Proceedings of the Institution of Mechanical

Engineers, Part B: Journal of Engineering

Manufacture
http://pib.sagepub.com/

Multi-objective optimization of electrochemical machining process parameters using a particle swarm

optimization algorithm
R V Rao, P J Pawar and R Shankar
Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture 2008 222: 949
DOI: 10.1243/09544054JEM1158

The online version of this article can be found at:

http://pib.sagepub.com/content/222/8/949

Published by:

http://www.sagepublications.com

On behalf of:

Institution of Mechanical Engineers

Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering
Manufacture can be found at:

Email Alerts: http://pib.sagepub.com/cgi/alerts

Subscriptions: http://pib.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://pib.sagepub.com/content/222/8/949.refs.html

>> Version of Record - Aug 1, 2008

What is This?

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

 949

Multi-objective optimization of electrochemical

machining process parameters using a particle
swarm optimization algorithm
R V Rao1*, P J Pawar2, and R Shankar3
1
Department of Mechanical Engineering, National Institute of Technology, Gujarat, India
2
Department of Mechanical Engineering, K K Wagh Institute of Engineering Education and Research, Maharashtra,
India
3
School of Management, Asian Institute of Technology, Pathumthani, Thailand

The manuscript was received on 3 March 2008 and was accepted after revision for publication on 1 May 2008.

DOI: 10.1243/09544054JEM1158

Abstract: The selection of optimum values of important process parameters of electrochemical

machining processes such as the tool feed rate, electrolyte flow velocity, and applied voltage play
a significant role in optimizing the measures of process performance. These performance mea-
sures generally include dimensional accuracy, tool life, material removal rate, and machining
cost. In this paper, a particle swarm optimization algorithm is presented to find the optimal com-
bination of process parameters for an electrochemical machining process. The objectives consid-
ered are dimensional accuracy, tool life, and the material removal rate subjected to the constraints
of temperature, choking, and passivity. Both single- and multi-objective optimization aspects are
considered. The results of the proposed algorithm are compared with the previously published
results obtained by using other optimization techniques.

Keywords: electrochemical machining, multi-objective optimization, particle swarm optimi-

zation

1 INTRODUCTION 5. Temperature rise and residual stresses in the

workpiece are not desirable or acceptable.
Traditional machining processes, such as turning,
These requirements have led to the development
grinding, drilling, milling, etc., remove material by
of chemical, electrochemical, thermal, electrothermal,
chip formation, abrasion, or microchipping. There
mechanical, and other means of material removal.
are situations, however, where these processes are
Beginning in the 1940s, these advanced methods
not satisfactory, economical, or even possible, for
are called non-traditional or unconventional machin-
the following reasons [1]: ing processes. Over the last four decades, there has
1. The hardness and strength of the material is very been a large increase in the number of non-traditional
high (typically above 400 HB) or the material is machining processes (NTMP). Today, non-traditional
too brittle. machining processes with vastly different capabilities
2. The workpiece is too flexible, slender, or delicate and specifications are available for a wide range of
to withstand the cutting or grinding forces, or the applications. These processes are classified according
parts are too difficult to fix. to the nature of energy employed in machining, as dis-
3. The shape of the part is complex. cussed below:
4. Surface finish and dimensional tolerance require-
ments are more rigorous than those obtained by (a) chemical and electrochemical processes like
other processes. chemical milling, electrochemical machining,
electrochemical grinding, electrochemical hon-
*Corresponding author: Department of Mechanical Engineer- ing, etc.;
ing, S V National Institute of Technology, Ichchanath, Surat, (b) thermal and electrothermal processes like elec-
Gujarat 395007, India. email: [email protected] tric discharge machining, laser beam machining,

