Material removal characteristics of microslot (kerf) geometry

DOI 10.1007/s00170-010-2645-z ORIGINAL ARTICLE Material removal characteristics of microslot (kerf) geometry in μ-wedm on aluminum Kodalagara Puttanarasaiah Somashekhar & Nottath Ramachandran & Jose Mathew Received: 19 January 2009 / Accepted: 29 March 2010 # Springer-Verlag London Limited 2010 Abstract This paper presents the formulation and solution of optimization of various process parameters for the selection of the best control settings on a microwire electrical discharge machining process. A factorial design model is used to predict the measures of performance as a function of various control settings. Analysis of variance is used to indicate the significant factors. Regression models relating the machining performance are established. The performance measures taken for the model are material removal rate (MRR), overcut, and surface roughness. At discharge energy of 2,645 μj, maximum MRR of 0.0428 mm 3 /min and an overcut value of 69 μm are observed. With the value of discharge energy changing from 32 to 4,500 μj, the Ra value of slot surface varied from 1.17 to 4.25 μm. The analysis gave the average erosion efficiency around of 27%, which showed high sensitivity to the selected discharge energy levels. Keywords μ-wedm. Discharge energy. Overcut. Surface roughness. Material removal rate. Erosion efficiency Nomenclature Adj. R 2 Adjusted R-squared correlation coefficient C Capacitance of the capacitor (μf) C p Specific heat (J/kg C) df Degrees of freedom DOE Design of experiment f Sparking frequency (Hz) F F test value h Thickness of the plate (mm) K. P. Somashekhar (*) : N. Ramachandran : J. Mathew Advanced Manufacturing Centre, MED, NITC, Calicut, Kerala 673 601, India e-mail: somashekar.kp@gmail.com l Length of slot (mm) L m Latent heat of melting (μj/kg) L v Latent heat of vaporization (μj/kg) MRR Material removal rate (mm 3 /min) MS Mean squares P Probability value Q s Energy supplied (μj/kg) Q m Energy utilized for melting (μj/kg) Q v Energy utilized for vaporization (μj/kg) Q e Average erosion energy (μj/kg) R 2 R-squared statistic correlation coefficient R a Average surface roughness (μm) SS Sum of squares SR Surface roughness T o Ambient temperature ( C) T m Melting point of material ( C) T b Boiling point of material ( C) t Machining time(min) V s Supplied voltage (V) V b Breakdown voltage (V) V slot Volume of slot (mm 3 ) W Width of slot (mm) x Regression variables Y Predicted response η Average erosion efficiency (%) ρ Density of material (kg/m 3 ) β Regression coefficient 1 Introduction Wire electrical discharge machining (WEDM) is a special form of non-traditional machining in which the electrode is a continuously moving conductive wire. EDM wire cutting

uses a metallic wire electrode to make a programmed contour in a workpiece. μ-wedm is used to produce microslots and microstructured dies and molds in high strength materials. The mechanism of material removal in μ-wedm involves the complex erosion effect from electric sparks generated by a pulsating direct current power supply, the sparks being generated between two closely spaced electrodes under the influence of a dielectric liquid. This is a process, which is nearly force-free. The technology's independence of hardness and strength of the workpiece material ensures the highest accuracy and surface qualities in the manufacturing of complex geometries. As μ-wedm is flexible and easy to automate, it is especially suited in the production of microparts in small and medium batches, but a clear-cut theory has not been established for this complex machining process. The high automation level allows the implementation in unmanned manufacturing systems. μ-wedm is also an alternative for the production of complex shapes and profiles like microgears, turbine blades, and intricate electronic components [1, 2]. The published works [3 8] report on the various aspects of the developments of electric discharge machining research, viz.: improving the performance measures, optimizing the process variables, monitoring and control of sparking process, simplifying the electrode design and manufacture, surface quality improvement of WEDM process, etc. Yuan et al. [9] discussed the development of reliable multiobjective optimization based on Gaussian process regression (GPR) to optimize the high-speed wire cut electric discharge machining process and showed the superiority of GPR models over the regression models in terms of accuracy and feature scaling and probabilistic variation. Kanlayasiri and Booming [10] investigated the effects of machining variables on the surface roughness of WEDM DC 53 die steel. The mathematical model developed used multiple regression method to formulate the pulse on-time and pulse peak current to the surface roughness. Gauri and Chakraborty [11] used the weighted principle method (WPC) to optimize the multiple response of WEDM process and showed that the WPC offers significantly better accuracy than the other approaches. Manna and Bhattacharya [12] optimized the CNC wire cut electrical discharge machining parameters in machining of aluminum-reinforced silicon carbide metal matrix composites using the Taguchi and Gauss elimination dual response approach method. Lin and Lin [13] reported the effect of current polarity, voltage, and spark on-time on EDM process by using Taguchi method. Mahapatra and Patnaik [14] developed a model and applied it for parametric optimization of WEDM machining parameters by Taguchi method. Hao et al. [15] analyzed the vibration-assisted servo scanning 3D μ-edm and showed that effective discharge ratio and machining stability could be improved. Qu et al. [16, 17] developed the cylindrical WEDM process and investigation of surface integrity and mechanical property of EDM surface layer and showed that better material removal rate (MRR) could be achieved in cylindrical WEDM than in 2D wire EDM. Effects of the key parameters, wire feed rate, pulse on-time, and part rotational speed on the surface roughness and roundness were explored. Mathew et al. [18] conducted statistical based experimental investigation to analyze the effect of different process parameters on overcut in the microslots produced using μ-wedm operation with transistor circuit on aluminum and stainless steel and revealed that the dominant process parameters influencing the machining performance were gap voltage, resistance, and pulse on-time. For commercial wire cut EDM, tables provided by the manufacturers provide the database to set up parameters for commonly used work electrode combinations. Since research works even in electric discharge machining at micron-scale is still in infant stage, there are not much standard references available for the selection of process parameters and the levels for optimizing the performance characteristics in μ-wedm. Hence, it is necessary to conduct an extensive experimental investigation to study the effect of different process parameters on the accuracy and surface finish of μ-wed machined components. An attempt is also made to obtain the machining performance with the analysis of variance (ANOVA) approach. The paper highlights the significance of process parameters and different machining conditions on MRR, overcut, and surface roughness of the microslots produced in aluminum using μ-wedm with resistance capacitance (RC) circuit. Mathematical models are developed to correlate the process parameters and performance measures. The study also estimates the erosion efficiency by considering melting and evaporation concepts of material removal characteristics during μ-wedm. These results aid in microslots machining and optimizing machining of complex shapes by integrating the μ-wedm with the CNC system. A schematic representation of the process is as in Fig. 1. 2 Experimentation Microslots of 10 mm length are machined on 3-mm thick aluminum plate by μ-wedm with zinc-coated brass wire of diameter 70 μm. A three-level three full factorial with two replications experimentation is developed using design of experiments. Typical input parameters like gap voltage, capacitance, and feed rate are taken based on experience and evidence from the earlier results by Mathew et al. [19]. The experiments are carried out in a μ-wedm system using the commercial multi-process micromachining center (DT110) by machining slots (as shown in Fig. 1) for the

