Materials and Manufacturing Processes

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/lmmp20

Tri-objective constrained optimization of pulsating

DC sourced magnetic abrasive finishing process
parameters using artificial neural network and
genetic algorithm

Shadab Ahmad, Ranganath M. Singari & R.S. Mishra

To cite this article: Shadab Ahmad, Ranganath M. Singari & R.S. Mishra (2021) Tri-objective
constrained optimization of pulsating DC sourced magnetic abrasive finishing process parameters
using artificial neural network and genetic algorithm, Materials and Manufacturing Processes, 36:7,
843-857, DOI: 10.1080/10426914.2020.1866196

To link to this article: https://doi.org/10.1080/10426914.2020.1866196

Published online: 17 Feb 2021.

Submit your article to this journal

Article views: 140

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=lmmp20
MATERIALS AND MANUFACTURING PROCESSES
2021, VOL. 36, NO. 7, 843–857
https://doi.org/10.1080/10426914.2020.1866196

Tri-objective constrained optimization of pulsating DC sourced magnetic abrasive

finishing process parameters using artificial neural network and genetic algorithm
a
Shadab Ahmad , Ranganath M. Singaria,b, and R.S. Mishraa
a
Department of Mechanical Engineering, Delhi Technological University, Delhi, India; bDepartment of Design, Delhi Technological University, Delhi,
India

ABSTRACT ARTICLE HISTORY

Owing to the exceptional mechanical properties of Ti-6Al-4V, it is widely utilized in numerous critical Received 12 October 2020
mechanical parts for the uncompromised factor of safety. However, performing machining operations on Accepted 8 December 2020
this alloy in close tolerance is a challenging task. Moreover, establishing a process for its efficient finishing KEYWORDS
has become the interest of researchers. In this research study, the magnetic abrasive finishing process MAF; FMAB; ANN; genetic;
(MAF) has been studied using the ANN-GA approach, where ANN has been used for modeling of input– algorithm; microhardness;
output relations, and GA has been used to optimize the MAF process. The experiments were conducted elastic; indentation; surface;
on a pulsating DC sourced MAF set-up, and SiC-based loosely bonded magnetic abrasive media was used roughness; SiC; ti-6Al-4V
for material removal. During experimentation, the current, machining gap, speed of rotation, abrasive
composition, and finishing time were taken as input parameters being arranged in an array of L16
orthogonal. In contrast, output parameters were changed in surface roughness, change in the microhard­
ness, and change in the modulus of elastic indentation. ANN-GA approach provides a set of optimal
solutions for obtaining suitable output values. Furthermore, loosely bound SiC-based magnetic abrasive
media and its composition is found to be a very critical factor for the performance of the finishing quality
on Ti-6Al-4 V.

Introduction
Air traveling became popular from the 1980s, and since then, Components should have a micro-cracks-free surface, the fine-
rising demand for commercial aircraft has never ended; textured surface, along with dimensional conformity.[11,12]
because of this market need, more aircraft were to be manu­ Also, finishing of Titanium in close tolerance without any
factured, but rising fuel prices was the parallel issue.[1] Now the residual stress is not suitable by conventional grinding and
manufacturing industry is compelled to reduce the weight of polishing operation due to its hardness approximately four
aircraft, but with lives were at stake; thus, compromisation with times greater than steel.[10,13]
the factor of safety could never be accepted. High heat hard­ Magnetic abrasive finishing (MAF) is a cutting-edge
ness, fracture toughness, corrosion resistance, high strength-to machining technique where magnetic abrasive particles work
-weight ratio are few mechanical properties of mechanical under the influence of the magnetic field, and the cutting forces
parts that are expected by the aerospace, aircraft, and automo­ produced are too less that plowing action is absent; thus, the
tive industries .[2,3] Ultimately more focus on research and defect-free surface is obtained.[14,15] MAF Set-up is a low cost
development had to go with different materials, alloys, and because it needs no special designing considerations, rigidity,
composites, e.g., aluminum, steel, composite, and carbon vibration-free structure, etc.[9,16,17] Many researchers investi­
fiber, glass.[4,5] Material industries can develop and produce gated MAF process aiming Mechanics of process, effects of
the above discussed advanced engineering materials enriched process factor, nature and quality of surface produced, material
with novel properties that enable them to work in the corro­ removal rate (MRR), hybridization of the process to improve
sion-prone environment and under high temperature, high- MRR.[18–21] Hence MAF process can be a better alternative for
stress application.[6–8] Aluminum alloy and carbon fiber were the finishing of Ti-6Al-4V/Titanium (Grade 5) material.[22,23]
suitable for the high strength-to-weight ratio.[2] Apart from the In the literature survey, various attempts have been made by
structural part, the engine’s components such as blades, discs, the researchers to achieve desired outputs based on surface
shafts, and engine casings for the front fan to the rear end are integrity, surface finish, minor surface defects such as micro­
manufactured using Titanium and its alloys are widely used for cracks, etc., and to enhance the process efficiency with the help of
aircraft as a replacement having lighter weight (density approx. different abrasive media, hybridization of MAF process, and
60% that of steel), high hardness, high strength, and excellent manipulating the magnetic source and its intensity and period.
corrosion resistance replaced steel.[5,9,10] These Titanium- Those few attempts are discussed here, e.g., Yan et al. (2004)
based engine parts must have few checks before deploying Investigated that the MAF can effectively better the surface
into engine assembly for the durable and safe functioning. quality using unbonded magnetic abrasives (UMA) and MAF

CONTACT Shadab Ahmad [email protected] Department of Mechanical Engineering, Delhi Technological University, Delhi 110042, India.
© 2020 Taylor & Francis
844 S. AHMAD ET AL.

