Available online at www.sciencedirect.

com

ScienceDirect
Materials Today: Proceedings 5 (2018) 7515–7524 www.materialstoday.com/proceedings

IMME17

Study of surface integrity and recast surface machined by Wire

electrical discharge machining
Bikash Choudhuri*, Ruma Sen, Subrata Kumar Ghosh, S. C. Saha
Mechanical Engineering Department, National Institute of Technology, Agartala, 799046, India

Abstract

The surface quality of wire electric discharge machining is determined by formation of recast layer and its roughness profile.
These characteristics has a great relationship with the process parameters. To attain an intense knowledge of the surface
roughness and recast layers of a surface machined by WEDM on stainless steel 316, the inputs and outputs correlation has been
constructed using artificial neural network model. Particle swarm optimization algorithm is used to extract the optimal set of
process parameters for multi responses. Thick white layers (13–16 µm) and rough surface (3.5-4 µm) was found at high pulse on
time while very thine white layer (3-5 µm) and finish surface (1.5-2 µm) can be observed at low pulse on time.
© 2017 Elsevier Ltd. All rights reserved.
Selection and/or Peer-review under responsibility of International Conference on Emerging Trends in Materials and
Manufacturing Engineering (IMME17).

Keywords: WEDM; AISI SS316; Roughness; Recast layer; PSO

1. Introduction

Wire electric discharge machining (WEDM) is an advance machining process useful to machine any metallic
materials irrespective of their mechanical properties, because there is no physical interaction between the tool
electrode and the workpiece during the process [1]. In WEDM removal of material occur by a thermo-electric
process where a series of sparks occur between two electrodes with in a second [2]. With the advantages of high-
performance price ratio, good cooling effect, and material removal ability, WEDM play a significant role in

* Corresponding author. Tel.: +919436997858

E-mail address: choudhurib@rediffmail.com

2214-7853 © 2017 Elsevier Ltd. All rights reserved.

Selection and/or Peer-review under responsibility of International Conference on Emerging Trends in Materials and Manufacturing Engineering
(IMME17).
7516 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524

different manufacturing industries mainly automotive, aerospace, and molding to machine electrically conductive
difficult-to-machine materials with accurate, intricate, and irregular shapes [3]. Due to generation of high
temperature and pulse discharge rate, this process faces some difficulties including formation of recast layer, crack,
defects, and high surface roughness which can damage the surface integrity. Therefore, to reduce surface defects it is
essential to study the effect of process parameters on surface characteristics that can control machining performance.
For its complex behavior, it is necessary to evaluate an optimal parametric setting to increases machinability of
WEDM process. Many literatures have already narrated on machinability and optimization of EDM/WEDM.
Sidhom et al [4] reported the effect of electro discharge machining on white layer formation and corrosion resistance
of AISI316L SS. Surface defects like pores, craters and the cracks were observed using Scanning electron
microscopy (SEM), energy dispersive microanalysis (EDX) analysis Punturat et al [5] studied the surface
characteristics and defects of monocrystalline silicon machined by wire-EDM. The EDX analysis revealed that low
spark density and low flushing rate of dielectric fluid were increased the deposition of electrode materials such as
copper, zinc, silicon on the machined surface. Ekmekci et al [6] examine the electro discharge machinability of mold
steel and micro cracks found on white layer due to residual stress. Sharma et al [7] found that pulse on time and
servo voltage play as a significant role in the formation of recast layer. Due to high discharge energy, larger craters
might be formed on the machined surface. Dewangan et al [8] reported that low flushing rate of dielectric fluid
unflushed molten material gets re-solidified and formed very hard and grained layer on the machined surface which
is called recast layer that leads to generation of cracks and other surface defects.
Modeling of machining process is necessary to interpret, predict and optimize the results, which are otherwise
hard to understand [9]. Optimization of the WEDM process parameters is also essential because identify optimal
machine settings plays a vital role in obtaining high productivity along with better surface quality. Several research
has been reported on modeling and optimizing WEDM using well known methods. Dewangan et al [10] made an
approach using fuzzy-TOPSIS to optimize performance characteristic such as recast layer thickness, surface
roughness, surface crack density, and over cut after machining AISI P20 tool steel in EDM. Maher et al [11]
introduced ANFIS for modeling the process parameters and performance characteristics. Kuriachen et al [12]
discuss fuzzy based particle swarm optimization technique for prediction and optimization of MRR, surface
roughness in micro-wire EDM of Ti-6Al-4V. In present investigation, artificial neural network (ANN) is used to
map the input/output relationships to reduce error prediction of the WEDMed surface and then model is hybridized
with particle swarm optimization (PSO) to search the best combination of WEDM parameters. Present study also
highlighted the surface topography, elemental dispersion and inclusion in WEDM.

