kinematic isotropy. Journal of Intelligent and Robotic Systems, 14(1):21– 41, 1995. [17] B. Paden and S. Sastry. Optimal Kinematic Design of 6R Manipulator. The International Journal of Robotics Research, 7(2):43–60, 1988 [18] M.A. Gonzalez-Palacios, J. Angeles, and F. Ranjbaran. The kinematic synthesis of serial manipulators with a prescribed Jacobian. In IEEE International Conference on Robotics and Automation, pages 450– 455, 1993. [19] C.J.J. Paredis and P.K. Khosla. Kinematic design of serial link manipulators from task specifications. International Journal of Robotics Research, 12(3):274–287, 1993. [20] M. Kircanski. Kinematic isotropy and optimal kinematic design of planar manipulators and a 3DOF spatial manipulator. The International Journal of Robotics Research, 15(1):61–77, 1996. [21] [J.A. Snyman and F. van Tonder. Optimum design of a three-dimensional serial robot manipulator. Structural and Multidisciplinary Optimization, 17(2):172–185, 1999. [22] M. Ceccarelli and C. Lanni. A multi-objective optimum design of general 3R manipulators for prescribed workspace limits. Mechanism and Machine Theory, 39(2):119–132, 2004. [23] G. Carbone, E. Ottaviano, and M. Ceccarelli. An optimum design procedure for both serial and parallel manipulators. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 221(7):829–843, 2007. [24] K.Y. Tsai and Z.W. Wang. The design of redundant isotropic manipulators with special link parameters. Robotica, 23(02):231–237, 2005. [25] RV Mayorga, B. Ressa, and AKC Wong. A kinematic criterion for the design optimization of robot manipulators. In Robotics and Automation, 1991. Proceedings., 1991 IEEE International Conference on, pages 578–583, 1991. [26] D. Oetomo, D. Daney, and J.P. Merlet. Design Strategy of Serial Manipulators with Certified Constraint Satisfaction. IEEE Transactions on Robotics, 25(1):1–11, 2009.

754


SEMI-AUTOMATED COLOUR DYEING SYSTEM (YARN AND READY PIECES) FOR SMALL TIME SONGKET MANUFACTURERS Mohd Shahrul Azmi Mohamad Yusoff1, Shaifull Azhar Othman1, & R Fairuz Indra Didi Indra Tjahya1 Industrial Automation and Robotics Section, SIRIM Berhad ABSTRACT For years the colouring of yarn has been one of the most important processes in the manufacturing of textile products be it in established industrial area or the low income manufacturing in and around the rural area. Studies have shown that the traditional method used for yarn colouring in rural handicraft manufacturing often exposed the operators to hazardous chemicals and working conditions. The expensive automated colouring machines available in the market are often beyond the affordability of small time songket manufacturers. Semi-automated colour dyeing system for yarn and ready pieces is a novel invention that reduce the danger to small time manufacturers of songket at an affordable price. A year of systematic approach in designing the semi automated machine was put to test in real manufacturing environment for the last 6 months. Testing and debugging of the design were performed at the industry and the result was a proud success. The semi automated machine is now ready to be mass produce with minor adjustment to the heat exchanger component and the adjustability in dyeing volume. With the low price tag of the semi automated machine and the reduction in operating complexity with comparison to other available machine in the market, the commercialisation potential of this semi automated machine is inevitable. 1.

INTRODUCTION The race to become the world premier textile exporter has increased the demand of transformation to the current traditional methods that are still widely used in rural areas in Malaysia. As it is, the local textile entrepreneurs are mostly located deep in rural areas and operating in a small scale basis making the implementation of heavy machinery difficult. However, there is a possibility of improvement to the current practice with the help from government agencies by providing grants for research and development. Whether it is from the government or private entities, simple solution such as a semi automation system can be implemented to enhance the quality and productivity of local textiles production. The key is to build the ease the manufacturing step simply by introducing automation for one process at a time, without sacrificing the textiles aesthetic and uniqueness. [1]

Studies have shown that the traditional method used for yarn coloring in rural handicraft manufacturing often exposed the operators to hazardous chemicals and working conditions. The expensive automated coloring machines available in the market are often beyond the affordability of small scale songket manufacturers. Semi-automated color dyeing system for yarn and ready pieces is a novel invention that reduces the danger to small time manufacturers of songket at an affordable price. [1], [2] 2.

