Int J Adv Manuf Technol

DOI 10.1007/s00170-015-7448-9

ORIGINAL ARTICLE

Multi-objective optimization of electrochemical

machining process
Hamed Sohrabpoor 1 & Saeed Parsa Khanghah 2 & Saeid Shahraki 3 & Reza Teimouri 4

Received: 18 February 2013 / Accepted: 15 June 2015

# Springer-Verlag London 2015

Abstract Electrochemical machining (ECM) is a potential 1 Introduction

nontraditional machining process that includes too many factors
contributing in process performance. Therefore, obtaining opti- Electrochemical machining (ECM) process has been exten-
mal factor combination for higher efficiency is really complex. sively used in machining of hard-to-cut materials such as tita-
In the present work, in order to investigate effects of electrolyte nium, stainless steel, high-strength temperature-resistant al-
concentration, electrolyte flow rate, applied voltage, and feed loys, ceramics, refractories, fiber-reinforced composites, super
rate on material removal rate (MRR), surface roughness (SR), alloys, etc. which are not suitable to be machined by the con-
and radial overcut (ROC), manually constructed neuro-fuzzy ventional machining processes because of their high hardness,
inference systems have been used for creation predictive models strength, brittleness, toughness, and low machinability proper-
based on experimental observations. Then, effects of process ties. Moreover, by application of ECM process, any kind of
factors on MRR, SR, and ROC have been analyzed through intricate shape can be generated with high accuracy and preci-
plots which were drawn using developed fuzzy models. Then, sion with minimum residual stress generation. However, the
principal component analysis (PCA) was applied on experimen- process is not exactly the substitutes of the conventional ma-
tal data to allocate appropriate weight factors to the responses. chining processes, but can only complement them [1, 2]. The
Hereafter, the developed objective function which was obtained ECM has complex nature due to various complex physico-
through the combination of neuro-fuzzy and PCA is associated chemical and hydrodynamic phenomena that occur in the ma-
with teaching cuckoo optimization algorithm to find optimal chining gap [3]. During the course of machining, the machin-
parameter combination causing maximum MRR as well as min- ing rate at any instant depends not only on the end gap but also
imum SR and ROC, simultaneously. on other process parameters [4]. The electrolyte flow velocity
plays an important role in surface formation [5]; moreover, the
Keywords Electrochemical machining process . increment in gap resistance due to various causes, e.g., electro-
Multi-objective optimization lyte heating, gas bubble generation sludge formation, etc.,
leads to an uneven current flow, causing overcut phenomena
that result in poor dimensional control of the workpiece [6]. So,
* Reza Teimouri optimal quality of workpiece and economical aspects of ECM
[email protected] can be achieved through combinational control of various pro-
cess parameters. According to the above explanation, genera-
1
Department of Mechanical Engineering, Dezful branch, Islamic tion of a certain physical model which provides a precise pre-
Azad University, Dezful, Iran diction of ECM process is actually difficult especially when the
2
Department of Mechanical Engineering, Tabriz Branch, Islamic process is characterized for machining of hard-to-cut materials.
Azad University, Tabriz, Iran According to the above explanations, in order to generate a
3
Department of Mechanical Engineering, University of Zabol, model which can predict the behavior of complex ECM pro-
Zabol, Iran cess, researchers focused on developing comprehensive pre-
4
Department of Mechanical Engineering, Babol University of dictive model based on statistical analysis and artificial intel-
Technology, Babol, Iran ligence. Methods such as response surface methodology
 Int J Adv Manuf Technol