Fig. 3 Scheme of measurement Fig. 1 Schematic view of μ-wire electrical discharge machining process [36] various combinations of controllable parameters. Experimental findings influencing performances like MRR, overcut, and surface roughness measures, together with a large number of input factors that affect them, are shown in Fig. 2 [20 22]. Investigations on the influence of process parameters on the machining of microslots are also done. The width of machined slot is measured with universal measuring microscope (Zeiss, Germany). Overcut is half of the difference between slot width and wire diameter. The sidewall average surface roughness (Ra) value of the slot (Fig. 3) is also determined using surface roughness tester (Surftest) with 2 μm stylus interfaced with SURFPAK software. Figure 4 shows the schematic diagram of experimental setup adopted for the work with a traveling wire as tool electrode (zinc-coated brass wire of Ф70 μm), flushed type dielectric (EDM-3 synthetic oil) bath between the workpiece and traveling wire electrode. Electrical power and controlling system is with servo-controlled RC circuit to control input process parameters. RC circuit ensures low discharge current with high frequency and thus is suitable in microscale WEDM. 2.1 Experimental design In micromachining using simple-shaped electrodes, there are various parameters which determine the accuracy and surface finish of the final micromachined shape. Different combinations of these process parameters are chosen based on the literature review findings, preliminary two-level investigations, and results of ANOVA of errors for the test of curvature. A series of experiments were conducted with three-level three full factorial design including gap voltage, capacitance, and feed rate, as shown in Table 1. These machining conditions were chosen based on recommended operating conditions of the machine by properly randomizing the experiments. The effects of extraneous factors or compounding variables, which may be present, are averaged out. The complex and stochastic nature of the μ-wedm process makes it difficult to obtain an analytical model based on the physics of the process. In the present study, an Fig. 2 Cause and effect (Ishikawa) diagram Fig. 4 Schematic of μ-wire electrical discharge machining setup

Table 1 Factors and levels for experiment Factor Name Low level ( 1) Middle level (0) High level (+1) A Gap voltage (V) 80 115 150 B Capacitance (μf) 0.01 0.1 0.4 C Feed rate (μm/s) 1 6 10 attempt has been made to model such stochastic process by the traditional multiple regression analysis. Statistical regression analysis is a potential tool for the modeling of such process. It can provide a relationship between the input process parameters and output performance based on some experimental results. Initially, a linear model has been proposed but was later rejected based on the ANOVA test results. Finally, the analysis is done to study the main effects and their interactions to explore the quadratic effects of the influence of parameters on the performances. General higher order polynomial as the quadratic model relating the response to the factors for full factorial design is thus expressed as Y ¼ b 0 þ Xk b i x i þ Xk b ii x 2 i þ X b ij x i x j þ " ð1þ i¼1 i¼1 i<j where Y is the corresponding response; β 0 is the parameter estimated by the average of the responses; β i represents linear effect of the ith factor; β ii represents the quadratic effect of ith factor; β ij represents the cross-product effect or interaction effect between ith and jth factors; X 1, X 2 X k are independent input parameters which influence the response; and ε is a random error which is normally distributed with zero mean according to the observed response. In applying ANOVA techniques, assumptions are checked by the analysis of residuals before interpreting and concluding the results. Interpreting the results from p values of the ANOVA table without a careful check of the assumptions is uncertain and unreliable. From the statistical point of view, it is highly recommended to examine these residuals for normality, independence, and constant variance when using ANOVA [23]. 3 Results and discussions The experimental results of all responses, MRR, overcut, and wall surface roughness are tabulated as in Table 2. 3.1 Effect of process parameters on the material removal rate of the microslot ANOVA for MRR is performed with the assumptions of normality, independence, and constant variance as in Eq. 1. The test methods mentioned earlier were employed again, and none of the assumptions were violated, showing the reliability of ANOVA. Table 3 summarizes the effects of process variables and the interactions for second order quadratic model for MRR. This model was also developed for 95% confidence level. The model F value of 17.05 implied that the model is significant, with a negligible influence of noise. By checking F values and P values, it is seen that the factor C (feed rate) has a significant effect on MRR. The P value of this factor is 99%, which shows its strong influence with a contribution of 44.82%. The value of Prob>F less than 0.0500 indicates that the model terms A, B, C, BC, A 2,B 2, and C 2 are significant. Values greater than 0.1000 indicate the model terms are not significant. The lack of fit F value of 1.84 implies that it is not significant compared to pure error. The Pred R- squared of 0.62 is in reasonable agreement with the Adj R-squared of 0.68. Adeq precision measures the signal to noise ratio. A value greater than 4 is desirable. The ratio of 15.17 indicates an adequate signal. This model can be used to navigate the design space. The developed statistical model for MRR in coded form is MRR ¼ 0:022 þ 1:92 10 3 A þ 1:752 10 3 B þ 7:62 10 3 C 4:159 10 4 BC 4:914 10 3 A 2 þ 5:833 10 3 B 2 4:313 10 3 C 2 Final MRR equation in terms of actual factors is MRR ¼ 0:046 þ 9:848 10 4 Gap voltage 0:070 Capacitance þ 3:438 10 3 Feed rate 4:011 10 6 Gap voltage 2 þ0:153 Capacitance 2 2:129 10 4 Feed rate 2 þ2:922 10 3 Capacitance Feed rate ð2þ ð3þ Figure 5 shows the three-dimensional response surface plot and contour plot for the response MRR in terms of capacitance and feed rate at a gap voltage of 115 V. Contour plot plays a very important role in the study of response surface. By generating contour plot using computer

