You are on page 1of 11

Alexandria Engineering Journal (2017) xxx, xxx–xxx

H O S T E D BY
Alexandria University

Alexandria Engineering Journal


www.elsevier.com/locate/aej
www.sciencedirect.com

ORIGINAL ARTICLE

Multi-objective parametric optimization of nano


powder mixed electrical discharge machining of
AlSiCp using response surface methodology and
particle swarm optimization
Shalini Mohanty a, Ankan Mishra b, B.K. Nanda a, B.C. Routara a,*

a
School of Mechanical Engineering, KIIT University, Bhubaneswar, India
b
Indian Institute of Technology, Guwahati, India

Received 6 August 2016; revised 10 February 2017; accepted 13 February 2017

KEYWORDS Abstract Powder mixed electrical discharge machining (PMEDM) has gained popularity in the
Powder mixed EDM; current era owing to its benefits of providing better material removal rate (MRR), less electrode
Nano particle; wear rate (EWR) and improvement in surface finish. The use of powders enhances the machining
Particle swarm optimization; characteristics of the EDM processes. Low voltage current (LVC), high voltage current (HVC),
Response surface pulse-on time (Ton), pulse-off time (Toff) and flushing pressure (FP) are the input variables on which
methodology certain machining parameters such as material removal rate (MRR), surface roughness (Ra) and
tool wear rate (TWR) are analysed. A copper electrode of 99.98% purity with a diameter of
12 mm was used to cut AlSiCp12% metal matrix composite (MMC) in EDM. Box Behnken design
was used for planning the experimental run. The parameters were optimized using desirability
approach as multi-objective optimization technique for predicting the significance of the parame-
ters. In addition to all this, particle swarm optimization (PSO) was implemented for predicting
the results and hence error analysis was done for the set of experiments. Moreover, a confirmatory
test was carried out with the parametric settings obtained from PSO and hence the error percentage
was determined. Validation tests showed close relationship of predicted and experimental results.
Ó 2017 Faculty of Engineering, Alexandria University. Production and hosting by Elsevier B.V. This is an
open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction application in modern industries such as aerospace, aircraft,


nuclear, and bio-medical. This process works on the principle
One of the most extensively used non-conventional machining of controlled erosion of material from the workpiece through a
processes is electrical discharge machine (EDM). It finds chain of electrical sparks generated between the workpiece and
the electrode. The excess thermal energy liberated from the
* Corresponding author. sparks leads to severe heating conditions resulting in the melt-
E-mail address: bcroutray@gmail.com (B.C. Routara). ing and vaporization of the material. Every now and then the
Peer review under responsibility of Faculty of Engineering, Alexandria problem of poor surface finishes and low machining efficiency
University. arises, which further restricts its application. Thus, the
http://dx.doi.org/10.1016/j.aej.2017.02.006
1110-0168 Ó 2017 Faculty of Engineering, Alexandria University. Production and hosting by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
2 S. Mohanty et al.

