0 Introduction

The tool and die, mold, and metalworking industries have seen an enormous change in the last fifty years because of the WEDM process. Because of the many benefits it offers over other non-traditional machining techniques, like the ability of contour cutting, curves, and cavities with good surface polish and dimensional accuracy, this non-traditional machining process has proven to be the most versatile machining process. The WEDM process has been proven to be the best or most distinctive machining method for precisely machining materials with better strength to weight ratios, such as composite materials and Ni-based super alloys^[1]. Many studies have attempted to analyze many machining performances of this technique, such as machining efficiency, surface integrity, and wire breakage phenomena etc. The degree of alloying of three P/M HSS (ASP 23) samples that were treated differently was determined by Huang et al.^[2] using the anodic polarization measuring technique. In the recast layer region, the metallographic data from the Scanning Electron Microscope (SEM) revealed a martensitic transition with alloyed surface containing the steel and electrode material. Additionally, the martensitic layer carried a passive current but not the tempered one^[2]. Using the Taguchi approach, Puri and Bhattacharya^[2] examined the impact of three different WEDM control parameters on cutting rate (V_C), surface roughness (R_a), and geometrical error brought on by wire lag (g). The most important factor that was shown to have a considerable impact on each of the three machining responses was pulse on time. Furthermore, it was discovered that the expected S/N ratio values and the experimental outcomes attained from the ideal parameter setting agreed rather well^[3]. Ribeiro et al.^[4] carried out dry turning operation using coated carbide tool tip on Ti-6Al-4V to find out the optimized cutting conditions in terms of cutting speed, feed and depth of cut. From the experimental results it was found that surface roughness value increased with increase in the cutting length since the work piece material was getting adhered to the tool and due to the improper flushing of chip^[4]. Sarkar et al.^[5] employed the Pareto optimization approach to optimize the performance of γ-TiAl alloy in the WEDM process. For predicting the correlation between various process parameters and the responses, such as V_C, R_a, and dimensional accuracy, additive models were developed. It was discovered that the anticipated values and the experimental findings agreed fairly well. The investigation revealed that pulse off time had no effect on dimensional deviation or surface roughness^[5].

In order to more precisely forecast the performance parameters of the WEDM process while machining Inconel 718, Ramakrishnan and Karunamoorthy^[6] used Artificial Neural Network (ANN) models. The MRSN technique was used to optimize the response. The results of the ANOVA study showed that the factors that had the greatest impact on the process's overall performance were ignition current, delay time, and pulse on time^[6]. Poroś and Zaborski^[7] experimentally investigated the machining efficiency of hard to machine work materials. From the analysis it was observed that materials with high thermal conductivity and heat capacity led to lower machining efficiency. Pulse on time was found to have significant effect towards the machining efficiency of the WEDM process through ANOVA analysis^[7]. Newton et al.^[8] carried out a series of experiments to study the impact of various WEDM cutting parameters on the formation of recast layer during the machining of Inconel 718. The thickness of the recast layer seemed to be influenced by the peak current, current pulse duration, and energy per spark. However, wire diameter and the T_OFF were found to be insignificant parameters^[8]. Alias et al.^[9] investigated the effects of various machining parameters (i.e., kerf width, MRR, surface roughness, and surface topography) with constant current during the machining performance of Ti-6Al-4V. It was discovered that the most important cutting parameter was machine feed rate^[9]. During the cutting of SiC_P/6061 Al MMC, Shandilya et al.^[10] employed Response Surface Methodology (RSM) for optimizing wire feed rate, gap voltage, T_ON, and T_OFF. The results of the ANOVA analysis indicated that the wire feed rate and servo voltage had a major impact on kerf width minimization. Better surface quality was attained at a combination of lower levels of process parameters, according to SEM photographs of the processed surfaces^[10].

