Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

www.springerlink.com/content/1738-494x
DOI 10.1007/s12206-014-0637-x

A hybrid Taguchi-artificial neural network approach to predict surface roughness

during electric discharge machining of titanium alloys†
Sanjeev Kumar1, Ajay Batish1,*, Rupinder Singh2 and T. P. Singh3
1
Department of Mechanical Engineering, Thapar University, Patiala-147004, Punjab, India
2
Department of Production Engineering, GNDEC, Ludhiana-141006, Punjab, India
3
Symbiosis Institute of Technology, Pune-411042, Maharashtra, India

(Manuscript Received October 6, 2013; Revised January 16, 2014; Accepted February 27, 2014)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Abstract

In the present study, electric discharge machining process was used for machining of titanium alloys. Eight process parameters were
varied during the process. Experimental results showed that current and pulse-on-time significantly affected the performance characteris-
tics. Artificial neural network coupled with Taguchi approach was applied for optimization and prediction of surface roughness. The
experimental results and the predicted results showed good agreement. SEM was used to investigate the surface integrity. Analysis for
migration of different chemical elements and formation of compounds on the surface was performed using EDS and XRD pattern. The
results showed that high discharge energy caused surface defects such as cracks, craters, thick recast layer, micro pores, pin holes, resid-
ual stresses and debris. Also, migration of chemical elements both from electrode and dielectric media were observed during EDS analy-
sis. Presence of carbon was seen on the machined surface. XRD results showed formation of titanium carbide compound which precipi-
tated on the machined surface.
Keywords: Electric discharge machining (EDM); Titanium; Machining; Taguchi; ANOVA; Surface roughness (SR); Artificial neural network (ANN)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

evaporates from both the workpiece and tool surfaces. The

1. Introduction
metal is removed in the form of small debris and flushed away
In modern manufacturing industries, titanium alloys have by the flowing dielectric fluid from the machining zone and
been successfully used in aerospace industries due to their the remaining liquid material solidifies in the form of layer.
excellent properties such as light weight, high strength to This deposited resolidified layer is called white layer [7].
weight ratio and corrosion resistance. The machining of tita- Conventional EDM process results in tool wear and formation
nium (Ti) and its alloys are difficult and uneconomical with of brittle and cracked white layer on the machined surface.
the conventional machining processes [1-3]. Dornfeld et al. [4] Tool wear can be minimized by selecting proper machining
reported poor machining characteristics on titanium alloys parameters. However, the surface integrity is affected during
using traditional machining. Non-conventional thermo-electric EDM by the formation of a recast layer. Both surface finish
spark erosion machining process commonly known as electric and surface integrity are important during an EDM process.
discharge machining (EDM) has been successfully used to The machining efficiency in EDM is a function of material
machine titanium and its alloys effectively regardless of their removal rate (MRR), tool wear rate (TWR), surface roughness
chemical and mechanical properties. Moreover, in EDM is (SR) and is dependent upon the input process parameters.
there is no direct contact between the workpiece and tool EDM is now widely used for machining of Ti and its alloys
which eliminate the problems encountered in conventional and the machining efficiency can be improved by introducing
machining like; mechanical stresses, tool chatter, vibration etc. cryogenic treatment of the workpiece and/or the tool (elec-
[5, 6]. During EDM, when the spark occurs between the anode trode). Gill and Singh [8] compared the machining effi-
and cathode in the presence of dielectric, the temperature in ciency/accuracy of holes drilled in two different deep cryo-
the small localized area goes up to approximately 8000°C to genically treated and non-treated Ti 6246 titanium alloys in
10000°C. As a result, a small amount of material melts and terms of surface roughness and overcut. The authors con-
cluded that superior finish is obtained in deep cryogenically
*
Corresponding author. Tel.: +91 1752393097, Fax.: +91 1752393005
treated Ti 6246 material as compared to non-treated titanium
E-mail address: abatish@thapar.edu
†
Recommended by Associate Editor Sung Hoon Ahn alloy. Abdulkareem et al. [9] introduced the cooling effect by
© KSME & Springer 2014 injecting the liquid nitrogen to the copper electrode during
2832 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

EDM of Ti-6Al-4V alloy to study the effect of different ma- heating based on laminated plate theory in combination with
chining parameters on tool wear and surface roughness. The Finite Element Method. Sidhu et al. [27] implemented ANN
results show that electrode wear ratio reduced up to 27% and model to predict the residual stresses during EDM of an Alu-
smoother surface finish was obtained. Venugopal et al. [10] minum metal matrix composite. Rahman [28] utilized radial
studied the effect of uncoated carbide inserts under dry, wet basis function neural network to develop ANN model for
and cryogenic cooling environments while turning Ti-6Al-4V MRR, TWR and SR during EDM of TI-6Al-4V and observed
alloy, and observed that cryogenic cooling by jet liquid nitro- a good agreement with the experimental results. Kao et al.
gen showed substantial improvement in tool life. [29] used Taguchi technique with grey relational analysis for
optimization of parameters for multiple performance charac-
teristics in EDM of TI-6Al-4V. A detailed literature review as
2. Literature review
described above show that optimization of process parameters
In the past few decades, various modeling tools correlating using ANN technique can be successfully implemented for
the input variables and surface finish have been developed. predicting machining performance.
Soni and Chakraverti [11] developed the general mathematical There have been several other studies reporting the machin-
model for the selected process variables during EDM of tita- ing efficiency of Ti with EDM. Tan et al. [30] reported an
nium, and observed higher surface roughness when rotating improvement in surface roughness by mixing SiC and Al2O3
electrode was used. Jain and Jain [12] optimized the machin- nano-powders with dielectric during micro-EDM process.
ing parameters for abrasive flow machining using neural net- Yadav and Yadvaa [31] investigated the improvement in sur-
work model. Wang et al. [13] and Bharti et al. [14] developed face roughness of metal matrix composites using diamond
and applied a hybrid ANN with genetic algorism (GA) meth- grinding along with EDM process. Yan et al. [32] reported
odology for modeling and optimization of input parameters that by adding the urea solution in water, surface roughness of
for MRR and SR in EDM. Kao and Tarng [15] applied the titanium deteriorated when peak current increased. Jabbari-
feed-forward neural network in EDM to establish the relation- pour et al. [33] carried out the experiments on Ti-6Al-4V at
ship for inter electrode gap signal and pulse. Fenggou and different values of pulse current and pulse-on-time to explore
Dayong [16] proposed a method to automatically determine their effects on surface topography and recast layer. Hascalik
and optimize the input parameters in EDM with the applica- and Caydas [34] achieved the EDMed damaged free surface
tion of ANN. Cao and Zhang [17] conducted experiments on on Ti-6Al-4V alloy using the abrasive electrochemical grind-
explosive electric discharge grinding and optimized the results ing process with EDM and reported that surface roughness has
using genetic algorithm approach with neural network. Tzeng the tendency to increase with increase in the current density
and Chen [18] used fuzzy logic technique coupled with Ta- and the pulse-duration. The machining performance of tita-
guchi methodology to optimize the process parameters. Assar- nium and its alloys can be improved by cryogenic treatment of
zadeh and Ghoreishi [19] developed a back propagation neural the workpiece and/or electrode. This technique improves the
network (BPNN) model to assess the MRR and surface physical, mechanical, and thermal properties of the materials
roughness during EDM using current, pulse period, and source such as metals, alloys, plastics and composites [35, 8].
voltage. Markopoulos et al. [20] used ANN models for pre- A comprehensive literature review of machining of Ti and
dicting surface roughness in EDM of different grade of steels. its alloys with EDM show no or minimal work has been done
Fonda et al. [21] used response surface methodology to find to examine the effect of shallow and deep cryogenic treatment
the best optimal solution for metal removal rate, electrode on titanium alloys as well as tool-electrodes. The present study
wear ratio, gap size and surface finish. The effect of the has been completed to explore the effect of cryogenic treat-
thermo-electrical properties of workpiece material titanium ment of different grades of titanium alloys and the electrode
alloy Ti-6Al-4V was investigated during EDM. Rao et al. [22] on surface roughness during EDM. Cryogenic treatment re-
developed hybrid model using ANN and GA techniques for quires very slow cooling of materials (@1°C/min) from room
optimizing the SR of different materials Ti6Al4V, HE15, temperature to the temperature of liquid nitrogen (LN2), fol-
15CDV6 and M-250. Rahman et al. [23] developed a mathe- lowed by soaking for a suitable time and heated back to the
matical model to establish the effect of process parameters room temperature. The cooling rate, heating rate and soak
(peak current, pulse on time, pulse off time) using copper- time are important parameters in cryogenic processing which
tungsten electrode during EDM of Ti-6Al-4V alloy. Patowari may alter the properties of the parts. Also, manganese (Mn)
et al. [24] utilized the ANN model for the surface modification and tungsten (W) powder was mixed with the dielectric fluid
using tungsten-copper powder metallurgy sintered electrodes to study their effect on surface roughness and their impact in
in EDM. Tzeng and Chen [25] applied a hybrid method improving the finish or material removal rate has not been
(BPNN, GA, and RSM) to determine optimal input parameter studied in the literature. The effect of input parameters such as
settings for EDM on MRR, relative electrode wear ratio pulse-on-time (Ton), peak current (A), pulse-off-time (Toff),
(REWR), SR and purposed that genetic algorism methodology and cryogenic treatment on surface roughness has been stud-
predict better. Nguyen et al. [26] used feed-forward ANN ied. In order to predict and generalize the results for surface
model to predict the positions and sizes of induction triangle roughness, artificial neural network (ANN) has been used.
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2833