introduction of powders into the dielectric fluid has substan- entry clearance, machining duration and number of short-
tially diffused this problem, and hence named as powder mixed circuits generated. Singh et al. [11] also used the same GRA
EDM (PMEDM). The powders thus introduced help in technique on EDM of Al10%SiCp composite for multi-
increasing the spark gap between the electrode and the work- optimization for multiple responses. The experimental data
piece, and reducing the insulating strength of the dielectric prove that there is visible improvement in EDM
medium [1]. This improves the MRR, makes the process stable characteristics.
and reduces the surface roughness. The electrical density is Muthuramalingam et al. [12] used Taguchi-grey relational
reduced on the machining spot with the addition of powders. based multi response optimization parameters to maximize
Since the addition of powders makes the process a bit complex, MRR and minimize surface roughness. The study shows that
hence RSM is used for modelling and analysing the experimen- peak current is the most significant factor. In another study,
tal data thereby studying the influence of parameters on Ramamurthy et al. [13] also used Taguchi-grey relational
responses. Moreover, PSO is used to obtain optimal parame- approach to obtain optimum input parameters using wire
ters for carrying out the confirmation tests. EDM during machining of Titanium alloy (Ti-6Al-4V). They
concluded that pulse off time was the most significant process
2. Literature review parameter due to the significant nature seen during deioniza-
tion of spark plasma.
It is evident from the literature that a very little work is
The improvement of process parameters has been carried out done in the field of nano powder mixed dielectric in EDM.
by many researchers pertaining to better surface finish and bet- To this aspect the present work has been carried out keeping
ter MRR. Yet, best fit machining parameter value selection is a the literature in view. The particle swarm optimization tech-
challenge till today. For obtaining proper parametric results, nique (PSO) has been used in addition to response surface
multi-objective optimization was adopted. To this regard Pad- methodology (RSM) to obtain more adequate results. Hence,
hee et al. [2] carried out experiments on PMEDM adopting the gap between the literature and the current research is some-
non-dominated sorted genetic algorithm (NSGA-II) for opti- what diminished.
mizing the responses. Shabgard and Khosrozadeh [3] used car-
bon nano tubes (CNTs) in dielectric to increase the efficiency
3. Methods and materials used
of machining Ti-6Al-4V alloy. They found that the surface fin-
ish was significantly high and the micro-cracks were reduced
3.1. Powder mixed electrical discharge machining (PMEDM)
with the use of powders, when seen under scanning electron
microscope (SEM). Kolli and Kumar [4] used boron carbide
powders in dielectric for machining of titanium alloys using The machining of materials in PMEDM is quite different from
EDM. They used various concentrations of powder into the that of the conventional one. A suitable powder, Al2O3 is
dielectric and subsequently, and they found that with the addi- selected to mix with the dielectric fluid (EDM oil) in a dedi-
tion of 15 g of boron carbide in one litre of dielectric provides cated set-up designed with a pump attached as shown in
the lowest surface roughness value. Fig. 1. The pump ensures that the particles do not settle in
Sengottuvel et al. [5] carried out a comparative study of per- the tank and continuous stirring is done. The powders are
formances of electrode by desirability approach and ANOVA mixed by means of a magnetic stirrer until the powders are
using Fuzzy model, in EDM and found that current was the completely mixed in the dielectric fluid. With the application
most influencing factor, followed by pulse on time and pulse
off time. Chiang et al. [6] presented a model and analysed
the rapidly resolidified layer of spheroidal graphite (SG) cast
iron in the EDM process using RSM approach. The results
of ANOVA indicated the quantity and area fraction of gra-
phite particle are the most influential factors on rapidly reso-
lidified layer thickness and ridge density in EDM process.
Layer thickness decreases with increase in diameter of graphite
particle. Dewangan et al. [7] optimized the parameters of EDM
using grey-fuzzy logic based hybrid optimization technique.
Results indicated that pulse-on-time is the most significant
parameter followed by discharge current. Sivasankar et al.
[8] studied the performance of tool materials (graphite, nio-
bium, titanium, tantalum, tungsten) upon ZrB2-SiC MMC
and optimized the performance parameters using particle
swarm optimization (PSO) algorithm. Agrawal et al. [9] used
artificial neural network (ANN) technique to develop a reliable
model to predict tool wear rate with least predictive error.
They used graphite powder in the dielectric to reduce the tool
wear rate during machining MMCs. Peak current was found to
be the most significant factor and then powder concentration.
To this aspect, Jung and Kwon [10] used grey relational
approach (GRA) and Taguchi, a combination of multi-
optimization to study the tool wear exit clearance, tool wear Figure 1 PMEDM set-up.

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
Multi-objective parametric optimization nano powder mixed electrical discharge machining 3

where Wf denotes the weight of work piece before


machining (g), Wi gives the weight of workpiece post
machining (g), q is the density of the material (g/mm3)
and t is the time of machining (min).
ii. Surface roughness (Ra): To determine the surface rough-
ness (Ra) of the metal matrix composite that has been
machined under EDM, Taylor Hobson surface rough-
ness tester is used.
iii. Electrode wear rate (EWR): The electrode wear rate
(EWR) evaluation is carried out by weight loss method.
The following formula is used to calculate electrode
wear rate:
 
Tf  Ti
EWR ¼  q mm3 =min ð2Þ
t
where Tf is the weight of electrode before machining (g),
Ti is the weight of electrode post machining (g), q is the
density of the electrode and t is the time of machining
(min).