2. Experimentation and measurements

Austenite standard chromium nickel steel AISI 316 of 12 mm thickness plate was used to machine using WEDM
(ELECTRONICA SPRINTCUT). The uncoated soft brass (CuZn36) (250 m in diameter) is used as tool electrode
deionized water as dielectric fluid. The specification of work material is shown in Table 1 and Table2. Wire feed
rate, servo voltage and flushing pressure are kept constant as 4.0 m/min, 1020 machine unit and 10.0 L/min
respectively during machining process.

Table 1. Chemical composition of AISI 316 stainless steel.

ELEMENTS C Cr Mn Mo Ni P S Si Fe
% 0.03 17 2 2.5 12 0.045 0.012 1 Balance

Table 2. Basic physical properties of AISI 316 stainless steel.

Density Melting Point Thermal Conductivity Electrical Resistivity Specific Heat Capacity Hardness
8.1 G/M3 1371ºC-1399ºC 16.2 W/M/K 0.74 Ohm-mm2/m 450 J/Kg-K 15-20 HRC
 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524 7517

2.1 Surface roughness (Ra) & Recast layer (RL)

The surface roughness (Ra) is measured for checking machine surface quality. It was measured at ten different
points and the average of them was for calculated analysis. During cooling re-solidification of molten material
which is not flushed away by the dielectric fluid, formed a layer, known as recast layer [13]. The area of
recast/white layer was measured using software (ImageJ) and then the cross-section area of recast layer was divided
by length of scanning electron microscopic images to measure the average thickness of recast layer (RL). In order to
observe the surface roughness, recast layer thickness, surface integrity, and elements present in the recast layer,
‘Talysurf (Model: Taylor Hobson)’ surface roughness tester, scanning electron microscopy (ZEISS-8113 SEM) and
energy dispersive spectrograph (EDS) associated with the SEM was used after WEDM.
The process variables and their units and range for conducting experiments have been reported in Table 3. The
maximum and minimum range of these significant parameters are shorted out by trial and error methods.
Experiments were planned using Taguchi’s L25 orthogonal array based factorial design which are considered for
training ANN model and 5 random experiments are used for testing the proposed model. The experimental design
and results are shown in Table 4.
Table 3. WEDM parameters used in the experimentation.

Parameters Levels
Pulse on time (µsec) 0.35 0.6 0.85 1.1 1.35
Pulse off time (µsec) 10 15 20 25 30
Current (A) 140 160 180 200 220
Servo voltage (V) 20 30 40 50 60
Wire Tension (kg) 0.4 0.6 0.8 1.0 1.2

Table 4. Design of experiments and results.

Sl no Ton Toff IP SV WT Ra (µm) RL (µm) Sl no Ton Toff IP SV WT Ra (µm) RL (µm)
1 0.35 10 140 20 0.4 2.028 9.48  14 0.85 25 140 40 1.2 2.611  11.89 
2 0.35 15 160 30 0.6 1.851 8.07  15 0.85 30 160 50 0.4 2.462  11.62 
3 0.35 20 180 40 0.8 1.919 6.79  16 1.10 10 200 30 1.2 2.891  13.84 
4 0.35 25 200 50 1.0 1.989 5.75  17 1.10 15 220 40 0.4 2.992  15.15 
5 0.35 30 220 60 1.2 2.013 5.26  18 1.10 20 140 50 0.6 3.029  14.16 
6 0.60 10 160 40 1.0 2.032 10.40  19 1.10 25 160 60 0.8 2.718  11.36 
7 0.60 15 180 50 1.2 2.637 9.56  20 1.10 30 180 20 1.0 3.742  12.52 
8 0.60 20 200 60 0.4 2.368 9.17  21 1.35 10 220 50 0.8 3.257  14.54 
9 0.60 25 220 20 0.6 2.401 10.40  22 1.35 15 140 60 1.0 3.17  13.79 
10 0.60 30 140 30 0.8 2.311 9.82  23 1.35 20 160 20 1.2 3.442  14.24 
11 0.85 10 180 60 0.6 2.693 11.45  24 1.35 25 180 30 0.4 3.383  14.83 
12 0.85 15 200 20 0.8 2.796 12.79  25 1.35 30 200 40 0.6 3.159  13.15 
13 0.85 20 220 30 1.0 2.897 12.17     