(a) (b)

Fig. 2. Cryogenic treatment temperature curve: (a) SCT; (b) DCT.

(c) (d) Fig. 2(a). Further, in the same pattern the temperature is lower
Fig. 1. Experimental setup for EDM of titanium alloys: (a) EDM ma- down to -184°C with cooling rate 1°C/min for DCT. This
chine; (b) fabricated set up for PMEDM; (c) dialing of electrode; (d) temperature was maintained in the cryo- processor for 24
EDM using side flushing. hours and then, temperature raise up to room temperature
gradually with 1°C/min. Moreover, tempering of materials
was carried out at temperature 150°C with a heating rate of
3. Detailed description of the experimental methodology
1°C/min and soaked at this temperature for two hours; subse-
3.1 Experimental setup quently it is brought back to room temperature for relieve the
developed stresses, as shown in Fig. 2(b). Tempering is re-
The experiments were performed on die- sinking CNC
quired only for titanium and its alloys, not for copper and its
EDM machine (OSCAR MAX S 645) as shown in Fig. 1(a).
alloys.
A separate small metal container of 09 liter capacity was fab-
ricated using 03 mm thick mild steel sheet for mixing of pow- 3.2 Material used in the experiments
der and was attached with a motorized stirrer for proper blend-
ing and to avoid settling of powder particles at the bottom of Three grades of titanium alloy were used as work material
the tank as well as between the workpiece and electrode as in this study: (i) Ti alloy, ASTM grade II, TITAN 15 (99.83%
shown in Fig. 1(b). Manganese and tungsten two powder were Ti, 0.015%C, 0.106%O, 0.0052%N, 0.0005H, 0.04Fe), (ii)
suspended in EDM oil with a concentration of 10g/liter. Ti-5Al-2.5Sn titanium alloy, ASTM grade VI, TITAN 21
Ferrolac 3M EDM oil with a flash point of 100°C, break down (91.8%Ti, 0.013%C, 0.0689%O, 0.0033%N, 0.00145%H,
voltage 70 KV (rms) was used as the dielectric fluid for the 0.025%Fe, 5.32%Al, 2.76%Sn) and (iii) Ti-6Al-4V titanium
experimentation. Fig. 1(c) shows the dialing of electrode and alloy, ASTM grade V, TITAN 31 (89.60% Ti, 0.012%C,
Fig. 1(d) shows EDM in dielectric fluid without powder using 0.01744%O, 0.0032%N, 0.0035%H, 0.07%Fe, 6.19%Al,
side flushing. 4.04%V). Three different types of electrode material with
Cryogenic-treatment was applied on both workpiece and diameter 18 mm and length 40 mm were used namely (i) cop-
tool-electrode at low temperature, to enhance the properties of per (99.5%Cu, 0.175%Zn, 0.0837%Fe); (ii) copper-
materials by reliving the residual stresses, promotes a more chromium (99.2%Cu, 0.605Cr, 0.0294%Fe) and (iii) cop-
uniform micro-structure and precipitates the eta- carbides in per-tungsten (29.27%Cu, 68.1%W, 0.46%Fe).
steels for increased resistant to wear. In present study, two
types of cryogenic- treatment is applied such as; shallow
3.3 Experimental design and data analysis
cryogenic treatment (SCT) and deep cryogenic treatment
(DCT). Based on the available information from the literature and a
The workpiece and electrode materials were cryogenically pilot study, total of eight machining parameters with their
treated at-110°C for SCT and -184°C for DCT by passing levels were identified for the present study. One factor pulse-
liquid nitrogen. Cooling rate and heating rate is the important off-time varied at two levels and the remaining factors (i) peak
issue which directly influenced the properties of cryogenically current, (ii) pulse-on-time, (iii) dielectric fluid, (iv) electrode
treated materials. In SCT material is cooled down from room material, (v) cryo-treatment of electrode material, (vi) work-
temperature with cooling rate 1°C/min to -110°C and hold piece material and (vii) cryo-treatment of workpiece material
(soaked) at this temperature for 6 hours and then heated back were varied at three levels each to understand their impact on
to room temperature slowly at the rate of1°C/min, as shown in the surface roughness. The selected factors, their symbols and
2834 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

Table 1. Selected factors with their levels for machining.