Figure 2 Schematic of PMEDM. 3.3. Experimental data collection

Experiments were carried out with ECOWIN CNC EDM


(ECOWIN PS 50ZNC) machine manufactured in Taiwan. A
of high potential difference between the electrode and the constant gap is maintained with the help of the servo head with
workpiece, electrical sparks are created which generates ther- the electrode having positive polarity. Al2O3 powders are sus-
mal energy. The electrical density of the machining spot is pended in commercial grade EDM oil (paraffin oil) by means
decreased due to the presence of powders. The thermal energy of a magnetic stirrer and used as the dielectric fluid. The aver-
thus produced makes the local temperature high leading to the age size of the particle varies from 20 to 30 lm. Al-SiCp12%
melting and vaporization of workpiece material. Short pulse MMC is chosen as the workpiece material. Copper electrode
electrical power is supplied as such excessive heating is with 99.98% purity and diameter of 12 mm is used for machin-
avoided. With the addition of powders, significant increase ing the MMC. A Box–Behnken design of experiment [14] was
in gap between the tool and the electrode is seen. A PMEDM used with 5 levels and 5 variables with 46 runs. The influence of
schematic is shown in Fig. 2. machining parameters upon TWR, MRR and Ra is studied.
Response surface methodology (RSM) was performed to
3.2. Input and output parameter selection establish the mathematical relationship between the dependent
variables and the independent variables. Since five numbers of
In the past research, there are many electrical, non-electrical, process parameters and two numbers of responses are taken
powder and electrode parameters that solely affect the into account, thus a large number of experiments are carried
PMEDM performances. The most important machining per- out. With the use of RSM, the number of experiments can
formances are material removal rate (MRR) and surface be reduced to some extent and hence, their interactions can
roughness (Ra). In the present study, the following input be studied effectively.
parameters range is considered: By the help of regression analysis and design of experiments
(DOE), a response for independent input parameters can be
i. Low voltage current (LVC): 6–18 A retrieved. In RSM, the independent process parameters can
ii. High voltage current (HVC): 1–2 A be represented quantitatively as follows:
iii. Pulse on time (Ton): 30–90 ls
iv. Pulse off time (Toff): 10–30 ls y ¼ fðx1 ; x2 ; x3 ; . . . xn Þ  e ð3Þ
v. Flushing pressure (FP): 0.2–0.6 kg/m2
where y is the desired response, f is response function, x1, x2,
x3, . . . xn are the independent input variables, and e is the fitting
The following output parameters are taken into
error.
consideration:
The expected responses are plotted and the requisite sur-
face obtained so is known as response surface. The suitability
i. Material removal rate (MRR): The material removal
of RSM is determined with the approximation of f. It is
rate (MRR) is calculated by weight loss ratio method
obtained using second order polynomial regression model,
i.e., difference of pre-machining weight of workpiece
also called quadratic model. The quadratic model of f is as
and post-machining weight to that of the machining
follows:
time. MRR is calculated by the following formula:
  X
n X
n X
n
Wf  Wi f ¼ a0 þ ai xi þ aii x2i þ aij xi xj þ e ð4Þ
MRR ¼  q mm3 =min ð1Þ
t i¼1 1¼1 i<j

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
4 S. Mohanty et al.

Table 1 Factors and their levels.


Factor symbol Parameter Levels
Low (1) Medium (0) High (+1)
A Low voltage current (A) 6 12 18
B High voltage current (A) 1 1.5 2
C Pulse on time (ls) 30 60 90
D Pulse off time (ls) 10 15 30
E Flushing pressure (kg/m2) 0.2 0.4 0.6

Table 2 Design of experiment for process parameters and result matrix for PMEDM.
Expt. No. LVC HVC Ton Toff FP MRR EWR Ra
1 12.00 1.50 90.00 30.00 0.40 0.503 1.031 4.256
2 12.00 2.00 30.00 20.00 0.40 0.901 0.963 4.263
3 12.00 1.00 60.00 20.00 0.20 0.798 0.961 4.302
4 12.00 1.50 30.00 20.00 0.60 0.858 0.982 4.251
5 12.00 1.50 60.00 20.00 0.40 0.813 1.063 4.209
6 12.00 1.50 30.00 10.00 0.40 0.742 1.082 4.251
7 6.00 2.00 60.00 20.00 0.40 0.787 1.071 4.201
8 6.00 1.50 60.00 10.00 0.40 0.818 0.965 4.242
9 18.00 1.50 60.00 20.00 0.60 0.668 1.058 4.298
10 6.00 1.50 60.00 20.00 0.60 0.744 1.060 4.298
11 12.00 2.00 60.00 30.00 0.40 0.830 0.960 4.261
12 12.00 1.50 60.00 20.00 0.40 0.782 1.083 4.264
13 12.00 1.00 90.00 20.00 0.40 0.825 0.987 4.265
14 6.00 1.50 60.00 30.00 0.40 0.803 1.053 4.295
15 6.00 1.50 30.00 20.00 0.40 0.830 0.973 4.260
16 12.00 1.50 30.00 20.00 0.20 0.881 0.985 4.249
17 12.00 2.00 60.00 10.00 0.40 0.899 0.994 4.253
18 18.00 1.50 30.00 20.00 0.40 0.880 1.079 4.204
19 18.00 2.00 60.00 20.00 0.40 0.693 1.085 4.300
20 12.00 1.50 60.00 10.00 0.60 0.789 0.983 4.260
21 12.00 1.00 60.00 10.00 0.40 0.768 1.097 4.229
22 12.00 1.50 60.00 20.00 0.40 0.559 0.987 4.260
23 18.00 1.00 60.00 20.00 0.40 0.788 0.977 4.265
24 12.00 1.00 60.00 30.00 0.40 0.780 0.979 4.253
25 12.00 1.50 60.00 30.00 0.20 0.558 0.996 4.246
26 12.00 1.50 90.00 20.00 0.20 0.798 1.067 4.212
27 18.00 1.50 60.00 10.00 0.40 0.767 1.097 4.257
28 12.00 1.50 60.00 30.00 0.60 0.789 0.973 4.260
29 12.00 1.50 60.00 20.00 0.40 0.762 1.050 4.203
30 12.00 1.00 60.00 20.00 0.60 0.843 0.949 4.298
31 12.00 1.50 60.00 10.00 0.20 0.535 1.035 4.257
32 6.00 1.50 90.00 20.00 0.40 0.796 0.972 4.289
33 12.00 1.50 60.00 20.00 0.40 0.728 1.046 4.265
34 12.00 2.00 60.00 20.00 0.20 0.776 1.060 4.210
35 12.00 2.00 90.00 20.00 0.40 0.534 0.972 4.210
36 6.00 1.00 60.00 20.00 0.40 0.795 1.020 4.233
37 18.00 1.50 90.00 20.00 0.40 0.789 0.953 4.230
38 18.00 1.50 60.00 20.00 0.20 0.789 0.953 4.240
39 12.00 2.00 60.00 20.00 0.60 0.569 0.985 4.259
40 12.00 1.50 30.00 30.00 0.40 0.572 0.993 4.256
41 12.00 1.50 90.00 10.00 0.40 0.789 0.953 4.260
42 12.00 1.50 90.00 20.00 0.60 0.796 0.981 4.246
43 18.00 1.50 60.00 30.00 0.40 0.536 1.081 4.299
44 6.00 1.50 60.00 20.00 0.20 0.786 1.039 4.272
45 12.00 1.00 30.00 20.00 0.40 0.833 1.078 4.265
46 12.00 1.50 60.00 20.00 0.40 0.679 0.973 4.260

