Research Journal of Applied Sciences, Engineering and Technology 4(10): 1295-1299, 2012
ISSN: 2040-7467
© Maxwell Scientific Organization, 2012
Submitted: December 09, 2011
Accepted: January 04, 2012
Published: May 15, 2012
The Investigation of EDM Parameters on Electrode Wear Ratio
Reza Atefi, Navid Javam, Ali Razmavar and Farhad Teimoori
Department of Mechanical Engineering, Dehaghan Branch, Islamic Azad University,
Dehaghan, Isfahan, Iran
Abstract: Electrical Discharge Machining (EDM) is a well-established machining option for manufacturing
geometrically complex or hard material parts that are extremely difficult-to-machine by conventional machining
processes. The non-contact machining technique has been continuously evolving from a mere tool and die
making process. In this study, the influence of different electro discharge machining parameters (current, pulse
on-time, pulse off-time, arc voltage) on the electrode wear ratio as a result of application copper electrode to
hot work steel DIN1.2344 has been investigated. Design of the experiment was chosen as full factorial.
Artificial neural network has been used to choose proper machining parameters and to reach certain electrode
wear ratio. Finally a hybrid model has been designed to reduce the artificial neural network errors. The
experiment results indicated a good performance of proposed method in optimization of such a complex and
non-linear problems.
Key words:
Artificial Neural Network (ANN), Electrical Discharge Machining (EDM), electrode wear ratio,
hybrid model
INTRODUCTION
The origin of Electro Discharge Machining (EDM)
dates back to 1770 when an English scientist Joseph
Priestly discovered the erosive effect of electrical
discharges. Pioneering work on electrical discharge
machining was carried out in 1943 during World War II
by two Russian scientists (B.R. and N.I. Lazarenko) at the
Moscow University (Kumar et al., 2009). Shaping of
materials for modern manufacturing industries with
stringent design requirements, such as high precision,
complex shapes, and high surface quality, is inevitable to
put them in use (Jain and Dixit, 2004). To achieve these
objectives, advanced machining processes are required
.Moreover, advanced machining techniques have
classified into four types (Jain, 2001). Mechanical,
thermal, chemical machining and electrochemical
machining, and biochemical machining processes. Among
these, EDM is a thermal process which has been widely
used to produce dies and molds (Abbas et al., 2007). This
high technology is developed in the late 1943s, which
supports about 7% of all machine tool sales in the world
(Moser, 2001). Its unique feature of using thermal energy
to machine electrically conductive parts regardless of
hardness has been its distinctive advantage in the
manufacture of mold, die, automotive, aerospace and
surgical components (Ho and Newman, 2003). Lee and Li
(2001) analyzed the effect of machining parameters on the
machining characteristics EDM of tungsten carbide. They
study the influence of different EDM parameters on the
surface roughness, Material Removal Rate (MRR) and
electrode wear ratio (EWR) (Lee and Li, 2001). Puertas et
al. (2004) analyzed the effective parameters on surface
roughness; MRR and EWR in EDM. They evaluate the
effect of current, pulse on-time and pulse off-time on
surface roughness, MRR and EWR on finishing stage.
They present proper second degree regression models for
predicting surface roughness, material removal rate and
electrode wear (Puertas et al., 2004). Salman and Kayacan
(2008) evaluated the effect of EDM parameters on surface
roughness, volumetric material removal rate and electrode
wear. They developed a mathematical model which based
on that they could predict surface roughness, MRR and
electrode wear by changing the pulse on-time, current and
pulse voltage (Salman and Kayacan, 2008).
In this study, the influence of different EDM
parameters (current, pulse on-time, pulse off-time, arc
voltage) on the EWR as a result of application copper
electrode to hot work steel DIN1.2344 has been
investigated. Design of the experiment was chosen as full
factorial. Artificial neural network has been used to
choose proper machining parameters and to reach certain
EWR. Finally a hybrid model has been designed to reduce
the artificial neural network errors. The experiment results
indicated a good performance of proposed method in
optimization of such a complex and non-linear problems.
Procedure: In this section, there will be a brief
description of the equipment and material used to carry
out the EDM experiments. Also, the design factors used
in this study will be outlined.
Corresponding Author: Reza Atefi, Department of Mechanical Engineering, Dehaghan branch, Islamic Azad University,
Dehaghan, Isfahan, Iran.
