You are on page 1of 17

Int J Adv Manuf Technol (2015) 79:31–47

DOI 10.1007/s00170-015-6797-8

ORIGINAL ARTICLE

Parametric modeling and optimization for wire electrical


discharge machining of Inconel 718 using response surface
methodology
Vivek Aggarwal & Sehijpal Singh Khangura & R. K. Garg

Received: 5 March 2014 / Accepted: 11 January 2015 / Published online: 28 January 2015
# Springer-Verlag London 2015

Abstract Inconel 718 is a high-nickel-content superalloy Keywords Wire electrical discharge machining . Inconel
which possesses excellent strength at elevated temperatures 718 . Response surface methodology . Cutting rate . Surface
and resistance to oxidation and corrosion. This alloy has wide roughness
applications in the manufacturing of aircraft engine parts such
as turbine disks, blades, combustors and casings, extrusion
dies and containers, and hot work tools and dies, but the in- 1 Introduction
herent problems in machining of superalloys with convention-
al techniques necessitate the use of alternative machining pro- The machining of superalloys is an active research area
cesses. The wire electrical discharge machining (WEDM) pro- because of the widespread increase in demand of this
cess has been recently explored as a good alternative of con- category of materials and characteristic problems related
ventional machining methods, but there is lack of data and to their machining. Inconel 718 is a high-strength temper-
suitable models for predicting the performance of WEDM ature-resistant (HSTR) nickel-based superalloy which ex-
process particularly for Inconel 718. In the present work, hibits good resistance to corrosion and oxidation along
empirical modeling of process parameters of the WEDM with high creep-rupture strength and fatigue endurance
has been carried out for Inconel 718 using a well-known limit [1, 2]. It is extensively used in the aerospace indus-
experimental design approach called response surface try for manufacturing of gas turbine engine components
methodology. The parameters such as pulse-on time, such as turbine disks, blades, combustors and casings,
pulse-off time, peak current, spark gap voltage, wire feed nuclear power plant components such as reactor and
rate, and wire tension have been selected as input variables pump, spacecraft structural components, medical devices,
keeping others constant. The performance has been measured food processing equipment, extrusion dies and containers,
in terms of cutting rate and surface roughness. The models casting dies, hot work tools and dies, etc. [3, 4]. Machin-
developed are found to be reliable representatives of the exper- ing of Inconel 718 with conventional techniques is ex-
imental results with prediction errors less than ±5 %. The opti- tremely difficult because of its high toughness, hardness,
mized values of cutting rate and surface roughness achieved work hardening tendency, low thermal conductivity, and
through multi-response optimization are 2.55 mm/min and presence of hard abrasive particles [5]. Therefore, non-
2.54 μm, respectively. conventional machining methods based on chemical,
electro-chemical, thermal, thermoelectric, and mechanical
energy are preferred over traditional methods for the ma-
chining of Inconel 718. Wire electrical discharge machin-
V. Aggarwal (*) : R. K. Garg
ing (WEDM) is a non-conventional, thermoelectric process
Department of Industrial and Production Engineering,
Dr B R Ambedkar National Institute of Technology, that can be used to cut complex and intricate shapes in all
Jalandhar 144011, Punjab, India electrically conductive materials used in tool and die, au-
e-mail: agarwalz_v@yahoo.com tomobile, aerospace, dental, nuclear, computer, and elec-
tronic industries with better precision and accuracy [6].
S. S. Khangura
Department of Mechanical Engineering, Guru Nanak Dev WEDM is a well-established process and its working is
Engineering College, Ludhiana 141006, Punjab, India duly described in the literature [7–10]. The most important
32 Int J Adv Manuf Technol (2015) 79:31–47

performance measures in WEDM are material removal rate width and material removal rate while machining AISI 4140
(or cutting rate), surface finish, kerf (cutting width), and steel in WEDM process using CuZn37 brass wire. Han et al.
wire wear rate. These measures, in turn, are influenced by [18] studied the effect of process parameters on surface rough-
numerous machining parameters such as peak current, ness during wire electrical discharge machining of Cr12 alloy
pulse-on time, pulse-off time, wire tension, wire feed rate, steel with brass wire. Mahapatra and Patnaik [19] established
spark gap voltage, servo feed setting, average working the optimized relationship between machining parameters and
voltage, and dielectric flushing condition [11]. Owing to responses like cutting rate, surface roughness, and cutting
a large number of process parameters and a complex na- width by employing non-linear regression analysis and geo-
ture of the process, even a highly skilled operator with a metric algorithm in WEDM operations of D2 tool steel with
state-of-the-art WEDM is rarely able to achieve the opti- zinc-coated copper wire. Kondayya and Krishna [20] applied
mal performance [12]. The improperly selected parameters an integration of two evolutionary approaches namely genetic
may also result in serious consequences like short- programming and non-dominated sorting genetic algorithm-II
circuiting of wire and wire breakage that in turn reduces in WEDM process for modeling and optimization of cutting
productivity. An effective way to solve this problem is to rate and surface roughness while machining AISI D3 steel
determine the relationship between the performance mea- with uncoated brass wire. Kuruvila and Ravindra [21] applied
sures and the controllable input parameters using a suit- Taguchi’s technique and genetic algorithm for determining
able modeling and optimization technique. parametric influence and optimizing dimensional error, sur-
face roughness, and volumetric material removal rate during
WEDM of hot die steel using molybdenum wire. Sadeghi
2 Literature review et al. [22] investigated the influence of various process pa-
rameters on the cutting rate and surface roughness during
The available literature on WEDM throws light on its appli- WEDM of AISI D5/DIN 1.2601 steel using CuZn37 brass
cation for machining of a vast range of materials such as wire in accordance with Taguchi’s experimental design
steels, composites, titanium alloys, and superalloys, but major method. Kumar and Agarwal [23] optimized the machining
studies are focused on machining of steels. Most of these conditions for achieving maximum cutting rate and surface
studies have been reported on modeling and optimization of finish using multi-objective optimization technique based on
machining parameters like pulse-on time, pulse-off time, peak non-dominated sorting genetic algorithm-II in WEDM pro-
current, wire feed rate, wire tension, dielectric flow rate, and cess. The high-speed steel (M2, SKH9) work material and
servo feed setting. A brief review of the techniques used by zinc-coated copper wire were used in the research work
various researchers is presented in this section. Scott et al. [11] for performing experiments. Sharma et al. [24] developed
formulated a mathematical model for predicting the material mathematical models to study the effect of different pro-
removal rate and surface finish while machining D2 tool steel cess parameters of WEDM on cutting speed and dimen-
at different WEDM conditions. To find the optimal machining sional deviation while machining high-strength low-alloy
parameters, the non-dominated point approach was applied, (HSLA) steel using brass wire.
using both the explicit enumeration and dynamic program- The literature further reveals the use of WEDM for other
ming principles. Tarng et al. [13] utilized a feedforward neural materials like titanium alloys, composites, etc. Kuriakose
network model and applied simulated annealing algorithm in et al. [25] modeled the WEDM process using a machine
order to predict and optimize the surface roughness and ma- learning-based data mining approach to study the effect of
chining speed during wire electrical discharge machining of various process parameters on the cutting speed and surface
SUS-304 stainless steel materials with brass wire taken as tool finish. The experiments were performed on titanium alloy
electrode. Spedding and Wang [14, 15] attempted to model (Ti-6Al-4V) using zinc-coated brass wire as tool electrode.
WEDM process through artificial neural networks for Sarkar et al. [26] formulated mathematical models to pre-
predicting and optimizing the cutting speed, surface rough- dict the cutting speed, surface finish, and dimensional de-
ness, and surface waviness parameters using a constrained viation as the function of various process parameters by
optimization model while machining AISI 420 steel using applying constrained optimization and Pareto optimization
brass wire. Huang et al. [16] experimentally investigated the algorithm in WEDM of γ-titanium aluminide alloy using
effect of various machining parameters on the gap width, sur- brass wire. Chen et al. [27] proposed an optimization approach
face roughness, and the depth of white layer on the machined that integrated back-propagation neural network and simulat-
workpiece (SKD11 alloy steel) surface with brass wire during ed annealing algorithm to optimize cutting velocity and sur-
WEDM process. The feasible-direction non-linear program- face finish in WEDM process during manufacture of pure
ming method was adopted for determination of the optimal tungsten profiles with CuZn35 brass wire. Yu et al. [28] in-
process settings. Tosun et al. [17] used regression analysis and vestigated the effects of different machining parameters on
simulated annealing searching to model and optimize kerf cutting speed, machining groove width, and surface roughness
Int J Adv Manuf Technol (2015) 79:31–47 33