RESEARCH PROCEDURES

2.1 Preliminary studies Preliminary studies concentrated on the manual methods practiced by local textile entrepreneurs located in three different states within Malaysia i.e. Terengganu, Kelantan and Sarawak. It is imperative to truly

755


understand the process flow of textile production in order to find the key areas that can help the local operators improving the textile quality and productivity. The first step taken is to understand the overall process flow of the songket weaving production. A process flow analysis was created based on the input from manufacturer as well as the operators. 2.2 Identifying the key area of improvement After documenting all the processes involve in the production of songket weaving, the team identified the key area of improvement i.e. the key area that can be maximized via the introduction of semi-automated machine. All the key processes went through a detail analysis of which all the mechanical and electrical requirements were documented as the guideline in designing/integrating the appropriate machines. 2.3 Underlining the urgency of each potential improvement The degree of hazards imposed to the operators and the ergonomics factor of each and every process in the key area were then identified. Priorities are given to the areas that require urgent improvement based on the needs and the obstacles encountered by the manufacturers as well as the operators. Processes that are on top of the priority list were taken into the research program of which the buildings of semi-automated machine were funded by the government. 2.4 Design and development In this stage the team performs the required design procedures. The team consists of all related disciplines such as mechanical, electrical and control system. Within the tight schedule the team came up with the conceptual design which then undergo the required assessment eventually only the best design are considered for fabrications. 2.5 Testing and debugging After the completion of fabrications and all other systems, the semi-automated machine was then delivered to the manufacturing premises. Testing and debugging of the machine were done in-situ based on the input from operators. 3.

RESULTS

3.1 Process Flow After studying the various methods practiced in different weave production areas, it is found that the production process is not much different from one another. They depend much on the skills of local workers, without using any type of fully or semi-automated machine. The process flow as found out from the studies is as shown in the chart below:

756


Process flow of weave production Relevant data from the processes were identified and taken into account for the next research stage. 3.2 Key area of improvement To reduce time and eliminate hazards to operators during colouring To ease the take-up and let down process

To reduce the number of operators and reduce the time for warping process

5 key areas of improvement

To reduce the time for the process of winding

To increase the ergonomics Of take up process

Diagram 3.2.1 By observing and studying the songket weaving production process flow, the team identified 5 key areas of improvement shown in core element diagram 3.2.1; i.

Take up and let down – This process require great concentration as it is tedious for the operators to identify each and every yarn, transferring the pattern from on paper design to the weaving machine can be simplified by the introduction of pc base controller.

ii.

Colouring of yarn – hazardous gas and high humidity condition are two of the many problems encountered by operators. Poor management of water waste can well be overcome with proper treatment system. The efficiency of heating can be improvised by the correct seletion of automation tools.

iii.

Warping process – during the study the research team founded that the number of personnel/operators for every warping exercise can be reduce by 50%. This would allow appropriate reduction to production cost.

iv.

Take up – In this process operators are exposed to ergonomics hazards of which the working position can lead to other health risks issues.

757


v.

Winding machine – winding process was still a manually completed process which can certainly be improvised by minor modification to the drive system.

3.3 Improvement A deeper look into the color dyeing process, its techniques and methods, will show that the job is not only tedious, but very hazardous to the workers. The process is divided into 4 main stages, which are: 1.

Water heating. Water is heated using liquefied petroleum gas (LPG), or in some areas fire woods, up to the temperature of ±70° Celsius. The temperature need to be controlled properly, as any overheating or under heating will affect the color quality. Once the required temperature is achieved, coloring mixture is added into the water.

2.

Yarn hanging. Yarns are hung onto a stick which will act as a dipping mechanism.

3.

Dipping. The yarns are then dipped into the pre heated coloring/water mixture repeatedly. This process may take up to 1.5 hour to complete in order to obtain perfect color uniformity on the yarns. During this process the temperature of the mixture have to be at ±70° Celsius. If a drastic drop of temperature happens, the operator will need to heat it back at the water heating area.

4.

Drying. After the dipping process is completed, and color uniformity is obtained, the yarns are then brought to the drying area where they will be hung on drying rack and left to dry at room temperature.

To maintain a consistent high quality colored yarns requires highly skilled workers, and as most of the yarn factories are located in the rural areas, acquiring them is always going to be difficult. Some would hire skilled workers from the city maybe, but the cost to employ them would later become unbearable to sustain. The only option left for them is to send the made yarns to yarn coloring factories, which sometimes are located hundreds of kilometers away from their yarn factory. In addition, the high cost of the yarn coloring process (RM300/kg), will always a problem, a burden they have to carry. All the findings from the study concluded that among all the weave production processes, color dyeing is the most tedious and hazardous to the workers, and cost wise, the most unbearable for the entrepreneurs to sustain. Thus, focusing onto this matter becomes imperative, and a system to aid the workers is designed and developed. 3.4 Design and review The main purpose of the system is to replace the manual hands on method of coloring yarn and ready pieces of weave. The manual method requires the worker to dip the yarn/ready pieces into a pre heated coloring mixture repeatedly to obtain the uniformity of color over all area of the yarn/ready pieces. This process is replicated by the system, which consists of five rotating iron, ten sprinkler pipes and a water tub that can accommodate up to 1500 litres of coloring mix/water. The coloring mix/water is heated by a heat exchanger provided by the system. The use of heat exchanger to replace the current usage of LPG gas as the water/coloring mixture heater reduces the operational risk for the worker. The yarn/ready pieces are hung on the rotating iron, which will rotate and dip them into the coloring mixture, and at the same time, the mixture will be sprayed to the yarn/ready pieces thru the sprinkler pipes. Entire process can be completed in less than 30 minutes, which is a distant improvement than the manual method which can take up to one and a half hour to complete. The system is also able to process up to 50 yarns or 5 ready pieces at the same time. It is able to improve the overall process of coloring yarn/ready pieces of weave by much, thus making it applicable to be used widely in the weaving industry.

