You are on page 1of 14

Materials and Manufacturing Processes

ISSN: 1042-6914 (Print) 1532-2475 (Online) Journal homepage: http://www.tandfonline.com/loi/lmmp20

Solution to industrial optimization problems


through differential evolution variants

Hira Zaheer & Millie Pant

To cite this article: Hira Zaheer & Millie Pant (2017): Solution to industrial optimization
problems through differential evolution variants, Materials and Manufacturing Processes, DOI:
10.1080/10426914.2017.1279300

To link to this article: http://dx.doi.org/10.1080/10426914.2017.1279300

Accepted author version posted online: 11


Jan 2017.
Published online: 11 Jan 2017.

Submit your article to this journal

Article views: 39

View related articles

View Crossmark data

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


http://www.tandfonline.com/action/journalInformation?journalCode=lmmp20

Download by: [Indian Institute of Technology Roorkee] Date: 18 March 2017, At: 05:16
MATERIALS AND MANUFACTURING PROCESSES
http://dx.doi.org/10.1080/10426914.2017.1279300

Solution to industrial optimization problems through differential evolution variants


Hira Zaheer and Millie Pant
Department of Applied Science and Engineering, Indian Institute of Technology, Roorkee, Uttarakhand, India

ABSTRACT ARTICLE HISTORY


Optimization forms the core of several industrial problems and demands suitable methods of solution for Received 18 August 2016
determining the best possible solution. However, it is often observed that the classical optimization Accepted 17 December 2016
algorithms available in literature may not be applicable in real-life scenarios where the nature of the KEYWORDS
problems is quite complex. Under such circumstances, nature-inspired algorithms (NIA) can be beneficial Equipments; finishing;
because of their generic nature. This is shown in the present study by obtaining the optimal solution of two optimization; roughing
fundamental industrial problems, viz. optimizing and controlling the level of noise in an industry having
multiple sources of noise; and optimizing the total production cost in a machining process through different
NIA including two newly proposed differential evolution (DE) variants. Comparison is also done through
classical methods. Both the problems are nonlinear in nature where the complexity increases by increasing
the number of variables. The first problem taken is unconstrained in nature, while the second problem is
constrained. A thorough comparison of all the methods is done through various performance measures and
it is observed that the proposed DE variants form an attractive alternative for dealing with such problems.

Introduction process parameters for 9Cr-1Mo steel is done using RSM and
GA. PSO has been used for programming of computer
Several instances are available in literature where solutions to
numerical control (CNC) milling machines in Simon et al.
industrial problems are given through optimization. However,
[8], for multi-objective optimization of grinding process para-
as the complexity of the problem increases, it becomes difficult
meters [9], for machining parameters optimization of titanium
to apply classical methods of optimization. Consequently, last
alloy [10]. A hybridized version of PSO and DE is used for
few decades have witnessed the emergence of several generic
surface-potential-based model parameter extraction for nanos-
type algorithms that are independent of the nature of the prob-
cale MOSFETs [11]. Application of GA to Liþ ions contained
lem and therefore have wider applicability. These algorithms
in carbon nanotube is another example where an investigation
are usually based on some natural analogy and are commonly
is carried out using DE and PSO along with molecular
known as nature-inspired algorithms (NIA). Some NIA that
dynamics [12]. In Rao et al. [13], for parameter optimization
become immensely popular include genetic algorithms (GA),
of ultrasonic machining process, different optimization tools
differential evolution (DE), particle swarm optimization
like harmony search algorithm, GA, PSO, and artificial
(PSO), etc. These algorithms have been successfully applied
bee colony algorithm are used. Some other examples that
to a wide range of problems. Considering the brevity of space,
can be quoted include a brief review for the fundamentals of
a few are mentioned here:
GA and use of GA into polymeric and nonpolymeric
In Chakraborti et al. [1], optimization of the fluid flow in
applications [14]. Mittal et al. [15] formed a novel hybrid
hydrocyclones is done with the help of multi-objective
optimization method for determining the total number of
GA. In Jourdan et al. [2], the multi-objective design of
turbines to be placed in wind farm and to find out their
conducting polymer composites is analyzed and the effect of
optimal locations. Mittal et al. [16] considered a problem of
multiple layers is studied. In Baraskar et al. [3], multi-objective
energy maximization and noise minimization, the model
optimization of electrical discharge machining process is
formed was multi-objective, they proposed a new hybrid
implemented using hybrid method, and in Giri et al. [4], an
approach which is a combination of multi-objective evolution-
application of genetic programming evolving through
ary algorithm (NSGA II) and a single objective gradient
bi-objective GA is proposed. Experimental investigation to
approach for solving series of integer and continuous
optimize tool life and surface roughness in Inconel 718
problems.
machining is done using non-dominated sorting in genetic
algorithms (NSGA) in Farshid et al. [5], another application
of GA is shown in Chockalingam et al. [6], where it is applied
Optimal noise control
for improving anisotropic strength of fused deposited ABS
parts. Another example of optimization in industries can be The present study is focused on determining the optimal
seen in Nagaraju et al. [7], where the optimization of welding solution for two common but challenging problems arising

