Available online at www.sciencedirect.

com

ScienceDirect
Materials Today: Proceedings 4 (2017) 2194–2202 www.materialstoday.com/proceedings

5th International Conference of Materials Processing and Characterization (ICMPC 2016)

NSGA-II Approach for Multi- Objective Optimization of Wire

Electrical Discharge Machining Process Parameter on Inconel 718
Anshuman Kumar*a, Himadri Majumdera, K.Vivekanandaa,K.P.Maityb
a
Mechanical Department, National Institute of Technology Rourkela, Rourkela, Odisha 769008, India
b
Professor, Mechanical Department, National Institute of Technology – Rourkela, Rourkela, Odisha 769008, India

Abstract

The present paper envisagesthe effect and optimization of machining parameters on the material removal rate (MRR) and surface
roughness (Ra). Application of NSGA-II used for the selection of optimal process parameters during machining of Inconel 718. The
machining parameters considered in the study are wire tension, wire speed, discharge current and Pulse on time. The experimental
design using 27 orthogonal array are conducted viaTaguchi design of experiment to predict the MRR and Ra properties using Inconel
718 super alloy as a workpiece and brass wire as tool. The level of importance of the machining parameters on the cutting Ra and
MRR is determined by applying ANOVA. Using the signal-to-noise (S/N) ratio optimum machining parameters combination has been
obtained. To validate a mathematical model between control factors and responses like MRR and Raare established by means of
nonlinear regression analysis. To achieve multi-response optimization, NSGA-II has been adopted. Optimal machining parameters
were obtained from the Pareto front graph.

©2017 Elsevier Ltd. All rights reserved.

Selection and peer-review under responsibility of Conference Committee Members of 5th International Conference of Materials
Processing and Characterization (ICMPC 2016).

Keywords:Wire electrical discharge machining, MRR,Ra,Taguchi, Non linear Regression ,NSGA-II

1. Introduction

In this technological world there has been continuing demand of small size products by non-traditional process for
accurate and cost-effective measurement of material properties. The technology of wire electrical discharge
machining (WEDM) has been significantly improving to achieve industrial need. WEDM is a thermo-electrical
process where material is eroded from the work material by the spark between workpiece and wire(Tool). The
workpiece and wire are separated by a thin film of dielectric fluid. Machining performance specifications of concern
include minimum feature size, tolerance, surface finish, and material removal rate (MRR) and applications of
advanced, which is very difficult-to machine materials. The WEDM is an important manufacturing process to meet

* Corresponding author. Tel.: +91-7809207120; fax: +91-661-2462033.

E-mail address: [email protected]

2214-7853©2017 Elsevier Ltd. All rights reserved.

Selection and peer-review under responsibility of Conference Committee Members of 5th International Conference of Materials Processing and
Characterization (ICMPC 2016).
 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202 2195

industrial demands. Wire electrical discharge machining (WEDM) technology has been widely used in production,
in manufacturing industries.

Tosun et al.[1]studied on the effect and optimization of machining parameters on the kerf and MRR in WEDM
operations. Experimental studies were done under varying pulse duration, voltage, wire speed and dielectric flushing
pressure. Form the Taguchi design of experiment method the settings of machining parameters were determined.
Mahapatra et al.[2]have developed relationships between various responses like material removal rate (MRR),
surface roughness (SF) and control factors and kerf by means of nonlinear regression analysis, resulting in a valid
numerical model and then apply the genetic algorithm (GA) to optimize the process . Scott et al.[3] studied WEDM
process for the solution and formulation of a multi-objective function problem to get better machining parameter.
The basic requirements for the model were surface roughness and best MRR. In their research, they predicted the
measures of performances as a function of a selection of machining parameters using factorial design model. Lin CL
et al.[4]studied the electrical discharge machining (EDM) process of GRA based on fuzzy-based Taguchi method
and an orthogonal array for optimizing the multi-response process without using the fuzzy logic analysis and S/N
ratio are used in an orthogonal array.The Taguchi method is good and powerful technique and has been used
somewhat successfully in most of the industrial purposes to optimize the machining process and in the design of
electrical and mechanical components Badkar et al.[5].Datta et al.[6] studied the relationships between various
control factors and responses like MRR, SF and kerf by means of the grey-Taguchi technique as multi-objective
optimizer in the field of wire EDM, resulting in a valid numerical model. The study determines that the WEDM
process parameters could be familiar to achieve better MRR, surface finish and kerf width simultaneously. Lok et
al.[7] studied the behaviour of machining in terms of MRR and surface finish by processing of two advanced
ceramics under different cutting conditions using WEDM.Shouvik et al.[8] analysed using Taguchi orthogonal array
for find out the wear behaviours of Al-7.5%Sic Metal Matrix Composite .chinnadurai T [9]et al analysed the
performance measure of wire cut EDM based on process parameters for AISI 4140. Two sets of wire materials made
up of zinc coated and uncoated wires are used for cutting the workpiece.