JEM1158
IMechE 2008 Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

950 R V Rao, P J Pawar, and R Shankar

plasma arc machining, electron beam machin- ECM process. A typical gap width then can be about
ing, ion beam machining, etc.; 0.4 mm.
(c) mechanical processes like ultrasonic machining, The ECM process can handle a large variety of mate-
abrasive jet machining, water jet machining, etc.; rials, limited only by their electrochemical properties
(d) hybrid processes like electrochemical discharge and not by their strength. This process is characterized
grinding, abrasive electrical discharge machin- by high metal removal rates for high-strength and dif-
ing, vibration-assisted electrochemical machin- ficult-to-machine alloys. Fragile parts that are not
ing, etc. easily machinable can be shaped by the ECM process.
Certain characteristics of the ECM process, such as the
Among various advanced machining processes ability to machine three-dimensional curved surfaces
mentioned above, the electrochemical machining without the striation marks, stress-free and burr-free
(ECM) process is one of the most highly developed machining, no thermal damage to the workpiece, and
processes and the present study is mainly focused ideally no tool wear, etc., make this process widely
on this process. The basis of the ECM process is applicable.
the phenomenon of electrolysis, whose laws were However, the main limitation of the ECM process is
established by Faraday in 1833. The principle and the high initial investment along with high power
equipment used in the ECM process are illustrated consumption and large floor space requirement.
in Fig. 1 [2]. The workpiece and tool are the anode Therefore use of this process is a costly affair. This
and cathode, respectively, of an electrolytic cell, problem is further compounded by the corrosion,
and a constant potential difference (usually about toxicity, and safety-related problems of the electro-
5–30 V) is applied across them, producing a high chemical machining process. Also, electrochemical
current density of 10–200 A/cm2. A suitable electro- machining is a complex process and it is difficult to
lyte (NaCl or NaNO3 aqueous solution) is chosen predict the changes that may occur in the interelec-
so that the cathode shape remains unchanged during trode gap. The electrolyte properties vary due to the
electrolysis. The electrolyte is pumped at a rate of emission of a considerable amount of heat and gas
3–60 m/s, through the gap between the electrodes, bubbles. In addition, hydrodynamic parameters,
to remove the machining waste (i.e. dissolved mate- such as pressure, also vary along the electrolyte flow
rial, usually metal hydroxide) and to diminish direction and make the analysis quite complicated.
unwanted effects such as those that arise with catho- It is therefore essential to make a careful decision
dic gas generation and electrical heating. The rate at during process planning before using electrochemi-
which metal is then removed from the anode is cal machining for practical purposes. A human pro-
approximately in inverse proportion to the distance cess planner selects proper machining process
between the electrodes. As machining proceeds, parameters using his or her own experience or from
and with the simultaneous movement of the cathode the handbooks. However, these parameters do not
at a typical rate, e.g. 0.02 mm/s towards the anode, give an optimal result. The selection of optimum pro-
the gap width along the electrode length will gradu- cess parameters plays a significant role in improving
ally tend to reach a steady state value. Under these the process performance and process economics by
conditions, a shape that is approximately a negative reducing various costs.
mirror image of the cathode will be reproduced on The next section presents a brief review of the
the anode as the cathode does not alter during the past research work done on the optimization of
ECM process parameters.

2 REVIEW OF PAST RESEARCH WORK

Bhattacharyya et al. [3] proposed a two-dimensional

interelectrode gap model in which maximization of
the metal removal rate was considered as the objec-
tive function with the tool feed rate and electrolyte
flow velocity as the design variables. The three con-
straints considered were temperature, passivity, and
choking. However, the authors had considered only
a single-objective optimization problem and solved
the same using a graphical solution technique,
which, in itself, was less accurate. This model was
Fig. 1 ECM principle and equipment [2] also based on many simplified assumptions, such as

Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture JEM1158
IMechE 2008

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

 Multi-objective optimization of electrochemical machining process parameters 951

the constant void fraction, electrolyte conductivity as objective functions and constraints. The optimization
a function of the void fraction only, and constant problem was then solved using a genetic algorithm.
electrolyte pressure along its flow path. Furthermore, However, the authors had considered only a single-
no variable bounds were used. objective optimization problem, i.e. to minimize
El-Dardery [4] proposed a cost model of the ECM the dimensional inaccuracy. Also the passivity con-
process considering various costs involved in the straint was violated in their approach. Furthermore,
process. The cost equation was arranged in terms the genetic algorithm has its own limitations, such
of decision variables, namely feed rate, electrolyte as the risk of replacement of a good parent string
flowrate, and voltage. The optimum values of the with the deteriorated child, less convergence speed,
decision variables were obtained by partial differ- and difficulty in selecting the controlling para-
entiation of the cost equation with respect to the meters such as population size, crossover rate, and
decision variables. However, as no constraints were mutation rate.
considered in this model, the values of decision It is observed from the review of past work that
variables obtained were not practical. the graphical solution technique and mathematical
Hewidy et al. [5] analysed the components of ECM programming techniques like goal programming,
cost (such as costs of power consumption, machin- partial differentiation, etc., had been used to solve
ing, electrolyte, and labour) with the objective to the problem of optimization of process parameters
set out the basic principles for selecting a suitable of electrochemical machining. However, these tradi-
electrochemical machine to meet the local produc- tional methods of optimization do not fare well over
tion requirements of a company. The authors men- a broad spectrum of problem domains. Moreover,
tioned the impossibility of having a generalized traditional techniques are not robust. Due to the
model for this purpose. In another work, Hewidy complex nature of the optimization problem, these
et al. [6] modelled the performance of ECM assisted techniques are not ideal for solving these problems,
by low-frequency vibrations using an analytical as they tend to obtain a local optimal solution. To
approach. overcome the drawbacks of traditional optimization
Acharya et al. [7] considered the multi-objective techniques, researchers are now using evolutionary
optimization model for the ECM process with maxi- optimization techniques. Evolutionary computation
mization of the material removal rate, minimization consists of a variety of methods including optimiza-
of dimensional inaccuracy, and maximization of tion paradigms that are based on evolution mechan-
tool life as three conflicting objectives. The decision isms such as biological genetics and natural
variables were the tool feed rate, electrolyte flow selection. These methods use the fitness information
velocity, and applied voltage. The constraints used instead of the functional derivatives, making them
in this model were temperature constraint, passivity more robust and effective. The most commonly
constraint, and choking constraint. The optimi- used non-traditional optimization technique is the
zation problem was solved by goal programming genetic algorithm. However, this method provides a
after linearizing the objective functions and con- near-optimal solution for a complex problem having
straint equations by regression analysis. This model large numbers of variables and constraints. This is
overcame the limitations of the model proposed by mainly due to difficulty in determining optimum
Bhattacharyya et al. [3]. However, it did not include controlling parameters. Therefore efforts are conti-
the variable bounds for feed rate and differences in nuing to use more recent optimization algorithms,
the interelectrode gap. which are more powerful, robust, and able to provide
The drawbacks of the model proposed by Acharya accurate solution. This paper is intended to apply
et al. [7] were overcome by Choobineh and Jain [8]. one of such recently developed optimization algo-
The authors had considered only two objective func- rithms, known as particle swarm optimization, for
tions, i.e. maximization of the material removal rate optimization of process parameters of electrochemi-
and maximization of dimensional accuracy. The third cal machining.
objective to maximize the tool life was eliminated as In the present work, an effort is made to verify
tool life is overachieved in most practical cases. whether any improvement in the solution is possible
They used the vertex method to find appropriate dis- by employing some other recent optimization techni-
tribution of the objective functions. The modified ques such as particle swarm optimization to the same
goal-programming problem was then solved in the optimization model. Particle swarm optimization
same manner as in Acharya et al. [7]. (PSO) is reported to be the better algorithm for con-
Jain and Jain [9] formulated the optimization tinuous optimization as well as discrete optimization
model based on the analysis given in Acharya et al. problems [10–12]. Hence, PSO is considered in this
[7] with certain modifications, i.e. expanding the work for single-objective optimization and multi-
variable bound ranges for the tool feed rate and objective optimization of electrochemical machining
electrolyte flow velocity but without linearizing the process parameters. The optimization model given in