(RSM), artificial neural network (ANN), fuzzy inference sys- 3. Applying appropriate weight factors of each response
tem, etc. are most commonly used for modeling of the using principal component analysis
manufacturing processes. In the case of ECM, Bhattacharyya 4. Construction objective function regarding integration of
and Sorkhel [7] applied response surface methodology for in- fuzzy models and PCA weight factors
vestigation and modeling of electrochemical machining pro- 5. Multi-objective optimization of process parameters by
cess. They developed mathematical models to correlate rela- cuckoo optimization algorithm
tionships between electrolytic concentration, electrolytic flow 6. Comparing results obtained through ANFIS-PCA-COA
rate, applied voltage, and interelectrode gap as process inputs to with those derived by confirmatory experiments
material removal rate and overcut as responses. Senthilkumar
et al. [8] used RSM to study effects of ECM parameters on
material removal rate (MRR) and surface roughness (SR). They
showed that increasing in applied voltage and feed rate leads to 2 Methodologies
higher MRR and lower surface roughness. Also, they indicated
that increasing in electrolytic concentration and flow rate result- 2.1 Neuro-fuzzy inference system
ed in a higher material removal rate and better surface finish.
Moreover, they optimize the process for achieving higher MRR An adaptive neuro-fuzzy inference system is a hybrid predic-
and lower Ra based on RSM. In the case of modified ECM, tive model which uses both of neural network and fuzzy logic
Baran Puri and Banerjee [9] studied the effects of voltage and concepts to create a mapping relationship between inputs and
cutting speed on current density, material removal rate, and outputs [12]. The structure of this model consists of five layers
surface finish in an electrochemical grinding process. They ap- in which each layer is constructed by several nodes. Such as a
plied regression analysis to develop a mathematical model for neural network, in an ANFIS structure, the inputs of each layer
each response. Then, they utilized desirability approach and are gained by the nodes from a pervious layer. Figure 2 de-
overlapping contour plots to optimize the process in the form scribes an ANFIS structure. It can be inferred from Fig. 2 that
of multiple responses problem. Furthermore, Taweel and Gou- the network includes m inputs (X1Xm), in which each one
da [10] proposed wire electrochemical turning process and ful- consists of n membership functions (MFs). Moreover, a layer
filled a feasibility study to use wire as tool. They investigated with R fuzzy rules and also an output layer are contributed to
effects of applied voltage, wire feed rate, wire diameter, work- the construction of this model. The number of nodes in the
piece rotational speed, and overlap distance on material remov- first layer can be calculated by a product of m as the number of
al rate, surface roughness, and roughness error. The experimen- inputs and n as the number of MFs (N=m.n). The number of
tal results were statistically analyzed and modeled through re- nodes in other layers (layer 24) relates to the number of fuzzy
sponse surface methodology. Although there are numbers of rules (R). For further information about implementation of the
publications that used RSM and statistical techniques in model- ANFIS, the interested readers can read the reference [13, 14].
ing of ECM process, there are a few works that use artificial
intelligence or analyzing ECM process. Teimouri and 2.2 Principal component analysis
Sohrabpoor [11] applied artificial evolutionary approaches to
maximize MRR and minimize SR in ECM process. In the The principal component analysis is a collection of statistic
present study, a series of experiments was designed and con- analysis that is used to find the importance of quality charac-
ducted through a central composite design. Then, neuro-fuzzy teristics in a certain system such as the ECM process. In mul-
systems were used to correlate intelligent relationships between tivariate studies, the information of observations overlaps due
electrolyte concentration, electrolyte flow rate, applied voltage, to many variables and their internal correlation to some extent.
and feed rate as input and MRR, SR, and ROC as output. Then, PCA can simplify this issue by dimension reduction to find
to construct an objective function, appropriate weight factors uncorrelated composed factors that reflect the original infor-
were allocated to each response based on principal component mation as much as possible to represent all of the original
analysis. Finally, the obtained objective function (association of variables. Further details about performing PCA on experi-
neuro-fuzzy models and weight factors of PCA) was associated mental data can be found in reference [15].
with cuckoo optimization algorithm to optimize the ECM pa-
rameters achieving highest quality characteristics. 2.3 Teaching learning-based optimization algorithm
The present work can be summarized as follows:
The cuckoo optimization algorithm inspired by the life of a
1. Design and conducting experiments based on four bird family called cuckoo [16]. A special lifestyle of these
factors-five levels central composite design birds and their characteristics in egg laying and breeding has
2. Creation predictive models of MRR, SR, and ROC using been the basic motivation for the development of this new
a neuro-fuzzy system evolutionary optimization algorithm. Similar to other
Int J Adv Manuf Technol

