Surface Roughness Prediction using Artificial Neural Network in Hard ...

5 downloads 26 Views 400KB Size Report
[6]: Vikram Kumar, CH R. and Ramamoorthy, B. Performance of coated tools ... roughness in CNC turning and its prediction using artificial neural networks, ...

Available online at www.sciencedirect.com

ScienceDirect Procedia Engineering 97 (2014) 205 – 211

12th GLOBAL CONGRESS ON MANUFACTURING AND MANAGEMENT, GCMM 2014

Surface Roughness Prediction using Artificial Neural Network in Hard Turning of AISI H13 Steel with Minimal Cutting Fluid Application B. Anuja Beatricea, E. Kirubakaranb, P. Ranjit Jeba Thangaiahc, K. Leo Dev Winsd,* a,c

Assistant Professor, dAssociate Professor, Karunya University, Coimbatore – 641114, India b Additional General Manager / (System SSTP,BHEL Tiruchirappalli – 620014, India.

Abstract Surface roughness plays an important role in manufacturing process and is a factor of great significance in the evaluation of cutting performance. In this paper an attempt was made to develop a model based on Artificial Neural Network to simulate hard turning of AISI H13 steel with minimal cutting fluid application. This model is expected to predict the surface roughness in terms of cutting parameters. Networks with different architecture were trained using a set of training data for a fixed number of cycles and were tested using a set of input / output data reserved for this purpose. The root mean square error was determined for the selected architectures. The model with 3-7-7-1 architecture gave the minimum RMSE value. The ability of ANN model to predict surface roughness (Ra) was analyzed. It was found that the predictions made by the ANN model matched well with the experimental results. © 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license © 2014 The Authors. Published by Elsevier Ltd. (http://creativecommons.org/licenses/by-nc-nd/3.0/). Selectionand and peer-review under responsibility of the Organizing of GCMM 2014. Selection peer-review under responsibility of the Organizing CommitteeCommittee of GCMM 2014

Keywords:

Artificial Neural Network; Hard turning; Minimal cutting fluid application; Surface roughness.

Nomenclature f feed rate v cutting speed d depth of cut Ra Surface roughness

* Corresponding author. Tel.: +91 98948 22791; fax: +91 422 2615615. E-mail address: [email protected]

1877-7058 © 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/3.0/). Selection and peer-review under responsibility of the Organizing Committee of GCMM 2014

doi:10.1016/j.proeng.2014.12.243

206

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211

1. Introduction Hardened steels have wide application in making tools, heavy machine parts and automobile components. By virtue of their high wear resistance and compressive strength, hardened steels have to some extent been able to meet the industrial demands for high performance components [1]. Lower equipment costs, shorter setup time, fewer process steps and greater part geometry are some of the advantages that make hard turning a much attractive option over conventional practices [2]. In the hard turning process, there is a high need to reduce cutting force, friction and cutting temperature so as to improve surface finish and to reduce tool wear. Hence hard turning is normally performed under flood cooling by applying huge quanty of cutting fluid. The elevated temperature at the cutting zone will instantly boil and vaporize the cutting fluids not only causing thermal distortions but also producing harmful fumes which cause adverse effects to operator health [3]. Dry machining avoids the problems associated with cutting fluids but it is very tough to implement dry machining on the existing shop floor as it needs very rigid machine tools and ultra hard cutting tools which will again add to the cost of machining [4]. In order to reduce the negative effects of cutting fluids, techniques like Minimal Quantity Lubrication (MQL) and Minimal Cutting Fluid Application (MCFA) have been evolved. In MCFA, extremely small quantity of cutting fluid is injected with the help of fluid injector in the form of ultra fine droplets at very high velocity (about 100 m/s) that tend to penetrate deep into the critical zones rather that float in the air as in most of the MQL applications. In MQL systems, compressed air rushing through a mixing chamber atomizes the cutting fluid into an aerosol of micron-sized particles. In MCFA, cutting fluid is applied to the cutting zone in the form of pulsing jet instead of a continuous jet as in MQL systems which improves the cutting performance. For all practical purposes, MCFA technique is similar to dry machining in achieving improved surface finish, reducing tool-chip contact length and tool wear by maintaining cutting forces and power at reasonable levels [5-8]. Manufacturing industries are developing attention on dimensional accuracy and surface roughness of finished products as they are very much important. Surface roughness is a vital parameter in assessing the desirable quality of the finished product especially when dealing with issues related to friction, lubrication and tool wear. Modeling techniques help in improving the effectiveness of machining operations and to reduce the manufacturing time and costs that will be beneficial to the manufacturing industry [9]. Models related to machining operations are nonlinear in nature. In order to get better and accurate results, analytical models are often subjected to simplifications and assumptions. In the recent times, Artificial Intelligence based models are becoming popular and these models are used by many researchers to develop models for near optimal conditions in metal cutting. Artificial intelligence based modeling approaches are advisable for real time applications out of which Artificial Neural Network (ANN) was found to be reliable, viable, and attractive [10]. The ANN based approach has been successfully implemented by many researchers and it produced acceptable results [11]. M. Nalbant et al. developed a model based on Artificial Neural Network to predict surface roughness in CNC turning of AISI 1030 steel [11]. Another ANN based model was developed by O¨ zel and Karpat to predict surface roughness and tool flank wear during finish dry hard turning of AISI H13 steel [12]. Leo and Varadarajan developed a model using ANN approach with fluid application parameters to simulate surface milling of hardened AISI4340 steel with minimal fluid application [13]. From the review of literature, it was found that ANN has been successfully utilized by researchers for modeling surface roughness in hard turning process but no work is reported on the modeling of surface roughness in terms of cutting parameters during hard turning of AISI H13 tool steel with minimal cutting fluid application. Hence, an effort was made to develop a model based on ANN to predict surface roughness (R a) in terms of cutting parameters during hard turning of AISI H13 tool steel with minimal cutting fluid application.