CONTACT Hira Zaheer hirazaheeriitr@gmail.com Department of Applied Science and Engineering, Indian Institute of Technology, Indira Bhawan,
Roorkee 247667, India.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/lmmp.
© 2017 Taylor & Francis
2 H. ZAHEER AND M. PANT

in industries through modified variants of DE. The results are Cost optimization of turning process
also compared with GA and through fmincon, a classical tech-
Turning process is relevant to manufacturing industries. Basi-
nique based on Nelder–Mead approach.
cally, manufacturing is the method that involves physical,
In the coming sections, a brief description of the problems
chemical, and mechanical processes for converting the geo-
considered in this article is given which is followed by a
metrical properties of given raw material to finished ones.
description of DE and its modified variants. Mathematical
The whole process consists of different mechanisms such as
descriptions of the problems are given in Materials and
milling, turning, drilling, grilling, finishing, and roughing
Methods section.
required for product manufacturing. The process could be
The various processes in manufacturing industries involve a
completed in either one pass or through in multiple passes.
number of noise sources like internal combustion engines,
In general, single-pass operation is not cost-efficient and
impact processes, electrical machines, vibrating panels, turbu-
therefore from economic point of view multi-pass approach
lent fluid flow, gears, fans, rotors, and stators, etc. Some
is carried out with one finish cut and several rough cuts. For
equipment and particularly noisy operations include metal
solving this problem, researchers have exploited several opti-
and dividing cutting like grinders, milling machine, lathes,
mization techniques. A few references are given as follows:
shearing and pressing, punching; machine tools for forming,
Gopalakrishnan and Al-Khayyal [23] employed a geometric
boiler making, electric furnaces, sandblasting, plasma jets,
programming technique for parameter selection and presented
tumbling barrels; pneumatic equipment, lathes, drilling,
objective function for minimizing the time and production
shake-out punch presses, blasting, riveting, crushing; as well
cost for machining optimization. Here, machining conditions
as textile machines, plant conveying systems, self-propelled
for single- and multi-pass operations were calculated using
working machines, hand-guided machines, drive units, com-
dynamic programming and Nelder–Mead simplex method.
pressors and pumps, print machines; beverage-filling
Prasad et al. [24] also used geometric programming method
machines; and transport vehicles. Some industries that have
and linear programming for optimizing different values
a high noise quotient include basic metal industries, fabricated
(production cost and time) in multi-pass turning operation.
metal products, chemical products, chemicals, paper products,
Hui et al. [25] optimized the cost in turning operation and
mills manufacturing paper, wood mills, and sawmill. An
analyzed machining condition’s effect on quality, and solved
example showing the generation of noise from a gas jet is given
the constrained, nonlinear optimization model with the help
in Fig. 1.
of gradient finite difference and quasi-Newton method.
Considering the health issues due to noise pollution, opti-
Saravanan et al. [26] suggested different techniques for solving
mal control of noise has become an important issue for the
CNC turning problem and minimizing the total production
industries. Despite being an important problem, relevant refer-
time. They determined parameters like feed rate and cutting
ences related to it in literature are limited. Researchers have
speed with respect to some machining constraints such as
worked on noise control devices [17–19]. For controlling
power, cutting force, temperature, roughness, etc.
noise, it is necessary to recognize the exact location of noise,
Sharma et al. [27] applied neural network and constructed
the value of sound pressure level (SPL), and sound power level
model for measuring different machining variables, viz. sur-
(Lw) of noisy sources [20]. It is impossible to recognize the
face roughness and cutting force by considering parameters
actual location of noise in a multi-noise plant by human ear.
like speed, feed rate, approaching angle, and depth of cut.
Also, sound meters do not give actual knowledge about SPL
Chandrasekaran et al. [28] proposed a review on the appli-
of such noise plants. Solution to such problems through opti-
cation of various methods like GA, ant colony optimization,
mization algorithms can be seen in References [21] and [22].
PSO, fuzzy sets, and neural network to analyze the machining
process of milling, turning, grinding, and drilling.
Yildiz et al. [29, 30] applied hybridization to optimize pro-
duction cost in multi-pass turning process. They hybridized
DE with Taguchi’s method and formulated new approach
called hybrid robust differential evolution (HRDE) and com-
pared its performance with other evolutionary algorithms.
Lu Chao et al. [31] worked on multi-objective model for
multi-pass turning operations by considering energy con-
sumption and machining quality.

