Advances in Production Engineering & Management Volume 10 | Number 2 | June 2015 | pp 59-72 http://dx.doi.Org/10.14743/apem2015.2.192 ISSN 1854-6250 Journal home: apem-journal.org Original scientific paper Modeling and optimization of parameters for minimizing surface roughness and tool wear in turning Al/SiCp MMC, using conventional and soft computing techniques Tamang, S.K.a, Chandrasekaran, M.a aMechanical Engineering Department, North Eastern Regional Institute of Science and Technology, Nirjuli, India A B S T R A C T A R T I C L E I N F O Aluminium alloy with silicon carbide particulate (Al/SiCp) reinforced metal matrix composite (MMC) are used within a variety of engineering applications due to their excellent properties in comparison with non-reinforced alloys. This presented work attempted the development of predictive modeling and optimization of process parameters in the turning of Al/SiCp MMC using a titanium nitride (TiN) coated carbide tool. The surface roughness Ra as product quality and tool wear VB for improved tool life were considered as two process responses and the process parameters were cutting speed v, feed f, and depth of cut d. Two modeling techniques viz., response surface methodology (RSM) and artificial neural network (ANN) were employed for developing Ra and VB predictive models and their predictive capabilities compared. Four different RSM models were tried out viz., linear, linear with interaction, linear with square, and quadratic models. The linear with interaction model was found to be better in terms of predictive performance. The optimum operating zone was identified through an overlaid contour plot generated as a response surface. Parameter optimization was performed for minimizing Ra and VB as a single objective case using a genetic algorithm (GA). The minimum Ra and VB obtained were 2.52 |m and 0.31 mm, respectively. Optimizations of multi-response characteristics were also performed employing desirability function analysis (DFA). The optimal parameter combination was obtained as v = 50 m/min, f = 0.1 mm/rev and d = 0.5 mm being the best combined quality characteristics. The prediction errors were found as 4.98 % and 3.82 % for Ra and VB, respectively, which showed the effectiveness of the method. © 2015 PEI, University of Maribor. All rights reserved. Keywords: Metal matrix composite Surface roughness Tool wear Response surface methodology Artificial neural network Genetic algorithm Desirability function analysis *Corresponding author: mchse1@yahoo.com (Chandrasekaran, M.) Article history: Received 28 July 2014 Revised 29 March 2015 Accepted 10 April 2015 1. Introduction The application and use of metal matrix composites (MMC) in manufacturing industries have now become increased due to its improved properties viz., high strength, low weight, high wear resistance, low heat of thermal expansion, etc. [1]. The matrix phase and reinforcement design of the material is responsible for the desired property of MMC. Among different types of MMC available, aluminium based SiC particulate (SiCp) reinforced MMC have found useful application as engineering material [2]. The conversion of these materials into an engineering part or component is obtained by machining through common conventional machining processes like turning, milling, drilling, and grinding. Turning is considered as foremost common machining method because of its ability to machine cylindrical surfaces faster with reasonably good surface finish. Due to hard and abrasive characteristic of reinforcement materials used in MMC the ma- 59 Tamang, Chandrasekaran chinability study, development of predictive modeling and optimizing the process parameters have attracted the researchers. Most of the research on MMC machining is concentrated on investigation of cutting tool wear, surface roughness of the machined product, delamination factor of drill holes produced, and metal removal rate during machining. Yuan and Dong [3] studied on surface finish in precision turning of MMCs using diamond tool. They considered spindle speed, feed rate, cutting angle, volume percentage of reinforcement material as investigating parameters. Davim [4] used Taguchi's orthogonal array and analysis of variance (ANOVA) to investigate the cutting characteristics of MMC (A356/20/SiCp-T6) in turning using polycrystalline diamond (PCD) cutting tool. Cutting velocity, feed rate, and cutting time are considered as input parameters and found that the cutting velocity has the highest physical and statistical influence on the tool wear and cutting