The impact of high speed brass wire and Zn-coated brass wire on the WEDM process's performance was compared by Nourbakhsh et al.^[11] Zn-coated brass wire produced an improved surface finish and a faster cutting speed, according to the experimental investigation. Uncoated wire created more melted drips, craters, and fissures on the machined surface, according to SEM analysis^[11]. The development of holes on the surface of monocrystalline silicon machined by WEDM was investigated by Pan et al.^[12] Their investigation revealed that gasification holes began to form on the surface when the plasma channel's power density reached a particular degree. Due to damaging thermal stress, the growth of these tiny holes caused the surface heat source to change to a body heat source, which quickly increased the depth of the crack propagation^[12].

In order to analyze the impact of various cutting parameters, namely pulse off time, peak current, wire tension, and wire feed rate, on the surface roughness, cutting rate, and material removal rate of Ti-6Al-4V work material, Saedon et al.^[13] integrated Taguchi's OA approach with the Grey relational analysis technique. Manjaiah et al.^[14] conducted experiments based on Taguchi's L27 OA design under various input conditions of cutting parameters to optimize MRR and R_a simultaneously while machining TiNi shape memory alloy. The results of the ANOVA study and experimental analysis indicated that T_ON was the most important parameter, with servo voltage and wire feed rate having little impact on overall performance. SEM pictures showed that a white layer formed on the machined surface at greater pulse off times^[14]. To minimize overcut when machining High Strength Low Alloy (HSLA) steel on WEDM, Sharma et al.^[15] created a mathematical model based on the RSM approach and implemented a genetic algorithm over the model. Based on the analysis, the ideal parameter settings were determined to be: SV of 49 V, IP of 180 A, WT of 6 gmf, T_ON of 117 μs, and T_OFF of 50 μs^[15]. The impact of cutting rate (V_C) on the craters created on the surface of 16MnCr5 steel that was machined in WEDM was examined by Mouralova et al.^[16] The process of diffusion between the wire electrode and the work material was investigated using EDX microanalysis. The study's findings showed that while several deep craters emerged on the machined surface at lower values of V_C, no significant craters were generated on the surface at higher values of V_C. According to the EDX microanalysis, 70% of the machined surface had Cu and Zn deposits that had melted from the wire electrode as a result of an intense diffusion process^[16]. In their experimental study, Roy and Mandal^[17] examined the surface integrity of a machined Nitinol-60 SMA specimen as a function of surface crack density, thickness of the recast layer, and average peak to valley heights. RSM was used to develop the relationship between the input parameters and the responses. For numerous runs, Monte-Carlo simulation was used to confirm these models' effectiveness. From the analysis it was observed that, SCD and RLT increased with increase in flow rate due to insufficient flushing. Duty factor followed by SV, and flow rate were found to be significantly affecting the WEDM process performance^[17].

In order to achieve the greatest MRR, surface quality, and dimensional accuracy during WEDM of Al/Gr10Cp MMC, Phate et al.^[18] adopted an ANN-based GRA approach. The findings of the ANOVA showed that WF followed by T_OFF, WT, and T_ON had the greatest influence on all three responses^[18]. When milling EN-31 alloy steel, Padhi and Tripathy^[19] used the Taguchi-GRA technique to enhance V_C and reduce R_a and dimensional deviation (D_D). The ANOVA results showed that, with a 55.45% contribution, T_ON had the most impact on all three performance metrics. Additional noteworthy determining factors were WF(5.85%), SF(10.30%), SV(15.21%), and WT (5.21%)^[19]. Farooq et al.^[20] machined concave and convex profiles on Ti-6Al-4V by using WEDM process to investigate the effect of four process parameters such as SV, WF, T_ON and T_OFF, on the geometrical inaccuracies of the cut profile. Experimental results revealed that the optimized parameter setting resulted in least geometrical deviation with a corner radius of 0.106 mm. Furthermore, the SEM research indicated that both the surface integrity and the profile accuracy were significantly impacted by the erosion phenomenon and discharge energy. The use of wire offset, which ranges from 0.169 to 0.173 mm, further reduced the geometric deviations of the actual machined profiles from the intended geometries in addition to optimum parameters^[20]. Pramanik et al.^[21] studied the recast layer phenomena occurring on the Ti-6Al-4V alloy surface machined through WEDM. It was found that the outermost surface had the maximum cooling rate, which was brought about by the work material's low thermal conductivity and the quenching action caused by the presence of dielectric, was what formed the top flaky layers. The melted material swiftly solidified without forming any grain boundaries at the outer surface, where recast layer was created at a lower cooling rate. Heat treatment occurred during the machining process, causing the heat-affected zone to appear somewhat different in colour^[21].