758


Comparative analysis for manual vs semi auto 4.

CONCLUSION To preserve the national heritage through the continuity of handicraft production in rural area, one has got to understand and proposed solutions to the obstacle faced by the industries. A full transformation from manual to automated machines will reduced the value of those crafts. However, drawing a line between handicraft and mass production will help researchers in bridging the gap between demand and the supply of songket products. The team will continue to look at other area of improvement that will benefit the weavers and entrepreneurs in the long run.

5.

REFERENCES [1] Bovard, James. The Fair Trade Fraud. New York: St. Martin‘s Press, 1991. [2] Cline, William. The Future of World Trade in Textiles and Apparel. Washington, D.C.: Institute for International Economics, 1987. [3] Lord, Montague. The Handbook of Latin American Trade in Manufactures. Northhampton, MA: Edward Elgar, 1998. [4] Macario, Carla. Export Growth in Latin America. Boulder, CO: Lynne Rienner Publishers, 2000. [5] ―Market Access for Developing Countries.‖ Finance & Development. September 2002. [6] Omondi, Vitalis. ―Fresh Hope for Cotton as Kenya Plans More EPZs.‖ East African 22 April 2002. 10 October 2002 [7] Tomkin, Robert. Trade Promotion Authority: CQ House Action Report. July 26, 2002. [8] U.S. Congress, Office of Technology Assessment, The U.S. Textile and Apparel Industry: A Revolution in Progress–Special Report, OTA-TET-332 (Washington, DC: U.S. Government Printing Office, April 1987). http://www.wws.princeton.edu/cgibin/byteserv.prl/~ota/disk2/1987/8733/873306.PDF [9] Beatrice Craig, Judith Rysiel & Elizabeth Turcotte – Survival or adaptation? Domestic rural textile production in eastern Canada in the later nineteenth centuries - 2001

759


DEVELOPMENT OF A HUMANOID HEAD ROBOT AMIR-III M.F. Alias, A.A. Shafie, and S.H. Hashim Department of Mechatronics, Faculty of Engineering International Islamic University Malaysia Kuala Lumpur, Malaysia [email protected] [email protected] [email protected] ABSTRACT This paper discusses the mechatronics development of a humanoid head robot by the International Islamic University Malaysia (IIUM) capable of displaying a range of facial expressions toward friendly interactions with humans. Such capability would further enhance the role of social robots particularly humanoids as they operate within human-existing environments. The robotic head, named as AMIR-III, is designed to be of iconic look so as to avoid the anthropomorphism pitfall as suggested in Mori‘s Uncanny Valley theory. Besides that, its unique yet simple mechanical mouth design forms part of its facial cues toward more recognizable facial expressions. AMIR-III is currently capable of portraying 5 basic facial expressions namely neutral, happy, angry, surprised, and sad. The performance of AMIR-III is evaluated in terms of the positioning accuracy of its actuators and facial action units (AUs) in which satisfactory outcomes are accomplished. Conclusively AMIRIII humanoid head robot is set to serve as a research platform for social robotics. Keywords: AMIR; humanoid robot; robotic head; human-robot interaction; facial expressions; social robotics. 1.

INTRODUCTION Advances in humanoids technology have paved the way for broader possible applications of humanoid robots in human‘s life. Humanoids are envisaged to serve humans at work as well as in personal life such as supporting the elderly [1]. Thus humanoids in the future would not only operate as machines but would also be capable of interacting with humans in friendly and emotional manner toward effective and affective communications. To instill emotional components into human-robot interactions, a humanoid should be capable of portraying its emotional state, particularly by possessing an emotionally expressive head. This has become the attention of our research in Project AMIR which is to develop emotionally expressive humanoid head robots. Since 2006, three humanoid head prototypes have been developed starting with AMIR-I [2], followed by AMIR-II [3] and currently the latest AMIR-III (see Fig. 1) as to be presented in this paper. Developing a humanoid head is a not a mere engineering challenge, instead knowledge from other fields such as computer science, psychology, and artistry is required in this effort to build such a socially interactive humanoid head. The appearance aspect of a humanoid head is indeed a subjective matter, yet the end results would consequently influence human‘s impression toward the robot itself. From anthropomorphism (human-likeness) scale, a humanoid head could be of highly realistic human look with flexible skins and hair. Examples of this case are Hiroshi Ishiguro‘s Geminoid HI-1 [4] or Hanson Robotics‘s Einstein Head [5]. Conversely, a humanoid head might as well possess an abstract or iconic look like Nexi MDS of MIT Media Lab [6] or Flobi of Bielefeld University [7]. Other robotic heads such as iCat [8] and Leonardo of MIT [9] exhibit another type of appearance theme inspired by animal or imaginary creature.

