An experimental investigation of tool nose radius and machining parameters on TI-6 AL-4 V ( ELI ) using grey relational analysis , regression and ANN models

Article history: Received: October 28, 2018 Received in revised format: December 25, 2018 Accepted: January 10, 2019 Available online: January 10, 2019 Ti-6Al-4V Extra Low Interstitial (ELI) exhibits superior properties because of controlled interstitial element of iron and oxygen. The effects of four cutting parameters namely cutting speed, feed, depth of cut and tool nose radius on responses like cutting force, average cutting temperature and surface roughness have been investigated for turning of Ti-6Al-4V (ELI). Total 81 experiments have been performed in dry environment. Grey Relational Analysis has been used for multi-objective optimization. Analysis of Variance test has been carried out to investigate contribution of input parameters. The model was found fit with R-Square value of 88.74%. Regression and ANN models are developed for prediction and compared. From the Grey relational analysis, it is clear that optimum parameters to minimize cutting force, cutting temperature and surface roughness while turning Ti-6Al-4V (ELI), are cutting speed as 140 rpm, Nose radius 1.2mm, Feed 0.051mm/rev and depth of cut is 0.5mm. In comparison of regression model, the ANN model is found to be more accurate with average error of 3.57%. © 2019 by the authors; licensee Growing Science, Canada.


Introduction
Superior and favorable mechanical properties have made titanium alloys, a perfect choice in the applications of aerospace, biomedical and marine applications.High strength to weight ratio, better corrosion resistance, and good fracture toughness are attractive properties possessed by titanium alloys.Despite having complimentary properties, titanium alloys fall under the category of difficult to cut materials because of poor thermal conductivity and rapid tool wear.The high cutting temperature is an issue which requires high attention as it is responsible for poor machinability (Narutaki et al., 1983).Ti-6Al-4V and Ti-6Al-4V ELI (Extra Low Interstitial) are basically developed to be used as structural material but it has found wide application as implant material too (Niinomi, 1998).The extra low interstitial (ELI) grade of Ti-6Al-4V exhibits higher ductility and improved fracture stiffness than grade5 Ti-6Al-4V.This is because of controlled interstitial element of iron and oxygen.The investigation on diffusion bonding of Ti-6Al-4V ELI was also carried out, which revealed that it is possible to have super-plastic forming and diffusion boding at lower temperature than conventional Ti-6Al-4V (Lee et al., 2007).The components to be used in aerospace field are expected to have better surface integrity and higher reliability.The investigation by Che-Haron and Jawaid (2005) revealed that the surface integrity is more affected by feed and tool nose radius while machining Ti-6Al-4V ELI.In order to understand fatigue behavior of implant, the investigation on the relation between fatigue damage and mechanical properties of Ti-6Al-4V ELI was carried out by Akahori and Niinomi (1998).To evaluate the oxygen effect on processing of Ti-6Al-4V, the shapes of stress-strain curves, the kinetic parameters, and the processing maps obtained and have been compared for two grades of material (Prasad et al., 2001).Titanium alloys are used as implant materials for bio medical and dental application because of their corrosion resistance and good bio-compatibility.The corrosion behavior of titanium alloys like Ti-6Al-4V ELI and Ti-6Al-7Nb in simulated body fluids have also been investigated (Tamilselvi et al., 2006).
Turning is highly significant manufacturing process, in which single point cutting tool removes material from cylindrical work-piece while it is rotated.There are three cutting forces produced during turning namely thrust force, which acts in direction of cutting speed, feed force in the direction of feed and radial force which is produced in the direction normal to cutting speed.Effect of parameters on cutting power has been investigated by researchers (Valera & Bhavsar, 2014).Many researchers have contributed their work on optimization of process parameters in order to improve machinability of titanium alloys.Significance of cutting parameters on Tool life and surface roughness of Ti-6Al-4V ELI was investigated (Sulaiman et al., 2013).The findings show that feed rate and cutting speed were highly influencing factors for surface roughness.Tool nose radius also affects the surface properties of the product (Yildiz, Irez, & Sur, 2016).It has been observed that cutting speed and feed have more influence on cutting temperature (Nath et al., 2017).The geometry of cutting tool is also significant.Xie et al. (2013) investigated the effect of micro-grooved tool on cutting temperature and cutting force while dry turning of titanium alloy, and reported the decrease in cutting temperature with decrease in micro groove depth.After prolonged machining of titanium alloy under dry environment, tearing and plastic deformation of machined surface were observed (Che-Haron & Jawaid, 2005).To improve tool life during machining of titanium alloy, use of solid lubrication is a better option as it can perform cooling and lubrication simultaneously (Moura et al., 2015).Investigation of effect of cutting speed, feed and depth of cut on cutting temperature while turning hardened steel EN-36 was carried out by researchers (Gosai & Bhavsar, 2016).Grey Relational Analysis (GRA) is an effective tool for multi objective optimization.Many researcher have used the GRA method for optimization of parameters (Maiyar et al., 2013;Sarıkaya & Güllü, 2015;Vinayagamoorthy & Anthony Xavior, 2014).It has been effectively used for optimization of thermally enhanced machining parameters while turning Inconel 718 (Ganta et al., 2017).Optimization of cutter geometric parameters while end milling of titanium alloy was also carried out (Ren et al., 2015).Investigation of drilling parameters on hybrid polymer composite revels the important significance of parameters on delamination, thrust force and torque (Anand et al., 2018).In recent times artificial intelligence has drawn attention of many researchers.Amongst various methods based on artificial intelligence, Artificial Neural Network has been widely used by many researchers to predict the responses.The prediction of surface roughness has been predicted using ANN model and multiple regression method by Asiltürk and Çunkaş (2011).They concluded that ANN model is powerful tool for prediction as compared to multiple regression model.Machining of AISI 1030 steel by PVD and CVD coated tool by varying feed rate and cutting speed has been investigated, and the surface roughness was predicted by ANN model with acceptable accuracy (Nalbant et al., 2009).As per the literature survey, very limited research work has been carried out on simultaneous effect of cutting parameters and tool geometry on surface roughness, cutting temperature and cutting force while turning Ti-6Al-4V (ELI).In this study, an attempt has been made to investigate the effect of cutting speed, feed, depth of cut and tool nose radius on the cutting temperature and cutting force.Total 81 experiments have been carried out.
The experimental results have been used to calculate Grey relational grade (GRG).Mathematical regression and ANN models are developed for the prediction of GRG and the predicted values are compared with calculated GRG.ANOVA tests have been carried out to evaluate contribution of parameters.

