Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013
Optimization of Inventory Pick-up Time in a Server Manufacturing Environment S Chivukula1, C Saha1, BR Schleich1, N Nagarur1, and B Wassink2 1
Systems Science and Industrial Engineering, State University of New York at Binghamton, Binghamton, New York 13902 2
Information Delivery Management System, IBM Corporation, Rochester, Minnesota 55901 Corresponding author's Email:
[email protected]
Abstract: In the wake of globalization, companies are emphasizing on being cost competent, which could also be seen in power server manufacturing industries featured by leading edge technologies. These industries have expensive and large variety of components involving high inventory storage and customer service costs. The complexities resulting from extensive quality assurance tests, multi-layer suppliers with varying supply lead times, and complex product configurations affect the entire supply chain. In addition, the power server manufacturing facility considered for this study has been facing a major challenge due to inefficient raw material pick-up process. A meta-heuristic optimization technique, Genetic Algorithm technique, capable of solving small (to medium) size instances is proposed in this study, to facilitate effective planning of zoning and routing for raw material movement. Prior data is analyzed to understand the complexity of the problem, and to study the floor plan using a process map to develop an optimization model to minimize inventory pick-up time. The proposed model can fit into a decision support system for operational planning in this server manufacturing environment and would also be beneficial for similar discrete manufacturing systems. Keywords: Inventory pick-up, Meta-heuristic Optimization, Genetic Algorithm, Server Manufacturing, Decision Support Systems, Discrete Manufacturing Systems
1. Introduction 1.1 Background Continuously growing market competition is forcing companies to be (cost) efficient, which requires continuous improvement in the design and operation strategies of a manufacturing shop floor. With the recent advancements in manufacturing strategies and supply chain management, philosophies such as Just-In-Time (JIT) and Lean Management strategies are facing more challenges in streamlining a company’s processes (De-Koster et al., 2007). In the case of Build-ToOrder (BTO) (or Make-To-Order) scenarios with large product variety and components, parts require effective and efficient floor plans for Raw Material Inventory Management (RMIM) to meet customer demand on time. RMIM includes current inventory level monitoring, raw material replenishment and operation strategies, and many more. To meet customer demand on time, it is very important to pay attention to the storage location assignment for raw material or the items need to be assembled in the job-shop. For an efficient production plant, raw material storage location assignment should be supportive to the production plant. Therefore, efforts need to be taken to focus on the batching, routing, sorting, and zoning of component items in the job-shop to minimize the time required for raw material pick-ups. This helps to reduce total production lead time for a product and will help in meeting the production takt-time.
1.2 Problem Statement For this research, a BTO scenario of a computer hardware manufacturing firm, which aims to release an order within 4 days, is considered. This company has a high volume – high variety product mix. Company uses an RFID based software for raw material tracking and storage. Raw material inventories are divided into three categories: frames, kits, and tools. Frames are stored on the manufacturing floor. Kits include components like memories, modems, DVDs, LAN cards, printed circuit boards, etc. These items are kept in a different storage area. Tools such as screws, wires, etc. are stored on the production floor and workers collect them when required. These tool items are replenished once in a week. The company follows Kanban process for pulling components from kits zone according to order requirement and sending them to fabrication zone on the manufacturing floor. Materials are stored at the kits zone according to part types. When an order is received, the floor management software identifies the parts required to fulfill the order and prepares a list incorporating part names and their locations in the aisles. Currently, the company is facing difficulties as the zoning of components and routing ISBN: 97819384960-1-1
317
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 are not efficient enough due to multiple raw material storage locations. Therefore, to meet the component requirement of a single product, multiple trips to the storage areas and production floor are undertaken, which is resulting in wastage of resources (i.e., time and labor). This problem becomes more acute during the end of a month or any particular quarter when high customer demands are experienced.
