Journal of Materials Processing Technology 102 (2000) 48±55

Optimization of the electrical discharge machining process

based on the Taguchi method with fuzzy logics
J.L. Lina,*, K.S. Wangb, B.H. Yanb, Y.S. Tarngc
a
Department of Mechanical Engineering, National Central University, Chung-Li, Taoyuan, Taiwan, ROC
b
Department of Mechanical Engineering, National Central University, Chung-Li, Taoyuan, Taiwan, ROC
c
Department of Mechanical Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, ROC

Received 9 October 1998

Abstract

In this paper, the application of the Taguchi method with fuzzy logic for optimizing the electrical discharge machining process with
multiple performance characteristics has been reported. A multi-response performance index is used to solve the electrical discharge
machining process with multiple performance characteristics. The machining parameters (the workpiece polarity, pulse-on time, duty
factor, open discharge voltage, discharge current and dielectric ¯uid) are optimized with considerations of the multiple performance
characteristics (electrode wear ratio and material removal rate). Experimental results are presented to demonstrate the effectiveness of this
approach. # 2000 Published by Elsevier Science S.A. All rights reserved.

Keywords: Electrical discharge machining; Taguchi method; Fuzzy logics

1. Introduction fuzzy logic. As a result, optimization of complicated multi-

ple performance characteristics can be transformed into the
The Taguchi method can optimize performance charac- optimization of a single multi-response performance index
teristics through the settings of process parameters and (MRPI). In this paper, the optimization of the electrical dis-
reduce the sensitivity of the system performance to sources charge machining process with multiple performance char-
of variation. As a result, the Taguchi method has become a acteristics has been investigated to illustrate this approach.
powerful tool in the design of experiment methods [1±3]. Electrical discharge machining (EDM) has been used
However, most published Taguchi applications to date have effectively in the machining of hard, high-strength, and
been concerned with the optimization of a single perfor- temperature-resistant materials. Material is removed by
mance characteristic. Handling the more demanding multi- means of rapid and repetitive spark discharges across the
ple performance characteristics is still an interesting gap between the tool and the workpiece. In electrical dis-
research problem [4,5]. charge machining, it is important to select machining para-
The theory of fuzzy logics, initiated by Zadeh in 1965 [6] meters for achieving optimal machining performance [7].
has proven to be useful for dealing with uncertain and vague Usually, the desired machining parameters are determined
information. In fact, the de®nition of performance charac- based on experience or handbook values. However, this does
teristics such as lower-the-better, higher-the-better, and not ensure that the selected machining parameters result in
nominal-the-better contains a certain degree of uncertainty optimal or near optimal machining performance for that
and vagueness. Therefore, optimization of the performance particular electrical discharge machine and environment. To
characteristics with fuzzy logic has been considered in this solve this task in the present study, the Taguchi method with
study. In this study, a fuzzy reasoning of the multiple fuzzy logic is used as an ef®cient approach to determine the
performance characteristics has been developed based on optimal machining parameters in the electrical discharge
machining process.
*
Corresponding author. Tel.: ‡886-39-771997-284;
In the following, optimization of multiple performance
fax: ‡886-39-771997-283. characteristics with fuzzy logic is introduced brie¯y and the
E-mail address: [email protected] (J.L. Lin) electrical discharge machining process is then described,

0924-0136/00/$ ± see front matter # 2000 Published by Elsevier Science S.A. All rights reserved.
PII: S 0 9 2 4 - 0 1 3 6 ( 0 0 ) 0 0 4 3 8 - 6
 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55 49