3. Results and Discussion

3.1 Analysis of WEDM parameters on surface roughness and recast layer

The influence of parameters was explored through effect plot (Fig.1) and ANOVA (Table 5) on surface
roughness and recast layer. ANOVA results indicate that the pulse duration (85.11%), peak current (5.55%), and
7518 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524

servo voltage (4.99%) have physical and statistical influence on surface roughness. Whereas pulse on time
(86.90%), pulse off time (4.32%), servo voltage (4.78%) and wire tension (2.38%) produce significant effect on
recast layer (Table 5). Lengthier pulse on time with peak current remove more amount of material from workpiece
surface and produce deeper, overlapping craters and debris. Some molten metal is not wash way by dielectric but
redeposited on the surface. This subsequently generates rough surface and deeper recast layer. On the other hand,
higher values of servo voltage and pulse off time result produce better surface quality and lesser recast formation.

Table 5. Analysis of variance analysis for surface roughness (Ra) & recast layer (RL).
surface roughness (Ra) recast layer (RL)
Factor DF SS MS F-Value Contribution DF SS MS F-Value Contribution
Ton 4 5.72741 1.43185 56.22 85.11% 4 163.681 40.9203 517.39 86.90%
Toff 4 0.09576 0.02394 0.94 1.42% 4 8.135 2.0338 25.72 4.32%
IP 4 0.37371 0.09343 3.67 5.55% 4 2.742 0.6855 8.67 1.46%
SV 4 0.33581 0.08395 3.30 4.99% 4 8.998 2.2496 28.44 4.78%
WT 4 0.09485 0.02371 0.93 1.41% 4 4.488 1.1219 14.19 2.38%
Error 4 0.10187 0.02547 1.51% 4 0.316 0.0791 0.17%
Total 24 6.72941 100% 24 188.361 100%

Fig. 1. Main effect plot of (a) roughness and (b) recast layer.

3.2. Multi-objective Optimization

3.2.1. Artificial neural network (ANN) Modeling

ANN has been implemented as a smart flexible tool for mapping non-linear multivariate model which possesses
the capability to learn from historical data. The feed-forward BP-ANN model is developed in MATLAB. The
structure of the model is shown in Fig. 2. The information received by the input layer from the source node is passes
to output layer through the hidden layer. Initially the weights and bias are selected randomly for training data. [14,
15].
 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524 7519

Fig. 2. The structure of the BP neural network. Fig. 3. performance evaluation of neural network.

Summation function used in this study is given in Eq. 1.

net j = ∑i =1 X i wij + bij

n
i=1, 2, . . ., n; j=1, 2, . . ., l (1)
where xi denotes the input variable i; wij denotes the connection weights between neuron i in the input layer and
the neuron j in hidden layer; f is the transitive function of hidden layer neurons; netj is the activation value of neuron
j. bij is the bias. The transfer function (logistic sigmoid) of output neuron of jth hidden layer is indicated as the

H j = f(net j ) = 1 (1 + e −net ) (2)

Hj representing the outputs of hidden layer neurons, f represents the incentive function of neurons.
The best configuration for the smart system is normally determined based on the minimum value of MSE and
maximum magnitude of R2. MSE is the mean of the predicted squares error. The smallest MSE value, shows better
efficiency of the model.
In this study, 5 input neurons, 1 hidden layer (having 20 neuron) and 2 output neurons include Ra and RL are
found to be the best architecture. The data sets are normalized between 0.1 and 0.9 in order to make the model
quickly converge. The training process of the BP neural network and its performance is shown in Fig. 3. The model
converges to 0.0000649 (MSE) at the 2000 iteration.