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
Multi-objective parametric optimization nano powder mixed electrical discharge machining 5

Table 3 ANOVA table for material removal rate (MRR).


Source Adj SS DF Adj MS F-value P-value
Model 0.50000 20 0.025000 35.310 <0.0001
A-LVC 0.003512 1 0.003512 4.990 0.0347
B-HVC 0.400000 1 0.400000 562.08 <0.0001
C-Ton 0.005619 1 0.005619 7.990 0.0091
D-Toff 0.002803 1 0.002803 3.990 0.0569
E-FP 0.011000 1 0.011000 15.350 0.0006
AB 0.000366 1 0.000366 0.520 0.4775
AC 0.0002126 1 0.0002126 0.300 0.5873
AD 0.006139 1 0.006139 8.730 0.0067
AE 0.003359 1 0.003359 4.780 0.0384
BC 0.0000104 1 0.0000104 0.015 0.9044
BD 0.003947 1 0.003947 5.610 0.0259
BE 0.0000141 1 0.0000141 0.020 0.8887
CD 0.0000456 1 0.0000456 0.065 0.8012
CE 0.00000207 1 0.00000207 0.0029 0.9571
DE 0.001279 1 0.001279 1.820 0.1896
A^2 0.0006768 1 0.0006768 0.960 0.3360
B^2 0.045000 1 0.045000 63.320 <0.0001
C^2 0.002406 1 0.002406 3.420 0.0762
D^2 0.001641 1 0.001641 2.330 0.1392
E^2 0.0002446 1 0.0002446 0.350 0.5606
Residual 0.018000 25 0.0007032
Lack of fit 0.007515 20 0.0003758 0.190 0.9972
Pure error 0.01000 5 0.002013
Cor. total 0.51000 45

Table 4 ANOVA table for electrode wear rate (EWR).


Source Adj SS DF Adj MS F-value P-value
Model 0.097000 20 0.004835 13.51 <0.0001
A-LVC 0.003600 1 0.003600 10.06 0.0040
B-HVC 0.0003303 1 0.0003303 0.920 0.3460
C-Ton 0.005085 1 0.005085 14.20 0.0009
D-Toff 0.002398 1 0.002398 6.700 0.0158
E-FP 0.011000 1 0.011000 31.21 <0.0001
AB 0.014000 1 0.014000 40.08 <0.0001
AC 0.002208 1 0.002208 6.170 0.0201
AD 0.002796 1 0.002796 7.810 0.0098
AE 0.0000567 1 0.0000567 0.160 0.6939
BC 0.0009535 1 0.0009535 2.660 0.1152
BD 0.0000173 1 0.0000173 0.048 0.8280
BE 0.0008681 1 0.0008681 2.420 0.1320
CD 0.0002818 1 0.0002818 0.790 0.3834
CE 0.007288 1 0.007288 20.360 0.0001
DE 0.003787 1 0.003787 10.580 0.0033
A^2 0.027000 1 0.027000 75.380 <0.0001
B^2 0.0003717 1 0.0003717 1.040 0.3180
C^2 0.007720 1 0.007720 21.570 <0.0001
D^2 0.0004422 1 0.0004422 1.240 0.2770
E^2 0.019000 1 0.019000 52.240 <0.0001
Residual 0.008950 25 0.000358
Lack of fit 0.008067 20 0.0004033 2.280 0.1833
Pure error 0.000883 5 0.0001767
Cor. total 0.11000 45

where ai represents the linear effect of xi, aii denotes quadratic the entire factor space. Each factor with their range is illus-
effect of xi and aij denotes the linear-by-linear interaction trated in Table 1.
between xi and xj. Thus, the response surface f consists of lin- The current PMEDM study is carried out by standard Box-
ear, squared and cross-product terms. This quadratic model Behnken design with total forty-six number of experiments
allows to locate the region of optimality besides investigating with factors at their designated levels as given in Table 2.