760


Figure 1: AMIR-III humanoid head

The subject of humanoid head anthropomorphism and its influence to human perception has long been hypothesized and discussed by many researchers. For example, Mori‘s Uncanny Valley theory [10] argued that if a humanoid head robot appears very much similar to human head, consequently it would induce uncanny feeling among those humans observing it. However, some robot designers put their efforts to create and test hyper-realistic humanoid head robots as mentioned above to dismiss such theory. While both contrasting hypotheses hold their own arguments, we tend to adapt Mori‘s theory to entirely avoid the suggested anthropomorphic pitfalls. Therefore, a fixed iconic face mask design would serve our purpose instead of the use of hair and flexible skins which characterize hyper-realistic appearance. In terms of mouth design, some humanoid heads do not possess any as in the case of Meka S1 and iCub humanoid heads. Contribution of other facial cues toward expressiveness might be highly emphasized as compared to mouth influence. However, according to [11], the mouth region is found to be more effective in conveying the emotions of the facial expressions than the eye region. Such statement implies that mouth design should not be underemphasized or at worse being omitted at all in designing an expressive humanoid head. On the other hand, fixed head shell implementation would raise the issue on how to produce different mouth expressions which could easily be performed by flexible skins. Several approaches have been demonstrated by other iconic humanoid heads. For example, Nexi possesses a 2-degree-of-freedom (DOF) lower mouth jaw capable of moving up and down as well as back and forth [6], whereas Flobi has flexible lips coupled with magnetic control points behind its mask [7]. The magnets are attached to sliding joints which will be moved accordingly to produce featured lip shapes. Both unique mouth designs not only form part of their emotional displays, but as well provide unique characters to the robots themselves. Thus, our AMIR-III mouth is designed for expressiveness in its own distinct way. 2.

DESIGN The kinematic configuration of AMIR-III is as illustrated in Figure 2. In total there are 18 DOFs within AMIR-III in which 4 DOFs are at its eyebrow part, 4 DOFs (eyelid part), 4 DOFs (eye part), 3 DOFs (mouth part) and 3 DOFs (neck part). All joints except the neck joints are actuated by Dynamixel AX-12+ servos. They are connected to each other in a daisy chain network. The main advantage of using Dynamixel servos is their sensory feedbacks in terms of position and speed. Three types of mechanisms are employed in AMIR-III mechanical system. The first one is the common direct actuation by the servo onto its load as implemented for the neck, eye and brow parts. Secondly, four-bar mechanisms are applied for tilting and rolling of the eyelids (double-rocker) as well as for jaw movements (double-crank). Third, the respective servos rotate the right and left lips by pushing and pulling the transmission cables through their tubes. The summary of AMIR-III specification is provided in Table 1 below.

761


2.1

Neck and Vision System The neck part consists of 3-DOF pan-tilt-roll joints in serial configuration. To support its heavy head particularly during neck tilting and lateral bending, high torque actuators are required. Thus we select Dynamixel RX-64 servos as our AMIR-III neck actuators which can sustain up to 64 kgf-cm holding torque. For its eyes, two Microsoft LifeCam Cinema USB cameras are embedded inside the eyeballs to provide stereo vision capability to the robot. Each eye has two DOFs for pan and tilt movements. The operating ranges of both joints are 30 degrees and 25 degrees respectively. Even though the cameras are capable of capturing high definition videos, the video resolution is set to 320 by 240 pixels format to minimize memory allocation for video streaming while maintaining sufficient details for image processing.

Figure 2: Kinematic configuration of AMIR-III

Table 1: Summary of AMIR-III specification Characteristic

Specification

Size Head breadth Head height Weight Degree of Freedom Eyebrow Eyelid Eye Mouth Neck Total DOF Neck operating ranges Neck flexion Neck extension Neck left-right rotation Eye field-of-view (FOV) ranges Horizontal FOV Vertical FOV

762

25.4 cm 24.0 cm 2 kg 4 4 4 3 3 18 35 degrees 30 degrees ±100 degrees 130 degrees 90 degrees


2.2

Face Casing To enhance the attractiveness of AMIR-III, its face casing is designed toward child-like cartoonish look characterized by large, round head and eyes as well as round cheeks. Based on anthropomorphic data by Ahlstrom and Longo [12], selected adult head characteristics i.e. head breadth, head height and eye-to-eye distance of the 99th percentile head are given as 16.5 cm, 20.9 cm and 7.4 cm respectively. Thus, we design our AMIR-III head to be larger and more rounded than the abovementioned size as shown in Figure 3(a). The face casing is designed using SolidWorks software, printed as a 3D part through rapid prototyping process and then painted with aerosol spray paint to produce its current look. To analyse the resultant facial expressions, 11 action units (AUs) are assigned to specific face points as shown in Figure 3(b). Such AU implementation is inspired by Ekman‘s Facial Action Coding System [13].