Doreswamy et al.^[22] experimentally investigated the effect of various input process parameters such as peak current, T_ON, T_OFF on the MRR, R_a, and kerf width of Ti-6Al-4V material machined on WEDM. The examination of grey relationship grades revealed that T_ON was having a major impact on the given process's overall performance^[22]. Paturi et al.^[23] experimentally studied how the ANN approach could be used to model the R_a, V_C, kerf width (K_w), and MRR of titanium alloy while machining using WEDM, based on various process parameters such as voltage, flushing pressure, wire tension, WF, arc-on time, arc-off time, and pulse-on time. A satisfactory fit for the developed model was produced using the suggested ANN technique, with R-value more than 0.99. The ANN model's mean and highest absolute error percentages were found to be 2.32% and 5.05%, respectively^[23]. Through the WEDM of Ti-6Al-4V, Altin Karatas and Biberci^[24] examined the effects of cutting parameters such as voltage, dielectric fluid pressure and wire feed speed, on the kerf width, MRR, R_a, R_q (root mean square deviation), and R_z (maximum height) values of the machined specimen. Voltage was found to be the most significant parameter which affected the overall performance of the process. Further, they used ANN model to develop prediction models for each response characteristic^[24]. Doreswamy et al.^[25] experimentally investigated the effects of current, T_ON and T_OFF on the R_a value of Ti-6Al-4V work material machined through WEDM. The results revealed that increases in current and T_ON value led to increased R_a value. Micro cracks were observed on the recast layer formed on the machined surfaces. Additionally, at the microscopic level, the existence of globules, wavy structures, ridge and crater formations, and voids had also been observed^[25].

It is evident from the previous research publications that conventional machining of Ti-6Al-4V alloy is not cost effective since it has poor machinability. Since titanium alloy is very reactive to chemicals, it gets welded near the cutting tool tip as it is being machined, which causes chipping and early tool breakdown.The tool's life is impacted by an increase in temperature at the tool/work piece interface because of the material's low thermal conductivity. Its low modulus of elasticity and strong strength retained at high temperatures also make it more difficult to machine^[26]. Also, machining complex and intricate shapes with high precision and accuracy is not possible through conventional machining. So, wire electrical discharge machining has been proved to be the best alternative for machining complex shapes in Ti-6Al-4V material. In view of this, various studies have been conducted to investigate the effects of WEDM process parameters on features of surface integrity and the dimensional accuracy phenomenon, individually. To obtain excellent surface integrity and maximum dimensional accuracy, it is also crucial to identify the best parametric combination. Although Ti-6Al-4V has various industrial uses, including medical implants and prostheses, turbine blades, aircraft parts, additive manufacturing, and many more, there have not been many reports on the material's surface integrity and dimensional accuracy during WEDM. Surface integrity refers to the quality and characteristics of the surface of a work piece after it has been machined. It includes factors such as surface roughness, residual stresses, microstructural alterations, recast layer thickness and any defects (such as surface cracks, voids etc.) introduced during the machining process. Study of surface integrity is crucial because it affects the performance, functionality and longevity of the machined component. In light of this, the current work attempts to examine the effects of various WEDM process parameters on several elements of surface integrity, including dimensional deviation (D_D) and surface roughness (R_a), white layer thickness (T_WLT), and surface crack density (D_SCD). In addition, Taguchi-based MRSN technique will be used to obtain the ideal parameter setting for improving dimensional accuracy and surface integrity simultaneously. MRSN technique is one of the most versatile multi-response optimization technique which does not involve any mathematical/computational complexity. Furthermore, non-linear regression models are developed to accurately predict the performance features.