1.3 Objective The objective of this study is to develop an optimization model to minimize the raw material pick-up time, by effective planning of zoning and routing for raw material movement to the job-shop. For this, efforts are taken to analyze the prior data to understand the complexity of problem, study the floor plan, and develop the optimization model. The rest of this paper is organized as follows: Literature review on server manufacturing systems, inventory order picking, and Genetic Algorithm for order picking processes are described in Section 2. The structural design and process map are discussed in Section 3. Experimental results are presented in Section 4. Finally, conclusions and future directions of the research are illustrated in Section 5.
2. Literature Review The challenges faced by companies in recent days are mainly arising from a single metric: customer satisfaction. Other process metrics such as delivery times, low product costs, and product qualities also support the metric of customer satisfaction. To ensure customer satisfaction, an effective supply chain network design is crucial. The major focus of this study is the power server manufacturing supply chain with focus on inventory management phase. Rest of this study is designed as follows: a brief discussion on supply chain management, followed by inventory and manufacturing phases of server manufacturing environment. Finally, discussion is extended on to the design and control of warehouse inventory order pickup process.
2.1 Supply Chain Management A Supply Chain Network (SCN) comprises of a series of processes and echelons that start from suppliers and end with customers while material, information, and cash flows in both directions throughout the process. A Manufacturing SCN includes suppliers, factories, subcontractors, warehouses, distribution centers, retailers, etc., through which raw materials are acquired, transformed, produced, and delivered to the end customers. Supply Chain Management (SCM) encompasses the planning and management of all activities involved in sourcing, procurement, conversion, and logistics management activities. Figure 1 shows basic stages in a supply chain with product, fund, and information flows. It can be noticed that the products flow from upstream to downstream and funds flow from downstream to upstream, while information flows in both directions. Due to continuous advancements in communication technology, a customer desires increased efficiency from any supply chain. This efficiency can be achieved by proper implementation of technologies for better integration and collaboration among the participants of the whole supply chain. Several techniques can be found in the existing literature to help in making decisions for a SCM system of a company. The decision making in SCM can be divided into three main categories: strategic, tactical, and operational (Teigen, 1997). There are four major areas: location, production/manufacturing, inventory, and distributions, where both strategic and operational decision support elements are required. Supply chain models are developed mostly by applying network designs, rough-cut methods, analytical methods, and Discrete Event Simulation (DES) models. Among them, DES model is one of the techniques which gained popularity in recent decades with the advancement of computer technologies (Beamon, 1998). Previously, SC models were developed for strategic or long term decision making. However, in recent decades, wide range of applications in SCM can be found in taking technical and operational decisions. It can also be observed that DES technique is applied to initiate bonding between operational and business models (Ramakrishnan, 2008).
2.2 Simulation Approach for Server Manufacturing Environment “Server manufacturing is characterized by high customer expectations, short turnaround times, wide variety of product base and long manufacturing cycle times, contributing to considerable supply chain complexities. Moreover, it also has unpredictable demand coupled with high fall-out rates and high capital investments” (Lendermann, 2006). To incorporate these features, and analyze the systems, simulation models can be used. Distributed simulation approach refers to the technologies which allow a simulation program or model to execute on a computing system, which contains multiple 318
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 processors in a communication network. This approach is tremendously popular in modeling and simulation domain where enormous computational efforts are required. The applications of distributed simulation can be found in server manufacturing supply chain network where large amount of information needs to be computed to obtain results (Ramakrishnan et al., 2002). To solve a problem of complex interdependence, multiple alternatives and policies, performance tradeoffs in a supply chain can be successfully assessed by DES. Many server manufacturing companies have their own SC simulator and this software is a combination of simulation and optimization functions, to model and resolve its own supply chain issues such as site location, replenishment, manufacturing, transportation, stocking levels, floor management, lead times, and customer response (Ramakrishnan et al., 2008). Complexities in server manufacturing are already discussed. However, for a high end server manufacturer these complexities are even more complex due to the two stage fabrication/fulfillment process, multi-echelon bills-of-materials, server box configurations, part tests with random yields, lead times, and so on. The model that is proposed by Chen et al. has ‘Days of Supply’ profile which helps to monitor daily production plans, identify optimization opportunities, and action plans when inventory level goes below a threshold level. In addition, it enables ‘what-if’ analysis for performance improvement. The input data required for this system is obtained from DB2 database servers. End product demand pattern, configuration features ratios, component test time distributions, BOM’s, and simulation output report which contains customer serviceability, inventory level at store and on the way are the data types required for this simulation tool. The simulation model is developed through IBM discrete event simulation library. Due to complex supply chain, typical commercial software like Arena, Simprocss, and Promodel those can incorporate process delays, queues, branching, splitting, merging, and resources are incapable of designing the system. Although, IBM Supply Chain Analyzer (SCA) has the capability to model and optimize these processes, it can only work as a standalone desktop application. Therefore, the authors applied Java Event delegation model (Object Oriented Programming) aligning with the SCA. The simulation model discussed above can help the inventory planner for decision making on inventory level as well as analyze ‘what-if’ situation for process improvement plan (Chen et al., 2006).