Differential evolution
DE, first proposed by Storn and Price [32, 33], is a robust,
stochastic technique used for solving global optimization
problems. Several variants of DE are available in literature
including self-adaptive approach [34] and memory-based
differential evolution [35]. DE having two difference vectors
in mutation and a guiding force constant [36, 37], a brief
survey on various aspects of DE like initialization vectors,
Figure 1. Generation of noise from gas jet. mutation crossover parameters, hybrid DE algorithm, DE for
MATERIALS AND MANUFACTURING PROCESSES 3

multi-objective algorithm, parallel DE, etc., was done by Das The method used for solving constraints is based on feasi-
and Suganthan [38]. bility of constraints briefly explained in Reference [39].

Proposed strategy 1 Materials and Methods


This variant of DE is called AIPDE as it is based on infor- The objective here is to minimize the noise impact in the plant
mation preserving adaptive strategy for selection (AIP). This by minimizing the sum of squared differences of sound power
variant starts with the initialization process which is same as level at trial point and the targeted actual noise level.
that of basic DE. Next, step mutation and crossover are also
performed accordingly. The difference occurs during the
selection process. During selection, a random number rs Case Study 1: Optimal noise control
(0,1) is generated in each iteration, if rs < Sr, where Sr is
Mathematically it can be written as
the selection probability fixed as 0.6 in the present study,
m
then we opt for selection strategy 1 otherwise selection strat- X �2
Min f ð X Þ ¼ Lpj½trial� Lpj½act� ð2Þ
egy 2 is opted. j¼1
Let NP be the population size, and Xi,G ¼ (x1,i,G, x2,i,G, … , where X ¼ (x1[trial], … , xn[trial]; y1[trial], … , yn[trial]; z1[trial], … ,
xD,i,G) be the target vector. Vi,Gþ1 ¼ (v1,i,G, v2,i,G, … , vD,i,G) be zn[trial]; LP1[trial], … , LPn[trial].
the mutant vector and Uj,i,G ¼ (u1,i,G, u2,i,G, … , uD,i,G) be the LPi[act] is the actual sound power level and Lpj[trial] is defined
trial vector. Let f(X) and f(U) be the objective function values below.
for all target vectors and trial vectors, respectively, and let With the help of m sound power level (Lw), m sets of sound
F be the union of f(X) and f(U), that is, F ¼ f(X)∪f(U). In pressure level (LP) can be determined. For locating the noise
selection strategy 1, best NP vectors out of F are taken for level, a trial point [(xi[trial], yi[trial], zi[trial]), LPi[trial]] is assumed
the next generation. While selection strategy 2 is same as that at ith noise source.
of the selection strategy of basic DE, given as The sound pressure level LP for a receiving point j for n
� �
Ui;Gþ1 if f Ui;Gþ1 � f Xi;G
� equipments is given as
Xi;Gþ1 ¼
Xi;G otherwise (
X n
)
LPji =10
Lpj ¼ 10 � log 10 ð3Þ
where Xi,G þ 1 is the target vector for the next generation. i¼1
� � �
Lpij ¼ Lpi wi Rij ; u; n 20 log Rij 11 ð4Þ
� 2�
Proposed strategy 2 Rij t
10 8

