Multi Response Optimization of ECDM Process Parameters for Machining of Microchannel in Silica Glass Using Taguchi–GRA Technique

In this work, machining of microchannel in silica glass was successfully carried out using electro chemical discharge machining (ECDM) process. The experiments were planned according to L27 orthogonal array with applied voltage, stand-off distance (SOD), electrolyte concentration, pulse frequency and pulse-on-time (TON) as control factors. The material removal rate (MRR), overcut (OC) and tool wear rate (TWR) were considered as response characteristics. In this study the effects of control parameters on MRR, OC and TWR have been investigated. The increase in applied voltage, electrolyte concentration and pulse on time lead to the improvement in the output characteristics which is attributed to formation of heavily crowded hydrogen bubbles and further coalescence of hydrogen bubbles promotes the occurrence of sparks which resulted in higher values of MRR, OC and TWR. The multi-objective optimization of ECDM was carried out through grey relational analysis (GRA) method. Optimal combination of process parameters achieved from GRA was 45 V applied voltage, 25 wt.% electrolyte concentration, 1.5 mm SOD, 400 Hz pulse frequency and 45 μs TON. ANOVA for GRG study revealed that the applied voltage (70.33%) was most significant factor affecting output responses followed by electrolyte concentration (11.69%), pulse frequency (4.98%) and SOD (4.13%). Furthermore, the regression equations were formulated for the optimum combination to predict the collaboration and higher-order effects of the control parameters. In addition, confirmation test was conducted for the optimal setting of process parameters and the comparison of experimental results exhibited a good agreement with predicted values. The microstructural observation of machined surface for the optimum combination was carried out.


Introduction
Nowadays, silica glass plays a significant role in potential areas like medical, MEMS, windows, optical lenses, crucibles, metrological instruments etc., attributed to its relatively higher thermal shock resistance and low coefficient of thermal expansion [1]. These applications call for the machining of microchannels which serve for the path for fluidic flow in microreactors, micropumps, bio medical devices and hydrogen fuel cells [2][3][4]. Howbeit, machining of microchannels in glasses through conventional techniques is difficult task owing to properties of glasses like high hardness and brittleness [5] and therefore non-traditional machining methods are called for machining these glasses. Several works in this regard have been reported for machining these glasses through USM, LBM, AJM, EDM, ECM, CHM, ECDM and WJM etc. Presently electro chemical discharge machining (ECDM) process is emerging as a cost effective substitute for ultrasonic machining, laser ablation and wet chemical etching to produce micro-features in non-conductive materials such as pyrex glass, borosilicate glass, silica glass, silicon wafers, quartz and ceramics [6]. ECDM process is a hybrid process which combines the features of ECM and EDM to produce microfeatures like deep holes, narrow channels and small slots in non-conductive materials [2,5]. Chemical and thermal phenomena are the primary mechanisms of this process to erode the material from the work material [7] thereby achieving the advantages of both; electrochemical dissolution as well as electric-discharge erosion from the work material [8].
In the past, several studies have been conducted to investigate the effects of the ECDM process parameters such as pulse frequency, applied voltage, current density, electrolyte concentration, anode to cathode distance, tool feed rate, pulseon-time, duty cycle, stand-off-distance, different electrolytes, various tool materials etc. Dhanvijay et al. [9] performed ECD machining of fiberglass reinforced plastic with an equal proportion of NaOH and KOH mixture considered as electrolyte. Input process parameters such as voltage, duty factor, electrolyte concentration and tool rotation were optimized so as to obtain better MRR. The authors focused on experimentation and GRA technique of ECDM process with and without flow of electrolyte and concluded that with continuous flow of electrolyte, MRR increased marginally. Bellubbi et al. [10] machined microchannel in silica glass using ECDM process and it was noticed that error between predicted and experimental values of MRR, machining depth and overcut were 4.22%, 1.92% and 3.46% respectively. Priyaranjan et al. [6] focussed on machining of deep micro-holes in alumina by ECDM process. Authors reported the effects of voltage and tool feed rate on geometrical characteristics like depth, overcut, side wall profile of micro-hole along with tool wear. Experimental results confirmed that machining of alumina is difficult than the glasses which require higher voltage for machining.
Optimization of control parameters plays a significant role and assists to identify the range of most effective parameter that controls the machining performance in ECDM. Several optimization methods were suggested to establish mathematical formulation between control parameters and response variables in ECDM process to produce micro-channels, micro-holes, texturing and grooves. Taguchi-GRA is one of the attractive optimization techniques employed by many investigators to establish a control over the uncertainty and optimize several responses in a process [20][21][22]. Taguchi-GRA is extensively employed for evaluating the degree of relationship between arrangements by grey relational grade (GRG). Several researchers have recently used GRA technique to optimize the combination of different process parameters on soda lime glass [23], quartz glass [24], fiber glass reinforced [9], borosilicate glass [25,26], CFRP composites [19], silica (quartz) material [24], silicon wafer [27]. Works reported hitherto prove the effectiveness of Taguchi-GRA for optimizing ECDM process parameters; nevertheless, the available literature reveals the paucity of published works reported on micromachining of silica glass and relevant multi response optimization through GRA. In this work, the influence of control factors; applied voltage, pulse frequency, pulse-on-time, stand-off-distance and electrolyte concentration on response variables; MRR, OC and TWR have been investigated. Furthermore, multi response optimization of ECDM of silica glass through Taguchi-GRA has been proposed and the significance of process parameters was determined through ANOVA. Subsequently, the confirmation test was carried out by considering the optimal combination of process parameters.