after which the experimental details of using the Taguchi may vary. As a result, the application of the Taguchi method
method with fuzzy logic to optimize the electrical discharge in a process with multiple performance characteristics can-
machining process to secure low electrode wear ratio (EWR) not be straightforward. In this paper, the use of fuzzy logic to
and high material removal rate (MRR) are given. Finally, the deal with the optimization of a process with multiple
paper concludes with a summary. performance characteristics is reported. First, several fuzzy
rules are derived based on the performance requirement of
the process. The loss function corresponding to each per-
2. Optimization of multiple performance formance characteristic is fuzzi®ed and then a single MRPI
characteristics with fuzzy logic is obtained through fuzzy reasoning on the fuzzy rules. The
MRPI can be used to optimize the process based on the
Experimental design methods [8] were developed origin- Taguchi approach.
ally by Fisher [9]. However, classical experimental design
methods are too complex and not easy to use. Furthermore, a
large number of experiments have to be carried out as the 3. The electrical discharge machining process
number of the process parameters increases. To solve this
important task, the Taguchi method uses a special design of The electrical discharge machining removes workpiece
orthogonal array to study the entire parameter space with by an electrical spark erosion process. The variations in the
only a small number of experiments. The experimental machining parameters, such as the workpiece polarity,
results are then transformed into a signal-to-noise (S/N) pulse-on time, duty factor, open discharge voltage, discharge
ratio. The S/N ratio can be used to measure the deviation current, and dielectric ¯uid, greatly affect the measures of
of the performance characteristics from the desired values. the machining performance, for example, the EWR and the
Usually, there are three categories of performance charac- MRR. Therefore, proper selection of the machining para-
teristics in the analysis of the S/N ratio: the lower-the-better, meters can result in better machining performance in the
the higher-the-better, and the nominal-the-better. Regardless electrical discharge machining process.
of the category of the performance characteristic, a larger
S/N ratio corresponds to better performance characteristic. 3.1. Machining parameter selection
Therefore, the optimal level of the process parameters is the
level with the highest S/N ratio. Furthermore, a statistical In this study, an EDM machine (KT-200) was used as the
analysis of variance (ANOVA) is performed to identify the experimental machine. Cylindrical pure copper with a dia-
process parameters that are statistically signi®cant. The meter of 8 mm was used for an electrode to erode a work-
optimal combination of the process parameters can then piece of SKD11 with a diameter of 8 mm. The schematic
be predicted based on the above analysis. Finally, a con- diagram of the experimental set-up is shown in Fig. 1. The
®rmation experiment is conducted to verify the optimal workpiece and electrode were separated by a moving dielec-
process parameters obtained from the parameter design. tric ¯uid such as kerosene. In the experiments, an additive
Basically, the Taguchi method is designed to handle the (aluminum powder or aluminum oxide powder) to the
optimization of a single performance characteristic. The kerosene used as the dielectric ¯uid was employed to
usual recommendation for the optimization of a process improve the machining performance [11]. Machining
with multiple performance characteristics is left to engineer- experiments for determining the optimal machining para-
ing judgment and veri®ed by con®rmation experiments [10]. meters were carried out by setting: negative polarity or
This is because several problems are encountered in the positive polarity of the workpiece; a pulse-on time in the
optimization of a process with multiple performance char- range of 20±300 ms; a duty factor in the range of 0.3±0.7; an
acteristics. For example: the category of each performance open discharge voltage in the range of 100±150 V; a dis-
characteristic may not be same; the engineering unit for charge current in the range of 1.5±6.0 A and three different
describing each performance characteristic may be differ- dielectric ¯uids. To perform the experimental design, three
ent; and the importance of each performance characteristic levels of the machining parameters (pulse-on time, duty

Table 1
Machining parameters and their levels

Symbol Machining parameter Unit Level 1 Level 2 Level 3

A Workpiece polarity Negative Positive

B Pulse-on time ms 20 150 300
C Duty factor 0.3 0.5 0.7
D Open discharge voltage V 100 120 150
E Discharge current A 1.5 4.0 6.0
F Dielectric fluid Kerosene Kerosene‡Al2O3 powder Kerosene‡Al powder
50 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55

Fig. 1. Schematic diagram of the EDM process.

