MULTI-RESPONSE CHARACTERISTICS OF EDM BY RSM, GRA and

ENTOPY MEASUREMENT METHOD

Dr. M. K. Pradhan
M.A. National Institute of Technology, Bhopal-462051, India,
Assistant Professor, Department of Mechanical Engineering
Email: [email protected]

This paper reports a new combination of Response Surface Methodology (RSM), Grey Relational Analysis (GRA) and entropy
measurement method have been suggested to optimize the process parameters that maximize productivity without compromising
the surface quality. The input parameters of Electrical Discharge Machining (EDM) considered for this analysis are Pulse
Current (Ip), Pulse Duration (Ton), Duty Cycle (Tau) and Discharge Voltage (V). The designed experimental results are used in
GRA, and the weights of the responses are determined by entropy measurement method. Based on optimization results, using the
RSM the interactive effects of the machining parameters on the responses were evaluated. It is found that the Grey Relational
Grade (GRG) was dominantly influenced by Ip and their interactions with the other parameters. This method is simple with easy
operability, provides a scientific reference to obtain useful information to control the parameter to ensure high productivity
without compromising the quality of the EDM surfaces.

Keywords: Electrical discharge machining (EDM), Response surface methodology; (RSM); Grey relational analysis; (GRA); Entropy
measurement Method; Material removal rate; (MRR); Surface roughness; (Ra).

manufacturing tools in mould industries. Consequently, the

1. INTRODUCTION
multiple objective optimization problems have been of
increasing interest to the investigators to minimize the
complexity. A trouble-free and reliable approach based on
Electrical discharge machining (EDM) is an
statistically designed experiments, and RSM approach has
extensively used non-conventional machining process, been adopted with a face-center cube experimental design,
uses thermal energy to machine electrically conductive as a special case of CCD, later it synergies with GRA and
hard material parts regardless of their geometry [1,2].
entropy measurement technique for optimizing the
Many automotive and aerospace components, as well as
machining parameters in order to optimize the aforesaid
moulds and dies are manufactured using EDM. EDM has
responses. Gray analysis transfers exceptional solution to
been a mainstay of manufacturing for more than six
uncertain, multi-input, and discrete data problems. As
decades, offering distinctive capacities to machine EDM process is of similar nature, thus, this approach is
“difficult to machine” materials with desire shape, size, extremely suitable in parameter optimization of such
and required dimensional accuracy.
experimental work.
A large number of factors to be considered for EDM
process, since it is influenced by many factors [3]. So
2. EXPERIMENTATION
based on experience, related literatures, and the working
characteristics of the machine, the prime parameters
chosen, in the present study are discharge current, spark
Experiments were conducted to study the effects of
on-time, duty cycle (defined as the ratio of pulse on time to
various machining parameters; such as Ip, Ton Tau and V
the total pulse period) and discharge voltage [4,5].
on MRR and Ra on AISI D2 tool steel work piece
Nevertheless, numerous efforts have been made to
machined on die sinking electro discharge machine (make:
improve the productivity and surface quality [5, 6, 7] but
Electronica Elektra plus PS 50ZNC). An electrolytic pure
attempts with the combination of RSM, GRA and entropy
copper with a diameter of 30 mm was used as a tool
measurement for finding out optimal setting on EDMed
electrode (positive polarity) and work piece material used
AISI D2 tool steel had been rarely attempted [8]. AISI D2
was 100 mm diameter with thickness 4 mm d. Commercial
tool steel has been growing ranges of applications in
grade EDM oil (specific gravity = 0.763, freezing point =
 94o C) was used as dielectric fluid with lateral flushing two factors can be measured discretely. When experiments
pressure of 0.4 kgf /cm2. The arrangement to conduct the are unclear or if the experimental method cannot be carried
experiments using a face centered CCD with four out accurately, grey analysis assists to reimburse for the
variables, having a total of 30 runs in three blocks [8]. The deficiency in statistical regression.
different levels of factor considered for this study are
illustrated in Table 1. The machining time for all Table 1 Input parameters and their levels
experiments was kept constant at 15 min, and the various Parameters Level
responses are measured and tabulated in Table 2. 1 2 3
Ip (A) 5 10 15
Ton (μs) 50 75 100
3. RESPONSE SURFACE METHODOLOGY Tau (%) 50 66.5 83
Voltage (V) 40 45 50

Response surface methodology is a collection of Start

statistical and mathematical techniques useful for
developing, improving, and optimizing processes and the Mesurement of Responses
objective is to find the correlation between the response,
and the variables investigated [8]. A Central Composite Normalize the Result