process, as well as micro-crack, can be avoided by applying the connected with the modern Taguchi approach, for example, it’s
same method.[24] Ahmad et al. (2017), studied the behavior of inability to be used for the simplification of multi-objective test
bonded magnetic abrasives produced by sintering process to sets.[33–37] Hence, the process variables of MAF were attempted
inspect the effects of process factors of the MAF process on to be optimized with iterative algorithms during the finishing of
Stainless Steel 202. Surface finish improvement was the prime the Ti-6Al-4V. In recent times some advanced algorithm has
motive of the study for which the magnetic abrasive particle size been developed combining ANN, for example evolutionary
significantly affects.[15] Amineh et al. (2013) Reported that MAF neural network algorithm (EvoNN), modified EvoNN, REVA,
could be a better choice for removal of the recast layer, which is Bi-objective Genetic programming (BioGP), etc.[38] Roy et al.
hard residues generated after WECDM. The most important (2020) have studied mechanical properties of alloyed steels using
highlight for this particular study was the subtraction of the evolutionary deep neural network.[39] Mondal et al. (2011) used
recast layer and micro-cracks improved by an increase of grit multi-objective genetic algorithm to optimize the most suitable
size.[25] Lin et al. (2007) that how the MAF process affected and condition for the separation of Cu-Zn .[40] The primary advan­
behaved during abrasion action on free-form surfaces of tages of these algorithms are their versatility to be applied to/
Stainless SUS304. Change of roughness was about 2.5 microns, versatility with regards to application over any type of con­
and the 10th part of microns was the order of finish achieved. It strained or unconstrained problems. Global search, optimiza­
was the order similar to mirror-like surfaces.[26] Yamaguchi tion, and generalization [,34,35,40] can be carried out by genetic
et al., 2012 worked on Uncoated carbide cutting tools for algorithms (GAs). A number of parametric issues in manufac­
Titanium (Grade 5) alloy machining. They had finished turing processes have also been solved by artificial neural net­
uncoated carbide tool rake and flank with magnetic abrasive works (ANNs), such as the recognition and learning patterns in
finishing.[27] Wang et al. (2009) performed MAF with a new any data set and the improvement of recognition capability,
abrasive media based on silicone gel, which was mixed with predict, cluster, or model patterns.[41,42] Here the feasibility of
Silicon Carbide (SiC) abrasive particles with ferromagnetic par­ integration ANN modeling with GA optimization was imple­
ticles and concluded that small mesh of particles produced a high mented to predict optimum MAF parameter conditions.
MRR. Still, simultaneously, an excellent surface finish was The development of artificial neural network and genetic
attained by the larger mesh number particles.[28] Jain et al. algorithm (ANN-GA), a method of modeling and optimization
(2008) studied the performance of pulsating current to make used to optimize a tri-objective problem, is defined in this
abrasive brush into a flexible one so that orientation keeps on research study and this optimization of the problem may be
changing for the use of the newer edge of abrasive. During static named as tri-objective genetic algorithm (TrioGA) optimiza­
mode of current, i.e., the Direct Current supplied to electro­ tion. The effect of process parameters on microhardness was
magnet provided static brush for all time when machining was studied, along with the surface roughness of the elastic inden­
done, due to this abrasive particle face dullness in the absence of tation module. Furthermore, the effect of loosely bound SiC-
orientation change, resulting in a low rate of finishing. based magnetic abrasive media and its composition on the
Performance enhancement in the process was concluded performance of the finishing quality on Ti-6Al-4 V has been
because of the regular orientation change of Flexible Magnetic studied.
Abrasive Brush (FMAB) during the switching mode of the
pulsating power supply.[29] The efficiency of the MAF process
Magnetic Abrasive Finishing (MAF)
was enhanced by Lee et al. (2013) using/with the help of the
vibrational motion to change the lay of the micro-machining MAF is a magnetic field-controlled operation capable of
mark. These marks were removed by altering the set-up to obtaining higher degrees of surface finish with very small/
vibration-assisted magnetic abrasive finishing (VAMAF) set narrow tolerances.[20] The idea behind this abrasive finishing
up.[30] Li et al. have developed a Mathematical model in which process relies on using a large amount of random cutting
MRR was a function of pressure and velocity. Their main focuses points with abrasive particles having an indefinite orientation
were the finishing media performance on finishing effects and and geometry. The abrasive particles inherently do not have
efficiency. The conclusions made were, for maximum flux den­ any magnetic character; subsequently, they are blended with
sity, the angle between poles should be at the right angle. Also, a ferromagnetic material, i.e., Iron powder, so they can be
there is a relation between ferromagnetic particles and abrasive suspended in the air utilizing a magnetic field for the formation
composition with the efficiency of the process .[31] of a Flexible Magnetic Abrasive Brush (FMAB) by which the
Conventional finishing methods such as grinding, honing are material surface is finished.[14] The motion of FMAB is used/
not capable to produce defect-free surfaces and it allows high manipulated for the effective exclusion of material with tinier
wear during the operation of the machine parts; it also results in chip sizes, as compared to the chips obtained during conven­
reduced component performance and life span.[32] Therefore, tional machining and finishing operation. The plane MAF
the use of an unconventional method like magnetic abrasive mechanism diagram is shown in Fig. 1.
finishing is becoming popular, and its parametric optimization In MAF, because of the normal force exerted by the magnetic
is even more crucial for the reduction of processing time and field, abrasive penetration action of the workpiece surface is
cost. There are different techniques used to optimize manufac­ obtained. The continuous revolution of the abrasive brush is
turing process parameters, such as Taguchi‘s methodology, sur­ further liable for material removal from the workpiece surface.
face response methodology, fuzzy logic, gray relations The finishing process requires the influence of a magnetic field
techniques, etc. Also in terms of convergence time, precision of between the electromagnet tool and target surface of workpiece.
tests, and so on, they have obvious disadvantages. The key issue These underlying forces (surrounding the finishing zone), which
 MATERIALS AND MANUFACTURING PROCESSES 845

Figure 1. Schematic diagram of the MAF process with loosely bonded magnetic abrasive media.

remove material from the surface of the workpiece, could be varying intensity at the pole end face. The electromagnet pole
clearly elucidated with the help of an equipotential line.[16] was positioned in such a way that their faces target the surface of
the workpiece, after that FMAB was formed by slowly inserting
the abrasive magnetic media in presence of the magnetic field in
Materials And Method the machining gap.[43,44] The separation between the facial por­
Workpiece material and abrasive media tion of the workpiece and the electromagnetic pole or face of the
tool is referred to as the working or machining gap.[18] Fig. 3(a).
Ti-6Al-4V/Titanium (Grade-5) was the workpiece material for shows the Radial drill-based MAF set-up and Fig. 2(b). shows the
the study to obtain the MAF performance over it. The elemental FMAB formed from the electromagnet used. During the
composition (wt.%) of the workpiece material was attained with machining process the very fine FMAB was produced which is
the help of the XRF (X-ray fluorescence) test, which was found to clearly illustrated in Fig. 3(b)., and Fig. 4 shows the magnetic
be below the recommended limit of the grade-5 category of abrasive media on the workpiece in the absence/dearth of the
Titanium and shown in Table 1. Also, the Vickers Hardness magnetic field. The flat workpiece was kept on a flat horizontal
test was done on the workpiece, and it was found around 387. fixture made by nylon and the electromagnet spindle was per­
Silicon Carbide and Iron particles with a lubricant mix were used mitted to rotate at speed of rotations of 112 rpm, 150 rpm,
as magnetic abrasive media.[14] This abrasive media comprises 220 rpm, and 300 rpm. For the Change in Surface Roughness
iron powder (300 mesh, a ferromagnetic material), Silicon (ΔRa) in “µm” change in the microhardness (ΔHV) in Vickers
Carbide (400 mesh), and lubricant (5W30, engine oil) to hold hardness, and change in the Modulus of Elastic Indentation
them together during the finishing. The various proportions of (ΔEIT) in “GPa,” were affected by variable factors and in the
the magnetic abrasive media in terms of abrasive composition in current MAF process, the process parameters are described in
weight percentage according to the design of experiments Table 3. The current, machining gap, speed of rotation, abrasive
employed in the current study has been reported in Table 2. composition percentage, and finishing time were considered as
the process parameters.

Experimental details
Artificial Neural NetworK (ANN)
In the current research study, an electromagnet with a pole end
face of 34 mm diameter connected by a DC pulse source of The change in Surface roughness values (ΔRa), change in the
power (for variation of current see Fig. 2). The electromagnet Microhardness (ΔHV), and change in the Modulus of Elastic
was provided a different current combination of 0.8 Amp, 1.2 Indentation (ΔEIT) between the initial and final finished sur­
Amp, 1.6, Amp, and 2.0 Amp generate a magnetic field with faces are modeled using the artificial neural network. The
ANNs are individually developed for the modeling of ΔRa,
ΔHV, and ΔEIT.
Table 1. The weight percent of Material composition (Ti-6Al-4V).
In this work, modeling for the MAF process using ANN
Elements Ti V Al Fe Mn Zn Mo architecture with a backpropagation algorithm was used pre­
% weight composition 92.04 4.43 3.01 0.33 0.19 0.031 0.026
cisely its output concerning input parameters. Modeling the
ANN works in stages: training, testing, and validation.[35,36]
The program codes were written for that purpose in MATLAB
Table 2. Composition of Magnetic abrasive media.
R2015a software. Experimental data were trained in the ANN
Constituents % weight concentration Mesh size
architecture, which is shown in Table 5. The neural network
Iron powder 60 50 40 30 300
Abrasive (SiO2) 40 50 60 70 400 contains three different layers of neurons, out of which the first
lubricant 10–20% of the total - layer includes neurons corresponding to input parameters.[41,45]
846 S. AHMAD ET AL.

Figure 2. Variation of current source recorded by DSOX2024A Oscilloscope: 200 MHz, 4 Analog Channels.