in WEDM process while machining polycrystalline silicon It is also observed that the available research work that has
using brass wire. Chiang and Chang [29] presented the appli- been carried out for WEDM of Inconel 718 investigated the
cation of grey relational analysis for optimizing surface rough- impact of a limited number of process parameters. The effect
ness and cutting rate during WEDM of Al2O3 particle- of spark gap voltage on performance is studied by very few
reinforced material (6061 alloy) with copper wire electrode. researchers, although it may play an influential role in WEDM
Manna and Bhattacharyya [30] developed mathematical process. Therefore, in this research work, WEDM of Inconel
models for material removal rate, surface roughness, spark 718 has been carried out in order to develop the empirical
gap, and gap current using the Gauss elimination method in models for investigating the impact of various process param-
WEDM for effective machining of aluminum-reinforced eters including spark gap voltage on performance measures to
silicon carbide metal matrix composite (Al/SiC-MMC) obtain the optimum machining conditions. The present work
using brass as wire material. Patil and Brahmankar [31] is largely focused on machining of Inconel 718 with reference
experimentally investigated the effect of different machining to the applications like manufacturing of extrusion dies
parameters on surface finish, cutting speed, and kerf width and containers, hot work tools and dies, casting dies, food
using Taguchi’s method in WEDM of alumina particulate- processing equipment, etc. The actual values of input pro-
reinforced aluminum matrix composites (Al/Al2O3p) with cess parameters have been considered to make the models
coated brass wire material. more realistic and useful.
Very few research results are reported on WEDM of super-
alloys particularly Inconel series. Hewidy et al. [32] devel-
oped mathematical models for correlating the various machin- 3 Experimental procedure
ing parameters with volumetric metal removal rate, wear ratio,
and surface roughness based on the response surface method- A series of experimental trials have been conducted as per
ology during WEDM of Inconel 601 using copper wire. response surface methodology (RSM). The details about the
Ramakrishnan and Karunamoorthy [33, 34] developed artifi- work material, experimental set-up and measuring apparatus,
cial neural network models and optimized material removal selection of process parameters and their range, design of
rate and surface roughness using multi-response signal-to- experiments, and reproducibility have been explained in
noise ratio. A brass wire was used as tool electrode to cut the following sections.
Inconel 718 using WEDM process; however, no empirical
models have been reported. 3.1 Work material
The extensive review of available literature on WEDM
reveals that most of the research work has been carried out There is a rising demand for high-strength and high-thermal-
on steels and some other materials/composites. However, as resistant materials which give superior performance under se-
far as Inconel 718 is concerned, comparatively less research vere conditions. Being a high-strength temperature-resistant
has been done. Inconel 718 has a vast range of applications, material, Inconel 718 is a popular superalloy used in a wide
but despite having such a large application domain, relatively range of applications. Thus, in the present study, a commer-
fewer papers have been reported on the use of WEDM for cially available Inconel 718 is selected as work material for
machining Inconel 718. This may be due to the reasons experimentation. In all the experimental trials, the height
that Inconel 718, a difficult-to-cut material, has typical of the work material selected is 20 mm and it is cut in a
characteristics like presence of hard abrasive carbides in width of 20 mm in each experiment. The chemical com-
the microstructure, low thermal conductivity and specific position of the selected work specimen provided by the
heat, and high cutting temperature. Therefore, the choice supplier and verified by electron probe micro-analyzer (EPMA)
of correct parametric combination of WEDM process for is shown in Table 1.
this material becomes a challenging task. There is no ac-
ceptable analytical model for WEDM process due to the 3.2 Experimental set-up and measuring apparatus
presence of a large number of variables and complicated
stochastic process mechanism which necessitates the use of In this research work, all experiments have been carried out on
empirical or experimental studies of WEDM process with a WEDM machine tool (Elektra Sprintcut 734). The experi-
the goal of achieving mathematical models to enhance the ments are aimed at studying the effects of several controllable
process performance. process parameters on cutting rate and surface roughness

Table 1 Chemical composition


of Inconel 718 Element Ni Fe Cr Nb Mn C Co Al Si Ti Mo Other

Weight (%) 51.05 19.43 18.70 5.7 0.07 0.04 0.2 0.56 0.08 1.01 3.1 0.06
34 Int J Adv Manuf Technol (2015) 79:31–47

taken as performance measures. A zinc-coated brass wire Table 3 Constant parameters and their values
with 0.25 mm diameter is used as wire electrode because Parameter Description or value
it gives the best performance in terms of material removal
rate as compared to other wire materials namely plain and Wire material Zinc-coated brass wire
diffused brass wires [35]. The wire is connected to nega- Wire diameter 0.25 mm
tive polarity whereas the workpiece is maintained at pos- Workpiece material Inconel 718
itive polarity. The wire diameter is kept constant during Workpiece hardness 439 HV
the process. A fixture is used to hold the workpiece on Dielectric fluid Deionized water
the machine table to eliminate any possibility of misalign- Conductivity of dielectric fluid 20 mho
ment. The workpiece and the wire electrode are separated Flushing pressure of dielectric fluid 15 kg/cm2
by deionized water which is used as the dielectric medi- Pulse peak voltage 110 V DC
um. The cutting rate is displayed digitally on the control Servo feed 2100 units
panel of the machine in millimeters per minute whereas Machining width 20 mm
Mitutoyo’s surftest instrument SJ-301 has been used to Machining height 20 mm
measure the surface roughness (Ra) of each machined
specimen in micrometers. The surface roughness is measured
in terms of arithmetic mean roughness. A sampling length of
2.5 mm in five steps has been selected for the measurement. design, to develop the mathematical relationship between pro-
cess parameters and each output response (cutting rate and
3.3 Selection of process parameters and their range surface roughness), and to analyze the effects of process pa-
rameters on these responses. Box and Wilson [36] originally
In the present work, the effect of various process parameters developed RSM to explore the potential of statistical design in
(factors) such as pulse-on time (Ton), pulse-off time (Toff), peak industrial experiments. The experimentation along with re-
current (IP), spark gap voltage (SV), wire feed rate (WF), and gression analysis facilitates the modeling of the desired re-
wire tension (WT) on the cutting rate and surface roughness sponse to several input process parameters. The experiment
(response parameters) has been investigated. These process is designed to allow estimation of interaction and even qua-
parameters and their range have been selected on the basis dratic effects and, thus, provides an idea of the local shape of
of the existing literature, pilot experimentation, manufac- the response surface. Hence, it is termed as response surface
turer’s manual, and machine capability. The independent pro- method design [37]. In RSM, a relationship is formed between
cess parameters and their levels in coded and actual values are the desired response and the independent input parameters
shown in Table 2. The constant parameters and their values are which can be represented by Eq. (1):
indicated in Table 3.
yi ¼ f ðx1 ; x2 ; x3 ; …::; xk Þ  ε ð1Þ
3.4 Design of experiments
where yi is the desired response, f is the response function and
Response surface methodology’s central composite design ap- x1, x2, …, xk are independent input process parameters. The
proach has been employed to perform the experimental fitting error, ε, also called residual, measures the experimental

