Experimental Investigation and ANFIS-Based Modelling During Machining of EN31 Alloy Steel

This research presents the parametric effect of machining control variables while turning EN31 alloy steel with a Chemical Vapor deposited (CVD) Ti(C,N) + Al2O3 + TiN coated carbide tool insert. Three machining parameters with four levels considered in this research are feed, revolutions per minute (RPM), and depth of cut (ap). The influences of those three factors on material removal rate (MRR), surface roughness (Ra), and cutting force (Fc) were of specific interest in this research. The results showed that turning control variables has a substantial influence on the process responses. Furthermore, the paper demonstrates an adaptive neuro fuzzy inference system (ANFIS) model to predict the process response at various parametric combinations. It was observed that the ANFIS model used for prediction was accurate in predicting the process response at varying parametric combinations. The proposed model presents correlation coefficients of 0.99, 0.98, and 0.964 for MRR, Ra, and Fc, respectively.


Introduction
Metal cutting plays a crucial role in the field of manufacturing. It was found that turning factors such as feed, depth of cut, and speed influence material removal rate (MRR) and surface roughness (RA) [1]. However, most of these research findings are established with three levels of parameters and are for specific levels of these parameters. Thus, there is adequate scope for extensive research which takes into account extended boundary conditions. The Taguchi orthogonal array has emerged as one of the most widely adopted experimental designs owing to its ability in providing the most optimal parametric combinations [2]. Generalizations of the behaviors of the factors mentioned above and the specified process performance are difficult because the findings of the earlier researchers were for specific tool-material combinations, such as the range of process variables and given turning conditions. The feed rate indicates an effect on cutting force (FC) and RA. Moreover, the influence on depth of cut was rather low while performing a turning operation on AA2219-TiB 2 /ZrB 2 in a setup of a metal matrix composite using uncoated tungsten carbide inserts. The developed linear regression equation also showed a reliable agreement with the confirmation experiment during the turning operation [3]. The surface quality was obtained at low feed and depth of cut, while the speed was accurately. The relationships amongst the parameters and strength are commonly nonlinear and well taken by the ANFIS [22]. The comparison of RSM with ANFIS model during friction stir welding of AA2024-AA5083 aluminum alloys in relation to ultimate tensile strength shows that the developed ANFIS model is a powerful method as compared to the RSM model [23]. ANFIS with fuzzy inference systems (FIS), such as subtractive clustering, grid partition, and fuzzy c-means (FCM), was utilized for determining the cetane number. The result shows that all the fuzzy inference systems can determine the cetane number of fuel (FCM), and the grid partition shows higher desirability [24]. The adsorption process modelled with ANFIS shows a minimal error of total average error and total average of absolute error, and the coefficients of determination of the training data set were found to be 0.9999 or 0.9823 respectively, when estimating the efficacy of lead adsorption with functional nanocomposite adsorbent of hydroxyapatite (HAp)/chitosan [25]. The FFA and GA utilized for optimizing ANFIS parameters during bench blasting show that both GA and FFA are capable optimizers for improving the ANFIS prediction [26]. The ANFIS model implemented for prediction of vapor compression refrigeration system shows good agreement with experimental data and shows better statistical prediction efficiency [27]. The introduction of FEA optimizer with ANFIS parades a parsimonious modelling for streamflow forecasting by integrating a small number of factors essential to return the relatively strengthen performance [28]. The ANFIS model developed for estimating carbon dioxide loading abilities of amino acid salt solutions depicts that the developed model is sufficient to estimate the loading capabilities of CO 2 of amino acid salt solution [29]. The shear impact of the FRP modeled with the help of ANFIS shows better performance when related with seven widely used prediction tips.
Further, the ANFIS model shows effective correlation with the experimental data [30]. The optimal combination obtained from utility function minimization multi criteria optimization approach shows higher material removal rate and lower feed and normal force. Further, the reduction in tool deflection and cutting time, and improvement in surface finish and tool life, were observed [31]. The cryogenic liquid nitrogen was found to be more efficient for specific energy and temperature reduction and improvement in surface quality during machining of Ti-6Al-4V. The Grey-Taguchi hybrid approach has been utilized for obtaining the optimal parametric combination [32].