2. Experimentation

The present paper envisages the effect of WEDM parameters such as wire tension, wire speed, discharge current, and
pulse on time on Inconel 718.

2.1 Experimental set up

Material

Inconel 718 alloys, Table 1, nickel chromium super alloys characterized by low thermal conductivity that increases the
thermal effects during machining.
Table 1 Chemical composition of INCONEL 718
Inconel 718
Element Ni+Co Cr Fe Nb+Ta Mo Ti Al
Content (%) 50-55 17-21 Bal 4.45-5.5 2.8-3.3 0.65-1.15 0.2-0.8

Inconel 718 processes density of 8.19Kg/m3

2196 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202

Schematic of specimen

Fig. 1: Setup of WEDM

The schematic diagram of wire EDM setup is shown in fig.1. In this investigation, the experiments were conducted on a
CNC WEDM machine, the manufacturer is AGIE, SWITZERLAND. Thedimension of the work material is 60mm ×
30mm × 1mm. A workpiece of 1mm thickness and a 0.20mm brass wire were used in the experiment. During the
experiment 10mm length is cut continuously.

2.2 Design of Experiment

Taguchi design of experiment approach has been used for planning the experiment design. The results of 27 experiment
runs are conducted via Taguchi design of experiment table to predict the MRR and Ra properties using Inconel 718
super alloy as a workpiece and brass wire as a tool.

Measuring apparatus

Kerfwidthmeasured using an optical microscope. The roughness (Ra) of the surface after WEDM machined from
Talysurf (Talysurf 50, Taylor Hobson Ltd., UK). The surface texture produced and the finishing mechanism after the
WEDM process were observed and investigated using a scanning electron microscope (SEM).

Experiment Details

The kerf width’s value is the average of five measurements made from the workpiece with 2 mm increments along the cut
length. MRR is calculated by using the formula can be express as:[10]

MRR=Κ × t ×Cs × ρ (2)

Here K is the kerf width, t is the thickness, Cs cutting speed and ρ is the density of the workpiece material (8.19 g/cm3).
The cutting speed (Cs) was directly obtained from the computer monitor of WEDM machine. The process factors and
factors level as shown in Table 2 and parametric details for WEDM as shown in Table 3

Table 2 Process factors and factors levels

Code Process Factors Level Unit

Low Medium High
A Wire Tension 12 14 16 N
B Wire speed 105 210 305 mm/s
C Discharge Current 10 15 20 A
D Pulse on Time 14 20 30 µs
 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202 2197

Table 3 Parametric details for WEDM

Sl No A B C D Kerf MRR S/N Ratio of MRR Ra S/N Ratio of Ra
1 12 105 10 14 0.224 0.00607 -44.3289 2.41 -7.27470
2 12 105 15 22 0.216 0.00953 -40.4095 4.02 -12.2638
3 12 105 20 30 0.227 0.01588 -35.9783 5.33 -14.3722
4 12 205 10 22 0.248 0.00930 -40.6245 3.63 -11.0101
5 12 205 15 30 0.256 0.01269 -37.9189 4.26 -12.3967
6 12 205 20 14 0.227 0.01583 -36.0050 5.45 -15.0647
7 12 305 10 30 0.224 0.00761 -44.8413 3.99 -12.5522
8 12 305 15 14 0.224 0.01085 -39.2899 3.93 -11.9001
9 12 305 20 22 0.237 0.01411 -35.5036 4.74 -13.1384
10 14 305 10 14 0.220 0.01066 -39.4413 3.40 -10.4415
11 14 305 15 22 0.235 0.00856 -41.9562 4.52 -12.9112
12 14 305 20 30 0.227 0.01588 -35.9824 4.20 -12.8818
13 14 105 10 22 0.222 0.00927 -40.6481 2.42 -8.19380
14 14 105 15 30 0.228 0.01457 -36.7284 4.10 -12.0469
15 14 105 20 14 0.237 0.01678 -35.5036 5.15 -13.9326
16 14 205 10 30 0.232 0.01365 -37.2922 2.80 -8.61370
17 14 205 15 14 0.253 0.01043 -39.6367 4.44 -13.1269
18 14 205 20 22 0.248 0.01448 -36.7724 4.74 -13.3287
19 16 205 10 14 0.232 0.01365 -37.2917 2.82 -9.46070
20 16 205 15 22 0.228 0.01196 -38.4444 4.49 -13.0572
21 16 205 20 30 0.225 0.01543 -36.2273 5.66 -14.7528
22 16 305 10 22 0.221 0.00835 -41.5646 5.42 -14.3184
23 16 305 15 30 0.229 0.01540 -36.2470 3.73 -11.6135
24 16 305 20 14 0.237 0.01678 -35.5036 5.15 -14.1229
25 16 105 10 30 0.232 0.01365 -37.2917 2.84 -8.75510
26 16 105 15 14 0.223 0.01069 -39.4147 3.93 -11.6963
27 16 105 20 22 0.237 0.01393 -37.1191 4.71 -13.8773