25.4 cm 8.9 cm 24.0 cm

(a)

(b)

Figure 3: (a) Face casing of AMIR-III with head height, head breadth and eye-to-eye distance (b) AU positions on AMIR-III face casing 2.3

Robot Programming A Pentium 4 PC with 1 GB of RAM is used to host and run control program of AMIR-III. Since AX-12+ and RX-64 servos are networked based on different protocols, i.e. RS-232 and RS-485, two separate USB2Dynamixel interfacing circuits are used to connect the networked servos to the PC. A Matlab-based GUI program is written to manually control the robot operation such as to display emotional expressions. Such program is an upgraded version of the one used for previous AMIR-II prototype [6].

3.

RESULTS AND DISCUSSIONS The range of facial expressions capable to be portrayed by AMIR-III humanoid head is as shown in Figure 4 below. Those basic expressions are neutral, happy, sad, angry and surprised. Generally, each expression could be easily identified except for the happy expression which might be perceived as a sad face instead. Further study on its expression recognisability will be conducted as part of our next development works. The performances of AMIR-III mechanical system are evaluated in terms of positioning accuracy for its servos and AUs. The angular displacements of each individual servo are measured at its minimum and maximum operating ranges. In Table 2, the results are displayed as percentage differences between the desired and actual positionings. The data for the neck roll joint for lateral bending is unavailable in the table since at this moment, the neck part is currently being set to operate as a 2-DOF neck. In general, during each expression, most joints demonstrate acceptable accuracies of less than five percent diversion from the desired displacements. However, the accuracy discrepancy for the mouth joints is slightly higher which are possibly contributed by its own design. First, the bent part of the tubes causes irregular friction for cable movement within the tubes thus affecting the angular displacement of the side lips. Secondly, the jaw and the side lips are operating in the same tight workspace, thus surface contacts are inevitable and more friction occurs.

763


(a) Neutral (b) Happy (c) Surprised (d) Angry (e) Sad Figure 4: Five basic facial expressions portrayed by AMIR-III humanoid head.

The AU positioning accuracies are given in Table 3 based on different facial expressions. Relatively good accuracy of less than two percent deviation is achieved by all AU positionings including the mouth joints. This might be due to the positioning of all AUs being set within the operating ranges rather than exactly at the minimum or maximum joint limits. Table 2: Angular Positioning accuracy of AMIR-III servos in terms of percentage deviations from desired displacements Mean % Difference of Accuracy

Motor ID

Joint Name

1

At Min. Operating Range

At Max. Operating Range

Neck Pan

0.82

0.39

2

Neck Tilt

2.54

3.79

4

Jaw

7.25

2.34

5

Right Lip

5.71

14.29

6

Left Lip

6.43

9.29

7

Right Eye Pan

1.37

1.37

8

Right Eye Tilt

1.56

0.39

9

Left Eye Pan

1.53

2.34

10

Left Eye Tilt

2.79

0.32

11

Right Eyelid Roll

1.10

0.50

12

Right Eyelid Tilt

3.30

0.36

13

Left Eyelid Roll

1.10

1.89

14

Left Eyelid Tilt

3.39

0.24

15

Right Eyebrow Tilt

0.83

2.83

16

Right Eyebrow Roll

1.56

1.56

17

Left Eyebrow Tilt

0.12

2.33

18

Left Eyebrow Roll

0.78

0.80

764


Table 3: Positioning accuracy of AMIR-III AUs in terms of percentage deviations from desired displacements Percentage Difference of AU Accuracy

Servo Joint Name

Happy

Sad

Angry

Surprised

4.

Neutral

ID

4

Jaw

1.00

0.76

0.89

0.85

0.85

5

Right Lip

0.22

0.49

0.02

0.02

0.74

6

Left Lip

0.77

0.67

1.63

1.76

0.38

11

Right Eyelid Roll

0.40

0.06

0.06

0.43

0.00

12

Right Eyelid Tilt

0.06

0.06

0.03

0.05

0.56

13

Left Eyelid Roll

0.00

0.39

0.39

0.06

0.20

14

Left Eyelid Tilt

0.70

0.69

0.54

0.54

0.45

15

Right Eyebrow Tilt

0.03

0.03

0.04

0.04

0.45

16

Right Eyebrow Roll

0.20

0.03

0.66

0.08

0.28

17

Left Eyebrow Tilt

0.36

0.36

0.33

0.33

0.00

18

Left Eyebrow Roll

0.39

0.39

0.08

0.39

0.09

CONCLUSIONS AND FUTURE WORK AMIR-III humanoid head is a work in progress. Thus far its mechanical system has been developed and tested in which the accuracy measurement of the AUs for AMIR-III humanoid head yields satisfactory results. Through its GUI program, the robot can manually be commanded to portray basic facial expressions. In our next development stage we would focus on developing its vision system which would be partially based previous works in AMIR-II [14]. Later, artificial intelligence theories are to be implemented into its program so that AMIR-III would be capable of deriving its own social responses during its interactions with humans.