JEM1158
IMechE 2008 Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

952 R V Rao, P J Pawar, and R Shankar

Acharya et al. [7] is also considered by expanding the 3.1.3 Maximization of tool life
variable bound ranges for the tool feed rate and elec-
Maximization of tool life is ensured by minimiz-
trolyte flow velocity.
ing the number of sparks per cm as given by the
equation
3 OPTIMIZATION MODEL OF f2 f
ELECTROCHEMICAL MACHINING Nmin ¼ a þ bEi þc ð4Þ
VU V

Formulation of the optimization model is the most where a, b, and c are constants and
important task in the optimization process. It invol-

ves identifying decision variables to be optimized, Aa r2w Zw F
Ei ¼ 1000 · ð5Þ
expressing the objective functions and constraints B Ki Mw hi
as functions of decision variables, setting up the
bounds for decision variables, and finally expressing 3.2 Constraints
the optimization problem as a mathematical model The following three constraints are considered in this
in a standard format that can be directly solved by optimization model.
the optimization algorithm. The optimization model
for the electrochemical machining process is formu-
lated in the present work based on the analysis 3.2.1 Temperature constraint
given by Acharya et al. [7]. The three decision vari- To avoid boiling the electrolyte, the electrolyte tem-
ables considered for this model are tool feed rate perature at the outlet should be less than the electro-
f (mm/s), electrolyte flow velocity U(cm/s), and applied lyte boiling temperature. Mathematically this can be
voltage V. expressed as
2 !1=2 3
3.1 Objectives 1 1 þ Sk f 2
Ti
41
 n 56Tb ð6Þ
The various objectives under consideration are a 1
a0max U
described below.
where
3.1.1 Maximization of the material removal rate
2ag2 L
This is the product of projected area and tool feed Sk ¼ ð7Þ
Ki re Ce Jcn
rate. Maximization of the tool feed rate would maxi-
mize the material removal rate (MRR) since the pro-
Zw Frw
jected area is constant. Thus g¼ ð8Þ
Mw h i
MRRmax ¼ fmax ð1Þ
3.2.2 Passivity constraint
3.1.2 Maximization of dimensional accuracy
Oxygen evolved during electrochemical machining
ECM is perhaps the only machining process that does forms an oxide film, which is the root cause of
not allow the workpiece dimensions to be checked in passivity. To avoid passivity, the thickness of the
the course of machining. Although few techniques oxygen gas bubble layer must be greater than the
such as ultrasonic measurement of the interelectrode passive layer thickness. Mathematically, this can be
gap can be used [13], it is necessary to predetermine expressed as
the control parameters to ensure the desired dimen-
sional accuracy. Dimensional accuracy depends f ðTo þ 273Þ
upon the difference in the interelectrode gap from Gt >1 ð9Þ
Ua0max
inlet Yi to outlet Yo, which is given by

 where
Ko K i M w hi V
Yo
Yi ¼
1 ð2Þ
Ki rw Zw Ff Gt ¼
Rrf Rf Lg
ð10Þ
P0 t p i
Kx  0
n
3.2.3 Choking constraint
¼ 1
ax ½1 þ aðTx
Ti Þ ð3Þ
Ki
Hydrogen evolved at the cathode during the ECM
with Ko ¼ Kx at the outlet. The objective of process can choke the electrolyte flow. To avoid
maximizing the dimensional accuracy is attained by choking the electrolyte flow, the maximum thickness
minimizing ðYo
Yi Þ of the hydrogen bubble layer should be less than the

Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture JEM1158
IMechE 2008

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

 Multi-objective optimization of electrochemical machining process parameters 953

equilibrium interelectrode gap. Mathematically, it The acceleration constants cl and c2 in equation

can be expressed as (13) represent the weighting of the stochastic accel-
eration terms that pull each particle towards ‘pBest’
Ht f 2 ðTo þ 273Þ and ‘gBest’ positions, where c1 represents the confi-
 n 61
VUa0max 1
a0max ½1 þ aðTo
Ti Þ ð11Þ dence the particle has in itself (cognitive parameter)
and c2 represents the confidence the particle has in
a swarm (social parameter). Thus, adjustment of
where
these constants changes the amount of tension in
Mh RLg2 the system. Low values of the constants allow parti-
Ht ¼ ð12Þ cles to roam far from target regions before being
Zh Po FKi
tugged back, while high values result in abrupt move-
The next section briefly describes the PSO algorithm. ment towards, or through target regions [12]. The
inertia weight w plays an important role in the PSO
convergence behaviour since it is employed to con-
4 PARTICLE SWARM OPTIMIZATION (PSO) trol the exploration abilities of the swarm. The large
inertia weights allow wide velocity updates to explore
Particle swarm optimization (PSO) is an evolutionary the design space globally while small inertia weights
computation technique developed by Kennedy and concentrate the velocity updates to nearby regions
Eberhart [10]. It exhibits common evolutionary com- of the design space. The optimum use of the inertia
putation attributes, including initialization with a weight w provides improved performance in a num-
population of random solutions and searching for ber of applications. The effect of w, c1, and c2 on
optima by updating generations. Potential solutions, convergence for standard numerical benchmark
called particles, are then ‘flown’ through the problem functions is provided by Bergh and Engelbrecht [14].
space by following the current optimum particles. Particle velocities on each dimension are confined
The particle swarm concept was originated as a simu- to a maximum velocity parameter Vmax, specified by
lation of a simplified social system. The original the user. If the sum of accelerations cause the velo-
intent was to simulate the graceful but unpredictable city on that dimension to exceed Vmax, then the velo-
choreography of a bird flock graphically. Each parti- city on that dimension is limited to Vmax.
cle keeps track of its coordinates in the problem Although the heuristics are developed to deter-
space, which are associated with the best solution mine the inertia weights and acceleration constants
(fitness) it has achieved so far. This value is called for guaranteed convergent trajectories, it is mainly
‘pBest’. Another ‘best’ value that is tracked by the applicable to single-objective optimization. It is very
global version of the PSO is the overall best value difficult to obtain the values of inertia weights and
and its location obtained so far by any particle in acceleration constants for multi-objective optimiza-
the population. This location is called ‘gBest’. The tion problems, due to the inherent conflicting nature
PSO concept consists of, at each step, changing the of objectives to be optimized. To overcome this pro-
velocity (i.e. accelerating) of each particle towards blem a time variant PSO was described by Tripathi et
its ‘pBest’ and ‘gBest’ locations (global version of al. [15]. The proposed algorithm was made adaptive
PSO). Acceleration is weighted by a random term in nature by allowing its vital parameters, i.e. inertia
with separate random numbers being generated for weights and acceleration constants, to change with
acceleration towards ‘pBest’ and ‘gBest’ locations. iterations. This adaptiveness helps the algorithm to
The updates of the particles are accomplished as explore the search space more efficiently. A mutation
per the following equations operator was also included to overcome the premature
  convergence. The performance of the algorithm was
Viþ1 ¼ wVi þ c1 r1 ðpBesti
Xi Þ þ c2 r2 gBesti
Xi then measured with respect to the main four perfor-
ð13Þ mance measures, i.e. convergence rate, diversity,
purity, and minimal spacing.
Unlike a genetic algorithm, the PSO algorithm does
Xiþ1 ¼ Xi þ Viþ1 ð14Þ
not need a complex encoding and decoding process
Equation (13) calculates a new velocity (Vi+1) for each or a special genetic operator. PSO takes the real num-
particle (potential solution) based on its previous ber as a particle in the aspect of the representation
velocity, the best location it has achieved (‘pBest’) so solution and the particles update themselves with
far, and the global best location (‘gBest’) the population internal velocity. In this algorithm, the evolution
has achieved. Equation (14) updates an individual looks only for the best solution and all particles
particle’s position (Xi) in the solution hyperspace. tend to converge to the best solution. In the imple-
The two random numbers r1 and r2 in equation (13) mentation process, particles randomly generated
are independently generated in the range [0,1]. at the beginning or generated by internal velocity