Table 2 Independent process parameters (factors), their description, units, notations, and levels

Process parameter Description Unit Notation Levels


(symbol)
−1.565 (−α) −1 0 1 1.565 (α)

Pulse-on time (A) It is the duration of time for which the current is flowing μs Ton 0.4 0.6 0.9 1.2 1.4
in each cycle
Pulse-off time (B) It is the time interval between two simultaneous sparks μs Toff 14 20 30 40 46
Peak current (C) It is the maximum value of current flowing through A IP 70 100 140 180 210
electrodes for a given pulse
Spark gap voltage (D) It indicates the theoretical voltage difference between V SV 16 28 48 68 80
wire electrode and workpiece during erosion
Wire feed rate (E) The rate at which wire moves through the wire guides m/min WF 2 4 7 10 12
and is fed continuously for sparking
Wire tension (F) It is the gram-equivalent load for keeping the g WT 400 600 1000 1400 1600
continuously fed wire under tension
Int J Adv Manuf Technol (2015) 79:31–47 35

Table 4 Experimental design matrix with set of process parameters and corresponding response parameters

Standard order Run order Location Process parameters (actual values) Response parameters

Ton (μs) Toff (μs) IP (A) SV (V) WF (m/min) WT (g) Cutting rate (mm/min) Surface roughness (μm)

1 30 Factorial 0.6 20 100 28 4 600 1.45 2.56


2 6 Factorial 1.2 20 100 28 4 1400 3.01 3.52
3 48 Factorial 0.6 40 100 28 4 1400 0.79 2.36
4 25 Factorial 1.2 40 100 28 4 600 1.70 3.26
5 23 Factorial 0.6 20 180 28 4 1400 1.56 2.57
6 45 Factorial 1.2 20 180 28 4 600 3.67 3.48
7 27 Factorial 0.6 40 180 28 4 600 0.94 2.55
8 35 Factorial 1.2 40 180 28 4 1400 2.15 3.33
9 4 Factorial 0.6 20 100 68 4 1400 0.73 1.50
10 32 Factorial 1.2 20 100 68 4 600 1.52 2.70
11 24 Factorial 0.6 40 100 68 4 600 0.40 1.51
12 28 Factorial 1.2 40 100 68 4 1400 0.83 2.38
13 19 Factorial 0.6 20 180 68 4 600 0.78 1.55
14 26 Factorial 1.2 20 180 68 4 1400 1.88 3.18
15 47 Factorial 0.6 40 180 68 4 1400 0.47 1.63
16 31 Factorial 1.2 40 180 68 4 600 1.14 3.10
17 21 Factorial 0.6 20 100 28 10 1400 1.46 2.54
18 7 Factorial 1.2 20 100 28 10 600 3.09 3.32
19 41 Factorial 0.6 40 100 28 10 600 0.83 2.40
20 44 Factorial 1.2 40 100 28 10 1400 1.70 3.38
21 1 Factorial 0.6 20 180 28 10 600 1.62 2.67
22 51 Factorial 1.2 20 180 28 10 1400 3.75 3.37
23 37 Factorial 0.6 40 180 28 10 1400 0.90 2.47
24 43 Factorial 1.2 40 180 28 10 600 2.21 3.53
25 3 Factorial 0.6 20 100 68 10 600 0.72 1.40
26 16 Factorial 1.2 20 100 68 10 1400 1.51 2.79
27 40 Factorial 0.6 40 100 68 10 1400 0.43 1.56
28 2 Factorial 1.2 40 100 68 10 600 0.88 2.40
29 50 Factorial 0.6 20 180 68 10 1400 0.79 1.57
30 36 Factorial 1.2 20 180 68 10 600 1.95 3.21
31 42 Factorial 0.6 40 180 68 10 600 0.46 1.58
32 20 Factorial 1.2 40 180 68 10 1400 1.12 3.06
33 29 Axial 0.4 30 140 48 7 1000 0.53 1.37
34 49 Axial 1.4 30 140 48 7 1000 2.36 3.61
35 34 Axial 0.9 14 140 48 7 1000 2.63 3.01
36 15 Axial 0.9 46 140 48 7 1000 0.98 2.74
37 8 Axial 0.9 30 70 48 7 1000 0.93 1.72
38 5 Axial 0.9 30 210 48 7 1000 1.53 2.76
39 46 Axial 0.9 30 140 16 7 1000 2.13 3.04
40 38 Axial 0.9 30 140 80 7 1000 0.64 1.80
41 33 Axial 0.9 30 140 48 2 1000 1.35 2.81
42 22 Axial 0.9 30 140 48 12 1000 1.38 2.85
43 9 Axial 0.9 30 140 48 7 450 1.36 2.64
44 12 Axial 0.9 30 140 48 7 1600 1.32 2.71
45 17 Center 0.9 30 140 48 7 1000 1.31 2.89
46 14 Center 0.9 30 140 48 7 1000 1.27 2.66
47 10 Center 0.9 30 140 48 7 1000 1.37 2.98
36 Int J Adv Manuf Technol (2015) 79:31–47

Table 4 (continued)

Standard order Run order Location Process parameters (actual values) Response parameters

Ton (μs) Toff (μs) IP (A) SV (V) WF (m/min) WT (g) Cutting rate (mm/min) Surface roughness (μm)

48 52 Center 0.9 30 140 48 7 1000 1.42 2.88


49 13 Center 0.9 30 140 48 7 1000 1.38 2.82
50 11 Center 0.9 30 140 48 7 1000 1.28 2.93
51 39 Center 0.9 30 140 48 7 1000 1.34 2.78
52 18 Center 0.9 30 140 48 7 1000 1.41 2.73

errors. The approximation of response function has been points to form a CCD with α=±1.565 for estimation of cur-
proposed using the second-order polynomial regression vature (experimental runs 33 to 44), and eight center points
model, also called quadratic model. The quadratic model (experimental runs 45 to 52) at zero level for replication to
yi for k input process parameters can be expressed by the estimate pure error. The experimental runs are randomized to
following Eq. (2): assure elimination of human biases. The design is generated
and analyzed using Design-Expert software (Stat-Ease, Inc.,
Xk Xk XXk Minneapolis, v 8.0.6.1). The experimental design matrix with
yi ¼ b0 þ bx
i¼1 i i
þ b x2
i¼1 ii i
þ b xx
i< j i j i j
ε ð2Þ a set of process parameters and corresponding response values
(mean) obtained from experimentation are shown in Table 4.