Design (CCD) is used since it gives a comparatively

Calculate the dediation sequence
accurate prediction of all response variable averages
related to quantities measured during experimentation [9]. Calculation of Grey relational co-efficient
In these methods, there is a possibility that the experiments
will stop with few runs and decide that the prediction Calculate the grey relational grade using
Entopy measurement method
model is satisfactory. In CCD, the limits of the
experimental domain to be explored are defined and are
Analysis of Grey relational Grade using RSM
made as wide as possible to obtain a clear response from
the model. The discharge current (A), pulse on time (B) Getting optimal Setting
and pulse pause time (C) are the machining variables
selected for this investigation. The different levels taken Confirmation tests
for this study are depicted in Table 1. The second-order
model is normally used when the response function is not End

known or nonlinear. In the present study, a second-order

model has been utilized. The experimental values are Figure 1 Flowchart for grey relational Grade calculation.
analyzed and the mathematical model is then developed
that illustrate the relationship between the process variable
and response. The following second-order model explains In Grey relational analysis, the complex multiple
the behavior of the system. response optimizations can be simplified into the
k k k optimization of a single response Grey relational grade. At
Y   0    i X i    ii X i2    ij X i X j   (1) the outset in data pre-processing is the method of
i 1 i 1 i , j 1, i  j transferring the original sequence to a comparable
where Y is the corresponding response, Xi is the input sequence, where the original data normalise to a range of 0
variables, X2ii and XiXj are the squares and interaction and 1. There are three different kind of data normalizations
terms, respectively, of these input variables. The unknown are generally carried out rendering, whether the lower is
regression coefficients are βo, βi, βij and βii and the error in better (LB), the higher is better (HB), or nominal the best
the model is depicted as . (NB). For “the-larger-the-better’ characteristics such as
productivity or MRR the original sequence can be HB and
the original sequence should be normalized as explained
4. GREY RELATIONAL ANALYSIS by Fung [11] as follows:

GRA is a decision making technique based on grey ( ) ( )

( ) ( )
system theory originally developed by Deng [10]. Where, ( ) ( )
black represents a system with deficient information,
whereas a white system stands for complete information. However, if the expectancy is “as small as” possible for
However, the grey relation is the relation with incomplete responses such as Surface roughness, then the original
information and is used to characterize the grade of sequence should be normalized as LB as follows:
association between two sequences so that the distance of
 ( ) ( ) distinguishing coefficient or identification coefficient
( ) ( ) varies from 0≤ ζ ≤1.
( ) ( )
The GRG is a weighting-sum of the grey relational
where, ( ) is the value after the grey relational coefficients and it is defined as
generation (data pre-processing), ( ) is the largest ( ) ∑ ( ) (5)
value of ( ), ( ) is the smallest value of ( )
and x0 is the desired value. where represents the weighting value of the
performance characteristic, and ∑ . In the
Table 2 Experimental layout present analysis, the weights are computed using entropy
Ip Ton MRR Ra measurement method discussed in the subsequent section.
Run Tau V
(A) (μs) (mm3/min) (μm) With reference to the application, different response may
1 10 75 66.5 45 10.35 5.55 have different keenness and thereby, different weightages
2 10 75 66.5 45 9.04 5.98 have to be allocated to different responses. If different
3 15 50 83 50 29.163 8.43 precedence weightages have been allocated to different
4 5 50 50 50 5.18 5.01 responses, the equation for calculating overall grey
5 5 100 83 40 5.245 5.03
relational grade becomes:
6 15 50 50 50 33.103 7.43
∑ ( )
7 5 50 50 40 4.606 4.59 ( )
8 5 50 83 40 8.872 4.71 ∑
9 15 100 50 50 51.09 8.1
10 10 75 66.5 45 8.95 6.12
11 5 100 50 40 4.349 4.89
5. ENTROPY MEASUREMENT METHOD
12 15 100 50 40 51.004 10.93
13 5 100 83 50 6.972 5.7 Conventionally, GRA is based on the geometric mean
14 15 100 83 40 33.023 12.49 of the sequence data in relational space, such operation may
15 5 50 83 50 14.12 5.19 make the measure biased if the evaluations are deficient
16 5 100 50 50 4.349 5.59 and inconsistent in quantity [12]. To improve this
17 15 100 83 50 33.11 9.01 deficiency, the entropy concept is generally used. These
18 15 50 50 40 29.737 10.49 dissimilarities in weight are computed by using entropy
19 10 75 66.5 45 8.42 6.54 measurement method, which is an objective weighting
20 15 50 83 40 20.002 12.01 method that can calculate the comparative importance
21 10 75 66.5 40 8.936 8.2 amongst the attributes through relating the entropy value
22 10 75 83 45 9.356 7.13 for each attribute. Wen et. al [13] recommend the grey
23 10 75 66.5 50 11.007 6.35 entropy based on the discrete type of entropy that
24 15 75 66.5 45 33.084 9.68 appropriately conducts weighting analysis, a method of the
25 10 50 66.5 45 9.182 5.87 grey entropy weighting contains following steps:
26 10 75 66.5 45 11.01 6.25
27 5 75 66.5 45 5.361 6.07 (i). Let X be a factor set of grey relation.
28 10 100 66.5 45 10.43 7.27 (ii). Compute the summation of each attribute’s value
29 10 75 50 45 9.25 5.92
for all sequences,
30 10 75 66.5 45 9.352 6.75