Materials and Methods
The experimentation was performed on Panther Geared Centre Lathe (Gujurat Lathe MFG.Co.Pvt.Ltd, Shapar, Gujarat, India). Cutting variables which include revolutions per minute (RPM), feed, and depth of cut (a p ) having four levels, were adopted for experimentation. The selection of the input parameters and their levels was done considering the literature survey and trial experiments. The diameter of the work material considered was 150 mm with a length of 1500 mm. The rpm values were selected based on the pilot experiment. More than 20 pilot experiments were conducted to choose the levels of each parameter. Table 1 depicts the parameters with their levels. The Taguchi L16 array design of experimental design was implemented for designing the experiments. The process response parameters, which include MRR, Ra, and Fc, were estimated at distinct parametric settings. The turning operation was performed in EN31 alloy steel using CVD Ti(C,N) + Al 2 O 3 + TiN coated carbide tool insert. Table 2 presents the physical properties of EN31 alloy steel. The MRR was evaluated with an analytical equation-MRR = πdDfN, where d remains the diameter of the work piece, a p is the depth of the cut, f is the feed, and N is the RPM. Ra was measured with a roughness tester, (Surftest SJ-210; Mitutoyo, Japan) the measurement was done at three different points in the finished sample with three repetitions, and the average of the same was utilized for further analysis; and Fc were measured using the dynamometer (model No-621C, N.K. Instrument, Thane, Maharashtra, India, having range-500 kgf). Table 3 provides experimental values. The experiment was repeated several times until the values converged and average values have been considered for the analysis.

Results and Discussion
The results achieved in this research work are presented in this section. The analysis of the MRR is presented in Section 3.1. The RA experimental results are presented in Section 3.2 and the evaluation of the FC tests is presented in Section 3.3. Figure 1 presents the effect of turning parameters on MRR. The figure defines that with the rise in the rotational speed of a workpiece, the MRR increases. This is associated with an increase in RPM that leads to a rise in material removal per unit time. Moreover, with the intensification in RPM, the MRR also increases. With the increment in feed, the MRR increases in the beginning, and then, it decreases. The feed rate increases the amount of material removal per unit time. With rise in the value of feed, the MRR decreases as an increase in feed requires added cutting force, which subsequently reduces the MRR. The plot also depicts that the increment in the depth of cut initially does not have a noteworthy effect on MRR. However, the MRR increases when the depth of cut also increases. Furthermore, this is because the tool plunges into more material, which increases MRR.

Analysis of Material Removal Rate
decreases. The feed rate increases the amount of material removal per unit time. With rise in the value of feed, the MRR decreases as an increase in feed requires added cutting force, which subsequently reduces the MRR. The plot also depicts that the increment in the depth of cut initially does not have a noteworthy effect on MRR. However, the MRR increases when the depth of cut also increases. Furthermore, this is because the tool plunges into more material, which increases MRR.    The ANOVA for the MRR is depicted in Table 4. The table presents the performance criteria with calculated F-values concerning the respective control factors.  The ANOVA for the MRR is depicted in Table 4. The table presents the performance criteria with calculated F-values concerning the respective control factors. It was observed that both rpm and feed have major contributions (69% and 24% respectively) to the MRR. The contribution of error is negligibly small and hence ignored. The other parameters (feed and depth of cut) are less significant. Percentage contributions of the process parameters on MRR as a pie chart are shown in Figure 3. The percentage contributions of the process variables on MRR as a chart are shown in Figure 7.

Analysis of Surface Roughness
The consequence of process variables on surface roughness is represented in Figure 3. It was perceived that with increment in the RPM, the roughness of a machined surface reduced. An increase in cutting speed leads to a decrease in the built-up edges, and thus, the friction force between the cutting tool and workpiece surface to be machined is much less. This is because the resistance offered by the cutting edge to cutting gets reduced with a reduction in contact area between the tool and the work surface [33]. The graph also depicts that the surface roughness increases with increases in feed rate as well. This is because during machining, the cutting edge of the tool is not able to cut uniform material from the upper surface of a rotating job. Besides, there is a reduction in the contact time during machining that leads to the material not being appropriately removed. The depth of cut does not show much significance up to a certain level. However, a greater depth of cut rises the Ra values. A higher value of depth shows a further reduction of material from the work surface, causing better surface finish as removal of material from the work surface takes place slowly. Figure 4 presents the interaction of surface roughness with other process parameters during turning operation. The figure indicates that the process parameters have huge influences on surface roughness values. Table 5 shows the ANOVA results for surface roughness and calculated F-values concerning the respective control factors. It can be observed that only cutting speed (rpm) has a statistically significant contribution (68.8%) to the surface roughness of the work-piece. The percentage contributions of the process parameters on surface roughness as a bar chart is shown in