Experimental Work
Recent developments in ECDM process comprises with micro-machining of ceramics, glass, composites, silicon wafer and quartz with enhanced output responses [13]. The performance of ECDM process in terms of width of cut, surface roughness, machining depth, TWR, MRR, heat affected zone, recast layer is mainly dependent on various factors like SOD, voltage, current density, electrolyte concentration, auxiliary electrode material, tool materials, electrolyte and pulse factors of supply.

Materials and Methods
In this research, silica glass of size 75 × 26 × 1.2 mm 3 was used as the work material for experimentation. Table 1 presents the physical properties of silica glass used in the present work. The main composition of silica glass is SiO 2 that increases the resistance to thermal shocks [28]. The required ECDM facility for the experimentation was developed as shown in Fig. 1a which consists of power regulated supply, tool holder equipped with 3-axis movement, fixture for holding workpiece, electrodes and electrolyte container. The tool head movement was controlled by GRBL software. A Tungsten carbide tool of 250 μm diameter was considered as cathode, graphite plate of 300 × 200 mm was used as auxiliary electrode (anode) and NaOH solution with varying concentrations was employed as electrolyte solution. NaOH electrolyte having most interesting properties compared to other electrolytes like NaCl, KOH, NaNO 3 , NaF, H 2 SO 4 and HCl [5]. Figure 1b shows typical photograph of the silica glass with micro machined groove.
The machining parameters during pilot experiments were selected based on literature review and initial experiments were conducted by one time approach. After performing the pilot experiments, it was noticed that the viable range of applied voltage was 45-55 V, the electrolyte concentration range 10-25 wt.%, the SOD 0.5-1.5 mm, the pulse frequency 200-400 Hz and the pulse on time 45-55 μs. The parameter design in present work is based on three levels for each machining parameter as shown in Table 2. The stand-off-distance was set using slip gauges between the work material and tool tip. Inter-electrode-gap and machining time were kept constant as 40 mm and 20 min respectively for all trials.
The most prominent output responses were identified and selected as MRR, OC and tool wear rate (TWR). MRR and TWR were determined in terms of material eroded per unit time. The samples and tools in each trial were precisely weighed before and after machining using a digital electronic balance (Make: Mettler Toledo, India) with 0.01 mg resolution. The overcut was measured using Leica microscope at various locations throughout the length of microchannel (Fig. 2) and the average values have been considered. Machined surfaces were also subjected to micrographic observations through scanning electron microscope; SEM (Model: JEO JSM-638OLA from JEOL, Japan). Figure 3 showing field emission scanning electron microscopy (FESEM) demonstrates the depth of machined surface of the experiment no. B5, machining depth observed in the range from 140 to 400 μm.