3.5 Reproducibility
where b0 is a constant and bi, bii, and bij represent the
coefficients of linear, quadratic, and cross product terms,
The reproducibility of the process is verified by performing
respectively [37]. The variable xi corresponds to the pro-
eight experiments at the center point of the input variables
cess parameter under study. The response surface, yi, may
using CCD technique. The results of experimental runs
contain linear terms, squared terms, and interaction terms.
(45 to 52) depicted in Table 4 are also shown independently
in Table 5. The percentage error mentioned in Table 5 shows
3.4.1 Central composite design
the variability of the response parameters with reference to its
average value (sum of all terms divided by the number of
Box and Hunter [38] proposed that the scheme based on
terms) and is calculated by the following Eq. (3):
central composite design (CCD) fits the second-order re-
 
sponse surfaces quite accurately. Also, CCD is the most Average value − Experimental value
Percentage error ¼  100
popular among the various classes of RSM designs due to Average value
its flexibility, ability to run sequentially, and efficiency in ð3Þ
providing the overall experimental error in a minimum
number of runs. Therefore, it has been selected in the The results shown in Table 5 indicate that the cutting rate
present work. In CCD, each factor is varied at five levels and surface roughness are reproduced within the acceptable
(−α, −1, 0, 1, α) for developing a second-order model as range of ±5 % [37]. In addition, the values of experimental
given in Eq. (2). When the number of factors (k) is five standard deviation and standard uncertainty have been calcu-
or greater, it is not necessary to run all combinations of lated for response parameters to depict their variability over a
factors. The factorial part of the design can be run using a repeated set of experiments [40]. The values of standard un-
fraction of the total number of available combinations. The certainty for cutting rate and surface roughness are calculated
possible designs options can either be regular fractional as 1.87 and 3.54 %, respectively (Appendix), which are well
factorials or a minimum run resolution of five. within the acceptable range.
Fifty-two sets of experiments have been conducted accord-
ing to the CCD of response surface methods using half repli-
cation for six factors with α=1.565 (α=k1/4). This is known as 4 Development of empirical models for response
practical α and is beneficial when the number of factors is parameters
larger than five [39]. The resolution of this design allows
estimating all the main effects, two-factor interactions, and The complex and stochastic nature of the WEDM process
quadratic effects. The present design of 52 experiments con- makes it difficult to obtain an analytical model based on the
sists of 32 factorial points (experimental runs 1 to 32), 12 axial physics of the process. Therefore, a multi-variable regression
Int J Adv Manuf Technol (2015) 79:31–47 37

Table 5 Reproducibility and percentage (%) error for cutting rate and surface roughness

Ton (μs) Toff (μs) IP (A) SV (V) WF (m/min) WT (g) Cutting rate (mm/min) Surface roughness (μm)

Experimental value % Error Experimental value % Error

0.9 30 140 48 7 1000 1.31 2.78 2.89 −1.98


0.9 30 140 48 7 1000 1.27 5.75 2.66 6.13
0.9 30 140 48 7 1000 1.37 −1.67 2.98 −5.16
0.9 30 140 48 7 1000 1.42 −5.38 2.88 −1.63
0.9 30 140 48 7 1000 1.38 −2.41 2.82 0.48
0.9 30 140 48 7 1000 1.28 5.01 2.93 −3.40
0.9 30 140 48 7 1000 1.34 0.56 2.78 1.90
0.9 30 140 48 7 1000 1.41 −4.64 2.73 3.66

model using response surface methodology has been devel- variance. If the variances are nearly same, the ratio will be
oped for each response which finds out the mathematical rela- close to 1 and it is less likely that model (or any of factor
tionship between process parameters and performance measure terms) has a significant effect on the response. A particular
based on the experimental results. Then, analysis of variance source of variation may be significant if the calculated F value
(ANOVA) is performed in order to statistically analyze the at a certain confidence level is greater than the tabulated F
results of the selected model. Significant process parameters value at the same confidence level. Confidence level is chosen
are identified and interaction effects of these process parame- to be 95 % in this study. If Prob>F value of the model is
ters on response parameters are studied with the help of re- considerably less than 0.05 (i.e., at 95 % confidence level),
sponse surface graphs. then the terms in the model have a significant effect on the
response [37].
4.1 Analysis of variance for cutting rate The model F value of 432.86 with its Prob>F value less
than 0.0001 as shown in Table 6 indicates that the model is
The experimental results for cutting rate shown in Table 4 significant for cutting rate as it demonstrates that the terms in
have been analyzed using Design-Expert software. To check the model have a significant effect on the response. There is
the adequacy of the model, three different tests, sequential only a 0.01 % chance that such a large model F value could
model sum of squares, lack-of-fit test, and model summary occur due to noise. Since values of Prob>F less than 0.0500
statistics, have been performed. The fit summary of these tests indicate the significance of model terms, therefore A, B, C, D,
recommends the adequacy of the quadratic model. The results AB, AC, AD, BD, CD, B2, and C2 are significant model terms
of the quadratic model analyzed using ANOVA reveal that for cutting rate with their contribution percentages of 38.64,
there are many insignificant terms in the model. Therefore, 20.78, 2.54, 27.93, 2.37, 0.9, 3.38, 1.87, 0.15, 1.4, and 0.23 %,
model reduction using the backward elimination process has respectively. These percentage contributions have been com-
been performed to improve the model. It eliminates the insig- puted from Table 6 by dividing the “individual (model) term”
nificant terms in order to adjust the fitted quadratic model sum of squares by “model” sum of squares. Results also show
while maintaining the hierarchy of the model. Table 6 gives that A, B, and D are the most significant model terms affecting
the results of pooled ANOVA after backward elimination. the cutting rate. Values of Prob>F greater than 0.1000 indicate
The normal probability plot of the residuals (Fig. 1a) shows the non-significance of model terms.
that 98 % of residuals are falling within three sigma limits and The lack-of-fit F value of 2.13 (Table 6) implies that it is
are lying on a straight line. This means that the errors are not significant relative to pure error. A non-significant lack of
normally distributed [37]. Further, it can be seen in Fig. 1b fit is good as it allows the model to fit the experimental data
that the actual values are following the predicted ones calcu- and suggests that there may be very less systematic variation
lated from the model. As both the plots (Fig. 1a, b) satisfy the that remains unaccounted for a particular model. In the present
error normality and prediction capability criteria, it is inferred case, there is a 15.06 % chance that a lack-of-fit F value this
that ANOVA results listed in Table 6 are reliable. large could occur due to noise. Hence, the developed model
The model F value is calculated as “model” mean square can be accepted.
divided by the “residual” mean square. Similarly, an F value To check further whether the fitted models actually de-
on any individual factor term is calculated as the term mean scribe the experimental data, determination coefficient (R2)
square divided by the residual mean square. The F value test is computed. The determination coefficient is defined as the
compares the model (or term) variance with the residual ratio of explained variation to the total variation and is a
38 Int J Adv Manuf Technol (2015) 79:31–47

Table 6 Pooled ANOVA results


for cutting rate after backward Source Sum of Degree of Mean square F value Prob>F
elimination squares freedom

Model 29.89 11 2.72 432.86 <0.0001 Significant


A 11.55 1 11.55 1840.27 <0.0001
B 6.21 1 6.21 988.81 <0.0001
C 0.76 1 0.76 121.38 <0.0001
D 8.35 1 8.35 1329.82 <0.0001
AB 0.71 1 0.71 112.78 <0.0001
AC 0.27 1 0.27 42.44 <0.0001
AD 1.01 1 1.01 160.59 <0.0001
BD 0.56 1 0.56 89.49 <0.0001
CD 0.045 91 0.045 7.17 0.0107
B2 0.42 1 0.42 66.66 <0.0001
C2 0.068 1 0.068 10.87 0.0021
Residual 0.25 40 0.006278
Lack of fit 0.23 33 0.006920 2.13 0.1506 Not significant
Pure error 0.023 7 0.003250
Cor total 30.14 51
Standard deviation 0.079 R2 0.9917
Mean 1.43 Adj R2 0.9894
CV % 5.54 Pred R2 0.9837
PRESS 0.49 Adeq Precision 83.784

measure of the degree of fit. When it approaches unity, the model is defined as the ratio of the standard deviation
the response model fits better to the actual data and shows to the mean. The lower value (5.54) of CV % given in
less difference between the predicted and actual values. Table 6 indicates improved precision and reliability of the
The determination coefficient for cutting rate is found to conducted experiments [41]. Further, the Adeq Precision
be 0.9917 (Table 6) which shows that the quadratic model measures the signal-to-noise ratio. Generally, a ratio greater
can explain the variation in the cutting rate up to the than 4 is desirable [42]. The Adeq Precision obtained for
extent of 99.17 %. On the basis of the high values of the model is 83.784, which is well above the desired value
the determination coefficient, it can be said that the pro- and thus indicates an adequate signal for the model. Hence,
posed model is adequate in representing the process. The this model can be used to navigate the design space and
other R2 statistics, the Pred R2 (0.9837), is in good agreement predict the values of the cutting rate within the limits of the
with the Adj R2 (0.9894). The coefficient of variation (CV) of factors studied.