power. Feed have high influence on the surface roughness of the component. Muthukrishnan and Davim [5] also conducted an experimental study on turning of Al/SiCp (20 %) MMC using the PCD tool for prediction of the surface roughness and found that the feed rate is a highly influencing parameter. Palanikumar and Karthikeyan [6] have studied on surface roughness using Taguchi method combined with RSM for minimizing the surface roughness in machining GFRP composites with PCD cutting tool. They concluded that fiber orientation and machining time are more influencing parameters on machining for obtaining better surface roughness. Rajasekaran et al. [7] also investigated the influence of surface roughness in turning CFRP composite using cubic boron nitride (CBN) cutting tool and applied fuzzy logic technique for modeling. They found that feed has the greater impact on surface roughness and fuzzy logic model predicts better. The influence of tool wear on machining glass fibre-reinforced plastics (GFRP) composites was investigated by Palanikumar and Davim [8] conducting series of experiments. They used ANOVA technique to assess the influencing parameters. Chandrasekaran and Devarasiddappa [9] used fuzzy logic for developing surface roughness model for end milling of Al/SiCp metal matrix composite with carbide cutter. They found that the model predicts with an average prediction error of 0.31 % when compared with experimental data. The surface roughness is influenced by feed rate and spindle speed while depth of cut has less influence. In comparing the performance of ANN model with RSM they found that ANN outperforms. Arokiadass et al. [2] also developed surface roughness prediction model for end milling of LM25Al/SiCp MMC using RSM technique. They also have taken influencing parameters as feed rate, spindle speed, depth of cut and SiCp percentage and found that feed rate is the most dominant parameter and depth of cut is of least influence on the surface roughness. Thiagarajan and Sivaramakrishnan [10] conducted an experimental study for investigating the grindability of Al/SiCp MMC in a cylindrical grinding process. They considered wheel velocity, work piece velocity, feed, depth of cut and SiCp volume fraction percentage as input parameters. They observed that the improved surface roughness and damage free surfaces are obtained at high wheel and workpiece velocity while using white A^O3 grinding wheels. A numerical model based GA optimization methodology has been applied by Davim et al. [11] for determination of optimal drilling conditions in A356/20/p metal matrix composites. The experimental study inferred that the surface finish of the drilled holes increase with increase in feed rate but does not change significantly with variation in cutting speed. Basavarajappa et al. [12] have studied the variation of surface roughness on the drilling of metal matrix composites using carbide tool. They also found that the surface roughness decreases with the increase in cutting speed and increases with the increase in feed rate. Chandrasekaran and Devarasiddappa [13] developed a surface roughness prediction model using artificial neural network (ANN) for grinding of MMC components. The input parameters are wheel velocity, feed, work piece velocity and depth of cut. They found that surface roughness is highly influenced by feed and wheel velocity but least effected by depth of cut Hocheng and Tsao [14] compared the RSM and radial basis function network (RBFN) for core-center drilling of composite materials. They concluded that for evaluating thrust force RBFN is more practical and predict better than the RSM method. Drilling CFRP composites have investigated by Tsao and Hocheng [15] using Taguchi and neural network methods. They conducted an experiment using Taguchi L27 orthogonal array of experiments with feed rate, spindle speed and drill diameter as input parameters. Thrust force and 60 Advances in Production Engineering & Management 10(2) 2015 Modeling and optimization of parameters for minimizing surface roughness and tool wear in turning Al/SiCp MMC, using ... surface roughness produced were output parameters and it has been found that the feed rate and drill diameter are most significant factors for predicting the thrust force. They also confirmed that RBFN model is found to be more effective than multiple regression analysis in predicting the output responses, i.e. surface roughness and thrust force. From review of above literatures the machining investigation on turning Al/SiCp MMC was performed by the