Analysis of Surface Roughness
The consequence of process variables on surface roughness is represented in Figure 3. It was perceived that with increment in the RPM, the roughness of a machined surface reduced. An increase in cutting speed leads to a decrease in the built-up edges, and thus, the friction force between the cutting tool and workpiece surface to be machined is much less. This is because the resistance offered by the cutting edge to cutting gets reduced with a reduction in contact area between the tool and the work surface [33]. The graph also depicts that the surface roughness increases with increases in feed rate as well. This is because during machining, the cutting edge of the tool is not able to cut uniform material from the upper surface of a rotating job. Besides, there is a reduction in the contact time during machining that leads to the material not being appropriately removed. The depth of cut does not show much significance up to a certain level. However, a greater depth of cut rises the Ra values. A higher value of depth shows a further reduction of material from the work surface, causing better surface finish as removal of material from the work surface takes place slowly. Figure 4 presents the interaction of surface roughness with other process parameters during turning  Table 5 shows the ANOVA results for surface roughness and calculated F-values concerning the respective control factors. It can be observed that only cutting speed (rpm) has a statistically significant contribution (68.8%) to the surface roughness of the work-piece. The percentage contributions of the process parameters on surface roughness as a bar chart is shown in Figure 7.

Analysis of Cutting Force
The significance of turning parameters on cutting force is represented in Figure 5. It is understood from the figure that the rise in cutting speed decreases the cutting force, which could be owing to a temperature increase at the tool-work contact. Additionally, the shear strength of the material at the cutting zone reduced at a higher value of spindle speed. The graph also shows that the increment in feed rate raises the cutting force. This is caused by the increase in chip load per tooth and the increment in feed rate [13]. The increment in the depth of cut for a specific value (0.1 to 0.2 mm) has resulted in a higher value of cutting force, and further, an additional increment in the depth of the cut decreases the force of cutting. Figure 6 depicts the interaction plot for cutting force.

Analysis of Cutting Force
The significance of turning parameters on cutting force is represented in Figure 5. It is understood from the figure that the rise in cutting speed decreases the cutting force, which could be owing to a temperature increase at the tool-work contact. Additionally, the shear strength of the material at the cutting zone reduced at a higher value of spindle speed. The graph also shows that the increment in feed rate raises the cutting force. This is caused by the increase in chip load per tooth and the increment in feed rate [13]. The increment in the depth of cut for a specific value (0.1 to 0.2 mm) has resulted in a higher value of cutting force, and further, an additional increment in the depth of the cut decreases the force of cutting. Figure 6 depicts the interaction plot for cutting force.   The ANOVA test for cutting force is presented in Table 6. Relating the F calculated and F standard, it is confirmed that rotating speed and feed process parameters are significant to the 90% confidence level. However, the depth of cut parameter is substantial at the 75% confidence level. The percentage contributions of the process parameters on c utting force are shown in Figure 7.  The ANOVA test for cutting force is presented in Table 6. Relating the F calculated and F standard, it is confirmed that rotating speed and feed process parameters are significant to the 90% confidence level. However, the depth of cut parameter is substantial at the 75% confidence level. The percentage contributions of the process parameters on c utting force are shown in Figure 7. The ANOVA test for cutting force is presented in Table 6. Relating the F calculated and F standard, it is confirmed that rotating speed and feed process parameters are significant to the 90% confidence Materials 2020, 13, 3137 9 of 15 level. However, the depth of cut parameter is substantial at the 75% confidence level. The percentage contributions of the process parameters on cutting force are shown in Figure 7.

Adaptive Neuro Fuzzy Inference System (ANFIS)-Based Prediction Model
ANFIS is a well-known hybrid model which was discovered by Jang in 1993 [13]. Indeed, it can set up a feasible rule, which coordinates with a robust neural network system to predict the output data, as it inherits those unique and efficient features from fuzzy logic and artificial neural networks. ANFIS is a five-layered system [14], which is based on the Takagi-Sugeno fuzzy interference system [1]. Every layer possesses a different function as mentioned below: Layer I: The most relevant layer of all the layers which converts the input set to fuzzy set takes place. The membership function plays a prime role to undergo a conversion of the input set to a fuzzy set. There are a set of adaptive nodes in a fuzzy layer whose functions are described as: P, Q, and R are input variable nodes. x, y, z, B, C, and D are labels connected with variable input nodes having µ(P), µ(Q) and µ(R) as membership functions. Due to the versatility, Gaussian shape is selected in this case out of all membership function.
Layer II: The output layer is a fixed function of input signal and the node function and is represented as:

Adaptive Neuro Fuzzy Inference System (ANFIS)-Based Prediction Model
ANFIS is a well-known hybrid model which was discovered by Jang in 1993 [13]. Indeed, it can set up a feasible rule, which coordinates with a robust neural network system to predict the output data, as it inherits those unique and efficient features from fuzzy logic and artificial neural networks. ANFIS is a five-layered system [14], which is based on the Takagi-Sugeno fuzzy interference system [1]. Every layer possesses a different function as mentioned below: Layer I: The most relevant layer of all the layers which converts the input set to fuzzy set takes place. The membership function plays a prime role to undergo a conversion of the input set to a fuzzy set. There are a set of adaptive nodes in a fuzzy layer whose functions are described as: P, Q, and R are input variable nodes. x, y, z, B, C, and D are labels connected with variable input nodes having µ(P), µ(Q) and µ(R) as membership functions. Due to the versatility, Gaussian shape is selected in this case out of all membership function.
Layer II: The output layer is a fixed function of input signal and the node function and is represented as: A2,x = Gx = µBx (P). µCy (Q). µDz (R), (for x,y and z = 1,2) (4) A2 represents the output of layer II, and Gx corresponds to the fuzzy strength rule. Layer III: It has a defined function node network like layer II and layer strength III depicted as: A3 represents the output of layer III, and Gl is the normalized strength rule for layer III. Layer IV: The nature of this is layer is inconsistent and varies spontaneously. Due to the nature of the layer node being adaptable, and the strength function is defined as: Here, Glx is the normalized weight of xth rule. fx is a fuzzy rule for x = 1,2. Therefore the fuzzy rule setup is evaluated as: Ux, Vx, and Wx are the defined parameters. Layer V: Total outcome is calculated in this layer. It has a fixed node, and the total output of the total system is evaluated by: The training of input values was done by applying the backpropagation method and was tested based upon the trained values.
The Gaussian membership function was selected and applied due to its versatility and reliability. The root mean square error (RMSE), correlation coefficient (R), and mean absolute percentage error (MAPE) of the predicted values were evaluated by using the following relations [1]: Here E V is the experimental value, P V is the predicted value, M EV is the mean of the experimental value, M PV is the mean of the predicted value, and T no is the number of experiments.
The experimental results and predicted values of MRR, Ra, and Fc are represented in Table 7.
The comparative plot of predicted values based upon the ANFIS model and experimental values is shown in Figures 8-10. It is witnessed from the plot that ANFIS-based predicted data show reliable responses concerning experimental values for all three outputs. Hence, the proposed ANFIS model is accurate for predicting the output values for varying parametric combinations.   Furthermore, Table 8 depicts the values of MAPE, RMSE, and R for all the experimental and predicted values.   Furthermore, Table 8 depicts the values of MAPE, RMSE, and R for all the experimental and predicted values.  Furthermore, Table 8 depicts the values of MAPE, RMSE, and R for all the experimental and predicted values. The parametric analysis was performed considering three process factors, namely, RPM, feed, and depth of cut, and three process performance variables-MRR, Ra, and Fc. It was observed that the cutting parameters influence the process characteristics. The RPM is the most promising factor for all the performance parameters, namely, MRR, Ra, and Fc. The correlation coefficient suggested that the prediction model is reliable and can be utilized for predicting the output responses, such as MRR, Ra, and Fc [1].
The ANFIS-based prediction was undertaken in this research to predict the process response values at different parametric combinations. The research concludes that the proposed ANFIS model used for prediction is accurate at predicting the process responses at varying variable combinations. The developed model is useful for predicting the values of performance criteria at different parametric combinations in a turning operation. Machining of EN31 alloy steel, which is high carbon alloy steel, has been a critical area of study in the of metal cutting domain in the past several years. Moreover, it is related to a wide range of applications, such as round bars, necessarily, the connectors between pipes in oil and gas systems, water supply systems, and plugs to resist pressure in pressure valves, which are made available at lengths of 100 mm to 6000 mm with a diameter range from 5 mm to 500 mm. Despite the wide range of applications, the machining of EN31 alloy steel has always been a challenge due to its physical properties, especially the hardness.

Conclusions
The research presents an experimental investigation and ANFIS-based modelling during the turning operation of EN31 alloy steel. The major findings of the research work are as follows. 1.
The proposed model presents R values of 0.99, 0.98, and 0.964 for MRR, Ra, and Fc, respectively.

2.
The RPM is the major contributing parameter for all performance responses, such as MRR, Ra, and Fc.

3.
The proposed ANFIS model is suitable for predicting the performance responses at all values of parametric combinations.

4.
Machining of EN 31 alloy steel, which is a high carbon alloy steel, has been an essential area of research in the field of metal cutting in the past several years owning to its wide range of applications, such as: round bars essentially as the connectors between pipes in oil and gas systems, water supply systems, and plugs to resist pressure in pressure valves. Moreover, this material is available at lengths from 100 mm to 6000 mm with a diameter range from 5 to 500 mm. Despite the wide range of applications, the machining of EN 31 alloy steel has always been a challenge due to its physical properties, especially the hardness. Therefore, machining parameter optimization becomes a critical area of research and this research has contributed to the body of knowledge in this area. The findings of this research may be useful for the manufacturers and future researchers in this area.