Fig. 1 a Normal probability plot of residuals for cutting rate. b Plot of predicted vs. actual response for cutting rate
Int J Adv Manuf Technol (2015) 79:31–47 39

As the obtained model presents higher value of R2 (0.9917) summary, the adequacy of the quadratic model has been
and Adeq Precision (83.784) for the cutting rate, therefore, recommended by Design-Expert software. In order to sta-
this mathematical model can be regarded as significant for tistically analyze the results of recommended quadratic
fitting and predicting the experimental results. model, ANOVA is performed. Table 7 shows the results
of pooled ANOVA after backward elimination of non-
4.1.1 Regression equation for cutting rate significant terms.
The normal probability plot of the residuals for the surface
In terms of actual factors, the final empirical relationship roughness is shown in Fig. 2a. A check on this plot re-
between cutting rate (response characteristic) and input veals that 98 % of residuals are falling within three sigma
process parameters can be expressed by the following limits. Further, it can be seen in Fig. 2b that the actual
second-order polynomial Eq. (4): values are following the predicted values calculated from
the model, therefore confirming the reliability of ANOVA
Cutting rate ¼ 1:24642 þ 3:69173  T on −0:12058  T off þ 0:00878 results (Table 7).
 IP−0:01040  SV−0:04958  T on  T off þ 0:00760 A model F value of 83.46 as shown in Table 7 implies that
 T on  IP−0:02958  T on  SV þ 0:00066  T off  SV the model is significant. There is only a 0.01 % chance that a
model F value this large could occur due to noise. Values of
−0:00005  IP  SV þ 0:00154  T 2off −0:00004  IP2
Prob>F less than 0.0500 indicate that the model terms are
ð4Þ significant. In the present case, the model terms A, B, C, D,
AC, AD, CD, B2, C2, and D2 are significant for surface rough-
ness with their contribution of 61.22, 0.47, 3.44, 27.35, 0.48,
The coefficients of the process parameters in Eq. (4) have 1.87, 6.61, 0.56, 2.17, and 0.76 %, respectively. These
been computed by Design-Expert software after analysis of percentage contributions have been computed from Table 7
the data shown in Table 4. by dividing the “individual (model) term” sum of squares
by “model” sum of squares. Results clearly show that A
4.2 Analysis of variance for surface roughness and D are the most significant factors affecting the surface
roughness.
In this section, the experimental results depicted in Table 4 for The lack-of-fit F value of 2.07 implies that it is not signif-
surface roughness have been analyzed. On the basis of fit icant relative to pure error. There is 15.95 % chance that a

Table 7 Pooled ANOVA results


for surface roughness after Source Sum of Degree of Mean square F value Prob>F
backward elimination squares freedom

Model 19.78 11 1.80 83.46 <0.0001 Significant


A 12.11 1 12.11 561.86 <0.0001
B 0.093 1 0.093 4.33 0.0438
C 0.68 1 0.68 31.54 <0.0001
D 5.41 1 5.41 251.17 <0.0001
AC 0.096 1 0.096 4.44 0.0414
AD 0.37 1 0.37 17.26 0.0002
CD 0.13 1 0.13 5.86 0.0201
A2 0.079 1 0.079 3.66 0.0629
B2 0.11 1 0.11 4.94 0.0320
C2 0.43 1 0.43 20.17 <0.0001
D2 0.15 1 0.15 6.80 0.0127
Residual 0.86 40 0.022
Lack of fit 0.78 33 0.024 2.07 0.1595 Not significant
Pure error 0.080 7 0.011
Cor total 20.64 51
Standard deviation 0.15 R2 0.9583
Mean 2.62 Adj R2 0.9468
CV % 5.61 Pred R2 0.9002
PRESS 2.06 Adeq Precision 32.143
40 Int J Adv Manuf Technol (2015) 79:31–47

Fig. 2 a Normal probability plot of residuals for surface roughness. b Plot of predicted vs. actual response for surface roughness

lack-of-fit F value this large could occur due to noise. in terms of input process parameters can be expressed by the
Non-significant lack of fit is good because it allows the following Eq. (5):
model to fit the experimental data. The value of determi-
Surface roughness ¼ 1:36163 þ 1:72341  T on −0:05407  T off
nation coefficient (R2) is 0.9583. It depicts that the qua-
dratic model can explain up to 95.83 % variation in the þ 0:02089  IP−0:02322  SV þ 0:00456
surface roughness. The Pred R2 value of 0.9002 is in  T on  IP þ 0:01797  T on  SV þ 0:00008
reasonable agreement with the Adj R2 of 0.9468. The  IP  SV−0:73962  T 2on þ 0:00082  T 2off
lower value (5.61) of the coefficient of variation reveals
−0:00009  IP2 −0:00024  SV2 ð5Þ
improved precision and reliability of the performed exper-
iments. Adeq Precision measures the signal-to-noise ratio.
A value of 32.143 for Adeq Precision specifies an ade-
quate signal for the model as a ratio greater than 4 is The coefficients of the process parameters in Eq. (5) have
desirable. Hence, this quadratic model can be used to been computed by Design-Expert software after analysis of
navigate the design space and considered significant for the data shown in Table 4.
fitting and predicting the experimental results.
4.3 Validation of models

4.2.1 Regression equation for surface roughness In order to validate the cutting rate and surface roughness
models, seven additional experiments have been conducted
In terms of actual factors, the final second-order regression using different parameter settings within the range of selected
equation for the surface roughness (response characteristic) parameters. The set of input parameter combinations used for

Table 8 Results of validation experiments for cutting rate and surface roughness

Experiment Ton (μs) Toff (μs) IP (A) SV (V) WF (m/min) WT (g) Cutting rate (mm/min) Surface roughness (μm)
number
Experimental Predicted Prediction Experimental Predicted Prediction
value value error (%) value value error (%)

1 0.9 28 170 60 7 1200 1.16 1.20 −3.45 2.66 2.58 3.01


2 1.3 24 90 40 6 600 2.45 2.40 2.04 3.05 3.11 −1.97
3 1.0 22 180 20 3 500 2.99 3.07 −2.67 3.20 3.27 −2.19
4 0.5 36 150 50 5 1000 0.55 0.53 3.64 1.94 1.86 4.12
5 0.6 18 190 25 10 700 1.81 1.89 −4.42 2.51 2.54 −1.19
6 1.1 42 140 75 4 1000 0.78 0.76 2.56 2.56 2.60 −1.56
7 0.8 30 120 35 8 1400 1.39 1.36 2.16 2.81 2.74 2.49
Int J Adv Manuf Technol (2015) 79:31–47 41

this purpose are different from the one used for developing the
regression models. The results of validation experiments are
shown in Table 8. The prediction error (%) listed in Table 8 is
calculated by the following Eq. (6):
 
Experimental result−Predicted result
Prediction error ð%Þ ¼  100
Experimental result
ð6Þ