researchers. They were mainly considered mainly single response and simultaneous modeling and optimization of surface roughness and tool wear were not attempted. These responses are important for manufacturing industries on the basis of job quality and longer tool life. In the area of modeling and optimization the researchers were carried out by a number of traditional and soft computing techniques. Application of GA found successful by number of researchers, Mukherjee and Ray [16], and Wang and Jawahir [17]. Oktem et al. [18] used RSM coupled with GA to optimize the cutting conditions for obtaining minimum surface roughness in milling of mold surfaces. For optimizing multi-response characteristics, various researchers use GRA as useful tool. The method does not require mathematical computation and can be applied easily for multi-response problems. Pawade and Joshi [19] have attempted to optimize the highspeed turning of Inconel 718 to optimize machining parameters using grey relational analysis considering cutting speed, feed, depth of cut and edge geometry as input parameters and surface roughness and cutting force as responses. Sahoo and Pradhan [20] carried out an experiment study based on Taguchi L9 orthogonal array in turning Al/SiC MMC using uncoated carbide tool. Three cutting parameters viz., cutting speed v, feed rate f and depth of cut d were optimized to obtain minimum flank wear and surface roughness. Low and high cutting speed was found as optimum parameter for VB and Ra, respectively. They also developed a linear mathematical model for VB and Ra and found statistically significant as P-value is less than 0.05. In another attempt, Sahoo et al. [21] performed turning experiments on Al/SiC MMC (10 % weight) produced by traditional casting process. Multi-layer coated carbide tool was used to investigate tool wear and surface roughness. They found that cutting speed is the most influencing machining parameter on flank wear and feed rate on surface roughness. They also carried out multi-objective optimization using grey relational grade and found optimum combination as cutting speed at 180 m/min, feed at 0.1 mm/rev, and depth of cut at 0.4 mm. Gopalakannan and Thiagarajan [22] investigated on Al/SiCp MMC using EDM process. Pulse current, gap voltage, pulse on time and pulse off time were considered as input parameters and metal removal rate, electrode wear rate and surface roughness were output parameters. The developed RSM models show good predictive capability. The parameters were optimized using desirability analysis for multiple objectives. The present work is envisaged to develop a modeling and optimization of machining parameters on the performance characteristics in turning of Al/SiCp MMC using TiN coated cutting tool. Predictive modeling was developed for surface roughness Ra and tool wear VB using RSM and ANN techniques. Machining parameters are optimized for single- and multi-objective case using GA and DFA for minimize Ra and VB or both simultaneously. 2. Development of RSM mathematical model The statistical tools such as multiple regression analysis, response surface methodology and Taguchi method are widely used for development of conventional predictive modeling. RSM is a collection of mathematical and statistical techniques for empirical model building. It is used for the problems in which an output parameter is influenced by several input parameters and the objective is to optimize the output response. In this work RSM model is developed in order to investigate the influence of machining parameters (i.e., cutting speed v, feed rate f, and depth of cut d on the surface roughness Ra and tool flank wear VB in turning Al/SiCp MMC. All the machining parameters were chosen as independent input variables while desired responses are assumed to be affected by the cutting parameters. The predicted surface roughness (response surface) of turning process can be expressed in term of the investigating independent variables as Ra = Cvxfydz (1) Advances in Production Engineering & Management 10(2) 2015 61 Tamang, Chandrasekaran where Ra is the predicted surface roughness in |j.m, v is the cutting speed in m/min, f is the feed in mm/rev, and d is the depth of cut in mm. C is the constant and x, y, and z are the exponents to be estimated from experimental results. Eq. 1 is linearized using logarithmic transformation and can be expressed as lnRa = xlnv + ylnf+ zlnd (2) Eq. 2 is re-expressed into generalized linear model as: 3 y = P0X0 +P1X1 +02*2 + &*3 = Po + ^ (3) i where y is true (measured) response