Modeling and Prediction of Surface Roughness During Dry Turning Process

Modeling and Prediction of Surface Roughness During Dry Turning Process

M. M. Koura, T. H. Sayed, A. S. El-Akkad

Design and production eng. Dept.

Ain shams university Cairo, Egypt

Abstract–Surface roughness () prediction model using artificial neural network (ANN) is developed in this work for dry turning of mild steel (St. 42) using carbide inserts. Cutting parameters (cutting speed, feed rate, and depth of cut) are used as network inputs. Also, investigations of the effect of cutting parameters are presented. The analysis reveals that increasing the cutting speed will decrease surface roughness, and increasing feed rate will increase surface roughness. The results show that the developed ANN model can predict surface roughness with regression value 98.641 % between measured and predicted values and average error 5.4%. Detailed experimentation and ANN network structure are presented in the paper.

Keywords: Surface roughness, artificial neural network, dry turning, cutting parameters

1. INTRODUCTION

   Turning is the primary operation in most of the production processes. Components, with critical features usually require specific surface finish. Selection of the proper cutting parameters in turning operation is a critical task and greatly affects surface roughness.

   So achieving the desired surface roughness greatly affects the functional behavior of the mechanical parts. Surface roughness has influence on several properties such as wear resistant, fatigue strength, coefficient of friction, wear rate, and corrosion resistance of machined parts [1].

   The surface roughness can be defined by ( ) which is the arithmetical average roughness. Theoretical equation was presented [2] to estimate this parameter:

   = 1000 2

   32

   Where is the average surface roughness (µm), f is the feed rate (mm/rev) and r is the tool nose radius (mm).

   However, the above theoretical equation does not take into consideration many parameters that can affect surface roughness as cutting speed, depth of cut, and others. It will be shown that all of these parameters play an important role on surface roughness of the finished components.

2. LITERATURE REVIEW

   Standard roughness measurement procedures depend on stylus instruments which have only limited flexibility in handling. Furthermore, this procedure is not suitable for automation and relatively slow. In recent years several researchers developed many models to predict surface roughness in terms of various process parameters during turning of different materials to overcome the disadvantages of standard roughness measurement.

   1. Artificial neural network model

      J. Paulo et al. [2] developed a surface roughness prediction model using artificial neural network to investigate the effect of cutting conditions during turning of free machining steel. It is found that the maximum absolute error between the predicted values and the measured values are 28.29% and 8.19% for and respectively, K. A. Risbood et al. [3] developed three artificial neural network models to predict surface roughness for dry and wet turning of commercially available steel using high cobalt-HSS and TiN coated carbide tools. It is found that the maximum absolute error between the predicted values and the measured values are 38.55%, 31.9%, and 22.93% for first model, second model, and third model respectively, Ilhan Asilturk et al. [4] developed artificial neural network and multiple regression (second order) approaches to predict surface roughness of AISI 1040 steel, using carbide inserts and they found that artificial neural network produces better results compared to multiple regression, C. Natarajan et al. [5] developed an artificial neural network to predict surface roughness for dry turning of brass C26000 using CNMG 120408 insert. The percentage deviation between actual roughness values and predicted roughness values is 24.4%, and the model could achieve an accuracy of 75.6%.

   2. Adaptive fuzzy network model

      Yue Jiao et al. [6] developed a fuzzy adaptive network to model surface roughness in turning operations, during turning of cold rolled steel using coated carbide tool, the results of the adaptive fuzzy model are compared with the results obtained from a regression model and they found that the adaptive fuzzy model outperforms the regression model. Shinn-Ying Ho et al. [7] proposed a method using

      an adaptive neuro-fuzzy inference system to accurately establish the relationship between the features of surface image and the actual surface roughness of S45C steel bars using tungsten carbide tool. The errors of the developed model are smaller than 4.6%.