Table 2 Design matrix of the experiments and measured responses Experiment no. Process parameter Responses Voltage (V) Capacitance (μf) Feed rate (μm/s) MRR (mm 3 /min) Overcut (μm) Wall SR (μm) 1 80.00 0.01 1.00 0.009 21 3.27 2 80.00 0.01 1.00 0.0088 23 3.13 3 115.00 0.01 1.00 0.0155 18 3.22 4 115.00 0.01 1.00 0.0207 16 2.47 5 150.00 0.01 1.00 0.0175 20.78 2.38 6 150.00 0.01 1.00 0.0136 14.5 3.1 7 80.00 0.10 1.00 0.0085 24.5 2.35 8 80.00 0.10 1.00 0.0087 18.01 1.91 9 115.00 0.10 1.00 0.0092 59.2 1.74 10 115.00 0.10 1.00 0.0084 35.5 2.78 11 150.00 0.10 1.00 0.0076 18.01 2.04 12 150.00 0.10 1.00 0.0088 24.5 1.5 13 80.00 0.40 1.00 0.01 27 2.12 14 80.00 0.40 1.00 0.0093 29 3.36 15 115.00 0.40 1.00 0.0129 42.6 2.12 16 115.00 0.40 1.00 0.01 33 1.61 17 150.00 0.40 1.00 0.0162 44.2 2.25 18 150.00 0.40 1.00 0.0166 34.5 2.5 19 80.00 0.01 6.00 0.0165 24 2.04 20 80.00 0.01 6.00 0.0145 16 3.18 21 115.00 0.01 6.00 0.0364 32 1.57 22 115.00 0.01 6.00 0.0288 30.5 3.86 23 150.00 0.01 6.00 0.0224 13 2.71 24 150.00 0.01 6.00 0.0241 11 3.1 25 80.00 0.10 6.00 0.0139 20 1.4 26 80.00 0.10 6.00 0.0168 34 2.6 27 115.00 0.10 6.00 0.0168 47.87 2.22 28 115.00 0.10 6.00 0.0231 29.5 1.2 29 150.00 0.10 6.00 0.0328 16 1.56 30 150.00 0.10 6.00 0.0214 18.5 1.24 31 80.00 0.40 6.00 0.0234 21.5 1.72 32 80.00 0.40 6.00 0.0246 20 1.62 33 115.00 0.40 6.00 0.0335 55 4.25 34 115.00 0.40 6.00 0.0356 56 3.68 35 150.00 0.40 6.00 0.0224 28.5 1.61 36 150.00 0.40 6.00 0.0242 32 1.61 37 80.00 0.01 10.00 0.0139 10.5 1.06 38 80.00 0.01 10.00 0.0149 13 3.59 39 115.00 0.01 10.00 0.0286 18.5 2.41 40 115.00 0.01 10.00 0.0233 16 1.57 41 150.00 0.01 10.00 0.0292 22.5 3.96 42 150.00 0.01 10.00 0.0304 21 3.99 43 80.00 0.10 10.00 0.0145 20.5 1.17 44 80.00 0.10 10.00 0.0326 22 1.95 45 115.00 0.10 10.00 0.019 26.5 1.8 46 115.00 0.10 10.00 0.0342 27.5 2.4 47 150.00 0.10 10.00 0.0153 30 1.46

Table 2 (continued) Experiment no. Process parameter Responses Voltage (V) Capacitance (μf) Feed rate (μm/s) MRR (mm 3 /min) Overcut (μm) Wall SR (μm) 48 150.00 0.10 10.00 0.0173 25.5 1.59 49 80.00 0.40 10.00 0.0305 23 2.42 50 80.00 0.40 10.00 0.0315 62 2.89 51 115.00 0.40 10.00 0.0428 65 2.57 52 115.00 0.40 10.00 0.0307 69 2.38 53 150.00 0.40 10.00 0.0271 57 1.74 54 150.00 0.40 10.00 0.0333 29.5 2.96 software for the response surface analysis, it is easy to characterize the shape of the surface and locate the optimum with reasonable precision. From the examination of the contour plot, it is noticed that MRR increases with increase in feed rate. As feed rate increases, the spark energy is more involved in material erosion, which also increases MRR till it reaches optimum. For further increase of feed rate from optimum, the MRR decreases due to the unflushed debris, which may be causing secondary sparks. Even though literature evidence was insufficient regarding this aspect, the secondary sparks are treated as the reason for MRR change. Further studies in this regard need to be done to substantiate this observation. The justification for the inference is as below. With increase in capacitance, high energy is dissipated, which erodes more work material with stronger sparks. With this material erosion, unexpelled debris trapped in between the machining zone causes unwanted sparks. Thus, a portion of the discharge energy is used to spark with debris. Hence, a lower amount of work material is eroded [24]. The surface plot as in Fig. 5 shows a saddle point, which is the stationary point but not an extremum. An extremum is a maximum or minimum. Functions with many extrema can be very difficult to trace. It was found that the stationary point lies at 0.142, 9.05, which is very close to the stationary point observed by visual examination of the contour plot in Fig. 5b. This stationary point is a saddle point. This is done by recognizing a saddle point by a surface that appears to intersect itself in a curve up, curve down representation, as explained in the Appendix. Figure 6 shows the variation of MRR with different levels of discharge energy at various feed rates. When the Table 3 Analysis of variance for main and interaction effects of parameters on material removal rate Source Sum of squares df Mean square F value Prob>F At 95% confidence level Percentage contribution Model 0.003226 7 0.0005 17.05 <0.0001 Significant A 0.00017 1 0.0002 6.30 0.0157 3.9 B 0.00011 1 0.0001 4.08 0.0492 2.5 C 0.002003 1 0.0020 74.10 <0.0001 44.82 A 2 0.00029 1 0.0003 10.72 0.0020 6.69 B 2 0.000188 1 0.0002 6.94 0.0114 4.3 C 2 0.000217 1 0.0002 8.02 0.0068 5.13 BC 0.000174 1 0.0002 6.43 0.0147 3.98 Residual 0.001243 46 0.0000 27.9 Lack of fit 0.000701 19 0.0000 1.84 0.0720 Not significant Pure error 0.000542 27 0.0000 Cor total 0.004469 53 Response: material removal rate Analysis of variance (ANOVA) for response surface reduced quadratic model ANOVA table (partial sum of squares) Std. Dev., 0.0052; mean, 0.0206; C.V., 25.2572; PRESS, 0.0017; R-squared, 0.72; Adj R-squared, 0.68; Pred R-squared, 0.62; Adeq precision, 15.17

