PREDICTION OF ROUGHNESS IN HARD TURNING OF AISI 4140 STEEL THROUGH ARTIFICIAL NEURAL NETWORK AND REGRESSION MODELS

Transcription

1 International Journal of Mechanical Engineering and Technology (IJMET) Volume 7, Issue 5, September October 2016, pp , Article ID: IJMET_07_05_022 Available online at Journal Impact Factor (2016): (Calculated by GISI) ISSN Print: and ISSN Online: IAEME Publication PREDICTION OF ROUGHNESS IN HARD TURNING OF AISI 4140 STEEL THROUGH ARTIFICIAL NEURAL NETWORK AND REGRESSION MODELS D. Rajeev Research Scholar, Hindustan University, Chennai, India. D. Dinakaran Professor and Group Lead, Hindustan University, Chennai, India. S. Muthuraman Professor, Department of Mechanical Engineering, Higher College of Technology, Muscat, Oman. ABSTRACT The surface roughness in hardened steel, which has wide industrial application, is crucial if not monitored properly. In this investigation the prediction of surface roughness during hard turning of AISI4140 steel (47HRC) is carried out. The cutting insert used is coated carbide tool. The experiments are carried out based on central composite design approach of Response surface methodology. The prediction model for roughness is developed using regression and artificial neural network (ANN).The cutting speed, feed rate and depth of cut are given as inputs and the roughness is the output. The models are tested for random cutting conditions. The results reveal that the accuracy is more in ANN model compared to regression model. Key words: ANN, Regression, Hard Turning, Roughness. Cite this Article: D. Rajeev, D. Dinakaran and S. Muthuraman, Prediction of Roughness in Hard Turning of AISI 4140 Steel through Artificial Neural Network and Regression Models. International Journal of Mechanical Engineering and Technology, 7(5), 2016, pp INTRODUCTION In a modern industrial scenario, the hard turning is slowly replacing grinding process in the finishing operation of hard steel. The Cubic Boron Nitride (CBN) inserts are widely used for hard turning [1, 2]. Various researches have been carried out to find the effectiveness of less expensive coated carbide tool for hard turning [3, 4]. But Roughness is a major problem when using carbide inserts, as the wear rate is higher in hard Turning. The decreased surface quality will lead to machine downtime, which in turn amounts to an increase in production cost [5]. For finding the optimum conditions in hard turning the estimation of thoroughness in terms of the process parameters (cutting speed, feed and depth of cut) are important

2 Prediction of Roughness in Hard Turning of AISI 4140 Steel through Artificial Neural Network and Regression Models Many authors have done research, in the prediction of Roughness during hard turning. Since the process is quite complex the developing of an analytical model is difficult. The empirical model which is based on the experimental data is well suited in such situations [6]. Regression model which estimate theroughness as function of cutting conditions have been developed by many authors. Zhang et al. [7] uses the cutting parameters and cutting force to develop a multi regression technique for finding surface roughness. Two sub systems for surface roughness evaluation and control have been developed and evaluated. E.Aslan et.al [8] have developed regression equation for estimating tool wear and roughness during machining of AISI4140 steel using ceramic inserts. The optimal cutting conditions for minimizing the wear were also derived.h.aouichi et.al [9] have developed regression equation to find the roughness and force components during hard turning of X38CrM0V5-1 steel. The cutting tool that is used is CBN tool. The optimal conditions to reduce the wear were also found out. Y.Sahin [10] has developed an empirical equation to compare the tool life between the CBN and ceramic inserts, during machining of hardened AISI bearing steel. CBN tools showed better performance. The artificial neural networks are widely employed whenever the input and output relations are nonlinear [11].The advantage of ANN being independent of the mechanism involved, depends only on, the mapping of data between the input and the output. Ozel and karpat have developed an ANN model to predict tool wear and surface roughness [12].Experimental results were used for training. Algorithm used for training is Bayesian regularization with Leven berg Marquardt training Algorithm. The developed model was capable of predicting surface roughness and wear in the range of training. The model was compared with empirical model and it was found to be better. They also observed that ANN with a single output gave better results compared with ANN with multiple outputs. Harun Akkus and II hanasilturk have modeled roughness using fuzzy, regression and ANN with process parameter as inputs. The Mean square error for each model was calculated and they found that fuzzy gave better results [13]. In their study D.E.Dimla Sr and PM Lister [14] have developed a multilayer perception network for predicting wear using force and acceleration as input. The wear states were classified and the scopes of developed tool condition monitoring were also assessed. The material taken in to consideration is AISI4340. Xiaoyu Wang et.al[15] have developed a tool wear estimator for CBN tools based on hard turning using Fully forward connected neural network. Training algorithm used is the extended Kalman filter algorithm. Based on the literature, even though there are many surface roughness prediction algorithms available in hard turning, most of them are limited to the roughness, prediction based on expensive CBN tools. The algorithm of surface roughness for different levels of process parameters which is required for online optimization and adaptive control in the hard turning using less expensive coated carbide tool is very much limited.in this work the roughness when using coated carbide tool, is predicted based on Regression and ANN, during the hard turning operation. The input parameters used to create the model are the process parameters. Comparison of predicted values with the measured values is also made. 2. MATERIALS AND METHODS In this study, AISI 4140 steel heat treated to 47 HRC is the work piece. Rough turning is done to remove oxide layers before heat treatment. The dimensions of the work piece are 80mm diameter and 250 mm length. Heavy duty kirloskar lathe is used for experimentation. The CVD coated Ti(C,N) +Al 2 o 3 carbide of SECO make is used as the cutting tool. PCL NR2525 M12 type is the tool holder. Mitutoyo make surface roughness tester (SJ-210) is used for the measurement of roughness. The machining length considered is 200mm.The details are shown in Fig.1and