∑ ( ) ( )
4.1. Grey Relational Coefficient & Grey Relational
Grade (iii). Compute the normalization coefficient K:
After normalizing the data, usually grey relational
coefficient is calculated to show the relationship between ( )
( )
the optimal and actual normalized experimental results.
The grey relational coefficient can be expressed as: (iv). Find the entropy for the specific attribute,

( ) ( ( ) ( )) (4) ∑ ( ) ( )
( )
i= 1,....n; k = 1, ......p. ( )
Where ( ) ( )
where ( ) | ( ) ( )| s the difference of the ( )
absolute value called “deviation sequence” of the reference
sequence x0(k) and comparability xi(k) is the
(v). Compute the total entropy value E:
 6. RESULT AND DISCUSSION
∑ ( )

At the outset, the experiments were conducted as

(vi). Determine the relative weighting factor
discussed in Sec 2 and the experimental results were
recorded in Table 2. Characteristically MRR is considered
( ) ( )
to be as large as possible and Ra is considered to be as
(vii). The normalized weight of each attribute can be small as possible. Therefore, all MRR values were
substituted in Eq. 2 and Ra was substituted in Eq. 3 to get
calculated as: ∑
( ) normalized values, which are presented in Table 3. As
mentioned by Deng [10], larger values of the normalised
Grey relational grade are calculated by multiplying results stand for better performance, and the maximum
grey relational coefficients with their corresponding normalized results that are equal to 1 specify the best
weight of quality characteristics. performance. The results shown in Table 3 were
substituted in Eq. 4 to compute grey relational coefficients
Table 3 Normalised value, deviation sequence, GRG and of the aforementioned responses. Subsequently, the
Rank of the experiments deviation sequences D0i are calculated. The distinguishing
Normalized Deviation Grey relational coefficient z is used to adjust the range of the comparison
Run values sequences coefficient
Order MRR Ra MRR Ra MRR Ra
GRG Rank environment, which can be substituted for the grey
relational coefficient in Eq. 4. For this study, the value of z
1 0.128 0.878 0.872 0.122 0.365 0.804 0.585 9
is selected as 0.5 and the observed values were
2 0.100 0.824 0.900 0.176 0.357 0.740 0.548 15
summarized in Table 4. The grey relational coefficient and
3 0.531 0.514 0.469 0.486 0.516 0.507 0.511 22 grade for each experiment of the experimental arrangement
4 0.018 0.947 0.982 0.053 0.337 0.904 0.621 7 calculated by applying Eq. 4 and 5. Table lists the values
5 0.019 0.944 0.981 0.056 0.338 0.900 0.619 8 of each the grey relational grade.The grey relational grads
6 0.615 0.641 0.385 0.359 0.565 0.582 0.573 10 are calculated by the grey relational coefficients using Eq
7 0.005 1.000 0.995 0.000 0.335 1.000 0.667 3
5.The weightage for each response is it reflects its
comparative significance in the grey relational analysis
8 0.097 0.985 0.903 0.015 0.356 0.971 0.663 4
which is decided in this study by entropy measurement
9 1.000 0.556 0.000 0.444 1.000 0.529 0.765 1 method proposed by Wen et al. [13] and described in
10 0.098 0.806 0.902 0.194 0.357 0.721 0.539 16 section 5. After pre-processing, the sequences of MRR and
11 0.000 0.962 1.000 0.038 0.333 0.929 0.631 5 Ra against different experimental runs are depicted in
12 0.998 0.197 0.002 0.803 0.996 0.384 0.690 2 Table 5, which has been calculated using Eq. 10. The
13 0.056 0.859 0.944 0.141 0.346 0.781 0.563 12
calculated weight for MRR and Ra has been found to 0.5,
and 0.5, respectively by using Equation. (6), (7), (8), (9),
14 0.613 0.000 0.387 1.000 0.564 0.333 0.449 28
(10), and (11) The obtained weights of each quality
15 0.209 0.924 0.791 0.076 0.387 0.868 0.628 6
characteristics have been further used to calculate the grey
16 0.000 0.873 1.000 0.127 0.333 0.798 0.566 11 relational grade by using Eq.(5a), thus, the multi-criteria