5.

ACKNOWLEDGEMENTS This research is supported by the IIUM under grant no. FRGS-0207-47.

6.

REFERENCES [1] Y.-H. Wu, C. Fassert, A.-S. Rigaud, ―Designing robots for the elderly: Appearance issue and beyond,‖ in Archives of Gerontology and Geriatrics, Feb. 2011, [Online]. Available: http://dx.doi.org/10.1016/j.archger.2011.02.003 [2] A.A. Shafie, M.N. Kasyfi, N.I. Taufiq, ―Humanoid robot head,‖ presented at the 3rd Intl. Conf. on Mechatronics (ICOM 2008), Kuala Lumpur, Malaysia, Dec. 2008. [3] A.A. Shafie, M.F. Alias, N.K. Rashid, ―Graphical user interface for Humanoid Head AMIR-II,‖ presented at the IEEE 3rd Intl. Conf. on Computer and Communications Engineering (ICCCE 2010), Kuala Lumpur, Malaysia, May 2010. [Online]. Available: http://dx.doi.org/10.1109/ICCCE.2010.5556758 [4] K. Ogawa, C. Bartneck, D. Sakamoto, T. Kanda, T. Ono, H. Ishiguro, ―Can an android persuade you?― in Proceedings of the IEEE 18th Intl. Symposium on Robot and Human Interactive Communication

765


(RO-MAN 2009), 2009, pp. 516-521. [Online]. Available: http://dx.doi.org/10.1109/ROMAN.2009.5326352 [5] D. Hanson, A. Olney, I.A. Pereira, M. Zielke, ―Upending the uncanny valley,‖ in Proceedings of the 20th Ntl. Conf. on Artificial Intelligence (AAAI 2005), vol. 4, 2005. [Online]. Available: http://www.aaai.org/Papers/Workshops/2005/WS-05-11/WS05-11-005.pdf [6] M.M. Lab, MDS: Head & Face, 2008. [Online]. Available: http://robotic.media.mit.edu/projects/robots/mds/headface/headface.html [7] I. Lutkebohle, et al.,‖The Bielefeld Anthropomorphic Robot Head ‗Flobi‘,‖ presented at the IEEE Intl. Conf. on Robotics and Automation, Anchorage, Alaska, May 2010. [Online]. Available: http://aiweb.techfak.uni-bielefeld.de/files/bielefeld-head.pdf [8] A.V. Breemen, X. Yan, B. Meerbeek, ―iCat: An animated user-interface robot with personality,‖ presented at the 4th Intl. Joint Conf. Autonomous Agents and Multiagent Systems (AAMAS 2005), Utrecht, Netherlands, July 2005. [Online]. Available: http://www.grapa.cs.huji.ac.il/course/2005/aisemin/articles2006/docs/ded31_143.pdf [9] C. Breazeal, G. Hoffman, A. Lockerd, ―Teaching and working with robots as a collaboration,‖ in Proceedings of the 3rd Intl. Joint Conf. on Autonomous Agents and Multiagent Systems (AAMAS 2004), pp. 1030-1037, July 2004. [Online]. Available: http://dx.doi.org/10.1109/AAMAS.2004.242646 [10] C. Bartneck, T. Kanda, H. Ishiguro, N. Hagita, "Is the uncanny valley an uncanny cliff?," in Proceedings of the IEEE 16th Intl. Symposium on Robot and Human interactive Communication (RO-MAN 2007), 2007, pp. 368-373, Aug. 2007. [Online]. Available: http://dx.doi.org/10.1109/ROMAN.2007.4415111 [11] Tomoko Koda, Z. Ruttkay, Y. Nakagawa, K. Tabuchi, ―Cross-cultural study on facial regions as cues to recognize emotions of virtual agents,‖ in Lecture Notes in Computer Science, Springer, vol. 6259/2010, pp. 16-27, 2010. [Online]. Available: http://dx.doi.org/10.1007/978-3-642-17184-0_2 [12] V. Ahlstrom, K. Longo, (2003). Human Factors Design Standard (HF-STD-001). Atlantic City International Airport, NJ: Federal Aviation Administration William J. Hughes Technical Center. [Online]. Available: http://hf.tc.faa.gov/hfds/hfds_pdfs/Ch14_Anthropometry_and_biomechanics_Oct2009.pdf [13] T. Wu, N.J. Butko, P. Ruvulo, M.S. Bartlett, J.R. Movellan, ‖Learning to make facial expressions,‖ presented at the IEEE 8th Intl. Conf. on Development and Learning, Shanghai, China, June 2009. [Online]. Available: http://dx.doi.org/10.1109/DEVLRN.2009.5175536 [14] A. Iqbal, A.A. Shafie, M.R. Khan, ―Visual tracking and servoing of human face for robotic head AmirII,‖ presented at the IEEE 3rd Intl. Conf. on Computer and Communication Engineering, Kuala Lumpur, Malaysia, May 2010. [Online]. Available: http://dx.doi.org/10.1109/ICCCE.2010.5556851