Figure 3. (a) Electromagnet mounted on a radial drill and (b) enlarged view of FMAB.

The input layer contains ‘5ʹ neurons synonymous to each of the hidden and output layers were Log-sigmoid and Tan-sigmoid,
input variables; the 2nd layer is called the hidden layer. For the respectively, which were calculated as given by Equation (1) and
determination of the number of hidden layers and the number of Equation (2) .[48–50] For the training, testing, and validation of
neurons in each of them for multilayer perceptrons, various ANN, 70%, 15%, and 15% data are used, respectively. Mean
hyperparameter tuning techniques may be used, ranging from square error algorithm was used for performance of the training.
sheer experimentation, referring to existing models for inspira­ Best validation performance has been attained at zero epoch
tion as well as optimization using algorithms such as Bayesian (shows number of passes the ANN training has performed
Algorithm.[46] In this case, we executed a for-loop that iterated over entire training dataset) with the value of 5.0058*10−7,
over different configurations of ANN architecture and plotted 4.5788*10−5 and 4.8021*10−12 for ΔRa, ΔHV and ΔEIT, respec­
the RMSE with respect to it. The configuration with the lowest tively, on mean square error vs epoch plot and the training
RMSE value was taken to be the most suitable for this case and persists until the 4th epoch for each training.
the final ANN was modeled based on this. Now, network has
10–8, 12–10, and 12–10 neurons, respectively, for (ΔRa) (ΔHV) 2
tansig ðnÞ ¼ 2n Þ
1 (1)
and (ΔEIT), as shown in Table 4. The 3rd layer has ‘1ʹ neuron ð1 þ e
corresponding to the single output value. For the present data
outputs, the “trainlm,” Levenberg-Marquardt training method 1
was used for quick supervised learning is easy, safe, and compu­ logsðnÞ ¼ (2)
ð1 þ e n Þ
tationally less expensive because of its adoptive learning and no-
line search technique.[47] The transfer functions selected for the where n is input for the function.
 MATERIALS AND MANUFACTURING PROCESSES 847

Figure 4. Media on the workpiece without the magnetic field.

Table 3. Parameters, units, definition, and levels.

Levels
Parameters Units Definition 1 2 3 4
Current (x1) Amp Current input to the electromagnet to alter the magnetic flux 0.8 1.2 1.6 2.0
Machining gap (x2) mm The separation between electromagnet tool face and target surface of workpiece 1.5 2.0 2.5 3.0
Speed of rotation (x3) rpm The relative motion between the workpiece and the tool 112 150 220 300
Abrasive composition (x4) wt. % The mixture of the abrasive and ferromagnetic particle 40 50 60 70
Finishing time (x5) min Finishing time for an individual experimental run 10 20 15 20

Table 4. ANN architecture for input to output modeling. Genetic Algorithm (GA) And Process Optimization
Input Output ANN architecture
It is observed from the ANN models developed and from the
x1, x2, x3, x4, x5 ΔRa 5–10-8-1
x1, x2, x3, x4, x5 ΔHV 5–12-10-1 literature that the output responses, [53] such as the ΔRa, ΔH,
x1, x2, x3, x4, x5 ΔEIT 5–12-10-1 and ΔEIT, are concurrent and harmonious in nature.
Therefore, the multi-optimal combination of process para­
meters would satisfy the objectives.[34,52,54] The Genetic
The ANN predicted, and experimental values ΔRa, ΔHV, Algorithm is able to offer better performance compared to
and ΔEIT comply to a very high degree with one another, and it traditional optimization techniques Because of its robustness,
is visible from Fig. 6,7, and 8. The maximum error comes out independence of gradient knowledge, and usage of intrinsic
between ANN predicted and experimental values for ΔRa, parallelism in design space searches.[55,56]
ΔHV, and ΔEIT equal to 0.1388%, 0.4387% 0.0353%, respec­ Genetic algorithm is a commercially viable, less complicated,
tively. The weights and biases of the developed model are and quicker for multi-objective optimization.[35] The optimization
stored after confirming with this high degree of maximum issue is resolved using this globally effective optimization method.
error. ANN architecture presented in Table 4 and the devel­ GA is a search and exploration algorithm similar to the mechan­
oped Neural Network architecture can be seen in Fig. 5. The ism of natural selection that belongs to the class of evolutionary
number of the hidden layers and the number of neurons in algorithm. GA optimizes the problems using strategies driven by
each hidden layer were chosen and then network efficiency was replication, mutation, crossover (recombination), and selection
calculated by percentage error .[51,52] The percentage error was phenomena.[57,58] To this intent, the ANN models developed in
measured as Equation (3) and the values determined are shown MATLAB were coupled with GA. Figure 9 describes the various
steps involved in the implementation of the ANN-GA model.
in Table 5. It is therefore guaranteed that the relationship
A random population of size ‘50ʹ was initialized by taking
between the values of input and output has been effectively
(5–10-8-1), (5–12-10-1), and (5–12-10-1) ANN models.
learned by the developed ANN model. It can thus be used to
optimize the operation.
Results And Discussion
Percentage error ¼
ðExperimental value ANN predicted valueÞ � 100 (3) The experimental results conducted during the investigation are
further used to develop a model of the MAF process using
Experimental value
artificial neural networks. The values of the output parameters
848 S. AHMAD ET AL.

Figure 5. ANN architecture.

Table 5. Experimental and ANN predicted results corresponding to L16 orthogonal array.
Input parameters and their values Experimental ANN Predicted Absolute error %
S.No x1 x2 x3 x4 x5 ΔRa (µm) ΔHV ΔEIT (GPa) ΔRa (µm) ΔHV ΔEIT (GPa) ΔRa (µm) ΔHV ΔEIT (GPa)
1 0.8 1.5 112 40 10 0.076 56.97 12.14 0.0759 56.9700 12.1400 0.1315 0 0
2 0.8 2 150 50 15 0.066 51.63 11.27 0.0660 51.6300 11.2699 0 0 0.0008
3 0.8 2.5 220 60 20 0.065 50.73 11.08 0.0650 50.7300 11.0800 0 0 0
4 0.8 3 300 70 25 0.065 51.18 11.18 0.0650 51.1800 11.1799 0 0 0.0008
5 1.2 1.5 150 60 25 0.118 89.29 19.50 0.1179 89.2900 19.4994 0.0847 0 0.0030
6 1.2 2 112 70 20 0.078 52.96 11.56 0.0779 52.9893 11.5600 0.1282 0.0553 0
7 1.2 2.5 300 40 15 0.090 80.71 17.26 0.0900 80.7004 17.2661 0 0.0118 0.0353
8 1.2 3 220 50 10 0.050 38.50 08.42 0.0499 38.5000 08.4199 0.2000 0 0.0011
9 1.6 1.5 220 70 15 0.109 83.18 18.17 0.1090 83.1800 18.1699 0 0 0.0005
10 1.6 2 300 60 10 0.095 79.58 17.38 0.0949 79.5800 17.3801 0.1052 0 0.0005
11 1.6 2.5 112 50 25 0.106 79.68 17.39 0.1060 79.6800 17.3900 0 0 0
12 1.6 3 150 40 20 0.087 70.14 15.12 0.0870 70.1400 15.1199 0 0 0.0006
13 2 1.5 300 50 20 0.160 133.59 29.15 0.1599 133.5900 29.1499 0.0625 0 0.0003
14 2 2 220 40 25 0.157 128.13 27.70 0.1570 128.1300 27.6999 0 0 0.0003
15 2 2.5 150 70 10 0.072 50.58 11.07 0.0719 50.3581 11.0716 0.1388 0.4387 0.0144
16 2 3 112 60 15 0.073 51.05 11.15 0.0729 51.0500 11.1500 0.1369 0 0

Figure 6. Comparison between experimental and ANN predicted ΔRa.

obtained after carrying out each experiment in the design of the input parameter on the output is observed. In section 6.1 the
experiment table are given in Table 5. The ANN models devel­ parametric analysis is summarized, whereas the findings of opti­
oped were used to predict the output for various input sets, and mization and related discussion are outlined in the next arti­
keeping other input parameters constant the influence of the cle 6.2.
 MATERIALS AND MANUFACTURING PROCESSES 849

Figure 7. Comparison between experimental and ANN predicted ΔHV.