2.3 Inventory Order Picking Strategies A task performed in the warehouse to retrieve material (inventory) from pre-determined or dynamic storage locations to meet single or multiple customer orders is known as an order picking process. These tasks are needed to be performed because incoming materials are stored in large quantities (mostly based on economic order quantities) and internal/external orders are usually requesting a small portion of received quantities. When tasks of order picking are under performed i.e., prolonged delivery times, incorrect shipments, etc., they result in dissatisfaction of customers and increased costs to the firm. These will negatively affect competitiveness of the storage area in the facility (De-Koster et al., 2007). Order picking has been recognized as highly labor intensive and also an expensive process in warehouse operations. It is estimated that order picking process solely contributes to 55% of the total warehouse operating expenditure. Therefore, any underperformance will result in high costs for the warehouse as well as for entire supply chain. Hence, there is a need for designing the process robustly and in an optimal manner (De-Koster et al., 2007). De-Koster et al. (2007) provided a literature overview on crucial decision problems in designing and controlling manual order picking processes. Similar to many other repetitive material handling tasks performed in a firm, order picking involves a great extent of manual labor. These systems can be categorized into picker to parts systems (where order pickers go to warehouses inside facilities and pull requested material) and parts-to-picker systems (where automated storage and retrieval systems bring parts to the order pickers waiting for material). For systems of this kind, planning problems can be categorized into three groups, at operational level: assigning incoming items to storage locations in warehouses; grouping customer orders into batches; and routing order pickers via warehouses (Henn and Wascher, 2012). Performance of order picking systems can be estimated based on batching, picking sequence, storage policy, zoning, layout design, picking equipment, design of picking information, etc. Many researchers have focused on investigating combined effects of several above mentioned factors to study the performance of order picking systems. From those researches, numerous picking, routing and storing methods have been validated to decide what combination of above factors is most efficient in terms of minimizing picking time. It was concluded that batching of picking orders leads to largest enhancement in picker travel distance performance, particularly when there are more numbers of small sized orders. Furthermore, enhanced storage policy will achieve considerably high improvement, being insensitive to batch size (Henn, 2012). Efficacy of the order-picking process will mainly depend on storage systems, locations of storage areas, and control mechanisms to retrieve the inventory. Researchers have proposed four different approaches to enhance efficiency of the order-picking process to optimize the order picking time or travel distances inside the warehouse. The first approach reduces travel time by planning order-picking routes. Second approach separates warehouses into zones and assigning zones to specific order pickers. Third approach stores inventories in storage locations selected based on the best usage of shelves. There exists a crucial relationship between storage assignment rule and the routing methods. Fourth and final approach 319
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 batches the orders and thus reduces the travel distance for order picking because all the orders assigned to the same batch will be picked at once. Researchers till now have mainly focused on storage assignments, batching orders, and order routings (Bukchin et al., 2012). It is widely believed that local optimization can be achieved if the focus of research is on layout design, storage assignment, and order batching, along with picker routing strategies (Hsu et al., 2005). Currently, Order picking is the most inefficient process at this server manufacturing company. Order picking is the process of retrieving several products for a customer order where customer can be internal or external. In case of server manufacturing, a very similar process of order processing to meet internal customers’ order to fulfill external customer demand, is inventory picking. As ordered servers are most of the time customized based on specific customer needs, raw materials or finished server components need to be picked before server manufacturing can start. Inventory picking is a very crucial process in server manufacturing and was identified by our stakeholder as the main concern, which should be analyzed and improved. In order to generate a better