1295
Res. J. Appl. Sci. Eng. Technol., 4(10): 1295-1299, 2012
Table 1: Details of work piece, tool and dielectric fluid
Electrode
Workpiece
Dielectric fluid
Copper (electrolytic grade)
Hot Work Steel: (Kerosene)
DIN 1.2344
Dimension: cylindrical shape
Composition-C:
with a diameter of 10 mm
0.39%; Cr:5.15%;
(10 mm×10 mm×25 mm)
Mo: 1.25%; V: 1%;
Si: 1%; Mn: 1%;
rest iron
De(density of copper) is
Dimension: cylindrical
shapewith a diameter of
8.9/103 g/mm3
25 mm( 25 mm×25 mm×
5 mm)
Dw(density of steel) is
7.8/103 g/mm3
Table 2: Experimental machining setting
Current Gap voltage Pulse on- Pulse off(I)
(V)
time (ton) time (toff)
4, 6,
40, 60,
25, 50,
25, 50,
100 :s
8A
80 v
100 :s
Electrode
polarity
Positive
(+)
Parameters explained above used as experimental
variables and it defined the value of EWR. MRR and
EWR can be calculated by the following equations
(Habib, 2009):
MRR =
Ww
VEW =
We
(mm3 / min)
ρet
(2)
100 × VEW
MRR
(3)
ρw t
EWR =
Jet
flushing
pressure
25 kPa
Equipment used in the experiment:
Die-sinking EDM machine: Die-sinking EDM machine
used in this experiment was Roboform 40 manufactured
by Charmilles Technologies Machine. It has 4 axial
movements (linear movement in X, Y and Z axis and
rotational movement in Z axis). Movement resolution of
EDM machine was 0.5 microns.
Digital weighing machine: Digital weighing machine
(used for checking the weight of samples) was model 100
manufactured by GB Co., USA (precision of 0.01 g).
MATERIALS AND METHODS
A new set of instrument (electrode and workpiece)
for each experiment has been used. The machining state
has been shown in Table 1.
The purpose of doing the experiment was the
electrode wear ratio in EDM finishing stage of hot work
steel DIN1.2344 and presenting an appropriate ANN for
the prediction of EWR. As the aim of experiment was
evaluation of EWR in finishing stage, the work pieces
have been selected to be drilled 0.2mm deep in the
surface. The most important parameters in EDM are pulse
current (I), pulse voltage (V), pulse on-time (Ton) and
pulse off-time (Toff) (Lee and Li, 2001; Salman and
Kayacan, 2008). This study employed a full factorial
design because ANN model needed a lot of data to obtain
an appropriate model for EWR prediction. Pulse current
3 to 8 Ampere was selected for EDM finishing and as a
result, pulse currents 4, 6, 8A were used. Pulse voltages
40, 60, 80 v were used based on available pulse voltages
EDM machine. Pulse on-times 25, 50, 100 :s were used.
The pulse-off duration is equal to the pulse-on therefore
pulse off-times 25, 50, 100 :s were used. Therefore, in
this study, 81 experiments were done on Work pieces.
The Experimental machining setting has been shown in
Table 2.
(min3 / min)
(1)
VEW is the volumetric electrode wear, Ww is the
workpiece weight loss in g, We is the electrode weight
loss in g, T is the machining time in min, Ds is the density
of workpiece(steel) and DCu is the density of electrode
(copper).
RESULTS AND DISCUSSION
All of the 81 electrode wear ratio values measured as
a result of the EDM based on parameters such as the
discharge current, pulse on-time, pulse off-time and gap
voltage have been indicated in Table 3.
Design of the artificial neural network model: Artificial
Neural Network (ANN) has been designed for the
prediction of EWR. For designing and training of ANN
model, the programming in Matlab software was used.
Training procedures were as follow:
C
C
C
C
C
C
Defining the inputs and outputs of the network
Defining error function of the network
Obtaining the trained output data for input vector
data
Comparing real outputs with test outputs
Correcting ANN weights based on error value
Repeating "Correct ANN weights based on error
value" to reach minimum error
The input parameters considered in the experiments
include discharge current (I), voltage (V), pulse-on time
(Ton) and pulse-off time (Toff). The output parameter
considered in experiments includes surface roughness
(Ra). Architecture of ANN model is shown in Fig. 1.