factor, open discharge voltage, discharge current, dielectric 4.1. Orthogonal array experiment
¯uid) were selected and are shown in Table 1. The initial
machining parameters were selected as follows: negative To select an appropriate orthogonal array for the experi-
polarity of the workpiece; a pulse-on time of 20 ms; a duty ments, the total degrees of freedom need to be computed.
factor of 0.5; an open discharge voltage of 120 V; a dis- The degrees of freedom are de®ned as the number of
charge current of 4.0 A; and an additive of aluminum oxide comparisons between process parameters that need to be
powder in the kerosene used as the dielectric ¯uid. made to determine which level is better and speci®cally how
much better it is. For example, a two-level process parameter
3.2. Machining performance evaluation counts for one degree of freedom. The degrees of freedom
associated with interaction between two process parameters
The machining performance is evaluated by the EWR and are given by the product of the degrees of freedom for the
the MRR. The EWR is de®ned as the ratio of the electrode two process parameters. In the present study, the interaction
wear weight (EWW) to the workpiece removal weight between the machining parameters is neglected. Therefore,
(WRW) and is usually expressed as a percentage, i.e. there are 11 degrees of freedom owing to one two-level
machining parameter and ®ve three-level machining para-
EWW
EWR …%† ˆ  100 (1) meters in the EDM process.
WRW Once the degrees of freedom are known, the next step is
The MRR is the WRW under a period of machining time select an appropriate orthogonal array to ®t the speci®c task.
in minutes (T), i.e. The degrees of freedom for the orthogonal array should be
greater than or at least equal to those for the process
WRW
MRR …g=min† ˆ (2) parameters. In this study, an L18 orthogonal array was used
T because it has 17 degrees of freedom more than the 11
In the experiments, the machining time for each work- degrees of freedom in the machining parameters. This array
piece is 30 min. Basically, the lower is the EWR in the EDM has 8 columns and 18 rows and it can handle one two-level
process, the better is the machining performance. However, process parameter and seven three-level process parameters,
the higher is the MRR in the EDM process, the better is the at most. Each machining parameter is assigned to a column
machining performance. Therefore, the EWR is the lower- and 18 machining parameter combinations are required.
the-better performance characteristic and the MRR is the Therefore, only 18 experiments are needed to study the
higher-the-better performance characteristic. entire machining parameter space using the L18 orthogonal
array. The experimental combinations of the machining
parameters using the L18 orthogonal array is presented in
4. Determination of the optimal machining parameters Table 2.

In this section, the use of the Taguchi method with 4.2. Signal-to-noise ratio
fuzzy logic to determine the machining parameters with
optimal machining performance in the EDM process is In the Taguchi method, a loss function is de®ned to
illustrated. calculate the deviation between the experimental value
 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55 51

Table 2
Experimental layout using an L18 orthogonal array

No. Machining parameter

Workpiece polarity (A) Pulse-on time (B) Duty factor (C) Discharge voltage (D) Discharge current (E) Dielectric fluid (F)
1 1 1 1 1 1 1
2 1 1 2 2 2 2
3 1 1 3 3 3 3
4 1 2 1 1 2 2
5 1 2 2 2 3 3
6 1 2 3 3 1 1
7 1 3 1 2 1 3
8 1 3 2 3 2 1
9 1 3 3 1 3 2
10 2 1 1 3 3 2
11 2 1 2 1 1 3
12 2 1 3 2 2 1
13 2 2 1 2 3 1
14 2 2 2 3 1 2
15 2 2 3 1 2 3
16 2 3 1 3 2 3
17 2 3 2 1 3 1
18 2 3 3 2 1 2