Experimentation
The following is the explanation of procedure adopted for the performance of experiments.Tool material, work piece material, instruments and tooling have been described here in this section.

Workpiece and Tool
The material used for experiment is Ti-6Al-4V ELI (round bar with 70mm diameter, 250mm length).The chemical composition of work material has been shown in Table 1.The cutting inserts which have been utilized are coated cemented carbide inserts with ISO designation as TNMG 160404, TNMG 160408 and TNMG 160412 with nose radius 0.4mm, 0.8mm and 1.2mm, respectively.

Machining Tests
All experiments were performed in dry environment using CNC turning center STC-200 with a maximum spindle speed of 3500 rpm and a power rating of 9 KW.The maximum turning length of turning center was 400mm and the maximum turning diameter was 200mm.The cutting forces have been measured using strain gauge type lathe tool dynamometer.The strain gauge type 3-channel lathe tool dynamometer was having resolution of 0.01 Kg and accuracy of ±5 percent.The range of force was 0 to 200 in all three directions i.e. axial, radial and tangential.Cutting temperature was measured using MECO made infrared pyrometer (model IRT550P) for the range -50 0 C to 500 0 C. The surface roughness was measured by Mitutoyo SJ 410 having measuring range 800µm/0.01µm.Fig. 1 shows the machine tool, cutting tool and equipment used for the purpose of experimentation.