evolutionary methods, the cuckoo optimization algorithm NaCl solution because of the fact that NaCl electrolyte has no
(COA) starts with an initial population. The cuckoo popula- passivation effect on the surface of the specimens.
tion, in different societies, is in two types: mature cuckoos and In order to calculate MRR, a WTB RADWAG electronic
eggs. The effort to survive among cuckoos constitutes the basis weighing device has been used for measurement of workpiece
of cuckoo optimization algorithm. During the survival competi- mass loss during experiment. In this case, the machining time
tion, some of the cuckoos or their eggs demise. The surviving was kept constant on 2 h. For evaluating surface roughness, a
cuckoo societies immigrate to a better environment and start MAHR MarSurf PS1 surface profile meter with a cutoff of
reproducing and laying eggs. Cuckoos survival effort hopefully 0.5 mm has been utilized. For each specimen, the roughness of
converges to a state that there is only one cuckoo society, all with a side wall was measured five times stochastically and the
the same profit values. Further detail about COA algorithm is average of them has been reported. Also, the radial overcut
found in literature [16]. is calculated by the difference of actual hole radius and ex-
pected hole radius through a digital micrometer.
The experimental observations were made by varying pre-
3 Experimentations dominant process parameters such as applied voltage, electro-
lyte concentration, electrolyte flow rate, and tool feed rate.
The KENNAMETAL die-sinking ECM machine with servo Table 2 presents process factors and their levels. Due to a wide
motorized vertical up/down movement of tool has been used range of factors, it was decided to use four factors, five levels,
for conducting experiments. Figure 1 depicts a schematic of and central composite design (CCD) matrix to optimize the
ECM process. The workpiece material is cemented tungsten experimental conditions. For this purpose, the Design-Expert
carbide in the shape of a plate with dimensions of 20 mm V8 statistical software has been employed to design experi-
20 mm10 mm. The faces of each specimen were ground to ments. Table 3 shows the 31 sets of coded conditions used to
generate accurate parallel sitting on the machine table. The form the design matrix. The first 16 experimental conditions
properties of a workpiece have been presented in Table 1. The are derived from full factorial experimental design matrix (24 =
tool was made up from copper in the form of cylinders with an 16). All the variables at the intermediate (0) level constitute the
outer diameter of 12 mm and an inner diameter of 6 mm. An center points while the combinations of each process variable
electrolyte was axially fed to the cutting zone through the central at either their lowest (2) or highest (+2) with the other three
hole of the tool. The electrolyte used for experiment was fresh variables of the intermediate levels constitute the star points.

Fig. 1 A schematic view of

experimental setup
 Int J Adv Manuf Technol

Table 1 Main properties of workpiece

Nominal Grain Hardness Density (kg/ Transverse strength Compressive strength Modules of elasticity
composition size (HV) m3) (MPa) (MPa) (GPa)

90 % WC-10 % Co Fine 13001800 14,600 3100 5170 620

4 Results and discussions tion, the value of surface roughness decreases and reaches to
a minimum value at center point. But by further increase in
In this section, the effects of process factors on ECM perfor- electrolyte concentration, the surface roughness increases,
mance are analyzed according to perturbation plots which accordingly. When the electrolyte flow rate is lower than a
obtained through the measured values of responses. The dis- critical value, the current density is lower and poor surface
cussion about the obtained results is presented as follows. finish is obtained. On the other hand, when going beyond a
critical value due to excessive current density, the tendency
4.1 Analysis of MRR