wi Rij ; u; n ¼ 7:4 ð5Þ
u
This strategy also starts like the usual DE algorithm by initia-
lizing the population, and sorting it as per the fitness value. where Lpij is the noise pressure level experienced at jth
Similarly, differential mutation occurs after initialization. Its receiving point evolved from ith noise source; Lpi is the sound
flow diagram will be similar to the above DE flow chart shown pressure level of ith noise source; wi is the sound absorbed by
in Fig. 1 with the modified crossover operation explained air at 20°C; ϕ is the humidity in the present air; ν is the fre-
below. We generate a new individual called Xav, by considering quency of sound; and Rij is the distance between the ith noise
all the individuals of the population. Then, in crossover step source and jth receiver, calculated by the distance formula
crossover probability (CR) is divided into two different cross- between two points.
over probabilities CR1 and CR2 such that CR2 ¼ 1 CR1, CR2 For identifying the noise level, three kinds of M point
will always be smaller than CR1, that is, the value of CR1 lies monitoring systems are installed separately by taking the value
between 0 to 0.5. The value of CR1 decides to which vector of M as 18, 36, and 66. The location of these points is shown in
maximum probability should be given among target, trial, Tables 1–3 and its pictorial representation is given in Figs. 2–4,
and average vectors. respectively, where MP refers to monitoring points.
Let m be the dimension of the problem and N be the The noise was identified by taking one, two, and four noise
population size then X ¼ {X1, X2, … , Xi, … , XN} be the total plants independently on a plant area of 20 m in length, that is,
population such that Xi ¼ (xi1, xi2, … , xii, … , xim), XN ¼ (xN1, x-axis, and 15 m in breadth, that is, y-axis.
xN2, … , xNi, … , xNm) then the average vector is denoted by
Xav and defined as Xav ¼ (xav1, xav2, … , xavi, … , xavm) where
xavi ¼ ðx1i þx2iNþ...þxNi Þ. Since the average vector will always be Nature of the problem
better than some of the individual, it contracts the search The problem is unconstrained nonlinear type in which com-
space which may result in fast convergence. Mathematically, plexity increases on increasing the number of noise equip-
crossover step is explained as: ments and monitoring points. Increase in the number of
8 noise sources in the system increases the number of variables
< vj;i;Gþ1 if rand � Cr1 8j ¼ jrand
uj;i;Gþ1 ¼ Xav Cr1 � if rand � Cr2 ð1Þ making it difficult to identify the exact location of noise equip-
:x otherwise ment. By increasing the monitoring points, objective function
j;i;G
becomes complex as the number of terms increases in the
4 H. ZAHEER AND M. PANT