Normalizing or Preprocessing of Data
The measured responses were normalized by Grey relational method.The measured values of cutting force, cutting temperature and surface roughness were pre-processed to a sequence between zero and one.For normalizing in "higher-the-better" characteristic, the following equation is used.* , and for "lower -the-better" characteristic, following equation is used.* , where, = original value, * = value after normalizing, = maximum value of and * = minimum value of .
Here, in this study all the responses are required to be minimized; Eq. ( 2) is used for preprocessing/normalizing the data.The normalized data is shown in Table 4.

Grey Relational Coefficient and Grey Relational Grade
After normalization the grey relational coefficient (GRCi(k)) is calculated by Eq. ( 3) as follows, where, and are the maximum and minimum values in the normalized sequence, in this study they are 0 and 1 respectively.
is the absolute difference between and * (k), for i =1 to 81 and k = 1to 3, it is also named as deviation sequence. is coefficient of distinguishing, generally taken as 0.5.By averaging the values of GRC, Grey relational grade (GRG)  can be calculated by Eq. ( 4), where is normalized weight for response k.Here in this study all responses are given equal weight, hence the Eq. ( 4) can be written as, ( The calculated values of GRC and GRG are tabulated in Table 4 In present study multiple response like cutting force, average cutting temperature and surface roughness were optimized for turning of Ti-6Al-4V (ELI).The influence of parameters cutting speed, feed, depth of cut and tool nose radius is analyzed.

ANOVA and Grey Relational Analysis
Many researchers have worked on single objective optimization.In this study multi objective optimization is done using Grey relational Analysis.From Table 5, it is clear that the experiment number 55 is having maximum GRG value of 0.833.The process parameters for experiment number 55 are cutting speed as 140 rpm, Nose radius 1.2mm, Feed 0.051mm/rev and depth of cut is 0.5mm.These values of parameters are considered as optimum process parameters for turning Ti-6Al-4V (ELI) among the 81 experiments.The influence of cutting parameters on Grey Relational Grade are analyzed (See Fig. 3).In Fig. 3(a), effect of parameters on Grey relational grade is shown.It can be said from Fig. 3(a) that feed and cutting speed are significant factors for multiple responses when studied simultaneously.Increase in nose radius increases grey relational grade whereas it is decreased with increase in depth of cut.ANOVA tests were performed for statistical analysis of effect of turning parameters on Grey relational Grade.
Table 7 shows the results of ANOVA tests.It can be interpreted from Table 7 that Feed rate is having maximum influence followed by cutting speed, nose radius and depth of cut.The R-Square value for the model developed for GRG is 88.74%For better understating the effect of input parameters on Grey relational Grade, 3D surface plots are obtained.The effect of feed and depth of cut on GRG is shown in figure 3 (b).Increase in feed rated reduces the GRG value.The GRG value is initially increases and highly decreased with increase in depth of cut.Fig. 3(c) shows the effect of nose radius and cutting speed on GRG.The increase in cutting speed makes the GRG to be decrease.The GRG value highly increases by increase in nose radius.Using Eq. ( 6), the GRG values are predicted and compared with calculated values of GRG.The comparison is shown in Table 5.The average error is 7.63 %.

Artificial Neural Network
In order to develop more precise model, the artificial neural network is used.Initially input data, sample data and corresponding output data are created in Matlab workspace.Using these data a network is created in workspace.In present study, the Feed forward back propagation model has been used.From the data fed in the workspace, 75% are used for training, 12% for testing and 12% for validation purpose.

Conclusion
The effects of cutting parameters like cutting speed, feed, depth of cut and tool nose radius on responses like cutting temperature, cutting force and surface roughness have been investigated for Ti-6Al-4V ELI in present study.Following points have been concluded from the experimental study which has been carried out.
 For multi objective optimization, Grey Relational Analysis was used. From experimental data, mathematical models are developed for Grey Relational Grade using Regression method and Artificial Neural Network method. The optimum parameters to minimize cutting force, cutting temperature and surface roughness while turning Ti-6Al-4V (ELI), are cutting speed as 140 rpm, Nose radius 1.2mm, Feed 0.051mm/rev and depth of cut is 0.5mm. ANOVA test revels the R-Sq value of model as 88.74%  Pareto chart and ANOVA table of GRG, indicate cutting speed and feed are significant parameters followed by cutting speed and depth of cut. The average error while comparing calculated GRG and Regression GRG was 7.63%  The GRG values predicted by ANN model were having average error of 3.75%