and the desired value. Usually, there are three categories of are different. To consider the two different performance
the performance characteristics in the analysis of the signal- characteristics in the Taguchi method, the S/N ratios corre-
to-noise ratio, i.e., the lower-the-better, the higher-the-bet- sponding to the EWR and MRR are processed by the fuzzy
ter, and the nominal-the-better. To obtain optimal machining logic unit.
performance, the minimum EWR and the maximum MRR
are desired. Therefore, the lower-the-better EWR and the 4.3. Fuzzy logic unit
higher-the-better MRR should be selected.
The loss function Lij of the lower-the-better performance A fuzzy logic unit comprises a fuzzi®er, membership
characteristic can be expressed as functions, a fuzzy rule base, an inference engine, and a
defuzzi®er. First, the fuzzi®er uses membership functions to
1X n
fuzzify the S/N ratios. Next, the inference engine performs a
Lij ˆ y2 (3)
n kˆ1 ijk fuzzy reasoning on fuzzy rules to generate a fuzzy value.
Finally, the defuzzi®er converts the fuzzy value into a MRPI.
where Lij is the loss function of the ith performance char-
acteristic in the jth experiment, n the number of tests, and yijk
is the experimental value of the ith performance character- Table 3
istic in the jth experiment at the kth test. Experimental results for the EWR and its S/N ratio
The loss function of the higher-the-better performance
No. EWR (%) S/N ratio (dB)
characteristic can be expressed as
1 33.6 9.510
1X n
1 2 29.9 10.486
Lij ˆ (4)
n kˆ1 y2ijk 3 47.1 6.356
4 6.70 23.380
The loss function is further transformed into an S/N ratio. 5 11.8 18.566
6 43.9 7.151
In the Taguchi method, the S/N ratio is used to determine the
7 41.0 7.682
deviation of the performance characteristic from the desired 8 4.91 26.013
value. The S/N ratio Zij for the ith performance characteristic 9 0.70 43.383
in the jth experiment can be expressed as 10 22.8 12.665
11 32.1 9.864
Zij ˆ ÿ10 log …Lij † (5) 12 29.2 10.688
13 15.5 16.083
Table 3 shows the experimental results for the EWR and 14 67.2 3.449
its S/N ratio based on the experimental parameter combina- 15 29.2 10.550
tions (Table 2). The experimental results for the MRR and its 16 38.3 8.327
S/N ratio are shown in Table 4. As shown in Tables 3 and 4, 17 24.1 11.935
18 121.2 ÿ1.681
the engineering units for describing the EWR and the MRR
52 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55

Table 4
Experimental results for the MRR and its S/N ratio

No. MRR (g/min) S/N ratio (dB)

1 0.00182 ÿ25.2606
2 0.00159 ÿ26.4851
3 0.00089 ÿ31.5212
4 0.00464 ÿ17.1720
5 0.00096 ÿ31.0124
6 0.00128 ÿ28.3449
7 0.00024 ÿ43.0697
8 0.00303 ÿ20.8298
9 0.00330 ÿ20.0926
10 0.00024 ÿ43.0255
11 0.00021 ÿ44.2724
12 0.00026 ÿ42.2474
13 0.00026 ÿ42.6933
14 0.00029 ÿ41.1814
Fig. 2. Membership functions for the EWR.
15 0.00033 ÿ40.6065
16 0.00053 ÿ36.3971
rules (Table 5) are derived directly based on the fact that
17 0.00035 ÿ40.4645
18 0.00031 ÿ40.6430 larger is the S/N ratio, the better is the performance char-
acteristic. By taking the max±min compositional operation
[12], the fuzzy reasoning of these rules yields a fuzzy output.
Supposing that x1 and x2 are the two input values of the fuzzy
In the following, the concept of fuzzy reasoning is described logic unit, the membership function of the output of fuzzy
brie¯y based on the two-input±one-output fuzzy logic unit. reasoning can be expressed as
The fuzzy rule base consists of a group of if±then control
rules with the two inputs, x1 and x2, and one output y, i.e. mC0 …y† ˆ …mA1 …x1 † ^ mB1 …x2 † ^ mC1 …y†† _ ::…mAn …x1 †
 Rule 1: if x1 is A1 and x2 is B1 then y is C1 else ^ mBn …x2 † ^ mCn …y†† (6)
 Rule 2: if x1 is A2 and x2 is B2 then y is C2 else where ^ is the minimum operation and _ is the maximum













































operation.















Finally, a defuzzi®cation method, called the center-of-
 Rule n: if x1 is An and x2 is Bn then y is Cn. gravity method [12], is adopted here to transform the fuzzy
Ai, Bi, and Ci are fuzzy subsets de®ned by the correspond- inference output mC0 into a non-fuzzy value y0, i.e.
P
ing membership functions, i.e., mAi , mBi and mCi . In this paper, ymC …y†
three fuzzy subsets are assigned in the two inputs (Figs. 2 y0 ˆ P 0 (7)
mC0 …y†
and 3). Seven fuzzy subsets are assigned in the output
(Fig. 4). Various degree of membership to the fuzzy sets In this paper, the non-fuzzy value y0 is called an MRPI.
is calculated based on the values of x1, x2, and y. Nine fuzzy Based on the above discussion, the larger is the MRPI, the

Fig. 3. Membership functions for the MRR.