Source of variation Symbol Level-1 Level-2 Level-3

Pulse-off-time(Toff) A A1 = 30 μs A2 = 45 μs ------
Peak current (Amperes) B B1 = 06 B2 = 10 B3 = 14
Pulse-on-time(Ton) C C1 = 90 μs C2 = 120 μs C3 = 150 μs
Dielectric fluid mixed with powder D D1 = No powder D2 = Mn powder D3 = W powder
Electrode material E E1 = Cu E2 = Cu-Cr E3 = Cu-W
Cryogenic of electrode material F F1 = WCT F2 = SCT F3 = DCT
Work piece material G G1 = TITAN 15 G2 = TITAN 21 G3 = TITAN 31
Cryogenic of workpiece material H H1 = WCT H2 = SCT H3 = DCT

their levels are listed in Table 1. Taguchi’s technique was used 3.3.2 Analysis of variance (ANOVA)
to select orthogonal array for assignment of different levels of The analysis of variance was used to breakdown total vari-
factors for experimentation [38]. The degree of freedom of ability of the experimental observed results into various com-
selected factors are 15, hence L18 mixed-level orthogonal ar- ponents of variance, to further assess their significance. The
rays (21×37) was identified as the most appropriate array for total variation observed in the experimental results is the sum
the present study. MINITAB statistical software, version 16 of variance due to the different controlled input parameters
was used for assigning factors to the array. The L18 mixed- and their interactions and variation due to experimental error.
level orthogonal arrays (21×37) contains 18 experimental runs Furthermore, information can be obtained about the controlled
at various combinations of input parameters. The assignment parameters that how much individual parameter is significant
of actual parameter values with trial run conditions is shown which affects the response characteristics. The ANOVA for
in Table 5. For performing the experimentation, positive po- S/N data and raw data was performed to identify the signifi-
larity was applied to the electrode and negative to the work cant parameters by comparing the calculated F-ratio value
material. Throughout the experimentation, side flushing with a with tabulated F-ratio (F0.05f1f2) value at 95% confidence level
pressure of 0.5 kg/cm2 was used. (see Table 3). Those parameters are called significant if the
calculated F-ratio grater than the tabulated F-ratio (F0.05f1f2).
3.3.1 Screening procedure The current was observed to be the most significant parameter
Taguchi’s screening procedure was used to identify the sig- with contribution 54.38% followed by pulse-on-time with
nificant parameters. Different methods are available for meas- 18% affecting surface roughness. Pulse-on-time is the time
uring the surface roughness values of the machined work- which indicates the time period for which energy is supplied
pieces. After, the experimentation, all the machined samples to the machining zone. The pulse-off-time shows their impact
were cleaned with acetone solution. The surface roughness of on surface roughness with percent contribution of 8.74%.
all the samples were measured with MITUTOYO Surface During this period machining cycle is completed. However, if
Roughness Tester (Model: Surftest SJ-400) at a cut of length the off-time is too short, the ejected workpiece material will
of 0.8 mm. All the experiments were repeated three times to not be swept away by the flow of the dielectric fluid and it will
minimize the effect of random error, which occurs due to the affect the surface finish. The types of workpiece material and
effect of noise and uncontrollable factors. During the present electrode material also influence the surface texture. Here,
study, surface roughness values in terms of arithmetic average work material affects the surface roughness with 7.59% and
deviation of the assessed profile (Ra), root mean square devia- electrode material with 4.38%. The other parameters such as
tion of the assessed profile (Rq) and average maximum height powder mixed dielectric fluid, cryogenic treatment of elec-
of the profile (Rz) was measured for machined samples after trode material contributes approximately 3% each. The indi-
each trial. However, the analysis has been done using the vidual influence of all the process parameters and their signifi-
arithmetic average roughness expressed in terms of Ra ex- cance is shown in Table 3 for S/N data and raw data.
pressed in microns. The three values of surface roughness
R1,R2 and R3 for Ra are shown in Table 2. Standard deviation 3.3.3 Assessment of the main effects
of the three roughness readings after each of the 18 experi- The main effect of the parameters on surface roughness was
ments was calculated with the help of MINITAB 16 and is studied by the level average response analysis of raw data and
given in Table 2. The variance and reliability coefficient S/N data. The analysis was carried out by considering the
(Cronbach Alpha = 0.891) was calculated using SPSS. Since, average of SR (raw data) and S/N ratios for each parameter at
surface roughness is a “smaller the better” response, the sig- levels 1, 2 and 3 as shown in Table 4. The obtained values
nal-to-noise ratio (S/N ratio) was also calculated using were plotted in graphical form. The main effects of raw data
MINITAB and the results are given in Table 2. R1, R2, and and the S/N ratio (dB) of surface roughness for each parame-
R3 shows repetition of the experiments. ter are shown in Figs. 3(a)-(h). The average of response levels
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2835

Table 2. Response surface roughness values in Rz, Rq and Ra.

Trial Surface roughness (Rz) Surface roughness (Rq) Surface roughness (Ra) Std. S/N ratios
No. Dev. (dB)
R1 R2 R3 R1 R2 R3 R1 R2 R3 Avg.

1 22.4 23.6 25.5 4.52 4.72 5.66 3.60 3.70 4.53 3.94 0.511 -11.96
2 32.3 29.8 27.1 7.27 7.03 5.64 5.53 5.71 4.46 5.23 0.676 -14.42
3 25.4 27.5 24.8 5.93 6.35 5.85 4.82 5.07 4.79 4.89 0.154 -13.79
4 24.4 26.0 31.7 5.59 4.86 6.35 4.60 3.67 4.86 4.38 0.626 -12.88
5 40.9 40.8 43.4 10.55 9.92 9.73 8.52 7.96 7.44 7.97 0.540 -18.05
6 31.4 28.2 43.1 7.93 6.16 11.46 6.33 4.79 9.42 6.85 2.358 -17.04
7 35.9 38.5 40.7 8.72 9.83 9.13 7.02 8.19 7.28 7.49 0.614 -17.52
8 29.9 38.3 30.4 6.58 8.40 7.08 5.19 6.62 5.66 5.82 0.729 -15.35
9 46.27 43.0 44.0 11.52 10.7 11.07 9.52 8.84 9.12 9.16 0.342 -19.24
10 26.6 27.3 19.8 4.95 5.66 3.80 3.66 4.37 3.01 3.68 0.680 -11.41
11 22.0 23.8 24.4 4.90 4.78 5.66 3.97 3.76 4.61 4.11 0.443 -12.32
12 27.2 26.9 28.7 5.21 5.69 5.96 4.13 4.36 4.43 4.31 0.157 -12.69
13 26.0 24.1 30.9 6.78 4.37 7.08 4.90 3.30 5.60 4.60 1.179 -13.44
14 37.4 31.3 34.4 7.93 7.31 7.27 6.14 5.93 5.54 5.87 0.304 -15.38
15 40.4 27.7 26.5 9.17 6.61 6.66 7.21 5.36 5.50 6.02 1.031 -15.68
16 31.2 28.1 30.7 6.91 6.62 6.94 5.42 5.48 5.54 5.48 0.060 -14.77
17 34.8 40.6 31.3 8.17 9.18 6.99 6.38 7.20 5.68 6.42 0.761 -16.19
18 37.0 41.0 33.1 9.13 9.20 7.27 7.59 7.18 5.66 6.81 1.017 -16.73

Table 3. Analysis of variance (ANOVA) of control parameters for surface roughness.