3. NSGA II

The multi-objective optimization required in many applications proposes where two to more objective function are
required to optimize simultaneously. Non-dominated Sorting Genetic Algorithm-II (NSGA-II) is a multi-objective
evolutionary algorithm based on a fast non-dominated sorting principle [11]. The detailed steps of NSGA II
algorithm has been depicted through a flow chart as shown in fig.2.
Initialize population P of size of N

Generation =1

Calculation fitness value

Stop
Sorting based on crowded comparison
Operator

If Generation > Max.

Selection, Crossover and Mutation to create Generation

offspring population Q of size N

Chose Population P of size N

Combination Population R=P+Q, of size 2N based on crowded-comparison
Operator
Fig.2: Flow chart of NSGA-II algorithm.
2198 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202

3.1 Mathematical model of Nonlinear Regression

Nonlinear regression models are another vital and helpful part of regression models. Nonlinear regression is a
learning approach for a nonlinear model between set of independent variables and depend variables. Unlike
traditional linear regression, which is restricted to estimating linear models, nonlinear regression can estimate
models with arbitrary relationships between independent and dependent variables[12, 13]. The proposed
mathematical model of various responses isshown in the equation (3):

Yi = C × Aw × B x × D y × E z (3)
Where C is the constant, A is the Wire tension, B is the wire speed, D is the discharge current, E is the pulse on time
andw,x,y and z are estimated coefficients of regression model. Statistical simulation software estimates the
parameters in nonlinear models using the SYSTAT 7.0 software. By using the machining parameters and responses
from the experiments, the following exponential models for MRR and Ra are developed and shown in equation (4-5)

MRR = 0.000502 × A( 0.476833) × B ( −0.007657 ) × D ( 0.605737 ) × E ( 0.114859 ) (4)

Ra = 0.458545 × A( 0.075469 ) × B ( 0.07933) × C ( 0.592991) × D ( −0.002877 ) (5)

3.2 Design of experiment based on Taguchi method

By using WEDM AGIECUTPROGRESS-2 machine, a limited set of input parameters values’ has been chosen from
a limited set of possible value. The WEDM process consists of three operations, a roughing, a finishing and a
surface finishing operation. Usually, performance of various types of cutting operation is judged by different
measures. The rough cutting phase is observed in this present work considering two performance objectives like
MRR and Ra. Taguchi proposed to acquire the characteristic data by using orthogonal array and to analyse the
performance measure from the data to decide the optimal process parameters. In this work, it is planned to study the
behaviour of four control factors viz., A, B, C, and D. The experimental observations are further transformed into
signal to noise (S/N) ratio. Taguchi used the signal-to-noise (S/N) ratio as the quality characteristic of choice. [14]There
are different S/N ratio available depending on type of characteristics like, lower is better (LB), nominal is best (NB) and
higher is better (HB). The characteristic that “HB” represents better machining performance, like MRR and inversely the
characteristic that “LB” represents better machining performance like Ra. Therefore “LB” for Ra and “HB” for MRR
were carefully chosen for obtaining optimum machining performance features. Following analysis were carried out on
the basis of S/N ratio calculations which are shown in table 3.
1 n 1
LHB = ∑
n i =1 yMRR
2 (6)

1 n 2
LLB = ∑ ykerf
n i =1
(7)

The S/N ratio can be calculated as a logarithmic transformation of the loss function as shown below:

S/N ratio for MRR = −10log10 ( LHB ) (8)

S/N ratio for Ra = −10log10 ( LLB ) (9)
 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202 2199

Fig 3:Microscope image after WEDM (45X) at A=14, B=305, C=15, D=22,

4. Result and discussion

In order to understand statistical visualization of impact of various factor and there interactions, analysis of variance
(ANOVA) has been performed. Table 4 suggests that discharge current is most significant factor for the analysis of
MRR and Table 5 shows that interaction term wire speed and discharge current are most significant factors in the
analysis. The coefficient of determination (R2) for MRR and Ra are 93.6% and 98.1% respectively.