involved in work material removal, which reduces MRR. This behavior is attributed to the high surface roughness. Higher feed rate is not recommended as this leads to frequent traveling wire breaking, causing higher inaccuracies in the microslots produced. The reason for the variation in material removal is the nonuniform discharge energy during machining provided by the RC-type pulse generator in lower energy levels. In the RC pulse generator, the capacitor stores the electrical energy and discharges during the machining. When machining starts with the breaking down of the dielectric, it discharges the stored charge in the capacitor. If the dielectric breakdown occurs before the capacitor is charged fully, it will not discharge the maximum energy. Instead, there will be variations in the discharge energy, leading to variation in MRR [25]. Another reason is that the gap voltage has an intrinsic relationship with the magnitude of the inter-electrode gap, i.e., the distance between the workpiece and electrode during the spark. Thus, at very low voltages, though the energy per pulse is low due to a smaller working gap, there is more possibility for short-circuiting and arcing. Moreover, proper flushing may not be possible if the working gap becomes too small, which in turn decreases the MRR. In addition to discharge energy, the thermal and electrical properties of the workpiece material have significant influence on MRR [26]. A part of the supplied energy is used to form a crater, which determines erosion efficiency. Erosion efficiency is the ratio of actual energy used to erode the material to the supplied energy. The erosion efficiency is also a contributing factor in determining the MRR, which depends on thermal and physical properties of the workpiece material [27]. 3.2 Effect of process parameters on the overcut of the microslot The ANOVA and F ratio test is performed to test the adequacy of the model as well as significance of model coefficients. Table 4 shows the ANOVA results for second Fig. 5 Estimated response surface and contour plot illustrating a surface with saddle point (minimax). a Response surface. b Contour plot capacitance is small, the MRR is higher. If the capacitance is small, the energy input by the discharge (0.5 CV 2 ) and the discharge time are also small. Thus, frequent sparks occur during machining. As a result, total energy entering into the workpiece is more than that during high capacitance. In general, high discharge energy results in high MRR, but the presence of high amount of debris in the machining zone often creates unwanted spark between the electrode and workpiece. Thus, only a portion of energy is Material Removal Rate (mm3/min) 0.0450 0.0400 0.0350 0.0300 0.0250 0.0200 0.0150 0.0100 0.0050 0.0000 At Feed rate 1 µm/s At Feed rate 6 µm/s At Feed rate 10 µm/s 32 66.125 112.5 320 661.25 1125 1280 2645 4500 Discharge energy (µj) Fig. 6 Variation of material removal rate with discharge energy

Table 4 Analysis of variance for main and interaction effects of parameters on overcut Source Sum of squares df Mean square F value Prob>F At 95% confidence level Percentage contribution Model 6,794.34 5 1,358.87 14.22 < 0.0001 Significant A 28.41 1 28.41 0.30 0.5881 0.25 B 3,922.99 1 3,922.99 41.06 < 0.0001 34.47 C 191.10 1 191.10 2.00 0.1638 1.68 A 2 2,005.01 1 2,005.01 20.98 < 0.0001 17.62 BC 608.56 1 608.56 6.37 0.0150 5.35 Residual 4,586.36 48 95.55 40.3 Lack of fit 2,666.07 21 126.96 1.79 0.0781 Not significant Pure error 1,920.29 27 71.12 Cor total 11,380.69 53 Response: overcut Analysis of variance (ANOVA) for response surface reduced quadratic model ANOVA table (partial sum of squares) Std. Dev., 9.77; mean, 29.03; C.V., 33.67; PRESS, 5,981.98; R-squared, 0.60; Adj R-squared, 0.56; Pred R-squared, 0.47; Adeq precision, 13.31 order quadratic model for overcut. The model was developed for 95% confidence level. The model F value of 14.22 implies that the model is significant. There is only a 0.01% chance that the model F value could have occurred due to noise on the model developed. By checking F value and P value, it is seen that factors B, BC, and A 2 are significant model terms. The P value of B and A 2 factors indicates the confidence level is more than 99%, which shows their very strong influence, with 34.47% and 17.62%, respectively. The value of interaction effect BC shows the confidence level is above 95% and thus shows very good influence on overcut with a contribution of 5.35%. The P value of factors A and C has no significant influence on overcut as it provides very high P values. The lack of fit F value of 1.79 implies that lack of fit is not significant compare to pure error. The predicted R 2 of 0.47 is in reasonable agreement with adjusted R 2 of 0.56. Adequate precision greater than 4 is desirable. The ratio 13.31 indicates an adequate measure of signal to noise ratio. This model is used to navigate the design space. Considering the significant terms, the developed statistical quadratic equation for overcut in coded form is Overcut ¼ 39:38 þ 0:89A þ 9:98B þ 2:35C 12:93A 2 þ 4:80BC Final equation for overcut in terms of actual factors is Overcut ¼ 110:280 þ 2:452 Gap voltage þ 21:095 Capacitance 0:599 Feed rate 0:011 Gap voltage 2 þ 5:469 Capacitance Feed rate ð4þ ð5þ Figure 7 shows the three-dimensional response surface and contour plot for the response overcut in terms of process parameters capacitance and feed rate at a gap voltage of 115 V. By examining these figures, it is noted that the process is more sensitive to changes in capacitance and feed rate. Overcut increases with increase in capacitance. With increase in capacitance, large energy is dissipated, which produces stronger sparks resulting in higher material erosion. As feed rate increases, there is less chance of dissipating heat to the surrounding, and hence, more heat is generated at spark gap leading to higher material removal and higher overcut. With further increase in feed rate and capacitance, overcut starts decreasing. At higher levels of capacitance and voltage, spark energy per pulse is greater. This high spark energy produces larger amount of debris. The debris sticks on the workpiece trap and may cause unwanted spark. The unwanted sparks, as explained earlier, result in tool material erosion, which results in less material removal, as the significant amount of spark energy is employed in sparking with debris, leading to less overcut. Figure 8 shows the variation of overcut with discharge energy at various levels of feed rates. Overcut is high at higher levels of capacitance and feed rates. This is due to the effect of high discharge energy (0.5 CV 2 ), where the discharge column will last for a longer time, concentrating on a smaller area. Moreover, the percentages of short circuits are higher during machining of high-aspect ratio microslots as flushing becomes difficult at the deep slots. It is observed that unlike normal WEDM, very high flushing pressure is not favorable during μ-wedm as it results in dimensional inaccuracy due to the deflection of wire electrode.