As the predicted results obtained from regression equations


are found to be in good agreement with the experimental
findings and also the prediction errors are less than ±5 %,
therefore, the developed models can be regarded as a reliable
representative of the experimental results.
Fig. 4 Response surface showing the interactive effect of pulse-on
time and pulse-off time on cutting rate at peak current=140.00 A,
spark gap voltage=48.00 V, wire feed rate=7.00 m/min, and wire
tension=1000.00 g
5 Results and discussion

In a WEDM process, cutting rate and surface roughness of the peak current (C) shows that it is less sensitive to cutting rate.
generated surface are important performance measures be- The reasons for these trends have been discussed while
cause of their crucial effect on industrial economy and surface explaining the interaction effects.
integrity. In this section, the effects of individual process pa-
rameters as well as their interactions on the performance mea- 5.1.1 Interactive effect of process parameters on cutting rate
sures have been discussed with the help of perturbation plot
and three-dimensional (3D) response curves. It is clear from Table 6 that the interactions which contribute
significantly to the model are those between the pulse-on time
5.1 Influence of process parameters on cutting rate and pulse-off time (AB), pulse-on time and peak current (AC),
pulse-on time and spark gap voltage (AD), pulse-off time and
The perturbation plot (Fig. 3a) shows the comparative effects spark gap voltage (BD), and peak current and spark gap volt-
of significant process parameters on the cutting rate of Inconel age (CD). The interaction plots corresponding to these are
718 cut by WEDM. By default, Design-Expert software sets shown in Figs. 4, 5, 6, 7, and 8, respectively.
the reference point at the midpoint (coded value 0) of all The response curves in Figs. 4, 5, and 6 depict the interac-
factors. A steep slope for pulse-on time (A), pulse-off time tive effects of pulse-on time with pulse-off time, peak current,
(B), and spark gap voltage (D) shows that the cutting rate is and spark gap voltage, respectively. Figure 4 shows the effect
highly sensitive to these factors while a relatively flat line for of pulse-on time and pulse-off time on cutting rate while

Fig. 3 Perturbation plots for a cutting rate and b surface roughness


42 Int J Adv Manuf Technol (2015) 79:31–47

Fig. 5 Response surface showing the interactive effect of pulse-on


time and peak current on cutting rate at pulse-off time=30.00 μs, Fig. 7 Response surface showing the interactive effect of pulse-off
spark gap voltage=48.00 V, wire feed rate=7.00 m/min, and wire time and spark gap voltage on cutting rate at pulse-on time=0.90 μs,
tension=1000.00 g peak current = 140.00 A, wire feed rate = 7.00 m/min, and wire
tension=1000.00 g

keeping all other input parameters at middle values. It illus-


trates that the cutting rate increases with an increase in pulse- The effect of peak current on cutting rate is more promi-
on time. The reason can be attributed to the fact that the nently visible at higher values of pulse-on time (Fig. 5). At
amount of material removed from the work specimen highly low values of pulse-on time, the peak current has less impact
depends on the discharge energy (the energy content of a on the cutting rate. The increase in peak current produces
single spark discharge), which is the product of pulse-on higher energy pulses at the discharge spot but the effect is
time and peak current. Figure 4 further indicates that the visible only at higher values of pulse-on time. This may be
cutting rate decreases with the increase in pulse-off time. due to the inherent characteristics of Inconel 718 such as pres-
The reason is that the higher the value of pulse-off time, ence of hard abrasive carbides in the microstructure, low ther-
the lesser is the number of discharges within a specific mal conductivity and specific heat, and high cutting tempera-
period of time which leads to reduced number of particles ture, which necessitate the application of peak current at high
dislodged near the surface of work material. It is impor- pulse-on time for melting and vaporizing the material in order
tant to note that the pulse-on time has a larger effect on to achieve higher cutting rate. The peak current of 170–180 A
the cutting rate at low pulse-off time than at high pulse- and pulse-on time of 1.10–1.20 μs seem to be suitable for
off time. The cutting rate is the maximum at 1.20 μs obtaining maximum cutting rate. Sharma et al. [24] also re-
pulse-on time and 20 μs pulse-off time. ported similar findings for WEDM of high-strength materials

Fig. 6 Response surface showing the interactive effect of pulse-on


time and spark gap voltage on cutting rate at pulse-off time= Fig. 8 Response surface showing the interactive effect of peak current and
30.00 μs, peak current=140.00 A, wire feed rate=7.00 m/min, and spark gap voltage on cutting rate at pulse-on time=0.90 μs, pulse-off time=
wire tension=1000.00 g 30.00 μs, wire feed rate=7.00 m/min, and wire tension=1000.00 g
Int J Adv Manuf Technol (2015) 79:31–47 43

such as HSLA where pulse-on time was the most prominent


factor for the responses. However, this justification may not be
applicable for other materials like steels where peak current
also plays a significant role apart from pulse-on time.
The response surface in Fig. 6 illustrates that the increase in
spark gap voltage leads to the decrease in cutting rate but the
decrease is steeper at higher value of pulse-on time. The spark
gap voltage determines the gap between the workpiece and the
wire electrode. The higher the spark gap voltage, the higher
will be the gap between wire and workpiece. This increases
the dielectric strength of the medium and as the open circuit
voltage of the machine is fixed, the discharge current during
machining decreases, resulting in less melting and evaporation
of the work material. As a result, the cutting rate is decreased.
Fig. 10 Response surface showing the interactive effect of peak current
The spark gap voltage in the range of 28–36 V is suitable and spark gap voltage on surface roughness at pulse-on time=0.90 μs,
for obtaining maximum cutting rate at higher values of pulse-off time=30.00 μs, wire feed rate=7.00 m/min, and wire tension=
pulse-on time (1.10–1.20 μs) in the present case. The 1000.00 g
interaction plot of spark gap voltage with pulse-off time
(Fig. 7) and peak current (Fig. 8) corroborates the infer- pulse-on time and peak current (AC), peak current and spark
ences already drawn from Figs. 4, 5, and 6. gap voltage (CD), and pulse-on time and spark gap voltage
(AD). The interaction plots corresponding to these are shown
5.2 Influence of process parameters on surface roughness in Figs. 9, 10, and 11, respectively.
It can be seen in Fig. 9 that the combination of low values
The perturbation plot for surface roughness is shown in of pulse-on time (0.60–0.70 μs) and peak current (100–110 A)
Fig. 3b. A steep slope for pulse-on time (A) and spark gap causes minimum surface roughness. It can be argued that
voltage (D) shows that the surface roughness is very sensitive when pulse-on time and peak current are high, the bigger
to these factors while relatively flat lines for pulse-off time (B) craters might be generated on the surface of the workpiece
and peak current (C) show comparatively less sensitivity to due to an increase in melting and evaporation rate.
surface roughness. Figure 10 shows the nature of interaction between peak
current and spark gap voltage while affecting surface
5.2.1 Interactive effect of process parameters on surface roughness. It is clear from Fig. 10 that low values of peak
roughness current (100–110 A) and higher values of spark gap voltage
(60–68 V) give rise to minimum surface roughness. It is con-
The ANOVA results (Table 7) reveal that the interactions sidered that at the higher spark gap voltage, the gap between
which contribute significantly to the model are those between the wire and workpiece will be higher. The discharge current

Fig. 9 Response surface showing the interactive effect of pulse-on time Fig. 11 Response surface showing the interactive effect of pulse-on time
and peak current on surface roughness at pulse-off time=30.00 μs, spark and spark gap voltage on surface roughness at pulse-off time=30.00 μs,
gap voltage=48.00 V, wire feed rate=7.00 m/min, and wire tension= peak current=140.00 A, wire feed rate=7.00 m/min, and wire tension=
1000.00 g 1000.00 g
44 Int J Adv Manuf Technol (2015) 79:31–47