JEM1158
IMechE 2008 Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

954 R V Rao, P J Pawar, and R Shankar

during the evolutionary process usually violate the Constraints:

system constraints, resulting in infeasible particles.
(a) Temperature constraint:
Therefore, the handling of system constraints, parti-
cularly non-linear equation constraints and the mea-  
surement and evaluation of infeasible particles, is 1
f 2:133007 U
1:088937 V
0:351436 e0:321968 > 0
very important. To cope with constrained problems ð16Þ
with evolutionary computation, various approaches
such as rejection of infeasible individuals, repair of (b) Passivity constraint:
infeasible individuals, replacement of individuals by 
0:844369
2:526076 1:546257 12:57697 
their repaired versions, and penalty function meth- f U V e
1> 0
ods can be adopted. Among them, the penalty func- ð17Þ
tion methods are particularly promising [12], as
evidenced by recent developments. (c) Choking constraint:
The next section presents two examples consider-
ing single-objective and multi-objective optimiza-  
1
f 0:075213 U
2:488362 V 0:240542 e11:75651 > 0 ð18Þ
tion to demonstrate and validate the proposed
PSO algorithm with constant values of inertia weight Parameter bounds:
and acceleration coefficient. Variable inertia weight
8 6 f 6 200 (mm/s)
and acceleration coefficients may be applied to the
300 6 U 6 5000 (cm/s)
multi-objective optimization problem [15]. However,
the multi-objective optimization problem considered 3 6 V 6 21 (V)
in the present work is limited to three objectives sub- Now, the proposed PSO algorithm is applied
jected to only three constraints; hence the constant to solve the above optimization problem. The follow-
values of inertia weight and acceleration coefficients ing parameters of optimization are selected after
are used. The values of inertia weight and accelera- various trials:
tion coefficients, for which the algorithm shows bet-
(a) maximum number of iterations: 50;
ter performance in terms of the convergence rate,
(b) inertia weight factor (w): 0.65;
are obtained through several trials with an initial
(c) acceleration coefficients: c1 ¼ 1.65 and c2 ¼ 1.75.
guess as given by Bergh and Engelbrecht [14].
Table 2 shows the optimum process parameter
data for Example 1, along with the previously pub-
lished results using other methods. The optimum pro-
5 EXAMPLES
cess parameter values are f ¼ 8 mm/s, U ¼ 300 cm/s,
and V ¼ 9.835 V. It is observed from the results that
To demonstrate and validate the proposed PSO
the solution obtained by PSO gives a significantly
algorithm, two examples are considered for the
smaller value of dimensional inaccuracy as com-
optimization of electrochemical machining process
pared to that of Acharya et al. [7], Choobineh and
parameters, based on the model given by Acharya
Jain [8], and Jain and Jain [9] when applied to the
et al. [7].
model of Acharya et al.[7]. This improvement is
mainly due to the use of a better optimization techni-
5.1 Example 1 que, PSO. It is also observed that the results obtained
by using a genetic algorithm [9] violate the passivity
This example presents the single-objective optimi-
constraint when applied to the model as Acharya
zation case for minimization of dimensional inaccu-
et al. [7]. Jain and Jain [9] had used the same model
racy subjected to the constraints of temperature,
as Acharya et al. [7] for optimization of ECM process
passivity, and choking. Data for this example
parameters. Even though Jain and Jain [9] had men-
are the same as those considered by Acharya et al.
tioned that all constraints were satisfied, when the
[7] and are given in Table 1. On substituting the
values obtained by them are substituted in the con-
values given in Table 1 in equations (1) to (12),
straint equations, the passivity constraint becomes
the following objective function and constraints are
violated. The same is the case for the value of dimen-
formulated:
sional inaccuracy (Z). Jain and Jain [9] mentioned
that they had obtained an optimum value of Z equal
Objective function:
to 7.4633. However, by putting the optimum values
Minimize Z ¼ f 0:381067 U
0:372623 V 3:155414 e
3:128926 of f, U, and V obtained by Jain and Jain [9] using the
ð15Þ genetic algorithm in the model proposed by Acharya
et al. [7], the value of Z turns out to be 33.62 with
where, Z ¼ dimensional inaccuracy (mm) violation of the passivity constraint.

Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture JEM1158
IMechE 2008

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

 Multi-objective optimization of electrochemical machining process parameters 955

Table 1 Values of the constants and decision variables used in the single-objective optimization problem

Notation Details Unit Value

a, b, and c Constants used in the tool life equation
2.05,
0.325, and 26.78
Aa Projected area cm2
B Width of workpiece cm
Ce Specific heat of electrolyte cal/g  C 0.997
f Feed rate mm/s
fmax Maximum feed rate mm/s
F Faraday’s constant coulombs 96500
i Ionic current density A/cm2 1.25
Jcn Joule’s constant J/cal 4.186
Ki Electrical conductivity of electrolyte at inlet S/cm 0.3333
Ko Electrical conductivity of electrolyte at outlet S/cm
Kx Electrical conductivity of electrolyte at a distance S/cm
x from inlet
L Length of workpiece cm 3
Mh Atomic weight of hydrogen g 1
Mw Atomic weight of workpiece g 56
n Exponent
Nmin Minimum number of sparks per cm
P Pressure of electrolyte MPa
Po Pressure of electrolyte at outlet MPa
R Gas constant g cm/g K 4.203 · 104
Rf Roughness factor 1.25
tp Time taken for film formation s 60

Tb Permitted electrolyte temperature C 65

Ti Room temperature C 27

To Electrolyte temperature at outlet C

Tx Electrolyte temperature at a distance x from inlet C
U Electrolyte flow velocity cm/s
V Voltage V
Yi Interelectrode gap at inlet mm 0.0002
Yo Interelectrode gap at outlet mm
Zh Valency of hydrogen 2
Zw Valency of workpiece 2
a Temperature coefficient of electrolyte conductivity / C 0.02
a0 Void fraction
a0 max Maximum void fraction 0.7
hi Current efficiency 0.95
re Density of electrolyte g/cm 3
1
rf Passive film density g/cm3 0.042
rh Density of hydrogen g/cm3 7.86
rw Density of workpiece g/cm3 7.86
s Slip ratio between electrolyte and hydrogen gas 1

Table 2 Results of single-objective optimization

Method Author(s) f(mm/s) U(cm/s) V(V) TC PC CC Z

GP Acharya et al. [7] 18.96 179 15 0.001 2.422 0.204 100

Fuzzy sets Choobineh and Jain [8] 12.75 400 21 0.841 0.0559 0.886 181.07
GA* Jain and Jain [9] 8 2978.45 16.5 0.992
0.993 0.999 33.62
PSO 8 300 9.835 0.895 0.001 0.810 15.452

TC ¼ value of temperature constraint; PC ¼ value of passivity constraint; CC ¼ value of choking constraint; Z ¼ dimensional inaccuracy
(single-objective); GP ¼ goal programming.
* For comparison purposes, this result is obtained by putting the optimum values of f, U, and V obtained by Jain and Jain [9] using a genetic
algorithm (GA) in the model proposed by Acharya et al. [7].

Optimality of the above-mentioned solution can be value of tool feed rate is selected. Hence the tool feed
confirmed from Figs 2 to 4. As shown in Fig. 2, the rate at the lower bound ( f ¼ 8 mm/s) is selected.
dimensional inaccuracy increases with the tool feed The variation of electrolyte flow velocity is shown
rate. Therefore, the smallest possible value of the tool in Fig. 3. As the dimensional inaccuracy decreases
feed rate will minimize the dimensional inaccuracy. with an increase in the electrolyte flow velocity,
Also the passivity constraint will be violated if a higher selection of a higher value of electrolyte flow

JEM1158
IMechE 2008 Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

956 R V Rao, P J Pawar, and R Shankar

Dimensional
30 inaccuracy

Dimensional Inaccuracy
25
20 Choking
constraint
15
10 Temperature

(␮m)
5 constraint
0 Passivity
-5 0 10 20 30 constraint
-10
-15
Tool feed rate (␮m/s)

Fig. 2 Variation of the objective function and the constraints with the tool feed rate

20 Dimensional
Dimensional Inaccuracy

15 inaccuracy
10
(microns)

Choking
5 constraint
0
Temperature
-5 0 500 1000 1500 constraint
-10
Passivity
-15 constraint
Electrolyte Flow velocity (cm/s)

Fig. 3 Variation of the objective function and the constraints with the electrolyte flowrate

200 Dimensional
inaccuracy
Dimensional Inaccuracy

150
Choking
constraint
100
(␮m)

Temperature
50 constraint

Passivity
0 constraint
0 5 10 15 20 25
-50
Applied voltage (V)

Fig. 4 Variation of the objective function and the constraints with the applied voltage

velocity is desirable. However, the value of electrolyte selection of a lower value of applied voltage is desir-
flow velocity at the lower bound (U ¼ 300 cm/s) is able. However, to ensure the non-negativity of the
obtained, for at any higher value than this the passiv- passivity constraint, the value of voltage (V) equal to
ity constraint is violated. This may be the reason 9.835 V is selected.
for violation of the passivity constraint for the
solution obtained by using the genetic algorithm [9]
5.2 Example 2
when used in the optimization model of Acharya
et al. [7]. This example presents a multi-objective optimization
Figure 4 shows that the dimensional inaccuracy case considering all three objectives, namely material
increases with an increase in voltage. Hence, removal rate, tool life, and dimensional inaccuracy

Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture JEM1158
IMechE 2008

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

 Multi-objective optimization of electrochemical machining process parameters 957

Table 3 Results of multi-objective optimization

Method f(mm/s) U(cm/s) V(V) TC PC CC Z1 Z2 Z3 Z

GP 18.96 179 15 0.001 2.422 0.204 100 51.79 18.96 18.22

Fuzzy set 12.75 400 21 0.841 0.0559 0.886 181.1 5.47 12.75 5.47
GA* 8 2978.45 16.5 0.992
0.993 0.999 33.62 1.94 8 1.23
PSO 8 300 13.225 0.905 0.583 0.799 39.34 3.39 8 1.811

Z ¼ normalized combined objective function.

* For comparison purposes, this result is obtained by putting the optimum values of f, U, and V obtained by Jain and Jain [9] using a genetic
algorithm (GA) in the model proposed by Acharya et al. [7].

subjected to the constraints of temperature, passivity, w1, w2, and w3 ¼ weightages assigned to the objec-
and choking. tive functions Z1, Z2, and Z3 respectively.
The first objective is to minimize the dimensional The values of weightages can be calculated by using
inaccuracy as given by the following expression the analytic hierarchy process [16]. However, in the
present example, equal weightages are assumed.
Z1 ¼ f 0:381067 U
0:372623 V 3:155414 e
3:128926 ð19Þ The following parameters of optimization were
selected after various trials:
where Z1 ¼ dimensional inaccuracy (mm).
The second objective is to maximize the tool life by (a) maximum number of iterations: 50;
minimizing the number of sparks per millimetre, (b) inertia weight factor (w): 0.65;
which is given by the following expression (c) acceleration coefficients: c1 ¼ 1.65 and c2 ¼ 1.75.

Z2 ¼ f 3:528345 U 0:000742 V
2:52255 e0:391436 ð20Þ The results of multi-objective optimizations along
with the previously published results using other meth-
where Z2 ¼ number of sparks per millimetre. ods are as given in Table 3. It is observed from the results
The third objective is to maximize the material that the combined objective function (Z) obtained
removal rate, which is given by the following expression by PSO shows substantial improvement over Acharya
Z3 ¼ f (21) et al. [7] and Choobineh and Jain [8]. The results
obtained by putting the optimum values of the genetic
where Z3 ¼ material removal rate ðmm=sÞ algorithm [9] in the model proposed by Acharya et al.
Decision variables, variable bounds, and con- [7] are also shown for comparison purposes for the three
straints are the same as specified in Example 1. The objectives and the combined objective function.
normalized combined objective function (Z) is for- Although the results obtained by the genetic algorithm
mulated by considering different weightages to all seem to be better than those obtained by PSO, they vio-
objectives and is given by the following equation late the passivity constraint and hence are not valid.
In the present work, both single objective and multi-
Z ¼ ðw 1 Z1 =Z1min Þ þ ðw2 Z2 =Z2min Þ
ðw3 Z3 =Z3max Þ
objective aspects of optimization are considered using
ð22Þ the PSO algorithm. The comparative performance of
single-objective and multi-objective optimizations is
where
shown in Table 4. As shown in Table 4, the solution
Z1min ¼ minimum value of dimensional inaccuracy
converges to a higher value of voltage using multi-
obtained when the single-objective opti-
objective optimization as compared to the results
mization problem considering only dimen-
obtained for single-objective optimization. As the vol-
sional inaccuracy as an objective was
tage increases, the dimensional inaccuracy (Z1)
solved for the given three constraints ¼
increases, but the number of spark (Z2) decreases
15.452 mm
(and hence increasing the tool life) substantially. The
Z2min ¼ minimum value of number of sparks per
value of the material removal rate (Z3) is the same
millimetre obtained when the single-
for both single-objective optimization and multi-
objective optimization problem consider-
objective optimization. The combined objective func-
ing only the tool life (in terms of number
tion (Z) therefore seems to be better.
of the sparks) as an objective was solved
for the given three constraints ¼ 1.055
Z3max ¼ maximum value of feed rate obtained when
the single-objective optimization problem 6 CONCLUSIONS
considering only the material removal rate
(in terms of the feed rate) as an objective The selection of proper values for the parameters
was solved for the given three constraints of an electrochemical machining process is crucial
¼ 25 mm/s. to the efficiency and high quality of the outcome of

JEM1158
IMechE 2008 Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014

958 R V Rao, P J Pawar, and R Shankar

Table 4 Comparison of results of single-objective optimization and multi-objective optimization using PSO algorithm