17 0.615 0.441 0.385 0.559 0.565 0.472 0.519 20 optimization problem has been transformed into a single
18 0.543 0.253 0.457 0.747 0.523 0.401 0.462 27 equivalent objective function optimization problem using
19 0.087 0.753 0.913 0.247 0.354 0.669 0.512 21 the combination of RSM, GRA and Entropy measurement
20 0.335 0.061 0.665 0.939 0.429 0.347 0.388 30
method. Higher is the value of Grey relational grade, the
corresponding factor combination is said to be close to the
21 0.098 0.543 0.902 0.457 0.357 0.522 0.440 29
optimal. The grey relational grades were also analysed in
22 0.107 0.678 0.893 0.322 0.359 0.609 0.484 25 the main effect analysis, and then the optimization
23 0.142 0.777 0.858 0.223 0.368 0.692 0.530 19 processing parameters of multiple quality characteristics
24 0.615 0.356 0.385 0.644 0.565 0.437 0.501 24 were obtained from the response table and response graph
25 0.103 0.838 0.897 0.162 0.358 0.755 0.557 13 for grey relational analysis. The GRG of each combination
26 0.143 0.790 0.857 0.210 0.368 0.704 0.536 17
are then ranked as per the value and it is found that a set of
optimal machining parameters based on the highest GRG
27 0.022 0.813 0.978 0.187 0.338 0.727 0.533 18
value of 0.765 peak value, which is obtained with the
28 0.130 0.661 0.870 0.339 0.365 0.596 0.480 26
Ip=15 A, Ton=100ms, Tau=50% and V=50 volt. The
29 0.105 0.832 0.895 0.168 0.358 0.748 0.553 14 assessment outcome provides a scientific reference to
30 0.107 0.727 0.893 0.273 0.359 0.646 0.503 23 obtain the minimal condition of MRR, and Ra.
 The normal probability plot of residuals for GRG is
illustrated in Fig. 2; this graph plots the residuals versus
their expected values when the distribution is normal. The
residuals from the analysis should be normally distributed.
It can be seen that the residuals are almost falling on a
straight line, which indicates that the errors are normally
distributed. Further, Fig. 3 represented is the residual plot
for the models and dataset on GRG, it can be seen it
depicts a random distribution, indicating that the residual
distribution of the regression equation follows normal and
independent patterns [4]. This recommends the high
adequacy of the quadratic models for evaluating the
response. A set of optimal process parameters based on
the peak value at each level of the factors are shown in
Fig.4. It can be seen that the optimal machining parameters Figure 3 Residual plots of quadratic model
are Ip1, Ton3, Tau1, and V3 for GRG. The contour plots
of the quadratic model with two variables kept at their Ip (A) Ton(μs)

central levels and the other two varying within the 0 .6

0

experimental ranges are, respectively, shown in Fig. 6. Fig. . 58

0
56
5(a) represents the contour plot for GRG in relation to the 0 .
4
machining parameters of Ip and Ton on GRG. It can be 0 .5
. 52
Mean of GRG
seen that the maximum GRG can be seen when Ip and Ton 0

10

15

50

75

100
at their lower level. Similar counter plots were also
observed in Fig. 5(b), (c) (d) (e) and (f). The corresponding Tau V
two-dimensional contours show a considerable curvature,
0
implying that these three factors were interdependent. In 0. 6
8
other words, there were significant interactive effects on 0. 5
. 56
GRG. Figure 6 shows the percentage contribution of 0
4
0. 5
factors and their interactions on the GRG. It can be seen 52
0.
that Ip, Tau and their interactions have the significant
50.0

66.5

83.0

40

45

50
influence on the GRG. Figure 4 Main Effect of factors on GRG

100 Ton(μs)*Ip (A) 83 Tau*Ip (A) 50 V*Ip (A) GRG

< 0.46
0.46 - 0.48
0.48 - 0.50
75 66 45 0.50 - 0.52
0.52 - 0.54
0.54 - 0.56
0.56 - 0.58
50 50 40
5 10 15 5 10 15 5 10 15 0.58 - 0.60
0.60 - 0.62
Tau*Ton(μs) 50 V*Ton(μs) 50 V*Tau
> 0.62
80