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
6 S. Mohanty et al.

4. Results and discussions

Design expert statistical software was used for experimental


data analysis. For all the three responses, a quadratic model
is obtained for the regression models for each of the responses
separately. The insignificant terms are eliminated and the final
equation for MRR in coded units is as follows:
MRR ¼ 7:7  101  1:5  102 A  1:6  101 B þ 1:9
 102 C  1:3  102 D  2:6  102 E þ 9:56
 103 AB þ 7:29  103 AC  3:9  102 AD
 2:9  102 AE  1:61  103 BC þ 3:1
 102 BD  1:88  103 BE þ 3:38  103 CD
þ 7:20  104 CE  1:8  102 DE  8:81
 103 A2  7:1  102 B2 þ 1:7  102 C2 þ 1:4
 102 D2  5:29  103 E2 ð5Þ
The ANOVA for material removal rate (MRR), electrode Figure 3 Normal probability plot of residuals for MRR.
wear rate (EWR) and surface roughness (Ra) is shown in Tables
3–5 respectively. The significance check is found from the F-
values. The P-value denotes the probability of F-value is greater
coefficient of determination (R2) and Adj. R2 values for EWR
than that of the calculated value because of noise. The signifi-
was found to be 0.9153 and 0.8475 respectively and that for
cance of a term is determined if the P-value is less than 0.05
Ra, it was found to be 0.9405 and 0.8923 respectively.
which means 95% confidence level. The lack of fit is deter-
The normal probability plot for MRR shown in Fig. 3 indi-
mined, which means P-value is greater than 0.05. It indicates
cates that the residuals follow normal distribution, which means
that the insignificant term is left away from the model and
that the errors are normally distributed along the straight line.
hence the model that is developed fits very well. It is clear from
Again the 3D surface plot shown in Fig. 4a indicates that when
the table that the developed model is a significant one. The coef-
the high voltage current (HVC) level is kept constant at low
ficient of determination values (R2) and Adj. R2 for MRR was
level, the response MRR varies along low voltage current
found to be 0.9658 and 0.9358 respectively. This denotes that
(LVC). Similarly from Fig. 4b, it is evident that by keeping
the relationship between the factors and the response (MRR)
Ton at high level, we find MRR increasing with flushing pressure.
is best explained by the regression model. Similarly, the

Table 5 ANOVA table for surface roughness (Ra).


Source Adj SS DF Adj MS F-value P-value
Model 0.032000 20 0.001607 19.75 <0.0001
A-LVC 0.031000 1 0.031000 377.82 <0.0001
B-HVC 0.00000375 1 0.00000375 0.046 0.8318
C-Ton 0.00001223 1 0.00001223 0.150 0.7015
D-Toff 0.0001341 1 0.0001341 1.650 0.2111
E-FP 0.0002973 1 0.0002973 3.650 0.0675
AB 0.000006225 1 0.000006225 0.076 0.7844
AC 0.00003502 1 0.00003502 0.430 0.5178
AD 0.00003392 1 0.00003392 0.420 0.5244
AE 0.0001238 1 0.0001238 1.520 0.2289
BC 0.000000036 1 0.000000036 0.0004436 0.9834
BD 0.000008248 1 0.000008248 0.100 0.7528
BE 0.00000818 1 0.00000818 0.100 0.7538
CD 0.0004895 1 0.0004895 6.020 0.0215
CE 0.00005652 1 0.00005652 0.690 0.4125
DE 0.00006269 1 0.00006269 0.770 0.3884
A^2 0.000002199 1 0.000002199 0.027 0.8707
B^2 0.00004329 1 0.00004329 0.530 0.4725
C^2 0.00003170 1 0.00003170 0.390 0.5382
D^2 0.00005920 1 0.00005920 0.730 0.4018
E^2 0.00002175 1 0.00002175 0.270 0.6097
Residual 0.0020340 25 0.00008137
Lack of fit 0.001152 20 0.00005762 0.330 0.9673
Pure error 0.0008819 5 0.0001764
Cor. total 0.034000 45

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
Multi-objective parametric optimization nano powder mixed electrical discharge machining 7

Figure 4 (a) Surface plot showing effect of HVC and LVC on MRR. (b) Surface plot showing effect of Ton and FP on MRR.