Working Principle of ECDM Process
The ECDM setup comprises of a primary tool electrode as cathode and a secondary metallic sheet as auxiliary electrode (anode) separated by an inter-electrode gap filled with electrolyte solution and regulated DC power supplied between both the electrodes.
The cathode tool is relatively much smaller than the auxiliary electrode [29]. The work sample is placed below the tool and submerged in electrolyte solution as in Fig. 4a. As the voltage is applied across the electrodes, the electrochemical reactions occur in the electrolyte solution when the applied voltage crosses the critical value which is dependent on geometry and electrolyte concentration [5,19]. Continued voltage supply and high current density boosts the electro chemical reactions and results in the formation of hydrogen bubbles in the vicinity of the tool [30] (Fig. 4b). The high density of hydrogen bubbles forms a casing around tool electrode (Fig.  4c). This casing behaves as dielectric between cathode tool and electrolyte solution, as the applied potential crosses the dielectric power of gas film, the spark is persuaded around tool electrode as shown in (Fig. 4d). Therefore, the work material beneath the tool electrode softens due to spark energy generated by these discharges probably contribute to erode the material by chemical etching and thermal erosion. As a result spark discharges promise a sequence of micro erosions in the layer of work piece, so that micro quantities of material takes place from work material continuously [14,15].

Genichi Taguchi Method
The Taguchi OA lessens the fewer number of trials [31] and hence, time and cost essential for the experimentation.
Commonly, based on the process parameters and their levels, more number of trials needs to be conducted as per traditional strategy. The intention behind selection of Taguchi-OA is to afford for the optimum level of each parameter so as to decide the relative contribution of individual parameter using ANOVA [20,32]. The necessary number of experiments is chosen based on the degrees of freedom (DF) of individual factors and the addition DF of their interaction factors. The number of rows of an orthogonal array must be at least equivalent or greater than DF united with the selected factors i.e. the control variables and their interactions. In the present study 23 experiments have to be conducted as per total DF however, the nearest OA of design is L 27 . Hence, L 27 OA is selected for the experimentation. The experimental design in the present work was planned using Minitab-17 software as shown in Table 3.

Grey Relational Analysis (GRA)
The GRA has provided the information that, any experimental analysis provides complete precise information called white; else the analysis contains lack of information named as black.
For instance the analysis is combination of both, then it is called grey and such analysis is termed as grey relational analysis [21,33]. While optimizing through Taguchi's approach, single response function can be optimized, GRA is the most suitable technique to obtain the optimal process parameters on multi response characteristics [34]. So that the obtained result gives better MRR, low OC and low TWR. Instead of having various optimized results, GRA provides single optimum combination of parameters for better MRR, OC and TWR. This technique involves normalization of the obtained results to find grey relational coefficients (GRCs) and then grey relational grades (GRGs) [35]. The combination of parameters with highest GRG is closest to optimum solution and considered as best possible setting of parameters. Furthermore, the ANOVA is executed to predict the optimum GRG. GRA have been employed by various researchers in their related field for multi-objective optimization of process parameters [36]. Figure 5 illustrates the steps involved in GRA.

Results and Discussion
Experimentation carried out to study the influence of control factors on MRR, OC and TWR according to design matrix shown in Table 3. Further, ANOVA was performed to study the significance and individual contribution of each control factors towards MRR, OC and TWR as shown in Table 4.
Consequently, GRA has been performed to achieve the best possible combination of process parameters resulting in optimal setting followed by confirmation test to ensure the same.

Influence of Process Parameters on Response Characteristics
In general, any increment in voltage enhances in magnification of the coalescence of gas bubbles that boosts the intensity of spark over the work material. As a result, higher material removal takes place due to larger thermal energy input. An increase in electrolyte produces the OH ions which escalates its electrical conductivity. Due to the high melting point of the Tungsten Carbide tool, the amount of molten material removed on the tool is very low compared to the workpiece [30]. As SOD increases, the spark generation between tool and workpiece reduces hence every discharge removes very little amount of material from tool electrode and work sample [37]. Influence of process parameters was analyzed based on the results obtained in Table 3. Also ANOVA was performed to find contribution of individual parameters on responses as shown in Table 4. Accordingly in the following sections the effect of process parameters and their individual contributions on MRR, OC and TWR has been elucidated.