ANOVA for S/N data ANOVA for raw data

Factors dof
Seq. SS Adj. SS F %PC Remarks Seq. SS Adj. SS F %PC Remarks
A 1 7.532 7.532 193.13 8.74 √√ 3.961 3.961 29.01 10.14 √√
B 2 46.872 23.436 600.92 54.38 √√√ 19.249 9.624 70.49 49.27 √√√
C 2 15.549 7.775 199.36 18.04 √√ 6.262 3.131 22.93 16.03 √√
D 2 2.609 1.305 33.46 3.03 √ 1.245 0.622 4.56 3.19 X
E 2 3.779 1.889 48.44 4.38 √ 2.639 1.319 9.66 6.75 X
F 2 2.532 1.266 32.46 2.94 √ 1.512 0.756 5.54 3.87 X
G 2 6.541 3.270 83.84 7.59 √ 3.573 1.787 13.09 9.15 X
H 2 0.697 0.348 8.92 0.81 x 0.359 0.179 1.32 0.92 X
Error 2 0.078 0.039 1.00 0.09 0.273 0.136 0.70
Total 17 86.191 100.00 39.07 100.00
dof- degree of freedom, Seq. SS- sequential sum of squares, Adj. MS- adjusted mean square(Variance),
F- variation ratio % PC- percent contribution
x- not significant, √ - less significant, √√- significant, √√√-most significant

help in studying the trend of the performance characteristic craters on the surface, which deteriorates the surface finish.
with respect to the variation of the selected factors. Similarly, pulse-on-time affects the SR. Longer on time re-
Fig. 3(a) shows that surface roughness improves as pulse- duces the surface finish because energy is applied for longer
off-time increased from 30 μs to 45 μs because debris particles time. At short pulse-on-time better surface finish is produced
flushed away from the gap with pressurized dielectric. Fig. as shown in Fig. 3(c). Fig. 3(d) explores the effect of powders
3(b) shows the trend of peak current. The curve shows that as on the surface finish. Tungsten powder mixed dielectric im-
current increases, surface roughness also increases. The mini- proves the surface finish as compared to manganese powder
mum surface roughness is observed at 6A current and maxi- mixed dielectric. The electrode material copper-tungsten pro-
mum at 14A current. At higher current 14A more energy is duced the poorer surface finish than copper, where copper-
applied to the machining zone resulted in larger and deeper chromium improves the surface finish as shown in Fig. 3(e).
2836 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 3. Effect of various input parameters on SR and its S/N ratios.

The alloy Ti-6Al-4V gives the better surface finish than other condition of input process parameters is (A2B1C1D3E2F1G3H3).
two grades of titanium alloy as shown in Fig. 3(g). The effect
of cryogenic treatment of work material and electrode material
4. Artificial neural network (ANN)
is less on the surface finish as presented in Figs. 3(f) and (h).
From Figs. 3(a)-(h) and Table 4 it was observed that the ANN is a flexible modeling tool which has the capabilities
second level of pulse-off-time (45 μs), first level of current of learning the mapping between the input parameters and the
(6A), first level of pulse-on-time (90 μs), third level of dielec- output characteristics to solve non- linear problems using a
tric fluid (mixed with tungsten powder), second level of elec- software computing technique. An artificial neural network
trode material (Cu-Cr), first level of cryogenic treatment of based multi layered algorithm is used for the validation of
electrode material i.e. without cryo-treatment (WCT), third surface roughness. The ANN validation has been developed
level of work material (Ti-6Al-4V) and third level of work using the neural network tool box of Matlab 7.12.0 software.
material’s cryogenic treatment i.e. deep cryo-treatment (DCT) A multilayer feed-forward network was used, where as back
gives the best value of surface roughness. Hence, the optimum propagation algorithm was used to train the network [22, 24,
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2837

Table 4. Response table for mean and S/N values of parameters for SR.

Average value
Process parameters Levels S/N Ratio (dB)
(Ra)
1 6.19 -15.58
Pulse-off-time (A)
2 5.25 -14.29
Fig. 4. A multi-layer tansig-purelin network with eight input neurons,
1 4.36 -12.77
one output neuron, and one hidden layer of twenty neurons.
Peak current (B) 2 5.95 -15.41
3 6.86 -16.63
1 4.93 -13.67
Pulse-on-time (C) 2 5.91 -15.28
3 6.34 -15.86
1 5.67 -14.71
Dielectric fluid (D) 2 6.07 -15.47
3 5.43 -14.63
1 5.85 -15.16
Electrode materials
2 5.21 -14.30
(E)
3 6.12 -15.36
1 5.32 -14.45
Cryogenics of elec-
trode materials (F) 2 5.85 -15.01
3 5.99 -15.36 Fig. 5. Configuration of the FFBP neural network model for the EDM
process on response SR.
1 6.12 -15.35
Workpiece materials
2 5.95 -15.38
(G)
3 5.10 -14.09
during training such that the difference between the sampled
1 5.71 -14.86
Cryogenics of work outputs and target is kept low. The adjustments are computed
materials (H) 2 5.91 -15.21 by the propagation algorithm. The network has been trained
3 5.56 -14.74 using Lavenberg-Marquardt algorithm, i.e. trainlm algorithm
of neural network tool box of Matlab 7.12.0 software. The
error goal and learning rate was set to be 0.001 and 0.1, re-
28]. A total of 18 experiments with two repetitions were per- spectively. The transfer functions which were preferred are
formed to measure change in surface roughness due to vari- tansig and purelin in hidden and output layer respectively.
able level setting of the eight input process parameters. Thus, The transfer function tansig is mathematically equivalent to
54 (18x3) experiments were performed for this study. Table 5 hyperbolic tangent (tanh) sigmoid transfer function, where as
shows the experimental conditions, actual experimental result, purelin is a linear transfer function. Tansig runs faster than the
ANN predicted results and % relative error between experi- MATLAB® implementation of tanh, but the results can have
mental and ANN predicted results for all the 54 observations. very small numerical differences. This function is a good
Out of these 54 observations, 38 (which are roughly 70% of tradeoff for neural networks, where speed is important and the
the total number of data sets) were taken for training purpose, exact shape of the transfer function is not. Trainlm is a net-
another 08 (about 15%) were taken for validation and the re- work training function that updates weight and bias values
maining 08 (15%) were taken for testing. The generated data according to Levenberg-Marquardt optimization. Trainlm is
was preferred for testing as well training of neural network for often the fastest back propagation algorithm in the toolbox,
feed-forward back propagation (FFBP) algorithm. The feed- and is highly recommended as a first-choice supervised algo-
forward back propagation algorithm consists of eight input rithm, although it does require more memory than other algo-
layer, twenty hidden layer and one output layer (SR). A two rithms. The tansig and purelin transfer functions shown by
layer transig - purelin neural network with eight input neurons, Eqs. (1) and (2) calculate their output as follows [24,
one output neuron, and a single hidden layer of twenty neu- MATLAB, 7.12.0 software]:
rons was used as shown in Fig. 5 for prediction of surface
roughness. 2
The neural network architecture used in this study is a feed- transig (n) = -1 (1)
1 + exp-2n
forward back propagation network is shown in Fig. 4. The
FFBP propagation algorithm considers gradient decent purelin (n) = n (2)
method for learning and to train the ANN process. The ANN
method automatically adjusts its threshold and weights values where n is input to the function.
2838 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

Table 5. Experimental conditions and results of experimental and ANN predicted for surface roughness.