understanding of order picking and how it can be optimized, literature specific ton server manufacturing and order picking is reviewed. De Koster et al. (2007) carried out a very thorough literature review on order picking. Five factors identified as essential for optimal order picking are optimal layout design, routing methods, storage assignment methods, zoning and order batching. It was also found that 70 % of the total time of order picking is accounted to travel and search. Order picking can have many different objectives, the most common objectives for order picking in the literature are to minimize throughput time, use of equipment, labor, and space and maximize accessibility. Therefore, it is not unusual to work with multi-objective optimization models where all the illustrated objectives are associated with minimizing cost (De Koster et al., 2007). Order picking is a proved NP-hard problem; therefore, obtaining an optimal solution is very difficult. However, in order to obtain a close to optimal solution with adequate computational time heuristic algorithms can be used. The authors used a GA to minimize travelling distance and time by forming batches. Algorithms for batching orders to minimize travelling distances also have been used by Henn and Wascher (2012). This study also showed that order batching is essential to minimize costs. A Markov decision process for optimal decision policies has been proposed for dynamic order picking processes (Bukchin et al. 2012). This model aims to reduce picker overtime costs and order tardiness. The picker has two options in this model. He can either pick an order immediately after order arrival or he has the option to wait for more orders to come in. The proposed model showed a superior result over other “naïve” heuristic algorithms and therefore, this might be interesting for consideration in this study. Yu and De-Koster (2009) included a totally different factor, zoning, what only few existing articles incorporate in their models. Zoning is the practice to divide the warehouse or picking area into zones and assign picker to zones in which they exclusively pick their orders. This review on order picking provides a much better insight on identifying the objective of the proposed model and also reveals interesting features which can be included in the model configurations.
2.4 Genetic Algorithm Approach for Optimizing Order Picking Process Genetic Algorithm (GA) is a meta-heuristic algorithm introduced by inspiration from the concepts of natural and biological evolutions. Developing GAs as optimization algorithms involves usage of notions of population genetics and evolution theories. GA tends to optimize fitness of populations of elements by using techniques of recombination and mutations of genes. Two areas, encoding potential issues and defining objective function that should be optimized, should be addressed before applying GA evolutionary approach to any real time optimization problem. A chromosome, which is a solution, will be encoded as a string composed of numerous genes, also called as components. For this, initial population of chromosomes (solution) will be either generated or chosen randomly, based on few principles. Then, the algorithm will execute an investigation of measure of fitness of possible solutions. Optimization with the help of GA can be accomplished by first selecting pairs of chromosomes which have probabilities proportional to their fitness values and then matching them to new offspring. After matching chromosomes, which is also called crossover, small mutations will be incorporated into a new offspring. This is performed to substitute bad solutions with new ones depending on few determined strategies. The above mentioned chromosomes evolve from successive repetitions known as generations. The process of assessment, optimization, and replacing bad solutions with new ones will be iterated till a stopping criterion is satisfied. Advantage of using GA is that, production orders in real world environment can be responded to at a faster pace (Seyedrezaei et al., 2012). When items from warehouse are requested by internal or external customers in a company or facility where manual order picking process is in place, order pickers ride through the distribution warehouse and collect materials. To perform these tasks in an efficient way, it will be useful if customer orders are consolidated into single/fewer picking order(s) which are restricted in size. Henn and Wascher (2012), for their research work on order picking processes, considered an order batching problem which deals with methods of consolidation of a given set of customer orders into fewer number of picking orders to minimize the total time spent to collect all the items requested in all the picker tours. Henn and Wascher (2012) 320