Error function network used mean square error
(MSE) procedure as shown in the following equation
(Mandal et al., 2007):
1296
1
MSE =
2N
N
m
∑ ∑ (Tj − O j )
i =1 j =1
2
(4)
Res. J. Appl. Sci. Eng. Technol., 4(10): 1295-1299, 2012
Table 3: Resault of the EDM experiment
Toff
EWR Ton
No I(A) V(v) (%)
(:s)
(:s)
1
4
40
25
25
11.95
2
4
40
25
50
9.64
3
4
40
25
100
7.38
4
4
60
25
25
13.30
5
4
60
25
50
11.69
6
4
60
25
100
9.29
7
4
80
25
25
9.29
8
4
80
25
50
6.82
9
4
80
25
100
6.01
10 4
40
50
25
21.43
11 4
40
50
50
16.95
12 4
40
50
100
15.03
13 4
60
50
25
27.74
14 4
60
50
50
17.90
15 4
60
50
100
15.35
16 4
80
50
25
24.93
17 4
80
50
50
20.43
18 4
80
50
100
18.71
19 4
40
100
25
29.79
20 4
40
100
50
27.03
21 4
40
100
100
23.24
22 4
60
100
25
31.73
23 4
60
100
50
25.14
24 4
60
100
100
24.06
25 4
80
100
25
34.07
26 4
80
100
50
32.91
27 4
80
100
100
28.24
No
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
I(A)
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
V(v)
40
40
40
60
60
60
80
80
80
40
40
40
60
60
60
80
80
80
40
40
40
60
60
60
80
80
80
Ton
(:s)
25
25
25
25
25
25
25
25
25
25
50
50
50
50
50
50
50
50
100
100
100
100
100
100
100
100
100
Toff
(:s)
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
EWR
(%)
32.72
31.40
26.31
33.83
27.38
23.22
34.97
31.76
24.61
37.21
35.88
40.00
39.60
36.12
31.90
41.90
40.46
38.11
44.84
43.73
42.71
45.71
44.84
44.55
48.44
46.04
44.60
No
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
I(A)
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
V(v)
40
40
40
60
60
60
80
80
80
40
40
40
40
40
40
80
80
80
40
40
40
60
60
60
80
80
80
Ton
(:s)
25
25
25
25
25
25
25
25
25
25
50
50
50
50
50
50
50
50
100
100
100
100
100
100
100
100
100
Toff
(:s)
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
25
50
100
EWR
(%)
50.31
47.69
45.60
51.52
51.52
41.24
51.73
51.34
43.67
53.94
53.02
46.69
56.03
53.87
51.60
57.68
55.04
51.69
61.83
59.76
56.74
62.95
61.79
60.08
68.66
66.36
63.86
prediction vale% = (actual value - predicted value/
actual value)×100
(5)
Fig. 1: Architecture of ANN model
N is the all number of training pattern (definition of
epoch in Matlab programming), m is the number of output
nodes, Tj is the target output of the jth neuron and Oj the
estimated value of the jth neuron (Puertas et al., 2004).
Designing the ANN model for EWR value estimation:
The number of data is 81.consequently, 9 out of 81 were
chosen for testing of the network and 72 for training the
network. The number of neurons was selected in hidden
layers, transportation function of each neuron, error
training method based on minimum error. For testing the
prediction ability of the prediction error model in each
output, node has been calculated as follows (Mandal
et al., 2007):
In this situation Mean prediction error has a minimum
value and network architecture is in the best situation. The
choose of the number of neurons in hidden layers,
transportation function of each neuron, learning method
and training method was based on trial and error to obtain
minimum error. The designed ANN had 4 inputs, 31
neurons in first hidden layer, 31 neurons in second hidden
layer and 1 neuron in output layer. The training of
network used trainrp (back propagation) method. 0.3 is
used as the value of MSE.
The maximum, minimum and mean prediction error
with different architectures network for selection neurons
has been shown in Table 4.
For the reduction of ANN errors and precise
estimation of EWR, a hybrid model was used (a
combination of statistical method and neural network(.
For this reason, by doing a statistical analysis, values
removed with high residuals in Table 3 (NO.6, 13, 26,
39). We have the value of 77 EWR which 68 values were
used for network training and 9 values for network test.