1 Methodology 1.1 Taguchi's Design of Experiment (DOE)

The Taguchi-based design approach arranges the cutting parameters and their levels that have an impact on the machining process using Orthogonal Arrays (OA). Three categories are used to group the response variables^[27]: nominal the best type, larger the better type, and smaller the better type. Signal-to-Noise (S/N) ratio that shows how far the response variables deviate from the intended value, is employed in the Taguchi technique. The following formulas are used to find the S/N ratio value for various response variable categories, i.e.

1) Smaller-the-better,

$ L_{i j}=\frac{1}{n} \sum\limits_{k=1}^n y_{i j k}^2 $

(1)

2) Larger-the-better,

$ L_{i j}=\frac{1}{n} \sum\limits_{k=1}^n \frac{1}{y_{i j k}^2} $

(2)

where n represents the number of repeated trials and y_ijk represents experimental result of the j^th response in i^th experimental run and at k^th replication.

1.2 Multi-Response Signal-to-Noise (MRSN) Ratio Method

For the analysis of multiple performance measures, following steps are followed in the MRSN technique^[28]:

Step 1: Calculation of scaled quality loss(S_ij).

The scaled quality loss for each response in each trial is calculated as follows:

$ S_{i j}=L_{i j} / \bar{L}_j $

(3)

where, S_ij is the scaled quality loss for j^th response in i^th trial, L_ij is the quality loss for j^th response in i^th trial computed using Eq. (1) and Eq. (2) as applicable, and $\bar{L}_j=\frac{1}{m} \sum\limits_{i=1}^m L_{i j}$ is the j^th response's average quality loss.

Step 2: Computation of total loss function(T_Li).

The total loss function for i^th trial is computed as stated below:

$ T_{\mathrm{L} i}=\sum\limits_{i=1}^m w_j \cdot S_{i j} $

(4)

where, w_j is the j^th response's weight decided by the decision maker, and $\sum\limits_{j=1}^p w_j=1$.

Step 3: Multi-response S/N (MRSN) ratio.

The multi-response S/N ratio value for the i^th trial is calculated as follows:

$ R_{\mathrm{MRSN} i}=-10 \log _{10}\left(T_{\mathrm{L} i}\right) $

(5)

2 Experimental Details

As shown in Fig. 1, the Elektra CNC wire electrical discharge machine has been used for conducting the experiments in this research work. The tool electrode is a 0.25 mm diameter Zn-coated brass wire, and the dielectric fluid is deionized water. Ti-6Al-4V slab of 12 mm thick has been utilized as the work piece specimen. Table 1 and Table 2 present this material's elemental composition and thermo-physical properties, respectively.

The servo voltage, wire feed rate, pulse on time, and pulse off time have been chosen as the machining factors after a thorough analysis of previous research works, since these parameters mostly affect the surface integrity and dimensional accuracy of the machined component. Table 3 lists the three levels at which each parameter can be varied. During the experimental trial, all other parameters remained unchanged. To reduce the number of trials conducted and, as a result, their expense, the current research effort has planned experiments in accordance with Taguchi's L₉ OA, which has been created using MINITAB software. As indicated in Table 4, nine set of experimental runs had been conducted using Taguchi's L₉ orthogonal array.

2.1 Measurement of Performance Characteristics

Nine experiments have been carried out using various combinations of the cutting parameters as indicated in Table 4. In each experimental run, a 17.5 mm contour cut is done and the machining profile of the work piece is represented in Fig. 2. Since the application of Ti-6Al-4V material included complex shapes, so in this study, contour cutting has been done rather than straight cutting to analyze the corner profile of the machined component.