Figure 2 is a perturbation plot, which illustrates the effects of Table 3 Design matrix and experimental results
all the factors at the center point in the design space. Accord- No. Process factors Responses
ing to this figure, it is seen that increase in all factors, i.e.,
electrolyte concentration, flow rate, applied voltage, and feed A B C D MRR (g/min) SR (m) ROC (mm)
rate causes increase in the value of material removal rate. A 1 1 1 1 1 0.301 7.899 0.255
higher electrolyte concentration causes improvement in elec- 2 1 1 1 1 0.217 6.782 0.325
trical conductivity of electrolyte and increases the current den- 3 1 1 1 1 0.321 6.781 0.27
sity in interior electrolyte gap (IEG); therefore, the MRR in- 4 1 1 1 1 0.342 5.984 0.315
creases, correspondingly. Also, when the electrolyte flow rate 5 1 1 1 1 0.361 6.889 0.30
increases, the higher mobility of charged ions improves and 6 1 1 1 1 0.428 6.748 0.34
leads to better chemical reactions. Moreover, the hydrogen 7 1 1 1 1 0.317 6.815 0.295
bubbles are removed from cathode grooves and it increases 8 1 1 1 1 0.561 6.891 0.345
ionic strength and leads to higher MRR on the anodic part. 9 1 1 1 1 0.323 6.678 0.315
Hence, at a high flow rate, higher MRR is obtained. 10 1 1 1 1 0.501 6.389 0.38
By increasing applied voltage, the current density in IEG 11 1 1 1 1 0.421 6.364 0.31
also increases and causes improvement in material removal 12 1 1 1 1 0.592 6.214 0.325
rate. Also, when the feed rate increases, the IEG decreases 13 1 1 1 1 0.656 6.234 0.32
accordingly and causes inducing current on a smaller area. 14 1 1 1 1 0.621 6.351 0.35
Therefore, the current density increases and causes high 15 1 1 1 1 0.781 6.489 0.325
MRR. 16 1 1 1 1 0.818 6.342 0.325
17 2 0 0 0 0.311 7.982 0.245
4.2 Analysis of SR 18 2 0 0 0 0.644 6.059 0.345
19 0 2 0 0 0.362 7.421 0.275
Figure 3 is a perturbation plot, which illustrates the effects of 20 0 2 0 0 0.582 6.235 0.25
all the factors at the center point in design space. From this 21 0 0 2 0 0.212 6.981 0.25
figure, it is seen that by an increase in electrolyte concentra-
22 0 0 2 0 0.421 7.981 0.395
Table 2 Process factors and their levels 23 0 0 0 2 0.502 6.578 0.305
24 0 0 0 2 0.981 5.210 0.345
Parameters Symbol Levels 25 0 0 0 0 0.358 5.021 0.295
26 0 0 0 0 0.218 5.124 0.35
2 1 0 1 2
27 0 0 0 0 0.481 5.034 0.32
Electrolyte concentration (g/l) A 10 15 20 25 30 28 0 0 0 0 0.331 5.482 0.315
Electrolyte flow rate (l/min) B 5 6 7 8 9 29 0 0 0 0 0.356 5.142 0.295
Applied voltage (V) C 12 13 14 15 16 30 0 0 0 0 0.318 5.111 0.345
Feed rate (mm/min) D 0.2 0.4 0.6 0.8 1 31 0 0 0 0 0.358 5.214 0.34
Int J Adv Manuf Technol

Fig. 2 Perturbation plot showing

effects of all the factors on MRR

of formation of pits on a machined surface increases and it density which removes pits from a machined surface. But by
causes a rougher surface. Also, from this figure, it is seen further increase in applied voltage, the surface roughness in-
that effects flow rate is similar to electrolyte concentration. creases drastically due to excessive heat generation on ma-
At low flow rate (lower than its center point), turbulence is chined surface that causes distortion of a machined surface.
being low, so the transported material moves slowly in a Increasing in feed rate causes smaller interior electrode gap
stratified form producing a streak on the surface and causes and causes higher current density which removes undesirable
a poor surface finish. On the other hand, when the flow rate pits from machined surface and improves surface finish.
goes beyond the center point, due to higher chemical reac-
tions, some defects such as pits may occur on a machined 4.3 Analysis of ROC
surface and damages the surface quality.
The applied voltage also shows the same behavior. By in- Figure 4 is a perturbation plot, which illustrates the effects of
creasing the applied voltage, the surface roughness decreases all the factors at the center point in design space. From this
and reaches to a minimum value due to an increase in current figure, it is seen that an increase in electrolyte concentration