Figure 8. Comparison between experimental and ANN predicted ΔEIT.

Parametric analysis using ANN models i.e., they decrease with the increase in the machining gap.
Their magnitude increases with the decline in the machin­
Here, the (5–10-8-1), (5–12-10-1), and (5–12-10-1) ANN mod­
ing gap. The reason for this observation is that on reducing
els developed for ΔRa, ΔHV, and ΔEIT, respectively, are used to
the machining gap, the magnetic field intensity amplifies,
comprehend the consequence of input process variables- cur­
which further results in the rise of the normal magnetic
rent, machining gap, speed of rotation, abrasive composition,
force and intensify the control over FMAB. Thus improving
and finishing time on the outputs i.e., ΔRa, ΔHV, and ΔEIT.
the finishing action. Since ΔHV and ΔEIT are harmonious
to ΔRa, the same effect has been observed on them, too.[59]
Effect of current
Figure 10, illustrates the effect of different current values on Effect of speed of rotation
ΔRa, ΔHV, and ΔEIT. It can be seen that as the current is Figure 12 shows the impact of speed of rotation on ΔRa, ΔHV,
incremented, the outputs ΔRa, ΔHV, and ΔEIT increase. This and ΔEIT. A notable rise in ΔRa can be observed with a minute
may be attributed to the subsequent increase in the magnetic decrease at a lower speed of rotation with the increase in the
field intensity, which helps in forming a stronger FMAB so that speed of rotation. An increase in the speed of rotation indicates
metal removing action by magnetic abrasive media becomes an improvement in machining efficiency. Moreover, it also
more active. A curve of the dome structure appears at 1.5 Amp signifies the removal of machined particles in the form of
in HV. This is because abrasive media are continually rubbing microchips, which in turn results in the decline of surface
and flushing the removed material. Precisely a heat treatment roughness value, i.e., an increment in the magnitude of ΔRa.
process is taking place.[43,44] At higher pressure (due to higher With the help of the trends, it can also be concluded that the
normal magnetic force), the workpiece interacts with a greater value of ΔHV and ΔEIT intensifies with the speed of rotation
number of abrasive particles, resulting in heat generation. At only up to 210 rpm, post which, their value starts to fall with
the same time, the heated abrasive media increasingly carries increasing the speed of rotation. It is also observed that the rate
out the heat produced. of decrement of ΔEIT is greater than ΔHV.

Effect of the machining gap Effect of the abrasive composition

A relationship between the machining gap with ΔRa, ΔHV, The impact of the abrasive composition on ΔRa, ΔHV, and ΔEIT
and ΔEIT is shown in Fig. 11. The inverse proportionality is shown in Fig. 13. It is observed that ΔRa attained its maximum
of the output parameters to the machining gap can be seen, value in the range 55–58 abrasive composition and later
850 S. AHMAD ET AL.

Figure 9. Flow chart of ANN-GA model.

Figure 10. Effect of current on ΔRa, ΔHV, and ΔEIT.

decreased on the increase in abrasive composition with a slight Effect of finishing time
decrease in lower values of abrasive compositions. For ΔHV, it is The impact of finishing time on ΔRa, ΔHV, and ΔEIT is illu­
noticed that on increasing abrasive composition, ΔHV value strated in Fig. 14. It is noted that as the finishing time increases,
rises. ΔEIT decreases on increasing the abrasive composition. ΔRa increases. Also, it is found that after 16 minutes of finishing
 MATERIALS AND MANUFACTURING PROCESSES 851

Figure 11. Effect of machining gap on ΔRa, ΔHV, and ΔEIT.

Figure 12. Effect of speed of rotation on ΔRa, ΔHV, and ΔEIT.

time, the ΔRa decreases abruptly because of higher grinding as per the objective, in MATLAB-15a for this purpose. The
efficiency that leads to significant material removal rates in goals of the present research are to optimize change in surface
confined zones, which further formed more profound valleys. roughness, change in microhardness, and change in elastic
Also ΔHV, and ΔEIT showed an increasing pattern with finish­ indentation modulus. The objectives are as shown -
ing time, where a slight decrease in ΔEIT initially on the lower Objective 1 = Minimize (1/ΔRa)
side of finishing time.[60] Objective 2 = Minimize (1/ΔHV)
Objective 3 = Minimize (1/ΔEIT)
The program is coded in MATLAB for the GA-based opti­
Multi-objective optimization using ANN-GA mization of the ANN model as per objectives. During this
To optimize (maximize) the change in surface roughness, optimization, GA was applied to the trained (5–10-8-1, 5–12-
change in microhardness, and change in the modulus of elastic 10-1, and 5–12-10-1) ANN models for the determination of the
indentation, the developed ANN models are fed to a genetic objectives function. The ANN component is responsible for
algorithm program that calculates the value of each input deciding the objective functions in the process of combined
parameter corresponding to the maximum output value optimization by the ANN-GA model, while the GA aspect is
achievable. This required goals were framed as objective func­ responsible for ranking and sorting the ANN-based solution.
tions, for optimizing the MAF process using GA. Furthermore, The developed ANN models provides the initial population of
the developed ANN models were linked with GA, for the size 50 in the first iteration of optimization, the crossover rate,
present tri-objective optimization. A GA program was coded, and the mutation rate is 0.8 and 0.01, respectively. List of
852 S. AHMAD ET AL.

Figure 13. Effect of abrasive composition on ΔRa, ΔHV, and ΔEIT.

Figure 14. Effect of finishing time on ΔRa, ΔHV, and ΔEIT.

parameters of selection for GA operation have been illustrated Optimal solutions presented in Table 7 have decision variables
in Table 6, and it is used in the subsequent iteration to measure (bound by finishing setup restriction limit). Optimized responses
the outputs viz. ΔRa, ΔHV, and ΔEIT for new offspring. are compared with experimental values of responses in the section
During optimization process, it is found that after 148 itera­ 6.2.1, section and section 6.2.2, section 6.2.3. It may hence be
tions, the optimum data are obtained. No noticeable improve­ concluded that the preference of one solution over another solely
ment in process parameters was observed after that if the depends on the product specifications, processing capabilities
iterations reached 148 iterations. resources, and the predilection of the process engineer. These 18
The Pareto fronts hence obtained were extracted for further solutions have been presented in Table 7.
analysis, and the optimal front of these 18 solutions are illustrated
in Fig. 15. Selected five decision variable are bound by upper and Comparison of experimental and optimal ΔRa
lower limit as per restriction of magnetic abrasive finishing set-up A juxtaposition of experimental ΔRa values and the values found
discussed in section 3, keeping in mind that all the responses have from Pareto optimal solution revealed that the maximum experi­
similar in nature as discussed in section 6.2 and in section 6.1.2. mental value of ΔRa is 0.160 μm at current = 2A, machining
gap = 1.5 mm, speed of rotation = 300 rpm, abrasive composi­
tion = 50%, and finishing time = 20 min, as shown in experiment
Table 6. List of GA parameters used. number 13 in Table 5. Whereas the maximum ΔRa obtained from
Population Crossover Mutation Max. Initial optimal solution is 0.160 μm at current = 1.99 A, machining
Type Fraction Rate Generations Population gap = 1.51 mm, speed of rotation = 223.55 rpm, abrasive compo­
Double 0.8 0.01 148 50 sition = 53.65% and finishing time = 20.77 min, as illustrated in
vector Table 6 (S.No 2). It can be concluded from these results that
 MATERIALS AND MANUFACTURING PROCESSES 853

Figure 15. Pareto optimal front (a) ΔRa vs ΔHV (b) ΔHV vs ΔEIT (c) ΔRa vs ΔEIT.