   3. Regression model

      I.A. Choudhury et al. [8] developed a first order prediction model, and a second order prediction model to predict surface roughness for turning EN 24T steel using uncoated carbide inserts. It indicated that the second order model is more adequate, Yusuf Sahin et al. [9] developed a first order and second order model to predict surface roughness for turning of mild steel using coated carbide tool. It is found that only the first order model for prediction is important. Vikas Upadhyay et al. [10] developed two regression models (first-order and second-order) to predict surface roughness of turning Ti-6Al-4V alloy using uncoated cemented carbide. The best 2 value of the models is 93.2% and the average error of this model found to be 3.5%. Sudhansu Ranjan Das et al. [11] developed multiple regression analysis to predict surface roughness of turning AISI 4340 steel using multilayer coated carbide inserts. It is observed that feed rate is the most significant parameters followed by cutting speed and it is found that the depth of cut did not impact the surface roughness in the studied range. Zahia Hessainia et al. [12] developed two multiple regression models to predict surface roughness when turning of 42CrMo4 hardened steel using 23/TiC mixed ceramic tool. The results indicate that the feed rate is dominant factor affecting the surface roughness.

   4. Abductive network model

      W. S. Lin et al. [13] developed two models: the first one is an abductive network model, the second model is regression analysis, for the surface roughness and cutting force for turning high carbon steel using carbide inserts. It is found that the percentage error under various cutting conditions is generally 10%, while it is 59.44% for regression model. B. Y. Lee et al. [14] developed a polynomial network using a self organizing adaptive modeling method to construct the relationships between the feature of the surface image and the actual surface roughness under a variation of turning operations of S45C steel bars using tungsten carbide tool. The developed model can predict surface roughness with reasonable accuracy (average percentage error is 6.2%), B. Y. Lee et al. [15] developed a system for measuring surface roughness of turned parts through computer vision system, the model use an adaptive network to predict surface roughness for turning of S45C steel bars using tungsten carbide tool. It is found that the maximum absolute error between the surface roughness measured by vision system and that measured by the stylus instrument is 14.96%,

   5. Experimental studies

      Thomas M. et al. [16] study and analyze of surface roughness and tool vibration data generated by lathe during dry turning of mild carbon steels using cemented carbide tool, Hasan Gokkaya [17] study and analyze cutting force, surface roughness, built-up edge, and built-up layer during turning of AA204 (T4) alloy by uncoated carbide tool, and N. Satheesh Kumar et al. [18] investigate the effect of process parameters in wet turning of carbon alloy steel using a carbide tip tool.

      The aim of this study is to:

      • analyze the effect of feed rate (f), cutting speed (v), and depth of cut (d) on surface roughness parameter such as average roughness ( ).

      • Develop a model using artificial neural network (ANN) to predict surface roughness using cutting parameters as network inputs.

3. EXPERIMENTAL SETUP

   Data sets are from experiments conducted on a conventional lathe in the workshop of Ain Shams University, Faculty of Engineering, Cairo, Egypt. The details of the machining experiments are given in table 1. Carbide insert was used for machining a mild steel bar of length 300 mm. After each experiment the surface roughness ( ) is measured using surface roughness tester (TAYLOR-HOBSON, surtronic 3). Roughness measurements are taken five times for each work piece then the average value is calculated and used. In order to avoid the effect of cutting tool wear and tool nose radius, each experiment was conducted with new, identical cutting edge. The cutting experiments were performed without coolant, and totally 27 experiments were performed according to full factorial design. Table 2 shows the experimental cutting parameters and the resulted surface roughness.

   Table 1: Cutting parameters details

   Parameter	Level 1	Level 2	Level 3
   Depth of cut (mm)	0.50	1.00	2.00
   Cutting speed (rpm)	765	955	1200
   Feed rate (mm/rev)	0.20	0.40	0.60
   Initial work piece diameter (mm)	25	26	28