surface on logarithmic scale, xo is dummy variable and its value is equal to 1, and xi, X2, and X3 are logarithmic transformation of input variables, i.e. cutting speed, feed, and depth of cut, respectively. fi0, p1, p2, and p3 are the parameters to be estimated. If s is the experimental error between estimated response y' and measured response y then y' = y-£ = b0x0 + bixi + b2x2 + b3x3 (4) where the b values are the estimate of (3 parameters. The linear model of Eq. 4 is extended as second-order polynomial response surface model (i.e., quadratic model) and is expressed as y' = y-£ = b0X0 +b1x1 + b2X2 + ¿3*3 + bltX^ +¿22*2 + ¿33*3 +¿12*1*2 +¿13*1*3 (5) +¿23*2*3 or 3 3 2 3 y' = ¿0 + ^ ¿i*i buxf ^ bijXiXj (6) i=l i=l i=lj=2 where bo is constant or free term, b, ba, and bij represent the coefficients of linear, quadratic, and cross product (i.e., interaction) terms. The Eq. 5 can be written as to build the relationship between turning parameters and responses (i.e., surface roughness and tool wear) as jRa =¿0 +¿1^ + ¿2/ + ¿3^ + ¿ll^2 +b22f2 +b33d2 +¿12*1*2 +¿13*1*3 +¿23*2*3 (7) yvB =b0+b1v + b2f + b3d + b11v2 + b22f2 +b33d2 + b12x1x2 +b13xtx3 + ¿23*2*3 (8) Where b0 is constant or free term, bi, ba, and bij represent the coefficients of linear, quadratic, and cross product (i.e., interaction) terms. The experimental work carried out by Kilifkap et al. [23] in turning Al/SiCp MMC using K10 TiN coated cutting tool for investigating surface roughness and tool wear is used in this work. For modeling and analysis of machining parameters RSM model is developed using MINITAB 15® statistical software. Table 1 show various machining parameters used at three levels. The RSM predictive model is developed using 20 data sets selected based on central composite design (CCD). The CCD experimental design matrix and responses are given in the Table 2. It is used for analyzing the measured response and determining the mathematical model with best fits. The fit summary for surface roughness and tool wear suggests that the quadratic relationship where the additional terms are significant and the model is not aliased. Table 1 Assignment of levels and parameters Factor Units Symbol Levels -l O l Cutting speed m/min v 5O 1OO 15O Feed mm/rev f O.l O.2 O.3 Depth of cut mm d O.5 l.O 1.5 62 Advances in Production Engineering & Management 10(2) 2015 Modeling and optimization of parameters for minimizing surface roughness and tool wear in turning Al/SiCp MMC, using ... Table 2 Experimental result Cutting speed, v (m/min) Tool feed, /(mm/min) Depth of cut, d (mm) Experimental responses Sl. No Code (A) Actual value Code (B) Actual value Code (C) Actual value Surface roughness, Ra (|im) Tool wear, VB (mm) 1 -1 50 1 0.3 1 1.5 4.13 0.601 2 1 150 1 0.3 1 1.5 3.17 1.050 3 -1 50 1 0.3 1 1.5 3.95 0.447 4 0 100 -1 0.1 0 1.0 3.21 0.603 5 0 100 1 0.3 0 1.0 4.03 0.702 6 1 150 1 0.3 -1 0.5 3.47 0.902 7 -1 50 -1 0.1 1 1.5 3.34 0.502 8 0 100 0 0.2 -1 0.5 3.47 0.630 9 0 100 0 0.2 0 1.0 3.40 0.651 10 -1 50 -1 0.1 -1 0.5 3.24 0.327 11 1 150 0 0.2 0 1.0 3.27 0.896 12 0 100 0 0.2 0 1.0 3.40 0.651 13 0 100 0 0.2 0 1.0 3.40 0.651 14 1 150 -1 0.1 0 1.0 3.17 0.623 15 0 100 0 0.2 1 1.5 3.43 0.698 16 1 150 -1 0.1 1 1.5 3.14 0.602 17 0 100 0 0.2 0 1.0 3.40 0.651 18 0 100 0 0.2 0 1.0 3.40 0.651 19 0 100 0 0.2 0 1.0 3.40 0.651 20 -1 50 0 0.2 0 1.0 3.68 0.477 Four different types of RSM mathematical models viz., linear, linear with interaction, and quadratic are obtained for prediction of surface roughnessyRa and tool wearyvB were obtained. a) Linear model: yRa = 3.367 - 0.0042v + 2.65/ - 0.018d yB = -0.0093 + 0.00344v + 1.045/ + 0.1045d b) Linear with interaction models: yRa = 2.382 + 0.00217v + 8.41/ + 0.313d - 0.034vf - 0.00009vd - 1.95/d yVB = 0.320 + 0.0018v - 1.63/ + 0.127d + 0.018v/ - 0.00149vd + 0.612/d c) Linear with square models: yRa = 3.28 - 0.0026v - 2.13/ + 0.88d -0v2 + 12.17/2 - 0.423d2 yB = -0.053 + 0.0037v + 2.46/ - 0.039d - 0v2 - 3.63/2 + 0.044d2 d) Quadratic models: yRa = 2.55 + 0.0022V + 4.086/ + 0.737d - 0.000v2 + 12.84/2 - 0.227d2 - 0.035v/ a - 0.0009vd - 2.47fd yVB = 0.103 + 0.0026V - 0.55/ + 0.288d - 4.114/2 - 0.066d2 + 0.0203v/ - 0.002vd + 0.877/d (9) (10) (11) (12) (13) (14) (15) (16) Advances in Production Engineering & Management 10(2) 2015 63 Tamang, Chandrasekaran where v, f, and d are cutting speed, feed and depth of cut, respectively. From these model equations, it is observed that the factor with highest value of coefficient posses the most dominating effect over the response. Feed has most significant effect over surface roughness and tool wear followed by the depth of cut and cutting speed. 