EWR ¼ 0:96  1:5  102 A þ 4:54  103 B  1:8


 102 C þ 1:2  102 D þ 2:6  102 E þ 6
 102 AB  2:3  102 AC þ 2:6  102 AD
 3:77  103 AE  1:5  102 BC  2:08
 103 BD  1:5  102 BE  8:39  103 CD
þ 4:3  102 CE þ 3:1  102 DE þ 5:6  102 A2
þ 6:53  103 B2 þ 3  102 C2 þ 7:12  103 D2
þ 4:6  102 E2 ð6Þ

Similarly, for Ra, Fig. 7 shows a clear picture of the resid-


uals distribution uniform along the normal distribution line.
Fig. 8 shows the effect of pulse on time and low voltage current
on surface roughness. It is seen that on keeping Ton at low
level, the surface roughness parameter varies directly with
LVC. Similarly Fig. 8b shows the effect of flushing pressure
and pulse off time on Ra where it is found that with high flush-
ing pressure, Ra is small. The second order model equation for
Figure 5 Normal plot for EWR residuals. Ra is given by the following:

Ra ¼ 4:25 þ 4:4  102 A  4:841  104 B  8:744


Fig. 5 shows the normal plot for EWR where the residuals
 104 C þ 2:895  103 D þ 4:311  103 E  1:247
are normally distributed along the straight line. The second
order regression equation for EWR is given by Eq. (6) below.  103 AB  2:959  103 AC  2:912  103 AD
The response surface interaction plots of A  C and B  D are  5:563  103 AE þ 9:5  105 BC  1:436
shown in Fig. 6a and b respectively. It can be seen from Fig. 6a
that with the increase in flushing pressure, electrode wear rate  103 BD  1:43  103 BE  1:1  102 CD
(EWR) decreases and shows marginal variation with respect to þ 3:759  103 CE  5:02  104 A2 þ 2:227
low voltage current (LVC). From Fig. 6b, it is evident that
 103 B2 þ 1:906  103 C2 þ 2:605  103 D2
EWR decreases with the increase in pulse on time. It is also
seen that EWR significantly increases with the increase in high þ 1:579  103 E2 ð7Þ
voltage current (HVC).

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
8 S. Mohanty et al.

Figure 6 (a) Effect of LVC and FP on EWR. (b) Effect of HVC and Toff on EWR.

8
>
< 1 if y < L
Table 6 Relative importance of responses. 
Uy w2
For minimization : di ¼ if L 6 y 6 U ð9Þ
Response Optimization Desirability function Relative
>
:
UL
0 if y > U
objective (di) with weightage importance
(wi = 1) (ri) 8
>
< 0 if y < L and y > U
Material Maximization d1 ¼ MRRMRRmax 3 
yL w1
MRRmax MRRmin For both sides : di ¼ if L 6 y 6 T ð10Þ
removal rate > TL
:  Uy w2
(MRR) UT
if T 6 y 6 U
EWRmax EWR
Electrode Minimization d2 ¼ EWRmax EWRmin
5
wear ratio where di = individual desirability for ith response,
Surface Minimization d3 ¼ Ramax Ra 4 y = response value at ith experiment, U = maximum value,
Ramax Ramin
roughness L = minimum value, and T = targeted value of the response.
The individual desirability (di) of each response (y) varies
between 0 and 1, where di = 0 stands for an undesirable
response, di = 1 represents a completely desirable or an ideal
response, and the intermediate values of di indicate more or
5. Multi-response optimization using PSO with desirability less desirable responses. The individual desirability is then
function combined to obtain the global desirability function (D) as
given below:
5.1. Desirability function
 P1 ri Yn
 ri P1 ri
D ¼ dr11  dr22  dr33      drnn ¼ di ð11Þ
Derringer and Suich [15] developed the technique of ‘‘desir- i¼1

ability function” that can be widely applied in industries for where ri is the relative importance given to individual response
simultaneous optimization of multiple basing on the principle for characterization of the process in a scale varying from one
that quality of a product or process possessing many features is plus (+) to five plus (+ + + + +) for the least and highest
completely unacceptable if it lies outside the desirable limit. importance as given in Table 6. If these values of relative
Then, Candioti et al. [16] optimized multiple responses by con- importance (ri) are same for all the ‘‘n” responses, then the glo-
sidering the solutions of Derringer and such to find the suitable bal desirability function is as follows:
operating conditions that satisfied the criteria of all the
1 Y
n
1
responses to provide a best value. The multi-response opti- D ¼ ½d1  d2  d3      dn n ¼ ðdi Þn ð12Þ
mization is obtained after converting the individual responses i¼1
into their corresponding desirability (di) for obtaining the Glo-
bal Desirability Function (D). The individual desirability (di) Global Desirability Function; DF ¼ ½d1  d2  d3 1=3 ð13Þ
of the responses MRR, Ra, and FD is calculated from the
experimental data on basis of the optimization objectives of 1
Fitness Function; Y ¼ ð14Þ
each response with the following formulae: 1 þ DF
8
>
< 0  if y < L To obtain the highest quality characteristics the objective
For maximization : di ¼ yL w1
if L6y6U ð8Þ was to choose an optimal setting of the EDM parameters that
>
:
UL
maximize the desirability function, DF or minimize fitness
1 if y > U
function Y.