207

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211

2. Experimentation 2.1. Selection of work material and tool H13 tool steel of hardness 45 HRC was selected as work material in this investigation. H13 tool steel has got wide range of applications in die casting. Bars of 70 mm diameter and 360 mm length were used in the present work. Table 1 presents the chemical composition of the workpiece material. Table 1. Chemical composition of H13 steel (weight percentage). C

Cr

Mo

Si

Mn

P

S

Fe

0.430

5.02

1.13

1.08

0.214

0.0330

0.0290

balance

Cutting tool inserts and the tool holder were selected as per the recommendations of M/s TaeguTec India (P) Ltd. Accordingly tool insert with a specification SNMG 120408 and tool holder with a specification PSBNR 2525 M12 were used. The experimental work was carried out with two replications on a Kirloskar Turn master-35 lathe and Fig. 1 shows the photograph of the experimental setup.

Fig. 1 Experimental set up

2.2. Selection of cutting parameters The cutting parameters namely feed rate, cutting speed and depth of cut were selected at semi finish range and their combinations were determined based on the results obtained through the preliminary experiments and the recommendations suggested by M/s TaeguTec India (P) Ltd. 2.3. Design of experiment A 27 run experiment was designed using Taguchi technique and varied at three levels [14] as shown in Table 2.

208

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211 Table 2 Input parameters and their levels Input parameter

Level 1

Level 2

Level 3

Feed rate (mm/rev)

0.05

0.075

0.1

Cutting speed (m/min)

75

95

115

Depth of cut (mm)

0.5

0.75

1

The fluid application parameters, namely pressure at the injector, rate of application of cutting fluid, composition of cutting fluid and frequency of pulsing were maintained in the optimized condition at 100 bar, 8 ml/min, 20% oil in water and 500 pulses/min respectively [5] and it was decided to predict surface roughness in terms of cutting parameters. Surface roughness (Ra) was recorded using Mitutoyo (SJ-210) portable surface roughness tester during each experiment. Table 3 shows the experimental conditions and results of L27 orthogonal array. Table 3: Experimental data collected during 27 run experiment S.No

f (mm/rev)

v (m/min)

d (mm)

Ra (μm)