2.1 Checking adequacy of the model The test of significance of all the models was carried out using analysis of variance (ANOVA) and their predictive capability is analyzed. ANOVA find the influence of machining parameters (v, f and d) on the total variance of the experimental findings. The test is performed by calculating the ratio between the regression mean square and the mean square error (i.e., F-ratio). The ratio measures the significance of the model in respect of variance of the parameters included in the error term for particular level of significance a. The analysis was carried out at 95 % confidence level and the result is presented in Table 3. The adequacy of the model is decided upon the value of S and coefficient of determination R2. S value being the measurement of error, it is the smaller value that gives better results. If R2 approaches unity the response model fits better with the actual data and less difference exists between predicted and actual data. To compare, more precisely adjusted R2 (Adj R2) is used, which is adjusted for the degrees of freedom. The closeness of the Adj R2 with R2 determines the fitness of the model. The higher value of R2 is obtained for linear with interaction model. This shows the predictive capability of linear with interaction model is found better and is selected among all models. The model equation used for predictionofsurface roughness and tool wearis given in Eq.11andEq.12, respectively. Table 3 Test of significance of RSM models Sl. RSM model S-Value R2 Adj R2 No. Ra VB Ra VB Ra VB 1 Linear 0.15 0.073 76.09 82.51 71.01 79.21 2 Linear with interaction 0.089 0.052 96.00 92.16 94.12 90.00 3 Linear with square 0.15 0.078 80.17 83.59 70.94 76.02 4 Full quadratic 0.089 0.046 94.86 95.63 89.78 91.69 2.2 Contour plots Fig. 1 shows two dimensional surface plot that shows the effect of influencing parameters on the output responses. Fig. 1(a) reveals that higher cutting speed and lower feed produces better surface finish. Increased feed increases the surface roughness value. This is due to rapid tool movement which deteriorates the quality of the machined surface. The analysis of contour plot shows improved surface roughness is obtained at higher v and lower f. The combination of parameters with cutting speed at 150 m/min, feed at 0.1 mm/rev, and depth of cut at 0.5 mm produces minimum surface roughness of 3.17 |j.m. The tool wear contour plots are shown in Fig. 1(b). Cutting speed is the influencing parameter followed by depth of cut and feed. Higher tool wear is noticed at increased v. This is due to increased temperature causing flank wear at tool nose. Tool wear plot shows reduced tool wear is obtained at lower values of v, f, and d. The combination of parameters with cutting speed at 50 m/min, feed at 0.1 mm/rev, and depth of cut at 0.5 mm produces tool wear less than 0.4 mm found as minimum. The comparison of experimental and RSM prediction for the parameters combination that produces minimum surface roughness and minimum tool wear are presented in the Table 4. However, the optimum region for combined minimization of surface roughness and tool wear is obtained by overlaying contour plot presented in the next subsection. 64 Advances in Production Engineering & Management 10(2) 2015 Modeling and optimization of parameters for minimizing surface roughness and tool wear in turning Al/SiCp MMC, using ... 0.30 Î 0.25 £ a 41* eS T3 CJ Ü — 0.20 0.15 0.10 Contour plot for R Vs v and/ Rj ■ < 3.2 ■ 12- 3.4 14- 3.6 E3 3.6 - 3.8 4,0 l 4.2 Mold Values d 03 0.30 0.25 0.20 0.15 a b 0.10 Contour plot for VB Vs v and/ 75 100 125 150 Cutting speed, v (m/min) - 50 75 100 125 Cutting speed, v (m/min) (a) For surface roughness (b) For tool wear Fig. 1 Contour plots for interaction effect (at d = 0.5 mm) 150 Table 4 Optimum parameter combination Sl. No. Turning parameters (v-f-d) Expt. RSM prediction Error (%) 1 For minimum Ra (150-0.1-0.5) 3.17 |m 3.18 |m 0.32 2_For minimum VB (50 -0.1- 0.5)_0.33 mm_0.38 mm_13.15 2.3 Overlaying contour plot for optimum operating zone Fig. 2 shows the region for the selection of optimum cutting speed and feed for different value of surface roughness with minimum tool wear. The range of cutting speed as 50-80 m/min and feed as 0.1-0.14 mm/rev with 0.5 mm depth of cut produce surface roughness less than 3.4 |m with tool wear less than 0.5 mm. It may be considered as optimum operating zone. Similar trend have been seen at all values of depth of cut. The method of overlaying