   Table – 2: detailed experimental data

   Test No.	v (m/min)	f (mm/rev)	d (mm)	(µm)
   1	98.06	0.2	2	2.93
   2	98.06	0.4	2	4.29
   3	98.06	0.6	2	8.09
   4	78.04	0.2	2	4.05
   5	78.04	0.4	2	5.61
   6	78.04	0.6	2	10.31
   7	62.51	0.2	2	4.95
   8	62.51	0.4	2	7.85
   9	62.51	0.6	2	10.91
   10	94.29	0.2	1	1.27
   11	94.29	0.4	1	4.25
   12	94.29	0.6	1	8.39
   13	75.04	0.2	1	3.09
   14	75.04	0.4	1	4.13
   15	75.04	0.6	1	8.97
   16	60.11	0.2	1	4.47
   17	60.11	0.4	1	8.47
   18	60.11	0.6	1	10.87
   19	92.40	0.2	0.5	3.05
   20	92.40	0.4	0.5	4.33
   21	92.40	0.6	0.5	7.57
   22	73.54	0.2	0.5	4.29
   23	73.54	0.4	0.5	4.59
   24	73.54	0.6	0.5	8.97
   25	58.91	0.2	0.5	4.09
   26	58.91	0.4	0.5	7.13
   27	58.91	0.6	0.5	10.45

4. ARTIFICIAL NEURAL NETWORK (ANN)

   In this study, ANN structure shown in figure 1 is used for modeling and predicting surface roughness ( ) in turning process. Many neural network structures with different

   number of neurons in each layer and different processing functions are tested and it is found that the network with the best results has 3 layers, the first layer is the input layer and it consists of 3 neurons for the 3 inputs of the network, the

   Figure – 1: ANN structure

   second layer is the hidden layer and it consist of 3 neurons, and the third layer is the output layer and it consists of 1 neuron.

   The processing function for the hidden layer is logsig, and for the output layer is tansig. Feed-forward back propagation ANN is used, Leven berg-Marquardt back propagation (TRAINLM) algorithm is used for network training and mean square error (MSE) is used as performance function, many neural network structures with different number of neurons in each layer and different processing functions are tested and it is found that this one has the best results

5. RESULTS AND DISCUSSION

   In this section, the results obtained from the ANN and the developed model are presented and discussed. Finally, the

   effect of different cutting parameters on surface roughness ( ) are explained.

   1. Artificial neural network results

      The experimental data set consists of 27 experiments of which 19 experiments are used for training the network and

      8 experiments are selected randomly for validation and testing the performance of the trained network.

      Table 3 shows the results obtained from the trained ANN for the 27 experiments; also it shows the error percentage between the measured value of surface roughness and the predicted value. Figure 2 shows the difference between the measured value and the predicted value of surface roughness.

      Table 3: ANN results

      Test No.	v (m/min)	f (mm/rev)	d (mm)	(µm)	(µm)	% error
      1	98.06	0.2	2	2.93	2.73	6.72
      2	98.06	0.4	2	4.29	4.22	1.70
      3	98.06	0.6	2	8.09	8.12	0.32
      4	78.04	0.2	2	4.05	3.30	18.48
      5	78.04	0.4	2	5.61	5.95	6.11
      6	78.04	0.6	2	10.31	10.07	2.33
      7	62.51	0.2	2	4.95	4.95	0.01
      8	62.51	0.4	2	7.85	7.85	0.01
      9	62.51	0.6	2	10.91	10.73	1.65
      10	94.29	0.2	1	1.27	1.67	31.88
      11	94.29	0.4	1	4.25	4.03	5.09
      12	94.29	0.6	1	8.39	7.76	7.52
      13	75.04	0.2	1	3.09	3.18	2.87
      14	75.04	0.4	1	4.13	4.14	0.14
      15	75.04	0.6	1	8.97	8.97	0.00
      16	60.11	0.2	1	4.47	4.03	9.77
      17	60.11	0.4	1	8.47	7.54	10.96
      18	60.11	0.6	1	10.87	10.63	2.19
      19	92.40	0.2	0.5	3.05	2.65	13.20
      20	92.40	0.4	0.5	4.33	3.95	8.82
      21	92.40	0.6	0.5	7.57	7.58	0.14
      22	73.54	0.2	0.5	4.29	4.28	0.21
      23	73.54	0.4	0.5	4.59	4.59	0.03
      24	73.54	0.6	0.5	8.97	9.78	9.05
      25	58.91	0.2	0.5	4.09	3.93	3.84
      26	58.91	0.4	0.5	7.13	7.29	2.20
      27	58.91	0.6	0.5	10.45	10.57	1.12

      *: surface roughness predicted by ANN

      The results show that the error percentage is less than 5% for the majority of experiments, also it shows that error percentage is near zero in many experiments, also it shows

      that the correlation coefficient between the measured values and predicted values is 0.98641, and the average error percentage is 5.4%.