Effect of Control Factors on MRR
It was noticed from Table 3 that, the MRR enhanced with increase in applied voltage. This is attributed to the formation of heavily crowded hydrogen bubbles along the edge of cathode tool. The coalescence of hydrogen bubbles promotes the occurrence of sparks which resulted in higher MRR. As the voltage  increases beyond the 55 V, the viscosity of glass decreases progressively with an increase in temperature which leads to remove more material from workpiece [38]. Likewise, the increase in concentration boosted the MRR. The high electrolyte concentration stimulates the kinetics of electrochemical reactions. As a result, the electrochemical reaction accelerates and leads to the increase in coalescence rate of hydrogen bubbles thereby producing more discharge energy in machining area. This high discharge produces melting and vaporizes work material. On the contrary, MRR decreases as SOD increases; this is because as the gap between tool and workpiece increases the kinetics of electrochemical reactions diminishes, resulting in decrease of sparking phenomenon. Further it was noticed that, with increase in pulse frequency, the length of the on-time reduces resulting in lower MRR. Whereas, shorter on-times erodes very little material and generate smaller craters [37]. These observations made are consistent with those investigated by other researchers [13,19]. From ANOVA study, it was noticed that applied voltage (70.13%) is the mostly influencing factor affecting MRR. The SOD (6.19%) and electrolyte concentration (5.90%) marginally influences the MRR as in Table 4.

Effect of Control Factors on Overcut
The overcut increases with rise in voltage supply and electrolyte concentration while, it decreases with increase in SOD and pulse frequency. As the applied voltage increases the thickness of hydrogen gas envelope increases around the cathode due to thicker hydrogen bubbles [30]. Therefore, sparking at the sides of cathode electrode upholds overcut as shown in Fig. 6 which depicts the typical cases observed with smaller and larger overcuts. The increase in electrolyte increases stray sparking over vicinity of tool electrode which leads to overcut. During the pulse interval, the regulated power supply is cut off and the hydrogen bubble formation is ceased thereby momentary extinction of sparks occurs which leads in smaller overcut. By choosing high frequency pulse parameters and lesser duty ratios, the surface roughness and overcut of machined area can be reduced [37]. From Table 4, it was observed from ANOVA that the applied voltage (70.22%) is the most significant factor influencing overcut, followed by the SOD (5.18%) and the electrolyte concentration (4.78%). In similar works, Garg et al. [19] reported that overcut rises with increase in voltage and concentration.

Effect of Control Factors on TWR
It was observed that the influence of control factors on TWR followed the trend which was similar to as that observed in case of MRR and OC. The TWR increased with respect to increase in applied voltage, electrolyte concentration and T ON . However, TWR decreased with higher values of SOD and  [39] reported that TWR reduced with the dropping of the electrolyte concentration. The heat transmitted to the workpiece increases with the electrolyte concentration which (heat transmitted) further is also accountable for the loss of tool material. The decrease in TWR with respect to increase in SOD is due to the fact that each spark occurred between the neighboring point of the cathode tool and work sample and each sparks melts a smaller quantity of material from both [37]. Table 4 shows that factor applied voltage (70.44%) is the most influential factor followed by standoff-distance (5.44%) and electrolyte concentration (5.10%) on overcut based on ANOVA by considering 95% confidence level. Figure 7 shows the SEM (Make: Zeiss) micrograph of ruptured WC tool tip. Since sparking occurs at the tip of tool, it is subjected to relatively higher temperature and thermal cycle compared to the stem portion of tool electrode and hence is more susceptible for failure. It was observed that micro size of tool material worn out due to increase in voltage and concentration. A part of discharge energy is transferred to the cathode surface which fuses and evaporates a small portion of the cathode material. Since, the tool tip is the sharpest edge, it receives large amount of thermal energy and therefore, subjected to higher wear compared to the stem portion of the tool [30].