Table 9 The constraints for input and response parameters responses and process parameters. The optimization process
Parameter Goal Lower limit Upper limit searches the optimum values of cutting rate and surface rough-
ness by maximizing cutting rate while minimizing surface
Pulse-on time (μs) In range 0.4 1.4 roughness taken simultaneously as well as individually. The
Pulse-off time (μs) In range 14 46 constraints for input and response parameters are given in
Peak Current (A) In range 70 210 Table 9.
Spark gap voltage (V) In range 16 80 The optimum operating conditions for input process pa-
Wire feed rate (m/min) In range 2 12 rameters and the corresponding responses have been comput-
Wire tension (g) In range 400 1600 ed with the help of software and are recorded in Table 10.
Cutting rate (mm/min) Maximize 0.4 3.75 While carrying out single response optimization, the other
Surface roughness (μm) Minimize 1.37 3.61 response has been ignored but for multi-response optimiza-
tion, both the responses have been considered and given equal
importance. The results given in Table 10 indicate remarkable
gets decreased which leads to lower melting of work material. improvement in cutting rate and surface roughness for Inconel
This causes decrease in surface roughness. Figure 11 indicates 718 cut by WEDM in comparison with the results available in
that the pulse-on time and spark gap voltage interact with literature [33, 34]. In order to validate the optimum results,
each other, and the results are similar to those observed from confirmatory experiments have been performed. It has been
Figs. 9 and 10. found that the values of response parameters obtained through
experimental procedures (Table 10) are in close agreement
with the predicted values.

6 Parametric optimization
7 Conclusions
Desirability function approach is one of the common ap-
proaches used for optimization of single and multiple quality On the basis of experimental study for machining of Inconel
characteristic problems [43]. In this approach, an estimated 718 with WEDM, the following conclusions have been
response is transformed into a scale-free value called de- drawn:
sirability. The desirable range varies from zero to one
(least to most desirable, respectively). The input process 1. The empirical models for cutting rate and surface rough-
parameter settings with maximum desirability are consid- ness of Inconel 718 machined by WEDM process have
ered to be the optimal parameter conditions. The approach been developed using response surface methodology and
has been used by some authors for manufacturing appli- are given below:
cations, and the detailed procedure has been duly docu-
mented and described [40, 44, 45]. Cutting rate ¼ 1:24642 þ 3:69173  T on −0:12058  T off
In the present work, the optimization module in the Design- þ 0:00878  IP−0:01040  SV−0:04958  T on
Expert software has been used to find out the combination of  T off þ 0:00760  T on  IP−0:02958  T on
input process parameters, i.e., pulse-on time, pulse-off time,
 SV þ 0:00066  T off  SV−0:00005  IP
peak current, spark gap voltage, wire feed rate, and wire ten-
sion, which satisfy the requirements imposed on each of the  SV þ 0:00154  T 2off −0:00004  IP2

Table 10 Optimal combinations of input process parameters for single and multi-response optimization and comparison with experimental results

Type of Objective Optimum parameters Response (predicted) Response Desirability


optimization (experimental)
Ton (μs) Toff (μs) IP (A) SV (V) WF WT (g)
(m/min)

Single response To maximize cutting 1.40 14.00 95.05 38.33 11.95 1599.11 3.78 mm/min 3.69 mm/min 1
rate
Single response To minimize surface 0.60 40.00 100.00 68.00 10.00 1400.00 1.31 μm 1.36 μm 1
roughness
Multi-response To maximize cutting 0.62 14.00 210.00 16.00 5.97 691.78 2.55 mm/min and 2.49 mm/min 0.553
rate and minimize 2.54 μm and 2.58 μm
surface roughness
simultaneously
Int J Adv Manuf Technol (2015) 79:31–47 45

Surface roughness ¼ 1:36163 þ 1:72341  T on −0:05407  T off with the increase in pulse-on time while it decreases with
þ 0:02089  IP−0:02322  SV þ 0:00456 the increase in spark gap voltage and pulse-off time. The
 T on  IP þ 0:01797  T on  SV þ 0:00008 highest cutting rate for Inconel 718 that has been achieved
is 3.78 mm/min.
 IP  SV−0:73962  T 2on þ 0:00082  T 2off
5. The pulse-on time and spark gap voltage are the major
−0:00009  IP2 −0:00024  SV2
contributors for affecting surface roughness with percent-
age contribution of 61.22 and 27.35 %, respectively. For
2. The models developed are considered reliable representa- surface roughness, the pulse-on time significantly inter-
tives of the experimental results with prediction errors less acts with peak current and spark gap voltage. The peak
than ±5 %. The standard uncertainty for cutting rate and current further interacts significantly with spark gap volt-
surface roughness is estimated to be 1.87 and 3.54 %, age. Surface roughness increases with the increase in
respectively. pulse-on time and decreases with the increase in spark
3. The pulse-on time is the most influencing factor for cut- gap voltage. The lowest surface roughness that has been
ting rate and surface roughness of Inconel 718. However, achieved is 1.31 μm.
the peak current does not play a significant role for 6. The multi-response optimization by providing equal im-
WEDM of Inconel 718. portance to both the responses achieved the highest cut-
4. Pulse-on time shows the highest percentage contribution ting rate as 2.55 mm/min and lowest surface roughness as
(38.64 %), followed by spark gap voltage (27.93 %) and 2.54 μm.
pulse-off time (20.78 %), to affect the cutting rate. The 7. Wire feed rate and wire tension have been found to be
pulse-on time significantly interacts with pulse-off time, insignificant factors for both the responses.
peak current, and spark gap voltage for cutting rate. The
spark voltage gap further interacts significantly with
pulse-off time and peak current. The cutting rate increases 8 Appendix

1:31 þ 1:27 þ 1:37 þ 1:42 þ 1:38 þ 1:28 þ 1:34 þ 1:41


Arithmetic mean f or cutting rate ¼ ¼ 1:347
8

Experimental standard deviation f or cutting rate


sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð1:31−1:347Þ2 þ ð1:27−1:347Þ2 þ ð1:37−1:347Þ2 þ ð1:42−1:347Þ2 þ ð1:38−1:347Þ2 þ ð1:28−1:347Þ2 þ ð1:34−1:347Þ2 þ ð1:41−1:347Þ2
¼ ¼ 0:053
8

0:053
Standard uncertainty f or cutting rate ¼ pffiffiffi ¼ 0:0187 ¼ 1:87%
8

2:89 þ 2:66 þ 2:98 þ 2:88 þ 2:82 þ 2:93 þ 2:78 þ 2:73


Arithmetic mean f or surface roughness ¼ ¼ 2:834
8

Experimental standard deviation f or surface roughness


sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð2:89−2:834Þ2 þ ð2:66−2:834Þ2 þ ð2:98−2:834Þ2 þ ð2:88−2:834Þ2 þ ð2:82−2:834Þ2 þ ð2:93−2:834Þ2 þ ð2:78−2:834Þ2 þ ð2:73−2:834Þ2
¼ ¼ 0:1
8

0:1
Standard uncertainty f or surface roughness ¼ pffiffiffi ¼ 0:0354 ¼ 3:54%
8
46 Int J Adv Manuf Technol (2015) 79:31–47