      1 4 7 101316192225

      experiment

      measured

      roughness

      ANN roughness

      15

      10

      5

      0

      roughness (µm)

      roughness(µm)

      Figure -2: Measured and predicted data for surface roughness

   2. Effect of cutting parameters

      Now the effect of each cutting parameter (f, v, d) on surface roughness ( ) will be discussed. Figures 3, 4, and 5 show the relation between feed rate and surface roughness ( ) for different cutting speed and different depth of cuts.

      feed and roughness, d=2mm

      roughness(µm)

      12

      10

      8

      feed and roughness, d=0.5mm

      15

      10

      v= 92 m/min

      5

      v=73.5 m/min

      0 v=59 m/min

      0 0.5 1

      feed (mm/rev)

      Figure – 5: Relation between feed and roughness for d= 0.5mm mm

      1. Effect of feed rate

         From the experiments it is shown that increasing feed rate will increase surface roughness for different cutting speeds and different depth of cuts. The reason for that is increasing the feed rate during machining causes the cutting forces to increase due to high frictional forces on the tool rake face.

      2. Effect of cutting speed

         6

         4

         2

         0

         0 0.5 1

         feed (mm/rev)

         v=98 m/min v=78 m/min v=62.5 m/min

         From the experiments it is shown that increasing cutting speed will decrease surface roughness for different depth of cuts and different feed rates. The reason for that is increasing the cutting speed during machining causes the cutting forces to decrease due to low frictional forces on the tool rake face at high cutting speeds.

      3. Effect of depth of cut

   From the experiments it is shown that increasing depth of

   Figure 3: Relation between feed and roughness for d= 2 mm

   feed and roughness, d=1mm

   roughness (µm)

   12

   10

   8

   cut from 0.5 mm to 1 mm for small and medium feed rate (

   0.2 and 0.4 mm/rev) will decrease surface roughness for different speeds. But further increasing of depth of cut will cause increasing in surface roughness; also it is shown that increasing depth of cut at high feed rate (0.6 mm/rev) will not have remarkable effect on surface roughness and this may be because that the feed rate become the dominant

   6

   4

   2

   0

   0 0.5 1

   feed (mm/rev)

   v= 94 m/min v=75 m/min v=60 m/min

   factor affecting the surface roughness.

6. CONCLUSIONS

In this study experiments are done on conventional lathe using different cutting parameters and the results of the experiments are used to build artificial neural network (ANN) model to predict surface roughness( ). The following conclusions are extracted from this study:

Figure – 4: Relation between feed and roughness for d= 1 mm

• The surface roughness ( ) could be effectively predicted by using depth of cut, cutting speed, and feed rate as the inputs variables.

• The developed artificial neural network prediction model can predict surface roughness ( ) accurately as correlation coefficient between the

measured values and predicted values is 0.98641, and the average error percentage is 5.4%.

• Artificial neural network can predict non-linear relationships accurately.

• Increasing feed rate will cause increasing of surface roughness for different cutting speeds and different depth of cuts.

• Increasing cutting speed will cause decreasing of surface roughness for different feed rates and different depth of cuts.

• Increasing depth of cut from for small and medium feed rate (0.2 and 0.4 mm/rev) will decrease surface roughness for different speeds. But further increasing of depth of cut will cause increasing in surface roughness. Also it is shown that increasing depth of cut at high feed rate will not have remarkable effect on surface roughness and this may be because that the feed rate becomes the dominant factor affecting the surface roughness.

REFERENCES

[1]. C. X. Feng, X. Wang, 2002 Development of empirical models for surface roughness prediction in finish turning. The international journal of advanced manufacturing technology (2002) 20: 348-356

[2]. J. Paulo Davim, V.N Gaitonde, S.R. Karnik, 2008 investigation into the effect of cutting condition on surface roughness in turning of free machining steel by ANN models. Journal of materials processing technology 205 (2008) 16-23.