3.2.1.1. Testing of ANN model

After testing of data as shown in Table 6, it can be concluded that the neural networks appear to constitute a
correct model for predicting the Ra, and RL as experimental results and ANN predicted results are very closed.

Table 6. Testing data of experimental result and neural network model.

Process Parameter Experimental Predicted Error (%)
Ton - Toff – IP – SV - WT Ra RL Ra RL Ra RL
0.85 – 14 – 150 – 40 - 0.55 2.768 10.523 2.736 9.891 1.16 6.01
1.1 – 14 – 150 – 45 - 0.85 3.285 12.658 3.143 11.894 4.32 6.04
1 - 17 – 150 – 45 - 0.85 2.875 11.452 2.991 11.221 -4.03 2.02
0.75 - 25 - 180 - 30 - 0.85 2.899 10.456 3.157 11.088 -8.9 -6.04
1 – 14 – 220 – 50 - 0.8 2.952 10.368 2.726 10.835 7.66 -4.5
7520 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524

3.2.1.2. Checking adequacy of model

Fig. 4 (a) & (b) displays the comparison between experimental value and ANN predicted value for training data
of Ra and RL respectively. The very small values of the RMSE and the high values of R2 and Adjusted R2, for both
the training and test set outputs are the indication of adequacy and accuracy of ANN model.

Fig. 4. Experimental (a) Ra and (b) RL Vs ANN predicted result for training data set.

3.2.2. Particle swarm optimization (PSO)

In order to predict the optimal structure, the BP-ANN fitness function is merged to PSO algorithm with objective
of maximum MPCI to solve the both conflicting objective problem. After ANN model was established, PSO is used
to optimized its input space. PSO initialize with random motion of intelligent swarm and then searches the optima
through updating generation [16]. The basic concept of PSO is based on finding Pbest value (particle’s best position)
and Gbest value (global best position). Each particle is updated in every iteration.

Vi t +1 = w tVi t + C1r1 ( pbest

t
− X it ) + C2 r2 ( g best
t
− X it ) i=1, 2, 3, …..t (4)

X it +1 = X it + Vi t +1 (5)

Where, and are the velocity and and are position of particle i at iteration t and t+1. w is the
inertia weight; C1 and C2 are cognitive (individual) and social (group) learning rate, r1 and r2 are random number
uniformly distributed within the range of 0 and 1, and are the best position in the current swarm and i
particle over generation t. The population is initialized with random data with swarm size of 20. The values of C1
and C2 are taken as 2. The initial weight and final weight are taken as 0.95 and 0.05 respectively. If the MPCI value
found in new iteration is better than old one, then PSO replace the old MPCI. Its position is taken as particle best
position. Among all best value the global best (Gbest) position has been chosen to reach optimal position very
rapidly and accurately. Fig.9 shows the trend of convergence of PSO to obtain the best objective value achieved
during every iteration. It is observed that 60 numbers of iteration are taken by PSO for convergence to the best
value. The highest value of MPCI is 0.7816 which is indicates that the optimum combination of parameter setting
within the experimental range. The results are shown in Table 7. The optimal values for the Ton, Toff, IP, SV, WT as
obtained from the model are calculated to be as, 0.35 µsec, 24 µsec, 170 Ampere, 60 volts and 1.2 kg respectively.
 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524 7521

Fig. 5. Objective function convergence of PSO for the optimization of WEDM process parameter.

Table 7. Optimal solutions for multi-objectives optimization problem.

Optimum condition (MPCI-0.7816) Optimum Performance ANN Predicted
Ton-Toff-IP-WT-SV Ra RL Ra RL
0.35-24-170-60-1.2 1.925 5.652 1.864 5.32

3.3. Surface morphology

3.3.1. Subsurface microstructure

The specimen was epoxied, gently polished, etched by Glyceregia (15 ml HCL+ 10 ml glycerol + 5 ml HNO3),
ultrasonic cleaned with ethyl alcohol, and air dried. Then, microstructure and surface characteristics obtained from
optical microscope and SEM is shown in Fig. 6, 8 & 9. The microstructure of the sub surface (HAZ) tried to
examine because the heat-affected zone is composed of several layers of different properties [17]. It contains a
different microstructure, impurities, micro-cracks, porosities and other undesirable features which can be causes part
failure.