Multi Response Optimization through Taguchi-GRA
The assessment of optimal process factors for a precise machining practice is an important assignment for researchers [19]. Most of the engineering systems involve multiple response characteristics to be determined however; a single combination of input factors may be optimal for single response but the same combination of input factors may not yield best results for the other responses. Thus, it is required    to obtain an optimal combination of the control factors to yield a system with optimum quality characteristics. In section 3.1, GRA was implemented to resolve the problem of multiple quality characteristics [40] since MRR, OC and TWR were not interconnected mutually and independent with process parameters. Consequently, optimization method was extended for the complete estimation of multiple quality characteristics by producing microchannel in silica glass using ECDM process through GRA. The performance of ECDM process was enhanced by defining the objective as to maximize the MRR and minimize the overcut and TWR of machined section. This method determined that experimental trial with highest GRG is closest to optimum solution. The various steps followed in this method as discussed in following sections.

Normalizing the Data
For normalization of data for the MRR, the larger-the-better performance characteristic was considered and is expressed in Eq. 1. On the other hand for OC and TWR, the smaller-thebetter performance characteristic was considered, and the same has been expressed in Eq. 2 [41].
where X i is the normalized value, Y i is the measured response value (i = 1, 2, 3….. 27 trials), Y min and Y max are minimum and maximum values of respective responses.

Calculation of Deviation Sequence
Deviation sequence is the difference between the reference value and normalized value obtained in the section 3.2.1, obtained value lies between 0 to 1. Deviation sequence was calculated using Eq. 3.
where, Δ i is the deviation sequence, X r is the reference value considered as one [21] and X i is the normalized value obtained.

Computation of Grey Relational Co-Efficient (GRC)
The GRCs of MRR, OC and TWR were computed by using the Eq. 4. ζ is distinguishing coefficient. Lesser the value of ζ, greater is the distinguishing capacity. In this study, value of ζ assumed in the range of 0 < ζ < 1 i.e. ζ = 0.5.
where ξ i is the GRC, Δ i is the deviation sequence of i th trial, Δ min and Δ max are smallest and largest values of Δ i for respective responses.

Computation of Grey Relational Grade (GRG)
GRG was computed by averaging the GRC values related to each performance characteristic. GRG values were computed using Eq. 5 [20].
Where ξ i is the GRC, n is number of response variables i.e. 3 (MRR, OC and TWR). The larger GRG provides better performance characteristics [20,21], which provides optimum MRR, OC and TWR. Further, GRG values are ranked from higher to lower. The results obtained from various steps of Taguchi-GRA are tabulated in Table 5.
From the grey analysis, it has been observed that seventh trial has maximum value of GRG and it is also shown in Fig. 8. Hence the experimental trial 7 was ranked as first. Therefore, in this work, out of 27 experiments trial 7 was considered as the best run. The process parameters of best run were applied voltage 45 V, electrolyte concentration 25 wt.%, stand-off-distance 1.5 mm, pulse frequency 400 Hz and pulse-on-time 45 μs (A 1 -B 3 -C 3 -D 3 -E 1 ). Therefore, the outcome of MRR, OC and TWR were 49.040 mg/h, 108.951 μm and 0.096 mg/h respectively.
However, in order to ensure the result obtained from Table 5, further evaluation of the multi-response characteristics for the optimal combination of the control factors and their levels was determined by averaging the GRGs.  Table 6 presents the optimum process parameters for getting good MRR, OC and TWR, hence the optimal combination from GRG is A 1 -B 3 -C 3 -D 3 -E 1 . Thus, a combination of 45 V applied voltage, 25 wt.% electrolyte concentration, 1.5 mm stand-off-distance, 400 Hz pulse frequency and 45 μs T ON which is same as that observed from Table 5 was achieved.
The graphical representation of the same is depicted through the main effects plot for GRG as shown in Fig. 9. It was noticed that lower applied voltage, higher electrolyte concentration, higher SOD, higher pulse frequency and lower pulse-on-time contributed in increasing the GRG. Generally, the higher the value of GRG, the closer would be ideal value of the product quality [40]. Thus, a higher value of GRG would be desirable for obtaining optimum performance. Therefore, the optimal parametric combination for the maximum MRR and minimum OC and minimum TWR is A 1 -B 3 -C 3 -D 3 -E 1 resulted in higher GRG (Fig. 9). Table 7 represents the ANOVA for MRR, OC and TWR which gives the effective level of selected factors on the desired responses. It was observed that applied voltage is the mostly influencing factor affecting response variables. The electrolyte concentration also marginally influences the MRR, OC and TWR. Furthermore, SOD and pulse frequency have slight effects on the output characteristics. This influence is identified from the percentage contribution calculated at 95% confidence. The interaction effect of all the parameters does not have major effect on the response variables; hence, it is pooled to error.