Table 7. Set of optimal solution corresponding to inputs and outputs.

Machining gap Speed of rotation Abrasive
S.No ΔRa (µm) ΔHV ΔEIT (GPa) Current (Amp.) (mm) (rpm) Composition (wt. %) Finishing time (min)
1 0.157 144.61 28.01 1.99 1.53 224.00 42.25 24.60
2 0.160 127.88 28.08 1.99 1.51 223.55 53.65 20.77
3 0.152 123.33 29.74 1.99 1.51 224.17 47.02 16.46
4 0.150 127.86 29.66 1.99 1.51 223.18 44.66 16.86
5 0.158 136.07 27.35 1.99 1.52 223.66 51.99 23.95
6 0.156 137.82 28.92 1.99 1.51 223.99 44.23 20.33
7 0.159 135.52 27.55 1.99 1.52 223.36 51.92 23.35
8 0.154 123.85 29.65 1.99 1.51 222.96 48.43 17.34
9 0.151 128.51 29.65 1.99 1.51 223.49 44.77 17.13
10 0.155 118.51 29.58 1.99 1.51 223.17 50.37 16.66
11 0.152 130.55 29.51 1.99 1.52 223.58 44.61 17.92
12 0.156 137.30 28.91 1.99 1.51 223.90 44.86 20.39
13 0.159 134.48 27.67 1.99 1.51 223.21 52.13 22.76
14 0.158 137.64 28.39 1.99 1.51 223.78 47.78 21.84
15 0.154 120.97 29.68 1.99 1.51 222.90 48.89 16.68
16 0.155 128.19 29.57 1.99 1.51 223.59 47.30 18.13
17 0.151 127.36 29.68 1.99 1.51 223.12 45.43 17.07
18 0.152 122.95 29.74 1.99 1.51 224.17 47.27 16.38

though the current, machining gap and finishing time are almost Comparison of experimental and optimal ΔHV
the same, there has been a significant drop (25.48%) in the speed When compared with the Pareto optimal solution, the max­
of rotation upon the change of the speed of rotation from 300rpm imum value of experimental ΔHV was 133.59 at current 2 A,
to 223.55rpm. Due to this, the abrasive composition has increased machining gap 1.5 mm, speed of rotation 300 rpm, abrasive
from 50% to 53.65% in the optimal solution, even though the ΔRa composition 50% and finishing time 20 min, corresponding to
obtained from the optimal solution is the same. This proves that the experiment number 13 in Table 5. Furthermore, the max­
a calculated selection of process parameters can provide huge imum ΔHV obtained from the optimal solution was 144.61 at
savings in energy consumption as speed of rotation is decreased 1.99 A current, 1.53 mm machining gap with 224.00 rpm speed
for the same output and effective abrasive composition. of rotation, and 42.25% abrasive composition taking a finishing
Furthermore, it can be seen that a combination of high current, time of 16.68 minutes as demonstrated in Table 6. It is to be
less machining gap, moderate speed of rotation, and abrasive observed that with the same current and machining gap,
composition, and sufficient finishing time is more suitable for a decrease of 25.34% in speed of rotation (from 300.00 rpm
achieving a higher ΔRa. It can be clearly noted from Table 5 to 224.00 rpm) and a marginal decrease in abrasive composi­
(Exp. 8) that a lower experimental value of ΔRa (=0.050 μm) can tion from 50.00% to 42.25% was there. However, the finishing
be attributed to a current = 1.2 A, machining gap = 3 mm, speed time was higher during the optimal solution (24.60 minutes) as
of rotation = 220 rpm, abrasive composition = 50% and finishing compared to the experimental one (20.00 minutes). A Prime
time = 10 min, while the lower ΔRa obtained from solution set is indication is that a careful selection of process parameters can
0.150 μm at current = 1.99 A, machining gap = 1.51 mm, speed of improve the ΔHV by 8.24%. The combination of high current,
rotation = 223.18 rpm, abrasive composition = 44.66% and finish­ machining gap, and finishing time, with moderate speed of
ing time = 16.86 min (Table 6, S.No 4). The results show that this rotation, and low abrasive composition has been found to be
value of lower ΔRa corresponding to the optimal solution is more appropriate for attaining higher ΔHV. In Table 5
significantly higher (nearly three times) than the experimental (Exp. 8), it has been demonstrated that the minimum experi­
value. This may be attributed to the formation of a strong mental value of ΔHV was 38.50 at a current of 1.2 A, machin­
FMAB with a higher current and lower machining gap. Also, an ing gap of 3.00 mm along with 220.00 rpm speed of rotation
increase in the speed of rotation improved the metal removal and 50.00% abrasive composition taking the finishing time
action, and 6.86 min more finishing time. 10.00 minutes. Moreover, it is shown in Table 6 (S.No. 15)
854 S. AHMAD ET AL.

that the minimum value of ΔHV found from the solution set set is 27.35 GPa at current = 1.99 A, machining gap = 1.52 mm,
was 120.97 at a current of 1.99 A, machining gap of 1.51 mm speed of rotation = 223.66 rpm, abrasive composition = 51.99%
with a speed of rotation of rotation of 222.90 rpm and 48.89% and finishing time = 23.95 min as shown in Table 6 (S.No. 5).
abrasive composition taking a finishing time of 16.68 minutes. The lowest ΔEIT thus obtained from the optimal solution is
Based on the data obtained, it can be concluded that the significantly higher (nearly 2.5 times) than the experimental
minimum value of ΔHV from the optimal solutions is around value. This significantly better result may be attributed to
three times higher than its experimental value. a slight increase in speed of rotation and 6.86 min longer
finishing time, which improved metal removal.[20]
Comparison of experimental and optimal ΔEIT
According to the solution from the Pareto Front, the maximum
Surface morphology and roughness
experimental ΔEIT value from Table 5 (Exp. 13) is 29.15 GPa at
current = 2 A, machining gap = 1.5 mm, speed of rotation = To comprehend the surface wear pattern at the micro-nano
300 rpm, abrasive composition = 50% and finishing time = level, the Scanning electron microscopy and Atomic force
20 min. Meanwhile, according to the optimal solution, the microscopy of the workpiece surface checked on various
maximum ΔEIT obtained is 29.74 GPa at current = 1.99 A, experimental conditions of the MAF process and shown in
machining gap = 1.51 mm, speed of rotation = 224.17 rpm, Fig. 16(a,b,c) and in Fig. 17(a,b,c) respectively. The FMAB
abrasive composition and finishing time is 47.02% and has substantially sheared off the peaks of the preprocessed
16.46 min or 47.27% and 16.38 min, respectively, as shown in ground workpiece, as can be seen in Fig. 16(b,c). The behavior
Table 6 (S.No 3 and 18). By comparing the optimal solution of the FMAB fluctuates more at lower current and less abrasive
with experimental values, it is noted that the current, machin­ composition. Hence, some pores and deep scratches that
ing gap, are almost the same. Still, the speed of rotation is damaged the finished surface are also observed in Sample 1 at
almost 25% less (from 300 rpm to 224.17 rpm), abrasive com­ voltage = 0.8 V, abrasive composition = 40 (Fig.s 16a), it can
position and finishing time are slightly lower (from 50% to 47% also be seen in same the figure that a groove begins to form
and 20 min to 16 min, respectively). The ΔEIT attained from because of the bulge of magnetic abrasive media in a confined
the optimal solution is much higher, which supports the thesis area. The reason for the formation of the bulge is concentrated
that selection of the right process parameters will enormously magnetic flux, which is either because of nonuniform coil
save on energy as speed of rotation is decreased for the same winding or non-homogeneous electromagnet core material.
output and effective abrasive composition.[15] In Table 5 (Exp. The surface texture obtained, as shown in (Fig. 16(b,c)), is
15), the lowest experimental value of ΔHV is 11.07 GPa at flatter with less groove depth resulting in a low Ra (finished
current = 2 A, machining gap = 2.5 mm, speed of rotation = surface) value. Owing to the variable intensity of FMAB and
150 rpm, abrasive composition = 70% and finishing time = abrasive material contact with workpiece surface points, the
10 min, whereas the lowest ΔEIT value obtained from solution peak heights are non-uniform. Figure 16(b) shows that the