Training/testing

1

0.05

75

0.5

1.28

Training

2

0.05

75

0.75

1.1

Training

3

0.05

75

1

1.58

Training

4

0.05

95

0.5

0.91

Training

5

0.05

95

0.75

1.03

Training

6

0.05

95

1

1.13

Training

7

0.05

115

0.5

1.38

Training

8

0.05

115

0.75

1.34

Training

9

0.05

115

1

1.45

Testing

10

0.075

75

0.5

1.19

Training

11

0.075

75

0.75

1.09

Training

12

0.075

75

1

1.48

Training

13

0.075

95

0.5

1.45

Training

14

0.075

95

0.75

1.86

Training

15

0.075

95

1

1.82

Testing

16

0.075

115

0.5

1.43

Training

17

0.075

115

0.75

1.7

Training

18

0.075

115

1

1.65

Training

19

0.1

75

0.5

1.64

Testing

20

0.1

75

0.75

1.13

Training

21

0.1

75

1

1.78

Training

22

0.1

95

0.5

1.77

Training

23

0.1

95

0.75

1.83

Training

24

0.1

95

1

1.53

Training

25

0.1

115

0.5

1.32

Testing

26

0.1

115

0.75

1.6

Training

27

0.1

115

1

1.79

Training

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211

As the quantity of cutting fluid used was extremely small and the cutting fluid needs to perform both cooling and lubrication, a specially formulated cutting fluid was employed. The base of the cutting fluid was mineral oil and the formulation contained other ingredients [15]. Then the formulated cutting fluid was mixed with sufficient quantity of water for the application into the cutting zone. 3. Artificial Neural Network The appropriate architecture for the artificial neural network was selected through an exhaustive examination of a number of network configurations. This was accomplished by changing the number of hidden layers and number of neurons in the hidden layer. A routine that utilizes a feed forward back propagation algorithm was used to develop the model as it is widely used by researchers and it was observed that the feed forward back propagation algorithm gave the most accurate results [16, 17]. Matlab ANN toolbox was used for easily updating the value of weights and biases of the algorithm. Networks with different architecture were trained for a fixed number of cycles and were tested using a set of input and output parameters. The guidelines given by Zhang et al.[18] were considered while selecting the network architecture along with 52 other configurations. According to Zhang et al., the recommended number of neurons in the hidden layer are ‘n/2’, ‘1n’, ‘2n’, and ‘2n + 1’ where n is the number of input nodes. Therefore, this study was applied to eight different architecture, which are 3–1–1, 3–3–1, 3–6–1, 3–7–1, 3–1–1–1, 3–3–3–1, 3–6–6–1 and 3–7–7–1, apart from 52 other configurations. In this present work, ‘learngdm’ was considered as the learning function and ‘trainlm’ as the training function. The transfer function of the ANN model was considered as “tansig” and the sigmoid function used in this experimentation is shown in equation (1) 1 (1) f x 1  ex The hidden layer type was altered progressively into single and multi layer types with different neuronal values using trial and error method [16] for the best configuration with least MSE value and better coefficient of determination. 3.1. Surface roughness prediction by ANN Out of the 27 readings in Table 3, 23 readings were used for training the model and 4 readings were reserved for testing and this ratio (85% : 15%) between the number of training values to the number of testing values was chosen based on similar work [16]. The RMSE was considered as quadratic scoring rule for the measurement of average magnitude of the error. In this study, 60 different architectures were trained and some of the configurations that gave lower MSEs are presented in Fig. 2

Fig. 2 Variation of limiting MSE with different layer configuration

209

210

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211

The test pattern used for validating the ANN model and its comparison with experimental results is shown in Table 4. It was seen that there is a good concordance between the predictions of ANN model and the experimental results. Table 4. Testing and comparison of surface roughness by ANN with experimental results Testing data S.No

Surface roughness Ra (μm)

f (mm/rev)

v (mm/min)

d (mm)

Experimental result

Prediction by ANN

% Error

1

0.05

115

1

1.45

1.35

6.89

2

0.075

95

1

1.82

1.74

4.39

3

0.1

75

0.5

1.64

1.58

3.65

4

0.1

115

0.5

1.32

1.26

4.54

4. Results and discussions In the present investigation, surface roughness (Ra) was considered as the performance parameter which is widely used for monitoring the surface texture. Out of the 60 configurations which were used for training, the architecture consisted of three neurons in the input layer, two hidden layers with seven neurons each and one neuron in the output layer (3-7-7-1) was found to have lowest MSE value of 0.008 at 10000 cycles. Coefficient of determination and standard error were chosen as performance indices for evaluating the ability of the models to predict surface roughness. In model analysis, coefficient of determination is used as a measure to consider how efficiently a model predicts the outcomes. It is expressed as a value between 0 and 1. The higher the value, the better the prediction. In the present investigation, coefficient of determination for 3-7-7-1 configuration was found to be 0.95962 and the standard error of 0.0950. It was also observed with ANN model that the predictions with considerable accuracy (error percentage of < 7%) is possible even with smaller number of training data. Conclusions In the present investigation, ANN approach was used for predicting surface roughness during hard turning of H13 tool steel with minimal cutting fluid application. The following conclusions were drawn: 1. 2. 3.

The ANN model developed can be highly useful in fixing the cutting parameters to achieve desired surface finish and to maintain the surface finish within the tolerance limits during automated hard turning of AISI H13 steel with minimal fluid application. The new scheme can be well acceptable to the industry as it does not require any major changes in the existing setup for its implementation on the shop floor. Minimal cutting fluid application technique promoted green environment in the shop floor, minimized the industrial hazard due to of harmful aerosols and usage of large quantity of cutting fluid.