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 suggested two different approaches that are based on Tabu Search technique. First of those is a classic Tabu Search, while second one is the Attribute-Based Hill Climber (ABHC) approach. By conducting a series of wide-ranging numerical experiments, the authors standardized the above mentioned techniques and benchmarked them against other existing solution methods. They also proved that their proposed methods are far more efficient than earlier methods proposed by other researchers, and claimed that their solutions allow more efficient tasks to be performed by distribution warehouses (Henn and Wascher, 2012). Ene and Ozturk (2012) worked on designing a new storage assignment and order picking system with the help of mathematical model and stochastic evolutionary optimization approach in an automotive industry. This was a two stage approach. During first stage, storage location assignment problem was addressed using class-based storage policy so that warehouse transmissions can be reduced. This was done using integer programming model. In the second stage, both batching as well as routing problems were considered so that travel cost in the warehouse operations can be reduced. For this work, an optimum solution was obtained using integer programming model. Since, more computational time will be required to solve an integer programming model, GA was developed by the authors to form optimal batches and routes for order picking. Results from this research work demonstrated that proposed approach with the aid of GA could be applied and incorporated into any type of warehouse layout in any automotive industry (Ene and Ozturk, 2012). Liang et al. (2007) worked on proposing a new model with capacity constraints and various objectives, based on nature of picking in an automated warehouse. Along with this, authors proposed a GA so as to address the issue of initial population depending on unachievable degree and work times. Simulation of the proposed model demonstrated that efficient solutions can be achieved with this approach. Bukchin et al. (2012) investigated the issue of batching orders in a dynamic and fixed scope environment to optimize order tardiness and overtime costs incurred due to pickers. This issue usually introduces trade-off between choosing a tour and picking the orders. Pickers will either have to wait for more orders to arrive or to go on a trip and pick orders by themselves. A picker waiting for orders can result in increased tardiness at the expense of less reaction time for dynamic and unexpected future orders. The authors used a Markov Decision Process (MDP) based approach to develop an optimal decision making policy. Since this optimal algorithm addresses issues raised by relatively small-scale concerns, a metaheuristic was developed by authors. This meta-heuristic was based on structure and properties of the optimal solutions. Computational results showed that proposed heuristic, MDP-H, was superior to immature heuristics used by researchers for several other works. When compared to optimal solution, MDP-H gave approximately optimal results for a slack of up to 40% (Bukchin et al., 2012). Hsieh and Huang (2001) developed two different heuristics (K-Means Batching – KMB and Self-Organization Map Batching – SOMB) to verify results obtained by simulation experiments. Both heuristics provided better results in terms of total travel distance and average picking vehicle utility, along with noticeable improvement in total CPU operation time. They also studied the whole performance of order picking systems amalgamating storage assignment, order batching, and picker routings to discover the optimal policy combinations under various types of orders. They also carried out sensitivity analysis to discriminate relative importance of different strategies to improve performance of process management. Seyedrezaei et al. (2012) presented a dynamic mathematical model for processing and storage of goods for order picking planning problem (OPP) in distribution centers. The authors worked on optimizing the number of accomplished orders with respect to the coefficients of each customer, meeting customers’ probabilistic requirements in the least possible time, and picking material to deliver to customers at the earliest so that material decomposing can be prevented. Authors’ used Lingo software to address small size real time problems, and proposed GA method to address the problems. Computational results demonstrated that GA proposed by the authors is more efficient than other techniques in searching near optimal solutions.
3. Methodology The following discussion about the structural design, system description using process map of server manufacturing production system, and order picking strategy will help to understand the server manufacturing process in detail along with the current order picking strategy.