Optimization using PSO f (mm/s) U(cm/s) V(V) Z1 Z2 Z3 Z

Single-objective 8 300 9.835 15.45 15.45

Multi-objectives 8 300 13.225 39.34 3.39 8 1.811

that process. The efficiency and quality of the 2 Rajurkar, K. P., Zhu, D., McGeough, J. A., Kozak, J.,
outcome may be measured based on one or more and De Silva, A. New developments in electro-chemical
objectives. These objectives, generally, are functions machining. Ann. CIRP, 1999, 48, 567–579.
of the operating parameters and often their func- 3 Bhattacharyya, A., Sur, B., and Sorkhel, S. K. Analysis of
optimum parametric combination in electro-chemical
tional forms are defined through careful deduction
machining. Ann. CIRP, 1973, 22, 59–60.
of physical laws of nature.
4 El-Dardery, M. A. Economic study of electro-
In the present work, both single-objective optimiza- chemical machining. Int. J. Mach. Tool Des. Res., 1982,
tion and multi-objective optimization aspects of 22, 147–158.
electrochemical machining process parameters are 5 Hewidy, M. S., Fattouh, M., and Elkhabeery, M. Some
considered using a PSO algorithm. The three objec- economical sepects of ECM processes. In Proceedings
tives considered are minimization of dimensional of the 1st AME Conference, Military Technical College,
inaccuracy, maximization of tool life by minimizing Cairo, Egypt, 1984, Paper TT8, pp. 87–94.
the number of sparks per millimetre, and maximiza- 6 Hewidy, M. S., Ebeid, S. J., El-Taweel, T. A., and
tion of the material removal rate subjected to the con- Youssef, A. H. Modelling the performance of
straints of temperature, passivity, and choking. It is ECM assisted by low frequency vibrations. J. Mater.
Processing Technol., 2007, 189, 455–472.
observed that the results obtained by using the PSO
7 Acharya, B. G., Jain, V. K., and Batra, J. L. Multi-
algorithm show significant improvement over other
objective optimization of ECM process. Precision Engng,
optimization techniques such as goal programm- 1986, 8, 88–96.
ing, fuzzy set theory, and genetic algorithms. When 8 Choobineh, F. and Jain, V. K. A fuzzy sets approach
the results of single-objective and multi-objective for selecting optimum parameters of an ECM process.
optimizations obtained by PSO are compared, the Processing of Adv. Mater., 1993, 3, 225–232.
combined objective function seems to be better. How- 9 Jain, N. K. and Jain, V. K. Optimization of electro-
ever, it can be observed that in both cases, the tool life chemical machining process parameters using
is overachieved, as indicated by very low values of the genetic algorithm. Machining Sci. Technol., 2007, 11,
number of sparks. Therefore, the solution obtained 235–258.
using single-objective optimization is preferred as it 10 Kennedy, J. and Eberhart, R. Particle swarm opti-
mization. In Proceedings of the IEEE International
gives maximum dimensional accuracy along with the
Conference on Neural networks, 1995, vol. 4, pp.
necessary tool life.
1942–1948.
In this paper, the performance of PSO in terms of 11 Elbeltagi, E., Tarek, H., and Donald, G. Comparison
convergence rate and accuracy of the solution is stu- among five evolutionary based optimization algorithms.
died. Compared to other non-conventional optimiza- Adv. Engng Informatics, 2005, 19, 43–53.
tion methods, few trials are required to predict the 12 Dong, Y., Tang, J., Xu, B., and Wang, D. An application
best and worst operating parameters of the PSO algo- of swarm optimization to nonlinear programming.
rithm. The proposed algorithm requires only 30 to Computer and Mathematics with Applications, 2005,
40 iterations for convergence to the optimal solution. 49, 1655–1668.
The algorithm can also be easily modified to suit 13 Clifton, D., Mount, A. R., Alder, G. M., and Jardine, D.
optimization of process parameters of other advanced Ultrasonic measurement of the inter-electrode gap in
electro-chemical machining. Int. J. Mach. Tools Mf.,
machining processes such as electrical discharge
2002, 42, 1259–1267.
machining, ultrasonic machining, abrasive jet machin-
14 Bergh, F., and Engelbrecht, A. P. A study of particle
ing, water jet machining, etc. Also, the proposed algo- swarm optimization particle trajectories. Information
rithm can handle the multi-objective optimization Sci., 2006, 176, 937–971.
models efficiently. 15 Tripathi, P., Bandopadhyay, S., and Pal, S. K. Multi-
objective particle swarm optimization with time variant
inertia and acceleration coefficients. Information Sci.,
2007, 177, 5033–5049.
REFERENCES
16 Rao, R. V. Vendor selection in a supply chain environ-
ment using analytic hierarchy process and genetic algo-
1 Kalpakjian, S. and Schmid, R. Manufacturing engineer- rithm methods. Int. J. Services and Ops Managmt, 2007,
ing and technology, 2000 (Addison Wesley Longman 3, 355–369.
(Singapore), New Delhi).

Proc. IMechE Vol. 222 Part B: J. Engineering Manufacture JEM1158
IMechE 2008

Downloaded from pib.sagepub.com at UNIV OF WINNIPEG on August 28, 2014