Fig. 3 Perturbation plot showing

effects of all the factors on SR
 Int J Adv Manuf Technol

Fig. 4 Perturbation plots

showing effects of all the factors
on ROC

causes a higher overcut. A higher electrolyte concentration 5 Modeling of MRR, SR, and ROC by ANFIS
leads to the formation of a greater volume of reaction prod-
ucts, e.g., sledges and precipitates and also inhibits the initia- As explained, the adaptive neuro-fuzzy inference system has
tion of gas bubbles, e.g., O2, H2, etc. These phenomenal ef- been used for prediction of MRR, SR, and ROC. For this
fects due to increased electrolyte concentration may, to a great purpose, the MATLAB R17 package (ANFIS toolbox) has
extent, lead to the possibility of the passage of stray current to been utilized.
the machining periphery. Also, the higher values of dielectric Prediction of the quality characteristics of the ECM process
flow rate causes a decrease in ROC. By increasing in the flow by ANFIS consists of two main stages, training and testing.
rate, the effects of stray current flow weaken gradually be- Hence, among 31 data sets cited in design matrix, the number
cause of the squeezing of the gas bubble diameters and the of 22 data has been selected stochastically for training of
quicker removal of reaction products and gas bubbles from the ANFIS network. Then, the trained network has been tested
machining zone. Therefore, the ROC decreases accordingly. by the other nine remaining data sets that were not contributed
The applied voltage is the factor having a great influence in training.
on overcut. According to Fig. 4, an increase in the voltage There are some important factors that contributed to pro-
causes a greater electrolyzing current to be available in the duce an accurate prediction by the ANFIS. They are type of
machining gap, as well as causing a greater stray current in- fuzzy-based rule, number of membership functions (MFs),
tensity. Hence, more material removed from the space be- and type of membership functions. In this paper, a first-order
tween the outer wall of tool and inner wall of workpiece. Also, TSK type fuzzy-based rule was used for creation of a predic-
from this figure, it is seen that an increase in feed rate causes a tive model. Then, the various numbers of membership func-
higher overcut. When the feed rate increases, the IEG in- tions have been tried. In order to compare the accuracy of all
creases accordingly and induces a higher current density in existing networks and select most accurate one, the value of
IEG space. Therefore, more material is removed from work- error goal (RMSE) was set 0.01 and the iteration number was
piece inner wall and ROC increases. 200 epochs. It means that the training epochs are continued

Table 4 Values of RMSE in

testing of MRR, SR, and ROC for Type of membership functions RMSEs of MRR RMSEs of SR RMSEs for ROC
various types of membership
function under 2-2-2-2 ANFIS Triangular 0.0463 0.2825 0.0622
structure Trapezoid 0.1321 1.3651 0.1472
Generalized bell 0.0731 0.4929 0.0842
Gaussian 0.0827 0.6822 0.0911

Numbers that written with italic related to most precise prediction

Int J Adv Manuf Technol

Fig. 5 Comparison between

measured and predicted values of
testing data for a MRR, b SR, and
c ROC

until the RMSE fell below 0.01 or the epochs go up 200. As

the RMSE criterion for all networks is the same, their actions Table 5 Eigenvalues and explained variations for principal component
are comparable. Then, their testing performances were com-
Principal component Eigenvalues Explained variations (%)
pared and the optimized model is selected based on its predic-
tive accuracy in response to new input data in the testing phase First 2.4315 81.05
when compared with experimental values. Second 0.1886 6.29
By testing various structures of ANFIS model for each Third 0.3798 12.66
response (MRR, SR, and ROC), it was obtained that structures
 Int J Adv Manuf Technol

Table 6 Eigenvectors for principal component

Quality characteristic Eigenvectors

First principal component Second principal component Third principal component

MRR 0.6010 0.7951 0.0809

SR 0.5599 0.3466 0.7526
ROC 0.5704 0.4976 0.6335

with numbers of 16 membership functions (2 MFs for each 6 Optimization

input or 2-2-2-2 topography) have the lowest values of
RMSE. Selection of a network with a larger number of MFs 6.1 Development objective function
leads to over-fitting and does not obtain the desired value of
RMSE. Another factor which is influential in the accuracy of In the present paper, in order to objectively reflect the relative
ANFIS model is the type of membership functions. In this importance for each performance characteristic in optimiza-
work, various types of MFs, namely, triangular, trapezoid, tion, principal component analysis is specially introduced here
generalized bell, and Gaussian have been practiced. Table 4 to determine the corresponding weighting values for each per-
presents RMSEs of ANFIS models in testing for MRR, SR, formance characteristic. The elements of the array for multiple
and ROC. From this table, it is inferred that functions under 2- performance characteristics listed in Table 3 represent the ob-
2-2-2 structure have been trained and their RMSEs were cal- tained value of each performance characteristic. These data
culated. Results indicated that for MRR, SR, and ROC. Tri- are used to evaluate the correlation coefficient matrix and
angular type of membership function leads to the lowest determine the corresponding eigenvalues, that is shown in
values of RMSE. Figure 5ac indicates agreement between Table 5. The eigenvector corresponding to each eigenvalue
measured and ANFIS predicted values of testing for cutting is listed in Table 6, and its square can represent the contribu-
velocity and surface roughness, respectively. According to the tion of the corresponding performance characteristic to the
figure, it can be inferred the developed ANFIS models can principal component. Table 7 shows that the contributions of
model MRR, SR, and ROC as well. MRR, SR, and ROC are 0.5031, 0.2735, and 0.2234, respec-
tively. Moreover, the variance contribution for the first princi-
pal component characterizing the whole original variables,
Table 7 The i.e., the three performance characteristics, is as high as
contribution of each Quality characteristic Contribution
81.05 %. Hence, for this study, the squares of its correspond-
individual quality
MRR 0.5031 ing eigenvectors are selected as the weighting values of the
characteristic for the first
principal component SR 0.2735 related performance characteristic and the coefficients w1, w2
ROC 0.2234 and w3 in for construction objective function are 0.5031,
0.2735, and 0.2234, respectively.