ANN
Cryogenic Work Cryogenic Exp.
Exp. Pulse-off- Peak Pulse-on- Dielectric Electrode Predicted % Relative
treatment Piece treatment Value
N0. time current time fluid material Value Error
of electrode material of workpiece (Ra)
(Ra)
1 30 6 90 oil Cu WCT Ti15 WCT 3.60 3.6028 -0.825
2 30 6 120 oil+Mn CuCr SCT Ti 21 SCT 5.53 5.5618 -0.116
3 30 6 150 oil+W CuW DCT Ti 31 DCT 4.82 4.8092 -0.753
4 30 10 90 oil CuCr SCT Ti 31 DCT 4.60 4.6020 -0.182
5 30 10 120 oil+Mn CuW DCT Ti15 WCT 8.52 8.5383 0.328
6 30 10 150 oil+W Cu WCT Ti 21 SCT 6.33 6.2950 0.202
7 30 14 90 oil Cu DCT Ti 21 DCT 7.02 6.9695 2.217
8 30 14 120 oil+Mn CuCr WCT Ti 31 WCT 5.19 5.1623 -0.401
9 30 14 150 oil CuW SCT Ti15 SCT 9.52 9.5327 -0.552
10 45 6 90 oil+W CuW SCT Ti 21 WCT 3.66 3.6734 0.615
11 45 6 120 oil Cu DCT Ti 31 SCT 3.97 3.9712 -0.591
12 45 6 150 oil+Mn CuCr WCT Ti15 DCT 4.13 4.1250 0.033
13 45 10 90 oil+Mn CuW WCT Ti 31 SCT 4.90 4.9093 0.041
14 45 10 120 oil+W Cu SCT Ti15 DCT 6.14 6.1619 0.726
15 45 10 150 oil CuCr DCT Ti 21 WCT 7.21 7.2917 -0.259
16 45 14 90 oil+W CuCr DCT Ti15 SCT 5.42 5.4099 0.447
17 45 14 120 oil CuW WCT Ti 21 DCT 6.38 6.3729 -0.267
18 45 14 150 oil+Mn Cu SCT Ti 31 WCT 7.59 7.6526 -0.716
19 30 6 90 oil Cu WCT Ti15 WCT 3.70 3.7043 -0.286
20 30 6 120 oil+Mn CuCr SCT Ti 21 SCT 5.71 5.7530 -0.421
21 30 6 150 oil+W CuW DCT Ti 31 DCT 5.07 5.0792 0.915
22 30 10 90 oil CuCr SCT Ti 31 DCT 3.67 3.6579 0.264
23 30 10 120 oil+Mn CuW DCT Ti15 WCT 7.96 7.9439 -0.395
24 30 10 150 oil+W Cu WCT Ti 21 SCT 4.79 4.6838 0.456
25 30 14 90 oil Cu DCT Ti 21 DCT 8.19 8.2228 -2.346
26 30 14 120 oil+Mn CuCr WCT Ti 31 WCT 6.62 6.6565 0.208
27 30 14 150 oil CuW SCT Ti15 SCT 8.84 8.7856 -0.120
28 45 6 90 oil+W CuW SCT Ti 21 WCT 4.37 4.3958 0.292
29 45 6 120 oil Cu DCT Ti 31 SCT 3.76 3.7588 -0.018
30 45 6 150 oil+Mn CuCr WCT Ti15 DCT 4.36 4.3582 -0.316
31 45 10 90 oil+Mn CuW WCT Ti 31 SCT 3.30 3.2760 0.006
32 45 10 120 oil+W Cu SCT Ti15 DCT 5.93 5.9454 -1.286
33 45 10 150 oil CuCr DCT Ti 21 WCT 5.36 5.3360 0.049
34 45 14 90 oil+W CuCr DCT Ti15 SCT 5.48 5.4946 0.251
35 45 14 120 oil CuW WCT Ti 21 DCT 7.20 7.2516 -0.340
36 45 14 150 oil+Mn Cu SCT Ti 31 WCT 7.18 7.2005 0.458
37 30 6 90 oil Cu WCT Ti15 WCT 4.53 4.5491 0.866
38 30 6 120 oil+Mn CuCr SCT Ti 21 SCT 4.46 4.4192 -0.825
39 30 6 150 oil+W CuW DCT Ti 31 DCT 4.79 4.7773 -0.116
40 30 10 90 oil CuCr SCT Ti 31 DCT 4.86 4.8792 -0.753
41 30 10 120 oil+Mn CuW DCT Ti15 WCT 7.44 7.4061 -0.182
42 30 10 150 oil+W Cu WCT Ti 21 SCT 9.42 9.6409 0.328
43 30 14 90 oil Cu DCT Ti 21 DCT 7.28 7.2649 0.202
44 30 14 120 oil+Mn CuCr WCT Ti 31 WCT 5.66 5.6668 2.217
45 30 14 150 oil CuW SCT Ti15 SCT 9.12 9.1196 -0.401
46 45 6 90 oil+W CuW SCT Ti 21 WCT 3.01 3.0105 -0.552
47 45 6 120 oil Cu DCT Ti 31 SCT 4.61 4.6246 0.615
48 45 6 150 oil+Mn CuCr WCT Ti15 DCT 4.43 4.4297 -0.591
49 45 10 90 oil+Mn CuW WCT Ti 31 SCT 5.60 5.6720 0.033
50 45 10 120 oil+W Cu SCT Ti15 DCT 5.54 5.5373 0.041
51 45 10 150 oil CuCr DCT Ti 21 WCT 5.50 5.4862 0.726
52 45 14 90 oil+W CuCr DCT Ti15 SCT 5.54 5.5588 -0.259
53 45 14 120 oil CuW WCT Ti 21 DCT 5.68 5.6540 0.447
54 45 14 150 oil+Mn Cu SCT Ti 31 WCT 5.66 5.6109 -0.267
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2839

Table 6. Mean square error and regression error for results of training, validation and testing.

Results Samples Mean square error (MSE) Regression error

Training 38 1.41729e-1 9.68488e-1
Validation 8 5.50307e-1 9.16105e-1
Testing 8 1.55300e-1 6.22211e-1

Table 7. Correlation between the experimental value and ANN predicted value.

Variables No. of exp. (N) Mean Standard deviation Standard error mean Correlation Significance
Experimental value 54 5.7250 1.6422 0.2235 1.00 0.000
ANN predicted value 54 5.7306 1.6557 0.2253

Table 8. Paired sample t- test for experimental value and ANN model value.

Paired differences
Pair
Standard error 95% confidence interval of the difference t-test Significance
(N = 54) Mean Standard deviation
mean Lower Upper
Experimental and ANN
-0.0056 0.0439 0.00598 -0.01759 0.00638 -0.938 0.353
predicted value

The performance graph of the result after training showed

the best results. Best validation performance of the network
was 0.1237 at epoch 06 and model is completed in 12 epochs
as shown in Fig. 6. Table 8 shows the percentage relative er-
rors of experimental and ANN predicted results for surface
roughness. The ANN model shows a good agreement between
the actual experimental results and the predicted results by the
FFBP neural network. Percentage relative error percentage
was calculated by using following Eq. (3):

æ Experimental Results -ANN Predicted Results ö

% Relative Error = çç ÷÷ ´ 100.
è Experimental Results ø
(3)

The comparison of the experimental results and the pre- Fig. 6. Performance result of FFBP algorithm developed for model of
dicted results by the FFBP neural network for surface rough- surface roughness.
ness is illustrated in Fig. 7 which shows a good agreement
between the experimental results and the ANN predicted re-
sults for SR. From the Table 5, it is clear that the maximum
error is -2.346% for trial number 25, minimum is 0.006 % for
trial number 31 and overall mean of % relative error is 0.502.
Fig. 8 shows the convergence or gradient of mean square error
(MSE) for surface roughness with the number of epochs dur-
ing training of the selected network. In addition, for analyzing
the capabilities of the network, a linear regression between the
network response and the experimental target value was per-
formed. For the present case, the entire data of SR was put
through for training, validation and testing to perform the re-
gression analysis. The obtained regression results are been
presented separately for the output, which are shown in Fig. 9.
The correlation coefficients (R) are 0.89 for training, 0.95 for
validation, 0.94 for testing and 0.89 for overall in simulating
the Ra. The network can be considered more accurate and Fig. 7. Comparison between the experimental results and the neural
powerful from statistical point of view if the value of correla- network predicted results.
2840 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

ANN predicted value by -0.0056, t (53) = 0.938, p =

0.353>0.05. Therefore, it is concluded that there are not sig-
nificantly difference between the observed experimental value
and ANN predicted value at 95% confidence level.