confidence. The model F value of 4.43 implies the model as significant. Values of Prob>F less than 0.0500 indicate that the model terms are significant. In this case, B 2 is the significant model term with a contribution of 23.6%. Values greater than 0.1000 indicate the model terms are not significant. The lack of fit value of 1.14 implies the lack of fit is not significant relative to the pure error. The high value of lack of fit 37.31% indicates that the model is fit, while the very low value of P, 0.0039, indicates that the model is significant. The specific power transformation was chosen within the confidence level, which was suggested by the design expert's software toolbox using Box Cox plot. In this case, natural log power transformation was suggested. The developed statistical quadratic equation for average surface roughness (Ra) in coded form is (a) Response surface Ln ðsurface roughnessþ ¼ 0:38 0:074B 0:034C þ 0:53B 2 þ 0:065BC ð6þ Final equation for SR in terms of actual factors is Ln ðsurface roughnessþ ¼ 1:168 6:503 Capacitance 0:023 Feed rate þ 13:931 Capacitance 2 þ 0:075 Capacitance Feed rate ð7þ Estimated response surface and contour plot for surface roughness are shown in Fig. 9. It is observed from the figure that the capacitance is strongly influencing the (b) Contour plot Fig. 7 Estimated response surface for overcut. a Response surface. b Contour plot 3.3 Effect of process parameters on the surface roughness of the microslot ANOVA for surface roughness is performed with the assumptions of normality, independence, and constant variance as in Eq. 1. The test methods mentioned earlier were employed again, and none of the assumptions was violated, showing the reliability of ANOVA. Table 5 summarizes the effects of process variables and their interactions in second order quadratic model for surface roughness. This model was also developed for 95% level of Overcut (µm) 80 70 60 50 40 30 20 10 0 At Feed rate 1 µm/s At Feed rate 6 µm/s At Feed rate 10 µm/s 32 66.125 112.5 320 661.25 1125 1280 2645 4500 Discharge energy (µj) Fig. 8 Variation of overcut with discharge energy

Table 5 Analysis of variance for main and interaction effects of parameters on surface roughness Source Sum of squares df Mean square F value Prob>F At 95% confidence level Percentage contribution Model 1.7412 4 0.4353 4.43 0.0039 Significant B 0.1979 1 0.1979 2.02 0.1622 3.0 C 0.0400 1 0.0400 0.41 0.5262 0.6 B 2 1.5481 1 1.5481 15.75 0.0002 23.6 BC 0.1133 1 0.1133 1.16 0.2882 1.7 Residual 4.8162 49 0.0983 71.1 Lack of fit 2.3144 22 0.1052 1.14 0.3731 Not significant Pure error 2.5018 27 0.0927 Cor total 6.5575 53 Response: surface roughness Transform: natural log Analysis of variance (ANOVA) for response surface reduced quadratic model ANOVA table (partial sum of squares) Std. Dev., 0.313512; mean, 0.79488; C.V., 39.44145; PRESS, 5.762144; R-squared, 0.56; Adj R-squared, 0.53; Pred R-squared, 0.52; Adeq precision, 10.07 surface roughness (Ra) with negligible influence of feed rate. With increase in capacitance from 0.01 to 0.2 μf, Ra value decreases, but for further increase of capacitance, Ra values increase. With increase in capacitance, large energy is dissipated, which erodes more material with stronger spark. This strong spark is eroding material with high amount of debris trapped in between the machining zone, causing unwanted spark. Thus, high amount of discharge energy is employed to spark with debris, while work material is effectively removed by a small portion of discharge energy. Thus, lower average surface roughness (Ra) is obtained. As the capacitance increased beyond its optimum value, the large energy is dissipated. The greater discharge energy causes more energy to be conducted into machining zone, which results in the formation of greater melted depth on the workpiece. Hence, greater discharge energy will produce a large crater, causing large surface roughness on the workpiece. Most related researches [28 33] on the performance characteristics of electric discharge machining processes using pulse-generating circuits confirm that the surface roughness depends on the size of spark crater. A large discharging energy will cause violent sparks and will result in a deeper erosion crater on the surface. Accompanying the cooling process after the spilling of debris, residues will remain at the periphery of the crater to form a rough surface. Discharge energy (0.5 CV 2 ) is proportional to voltage in RC circuit. Higher voltages will cause the carbon formation on surfaces to be machined, and it affects surface roughness, which leads to the components' inaccuracies. The improvement is limited because finishing process becomes more difficult due to the occurrence of short circuit attributed to wire deflection and vibration when the energy is gradually lowered. Variation of surface roughness at different levels of discharge energy is depicted in Fig. 10. Surface roughness increases with increase in level of discharge energy. This is attributed to the fact that discharge energy per pulse increases with increase in capacitance value, which produces a deeper crater resulting in higher surface roughness. At high discharge current, the impact of discharge energy on the surface roughness of workpiece becomes greater, and thus, resulting erosion leads to the increase in deterioration of surface roughness. Heating and cooling during machining process result in the formation of thermally affected layer on the surface of the workpiece, thus affecting the surface roughness values. Summary of the significant factors that showed best and worst results in each response achieved from μ- WEDM process is shown in Table 6. These include all the responses investigated. During the μ-wedm process, the discharge energy supplied by the pulse generator is converted into thermal energy in order to raise the temperature of the workpiece materials to remove materials by means of melting and evaporation. With the increase of discharge energy, more materials are removed from the workpiece, thus increasing the MRR. The increasing voltage caused higher energy to discharge, vaporize, and melt the machined area. It will also create a larger impulsive force of discharge, which resulted in higher MRR. MRR increases with increase in feed rate till it reaches optimum value of feed rate. Once the optimum feed rate is reached, MRR starts decreasing; this is because at longer pulse, interval results in smaller gap with constant feed rate. The small gaps lead to concentra-