Table 8 Setup parameters of COA for implementation

Parameter Value/function Definition

X[0] [0 0 0 0] Initial population

NC 50 Number of cuckoos
H(X) Eq. 1 ANFIS models of MRR, Ra, and ROC with their PCA weight factors
NEmin 2 Minimum number of eggs
NEmax 5 Maximum number of eggs
NCL 2 Number of clusters
0.9 Motion coefficient
NCmax 200 Maximum number of cuckoos that can live at the same time
RC 50 Radius coefficient (control parameter of egg laying)
PV 1e-13 Stop condition (population variance that cut optimization)
Int J Adv Manuf Technol

Table 9 Obtained optimal results based on various weight factors

Optimal setting Responses

Electrolyte concentration (g/l) Electrolyte flow rate (l/min) Applied voltage (V) Feed rate (mm/min) MRR (g/min) SR (m) ROC (mm)
20.68 8 14.17 0.8 0.6816 5.386 0.3265

Therefore, the objective function for maximizing MRR The optimum results which were obtained through inte-
as well as minimizing SR and ROC is presented as the gration of ANFIS, PCA, and COA for has been presented
following equation. in Table 9.
In order to verify the optimal solution which was obtain-
^
F W1 MRRW ^ ^
2 SRW3 ROC 1 ed through optimization, a confirmatory experiment with
optimal setting of factors should be conducted. If the ob-
where W1, W2 and W3 are the weighing factor related to each tained values of MRR, SR, and ROC in Table 9 are near to
output according to its importance in the process. those derived from confirmatory experiment, the modeling
^ SR,
MRR, ^ and ROC ^ are the normalized values of MRR, and optimization is performed successfully. Table 10 pre-
SR, and ROC which obtained by following equations: sents comparison between optimal MRR, SR, and ROC of
confirmatory experiment with results of Table 9. It is seen
^ MRRMRRmin
MRR 2 that an acceptable agreement exists between these two sets
MRRmax MRRmin
of results showing the efficiency of proposed method in the
^ SRSRmin
SR 3 present study.
SRmax SRmin

^ ROCROC min
ROC 4
ROC max ROC min
7 Conclusion
where, MRRmin and MRRmax are the minimum and maximum
values of MRR, respectively. Also, SRmin and SRmax are the The present work focused on multi-objective optimization of
minimum and maximum values of SR, respectively. More- ECM process of tungsten carbide WC-Co. The process main
over, ROCmin and ROCmax are the minimum and maximum factors were electrolyte concentration, electrolyte flow rate,
values of ROC, respectively. applied voltage, and feed rate. Also, the main responses were
MRR, SR and ROC. In order to perform the multi-objective
optimization, the cuckoo optimization algorithm was used
6.2 Optimization by COA where the objective function was constructed by integration
of the ANFIS model each response and their weight factors
Before performing the optimization by COA, the optimization which were obtained through principal component analysis.
criteria should be defined. The criterion for each factor is The important results can be summarized as follows:
presented as follows:
1. In modeling of MRR, SR, and ROC by ANFIS, the 2-2-2-
Electrolyte concentration: 10 to 30 (g/lit) 2 ANFIS structure with triangle type of membership func-
Electrolyte flow rate: 5 to 9 (lit/min) tion was selected as the best topography due to its lowest
Applied voltage: 12 to 16 (volt) prediction error and faster performance.
Feed rate: 0.2 to 2 (mm/min) 2. In optimization of process parameters by ANFIS-PCA-
COA, electrolyte concentration of 20.68 g/l, flow rate of
Such as other evolutionary algorithms, the COA needs its 20 l/min, voltage of 14.17 V, and feed rate of 0.8 mm/min
own setup parameters for implementation accurately. Table 8 were selected as the optimal solution that caused MRR of
presents the setup parameters for COA. 0.681 mg/min, SR of 5.386 m, and ROC of 0.326 mm.