3.1 Structural Design Figure 2 shows the structural design of the proposed study. Understanding the server manufacturing environment was the first step of this study. Then literature review and expert decisions were taken to identify the important factors and key performance measures of an order picking process. Next data were collected for 100 orders, which is a combination of 96
321
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 component parts. Process map of the manufacturing process was developed to fulfill an order to understand the importance of efficient order pick-up strategy by optimizing inventory storage location.
Understand server manufacturing environment
Review literature and discuss with experts to determine inputs and key performance measures
Collect data for inputs and process layout
Develop process map
Meta-heuristics optimization procedure: Genetic Algorithm
Result analysis
Figure 1. Structural design of the proposed study
3.2 System Description The server manufacturing environment under this study is a two-stage process known as fabrication-fulfillment strategy. Fabrication processes follow build-to-plan approach, which can be referred to as a push system. On the other hand, fulfillment processes follow build-to-order approach, which is known as pull system. Therefore, the manufacturing systems under consideration are a combination of both push and pull strategies. Figure 2 represents facility layout of the server manufacturing production system. The white blocks represent four different production areas. Blue boxes represent processes/activities, yellow boxes represent raw material inventory areas and silver rectangles represent work-in-process inventories. The company practices pull system to get raw material from suppliers whenever needed. Suppliers have their warehouse inside the company’s manufacturing area. Hence, raw material is vendor managed inventory. Whenever, raw material is pulled from suppliers’ warehouses, suppliers replenish those items and quantities in their warehouse locations. Sub-assembled items are kept in the kitting zone and sent to assembly area whenever there is a confirmed order to build a final assembly. Frames for the servers are stored in the frame zone and workers collect frame from this area for the confirmed orders. In the tools zone, screws, wires, drivers are stocked and these inventories are not monitored using any MRP system, but are replenished based on a fixed schedule every week. Once assembled and fulfilled, every customer order is either sent to testing or integration area depending on the product requirement. In the integration area, multiple frames are combined together in racks and the racks are then moved to the waiting area for testing. After testing, the items are sent to the clean & claim area for cosmetic refinement. Finally, all items are sent to packaging and shipment area for order fulfillment. Each processing area has multiple work stations and each station is capable of fulfilling all the activities required for that particular stage.
322
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013
Frame Zone
Rack Zone
Assembly
Waiting Area for Testing
Packaging & Shipment
Integration Clean & Claim Tools Zone
Tools Zone
Testing Tools Zone
Fabrication
Kitting Zone
Sub-assembly
Tools Zone
Tools Zone
Warehouse
Figure 2. Process map of server manufacturing production system
The fabrication process takes place in sub-assembly and fabrication areas. Here, based on the aggregate demand level and the plant’s component/parts capacity, PCB boards and processors are built and sent to kitting zone. When a confirmed order is received, items are released from kitting zone and will be directly sent to the assembly area. Based on the type of order, some components can be built in sub-assembly and fabrication zones and finally, those will be moved to the assembly area rather than to the kitting zone.
3.3 Order Picking Strategy In the assembly, integration, and testing areas, there are several work stations and each work-station is fully capable of assembling any type of server model. For a particular order, an order picker collects the frame from the frame zone and pre-processed items (i.e., PCB boards, CD-ROMs, hard disks, etc.) are delivered to the respective assembly work station following a Kanban pick-up process. After completing the assembly, the ordered item is either sent to testing area or to the integration area based on the requirement. In the integration area, several assembly items are integrated in a rack to produce larger or complicated servers (i.e., combination of multiple blades). After testing, they will be sent for packaging and are kept in the finished goods storage area. Sensitive components are kept in the knitting zone. When an order is placed, kitting report of that particular order will be generated at the kitting zone and this includes part name, quantity, inventory status, and location. An order picker collects all the parts in a box and places them in a Kanban. When the Kanban is full, it is sent to the assembly area with the help of a trolley or forklift. In the assembly area, each box is sent to the respective work station by the order picker.