Figure 16. SEM images in various experimental conditions (a) voltage = 0.8 V; machining gap = 1.5 mm; speed of rotation =112 rpm; abrasive composition. =40;
time = 10 min; (b) voltage = 1.6 V; machining gap =1.5 mm; speed of rotation =220 rpm; abrasive composition. =70; time = 15 min; (c) voltage = 2.0 V; machining
gap =1.5 mm; speed of rotation =300 rpm; abrasive composition = 50; time = 20 min.

Figure 17. AFM images of workpiece in various experimental conditions (a) voltage = 0.8 V; machining gap = 1.5 mm; speed of rotation =112 rpm; abrasive
composition = 40; time = 10 min; (b) voltage = 1.6 V; machining gap = 1.5 mm; speed of rotation = 220 rpm; abrasive composition = 70; time = 15 min; (c)
voltage = 2.0 V; machining gap = 1.5 mm; speed of rotation = 300 rpm; abrasive composition = 50; time = 20 min.
 MATERIALS AND MANUFACTURING PROCESSES 855

potential finishing mechanism for Ti-6Al-4 V is the formation decrease with a further increase in the speed of rotation.
of microchips by shearing and also, to some degree, a brittle For ΔHV, it is noticed that with an increase in the
fracture. Figure 16(b,c) show that there is leftover abrasive percentage of abrasive in the mixture, ΔHV value
media that is micro-welded. The reason for such leftover rises. It was noted that ΔRa increases with the increase
micro-welded magnetic abrasive media is the generation of of finishing time. Also, ΔHV showed an increasing
enough heat in concentrated regions at higher speed of trend with finishing time.
rotations.[35,42] Fig. 18. Shows an improvement in surface (4) ΔEIT increases with increasing value of current and
roughnesses at various stages. decreases with increasing machining gap. In addition
to this, the magnitude of all the output parameters is
more when the machining gap is lower. It is experi­
Conclusion enced from the results and discussion that ΔEIT value
In this research work, the ANN-GA approach has been used increases with the speed of rotation (up to 210 rpm) and
for tri-objective optimization of the pulsating DC sourced then begins to decrease as the speed of rotation
magnetic abrasive finishing process parameters. After increases further. Furthermore, if the abrasive composi­
a critical observation of the results obtained, it can be surmised tion is increased, the ΔEIT decreases. It has been found
that: that, with the rise in finishing time, ΔEIT increases.
ΔEIT showed an increasing trend with finishing time,
(1) The ANN model, which has been developed, can pre­ where there was initially a slight decrease in ΔEIT on
dict ΔRa, ΔHV, and ΔEIT with the maximum error the lower side of finishing time, but a constant upward
equal to 0.1388%, 0.4387% 0.0353%, respectively. trend was observed thereafter.
(2) ΔRa is directly proportional to current and inversely (5) It is observed that the ANN models could predict the
proportional to the machining gap. It is noticed that behavior of the output parameters with very high accu­
although co-Ra increases significantly on increasing the racy and efficiency. Upon further optimization with
speed of rotation, co-Ra at lower speed of rotation, GA, the obtained ANN models were highly optimized
however, decreases slightly. Also, It has been seen that and provided a better approach for the prediction of
ΔRa attains its maximum value in the range of 55–58 higher values of ΔRa, ΔHV, and ΔEIT than their experi­
abrasive composition and later decreases when the mental. ANN-GA-based modeling and optimization
abrasive composition was increased with a slight suggest a direct relation between choosing process para­
decrease in lower values of abrasive compositions. meters correctly and enhancing machining
Moreover, ΔRa increases with an increase in the finish­ performance.
ing time, and on the further increase, after 16 minutes (6) Loosely bound SiC-based magnetic abrasive media is
of finishing time, the ΔRa value decreases abruptly. found to be a reasonable abrasive media for the finish­
(3) ΔHV increases with an increment in the current and ing of Ti-6Al-4V by the MAF process. The composition
decreases with an increase in the machining gap. of ferromagnetic material and abrasive particles in the
Furthermore, the magnitude of all the output para­ media is found to be a very critical factor for the per­
meters is more when the machining gap is lower. It is formance of the finishing quality.
observed that ΔHV and ΔEIT values increase with the (7) Pareto’s optimal front offers a set of optimal solutions
speed of rotation (up to 210 rpm), and then they start to (Table 6). The table acts as a convenient solution for/

Figure 18. Comparison between surface roughnesses.