Table 10 Comparison of MRR, SR, and ROC of confirmatory experiments with those derived from ANFIS-PCA-COA

MRR (g/min) SR (m) ROC (mm)

Measured Predicted Error (%) Measured Predicted Error (%) Measured Predicted Error (%)
0.65 0.6816 4.86 5.71 5.386 56 0.35 0.326 6.8
 Int J Adv Manuf Technol

3. The proposed methodology of ANFIS-ICA can be ap- 8. Senthikumar C, Ganesan G, Karthikeyan R (2009) Study of elec-
trochemical machining characteristicsof Al/SiCp composites. Int J
plied for solving multi-responses optimization problems
Adv Manuf Technol 43:256263
of manufacturing process in which the responses are high- 9. Puri AB, Branjee S (2012) Multiple-response optimisation of elec-
ly correlated to each other. trochemical grindingcharacteristics through response surface meth-
odology. Int J Adv Manuf Technol. doi:10.1007/s00170-012-4065-8
10. El-Taweel TA, Gouda SA (2011) Performance analysis of wire
electrochemical turning processRSM approach. Int J Adv
Manuf Technol 53:181190
11. Teimouri R, Sohrabpoor H (2013) Application of adaptive neuro-
References fuzzy inference system and cuckoo optimization algorithm for an-
alyzing electro chemical machining process. Front Mech Eng 8(4):
1. Tipton H. (1964). Dynamics of ECM process. Proc. 5th Int.MTDR 429442
Conf. Birmingham, UK, Pergamon, Oxford,505-522 12. Teimouri R, Baseri H (2013) Optimization of magnetic field
assisted EDM using the continuous ACO algorithm. Appl Soft
2. McGeough JA (1988) Advanced methods of machining.
Comput 14:381389
Chapmanand Hall, London
13. Teimouri R, Baseri H, Moharami R (2013) Multi-responses optimi-
3. Amalnik MS, McGeough JA (1996) Intelligent concurrent manu- zation of ultrasonic machining process. J Intell Manuf. doi:10.1007/
facturability evaluation of design for electrochemical machining. J s10845-013-0831-1
Mater Process Technol 61:130139 14. Babajanzade-Roshan S, Behboodi-Jooibari M, Teimouri R,
4. Thorpe JF (1971) A mathematical model of electrochemical Asgharzade-Ahmadi G, Falahati-Naghibi M, Sohrabpoor H
machiningprocess, 3rd Int. Sem. on Optimisation of (2013) Optimization of friction stir welding process of AA7075
Manufacturing Systems. CIRP, Pisa, Italy. CAP19 aluminum alloy to achieve desirable mechanical properties using
5. Chetty OVK, Murthy R, Radhakrishnan V (1981) On some aspects ANFIS models and simulated annealing algorithm. Int J AdvManuf
of surface formation in ECM. J Eng Ind ASME 103:341348 Technol. doi:10.1007/s00170-013-5131-6
6. Bhattacharyya B, Sorkhel SK. (1989). Computer-aided design of 15. Lu HS, Chang CK, Hwang NC, Chung CT (2009) Grey relational
toolsin ECM for accurate job machining, Proc. ISEM9, Japan,pp. analysis coupled with principal component analysis for optimiza-
240243 tion design of the cutting parameters in high-speed end milling. J
7. Bhattacharyya B, Sorkhel SK (1999) Investigation for controlled Mater Process Technol 209:38083817
ellectrochemical machining throughresponse surface methodology- 16. Rajabioun R (2011) Cuckoo optimization algorithm. Appl Soft
based approach. J Mater Process Technol 86:200207 Comput 11:55085518