(a) (b)

Fig. 6. Microstructures of AISI SS316 (a) base metal and (b) subsurface (HAZ) of WED machine.

3.3.2. Surface Integrity:

The surface characteristics of stainless steel 316 alloy after WED machined are analyse through the surface profiler,
scanning electron microscope (SEM), EDS analysis. Fig. 7 (a) and (b) shows the 3D profile of WEDMed surface
throughout finish cutting condition and rough cutting condition.
7522 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524

Fig. 7. 3D view of machine surface in (a) finish cut (b) rough cut mode.

It is observed from experimental investigation that the average surface roughness and average thickness of recast
layer for stainless steel 316 is found to be 2.26 µm and 10.15 µm respectively, which is comparatively low because
of its lower thermal conductivity. Rise in pulse on time and drops in servo voltage results thicker recast layer
because of high spark intensity and transfer of more amount of thermal energy to the material. This, in turn, causes
more volume of molten material to resolidify on the machined surface leading to thick recast layer on the machined
surface (Fig. 8).

Fig. 8. Recast Layer of finish cut (Exp. No: 5) and and rough cut (Exp. no: 23).

The surface textures comparison in fig 9 (a) & (b) reveals that the volume of the craters is expressively affected
by the pulse on time. The volume of craters formed in rough cut mode are much larger and deeper than those formed
in finish cut mode. From Fig. 9 (b), it is clearly evident from exp. 23 that, the machining with high pulse energy
mode exhibits rough and poor surface quality whereas low pulse on time machining provides smoother surface. It is
also observed from Fig. 9 that the work specimen with high pulse on time has more number of melted droplets,
globules of debris, craters, on the machined surface. However, low pulse energy machining has better surface
quality due to less melted drops less numbers of craters, fine pock-marks and small blow holes as illustrated in Fig.
9 (a).
 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524 7523

Fig. 9. Surface intigrity of a. finish cut (Exp. No: 5) b. rough cut (Exp. no: 23).

EDS analysis of two different machining conditions was used to check the element composition and its
percentage on the recast layer after WEDM. Fig. 10 (a) & (b) and Table 8 compares relative percentages of different
materials on W-EDMed surfaces by rough cut and finish cut modes. The elements present in the recast layer are
clearly indicated by the peaks of graph corresponding to their energy levels. Fig. 10, signifying that the carbon and
oxygen depositions are largely changed under the both machining condition. The depositions of O and C in finish
cut (exp. no 5) were found to be about 3.91% and 10.20%, respectively. Whereas, in rough cut (exp. no. 23) the
oxidation is very high which is reach to 17.03%. More oxides exist on this samples surface because machining is
performed with longer pulse on time. In general, ingress carbon from material and decomposed oxygen from water
vapour chemically react with the molten work material and form as a carbonized and oxidized recast layer.
It is evident from Fig. 10 (a) & (b) and Table 8 that the high contamination of foreign elements (Cu & Zn) was
appeared related to the recast layer which migrated from the brass electrode wire. Amount of Zn proportion was
much lower than that of Cu because of its minor ratio in the brass wire. Although mostly of the soften material like
Cu and Zn can be washed away by the dielectric flushing, but some still remained on the machined surface as a
recast layer. The inadequate hydrodynamics and recoil pressure of the layer allow the electrode debris to deposit in
and recast on the machined surface. Cu content was as high as 17.15% at high pulse on time and low servo voltage
(rough cut). Because of rise in discharge energy most of the molten material from the electrode brass was ejected by
high bubble pressure, the splashed molten and/or vaporized material from both the work-piece and the electrode
mixed with each other and redeposited on the machined surface. Cu content was less than 7% in the finish cut mode
because of drops in pulse on time and rise in servo voltage due to less erosion of wire electrode. At low discharge
energy, bubble pressure is low and the mixing of molten materials is less resulted in lower Cu and Zn deposition.