856 S. AHMAD ET AL.

facilitates a process engineer for the selection of the [15] Ahmad, S.; Gangwar, S.; Yadav, P. C.; Singh, D. K. Optimization of
optimum values of control parameters depending on Process Parameters Affecting Surface Roughness in Magnetic
the product requirements. Abrasive Finishing Process. Mater. Manuf. Process. 2017, 32(15),
1723–1729. DOI: 10.1080/10426914.2017.1279307.
[16] Fox, M.; Agrawal, K.; Shinmura, T.; Komanduri, R. Magnetic
Abrasive Finishing of Rollers. CIRP Ann. 1994, 43(1), 181–184.
Acknowledgments DOI: 10.1016/S0007-8506(07)62191-X.
[17] Barman, A.; Design, D. M. Fabrication of a Novel Polishing Tool
The authors thank the “Council of Scientific and Industrial Research for Finishing Freeform Surfaces in Magnetic Field Assisted
(CSIR) - Government of India,” for financial support in the form of SRF. Finishing (MFAF) Process. Precis. Eng. 2017, 49, 61–68. DOI:
10.1016/j.precisioneng.2017.01.010.
[18] Jain, V. K.; Kumar, P.; Behera, P. K.; Jayswal, S. C. Effect of
ORCID Working Gap and Circumferential Speed on the Performance of
Magnetic Abrasive Finishing Process. Wear. 2001, 250(1–12),
Shadab Ahmad http://orcid.org/0000-0001-7468-8686 384–390. DOI: 10.1016/S0043-1648(01)00642-1.
[19] Chang, G.-W.; Yan, B.-H.; Hsu, R.-T. Study on Cylindrical
Magnetic Abrasive Finishing Using Unbonded Magnetic
Abrasives. Int. J. Mach. Tools Manuf. 2002, 42(5), 575–583. DOI:
References 10.1016/S0890-6955(01)00153-5.
[20] Singh, D. K.; Jain, V. K.; Raghuram, V. Parametric Study of
[1] Pollock, T. M.;. Alloy Design for Aircraft Engines. Nat. Mater. Magnetic Abrasive Finishing Process. J. Mater. Process. Technol.
2016, 15(8), 809–815. DOI: 10.1038/nmat4709. 2004, 149(1–3), 22–29. DOI: 10.1016/j.jmatprotec.2003.10.030.
[2] Dursun, T.; Soutis, C. Recent Developments in Advanced Aircraft [21] Chaurasia, A.; Rattan, N.; Mulik, R. S. Magnetic Abrasive
Aluminium Alloys. Mater. Des. 1980-2015, 2014(56), 862–871. Finishing of AZ91 Magnesium Alloy Using Electromagnet.
doi:10.1016/j.matdes.2013.12.002. J. Braz. Soc. Mech. Sci. Eng. 2018, 40(10), 482. DOI: 10.1007/
[3] El-Eskandarany, M. S.;. Mechanical Alloying for Fabrication of s40430-018-1399-7.
Advanced Engineering Materials; Noyes Publications: Norwich, [22] Zhou, K.; Chen, Y.; Du, Z. W.; Niu, F. L. Surface Integrity of
N.Y, 2001. Titanium Part by Ultrasonic Magnetic Abrasive Finishing. Int.
[4] Barthelat, F.;. Architectured Materials in Engineering and Biology: J. Adv. Manuf. Technol. 2015, 80(5–8), 997–1005. DOI: 10.1007/
Fabrication, Structure, Mechanics and Performance. Int. Mater. Rev. s00170-015-7028-z.
2015, 60(8), 413–430. DOI: 10.1179/1743280415Y.0000000008. [23] Srivastava, A.; Kumar, H.; Singh, S. Investigations into Internal
[5] Srinivasu, G.; Natraj, Y.; Bhattacharjee, A.; Nandy, T. K.; Surface Finishing of Titanium (Grade 2) Pipe with Extended
Nageswara Rao, G. V. S. Tensile and Fracture Toughness of High Magnetic Tool. Procedia Manuf. 2018, 26, 181–189. DOI:
Strength β Titanium Alloy, Ti–10V–2Fe–3Al, as a Function of 10.1016/j.promfg.2018.07.025.
Rolling and Solution Treatment Temperatures. Mater. Des. 2013, [24] Yan, B.-H.; Chang, G.-W.; Chang, J.-H.; Hsu, R.-T. Improving
47, 323–330. DOI: 10.1016/j.matdes.2012.11.053. Electrical Discharge Machined Surfaces Using Magnetic Abrasive
[6] Jain, V. K.;. Advanced Machining Processes; Allied Publishers: New Finishing. Mach. Sci. Technol. 2004, 8(1), 103–118. DOI: 10.1081/
Delhi, 2013. MST-120034246.
[7] Kovaliova, S.; Šepelák, V.; Grigoreva, T.; Zhornik, V.; Kiseleva, T.; [25] Khalaj Amineh, S.; Fadaei Tehrani, A.; Mohammadi, A. Improving
Khomich, M.; Devyatkina, E.; Vosmerikov, S.; Vityaz, P.; the Surface Quality in Wire Electrical Discharge Machined
Lyakhov, N. Mechanosynthesis of Composites in Chemically Specimens by Removing the Recast Layer Using Magnetic
Non-Reacting and Exothermically Reacting Systems for Magnetic Abrasive Finishing Method. Int. J. Adv. Manuf. Technol. 2012, 66
Abrasive Media. J. Mater. Sci. 2018, 53(19), 13560–13572. DOI: (9–12), 1793–1803. DOI: 10.1007/s00170-012-4459-7.
10.1007/s10853-018-2463-5. [26] Lin, C.-T.; Yang, L.-D.; Chow, H.-M. Study of Magnetic Abrasive
[8] Mozammil, S.; Karloopia, J.; Verma, R.; Jha, P. K. Effect of Varying Finishing in Free-Form Surface Operations Using the Taguchi
TiB2 Reinforcement and Its Ageing Behaviour on Tensile and Method. Int. J. Adv. Manuf. Technol. 2007, 34(1–2), 122–130.
Hardness Properties of In-Situ Al-4.5%Cu-XTiB2 Composite. DOI: 10.1007/s00170-006-0573-8.
J. Alloys Compd. 2019, 793, 454–466. DOI: 10.1016/j. [27] Yamaguchi, H.; Srivastava, A. K.; Tan, M. A.; Riveros, R. E.;
jallcom.2019.04.137. Hashimoto, F. Magnetic Abrasive Finishing of Cutting Tools for
[9] Guo, J.; Au, K. H.; Sun, C.-N.; Goh, M. H.; Kum, C. W.; Liu, K.; Machining of Titanium Alloys. CIRP Ann. 2012, 61(1), 311–314.
Wei, J.; Suzuki, H.; Kang, R. Novel Rotating-Vibrating Magnetic DOI: 10.1016/j.cirp.2012.03.066.
Abrasive Polishing Method for Double-Layered Internal Surface [28] Wang, A. C.; Lee, S. J. Study the Characteristics of Magnetic
Finishing. J. Mater. Process. Technol. 2019, 264, 422–437. DOI: Finishing with Gel Abrasive. Int. J. Mach. Tools Manuf. 2009, 49
10.1016/j.jmatprotec.2018.09.024. (14), 1063–1069. DOI: 10.1016/j.ijmachtools.2009.07.009.
[10] Kotobi, M.; Honarpisheh, M. Through-Depth Residual Stress [29] Jain, V. K.; Singh, D. K.; Raghuram, V. Analysis of
Measurement of Laser Bent Steel–Titanium Bimetal Sheets. Performance of Pulsating Flexible Magnetic Abrasive Brush
J. Strain Anal. Eng. Des. 2018, 53(3), 130–140. DOI: 10.1177/ (P-FMAB). Mach. Sci. Technol. 2008, 12(1), 53–76. DOI:
0309324717753212. 10.1080/10910340701883538.
[11] Carter, T. J.;. Common Failures in Gas Turbine Blades. Eng. Fail. [30] Lee, Y.-H.; Wu, K.-L.; Jhou, J.-H.; Tsai, Y.-H.; Yan, B.-H. Two-
Anal. 2005, 12(2), 237–247. DOI: 10.1016/j.engfailanal.2004.07.004. Dimensional Vibration-Assisted Magnetic Abrasive Finishing of
[12] Fu, Y.; Gao, H.; Wang, X.; Guo, D.; Machining the Integral Stainless Steel SUS304. Int. J. Adv. Manuf. Technol. 2013, 69(9–12),
Impeller and Blisk of Aero-Engines: A Review of Surface 2723–2733. DOI: 10.1007/s00170-013-5242-0.
Finishing and Strengthening Technologies. Chin. J. Mech. Eng. [31] Li, W.; Li, X.; Yang, S.; Li, W. A Newly Developed Media for
2017, 30(3), 528–543. DOI:10.1007/s10033-017-0123-3. Magnetic Abrasive Finishing Process: Material Removal Behavior
[13] Yadroitsev, I.; Yadroitsava, I. Evaluation of Residual Stress in and Finishing Performance. J. Mater. Process. Technol. 2018, 260,
Stainless Steel 316L and Ti6Al4V Samples Produced by Selective 20–29. DOI: 10.1016/j.jmatprotec.2018.05.007.
Laser Melting. Virtual Phys. Prototyp. 2015, 10(2), 67–76. DOI: [32] Zhu, Y.; Chen, X.; Zou, J.; Yang, H. A. Study on the Influence of
10.1080/17452759.2015.1026045. Surface Topography on the Low-Speed Tribological Performance
[14] Singh, D. K.; Jain, V. K.; Raghuram, V. On the Performance of Port Plates in Axial Piston Pumps. Wear. 2015, 338–339,
Analysis of Flexible Magnetic Abrasive Brush. Mach. Sci. Technol. 406–417. DOI: 10.1016/j.wear.2015.07.016.
2005, 9(4), 601–619. DOI: 10.1080/10910340500398217.
 MATERIALS AND MANUFACTURING PROCESSES 857