Surface roughness (µm) 4.5 4 3.5 3 2.5 2 1.5 1 At feed rate 1 µm/s At feed rate 6 µm/s At feed rate 10 µm/s 0.5 0 32 66.125 112.5 320 661.25 1125 1280 2645 4500 Discharge energy(µj) Fig. 10 Effect of discharge energy on surface roughness (a) Surface plot (b) contour plot Fig. 9 Estimated response surface for SR. a Surface plot. b Contour plot tion of spark debris in the spark gap, and thus, increase of short circuits led to decrease in MRR. However, some variations of MRR have been observed with the voltage; though the energy per pulse is low due to a smaller working gap, there is more possibility of occurring short circuits and arcing. Moreover, proper flushing may not be possible if the working gap becomes too small, which in turn decreases the MRR. The overcut affects the ability of a material to achieve good dimensional accuracy and good finishes. The lower and consistent is the size of the overcut, the more predictable will be the resulting dimension. Capacitance was the most significant factor that affects the overcut. Increases in capacitance increase in overcut. This may be due to increase in energy, which formed a bigger gap and directly affected the amount of overcut. It has been proven that capacitance was the most significant factor influencing the surface roughness. At higher capacitance, impact of discharge energy on the surface of workpiece becomes more intense, and resulting erosion led to increase in surface roughness. The test method summarizes the results of fitting regression models relating MRR, overcut, and surface roughness to the predictive machining factors. The adequacy of the proposed model based on R 2 and ANOVA test is observable. Since P value is less than 0.05, there is significant relationship between MRR, overcut, surface roughness, and the predicted variables at 95% confidence level. Precision values in all machining characteristics were all above 4, indicating the adequate model discrimination. Observations on all normal probability plots showed that residual plot variation was along a straight line, implying that the error was normally distributed. The predicted R 2 values are in reasonable agreement with adjusted R 2 for all the performances considered. Hence, this analysis is substantiating the model developed [23]. Table 6 Summary of significant factors in μ-wire electrical discharge machining experiments Response Significant factor Hierarchy factor MRR C A 2 C 2 B 2 BC A B Overcut B A 2 BC A C SR (wall) B 2 A gap voltage, B capacitance, C feed rate

4 Erosion efficiency of μ-wedm 4.1 Assumptions The following assumptions are made to elucidate the efficiency through a simple theoretical modeling using the experimental data: & & & & & & The workpiece material is homogenous and isotropic. The maximum amount of energy stored in the capacitor is completely discharged. The energy is transferred to the workpiece in the form of heat. The duration time of the discharge is equal to the total machining time in RC-pulse time. The enthalpies of phase transition (solid liquid and liquid vapor interface) are negligible. The thermophysical properties of the materials are constant and equal in all three phases from solid to liquid and to vapor, which apply over the whole temperature range. These are Density, ρ=2,700 kg/m 3 Specific heat capacity, C p =900 J/kg C Ambient temperature, T o =20 C Melting temperature, T m =657 C Boiling temperature, T b =2,519 C Latent heat of melting, L m =390 kj/kg Latent heat of vaporization, L v =6,259 kj/kg 4.2 Theoretical modeling For maximum power delivery in the RC circuit, the breakdown voltage at the sparking zone was considered to be equal to 72% of supply voltage. The supply voltage V s and breakdown voltage V b are related as [34] h i V b ¼ V s 1 erc t ð8þ will be equal to the energy lost in the anode and dielectric medium. The determination of the distribution of energy from the supplied energy is very complex, and it depends on the mechanism of material removal. When the energy is supplied, the erosion of material occurs first by melting and then by vaporization because prior to reaching the boiling temperature of the material, the electrode has to reach the melting temperature. The amount of material removed during melting is found using Eq. 12 and the vaporization action to erode the material by Eq. 13. The erosion efficiency of μ-wedm is modeled based on the thermal concepts by considering the melting and vaporization [27, 35]. Average erosion efficiency (η) is h ¼ Q e Q s : Average energy required is Q e ¼ Q m þ Q v 2 The energy required during melting is Q m ¼ V slot: r C p ðt m T o ÞþL m ð10þ ð11þ ð12þ The energy required for vaporization is Q v ¼ V slot: r C p fðt m T o ÞþðT b T m Þgþ L m þ L v ð13þ Machining time per pulse is given by " # 1 t ¼ RC log e 1 V b V s f ¼ 1 t ð14þ ð15þ In an RC circuit, the total discharge energy (Q ds ) gained after charging is delivered until the energy is stored in the capacitor is completely discharged through the gap. Thus, the supplied energy (Q ds ) or the energy delivered to the gap (Q s ) during machining is 1 Q S ¼ ft 2 CV b 2 ð9þ where f is the sparking frequency. Volumetric MRR is calculated as MRR ¼ V slot Machining time where the volume of slot is given by V slot ¼ W :l:h ð16þ ð17þ A part of supplied energy from the spark forms the microslot, which determines the erosion efficiency (η), i.e., the ratio of the actual energy (Q e ) used to erode the material to the supplied energy in the gap (Q s ) as in Eq. 10. The remaining energy supplied between the inter-electrode gap Figure 11 shows the variation of average erosion efficiency with discharge energy. At higher feed rates of 10 μm/s, average erosion efficiency varies between 7% and 46%, and it is 9.57 35% for intermediate feed rate of 6 μm/ s, whereas at lower feed rate of 1 μm/s, it is between 3.26%