323
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013 The bottle-neck of this order picking strategy is at the kitting zone where larger numbers of sensitive components are needed to be picked, combined, and sent to the assembly area. Therefore, efforts are needed to establish an efficient order picking policy for raw materials. To solve the order picker’s dilemma at kitting zone, Genetic Algorithm (GA) approach was used. First, possible component storage setups were generated, which is referred to as population pool and initial parentsolutions. A single solution is referred to as a chromosome and the single storage locations of the chromosomes are referred to as genes. The main objective function of this study is to minimize the order picker’s total travelling distance for all orders. New order picking solutions were generated by crossing over operation of two parent-solutions. Additionally, mutation operations were used to switch certain gene information within the child solutions. Multiple child solutions were originated from two parent solutions. In a GA approach, there is always a possibility to retain all the solutions or to replace them using different replacement strategies. For this study, a total replacement strategy, which reduces computational time significantly, was chosen. This strategy helps to reduce the solution pool to its original population size by only retaining the fittest ones.
4. Experimental Results GA to optimize the storing location for the order picking was written in Java programming language. The chromosomes used for the heuristic in the base case were coded with 96 genes as the model was dealing with 100 orders totaling 96 server components. For the crossover operation a single point crossover was used and mutation was coded as a gene location change. After new children solutions were generated from the parent solutions, a total replacement strategy was chosen. The base model was programmed to have a population of 100 initial solutions. For testing and evaluation purposes, the program was run for 1,000 iterations and this was repeated 10 times with different random seeds and averages of those ten runs for each performance measure were taken. Analyses were based on comparing performance of the GA by varying crossover and mutation probability, population size and number of child solutions per parent crossover. The objective was to minimize distance travelled for order pickers by optimizing the storage location of each component. The base case parameter setups for the GA are crossover probability 1, mutation probability 0.05, population size 100, and number of genes 30. Table 1 illustrates parameter settings and the different program outcomes.
Table 1. Experimental results analysis Crossover probability (prob.) Parameter setting Difference to base case (PS) (D2BC) 1 0 0.95 4.6% 0.9 8.3% 85 11.2%
Mutation prob.
Population size
Child solutions
PS
D2BC
PS
D2BC
PS
D2BC
0.1 0.05 0.015 0.001
5.4% 0 -6.4% -8.8%
100 200 50 20
0 -6.3% 9.7% 17.1%
10 8 6 4
-8.9% -5.4% 0.0% 14.6%
4.6% in the fourth row and second column indicate that the total distance travelled for all orders was 2.6% higher when decreasing the crossover probability from 100% to 95%.It is observed that the performance of GA is highly correlated with the parameter setting. A summary of parameter setting modification and performance outcome is illustrated in table 2.
Table 2. Summary table Parameter Modification Performance
Crossover probability Raise Decrease Improve Decline
Mutation probability Raise Decrease Decline Improve
324
Population Size Raise Decrease Improve Decline
Child solutions Raise Decrease Improve Decline
Proceedings of the 2nd Annual World Conference of the Society for Industrial And Systems Engineering Las Vegas, NV, USA November 5-7, 2013
5. Conclusion and Future Works In this study, order picking in the server manufacturing industry was analyzed and a heuristic method namely GA was used to optimize order picking process in the knitting zone. The objective of this research was to minimize the distance travelled by the order picker by finding a near optimal server component storage location setup. The proposed model based on GA is able to arrange component storage to minimize distance travelled and it is observed that the performance of the designed heuristic highly depends on the parameter settings. In future, the proposed model can be solved using other heuristic methods and research can be further extended towards a comparative study. Another possibility of extending this study is to minimize the travelling distance by rearranging the order of the components for pick up for each order which is a travelling salesman problem itself for each order.