2.1.1 Surface roughness

The roughness value of the machined surface has been measured using a 2D portable surface profilometer with 0.8 mm cut off value and 4 mm of sample length. The R_a value has been measured at three different places (R_a1, R_a2, R_a3) on the machined surface and the results are listed in Table 5.

2.1.2 Surface crack density

The top surface morphology of the WEDMed specimen has been examined using Scanning Electron Microscopy (SEM) at a 1.5KX magnification in order to calculate the density of surface cracks. The same software has been used to measure the length of the cracks occurring on the specified region. The following formula is used to get the average crack length for each machined surface.

$ D_{\mathrm{SCD}}=\left(l_1+l_2+l_3+\cdots+l_n\right) / A $

(6)

where l₁, l₂, …, l_n are length of cracks occurring on the machined surface in μm and, A= 45720 μm², represents the area of the region focused under SEM.

Fig. 3 shows SEM micrographs of machined surface of Ti-6Al-4V at different experimental runs.

2.1.3 White layer thickness

The recast layers exhibit voids and micro cracks, and the machined surfaces exhibit ridges, craters, micro globules development, and heat effects. Every machined specimen is scanned under SEM to determine the thickness of the recast layer or white layer formed on the machined surface. The white layer thickness (T_WLT) value for a given specimen is determined by averaging the lengths of the white layer measured at several points within the region as follows:

$ T_{\mathrm{WLT}}=\left(l_1+l_2+l_3+\cdots+l_n\right) / n $

(7)

where, l₁, l₂, …, l_n are length of T_WLT measured at different points on the machined surface in μm. Fig. 4 represents the SEM micrographs showing white layer thickness deposited on the machined surface of Ti-6Al-4V.

2.1.4 Dimensional deviation

Each machined specimen has been focused under SEM and dimension of the cut profile is measured as shown in Fig. 5. Dimensional deviation is the measure of dimensional accuracy calculate in percentage using the following expression:

$ D_{\mathrm{D}}=\left[\begin{array}{lll} \mid & V_{\text {measured value }}-V_{\text {actual value }} \mid / V_{\text {actual value }} \end{array}\right] \times 100 $

(8)

here, V_{measured value} represents the measured value; V_{actual value} is the actual value. The actual value of dimension is the value given during the programming i.e. 17.5 mm. But due to several machining factors/noise factors, there will be slight deviation in the actual cutting length. So, in the present work, dimensional deviation has been included as a performance characteristic since it will greatly affect the working behavior/functionality of the machined component.

3 Results and Discussion

The MRSN approach is then used to assess the experimental data that are provided in Table 5^{[6, 28]} as described in the earlier section. In this research work, all the four responses/performance characteristics i.e. R_a, D_SCD, T_WLT, and D_D are considered as smaller-the-better type responses. First, the loss function for all the responses has been calculated using Eq. (1). Every response in this study is given the same attention and weight, i.e. the weight value for each response is 0.25. Finally, the MRSN value (R_MRSN) for each experimental run is calculated using Eq.(5) and listed in Table 6.

In order to analyze the impact of T_ON, T_OFF, SV, and WF on the overall performance of the process i.e. MRSN ratio, main effects plot is generated (using MINITAB software) as shown in Fig. 6. It is evident from the first graph increasing value of pulse on time resulting in increased MRSN ratio value^[6]. Because at the increased level of pulse on time, equal distribution of discharge energy over the machined surface is observed resulting in better surface finish and dimension. Also, proper flushing helps in lesser amount of molten metal to be solidified, hence decreasing the white layer thickness value. Similarly, at higher value of T_OFF, better performance is observed in terms of surface integrity and dimensional accuracy. This could be due to the fact that number of discharges occurring per spark within a given period of time decreases at higher value of T_OFF, hence reducing the surface roughness value and prevents the creation of a re-solidification layer by allowing longer time for the molten material to disappear from the machined surface^[25]. However, low value of SV indicates better performance. With increase in SV, the inter electrode gap also increases resulting in unstable machining at the gap and wire breakage. This produces rough surface with larger cracks^[21]. From the last graph, it can be seen that variation in WF is not affecting the machining process much. At higher value of WF rate, larger MRSN ratio is observed due to proper flushing of debris and efficient cutting at faster moving wire electrode.