5. Analysis of surface topography

To scrutinize the surface properties of the machined surface,
samples were cleaned with liquid acetone and after that etched
with Kroll’s reagent solution followed by drying in air. Micro-
structural analysis was carried out on a scanning electron mi-
croscopy (SEM) analyzer (JEOL, JSM-6610LV, JAPAN) to
examine the surface characteristics after EDM. Weight per-
centage of various chemical elements comes through energy
dispersive spectroscopy (EDS) analyzer. To investigate the
possibility of formation of various compounds on the ma-
chined workpiece surface, X-Ray diffraction (XRD) analyses
Fig. 8. Learning behavior of the FFBP neural network model. was performed using PANALYTICAL X- ray diffractometers,
Netherlands along with X’Pert High Score Plus and Origin
Pro 8 software.

5.1 Microstructural analysis by SEM

The machined surface in EDM can be distinguished by the

surface characteristics such as; surface roughness, residual
stresses, micro-cracks, surface cracks, recast white layer, mi-
cro-pores, micro-hardness, micro-structural changes, deposi-
tion of carbon and other elements. The quality of a surface is
the most important characteristics which affects the functional
efficiency of the product. Due to the requirement of accurate
and precise size of the products, the quality of the machined
surface becomes more significant. It is a general and well
known conclusion given by the various researchers and, also
from the results of the present study it was concluded that
current and the pulse-on-time are the most significant factor
which influence surface roughness. Higher values of current
and pulse-on-time means higher discharge energy during the
spark for long time and vice-versa.
Two EDMed samples of titanium alloy TITAN 15 (Ti) and
TITAN 21(Ti-5Al-2.5Sn) of minimum and maximum Ra
Fig. 9. Linear regression analysis between the experimental values and
value were selected for analysis of the surface characteristics.
the FFBP-ANN predicted values of surface roughness (Ra) for train-
ing, validation, testing and overall. The machined sample which had maximum Ra value 9.16 μm
(experiment number 9, Table 2) was machined at current 14A,
pulse on and off time 150 and 30 μs respectively, SCT tita-
tion coefficients (R) approaches near to one. Table 6 shows nium (Ti) workpiece with SCT copper-tungsten electrode in
the values for mean square error and regression error of 38 presence of EDM. oil. As current increases it resulted in in-
samples for training, 08 samples for validation and 08 samples creased density of discharge energy as well as impulsive force.
for testing. The large discharge energy created more depth on the work-
Other statistical analysis namely correlation test and paired- piece. Hence, larger and deeper craters were formed, which
sample t-test were performed to measure the variation be- led to the increase in SR (refer SEM Fig. 10(a)). Furthermore,
tween the experimental and ANN predicted value obtained as pulse-on-time increases, expansion of plasma channel takes
from the network. The observed results are listed in Tables 7 place, resulting in decrease in the plasma flushing efficiency.
and 8. Table 7 shows that experimental value and ANN pre- Due to this phenomenon, both energy density and impulsive
dicted value are positively correlated, r (N = 54) = 1.00. Fur- force are reduced. Thus, the size of molten crater increases as
ther, from Table 8, it can be concluded that the mean surface it could not be flushed away because of the reduction of the
roughness value decreased from the experimental value to the impulsive force. This results in re-solidification which causing
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2841

(a) (b) (a) (b)

6000
5000 TiC TiC
TiC 5000
4000 TiC TiC
4000 Al2Ti4C

Counts
3000 TiC
Counts

TiC 3000 Al2Ti4C

2000 TiC 2000
TiC TiC
TiC Al2Ti4C TiC
1000 TiC 1000 TiC TiC Al Ti C
Al2Ti4C TiC 2 4
0
0
20 40 60 80 100 120
20 40 60 80 100 120 Position (2Theta)
Position (2Theta)
(c)
(c)
Fig. 11. (a) SEM micrographs; (b) EDS spectrum; (c) XRD pattern of
Fig. 10. (a) SEM micrographs; (b) EDS spectrum; (c) XRD pattern of WCT Ti-5Al-2.5Sn alloy machined with current 6A, Ton = 90 μs, Toff =
SCT Ti machined with current 14A, Ton = 150 μs, Toff = 30 μs, SCT Cu- 45 μs, SCT Cu-W electrode in tungsten powder mixed dielectric.
W electrode in EDM oil dielectric fluid.

formation of white recast layer on the machined surface dur- in yield stresses which results in formation of cracks [1]. Figs.
ing the cooling period thus degrading the SR [36, 37]. The 10(a) and 11(a) show cracks developed on the machined sur-
formation of craters of smaller and larger dimensions, recast face. In addition to that, machining at 14A current along with
layer and un-flushed debris particles are clearly visible in the pulse-on-time 150 μs, the width of the surface cracks in-
Fig. 10(a) which makes the surface finish poorer. On the other creased (see Fig. 10(a)) as compared to 6A current and 90 μs.
hand, machining with low current gives a better surface finish
because of the low discharge energy in the machining zone.
5.2 Chemical composition analysis by EDS
The machined surface with minimum Ra value 3.68 μm for
trial number 10 (Table 2) of L18 array was machined at current The EDS analysis was performed to provide information
6A, pulse-on 90 μs, pulse-off 45 μs, and WCT Ti-5-2.5 alloy about the constituents of the machined surface. The EDS
with SCT Cu-W electrode in the presence of suspended tung- analysis shows the presence of various elements on the work
sten powder in dielectric. Due to low discharge energy (cur- materials, because of the elements that migrate from the elec-
rent and pulse-on-time) smaller size of craters were formed on trode materials as well as from dielectric media. Presence of
the surface, which resulted in improved surface finish (refer the different elements is represented by peaks on the EDS
Fig. 11(a)). spectra. Thus, high peaks represent the higher values of that
During the EDM process, heat energy is applied in the form particular element present in the spectrum. Fig. 10(b) presents
of bombardment of electrons between the gaps for each and the EDS spectrum of machined Ti titanium and shows the
every discharge pulse. Due to this mechanism, temperature presence of different chemical elements such as carbon (C),
goes very high usually within the range of 8000°C to 12000°C titanium (Ti), copper (Cu) and tungsten (W). In this case, Ti
at the closet point between the two electrodes. This high tem- element is the base material, Cu and W elements migrated
perature is responsible for developing the thermal / residual from the Cu-W electrode. Carbon transferred from the dielec-
stresses at the machined surface during the EDM process. tric fluid after its decomposition into chemical components
Further, when the magnitude of these developed stresses ex- due to high temperature. As FERROLAC 3M EDM dielectric
ceeds the ultimate tensile strength of the workpiece materials, oil is hydrocarbon composition, thus after decomposition car-
cracks areformed on the surface. The continuous heating and bon precipitated on the surface and is responsible for the for-
cooling of the machined surface contributes in rapid increase mation of titanium carbide compound on the machined sur-
2842 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