 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55 53

Fig. 4. Membership functions for the MRPI.

better is the performance characteristic. Table 6 shows the variability of the multi-response performance indexes,
experimental results for the MRPI using the experimental which is measured by the sum of the squared deviations
combinations of Table 2. from the total mean of the MRPI, into contributions by each
Since the experimental design is orthogonal, it is then of the process parameter and the error. First, the total sum of
possible to separate out the effect of each machining para- the squared deviations SST from the total mean of the MRPI
meter at different levels. For example, the mean of the MRPI Zm can be calculated as
for the workpiece polarity at levels 1 and 2 can be calculated
X
p
by averaging the multi-response performance indexes for SST ˆ …Zj ÿ Zm †2 (8)
experiments 1±9 and 10±18, respectively (Table 2). The jˆ1
mean of the MRPI for each level of the other machining
parameters can be computed in a similar manner. The mean where p is the number of experiments in the orthogonal array
of the MRPI for each level of the machining parameters is and Z is the mean of the MRPI for the jth experiment.
summarized and called the MRPI table (Table 7). In addi- The total sum of the squared deviations SST is decom-
tion, the total mean of the MRPI for the eight experiments is posed into two sources: the sum of the squared deviations
also calculated and listed in Table 7. Fig. 5 shows the MRPI SSd due to each process parameter and the sum of the
graph, where the dashed line indicated in this ®gure is the squared error SSe. The percentage contribution by each of
value of the total mean of the MRPI. Basically, the larger is the process parameter in the total sum of the squared
the MRPI, the smaller is the variance of the performance deviations SST can be used to evaluate the importance of
characteristics around the desired value. However, the rela-
tive importance amongst the machining parameters for the Table 6
Results for the MRPI
multiple performance characteristics still needs to be known
so that the optimal combinations of the machining parameter No. MRPI
levels can be determined more accurately. 1 0.482
2 0.468
4.4. Analysis of variance 3 0.339
4 0.750
The purpose of the ANOVA is to investigate which 5 0.770
6 0.390
process parameters signi®cantly affect the performance 7 0.218
characteristics. This is accomplished by separating the total 8 0.713
9 0.863
Table 5 10 0.250
Fuzzy rule table 11 0.325
12 0.275
MRPI S/N for EWR 13 0.271
Small Middle Large 14 0.241
15 0.301
S/N for MRR Small Very small Small Middle 16 0.338
Middle Small Middle Large 17 0.309
Large Middle Large Very large 18 0.168
54 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55

Table 7
MRPI table

Symbol Machining parameter MRPI

Level 1 Level 2 Level 3 Max±min

A Workpiece polarity 0.522 0.253 0.269
B Pulse-on time 0.336 0.408 0.418 0.082
C Duty factor 0.384 0.390 0.387 0.006
D Open discharge voltage 0.473 0.314 0.375 0.159
E Discharge current 0.286 0.469 0.408 0.183
F Dielectric fluid 0.388 0.457 0.318 0.139
Mean value of MRPIˆ0.396 dB

the process-parameter change on the performance charac- 4.5. Con®rmation tests

teristics. In addition, the F-test named after Fisher [9] can
also be used to determine which process parameters have a Once the optimal level of the process parameters has been
signi®cant effect on the performance characteristic. Usually, selected, the ®nal step is to predict and verify the improve-
the change of the process parameter has a signi®cant effect ment of the performance characteristic using the optimal
on the performance characteristic when the F value is large. level of the process parameters. The estimated S/N ratio ^Z
The results of ANOVA (Table 8) indicate that workpiece using the optimal level of the process parameters can be
polarity, discharge current, and open discharge voltage are calculated as
the signi®cant machining parameters in affecting the multi- Xq
ple performance characteristics, with the workpiece polarity ^Z ˆ Zm ‡ …Zi ÿ Zm † (9)
being the most signi®cant. Based on the above discussion, iˆ1
the optimal machining parameters are the workpiece polar- where Zm is the total mean of the MRPI, Zi the mean of the
ity at level 1, pulse-on time at level 3, duty factor at level 2, MRPI at the optimal level, and q is the number of the process
open discharge voltage at level 1, discharge current at level 2 parameters that signi®cantly affect the multiple performance
and dielectric ¯uid at level 2. characteristics.