766


A POTENTIAL FIELD METHOD FOR AUTONOMOUS LUNAR ROVER NAVIGATION IN 3D TERRAIN Parth Nanadikar1, Rahul Shome2 and Ashish Dutta1 1

2

Department of Mechanical engineering, Indian Institute of Technology Kanpur, Kanpur 208016, India. (e-mail: parthnan/[email protected])

Department of Computer Science and Engineering, National Institute of Technology Durgapur, West Bengal 713209, India.

ABSTRACT The development of planetary rovers is essential for the success of planetary missions. This paper discusses an algorithm based on the potential field method for navigation of a rover in an unknown 3D terrain containing obstacles. A 3D map of the terrain is generated using a structured light system, and the terrain is then divided into square grids having gradients. Assigning positive and negative gradients to the goal, obstacles, rover and grids we generate a potential field function. Using this function the rover finds the best path to reach a goal point. Unlike potential field functions in 2D this method works in 3D and also considers the rover kinematics. Keywords: lunar rover, 3D path planning, potential field. 1.

INTRODUCTION In the last decade, with the launch of new planetary missions there are further developments being made in planetary rovers. The developments of new mechanical designs and software algorithms are aimed at enhancing the exploration capabilities of planetary rovers. The New generation planetary rovers make possible the performance of more tedious scientific experiments in situ on other planets. The rovers with rocker bogie mechanism are tried and tested designs, their success has been established by several Mars missions. This paper deals with development of path planning algorithm for the lunar rover, taking into consideration the kinematic constrains on lunar rover due to its mechanical design and 3D nature of ground profile of the terrain. The surface of the moon has boulders acting as obstacles to the rover and the regolith with certain gradients which make maneuvering of the rover susceptible to obstruction or the over tipping. Hence there is a need to develop a path planning algorithm which will take care of kinematic constraints of the moon rover. The moon rover obtains the 3D local map from structured light based laser source system. The 3D map is converted into a 2D grid value map where each grid has a gradient and/or obstacles, based on which the path is generated. In the simulation developed the 2D grid value map with obstacles is manually entered to the simulator. Earlier potential field approaches considered flat terrains in 2D with obstacles for path planning. We propose a newer method in which the gradient information of the terrain as well as the kinematic constraints of the rover are also incorporated to find the best path. The potential field methods are widely studied and implemented methods in robot path planning. Since O. Khatib [1] suggested the theory of potential field first time in 1986. Khatib proposed a method of real time obstacle avoidance approach for manipulators and mobile robots based on the artificial potential field concept. Much work has been done in the field since then. McFetridge and Ibrahim[2] have summarized the old potential field approach and the new hybrid adaptive Fuzzy – Potential field approach.Volpe and Khosle [3] suggested the potential function based on superquadrics which closely models a large class of object shapes. Shrikant Parakh et al. [4] discuss the motion of the rover on uneven terrain that can take into account the loss of contact of wheel with the ground. Their approach can be incorporated in the algorithm for field trials. Takeshi Ohki et al. [5,6] have demonstrated, in simulations and in actual field trials, the path planning method for autonomous robots considering the instability of attitude maneuvers on rough terrains. They generated more than one path using general graph search method. This paper also tries to address the same concern of instability of the mobile robot during attitude maneuvers, but the path is generated through potential field methods. But most of them discuss about methods of overcoming the local minima‘s once the robot is trapped in it. These methods put displacement criteria to identify the local minima problem. In these methods the rover is said to be trapped in a local minima if at any iteration its displacement is below a

767


certain critical value. We propose a method where we identify the local minima in advance and try to circumvent it without getting trapped in it. But most of these methods are for path generation in 2D. This paper describes the way to generate a path in 3D. The proposed method is discussed in section 2 in detail. The equations and the values of constants selected are shown in this section. A flowchart of the simulation is shown in this section. Section 3 discuses implementation and evaluation of the proposed path planning algorithm. The assumptions made for the purpose of simulation are discussed in this section. Section 4 discuses the results of the simulation and the path extracted are shown for each result. Section 5 discusses the conclusion. 2.