face. In another case, when WCT Ti-5Al-2.5Sn alloy was The other phases of aluminum titanium carbide (Al2Ti4C2)
machined with SCT copper-tungsten electrode in the presence were also observed at the different 2θ position. The maximum
of tungsten powder, different elements were observed on the count of Al2Ti4C2 was obtained 18.164 at 2θ position 53. 580.
surface. Peaks of titanium (Ti), aluminum (Al), Tin (Sn), car- The number of counts of Al2Ti4C2 was very less as compared
bon (C) and tungsten (W) were noticed on the surface. Migra- to TiC. Further, some other chemical compound also observed
tion of small amount of Cu, from the tool material and W such as tin titanium carbide (SnTi2C), tin titanium tungsten
from electrode as well as tungsten powder mixed dielectric oxide (Sn2TiWO7), copper titanium oxide (Cu2TiO3), gallium
was also noticed. Large percentage of carbon migrated from titanium carbide (Ga2Ti4C2), titanium zinc carbide (Zn2Ti4C),
the dielectric fluid was observed. Fig. 11(b) shows the EDS rutile (TiO2) and titanium oxide (TiO). Due to their very low
spectra of machined Ti-5-2.5 alloy. weightage, these compounds are not shown in the XRD pat-
The duration for which element remain in molten state de- tern.
pends on the temperature of the elements. The higher meting The various phases and compounds discussed above were
temperature elements remain for longer time as compared to formed on the machined surface due to interaction with the
the lower melting temperature element in molten state. There- parent metal elements of workpiece and dielectric fluid with-
fore, it is clear that element of lower range of melting tem- out powder or with suspended powder particles.
perature solidify first [11]. All these elements which are ob-
served in EDS analysis contribute to formation of different
6. Conclusions
types of compounds. The major compound titanium carbide
(TiC) was formed on the machined surface due to the large In the present study, the objective function was to study the
amount of transferred carbon from EDM oil dielectric [34]. surface characteristics of three types of cryogenically treated
Further, the developed layer of hard and brittle TiC required titanium alloys. A total of eight factors were considered for
more dicharge energy to melt it, because melting temperature studying their impact on surface roughness. A combination of
of TiC is very high. Hence, it is responsible for low machining optimum input parameters (A2B1C1D3E2F1G3H3) for minimiz-
efficiency with degradation of surface finish. ing surface roughness were identified using Taguchi approach.
ANOVA analysis has been conducted to know the impact of
each parameter on SR. Peak current was the most significant
5.3 Analysis using X-ray diffraction (XRD)
factor followed by pulse-on- time and pulse-off -time and
X-Ray Diffraction analyses were conducted on the ma- workpiece material. The contribution of dielectric fluid (Mn
chined workpiece surface to find out the chemical compounds and W powder), electrode material and their cryogenic treat-
precisely. The XRD analysis was performed between 2θ range ment was observed as less significant. The effect of cryogenic
of 20.0084 to 119.9854 with step size (2θ) 0.0170 and scan treatment of workpiece material was found to be negligible.
time 20.9550 sec using Cu K-Alpha anode material on 10 mm The structural design of neural network was selected, trained,
specimen length. The outcome of XRD analysis (experiment validated, tested and then used for simulation to optimize the
9) for machined SCT Ti grade II alloy surface is presented in surface roughness. Experimental results and ANN predicted
Fig. 10(c). Due to the penetration of carbon on the machined results showed good agreement. The maximum percentage
Ti surface, major formation of titanium carbide (TiC) was relative error of 2.346% and a minimum of 0.006% were ob-
noticed on the workpiece surface. The pattern shows the tained. Surface topography analysis was carried out using
phases of TiC at position-2 theta values 36.100, 41.891, SEM, EDS and XRD on two grades of titanium material. The
60.626, 72.521, 76.286, 90.911, 101.843 and 105.605. The SEM pictures show that high discharge energy was responsi-
highest peak of titanium carbide with maximum counts 4490 ble for the surface defects such as; surface or thermal cracks,
was observed at 2θ values 41.891 (Cu K-alpha). The TiC craters, thick recast layer, micro pores, pin holes, residual
compounds have the highest score of 71. Moreover, some stresses and debris. Migration of different chemical elements
other compounds were also formed such as germanium tita- from electrode and dielectric media were observed in EDS
nium carbide (Ge2Ti4C2) and titanium oxide (TiO), but due to spectrum. Presence of large % of carbon was noticed in EDS
a very low score, these compounds are not shown on the XRD analysis. XRD results shows major formation of TiC com-
plot. pound which precipitated on the machined surface. The study
For another case, Fig. 11(c) shows the XRD pattern of provides an insight into the migration of different elements
without cryo-treated titanium alloy Ti-5Al-2.5Sn machined in from the tool and the powder mixed dielectric fluid on to the
the presence of tungsten powder mixed dielectric fluid with machined surface thereby improving the surface properties.
shallow cryo-treated Cu-W electrode, 6A current, 90μs pulse-
on-time and 45 μs pulse-off-time. The carbides of titanium
References
were exposed on the XRD pattern at different 2-theta position
36.091, 40.536, 41.874, 60.683, 72.480, 76.263, 90.896, [1] A. Hascalik and U. Caydas, Electric discharge machining of
101.801, 102.246 and 105.615. TiC compound has maximum titanium alloy (Ti-6Al-4V), Applied Surface Scence, 253
peak height with 5002.6 counts obtained at 41.874 2θ position. (2007a) 9007-9016.
 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844 2843