Average erosion efficiency (%) 50 45 40 35 30 25 20 15 10 5 0 At feedrate 10 µm/s At feedrate 6 µm/s At feedrate 1 µm/s 32 66.125 112.5 320 661.25 1125 1280 2645 4500 Discharge energy (µj) Fig. 11 Variation of average erosion efficiency with discharge energy and 15.08%. Efficiency at lower energies is found to be more than that at higher energies. This indicates that energy is more efficiently used to form microslot when the applied energy is lower. At lower energy levels, pulse duration will be lesser, and the possibility of the heat transfer to the surrounding medium is reduced. Thus, temperature could raise high at the point of discharge and may result in the vaporization of the material. If feed value is greater, there is possibility for frequent successive sparks without more time lags for wire to move to the required inter-electrode gap. plasma zone, and there is less time for the heat transfer to the surrounding medium. 5.2 Compositions of the machined slot surface The workpieces were also subjected to energy dispersive X- ray (EDX) with SU6600 SEM to investigate how the structure and composition got altered during the μ-wedm processes. Figure 13a and b shows the SEM energy dispersive spectra (EDS) of the affected surfaces, which are obtained by an accelerating voltage of 3 kv. Through EDX analysis, the residuals of Cu, Zn, C, O, etc. could be found. The relative ratio of each composition is shown in Fig. 13. Among the compositions, Cu and Zn are the elements belonging to wire electrode. The carbon content C is from dielectric. Some percentage of O was also found in the EDX analysis of aluminum after machining. In μ-wedm, sometimes, there is oxidation of debris during solidification, hence the oxygen content. For comparison, EDX analyses were also conducted on the workpiece surface before subjecting to μ-wedm, as shown in Fig. 13a. 5 SEM analyses 5.1 Surface topography of the machined slot After μ-wedm operations, the machined slots were examined using scanning electron microscope (SEM), and then the slots were cut into halves so as to expose the section of the surface layer. SU6600, HITACHI, SEM is used for the analysis. Secondary electron imaging was carried out. SEM images of the microslots at high and low discharge energy conditions in aluminum plates are shown in Fig. 12. It has been found that with high discharge energy, overcut of the slot is greater because quantum of electrons released from the negative poles will collide with the neutral particles in the dielectric fluid, resulting in greater ionization effect. The greater the number of electrons and ions colliding with the workpiece, the bigger the microslot expansion [35]. SEM image of the microslot with low discharge energy (Fig. 12b) is more accurate than with high discharge energy, because of the concentration of discharge energy in a small (a) At high discharge energy condition (b) At low discharge energy condition Fig. 12 Scanning electron microscope images of microslots. a At high discharge energy condition. b At low discharge energy condition

(a) EDX spectrum of the polished surface before machining (b) EDX spectrum of the micro slot cut by µ-wedm Fig. 13 Energy dispersive X-ray (EDX) analysis result for the μ- wire electrical discharge machined (WEDM) surface under the conditions V=80 V, C=0.4 μf, and F=10 μm/s. a EDX spectrum of the polished surface before machining. b EDX spectrum of the microslot cut by μ-wedm 6 Conclusions 2. Overcut of the microslots is significantly affected by the discharge condition. 3. During μ-wedm, wire feed rate showed a significant effect on the performance measure of μ-wedm operation. As feed rate increased, there is less heat dissipation to the surrounding, and hence more heat is generated at spark gap leading to higher material removal and higher overcut. 4. The maximum MRR of 0.0428 mm 3 /min occurs at discharge energy of 2,645 μj with higher feed rate and high capacitance value. 5. Maximum overcut value around 69 µm is observed at discharge energy of 2,645 μj. 6. The Ra value of slot surface varied from 1.17 to 4.25 μm, when the value discharge energy changed from 32 to 4,500 μj. 7. Mathematical models developed to predict the various machining characteristics are statistically valid and sound within the range of the factors investigated. 8. The erosion efficiency is a contributing factor in determining the MRR, which depends on the thermal and physical properties of the workpiece material. 9. The maximum erosion efficiency of 46% is at discharge energy of 1,280 μj at a feed rate 10 μm/s when compared to 35% observed at an intermediate feed rate of 6 μm/s and 15% for lower feed rate of 1 μm/s. 10. The average erosion efficiency is around 27%, which shows high sensitivity to the discharge energy and to the level of feed rates. 11. Melting point, boiling point, and heat capacity of materials showed significant effect on the process performance. 12. The study showed that the level of discharge energy played an important role in the performance characteristics of μ-wedm, and lower discharge energies produced more accurate and consistent slots with high efficiency. 13. After microslot machining by μ-wedm, EDS analysis showed the residuals consisting of Cu, Zn, C, O, etc. on the workpiece. Appendix Influences of μ-wedm machining parameters on MRR, overcut, and surface roughness of aluminum are investigated. The following are the inferences: 1. Of all the machining parameters investigated, capacitance was found to be the most significant factor. Higher capacitance produced higher MRR and wall SR. MRR ¼ 0:046 þ 9:848 10 4 A 0:070 B þ 3:438 10 3 C 4:011 10 6 A 2 þ 0:153 B 2 2:13 10 4 C 2 þ 2:922 10 3 BC ð18þ