FigFig. 1 .
Fig. 1.(e) CNC lathe Fig. 1.Machine and equipment used for experimentation Four different cutting parameters have been chosen for experimentation.Cutting speed, feed and depth of cut are process parameters and tool nose radius is the parameter of cutting tool geometry.Table2indicates cutting parameters and their levels which have been set to carry out experiments.In this study, experiments have been planned for four different parameters with three levels.According to full factorial design for four parameters having 3 levels, total of 3 4 = 81 experiments have been performed.The levels of parameters have been selected based on cutting tool supplier manual, trial runs of experiments and literature survey.The cutting parameters and measured responses have been presented in Table3.The effect of input parameters on responses like cutting force, cutting temperature and surface roughness were analyzed by main effect plots developed using Minitab-17.In order to investigate significance and contribution of individual parameters on multiple responses, Grey relational analysis is used; ANOVA test has been carried out on calculated Grey relational grade (GRG).ANOVA has also been utilized to model GRG.Regression equation is developed for the prediction of GRG.The effects of all input parameters on GRG are potted using 3D surface plots.The comparison of calculated and predicted values of GRG reveled the average error of 7.63%.Using the measure responses values, ANN model was developed for the prediction of GRG.The ANN model predicted the values of GRG with average error of 3.75%.The regression and ANN models are compared on a common graph.

Fig. 2
Fig. 2(a).Mean effect plots for Temperature Fig. 2(b).Mean effect plots for Cutting Force The mean effect plots for cutting parameters on temperature, cutting force and surface roughness are shown in Fig. 2(a), Fig. 2(b) and Fig. 2(c) respectively.From Fig. 2(a) it can be observed that increase in tool nose radius initially decreases the temperature and then the temperature increases with increase in nose radius.With the increase in cutting speed and feed the average cutting temperature rises.The depth of cut has lesser influence of cutting temperature.From Fig. 2(b), it can be interpreted that nose radius is having least influence on cutting force.The cutting force increases with increase of cutting speed and feed rate.The depth of cut is the maximum influence on cutting force while turning.

Fig. 2
Fig. 2(c).Mean effect plots for Surface Roughness Fig. 3(a).Mean effect plots for Grey relational grade Fig. 2(c) shows that surface roughness is highly influenced by tool nose radius.The increase in nose radius highly decreases the surface roughness.The effect on cutting force with change in depth of cut is high as compared to change in cutting speed and feed.

Fig. 3
Fig. 3(b).Surface plot of Grey relational grade v/s feed and depth of cut Fig. 3(c).Surface plot of Grey relational grade v/s cutting speed and nose radius 'TRAINLM' function was used for training and 'PURELIN' function was used as transfer function.Then the developed network was ready for training.The training was done until the predicted value matches nearer to the actual experimental values.The graphs for mean square error values for training, testing, validation and overall target data are shown in Figure 5.The predicted values of GRG by ANN are compared with calculated GRG values and are tabulated in

Fig. 6 .
Fig. 6.Comparative graph for calculated GRG v/s predicted values of GRG The models developed for GRG using Regression and ANN are compared and shown in Fig. 6.The calculated GRG, Re-GRG and ANN GRG are shown in comparative graph.It can be easily interpreted

Table Full -
Factorial Design of Experiments and Responses

Table 7
ANOVA test for GRG

Table 6
Mean response table for Grey relational gradeFrom Table6, it is clear that the cutting speed is significant factor while considering multiple responses simultaneously.The feed, depth of cut and nose radius are having influence in decreasing order.Pareto chart is prepared to analyze the contribution of cutting parameters on multiple responses.Figure4shows the Pareto chart for machining parameters on GRG value.It can be interpreted that feed and cutting speed pay significant contribution on measured responses like cutting temperature, cutting force and surface roughness while optimizing simultaneously.

Table 6 .
The average error is 3.75%.The fitness for ANN model is 95.54%.