The designed ANN had 4 inputs, 20 neurons in first
hidden layer, 21 neurons in second hidden layer and 1
neuron in output layer. The maximum, minimum and
mean prediction errors for this network are 6, 0.05 and
2%, respectively. Mean prediction error has been
calculated by taking the average of all the individual
errors, for all the testing patterns. The maximum,
1297
Res. J. Appl. Sci. Eng. Technol., 4(10): 1295-1299, 2012
Table 4: Different architectures network for ANN model
Network
Minimum
Serial no
architecture
prediction error (%)
1
4-24-24-1
0.4
2
4-25-25-1
0.8
3
4-26-26-1
3
4
4-27-27-1
3
5
4-28-28-1
0.4
6
4-29-29-1
4
7
4-30-30-1
0.01
8
4-31-31-1
1
9
4-32-32-1
2
Maximum
prediction error (%)
12
28
11
20
11
10
15
8
12
Mean prediction
error (%)
5.7
8
6
10
4.4
6.6
7
3.1
5.5
Table 5: Different architectures network for hybrid model
Network
Minimum
Serial no
architecture
prediction error (%)
1
4-15-15-1
1.00
2
4-16-16-1
0.20
3
4-17-17-1
8.00
4
4-18-18-1
0.10
5
4-19-19-1
0.08
6
4-20-20-1
0.05
7
4-21-21-1
0.50
8
4-22-22-1
1.00
Maximum
prediction error (%)
10
9
22
6
11
6
20
21
Mean
prediction error (%)
4.0
4.5
13
5.0
3.9
2.0
10
7.8
minimum and mean prediction error with different
architectures network for selection neurons has been
shown in Table 5.
Using hybrid model caused mean error reach to 2%
which showed 1.1% less error in compared to the
experiments that ANN was used. The results show good
performance of proposed model when we optimize such
a complex and non-linear problems.
CONCLUSION
In this study, the influence of different EDM
parameters (current, pulse on-time, pulse off-time, pulse
voltage) in finishing stage on the surface quality (Ra) as a
result of application copper electrode to a work piece (hot
work steel DIN1.2344) has been investigated. ANN has
been designed for the prediction of EWR in finishing
stage of hot work steel DIN1.2344. Finally for reducing
the error in ANN, a hybrid model (a combination of
statistical analysis and ANN model) has been designed
and following results has been obtained:
C
C
C
C
Application of ANN to predict EWR is a scientific
method which makes industries free from complex
traditional trial and error methods.
By using ANN and correct training of it, without
doing any test, we can precisely predict EWR by
changing current, pulse on-time, pulse off-time and
arc voltage.
Designed ANN has mean error of 3.1% and
maximum error of 8%.
By using a hybrid model, mean error of ANN had
reduced to 1.1% and reached to 2%.
The results show good performance of proposed
method in optimization of complex and non-linear
problems. This error level shows a good precision for
EWR.
REFERENCES
Abbas, N.M., D.G. Solomon and M.F. Bahari, 2007. A
review on current research trends in Electrical
Discharge Machining (EDM). Int. J. Mach. Tool.
Manuf., 47: 1214-1228.
Habib, S., 2009. Study of the parameters in electrical
discharge machining through Response surface
methodology approach. Appl. Math.l Mode., 33:
4397-4407.
Ho, K.H. and S.T. Newman, 2003. State of the art
Electrical Discharge Machining (EDM). Int. J. Mach.
Tool. Manuf., 43: 1287-1300.
Jain, V.K., 2001. Advanced Machining Processes. Allied
Publisher, Bombay.
Jain, P.M. and Dixit, 2004. Parametric study of
temperature distribution in electrodischarge diamond
grinding. Mater. Manuf. Process., 19: 1071-1101.
Kumar, S., R. Singh, T.P. Singh and B.L. Sethi, 2009.
Surface modification by electrical discharge
machining: A review. J. Mater. Proc. Technol.,
209(8): 3675-3687.
Lee, S.H. and X.P. Li, 2001. Study of the effect of
machining parameters on the machining
characteristics electrical discharge machining of
tungsten carbide. J. Mater. Proc. Technol., 115:
344-358.
1298
Res. J. Appl. Sci. Eng. Technol., 4(10): 1295-1299, 2012
Mandal, D., S.K. Pal and P. Saha, 2007. Modeling of
electrical discharge machining process using back
propagation neural network and multi-objective
optimization using non-dominating sorting genetic
algorithm-II. J. Mater. Proc. Technol., 186: 154-162.
Moser, H., 2001. Growth industries rely on EDM. Manuf.
Eng., 127: 62-68.
Puertas, I., C.J. Luis and L. Álvarez, 2004. Analysis of
the influence of EDM parameters on surface quality,
MRR and EW of WC-Co. J. Mater. Proc. Technol.,
154: 1026-1032.
Salman, O.Z. and M.C. Kayacan, 2008. Evolutionary
programming method for modeling the EDM
parameters for roughness. J. Mater. Proc. Technol.,
200: 347-355.
1299