To get the stationary point, partially differentiate Eq. 18 w.r.t. B and C and equate to zero, @f @B ¼ 0:070 þ 0:306 B þ 2:922 10 3 C @f @C ¼ 3:438 10 3 þ 4:26 10 4 C þ 2:922 10 3 B @f @B ¼ @f @C ¼ 0 0:306 B þ 2:922 10 3 C ¼ 0:070 2:922 10 3 B 4:26 10 4 C ¼ 3:438 10 3 Solving, we get B ¼ 0:142 C ¼ 9:05 This point (0.142, 9.05) is a stationary point. " # @ 2 f @ 2 f @B 2 @B@C 0:306 2:922 10 ¼ 3 2:922 10 3 4:26 10 4 0 @ 2 f @C@B @ 2 f @C 2 Determinant of Hessian matrix, i.e., Δ<0, confirms that the point (0.142, 9.05) is a saddle point. References 1. http://www.mikrotools.com. Accessed 18 May 2007 2. http://www.milcowireedm.com. Accessed 25 July 2008 3. Ho KH, Newman ST (2003) State of the art electrical discharge machining (EDM). Int J Mach Tools Manuf 43:1287 1300 4. Pham DT, Demov SS, Bigot S, Ivanov A, Popov K (2004) Micro- EDM recent developments and research issues. J Mater Process Technol 149:50 57 5. Ho KH, Newman ST, Rahimifard S, Allen RD (2004) State of the art in wire electrical discharge machining (WEDM). Int J Mach Tools Manuf 44:1247 1259 6. Rajurkar KP, Wang WM (1993) Thermal modeling and on-line monitoring of wire EDM. J Mater Process Technol 38(1 2):417 430 7. Sadiq MA, Rahman M, Lim HS (2008) Study of WEDM parameter phenomena for micro fabrication. Int J Manuf Technol Manag 13(2 4):226 240 8. Miller SF, Kao C-C, Shih AJ, Qu J (2005) Investigation of wire electrical discharge machining of thin cross-sections and compliant mechanisms. Int J Mach Tools Manuf 45:1717 1725 9. Jin Y, Wang K, Tao Y, Fang M (2008) Reliable multi-objective optimization of high speed WEDM process based on Gaussian process regression. Int J Mach Tools Manuf 48:47 60 10. Kanlayasiri K, Booming (2007) Effect of machining variables on the surface roughness of wire EDMed DC53 die steel: design of experiments and regression model. J Mater process Technol 192:459 464 11. Gauri SK, Chakraborty S (2009) Optimize the multiple response of WEDM process using weighted principle components. Int J Adv Manuf Technol 40:1102 1110. doi:10.1007/s00170-008- 1429-1 12. Manna A, Bhattacharya B (2006) Taguchi and Gauss elimination method: a dual response approach for parametric optimization of CNC wire cut EDM of PRAISiCMMC. Int J Adv Manuf Technol 28:67 75 13. Lin JL, Lin CL (2005) The use of orthogonal array with gray relational analysis to optimize the electrical discharge machining process with multiple performance characteristics. Int J Mach Tools Manuf 42:237 244 14. Mahapatra SS, Patnaik A (2006) Parametric optimization of wire electric discharge machining (WEDM) process using Taguchi method. J Braz Soc Mech Sci Eng 58(4):423 429 15. Hao T, Wang Y, Li Y (2007) Vibration assisted servo scanning 3D micro EDM. J Micromech Microeng 18:25011 25019 16. Qu AJ Shih, RO S (2002) Development of the cylindrical wire electrical discharge machining process: part І: concept, design and material removal rate. ASME J Manuf Sci Eng 124(3):702 707 17. Qu AJ Shih, RO S (2002) Development of the cylindrical wire electrical discharge machining process: part ІІ: surface integrity and roundness. J Manuf Sci Eng 124(4):708 714 18. Mathew J, Suresh Kumar VB, Somashekhar KP (2007) Investigation into the influence of process parameters on Micro Wire EDM. International Conference on Precision, Meso, Micro & Nano Engineering (Copen 2007), pp 287 294 19. Mathew J, Somashekhar KP, Sooraj VS, Subbarao N, Ramachandran N (2008) Effect of work material and machining conditions on efficiency and accuracy of micro electric discharge drilling. Proceedings of eighth APCMP, China, June 2008, pp 550 558 20. Somashekhar KP, Subbarao N, Mathew J (2008) Effect of discharge conditions on the performance of micro electric discharge machining. Proceedings of AIMTDR, Madras, 2008, pp 615 620 21. Somashekhar KP, Mathew J (2008) Fabrication of microelectrode for micro EDM operation using micro WEDG. Proceedings of AIMTDR, Madras, 2008, pp 639 644 22. Somashekhar KP, Ramachandran N, Mathew J (2009) Modeling and optimization of process parameters in micro wire EDM by genetic algorithm. Adv Mat Res 76 78:566 570 23. Montgomery DC (2001) Design and analysis of experiments, 5th edn. Wiley, New York, pp 427 500 24. Mehfuz R, Ali MY (2009) Investigations of machining parameters for multiple-response optimization of micro electric discharge milling. Int J Adv Manuf Technol 43:264 275 25. Jahan MP, Wong YS, Rahman M (2010) A comparative experimental investigation of deep-hole micro-edm drilling capability for cemented carbide (WC-Co) against austenitic stainless steel (SUS 304). Int J Adv Manuf Technol 46:1145 1160 26. Lim HS, Wang YS, Rahman M, Edwin MK (2003) A study on machining of high aspect ratio micro structures using micro EDM. J Mater Process Technol 140:318 325 27. Wong YS, Rahman M, Lim HS, Han H, Ravi N (2003) Investigation of micro-edm using single RC-pulse discharges. J Mater Process Technol 140:303 307 28. Han F, Wachi S, Kunieda M (2004) Improvement of machining characteristics of micro-edm using transistor type isopulse generator and servo feed control. Precis Eng 28:378 385 29. Amorim FL, Weingaertner WL (2005) The influence of generator actuation mode and process parameters on the performance of finish EDM of tool steel. J Mater Process Technol 166:411 416