contour plot pictorially obtains the optimum operating zone and easy selection of cutting parameters for different values of Ra. Fig. 2 Optimum operating region Advances in Production Engineering & Management 10(2) 2015 65 Tamang, Chandrasekaran 3. Multi-response artificial neural network modeling Artificial neural network (ANN) is the system that acquire, store and utilize knowledge gained from experience. It is motivated by the biological neurons that work in human brain. Researchers have employed ANN for modeling of machining processes and found that ANN provides reasonable accuracy. The network is built with number of layers (input, hidden and output) having specific number of neurons (also called nodes). All the neurons are interconnected with weights and bias is added at each node. The number of neurons in the input and output layers depend upon input and output parameters of the proposed model. The number of neurons of the hidden layer is decided during network training. The network architecture is trained with the number of real life experimental datasets. Each dataset consists of input parameters and the corresponding output responses. The optimum network is obtained with the selection of appropriate transfer functions and number of neurons in the hidden layer. The mean square error between the experimental response and ANN prediction is the criteria for deciding the optimum network architecture. Once network is trained then it is ready for prediction. The trained network is tested with unseen datasets for model validation and the predictive results are compared with experimental results. The size and selection of training and testing datasets are very crucial in the design of ANN model. There is no well- established formula for finding out the number of training and testing data [24]. Kohli and Dixit [25] have used 19 datasets for training 9 datasets for testing in developing ANN model used for prediction surface roughness in turning process. Nearly 66 % of total experimental data sets are selected is the training phase. The data sets are selected appropriately including extreme datasets (i.e., Vmin, fmin, and dmin; Vmax, fmax, and dmax). The remaining 34 % datasets were used in the testing phase. The predictive results of the tested data sets are compared with experimental datasets. In this work, a soft computing based artificial neural network model for predicting surface roughness and tool wear as a function of three input parameters viz., cutting speed, feed, and depth of cut is developed. The multi-layer perceptron (MLP) network comprised of an input layer with three neurons, a hidden layer, and an output layer with two neurons. The networks with neurons (nodes) in each layer are interconnected with nodes of the subsequent and preceding layer with synaptic weights. Additionally a bias is added to each neurons of the hidden and output layer. The output of each neuron is obtained by summing up weighted inputs of neuron in preceding layer and its own bias. The output of each neuron in the hidden or output layer is computed by the equation where Wij is the associated weights with jth neurons of the layer and zth neurons of the preceding layer, bj is the bias of jth neurons, n is the total number of neurons of the preceding layer and f is the appropriate transfer function used. In this work, the ANN model is trained with 19 experimental datasets and tested with eight unseen datasets. Fig. 3 shows the architecture of two layered feed forward neural network system used in this work. The network is modeled with neural network tool box available in MATLAB® that working on back propagation learning algorithm. The algorithm use gradient decent technique and minimize mean square error (MSE) between actual network outputs with desired output pattern. (17) 66 Advances in Production Engineering & Management 10(2) 2015 Modeling and optimization of parameters for minimizing surface roughness and tool wear in turning Al/SiCp MMC, using ... Cutting speed : I, Feed rate : 1, Depth of cut: I, I_I Input layer Hidden layer Output layer Wij and b, are weights and bias of hidden layer, respectively Vij and Ci are weights and bias of output layer, respectively Fig. 3 ANN architecture The network is optimized with varying number of neurons in the hidden layer and activation transfer function used so as to obtain minimum MSE. The network architecture with five hidden layer neurons with tansig transfer function obtains least MSE of 0.0001 and is considered as optimum network. The output layer uses purelin transfer function to evaluate the estimated outputs of surface roughness and tool wear. The validation of the network is performed by predicting surface roughness and tool wear for unseen data sets and ANN prediction is compared with experimental result. 