Fig. 5. MRPI graph.

 J.L. Lin et al. / Journal of Materials Processing Technology 102 (2000) 48±55 55

Table 8
Results of the ANOVA

Symbol Machining parameter Degrees of freedom Sum of squares Mean square F Contribution (%)

A Workpiece polarity 1 0.328 0.328 19.22 47.23

B Pulse-on time 2 0.024 0.012 0.71 3.48
C Duty factor 2 0.0001 0.00005 0.003 0.01
D Open discharge voltage 2 0.077 0.039 2.25 11.08
E Discharge current 2 0.105 0.063 3.07 15.11
F Dielectric fluid 2 0.058 0.029 1.70 8.34
Error 6 0.102 0.017 14.74
Total 17 0.693 100

Table 9
Results of the con®rmation experiment

Initial machining parameters Optimal machining parameters

Prediction Experiment

Setting level A1B1C2D2E2F2 A1B3C2D1E2F2 A1B3C2D1E2F2

ERR (%) 29.9 2.07
MRR (g/min) 0.00159 0.00383
S/N ratio 0.468 0.733 0.781
Improvement MRPIˆ0.313

Table 10 and MRR can be improved through this approach. An

S/N ratio for the initial machining parameters and the optimal machining experiment was conducted to con®rm this approach. As a
parameters
result, the optimization methodology developed in this study
EWR MRR is useful in improving multiple performance characteristics
Initial machining parameters (A1B3C2D3E2F1) 10.486 ÿ26.485 in the electrical discharge machining operation.
Optimal machining parameters (A1B3C2D1E2F2) 33.447 ÿ18.786
Improvement S/N ratio (dB) 22.961 7.699
References

[1] P.J. Ross, Taguchi Techniques for Quality Engineering, McGraw-

Based on Eq. (9), the estimated MRPI using the optimal Hill, New York, 1988.
[2] A. Bendell, J. Disney, W.A. Pridmore, Taguchi Methods: Applica-
machining parameters can then be obtained. Table 9 shows
tions in World Industry, IFS Publications, Bedford, UK, 1989.
the results of the con®rmation experiment using the optimal [3] G. Taguchi, Introduction to Quality Engineering, Asian Productivity
machining parameters. As shown in Table 9, the EWR is Organization, Tokyo, 1990.
decreased from 29.9 to 2.07% and the MRR is increased [4] E.A. Elsayed, A. Chen, Optimal levels of process parameters for
from 0.00159 to 0.00383 g/min. The comparison of the S/N products with multiple characteristics, Int. J. Prod. Res. 31 (5) (1993)
1117±1132.
ratios between the initial machining parameters and the
[5] Y.S. Tarng, W.H. Yang, Application of the Taguchi method to the
optimal machining parameters is also shown in Table 10. optimization of the submerged arc welding process, Mater. Manuf.
It is shown clearly that the EWR and the MRR are greatly Process. 13 (3) (1998) 455±467.
improved through this study. [6] L. Zadeh, Fuzzy sets, Inform. and Control 8 (1965) 338±353.
[7] J.A. McGeough, Advanced Methods of Machining, Chapman & Hall,
New York, 1988.
[8] D.C. Montgomery, Design and Analysis of Experiments, Wiley,
5. Conclusions Singapore, 1991.
[9] R.A. Fisher, Statistical Methods for Research Worker, Oliver &
The paper has presented the use of fuzzy logics to the Boyd, London, 1925.
Taguchi method for the optimization of the electrical dis- [10] M.S. Phadke, Quality Engineering Using Robust Design, Prentice-
Hall, Englewood Cliffs, NJ, 1989.
charge machining process with multiple performance char-
[11] Q.Y. Ming, L.Y. He, Power suspension dielectric ¯uid for EDM, J.
acteristics. A fuzzy reasoning of the multiple performance Mater. Process. Technol. 52 (1995) 44±54.
characteristics has been performed by the fuzzy logic unit. [12] H.J. Zimmermann, Fuzzy Set Theory and its Applications, Kluwer
As a result, the performance characteristics such as EWR Academic Publishers, London, 1985.