Confirmation Test
The confirmation test for the achieved optimum combination was conducted once again to confirm the quality features of producing microchannel in silica glass using ECDM process. Table 5 presents the highest GRG highlighting the combination of control factors as A 1 -B 3 -C 3 -D 3 -E 1 . Regression equations were formulated as in Eq. 6-8 using MINITAB software to associate the interaction and higher-order effects of the previously mentioned factors, employing the related experimental data as noticed during the course of micro-machining.   Table 8 represents comparison between experimental results and predicted results for optimum combination (A 1 -B 3 -C 3 -D 3 -E 1 ). Confirmation experimental results designate a good agreement between the experimental results and predicted values.

Microstructure Study of Machined Surface
The microstructure investigation of SEM images was carried out on machined surface for optimum combination (A 1 -B 3 -C 3 -D 3 -E 1 ) setting. In every discharge, removal of material from the work piece takes place due to a portion of thermal energy generated in electrolyte concentration. Figure 10a shows the SEM micrograph of microchannel formed in silica glass with a small amount of overcut. Figure 10b, depicts the magnified view in which sharp edges and deep grooves confirmed the material removal in large chunks due to the combination of the applied voltage, concentration and pulse frequency [36]. The relatively smoother surface indicates an improvement in surface roughness. The observed surface roughness was in the range of 1.60 to 3.10 μm. The reproduced micrographs were initial findings, where in the primary focus of study was to determine material removal rate, tool wear rate and overcut. Deep grooves are obtained at several locations due to interlinking of sparks generated during machining. Furthermore, the presence of deeper grooves on the surface due to increase in applied voltage and effect of pulse parameters increases indentation of sparks on machined surface [27,42]. The presence of fine cracks originating from the shallow grooves confirmed the erosion of material from workpiece. The microstructure examination on machined surface clearly revealed that the dominance of brittle mode of material removal with presence of grain pull-out and fine cracks. Further, the machined surface also appears to be irregular which is attributed to the deposition of re-solidified particles on the previously machined surface which leads to the formation of sharp edges.

Conclusion
In the present work ECDM setup has been developed and machining of micro-channel in silica glass using 250 μm tungsten carbide was successfully carried out followed by the optimization of process parameters using GRA. Based on the experimentation results, following conclusions have been drawn.
(i) MRR, OC and TWR to a great extent depend on voltage and electrolyte concentration due to formation of heavily crowded hydrogen bubbles and further coalescence of hydrogen bubbles promotes the occurrence of sparks around the tool vicinity. (ii) As the stand-off-distance increases MRR, OC and TWR decrease due to the diminished kinetics of electrochemical reactions, resulting in decrease of sparking phenomenon.  (iii) The optimum combination of control factors obtained through GRA was observed to be A 1 -B 3 -C 3 -D 3 -E 1 (45V -25 wt.% -1.5 mm -400 Hz -45 μs). Further, confirmation experiment results and predicted values were found to be in good agreement with an error 3.65%, 4.93% and 5.797% for MRR, OC and TWR respectively. (iv) Based on ANOVA results of GRG it was noticed that, applied voltage has major influence on MRR, OC and TWR followed by electrolyte concentration, pulse frequency and SOD.