3.1 Comparison of RSM and ANN model performance The ANN and RSM predicted values for surface roughness and tool wear is compared with the experimental values. The comparison of predictive performance of both the models with the experimental value is given in Table 5. The prediction accuracy PA of each datasets was calculated using Eq. 18. PA = abs(Expt_valuei — Model_predt) Expt_valuei x 100 (18) Finally, the model accuracy MA is computed as the average of individual accuracy on confirmation data set. It is obtained using Eq. 19. MA n ¿ = 1 x 100 (19) The model accuracy of the ANN and RSM model are 95.38 % and 92.90 % for surface roughness and 92.16 % and 91.56 % for tool wear. It can be concluded that the correlation between the prediction of developed models and experimental result is very good. The prediction accuracy in ANN for surface roughness and tool wear is more than 95.00 %. The prediction accuracy for RSM based on linear with interaction model found more than 91.00 % for predicting surface roughness with a maximum PA of 99.69 %. While for tool wear PA is more than 90.0 % with the maximum of 98.64 %. This shows that neural network based prediction model has been found better than the response surface model for turning Al/SiCp metal matrix composite using coated TiN tool. Advances in Production Engineering & Management 10(2) 2015 67 Tamang, Chandrasekaran Table 5 Comparison of ANN and RSM predictive model Sl. No. Surface roughness, Ra Tool wear, VB ANN RSM Expt. (mm) ANN RSM Expt. (im) Pred. (im) Pred. acc. (%) Pred. (im) Pred. acc. PA (%) Pred. (mm) Pred. acc. PA (%) Pred. (mm) Pred. acc. PA (%) 1 3.27 3.48 93.96 3.28 99.69 0.508 0.405 79.72 0.45 88.58 2 3.87 4.16 93.02 3.79 97.93 0.400 0.453 88.30 0.35 87.50 3 4.67 4.49 96.15 4.20 89.93 0.521 0.493 94.63 0.43 82.53 4 4.04 3.59 88.86 3.68 91.08 0.799 0.783 97.99 0.81 98.64 5 4.16 4.37 95.88 3.96 95.19 0.685 0.707 96.89 0.63 91.97 6 3.08 3.00 97.40 3.14 98.08 0.653 0.677 96.46 0.66 98.93 7 3.79 3.78 99.74 3.32 87.59 0.750 0.792 94.70 0.81 92.59 8 4.06 4.02 99.01 3.41 83.99 0.951 0.842 88.54 1.04 91.44 Model accuracy 95.50 92.94 Model accuracy 92.15 91.52 4. Optimization of cutting parameters The selection of best or right combination of cutting parameters for obtaining optimum process response is still the subject of many studies. In this work the parameter optimization for single as well as multiple objectives is carried out. Optimization for minimum Ra and minimum VB are performed using the non-traditional techniques of genetic algorithm (GA). The optimum parameters are also obtained for simultaneous optimization of Ra and VB using desirability function analysis (DFA). 4.1 Single-objective optimization with GA GA is one of the popular optimization technique performed by the natural evolution process inspired on the principle of survival of fitness [26]. GA works on the mechanism of genetics and evolution and has been found as a very powerful algorithm for obtaining global minima by Chandrasekaran et al. [27]. In GA the different process parameters are represented either binary or decimal numbers, called as string or chromosome. A set of chromosomes is called population. A population is evolved through several generations using different genetic operations such as reproduction, crossover, and mutation. The best chromosome in the population is identified by the closeness of fitness value with the objective function. The process is repeated till the optimization function converges to the required accuracy after many generations and optimum parameter is obtained. Researchers have found GA as powerful optimization tool/procedure to obtain global optima and the mathematical derivative of the function is not required in this procedure. In this work, the fitness/objective function of the optimization problem is formulated using the best regression model given in Eq. 20 and Eq. 21 for surface roughness and tool wear, respectively. The formulated single-objective optimization function is given as follows: Minimize Ra (v, f,d) = Min{2.382 + 0.00217v + 8.41/ + 3.313d - 0.034vf - 0.00009vd - 1.95/d) Minimize VB{y,f,d) = Min(0.320 + 0.0018v - 1.63/ + 0.127d + 0.018vf - 0.00149vd + 0.612/d) (20) (21) The variables of the function are limited by its upper and lower bounds and are given as 50 < v < 150 (22) 0.1