(a) (b)

Fig. 10. Elemental analysis of (a) finish cut (Exp. No: 5) and (b) rough cut (Exp. no: 23).
7524 Bikash Choudhuri et al/ Materials Today: Proceedings 5 (2018) 7515–7524

Table 8. Elemental percentage of finish surface (Exp. No: 5) and rough surface (Exp. no: 23)

Experiment no: 5 Experiment no: 23

Element Weight% Atomic% Element Weight % Atomic %
CK 3.91 15.75 CK 3.86 11.66
OK 10.20 25.94 OK 17.03 38.60
Cr K 13.45 9.96 Cr K 10.64 7.42
Ni K 6.51 4.27 Ni K 5.50 3.40
Cu K 6.63 4.02 Cu K 17.15 9.79
Zn K 2.00 1.17 Zn K 3.58 1.99
Mo L 1.69 0.68 Mo L 0.94 0.35
Mn K 0.73 0.51 Mn K 0.96 0.63
Si K 0.37 0.51 Co K 0.24 0.15
Fe K 54.48 38.17 Fe K 40.08 26.02
Totals 100.00 Totals 100.00

4. Conclusions

An Investigation of average surface roughness, and recast layer on the wire electric discharge machining of SS-
316 steel is carried out. Based on experimental study and data analysis, the effects of major control process
parameters (such as Ton, Toff, current, servo voltage, and wire tension) on each performance characteristics (Ra and
RL) are systematically investigated using ANOVA and optimize through ANN-PSO hybrid decision making tool.
The implication of this proposed approach under multi-response consideration can effectively assist engineers in the
modelling and optimizing for this stochastic type WEDM process. It was found that Ton of 0.35 µsec, Toff of 24 µsec,
pulse current of 170 Amp, voltage of 60 Volt and wire tension of 1.2 kg are the optimal combination of WEDM
parameters for the particular steel. Based on the ANOVA result, pulse on time was found to be the most dominating
factor for both roughness and recast layer. Higher the pulse on time results the rough surface, and thick recast layer.
Pulse off time and servo gap voltage were also found out to be important factors affecting recast layer, i.e. with
increased pulse off time and servo voltage the recast layer will decrease. The rough surfaces, craters, micro defects,
thick recast layers and inclusion of foreign particle on the recast surfaces were clearly observed in the high energy
rough cut mode.
Reference:
[1] K.Y. Song, C.N. Chu, Int. J. Precis. Eng. Manuf;, 14, (2013) 2061–2066.
[2] T.R. Newton, S.N. Melkote, T.R. Watkins, R.M. Trejo, L. Reister, Mater Sci Eng A-Struct; 13(2009), 208–215.
[3] M. Rajarshi, C. Shankar, S. Suman, Appl Soft Comput, 12 (2012), 2506–2516.
[4] S. Habib, G. Farhat, T. Amadou, G. Gonzalez, C. Braham, Int J Adv Manuf Technol, 65 (2013),141–153.
[5] J. Punturat, V. Tangwarodomnukun, C. Dumkum, Applied Surface Science, 320 (2014), 83–92.
[6] Ekmekci B. Appl. Surf. Sci. 253 (2007), 9234–9240.
[7] P. Sharma, D. Chakradhar, S. Narendranath, Materials and Design, 88 (2015), 558–566.
[8] S. Dewangan, S. Gangopadhyay, C.K. Biswas, Measurement, 63 (2015), 364–376.
[9] E. Weingärtner, F. Kuster, K. Wegener, Procedia CIRP, 2 (2012) 74 – 78.
[10] S. Dewangan, S. Gangopadhyay, C. K. Biswas, Measurement, 63 (2015) 364–376.
[11] I. Maher, A.D. Sarhan, M.M. Barzani, M. Hamdi, Journal of Cleaner Production, 108 (2015), 247-255.
[12] B. Kuriachen, K.P. Somashekhar, J. Mathew, Int J Adv Manuf Technol, 76 (2015), 91–104.
[13] P. Sharma, D. Chakradhar, S. Narendranath, Materials and Design, 88 (2015) 558–566.
[14] R.N. Yadav & V. Yadava,, Trans Indian Inst Met., 68 (2015) 981.
[15] D. Karaye, Journal of material processing technology, 209 (2009) 3125-3137.
[16] P. Sathiyaa, S. Aravindan, A.N. Haq, K. Paneerselvam, Journal of materials processing technology, 209 (2009) 2576–2584.
[17] T.R. Newton, S.N. Melkote, T.R. Watkins, R.M. Trejo, R. Reister, Materials Science and Engineering A, 513–514 (2009) 208–215.