Conflict of interest The authors declare that they have no known actual 19. Mahapatra SS, Patnaik A (2007) Optimization of wire electrical dis-
or potential conflict of interest including any financial, personal, or other charge machining (WEDM) process parameters using Taguchi meth-
relationships with other people or organizations within 3 years of begin- od. Int J Adv Manuf Technol 34:911–925
ning the submitted work that could inappropriately influence, or be per- 20. Kondayya D, Krishna AG (2011) An integrated evolutionary
ceived to influence, this publication. approach for modelling and optimization of wire electrical
discharge machining. Proc Inst Mech Eng B J Eng Manuf
225:549–567
21. Kuruvila N, Ravindra HV (2011) Parametric influence and opti-
References mization of wire EDM of hot die steel. Int J Mach Sci Technol
15:47–75
22. Sadeghi M, Razavi H, Esmaeilzadeh A, Kolahan F (2011)
1. Wang ZY, Rajurkar KP, Fan J, Lei S, Shin YC, Petrescu G (2003) Optimization of cutting conditions in WEDM process using regres-
Hybrid machining of Inconel 718. Int J Mach Tool Manuf 43: sion modelling and Tabu-search algorithm. Proc Inst Mech Eng B J
1391–1396 Eng Manuf 225:1825–1834
2. Thakur DG, Ramamoorthy B, Vijayaraghavan L (2009) Study on the 23. Kumar K, Agarwal S (2012) Multi-objective parametric optimization
machinability characteristics of superalloy Inconel 718 during high on machining with wire electric discharge machining. Int J Adv
speed turning. Mater Des 30:1718–1725 Manuf Technol 62:617–633
3. Sharman ARC, Hughes JI, Ridgway K (2006) An analysis of the 24. Sharma N, Khanna R, Gupta RD, Sharma R (2013) Modeling and
residual stresses generated in Inconel 718TM when turning. J Mater multiresponse optimization on WEDM for HSLA by RSM. Int J Adv
Process Technol 173:359–367 Manuf Technol 67:2269–2281
4. Thomas A, El-Wahabi M, Cabrera JM, Prado JM (2006) High tem- 25. Kuriakose S, Mohan K, Shunmugam MS (2003) Data mining applied
perature deformation of Inconel 718. J Mater Process Technol 177: to wire-EDM process. J Mater Process Technol 142:182–189
469–472 26. Sarkar S, Mitra S, Bhattacharyya B (2005) Parametric analysis and
5. Shaw MC, Nakayama K (1967) Machining high strength materials. optimization of wire electrical discharge machining of γ-titanium
CIRP Ann 15:45–59 aluminide alloy. J Mater Process Technol 159:286–294
6. Kinoshita N, Fukui M, Gamo G (1982) Control of wire-EDM 27. Chen HC, Lin JC, Yang YK, Tsai CH (2010) Optimization of wire
preventing electrode from breaking. CIRP Ann Manuf Technol 31: electrical discharge machining for pure tungsten using a neural net-
111–114 work integrated simulated annealing approach. Expert Syst Appl 37:
7. Benedict GF (1987) Electrical discharge wire cutting (EDWC). In: 7147–7153
Nontraditional manufacturing processes. Marcel Dekker, New York, 28. Yu PH, Lee HK, Lin YX, Qin SJ, Yan BH, Huang FY (2011)
pp 231–246 Machining characteristics of polycrystalline silicon by wire electrical
8. Kalpakjian S, Schmid SR (2008) Material-removal processes: abra- discharge machining. Mater Manuf Process 26:1443–1450
sive, chemical, electrical, and high-energy beams. In: Manufacturing 29. Chiang KT, Chang FP (2006) Optimization of the WEDM process
processes for engineering materials, 5th edn. Pearson Education, pp of particle-reinforced material with multiple performance charac-
561–565 teristics using grey relational analysis. J Mater Process Technol
9. Pandey PC, Shan HS (2010) Thermal metal removal processes. 180:96–101
In: Modern machining processes. Tata McGraw-Hill Education, 30. Manna A, Bhattacharyya B (2006) Taguchi and Gauss elimination
pp 84–113 method: a dual response approach for parametric optimization of
10. Ho KH, Newman ST, Rahimifard S, Allen RD (2004) State of the art CNC wire cut EDM of PRAISiCMMC. Int J Adv Manuf Technol
in wire electrical discharge machining (WEDM). Int J Mach Tool 28:67–75
Manuf 44:1247–1259 31. Patil NG, Brahmankar PK (2010) Some studies into wire electro-
11. Scott D, Boyina S, Rajurkar KP (1991) Analysis and optimization of discharge machining of alumina particulate-reinforced aluminum
parameter combinations in wire electrical discharge machining. Int J matrix composites. Int J Adv Manuf Technol 48:537–555
Prod Res 29:2189–2207 32. Hewidy MS, El-Taweel TA, El-Safty MF (2005) Modeling the ma-
12. Williams RE, Rajurkar KP (1991) Study of wire electrical discharge chining parameters of wire electrical discharge machining of Inconel
machined surface characteristics. J Mater Process Technol 28: 601 using RSM. J Mater Process Technol 169:328–336
127–138 33. Ramakrishnan R, Karunamoorthy L (2008) Modeling and multi-
13. Tarng YS, Ma SC, Chung LK (1995) Determination of optimal cut- response optimization of Inconel 718 on machining of CNC
ting parameters in wire electrical discharge machining. Int J Mach WEDM process. J Mater Process Technol 207:343–349
Tool Manuf 35:1693–1701 34. Ramakrishnan R, Karunamoorthy L (2009) Performance studies of
14. Spedding TA, Wang ZQ (1997) Parametric optimization and surface wire electro discharge machining (WEDM) of Inconel 718. Int J
characterization of wire electrical discharge machining process. Mater Prod Technol 35:199–215
Precis Eng 20:5–15 35. Garg RK, Aggarwal V, Singh S (2014) Effect of wire materials on
15. Spedding TA, Wang ZQ (1997) Study on modeling of wire EDM cutting performance of WEDM for machining of Inconel superalloy.
process. J Mater Process Technol 69:18–28 Appl Mech Mater 624:124–128
16. Huang JT, Liao YS, Hsue WJ (1999) Determination of finish-cutting 36. Box GE, Wilson KB (1951) On the experimental attainment of opti-
operation number and machining-parameters setting in wire electrical mum conditions. J Roy Stat Soc Ser B 13:1–45
discharge machining. J Mater Process Technol 87:69–81 37. Montgomery DC (2009) Design and analysis of experiments, 7th
17. Tosun N, Cogun C, Tosun G (2004) A study on kerf and material edn. Wiley, New York
removal rate in wire electrical discharge machining based on Taguchi 38. Box GE, Hunter JS (1957) Multi-factor experimental designs for
method. J Mater Process Technol 152:316–322 exploring response surfaces. Ann Math Stat 28:195–241
18. Han F, Jiang J, Yu D (2007) Influence of machining parameters on 39. Anderson MJ, Whitcomb PJ (2005) RSM simplified: optimizing pro-
surface roughness in finish cut of WEDM. Int J Adv Manuf Technol cesses using response surface methods for design of experiments.
34:538–546 Productivity Press, New York
Int J Adv Manuf Technol (2015) 79:31–47 47

40. Bhushan RK (2012) Optimization of cutting parameters for minimiz- 43. Derringer G, Suich R (1980) Simultaneous optimization of several
ing power consumption and maximizing tool life during machining of response variables. J Qual Technol 12:214–219
Al alloy SiC particle composites. J Clean Prod 39:242–254 44. Kuppan P, Rajadurai A, Narayanan S (2008) Influence of EDM pro-
41. Kuehl RO (2000) Design of experiments: statistical principles of research cess parameters in deep hole drilling of Inconel 718. Int J Adv Manuf
design and analysis, 2nd edn. Duxbury/Thomson Learning, Pacific Grove Technol 38:74–84
42. Myers RH, Montgomery DC (2002) Response surface methodology: 45. Sachdeva A, Singh S, Sharma VS (2013) Investigating surface
process and product optimization using designed experiments, 2nd roughness of parts produced by SLS process. Int J Adv Manuf
edn. Wiley, New York Technol 64:1505–1516

You might also like