Table 4 ANOVA table for MRR of WEDM process

Source DF Seq SS Adj MS F P % Contribution

A- wire Tension 2 14.546 7.273 3.76 0.087 8.049806
B- Wire Speed 2 06.029 3.014 1.56 0.285 3.336469
C- Discharge Current 2 86.074 43.037 22.25 0.002 47.63365
D- pulse on time 2 11.767 5.884 3.04 0.122 6.511898
A×C 4 27.147 6.787 3.51 0.083 15.02324
B×C 4 14.647 3.662 1.89 0.231 8.1057
C×D 4 08.884 2.221 1.15 0.418 4.916436
Error 6 11.606 1.934
Total 26 180.70 100

Table 5 ANOVA table for Ra of WEDM process

Source DF Seq SS Adj MS F P % Contribution

A- wire Tension 2 02.266 1.1332 2.98 0.126 1.908741
B- Wire Speed 2 07.831 3.9157 10.31 0.011 6.596359
C- Discharge Current 2 68.130 34.0648 89.71 0.010 57.38858
D- pulse on time 2 01.616 0.8082 2.13 0.200 1.36122
A×C 4 04.596 1.1490 3.03 0.109 3.871392
B×C 4 24.189 6.0472 15.92 0.002 20.37535
C×D 4 07.264 1.8160 4.78 0.045 6.118753
Error 6 02.278 0.3797
Total 26 118.171 100

6% 2%2%
5% 3% 6% 7%
8% 8% A- wire Tension
A- wire Tension
B- Wire Speed
C- Discharge Current 20% B- Wire Speed
15% D- pulse on time
A×C 4% 58% C- Discharge
48% B×C Current
C×D 1%
Error D- pulse on time
7%

Fig. 4: Percentage Contribution of MRR Fig. 5: Percentage Contribution of Ra

2200 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202

Fig. 6: Main effect plot of MRR Fig. 7: Main effect plot of Ra

Percent contribution (Figs. 4-5) shows the relative combination of each factor. The image of micro-slot as shown in Fig
3 after WEDM (45X) results of the verification experiments. The main effect of machining parameters on MRR is
shown in (Fig.6). The optimum machining performance for the MRR was found at 12N (Level 1), 305 mm/s (Level
3), 10A (Level 1) and 22 µs (level 2) settings. The main effects plot of machining parameters on the Ra shown in
Fig. 7. From S/N ratio analysis, the optimal machining performance for the Ra was obtained at 14N (Level 2),
105mm/s (Level 1), 10A (Level 1) and 14µs (level 1) settings.
To attain best fitness values for the objective function. NSGA II has been run in MATLAB® 12.0. The initial
parameter setting for this algorithm is given in table 6
Table 6 Parameter setting of NSGA II
Population size = 200
Crossover=0.9
Migration interval=20
Pareto fraction= 0.35

The obtained objective fitness values and their decision variables of the non-dominated solution were enlisted in the
Table 7 and pareto-front is shown in Fig. 8 In the set of non-dominated solutions none of the single solution is
superior to the other, so it depends on the decisions of the manufacturer which set of data is to be selected.Both
objectives are individually treated and predicted by using trained NSGA model

Fig.8: Pareto graph between MRR versus Ra

 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202 2201

Table 7 Results of NSGA II (only first 35 solutions were taken)