References [1] [2] [3] [4] [5] [6]

H.K. Tönshoff, C. Arendt, R. Ben Amor, Cutting of hardened steel, CIRP Annals–Manufacturing Technology, 49(2): (2000) 547–566. König W, Hochschule T, Komanduri R, Schenectady D, Tönshoff HK, Machining of hard materials, CIRP Annals, 33(2): (1984) 417– 427. NIOSH ‘Metal Working Fluids - Recommendation for Chronic Inhalation Studies’, National Institute for Occupational Safety and Health (NIOSH), Cincinnati, OH USA 45226, 2001. K. Leo Dev Wins, A.S. Varadarajan, An Environment Friendly Twin-jet Minimal Fluid Application Scheme for Surface Milling of Hardened AISI4340 Steel, International Journal of Manufacturing Systems (2011) 30-45. Philip, P. K., Varadarajan, A. S., Ramamoorthy, B. Influence of cutting fluid composition and delivery variables on performance in hard turning using minimal fluid in pulsed jet form, Journal of the Institution of Engineers (India), (2000) 68-72. Vikram Kumar, CH R. and Ramamoorthy, B. Performance of coated tools during hard turning under minimum fluid application, Journal of Materials Processing Technology, 185 (2007) 210–216.

B. Anuja Beatrice et al. / Procedia Engineering 97 (2014) 205 – 211 [7]

[8] [9]

[10] [11]

[12] [13]

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

Vikram Kumar, CH R., Kesavan Nair, P. and Ramamoorthy, B. Performance of TiCN and TiAlN tools in machining hardened steel under dry, wet and minimum fluid application, International Journal of Machining and Machinability of Materials, 3 (1/2): (2008) 133-142. Varadarajan, A.S., Philip, P.K. and Ramamoorthy, B. Investigations on hard turning with minimal cutting fluid application (HTMF) and its comparison with dry and wet turning, International Journal of Machine Tools and Manufacture, 42 (2002) 193–200. C.A. van Luttervelt, T.H.C. Childs, I.S. Jawahir, F. Klocke, P.K. Venuvinod, Y. Altintas, E. Armarego, D. Dornfeld, I. Grabec, J. Leopold, B. Lindstrom, D. Lucca, T. Obikawa, Shirakashi, H. Sato, Present Situation and Future Trends in Modelling of Machining Operations Progress Report of the CIRP Working Group ˄Modelling of Machining Operations, CIRP Annals - Manufacturing Technology, 47 (2): (1998), 587ˀ626. Chryssolouris G ,Guillot M., A comparison of statistical and AI approaches to the selection of process parameters in intelligent machining, ASME Journal of Engineering for Industry, 112 (1990) 122–131. Muammer Nalbant, Hasan Gökkaya, İhsan Toktaş, Gökhan Sur, The experimental investigation of the effects of uncoated, PVD- and CVD-coated cemented carbide inserts and cutting parameters on surface roughness in CNC turning and its prediction using artificial neural networks, 25 (1), : (2009), 211ˀ223. O¨ zel T, Karpat Y, Predictive modeling of surface roughness and tool wear in hard turning using regression and neural networks, International Journal of Machine Tools and Manufacture, 45 (2005) 467–479. K. Leo Dev Wins, A.S Varadarajan, ‘Prediction of surface roughness during surface milling of hardened AISI4340 steel with Minimal cutting fluid application using Artificial Neural Network.’ International journal of Advances in Production Engineering & Management, 7(1): (2012), 51- 60. Lochner, R. H., Matar, J.E. ‘Designing for Quality’, 20th Edn., Chapman and Hall, London, 1990. Varadarajan, A.S., Ramamoorthy, B. and Philip, P.K., Formulation of a Cutting fluid for Hard Turning with Minimal Fluid Application’, 20th AIMTDR conference, Birla institute of Technology Ranchi, India, (2002) 89-95. Valluru Rao, Hayagriva Rao. C++, Neural Networks and Fuzzy Logic, BPB Publications, New Delhi, 1996. Azlan Mohd Zain, Habibollah Haron, Safian Sharif, ‘Prediction of surface roughness in the end milling machining using Artificial Neural Network’, International journal of Expert Systems with Applications, 37 (2): (2010), 1755ˀ1768. Zhang, G, Patuwo, B. E, Hu, M. Y., Forecasting with artificial neural networks: The state of the art. International Journal of Forecasting, 14 (1998) 35–62.

211

Suggest Documents