3 D. Rajeev, D. Dinakaran and S. Muthuraman 3. EXPERIMENTAL PPROCEDURE The design of the experiments is formulated based on the response surface methodology. The Central composite design (CCD) approach is followed. As per the requirement of CCD, a total of 20experiments were carried out, based on the levels of cutting parameters shown in Table1.In CCD the levels are taken -1; 0 ;and+1and the α value is The average roughness (Ra) is taken as the response. The advantage of CCD is only less number of experiments are expected to be carried out when compared to full factorial design. The details are depicted in Table2.The average roughness is based on the relation given below. Ra =1/l Where l-length considered f(x)-roughness profile function. (1) Figure 1 Operational block diagram Figure 2 Experimental Setup 202

4 Prediction of Roughness in Hard Turning of AISI 4140 Steel through Artificial Neural Network and Regression Models Table 1 Levels of Process Parameters Level V(m/min) f(mm/rev) d(mm) Table 2 Experimental values. Machining parameter Response factors Run No V(m/min) f(mm/rev) d (mm) R a (µm)

5 D. Rajeev, D. Dinakaran and S. Muthuraman 4. REGRESSION BASED MODEL The regression equation is a nonlinear quadratic equation as given below. Y= a 0 + a X i +,a X i X j + a X $ (2) Wherea 0,a i,a ij,a ii are the coefficients, Y is the output and X i, X j are the inputs. In this work Roughness is the output and the cutting parameters are the input. Based on the experimental dataa given in Table 2, the following quadratic equation is derived. R a % V' f d % V d f d' , V $ f $ ' d $ (3) Prediction of the Roughness values for any value within the range specified in Table1 is given by the above equation. The prediction ability is specified by the R 2 value. For the above regression equation the R 2 value is Closeness to 1 indicates good correlation with measured value. The above equation is validated for the condition shown in Table ANN BASED MODEL Figure 3 ANN Roughness Estimator The accuracy of regression model formulated above is based on the orderr of the polynomial. It is bit complicated whenever the relations are nonlinear in nature. In such cases artificial neural networks are widely used. The ANN does not take in to account the complex process involved during model formation. It is purely based on the mapping of input and output data. The feed forward neural network topology is widely used one and consists of three layers as in Fig3.The network is trained based on the input drawn from experiments. Training is carried out by Leven berg-marquardt (LM) back propagation algorithm. The training is stopped when the mean square error (MSE) reaches a minimum value specified in convergence criteria or based on the maximumm iterations (epoch).it is shown in Figure 4. The MSE is calculate das below. (3 MSE1/N ek $ Where N=Total number of epochs (iterations) i=epoch number e(k)=error between the network output and the target output. 3) 204

6 Prediction of Roughness in Hard Turning of AISI 4140 Steel through Artificial Neural Network and Regression Models Figure 4 The ANN training process In this work the cutting speed, feed and depth of cut are taken as input. The roughness is the output. After the training process using LM algorithm, the network finalized is For this network the training is stopped based on the maximum value of iterations. The MSE plot for this network is shown in Figure 5. Figure 6 shows the overall regression plot. The network is validated based on the condition given in Table Best Training Performance is e-05 at epoch Train Best Goal Mean Squared Error (mse) Epochs Figure 5 MSE plot of selected network 205

7 D. Rajeev, D. Dinakaran and S. Muthuraman Training: R= Data Fit Y = T Output ~= 1*Target Target Figure 6 Overall Regression plot 6. RESULTS AND DISCUSSION The predicted values from Regression and ANN models are compared with the experimental values and the results are shown in Table3, 4.The validation is based on the data drawn from random cutting conditions. The errors (%) for the predicted values are also calculated based on the deviation. The average error of regression model is %, whereas the error percentage of ANN model is as low as 3.94%.The lower error values indicates higher prediction ability. The ANN model prediction ability is higher compared to the regression. The details are shown in Figure7. Training with more data can make the prediction ability higher. The error is calculated based on the relation Where R aexp = Experimental Roughness Ra predicted =Predicted Roughness Error= :; x100 (4) Table 3 Validation data and Predicted values-regression model S.No Input Parameters Surface Roughness Exp. Predicted Error V(m/min) f(mm/rev) d(mm) R a (µm) Ra(µm) (%) Average Error