Sl No A B C D MRR Ra
1 12.00 105.00 10.00 14.37 0.008680 3.110
2 15.93 131.44 16.42 27.90 0.014461 4.333
3 15.20 123.45 10.79 23.35 0.010747 3.351
4 15.90 134.27 13.98 28.31 0.013121 3.944
5 15.71 136.69 12.70 29.32 0.012361 3.728
6 15.86 134.60 18.62 27.52 0.015540 4.675
7 15.98 132.75 18.44 28.04 0.015539 4.645
8 15.77 132.45 18.82 26.85 0.015559 4.698
9 15.92 132.68 17.97 27.86 0.015261 4.574
10 15.33 124.43 11.23 23.70 0.011078 3.437
11 15.13 122.93 11.41 23.13 0.011082 3.462
12 15.92 141.85 19.75 27.34 0.016116 4.863
13 15.96 145.52 19.93 27.09 0.016207 4.901
14 15.86 135.38 13.39 28.72 0.012789 3.846
15 15.97 132.49 19.01 27.85 0.015815 4.730
16 15.66 135.95 12.18 29.46 0.012039 3.633
17 15.22 123.65 10.93 23.47 0.010846 3.378
18 15.98 132.32 19.40 27.87 0.016016 4.787
19 15.96 145.08 19.80 27.12 0.016146 4.881
20 15.91 132.15 15.87 28.05 0.014164 4.248
21 15.21 123.44 10.81 23.34 0.010762 3.355
22 15.58 134.01 12.94 28.46 0.012415 3.762
23 15.68 135.62 12.26 29.44 0.012094 3.647
24 15.98 132.46 19.38 27.85 0.016002 4.784
25 12.74 109.30 10.30 16.35 0.009228 3.189
26 15.30 133.07 11.29 28.14 0.011313 3.462
27 12.49 107.85 10.12 15.76 0.009008 3.148
28 15.62 136.24 11.76 29.69 0.011781 3.558
29 15.86 132.15 15.65 27.97 0.014018 4.212
30 15.54 134.18 12.41 28.71 0.012101 3.670
31 15.60 136.11 11.59 29.68 0.011671 3.528
32 15.76 136.15 13.26 29.11 0.012693 3.824
33 15.87 136.42 14.08 28.55 0.013179 3.965
34 15.94 133.07 16.86 28.13 0.01471 4.406
35 15.80 132.53 14.67 27.73 0.013446 4.054

5. Conclusions

This work proposes an effective process parameter optimization approach of Taguchi method couple with multi
objective Non-dominated Sorting Genetic Algorithm-II (NSGA-II) of WEDM process on machinability of Inconel
718 alloy. The proposed approach can effectively assist decision maker in determining the optimal process
parameter settings for the WEDM process under multi-response consideration. The solutions obtained are true
representation of the given range of the input parameters and are not biased. On the basis of experimental data and
observation, the results obtained in the illustrative example are summarized as follows:
1. From the ANOVA table, the level of importance of the machining parameters on responses has been
analysed. Based on this, the highly effective parameters on both the kerf and the MRR were found as
voltage and Ton variable, whereas current was less effective factors among all.
2. On the basis of Taguchi approach, the optimum condition has obtained as wire tension 16N, wire speed
205mm/s and discharge current 20A and pulse on time 30 µs for MRR andwire tension 16N, wire speed
305mm/s, discharge current 20A and pulse on time 22 µs for Rarespectively.
As the WEDM in Inconel 718 workpiecewith 0.20mmbrass wire electrode is a costlier and time consuming process,
so there is requirement of optimization of process parameters that will benefit in growth of production rate by
minimizing the machining time and cost. The NSGA II has been applied to obtain a set of optimal combination of
parameters from the Pareto-optimal solution set to enhance the machining conditions.
2202 Anshuman Kumar et al. / Materials Today: Proceedings 4 (2017) 2194–2202

References

[1] N. Tosun, C. Cogun,A. Inan, Machining science and technology.7 (2003) 209-219.
[2] S. Mahapatra,A. Patnaik, The International Journal of Advanced Manufacturing Technology. 34 (2007) 911-925.
[3] D. Scott, S. Boyina,K. Rajurkar, The International Journal of Production Research. 29 (1991) 2189-2207.
[4] C. Lin, Materials and manufacturing processes. 19 (2004) 209-220.
[5] D. S. Badkar, K. S. Pandey,G. Buvanashekaran, The International Journal of Advanced Manufacturing Technology. 52 (2011) 1067-
1077.
[6] S. Datta,S. Mahapatra, International Journal of Engineering, Science and Technology. 2 (2010) 162-183.
[7] Y. Lok,T. Lee, Journal of materials processing technology. 63 (1997) 839-843.
[8] S. Ghosh, P. Sahoo,G. Sutradhar, Advanced Materials Manufacturing & Characterization. 4 (2014) 93-99.
[9] T. Chinnadurai,S. A. Vendan, Materials Today: Proceedings. 2 (2015) 1067-1073.
[10] R. T. Yang, C. J. Tzeng, Y. K. Yang,M. H. Hsieh, The International Journal of Advanced Manufacturing Technology. 60 (2012) 135-
147.
[11] K. Deb, A. Pratap, S. Agarwal,T. Meyarivan, IEEE Transactions on. 6 (2002) 182-197.
[12] F. Cus,J. Balic, Robotics and Computer-Integrated Manufacturing. 19 (2003) 113-121.
[13] F. Cus,U. Zuperl, Journal of Materials Processing Technology. 173 (2006) 281-290.
[14] M. S. Phadke, Springer Quality engineering using design of experiments (1989) 31-50.