PROPOSED METHOD In this section path generation is described using the potential field methods with additional algorithms described in subsequent paragraphs. It is assumed that the rover is inscribed in a circle within a radius of 391 mm and this circle is called rover circle (figure.1). Circumference of rover circle and the surface of obstacles have like electrical charges. Hence they repel each other with force which is inversely proportional to the square of the shortest distance between them. The rover circle and the target have unlike charges hence they attract each other. The gradient cells in vicinity of the rover exerts a repulsive force depending upon their gradient. Each grid is of size 20x20mm. In each iteration the vector summation of all forces acting on the rover is calculated. We treat the rover path planning as problem as one in which the rover moves very slowly so only rover velocity is considered during simulation.

Fig. 1: The basic dimensions of the rover and the rover circle.

In the simulation there is a virtual positive electric charge on the rover of 10 units. The target is given a virtual negative electric charge of -1000 unit. The point on the obstacle surface, which is closest to the rover, is given a virtual positive charge of +1 unit. The value of the electrostatic constant is taken as 25. The charges on the rover and the target were so chosen that the velocity of the rover at the starting point is not 1 less than 19.5 mm/ sec. The charge on the obstacle is taken as of the charge on the rover for our case. 10

The force acting between two electrical charges in the potential field method is given by : F=

k∗𝑘 1 ∗𝑘 2 d²

Where,

768


K is the electrostatic constant 𝑘1 is the first charge 𝑘2 is the second charge d is the distance between the two charges. Hence the virtual electrostatic attractive force acting in between the rover and the target is given by Fatt =

25∗10∗1000 d²

And the virtual electrostatic repulsive force acting between the rover and the obstacles is given by Frep =

25∗10∗1 d²

A particular trap situation is shown in figure 2. We neglect the ground profile for this example. The target is exactly behind the Obstacle and the obstacle surface facing the rover circle is exactly perpendicular to the line joining the rover circle centre and the target. Figure 2(b) shows the direction of net resultant force acting at each point on the map, if rover is placed at these points. The small dot like circles indicate the centre of rover circle and the normalized line originating from these centres indicate the direction of resultant force acting at those points. The figure 2(b) shows the trapped robot as a result at the local minima, as the attractive and repulsive forces are in line. Here the obstacle surface ‗ab‘ which causes the trap situation is seen as a line view in top view. The surface has two end vertices, ‗a‘ and ‗b‘. The algorithm checks which of these two end vertices are closer to the rover at every iteration. The vertex which is closer (in the case of figure 2(c) it is a) is joined to the centre of rover circle by a straight line. Then the reflection of the force of attraction is taken about this line. When the reflected force of attraction is used to calculate the resultant force it guides the rover away from the local minima. The figure 2(c) shows the method of reflection of forces applied to overcome such trap situation. In figure 2(d) the rover is able to avoid the local minima without getting trapped into it.

(a)

(b)

(c)

(d)

Fig. 2(a-d) . (a) The direction of net resultant force acting at each point in the map. (b) As a result, the robot is trapped at the local minima (c) The method of reflection of forces implied to overcome such trap situation (d) The final path.

769


The rover should be guarded from over tipping. To achieve this, it is desirable that the path generated passes as far as possible through the cells which have lower radial gradients. In the figure. 3 it is shown how the rover will look at two different gradients.

Fig 3: Maximum longitudinal and lateral slopes allowed in each grid considering stability The map is now divided into cells. The resolution is decided by the laser scanning capacity of the rover. The structured light used is capable of generating a grid of 2 square cm up to 2m distance. The algorithm has adaptability to different resolutions, but higher resolution means higher number of cells hence more computational time. Each cell has gradient value in radial outward direction from the rover. A cell exerts the repulsive force on the rover circle from its centre which is given by, ∗ Frep-cell = Kcg*Kad∗ 𝐹𝑎𝑡𝑡 (1)

Where, F rep-cell is repulsive force due to a single grid cell. Kcg is cell gradient factor for the particular cell Kad is the angle distance factor for the particular cell ∗ 𝐹𝑎𝑡𝑡 is the maximum force of repulsion acting on the rover from the centre of the gradient cell. This factor is proportional to, Fatt by relation :

∗ 𝐹𝑎𝑡𝑡 = k‘ * Fatt ; 0 < k‘ < 1

(2)

Fatt is force of attraction by target acting on The rover if it were placed at the centre of the cell The different factors are calculated as follows,

Kcg =

gradient / tan (max slope allowed in degree )

;

in case of the rover maximum allowed slope is taken 35°

770


hence Kcg becomes, Kcg =

gradient

(3)

0.7002

Kad = 0.75 cellwidth

if 20°=< cellcentreangle < 138° and if 150 mm < cellcentrerobodistance