8 Prediction of Roughness in Hard Turning of AISI 4140 Steel through Artificial Neural Network and Regression Models S.No V(m/min) Table 4 Validation data and Predicted values-ann model Input Parameters Exp. Surface Roughness Predicted Error f(mm/rev) d(mm) R a (µm) Ra(µm) (%) Average Error CONCLUSION In this work, the flank wear prediction model has been formulated based on the data obtained during hard turning of AISI4140.The models generated include multi regression and ANN. The prediction ability of both models are compared to various cutting conditions randomly chosen, even though both models give reasonably good results within the range of data considered, The ANN is comparatively better. The prediction model based on cutting conditions will pave the way for online optimization and adaptive control of cutting parameters, during real time machining. The accuracy indicates that the models can be considered for industrial use. REFERENCE Figure 7 Predicted Vs Measured Plot [1] Bouacha K, Yallese M A, Mabrouki T and Rigal J-F, Statistical Analysis of Surface Roughness and Cutting Forces using Response Surface Methodology in Hard Turning of AISI Bearing Steel with CBN tool,international Journal of Refractory Metals and Hard Materials, Vol. 28(2010): [2] H.M. Lin, Y.S. Liao and C.C. Wei, Wear Behavior in turning High Hardness Alloy Steel by CBN Tool, Wear,Vol. 264 (2008)

9 D. Rajeev, D. Dinakaran and S. Muthuraman [3] M.A. Yallese, K. Chaoui, N. Zeghib, L. Boulanouar and J.F. Rigal, Hard Machining of Hardened Bearing Steel using Cubic Boron Nitride Tool, Journal of Materials Processing Technology,Vol. 209 (2009) [4] U. Çaydas, Machinability Evaluation in Hard Turning of AISI 4340 Steel with Different Cutting Tools using Statistical Techniques, Journal of Engineering Manufacture,Vol. 224 (2009) [5] Ashok Kumar Sahoo and BidyadharSahoo, Performance Studies of Multilayer Hard Surface Coatings (TiN/TiCN/Al2O3/TiN) of Indexable Carbide Inserts in Hard Machining: Part-II (RSM, grey relational and techno economical approach), Measurement,Vol. 46 (2013) [6] C. Scheffer, H. Kratz, P.S. Heyns, F. Klocke, Development of a tool wear-monitoring system for hard turning, International Journal of Machine Tools & Manufacture,Vol. 43 (2003) [7] Zhang J, Chen J, The development of an in-process surface roughness adaptive control system in end milling operations,int. J.Adv. Man. Technol., Vol. 31 (2007) [8] Ersan Aslan, Necip Camuscu and Burak Birgoren, Design Optimization of Cutting Parameters when Turning Hardened AISI 4140 Steel (63 HRC) with Al2O3 + TiCN Mixed Ceramic Tool, Materials and Design,Vol. 28 (2007) [9] Haouici and M.A Yallese, Experimental Investigation of Cutting Parameters influence on Surface Roughness and Cutting forces in Hard Turning of X38CrMoV5-1 with CBN Tool, Sadhana,Vol. 38 (2013) [10] Y. Sahin, Comparison of Tool Life between Ceramic and Cubic Boron Nitride (CBN) Cutting Tools when Machining Hardened Steels, Journal of Materials Processing Technology, Vol. 209 (2009) [11] P. Subhash Chandra Bose, C S P Rao, Modelling and Prediction of Surface Roughness, Cutting Force and Temperature While Machining Nimonic-75 and NICROFER C-263 Super Alloys Using Artificial Neural Network (ANN). International Journal of Mechanical Engineering & Technology (IJMET), 3(3), 2012, pp [12] Saeed Zare Chavoshi& Mehdi Tajdari, Surface roughness modelling in hard turning operation of AISI 4140 using CBN cutting tool,int J Mater Form,Vol. 3 (2010) [13] Tugrul O Zel, Yigit Karpat, Predictive modeling of surface roughness and tool wear in hard turning using regression and neural networks, International Journal of Machine Tools & Manufacture, Vol. 45 (2005) [14] Harun Akkus and Đlhan Asilturk, Predicting Surface Roughness of AISI 4140 Steel in Hard Turning Process through Artificial Neural Network Fuzzy Logic and Regression Models, Scientific Research and Essays,Vol. 6 (2013) [15] P. R. Prabhu, S. M. Kulkarni, S. S. Sharma and Jagannath K, Surface Layer Alterations in AISI 4140 Steel from Turn-Assisted Deep Cold Rolling Process. International Journal of Mechanical Engineering & Technology (IJMET), 5(9), 2014, pp [16] D.E. Dimla Sr., P.M. Lister, On-line metal cutting tool condition monitoring II: tool-state classification using multi-layer perceptron neural networks,international Journal of Machine Tools & Manufacture,Vol. 40 (2000) [17] Xiaoyu Wang, Wen Wang, Yong Huang, Nhan Nguyen, Kalmanje Krishna kumar, Design of neural network-based estimator for tool wear modelling in hard turning, J IntellManuf, Vol. 19 (2008)