70 Hold Values
45 45 Ip (A) 10
Ton(μs) 75
60
Tau 66.5
V 45
50 40 40
50 75 100 50 75 100 50 66 83
Figure 2 Normal probability plot of residuals for GRG
Figure 5 Two-dimensional contour plots for GRG: effect
of (a) Ip and Ton (b) Ip and Tau (c) Ip and V (d) Ton and
Tau (e) Ton and V (f) Tau and V
 Reference

1. R. Snoeys, F. Staelens, and W. Dekeyser, “Current

trends in nonconventional material removal processes,
ANN. CIRP, vol. 35(2), p. 467 480, 1986.
2. Kunieda, M., Lauwers, B., Rajurkar, K. P. and
Schumacher, B. M., Advancing EDM through
fundamental insight into the process, Annals of CIRP,
2005, 54(2), 599-622.
3. K. H. Ho and S. T. Newman, State of the art electrical
discharge machining (EDM),” Int. Journal of Machine
Tools and Manuf., vol. 43, pp. 1287–1300, 2003.
4. Pradhan, M. K. Experimental investigation and
Figure 6 Percentage contributions of process parameters
modelling of surface integrity, accuracy And
Productivity aspect in EDM of AISI D2 Steel. Ph.D
7. CONCLUSION thesis, NIT, Rourkela, 2010.
5. Pradhan, M. K. and C. K. Biswas, “Investigating the
effect of machining parameters on EDMed components
This investigation purposes a hybrid, integrated A RSM approach”, Int. J of Mechanical Engg,
approach of grey relational analysis coupled with entropy Universiti Teknologi MARA, Shah Alam, Malaysia.
measurement method for the optimization of the vol. 7, No-1 pp.47-64, 2010.
machining parameters of EDM process. Grey relational 6. Pradhan, M. K. and Biswas, C. K. (2009).
analysis transforms optimization of the multiple responses Investigations into the effect of process parameters on
into optimization of a single response problem the grey MRR in EDM of AISI D2 steel by response surface
relational grade. Factors having a significant effect are methodology. Journal of Mechatronics and Intelligent
retrieved using ANOVA and by regulating these process Manufacturing, vol. 1 No. (3-4).
parameters, optimization of responses was carried out. It 7. El-Taweel, “Multi-response optimization of EDM with
has also been found that the Ip is the most important al-cu-si-tic p/m composite electrode,” International
process parameter due to its highest percentage Journal of Advanced Manuf. Technology, pp. 1–14,
contribution of 11.62 % followed by Tau with 14.37 % 2008.
Ton and V with 1.3 and 2.07 %, respectively. However, 8. Pradhan, M. K, Multi-objective optimization of MRR,
the percentage contribution of Interaction terms such as TWR and Radial Overcut of EDMed AISI D2 tool steel
has 15.48, using response surface methodology, grey relational
12.18, 11.01 and 4.44 % respectively. The optimal analysis and entropy measurement” International J. for
operational conditions established by grey analysis Manuf. Science & Production, Vol-12, (1), 2012.
approach are as follows: a pulse current 15 A, pulse 9. Douglas C. Montgomery. Design and Analysis of
duration 100ms, duty cycle = 50% and discharge voltage Experiments. Wiley, New York, 2001.
50 volt. This research will help in recognizing the 10. Robert L. M., Richard F. G., Dallas, Texas, .James
significant factors that are efficiently adjusted to L. Hess., Statistical Design and Analysis of
diminishing error, time consumption, cost and to Experiments With Applications to Engineering and
increasing quality and productivity of the process. This Science”, 2nd Edition, John Wiley & sons pub., (2003).
research may facilitate an effective guideline for the 11. C P Fung. Manufacturing process optimization for
experimenter and practitioners to select optimum wear property of Fiber-reinforced polbutylene
parameter settings to obtain desired the MRR, and Ra terephthalate composites with grey relation analysis.
during EDM die sinking of AISI D2 tool steel. This Wear, 254:298–306, 2003
technique can also be useful for optimizing the processing 12. C.L. Chang, C.H. Tsai, and L. Chen. Applying grey
parameters of other manufacturing processes, to boost relational analysis to the decathlon evaluation model.
manufacturing efficiency. Int. J. of Computer, Internet and Management,11:
(3):54–62., 2003.
13. Kun-Li Wen, Ting-Cheng Chang, and Mei-Li You.
The grey entropy and its application in weighting
analysis. IEEE International Conference on Systems,
Man, and Cybernetics, 2:1842–1844, 1998.