Table 1. Location of monitoring points for 18 and 36 points monitoring system. Table 2. Location of points for 66 points monitoring system.
MP x y z MP x y z MP X y z MP x y z
R1 0 0 1 R1 0 0 1 R19 12 0 1 R1 0 0 1
R2 0 7.5 1 R2 0 3 1 R20 12 3 1 R2 0 3 1
R3 0 15 1 R3 0 6 1 R21 12 6 1 R3 0 6 1
R4 4 0 1 R4 0 9 1 R22 12 9 1 R4 0 9 1
R5 4 7.5 1 R5 0 12 1 R23 12 12 1 R5 0 12 1
R6 4 15 1 R6 0 15 1 R24 12 15 1 R6 0 15 1
R7 8 0 1 R7 4 0 1 R25 16 0 1 R7 2 0 1
R8 8 7.5 1 R8 4 3 1 R26 16 3 1 R8 2 3 1
R9 8 15 1 R9 4 6 1 R27 16 6 1 R9 2 6 1
R10 12 0 1 R10 4 9 1 R28 16 9 1 R10 2 9 1
R11 12 7.5 1 R11 4 12 1 R29 16 12 1 R11 2 12 1
R12 12 15 1 R12 4 15 1 R30 16 15 1 R12 2 15 1
R13 16 0 1 R13 8 0 1 R31 20 0 1 R13 4 0 1
R14 16 7.5 1 R14 8 3 1 R32 20 3 1 R14 4 3 1
R15 16 15 1 R15 8 6 1 R33 20 6 1 R15 4 6 1
R16 20 0 1 R16 8 9 1 R34 20 9 1 R16 4 9 1
R17 20 7.5 1 R17 8 12 1 R35 20 12 1 R17 4 12 1
R18 20 15 1 R18 8 15 1 R36 20 15 1 R18 4 15 1
R19 6 0 1
R20 6 3 1
R21 6 6 1
function. The problem is solved for 1, 2, and 4 equipments R22 6 9 1
consisting of 4, 8, and 16 variables, respectively. R23 6 12 1
R24 6 15 1
R25 8 0 1
R26 8 3 1
Case Study 2: Cost optimization of turning process R27 8 6 1
R28 8 9 1
Cutting of metal is an industrial process where metals are R29 8 12 1
shaped by removing extra unwanted parts. In turning, the R30 8 15 1
metal part is rotated and a tool is moved parallel to the axis R31 10 0 1
R32 10 3 1
of rotation. By turning metals can be shaped in different ways. R33 10 6 1
that is, straight, curve, conical, cylindrical, etc. Different R34 10 9 1
machining parameters such as depth of cut, number of passes, R35 10 12 1
R36 10 15 1
speed of cutting metals affect the production costs, machining R37 12 0 1
quality, and cutting forces. Some turning process parameters R38 12 3 1
are shown in Fig. 5. R39 12 6 1
R40 12 9 1
For minimizing the production cost, proper selection of R41 12 12 1
these parameters is needed. R42 12 15 1
The model of problem is taken from Shin and Joo [41] in R43 14 0 1
R44 14 3 1
which the motive is to optimize the total production cost R45 14 6 1
and to calculate the optimal depth of cut by optimizing R46 14 9 1
number of passes. Here, all machining parameters are used R47 14 12 1
R48 14 15 1
simultaneously. R49 16 0 1
The total production cost in machining process is the sum R50 16 3 1
of finishing cost and roughing cost, mathematically given as R51 16 6 1
R52 16 9 1
n
X R53 16 12 1
Min U ¼ UCs þ UCri þ A5 ð6Þ R54 16 15 1
i¼1 R55 18 0 1
R56 18 3 1
where UCs is the production cost for finishing, UCri is the R57 18 6 1
R58 18 9 1
production cost for roughing for ith pass, A5 is constant for R59 18 12 1
machining time, and U is the total cost. R60 18 15 1
R61 20 0 1
UCs is defined as UCs ¼ A2 þ A1 fsðq=p 1Þ r=p
ds ð7Þ R62 20 3 1
R63 20 6 1
where R64 20 9 1
R65 20 12 1
R66 20 15 1
pDLk0 Ts 1=p
� �
A1 ¼ ðTs þ nðte þ kt =k0 ÞÞ ð8Þ
1000 Tp C0
A 2 ¼ k 0 ð h1 L þ h2 Þ ð9Þ
Tp
and Ts is given by Ts ¼ and constraints for finishing
n
feed rate:
objective function are defined as:cutting speed:
Vs L � Vs � VsU ð10Þ fsL � fs � fsU ð11Þ
MATERIALS AND MANUFACTURING PROCESSES 5

Table 3. Value and bounds of various parameters for all three cases.
Cases Case I Case II Case III
Parameters One equipment plant Two equipment plants Four equipment plants
Dimension 4 8 16
Equipment point(s) (5, 7.5, 2) (5, 7.5, 2)(18, 7.5, 2) (5, 7.5, 2)(15, 7.5, 3)(11, 4.5, 2)(9, 10.5, 1.5)
Sound power levels Lw1(db) 105 105, 102 105, 90, 100, 85
Axes bounds x-axis [0, 15] [0, 15] [0, 15]
y-axis [0, 20] [0, 20] [0, 20]
z-axis [0, 3] [0, 3] [0, 38]
Range of Lw Lw [80, 120] [80, 120] [80, 120]

depth of cut: fs ¼ ð8RRmax Þ0:5 ð16Þ


dsL � ds � dsU ð12Þ It may be combined with feed rate bound (Eq. 11) and be
The above three constraints represent the variable written as
restrictions.
In finishing process, the machining conditions very much fsL � fs � minffsU ; ð8RRmax Þ0:5 g ð17Þ
affect the life of machining tool. Tool life constraint is defined as: Further there is cutting force constraint which limits the
C0 deflection of cutting tool, expressed as
Tmin � Ts � Tmax ; where Ts ¼ p q r ð13Þ
Vs fs ds
Fs ¼ k1 ðfs Þl ðds Þv � Fmax ð18Þ
Surface finishing also plays important role in finishing
operation because it affects the machining quality. This may be And finally there is power constraint, which ensures that
affected by feed rate and noise radius, mathematically defined as the cutting power should not exceed the total available power.