[2] E. O. Ezugwu and Z. M. Wang, Titanium alloys and their parameters optimization of increased explosive electric dis-
machinability - a review, Journal of Materials Processing charge grinding process for large area polycrystalline dia-
Technology, 68 (1997) 262-274. mond, Journal of Materials Processing Technology, 149
[3] S. Pervaiz, I. Deiab and B. Darras, Power consumption and (2004) 106-111.Y.
tool wear assement when machining titanium alloys, Inter- [18] Tzeng and F. Chen, Multi-objective optimisation of high-
national Journal of Precision Engineering Manufacturer, 14 speed electrical discharge machining process using a Ta-
(6) (2013) 925-936. guchi fuzzy-based approach, Materials and Design, 28
[4] D. A. Dornfeld, J. S. Kim, H. Dechow, J. Hewsow and L. J. (2007) 1159-1168.
Chen, Drilling burr formation in titanium alloy Ti-6Al-4V, [19] S. Assarzadeh and M. Ghoreishi, Neural-network-based
Annals of CIRP., 48 (1999) 73-76. modeling and optimization of the electro-discharge machin-
[5] M. A. Norliana, D. G. Solomon and Md. F. Bahari, Areview ing process, International Journal of Advanced Manufactur-
on current research trends in electrical discharge machining ing Technology, 39 (2008) 488-500.
(EDM), International Journal of Machine Tools & Manufac- [20] A. P. Markopoulos, D. E. Manolakos and N. M. Vaxevani-
ture, 47 (2007) 1214-1228. dis, Artificial neural network models for the prediction of
[6] B. H. Yoo, B. K. Min and J. S. Lee, Analysis of the machin- surface roughness in electrical discharge machining, J. Intell
ing characteristics of EDM as functions of the mobilities of Manuf., 19 (2008) 283-292.
electrons and ions, International Journal of Precision Engi- [21] P. Fonda, Z. Wang, K. Yamazaki and Y. Akutsu, A funda-
neering Manufacturer, 11 (4) (2010) 629-632. mental study on Ti-6Al-4V’s thermal and electrical proper-
[7] K. H. Ho and S. T. Newman, State of the art electrical dis- ties and their relation to EDM productivity, Journal of Mate-
charge machining, International Journal of Machine Tools rials Processing Technology, 202 (2008) 583-589.
& Manufacture, 43 (2003) 1287-1300. [22] G. Rao, G. Krishna Mohana, D. Rangajanardhaa, H. Rao
[8] S. S. Gill and J. Singh, Effect of deep cryogenic treatment on and M. S. Rao, Development of hybrid model and optimiza-
machinability of titanium alloy (Ti-6246) in electric dis- tion of surface roughness in electric discharge machining us-
charge drilling, Materials and Manufacturing Processes, 25 ing artificial neural networks and genetic algorithm, Journal
(2010) 378-385. of Materials Processing Technology, 209 (2009) 1512-1520.
[9] S. Abdulkareem, A. A. Khan and M. Konneh, Reducing [23] M. M. Rahman, Md. Khan, A. Rahman, K. Kadirgama, M.
electrode wear ratio using cryogenic cooling during electri- M. Noor and R. A. Bakar, Modeling of material removal on
cal discharge machining, International Journal of Advanced machining of Ti-6Al-4V through EDM using copper tung-
Manufacturing Technology, 45 (2009) 1146-1151. sten electrode and positive polarity, International Journal of
[10] K. A. Venugopal, S. Paul and A. B. Chattopadhyay, Tool Mechanical Material Engineering, 1 (3) (2010) 135-140.
wear in cryogenic turning of Ti-6Al-4V alloy, Cryogenics, [24] P. K. Patowari, P. Saha and P. K. Mishra, Artificial neural
47 (2007) 12-18. network model in surface modification by EDM using tung-
[11] J. S. Soni and G. Chakraverti, Machining characteristics of sten-copper powder metallurgy sintered electrodes, Interna-
titanium with rotary electro-discharge machining, Wear, 171 tional Journal of Advanced Manufacturing Technology, 51
(1994) 51-58. (2010) 627-638.
[12] R. K. Jain and V. K. Jain, Optimum selection of machining [25] C. J. Tzeng and R. Y. Chen, Optimization of electric dis-
conditions in abrasive flow using neural network , Journal of charge machining process using the response surface meth-
Materials Processing Technology, 108 (2000) 62-67. odology and genetic algorithm approach, International
[13] K. Wang, H. L. Gelgele, Y. Wang, Q. Yuan and M. Fang, Journal of Precision Engineering Manufacturer, 14 (5)
hybrid intelligent method for modelling the EDM process, (2013) 709-717.
International Journal of Machine Tools & Manufacture, 43 [26] T. T. Nguyen, Y. S. Yang and J. W. Kim, An artificial
(2003) 995-999. neural network system for predicting the deformation of
[14] P. S. Bharti, S. Maheshwari and C. Sharma, Multi-objecttive steel plate in triangle induction heating process, Interna-
optimization of electric-discharge machining process using tional Journal of Precision Engineering Manufacturer, 14
controlled elitist NSGA-II, Journal of Mechanical Science (4) (2013) 551-557.
and Technology, 26 (6) (2012) 1875-1883. [27] S. S. Sidhu, A. Batish and S. Kumar, Neural-network-based
[15] J. Y. Kao and Y. S. Tarng, A neutral-network approach for modeling to predict residual stresses during electric dis-
the online monitoring of the electrical discharge machining charge machining of Al/SiC-MMCs., Proceedings of the In-
process, Journal of Materials Processing Technology, 69 (1- stitution of Mechanical Engineers Part B-Journal of Engi-
3) (1997) 112-119. neering Manufacture (2013) doi: 10. 7/0954405413492505.
[16] Fenggou and Y. Dayong, The study of high efficiency and [28] M. M. Rahman, Modeling of machining parameters of Ti-
intelligent optimization system in EDM sinking process, 6Al-4V for electric discharge machining: A neural network ap-
Journal of Materials Processing Technology, 149 (1-3) proach, Scientific Research and Essays, 7 (8) (2012) 881-890.
(2004) 83-87. [29] J. Y. Kao, C. C. Tsao, S. S. Wang and C. Y. Hsu, Optimi-
[17] Fenggou and Q. Zhang, Neural network modeling and zation of the EDM parameters on machining Ti-6Al-4V with
2844 S. Kumar et al. / Journal of Mechanical Science and Technology 28 (7) (2014) 2831~2844

multiple characteristics, International Journal of Advanced ters on the performance of EDM process with ultrasonic as-
Manufacturing Technology, 47 (2010) 395-402. sisted cryogenically cooled electrode, Journal of Manufac-
[30] P. C. Tan, S. H. Yeo and Y. V Tan, Effects of nano powder turing Processes, 14 (3) (2012) 393-402.
additives in micro electrical discharge machining, Interna- [37] H. T. Lee and T. Y. Tai, Relationship between EDM pa-
tional Journal of Precision Engineering Manufacturer, 9 rameters and surface crack formation, Journal of Materials
(3) (2008) 22-26. Processing Technology, 142 (3) (2003) 676-683.
[31] R. N. Yadav and V. Yadava, Experimental study of erosion [38] P. J. Ross, Taguchi techniques for quality engineering,
and abrasion based hybrid machining of hybrid metal matrix USA 2nd ed., McGraw hill, New York (1996).
composite, International Journal of Precision Engineering
Manufacturer, 14 (8) (2013) 1293-1299.
[32] B. H. Yan, H. C. Tsai and F. Y. Huang, The effect in EDM Sanjeev Kumar received B.E. (1996)
of a dielectric of a urea solution in water on modifying the degree in Mechanical Engineering from
surface of titanium, International Journal of Machine Tools NIT, Kurukshetra, India and M.Tech.
& Manufacture, 45 (2005) 194-200. (2007) in Production Engineering from
[33] B. Jabbaripour, M. H. Sadeghi, S. Faridvand and M. R. GNDEC, Ludhiana, India. He is pursu-
Shabgard, Investigating the effects of EDM parameters on ing his Ph.D. in the Department of Me-
surface integrity, MRR and TWR in Machining of Ti-6Al- chanical Engineering at Thapar Univer-
4V, Machining Science and Technology: AnInternational sity (Deemed University), Patiala, India.
Journal, 16 (2012) 419-444.
[34] A. Hascalik and U. Caydas, A comparative study of surface Ajay Batish is working as a Professor
integrity of Ti-6Al-4V alloy machined by EDM and AECG, and Head in the Department of Me-
Journal of Materials Processing Technology, 190 (2007b) chanical Engineering Thapar Univer-
173-180. sity (Deemed University), Patiala, India.
[35] S. Kalia, Cryogenic processing: a study of materials at low
temperatures, Journal of Low Temperature Physics, 158
(2010) 934-945.
[36] V. Srivastava and M. P. Pandey, Effect of process parame-