[33] Roy, R. K. A.;. Primer on the Taguchi Method, 2nd ed.; Society of Bi, Y., Bhatia, R., Kapoor, S., Eds.; Springer International
Manufacturing Engineers: Dearborn, MI, 2010. Publishing: Cham, 2020; pp 490–500.
[34] Ajith Arul Daniel, S.; Pugazhenthi, R.; Kumar, R.; [47] Transtrum, M. K.; Machta, B. B.; Sethna, J. P. Geometry of
Vijayananth, S. Multi Objective Prediction and Optimization of Nonlinear Least Squares with Applications to Sloppy Models and
Control Parameters in the Milling of Aluminium Hybrid Metal Optimization. Phys. Rev. E. 2011, 83(3), 036701. DOI: 10.1103/
Matrix Composites Using ANN and Taguchi -grey Relational PhysRevE.83.036701.
Analysis. Def. Technol. 2019, 15(4), 545–556. DOI: 10.1016/j. [48] Gluck, M. A.; Myers, C. E. Gateway to Memory: An Introduction to
dt.2019.01.001. Neural Network Modeling of the Hippocampus and Learning; Issues
[35] Singh, R. K.; Gangwar, S.; Singh, D. K.; Pathak, V. K. A Novel in Clinical and Cognitive Neuropsychology; MIT Press: Cambridge,
Hybridization of Artificial Neural Network and Moth-Flame Mass, 2001.
Optimization (ANN–MFO) for Multi-Objective Optimization [49] Shanmuganathan, S. Artificial Neural Network Modelling: An
in Magnetic Abrasive Finishing of Aluminium 6060. J. Braz. Introduction. In Artificial Neural Network Modelling;
Soc. Mech. Sci. Eng. 2019, 41(6), 270. DOI: 10.1007/s40430-019- Shanmuganathan, S., Samarasinghe, S., Eds.; Springer
1778-8. International Publishing: Cham, 2016; pp 1–14. DOI: 10.1007/
[36] Ahmad, S.; Singari, R. M.; Mishra, R. S. Modelling and 978-3-319-28495-8.
Optimisation of Magnetic Abrasive Finishing Process Based on a [50] Rasmussen, C. E.; Williams, C. K. I. Gaussian Processes for Machine
Non-Orthogonal Array with ANN-GA Approach. Trans. IMF. Learning; Adaptive Computation and Machine Learning; MIT
2020, 98(4), 186–198. DOI: 10.1080/00202967.2020.1776966. Press: Cambridge, Mass, 2006.
[37] Candan, G.; Yazgan, H. R. Genetic Algorithm Parameter [51] Stojanović, B.; Vencl, A.; Bobić, I.; Miladinović, S.; Skerlić, J.
Optimisation Using Taguchi Method for a Flexible Experimental Optimisation of the Tribological Behaviour of Al/
Manufacturing System Scheduling Problem. Int. J. Prod. Res. SiC/Gr Hybrid Composites Based on Taguchi’s Method and
2015, 53(3), 897–915. DOI: 10.1080/00207543.2014.939244. Artificial Neural Network. J. Braz. Soc. Mech. Sci. Eng. 2018, 40
[38] Roy, S.; Chakraborti, N. Development of an Evolutionary Deep (6), 311. DOI: 10.1007/s40430-018-1237-y.
Neural Net for Materials Research. In TMS 2020 149th annual [52] Yadav, R. N.; Yadava, V.; Singh, G. K. Multi-Objective
meeting & exhibition supplemental proceedings, Springer Optimization of Process Parameters in Electro-Discharge
International Publishing: Cham, 2020; pp817–828. Diamond Face Grinding Based on ANN-NSGA-II Hybrid
[39] Roy, S.; Saini, B. S.; Chakrabarti, D.; Chakraborti, N. Technique. Front. Mech. Eng. 2013, 8(3), 319–332. DOI: 10.1007/
Mechanical Properties of Micro-Alloyed Steels Studied Using s11465-013-0269-3.
a Evolutionary Deep Neural Network. Mater. Manuf. Process. [53] Chen, L.; Ahadi, A.; Zhou, J.; Ståhl, J.-E. Modeling Effect of Surface
2020, 35(6), 611–624. DOI: 10.1080/10426914.2019.1660786. Roughness on Nanoindentation Tests. Procedia CIRP. 2013, 8,
[40] Mondal, D. N.; Sarangi, K.; Pettersson, F.; Sen, P. K.; Saxén, H.; 334–339. DOI: 10.1016/j.procir.2013.06.112.
Chakraborti, N. Cu―Zn Separation by Supported Liquid [54] Pasandideh, S. H. R.; Niaki, S. T. A. Multi-Response Simulation
Membrane Analyzed through Multi-Objective Genetic Optimization Using Genetic Algorithm within Desirability
Algorithms. Hydrometallurgy. 2011, 107(3–4), 112–123. DOI: Function Framework. Appl. Math. Comput. 2006, 175(1),
10.1016/j.hydromet.2011.02.008. 366–382. DOI: 10.1016/j.amc.2005.07.023.
[41] Deniz, E.; Aydogmus, O.; Aydogmus, Z. Implementation of [55] Datta, S.; Garai, C.; Das, C. Efficient Genetic Algorithm On Linear
ANN-Based Selective Harmonic Elimination PWM Using Hybrid Programming Problem For Fittest Chromosomes. J. Glob. Res.
Genetic Algorithm-Based Optimization. Measurement. 2016, 85, Comput. Sci. 2012, 3(6), 8.
32–42. DOI: 10.1016/j.measurement.2016.02.012. [56] Ali-Tavoli, M.; Nariman-Zadeh, N.; Khakhali, A.; Mehran, M.
[42] Oktem, H.; Erzurumlu, T.; Erzincanli, F. Prediction of Minimum Multi-Objective Optimization of Abrasive Flow Machining
Surface Roughness in End Milling Mold Parts Using Neural Processes Using Polynomial Neural Networks and Genetic
Network and Genetic Algorithm. Mater. Des. 2006, 27(9), Algorithms. Mach. Sci. Technol. 2006, 10(4), 491–510. DOI:
735–744. DOI: 10.1016/j.matdes.2005.01.010. 10.1080/10910340600996126.
[43] Singh, R. K.; Gangwar, S.; Singh, D. K. Experimental Investigation [57] Ghosh, A.; Tsutsui, S. Advances in Evolutionary Computing Theory
on Temperature-Affected Magnetic Abrasive Finishing of and Applications; Springer Berlin: Berlin, 2013.
Aluminum 6060. Mater. Manuf. Process. 2019, 34(11), [58] Deb, K.;. Multi-Objective Optimization Using Evolutionary
1274–1285. DOI: 10.1080/10426914.2019.1628263. Algorithms; John Wiley & Sons: New York, 2001.
[44] Mishra, V.; Goel, H.; Mulik, R. S.; Pandey, P. M. Determining [59] Puranto, P.; Hamdana, G.; Pohlenz, F.; Langfahl-Klabes, J.; Daul, L.;
Work-Brush Interface Temperature in Magnetic Abrasive Li, Z.; Wasisto, H. S.; Peiner, E.; Brand, U. Indentation Modulus and
Finishing Process. J. Manuf. Process. 2014, 16(2), 248–256. DOI: Hardness Investigation of Crystalline Silicon Surfaces Treated by
10.1016/j.jmapro.2013.10.004. Inductively Coupled Plasma Reactive Ion Etching. J. Phys. Conf.
[45] Bishop, C. M.;. Neural Networks for Pattern Recognition; Oxford Ser. 2019, 1319, 012008. DOI: 10.1088/1742-6596/1319/1/012008.
university press: New York, 1995. [60] Teng, X.; Zhang, G.; Zhao, Y.; Cui, Y.; Li, L.; Jiang, L. Study on
[46] Ibnu Choldun, R. M.; Santoso, J.; Surendro, K. Determining the Magnetic Abrasive Finishing of AlSi10Mg Alloy Prepared by
Number of Hidden Layers in Neural Network by Using Principal Selective Laser Melting. Int. J. Adv. Manuf. Technol. 2019, 105
Component Analysis. In Intelligent Systems and Applications; (5–6), 2513–2521. DOI: 10.1007/s00170-019-04485-5.