Figure 4. Location of 66 points in 15 � 20 plant area.


Figure 2. Location of 18 points in 15 � 20 plant area.

Figure 3. Location of 36 points in 15 � 20 plant area. Figure 5. Basic parameters in turning process [40].
6 H. ZAHEER AND M. PANT

This is defined as Ant colony optimization (ACO) [46]


1=p
Gupta, Batra, and Lal model [47]
6120Tp Pmax g Shin and Joo model [41]
fsðl q=pÞ dsv r=p � 1=p
ð19Þ
K1 C0 The rationale of choosing the selective variants of DE for
comparison is that these have been already applied successfully
The roughing operation objective function is also defined in a for dealing with these problems. GA is chosen because it is still
similar manner in which s is replaced by ri, that is, all identities one of the most popular nontraditional optimization algo-
are same. Also, the constraints are same as for finishing oper- rithms. Besides using the algorithms mentioned above, both
ation. In addition to these, there is a depth of cut constraint, the problems are also solved with fmincon (MATLAB
defined as the total depth of cut for n passes, which should be Toolbox, having a collection of optimization methods for con-
equal to the depth of cut during finishing and in n rough pass. strained and unconstrained problems). Based on the nature of
n
X the problems, Nelder–Mead approach and interior point
dt ¼ dri þ ds ð20Þ method are used for solving problems 1 (unconstrained) and
i¼1 2 (constrained), respectively.
*For both the problems, GA is applied through “GA
Nature of the problem toolbox” of MATLAB, using the command “optimtool” with
default experimental setting provided by “GA toolbox.” How-
It is a constrained, nonlinear optimization problem. Other
ever, population size is taken as NP ¼ 100 and number of
than variable bounds, there are four inequality constraints in
function evaluations are taken as 1000 � D and NFE ¼ 600
finishing operation and four in roughing operation, also we
for problems 1 and 2, respectively.
have an equality constraint which is common with respect
to both operations. Complexity of problem increases on
increasing the numbers of multi-pass. There are three finish- Result and Discussion
ing variables and three roughing variables for one pass. So,
Case Study 1: Optimal noise control
in the problem for n ¼ 1 there are total six variables and nine
constraints. Whereas for two passes, there are three finishing Results are computed on the basis of various performance
and six roughing variables, that is, for n ¼ 2 there are 9 measures like:
variables and 13 constraints. In third case for n ¼ 3, number a) Fitness value for measuring the results in terms of best
of variables are 12 and number of constraints are 17. value, worst value, average value, and standard deviation.
Table 4 is showing the nomenclature. b) Accuracy of monitoring points in terms of average
location and sound power level.
c) Feasibility for obtaining the optimal solution within the
Algorithms Used for Comparison
feasible domain.
For problem 1: d) Nonparametric Friedman test for measuring the perfor-
Genetic algorithm (GA)* mance of algorithms in terms of rank. Parameter setting
Differential evolution (DE) [32] is shown in Table 5. Best fitness values are highlighted
Trigonometric mutation differential evolution (TDE) [42] in boldface in Tables 6, 8, 10, 11, 12 and most accurate
Differential evolution algorithm with random localization values are highlighted boldface in Tables 7, 9 and 13.
(DERL) [43]
Modified random localization differential evolution
Case I: For N ¼ 1
(MRLDE) [44]
Self-adaptive mutation strategy for DE (SaMDE) [44] Fitness Value
For problem 2: For one equipment plant, there are three monitoring systems
Totally disturbed particle swarm optimization (TDPSO) [28] with M ¼ 18, 36, and 66. For M ¼ 18, it can be seen from
Genetic algorithm (GA)* Table 6 that AIPDE is performing best in terms of best value
Simulated annealing (SA) [45] and average value of fitness function. In terms of worst value