6. References De-Koster, R., Le-Duc, T., and Roodbergen, K. J. (2007). Design and Control of Warehouse Order Picking: A literature Review. European Journal of Operational Research, 2007; 182(2): 481-501. Teigen, R. (1997). Information Flow in a Supply Chain Management System. Ph.D. Thesis, Dept. of Industrial Economics and Technology Management, Trondheim University, Sweden, 1997. Beamon, B. M. (1998). Supply Chain Design and Analysis: Models and Methods. International Journal of Production Economics, 1998; 55(3): 281-294. Ramakrishnan, S. (2008). A Simulation-based Framework for Integrating Operational Models with Business Process Models in A Global Business Environment. Ph.D. Thesis, Dept. of Systems Science and Industrial Engineering, Binghamton University, New York, 2008. Lendermann, P. (2006). About the Need for Distributed Simulation Technology for the Resolution of Real-world Manufacturing and Logistics Problems. Proc. of the 2006 Winter Simulation Conference, Monterey, CA, December 3-6, 2006; 1119-1128. Ramakrishnan, S., Lee, S., and Wysk, R. A. (2002). Implementation of a Simulation-based Control Architecture for Supply Chain Interactions. Proc. of the 34th Winter Simulation Conference, San Diego, CA, December 8-11, 2002; 16671674. Ramakrishnan, S., Tsai, P, -F., Srihari, K., and Foltz, C. (2008). Using Design of Experiments and Simulation Modeling to Study the Facility Layout for a Server Assembly Process. Proc. of the 2008 Industrial Engineering Research Conference, British Columbia, Canada, May 17-21, 2008; 601-606. Chen, Y., Fowler, J., Wu, T., Callarman, T., Ambrose, E., and Hargaden, V. (2006). An Adaptive Distributed Simulation Framework for a Server Fulfillment Supply Chain. Proc. of IEEE International Conference on Automation Science and Engineering, Shanghai, China, October 8-10, 2006; 649-655. Henn, S. and Wascher, G. (2012). Production, Manufacturing and Logistics: Tabu Search Heuristics for the Order Batching Problem in Manual Order Picking Systems. European Journal of Operational Research, 2012; 222 (3): 484-494. Henn, S. (2012). Algorithms for on-line order batching in an order picking warehouse. Computers and Operations Research, Vol. 39(11), 2012, pp. 2549-2563. Bukchin, Y., Khmelnitsky, E., and Yakuel, P. (2012). Production, Manufacturing and Logistics: Optimizing A Dynamic Order-picking Process. European Journal of Operational Research, 2012; 219(1): 335-346. Hsu, C.M., Chen, K.Y., and Chen, M.C. (2005). Batching orders in warehouses by minimizing travel distance with genetic algorithms. Computers in Industry, Vol. 56(2), 2005, pp. 169-178. Yu, M., and De-Koster, R. (2009). The impact of order batching and picking area zoning on order picking system performance, European Journal of Operational Research, Vol. 198(2), 2009, pp. 480-490. Seyedrezaei, M., Najafi, S. E., Aghajani. A., and Valami. H. B. (2012). Designing a Genetic Algorithm to Optimize Fulfilled Orders in Order Picking Planning Problem with Probabilistic Demand. International Journal of Research in Industrial Engineering, 2012; 1(2): 39-56. Ene, S. and Ozturk, N. (2012). Storage Location Assignment and Order Picking Optimization in The Automotive Industry. International Journal of Advanced Manufacturing Technology, 2012; 60(5-8): 787-797. Liang. C. F., Xiao. L. Z., Zheng. X., and Dong. L. D. (2007). Research on Order Picking Optimization Problem of Automated Warehouse. Systems Engineering – Theory & Practice, Vol. 27, Issue. 2, February 2007, pp. 139 – 143. Hsieh, L. F. and Huang, Y. C. (2001). New Batch Construction Heuristics to Optimize the Performance of Order Picking Systems. International Journal of Production Economics, 2001; 131(2): 618-630.
325