[3]. K. A. Risbood. U. S. Dixit, A. D. Sahasrabudhe, 2003 Prediction of surface roughness and dimensional deviation by measuring cutting forces and vibraion in turning process. Journal of material processing technology 132 (2003) 203-214.

[4]. Ilhan Asilturk, Mehmet Cunkas, 2011 Modeling and prediction of surface roughness in turning operations using artificial neural network and multiple regression method. Expert system with applications 38 (2011) 5826-5832.

[5]. C. Natarajan, S. Muthu, P. Karuppuswamy, 2011 Investigation of cutting parameters of surface roughness for a non-ferrous material using artificial neural network in CNC turning. Journal of mechanical engineering research Vol. 3(1), pp. 1-14, January 2011

ISSN 2141-2383.

[6]. Yue Jia, Shuting Lei, Z. J. Pei, E.S. Lee, 2004 Fuzzy adaptive network in machining process modeling: surface roughness prediction for turning operations. International journal of machine tools & manufacture 44 (2004) 1643-1651.

[7]. Shinn-Ying Ho, Kuang-Chyi Lee, Shih-Shin Chen, Shinn-Jang Ho, 2002 Accurate modeling and prediction of surface roughness by computer vision in turning operations using an adaptive neuro-fuzzy inference system. International journal of machine tools & manufacture 42 (2002) 1441-1446.

[8]. I. A. Choudhury, M. A. El-Baradie, 1997 Surface roughness prediction in the turning of high-strength steel by factorial design of experiments. Journal of materials processing technology 67 (1997) 55-61.

[9]. Yusuf Sahin, A. Riza Motorcu, 2005 Surface roughness model for machining mild steel with coated carbide tool. Materials and design 26 (2005) 321-326.

[10]. Vikas Upadhyay, P. K. Jain, N. K. Mehta, 2013 In-process prediction of surface roughness in turning of Ti-6Al-4V alloy using cutting parameters and vibration signals. Measurements 46 (2013) 154-160.

[11]. Sudhansu Ranjan Das, Amaresh Kumar, Debabrata Dhupal, 2013 Effect of machining parameters on surface roughness in machining of hardened AISI 4340 steel using coated carbide inserts. International journal of innovation and applied studies ISSN 2028- 9324 Vol. 4 Apr. 2013. Pp. 445-453.

[12]. Zahia Hessainia, Ahmed Belbah, Mohamed Athmane Yallese, Tarek Mabrouki, Jean-Francois Rigal, 2013 on the prediction of surface roughness in the hard turning based on cutting parameters and tool vibrations. Measurment 46 (2013) 1671-1681.

[13]. W.S. Lin, B. Y. Lee, C. L. Wu, 1999 Modeling the surface roughness and cutting force for turning. Journal of materials processing technology 108 (2001) 286-293.

[14]. B. Y. Lee, Y. S.Tarng, 2001 Surface roughness inspection by computer vision in turning operations. International journal of machine tools & manufacture 41 (2001) 1251-1263.

[15]. B. Y. Lee, S. F. Yu, H. Juan, 2004 The model of surface roughness inspection by vision system in turning. Mechatronics 14 (2004) 129- 141.

[16]. Thomas M, Beauchamp Y., Youssef A. Y., Masounave J., 1996 Effect of tool vibrations on surface roughness during lathe dry turning process. Computer ind. Engng S0360-8352(96)00235-5.

[17]. Hasan Gokkaya, 2010 The effect of machining parameters on cutting forces, surface roughness, built-up edge (BUE) and built-up layer (BUL) during machining AA2014 (T4) alloy, Journal of mechanical engineering 56 (2010)9, 584-593.

[18]. N. Satheesh Kumar, Ajay Shetty, Ashay Shetty, Ananth K., Harsha Shetty, 2012 Effect of spindle speed and feed rate on surface roughness of carbon steels in CNC turning, International conference on modeling, optimization and computing 38 (2012) 691-697.