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
Multi-objective parametric optimization nano powder mixed electrical discharge machining 9

profile of the ith particle, then the initial cluster of particles


is generated by using the equations:
Position
Si0 ¼ Smin þ randðSmax  Smin Þ and Vi0 ¼
time
Xmin þ randðXmax  Xmin Þ
¼ ð15Þ
Dt
where Xmax is the upper limit of the input factor, Xmin is the
lower limit of the input factor and Dt is the design space time.
The value of the fitness function with the help of generated
particles is used to evaluate the next iteration i.e. the personal
best pbest and the global best gbest. The velocity profile of the
particles is given by the equations.
pbesti ðtÞ  xi ðtÞ
vi ðt þ 1Þ ¼ wvi ðtÞ þ c1 rand þ c2 rand
Dt
gbesti ðtÞ  xi ðtÞ
 ð16Þ
Dt
where vi (t + 1) represents velocity of the particle i at time
t + 1, wvi(t) is the current motion, c1rand (pbesti(t)  xi(t))/dt
Figure 7 Normal plot for surface roughness. denotes the particle memory influence, and c2rand (gbesti(t)
 xi(t))/dt represents the swarm influence
Si ðt þ 1Þ ¼ Si ðtÞ þ vi ðt þ 1Þ ð17Þ
5.2. Implementation of particle swarm optimization (PSO)
wmax  wmin
wi ¼ wmax  ð18Þ
imax
The basic of PSO technique is the interaction among the group
members or particles to perform optimization for the group where wmax and wmin are the first and last values of the inertia
activity where each particle informs about its time duration weight respectively and imax is the maximum number of itera-
to complete a work. It is an important optimization tool for tions used in PSO. Commonly wmax and wmin are taken as 0.9
the collective intelligence of the socio-biological group of and 0.4 respectively. Then, pbest 0.6 and gbest are compared with
organisms such as a cluster of birds, fish, ants. Here, the par- the particle values at each iteration by updating the velocity
ticles update their paths with that particle having the least time profile and at the end of iterations, the minimal value is
duration of completing the work so that the target is achieved selected according to the desirability function, and the corre-
in minimum duration of time. Generally PSO algorithm com- sponding input factors as the optimal conditions. Finally,
prises of the following steps: (i) generating the velocity profiles multi-objective desirability based particle swarm optimization
for the cluster of particles from their minimum and maximum code is developed and run in MATLAB interface. Here, the
value of the input factors, (ii) updating velocity of each particle initial swarm size is set at 50 with maximum number of itera-
with respect to its position in each iteration, (iii) evaluating the tion of 200. The constants C1 and C2 are taken as 2 with initial
fitness function value and (iv) updating the pbest and gbest for inertia weight of 0.4. The iteration is carried out and the
the particles. Let Si0 and Vi0 are the starting point and speed predicted output is given in Table 7.

Figure 8 (a) Effect of Ton and LVC on Ra. (b) Effect of FP and Toff on Ra.

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
10 S. Mohanty et al.

Table 7 Deviation of the predicted optimal and experimental response values.


Type of model Process parameters Response values
LVC (A) HVC (A) Ton (ls) Toff (ls) FP (kg/cm2) MRR (mm3/min) EWR (mm3/min) Ra
Predicted 18 2 30 30 0.6 0.569 1.041 3.827
Experimental 15 2 100 30 0.6 0.465 1.418 4.125
Error% 2.48 4.18 3.13

values show to be in good agreement with each other with


low percentage error.
The scope of composites is quite vast. Yet, machining the
same has challenging options. The priority is given to those
machining techniques that give efficient results in less cost.
Though the estimation analysis has not been done in the cur-
rent research yet, it is a challenge for researchers to obtain best
out of less cost. The present work has been done keeping cost
in mind though the estimation has not been discussed. To this
aspect very minimal cost is involved with better results.