Hence, from this analysis, the optimal combination of cutting parameters is found to be A₃B₃C₁D₃ (which means high level of T_ON, high level of T_OFF, low level of SV, and high level of WF) i.e. T_ON of 115 μs, T_OFF of 30 μs, SV of 25 V, and WF of 8 mm/min.

3.1 Statistical Analysis of Variance (ANOVA)

The significant impact of each machining parameter on the overall process outcome is then determined by considering R_MRSN from Table 6 and by using statistical analysis of variance technique, as indicated in Table 7. An increased F-ratio value signifies that the corresponding cutting parameter is significantly impacting the overall performance of the machining process^[29]. According to the findings, T_ON has the greatest percentage contribution (78.19%) towards all the performance characteristics, making it the most influential parameter^{[14, 22-23]}. Whereas, wire feed rate has negligible impact on the process performance with percentage contribution of 0.67.

3.2 Regression Analysis

Multiple regression analysis was used to create the prediction models for R_a, D_SCD, T_WLT, and D_D to better understand the relationship between the cutting parameters and response characteristics. The following regression equations are obtained from MINITAB software:

$ \begin{aligned} R_{\mathrm{a}}= & 4.46-0.0177 T_{\mathrm{ON}}-0.0043 T_{\mathrm{OFF}}-0.0147 \mathrm{SV}- \\ & 0.0125 \mathrm{WF} \end{aligned} $

(9)

$ \begin{aligned} D_{\mathrm{SCD}}= & 0.00173+0.000015 T_{\mathrm{ON}}-0.000044 T_{\text {OFF }}- \\ & 0.000052 \mathrm{SV}+0.000079 \mathrm{WF} \end{aligned} $

(10)

$ \begin{aligned} T_{\mathrm{WLT}}= & 7.6-0.115 T_{\mathrm{ON}}+0.151 T_{\mathrm{OFF}}+0.719 \mathrm{SV}+ \\ & 0.396 \mathrm{WF} \end{aligned} $

(11)

$ \begin{aligned} D_{\mathrm{D}}= & 16.2-0.118 T_{\text {ON }}-0.068 T_{\text {OFF }}+0.0157 \mathrm{SV}- \\ & 0.052 \mathrm{WF} \end{aligned} $

(12)

Using the above regression equations, corresponding predicted values for R_a, D_SCD, T_WLT, and D_D are calculated and compared with the results of confirmatory experiment done at the optimal parameter combination i.e. A₃B₃C₁D₃. Table 8 represents the result of the confirmatory experiment. It is evident from the results that the developed regression equations can predict the values of the responses with good accuracy.

4 Conclusions

The important findings of the current study lead to the following conclusions:

1) The Taguchi-based MRSN technique is an incredibly powerful and user-friendly optimization tool that helps in identifying the optimum combination of input parameters to get the highest possible overall performance in the specified machining process.

2) Based on ANOVA results, pulse on time is found to have significant effect towards all the performance characteristics with 78.19% of contribution followed by servo voltage and pulse off time, respectively.

3) A good correlation was observed between the developed mathematical models and experimental results obtained at the optimal parameters settings, since the overall deviation is 3.67% which is below 5%. Thus, this work proposes a very useful multi-objective and predictive optimization tool for the WEDM process.

4) The optimum cutting parameters for this WEDM process are suggested to be: T_ON of 115 μs, T_OFF of 30 μs, SV of 25 V, and WF of 8 mm/min.

Analysis of impact of additional cutting factors, such as wire tension, wire diameter, wire material types, on the machining performance can be included in future research. The current study is restricted to the usage of brass wire with a zinc coating alone.

Acknowledgement

The Department of Mechanical Engineering at Veer Surendra Sai University of Technology, Burla, is acknowledged by the authors for their help and cooperation in the successful completion of this research work.