References

[1] H.M. Chow, B.H. Yan, F.Y. Huang, J.C. Hung, Study of added
powder in kerosene for the micro-slit machining of titanium
alloy using electro-discharge machining, J. Mater. Process.
Technol. 101 (1) (2000) 95–103.
[2] S. Padhee, N. Nayak, S.K. Panda, P.R. Dhal, S.S. Mahapatra,
Figure 9 Graph between desirability function value vs. number Multi-objective parametric optimization of powder mixed
of iterations. electro-discharge machining using response surface
methodology and non-dominated sorting genetic algorithm,
Sadhana 37 (2) (2012) 223–240.
[3] M. Shabgard, B. Khosrozadeh, Investigation of carbon
It is clear from Fig. 9 that the optimal condition is obtained nanotube added dielectric on the surface characteristics and
in lesser number of iterations by applying desirability based machining performance of Ti–6Al–4V alloy in EDM process, J.
particle swarm optimization method. An experiment is carried Manuf. Process. 25 (2017) 212–219.
out as per the predicted conditions to validate the predicted [4] M. Kolli, A. Kumar, Effect of boron carbide powder mixed into
responses at optimal experimental conditions. The deviations dielectric fluid on electrical discharge machining of titanium
among the predicted optimal response values and experimental alloy, Procedia Mater. Sci. 5 (2014) 1957–1965.
response values are shown in Table 4 and it is seen that the [5] P. Sengottuvel, S. Satishkumar, D. Dinakaran, Optimization of
error obtained between these two is very less; hence, the opti- multiple characteristics of EDM parameters based on
desirability approach and fuzzy modeling, Procedia Eng. 64
mality of the machining condition can be validated.
(2013) 1069–1078.
[6] K.T. Chiang, F.P. Chang, D.C. Tsai, Modeling and analysis of
6. Conclusion the rapidly resolidified layer of SG cast iron in the EDM process
through the response surface methodology, J. Mater. Process.
Technol. 182 (1) (2007) 525–533.
The experimental research work carried out in this paper states
[7] S. Dewangan, S. Gangopadhyay, C.K. Biswas, Study of surface
the use of nano powders in EDM technology with a large num- integrity and dimensional accuracy in EDM using fuzzy TOPSIS
ber of experiments using different levels of factors such as low and sensitivity analysis, Measurement 63 (2015) 364–376.
voltage current, high voltage current, pulse off time, pulse on [8] S. Sivasankar, R. Jeyapaul, P.K. Kunhahamed, Performance
time and flushing pressure. The responses material removal study of tool materials and optimisation of pulse duration on
rates (MRR), electrode wear rate (EWR) and surface rough- EDM of zirconium di boride, Int. J. Mach. Mach. Mater. 14 (2)
ness (Ra) are measured for each setting. It is evident from (2013) 123–141.
the results that the powders enhance the machining rate and [9] A. Agrawal, A.K. Dubey, P.K. Shrivastava, Modeling and
surface finish. It is due to the fact that the powder helps in optimization of tool wear rate in powder mixed EDM of MMC,
in: 2nd International Conference on Mechanical and Robotics
increasing the spark gap and, also the breakdown characteris-
Engineering (ICMRE’2013), 2013, pp. 17–18.
tics of the dielectric fluid are improved. Furthermore, statisti-
[10] J.H. Jung, W.T. Kwon, Optimization of EDM process for
cal validation of the mathematical models developed using multiple performance characteristics using Taguchi method and
response surface methodology was performed. The use of nano Grey relational analysis, J. Mech. Sci. Technol. 24 (5) (2010)
particles shows significant variation in surface finish and mate- 1083–1090.
rial removal rate. It means MRR increases and surface rough- [11] P.N. Singh, K. Raghukandan, B.C. Pai, Optimization by grey
ness decreases. The predictive results and the experimental relational analysis of EDM parameters on machining Al–10%

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006
Multi-objective parametric optimization nano powder mixed electrical discharge machining 11

SiC P composites, J. Mater. Process. Technol. 155 (2004) 1658– [14] G.E. Box, D.W. Behnken, Some new three level designs for the
1661. study of quantitative variables, Technometrics 2 (4) (1960) 455–
[12] T. Muthuramalingam, B. Mohan, Application of Taguchi-grey 475.
multi responses optimization on process parameters in electro [15] G. Derringer, R. Suich, Simultaneous optimization of several
erosion, Measurement 58 (2014) 495–502. response variables, J. Qual. Technol. 12 (4) (1980) 214–219.
[13] A. Ramamurthy, R. Sivaramakrishnan, T. Muthuramalingam, [16] L.V. Candioti, M.M. De Zan, M.S. Camara, H.C. Goicoechea,
Taguchi-Grey Computation Methodology for Optimum Experimental design and multiple response optimization. Using
Multiple performance Measures on Machining Titanium Alloy the desirability function in analytical methods development,
in WEDM Process, 2015. Talanta 124 (2014) 123–138.

Please cite this article in press as: S. Mohanty et al., Multi-objective parametric optimization of nano powder mixed electrical discharge machining of AlSiCp using
response surface methodology and particle swarm optimization, Alexandria Eng. J. (2017), http://dx.doi.org/10.1016/j.aej.2017.02.006

You might also like