Table 4. Nomenclature.
UCs,UCri Total production cost for finishing and roughing, respectively ($)
Vs, Vr Cutting speeds in finishing and roughing operations, respectively (m/min)
fs, fr Feed rates in finishing and roughing operations, respectively (mm/rev)
ds, dr Depth of cut in rough and finish machining, respectively (mm)
VsL, VsU, VrL, VrU, Lower and upper bound of cutting speed, depth of cut, and feed rate in finish and rough machining, respectively (m/min)
dsL, dsU, drL, drU,
fsL, fsU, frL, frU
dt Total depth of material to remove (mm)
D, L Length and diameter of work piece (mm)
k0 Direct labor cost, including overheads ($/piece)
kt Cutting edge cost ($/piece)
tmr, tms, tm Roughing, finishing, and actual machining time, respectively (min)
tc, tv, ti Preparation time (loading and unloading, etc.), idle tool motion time (tool travel and tool approach/depart time), and total machine
The numerical values of the above parameters are taken from Reference [40].
MATERIALS AND MANUFACTURING PROCESSES 7

Table 5. Parameter setting for both the problems.


Population size 100 Crossover probability, Cr 0.9 Scaling factor 0.5. Number of function evaluation (NFE) Dimension � 1000
Software used: Matlab, 7.0.1 version, and DEV Cþþ 4.9.9.0. Results are computed on the laptop with configuration: Intel i3 processor, 4 GB of RAM, Windows 10.
Average result of 50 runs is taken for each algorithm.

Table 6. Results for N ¼ 1 for fitness value and average location (x, y, z, Lw) (maximum NFE is taken as 4000).
Terms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
M ¼ 18 Best 1.43E – 01 3.76E-03 5.52E – 04 1.47E – 04 5.42E – 05 1.22E – 07 4.56E – 07 1.608E – 09
Worst 3.6095 4.12E-02 6.45E – 02 4.33E – 04 9.70E – 05 5.54E – 07 4.56E – 06 6.024E – 06
Average 1.4984 2.54E-02 7.82E – 03 2.23E – 04 4.87E – 05 3.22E – 07 7.58E – 07 1.680E – 07
SD 1.4490 1.08E-02 2.46E – 03 1.41E – 04 5.31E – 05 2.61E – 07 4.89E – 07 8.490E – 07
Location (x,y,z) 4.462, 7.512, 5.000, 7.498, 4.997, 7.496, 5.001, 7.501, 4.999, 7.499, 4.999, 7.499, 5.000, 7.500 4.999, 7.500,
2.294 1.601 1.983 1.998 2.000 2.000 0.000 1.920
Lw(db) 105.05 104.975 104.997 104.999 105.00 105.00 105.00 105.00
M ¼ 36 Best 0.4891 1.11E-02 4.23E – 03 5.20E – 05 3.02E – 06 1.59E – 07 2.79E – 07 1.548E – 09
Worst 5.3975 4.75E-02 6.91E – 03 7.51E – 04 4.61E – 05 9.65E – 06 1.65E – 05 7.159E – 06
Average 2.3113 1.74E-02 5.29E – 03 8.71E – 05 2.29E – 05 5.24E – 07 2.02E – 06 2.854E – 07
SD 1.9801 1.13E-02 2.94E – 03 8.31E – 04 1.06E – 05 9.32E – 07 3.58E – 07 1.240E – 07
Location (x,y,z) 5.076, 7.539, 4.998, 7.500, 4.995, 7.500, 5.000, 7.499, 5.000, 7.499, 5.000, 7.499, 5.000, 7.500 5.000, 7.500,
2.227 1.797 2.000 1.999 2.000 2.000 0.000 1.960
Lw(db) 105.219 104.996 104.998 105.00 105.00 105.00 105.00 105.00
M ¼ 66 Best 0.2895 4.01E-03 3.07E – 04 9.04E – 06 1.47E – 06 3.33E – 08 4.37E þ 02 2.068E – 09
Worst 2.8711 5.45E-02 1.39E – 02 8.53E – 05 4.35E – 05 3.15E – 06 6.86E þ 02 6.2396E – 06
Average 1.4776 1.54E-02 2.94E – 03 6.07E – 05 1.01E – 05 2.11E – 07 5.33E þ 02 1.480E – 07
SD 1.0638 1.76E-02 4.67E – 03 2.37E – 05 4.13E – 06 1.08E – 07 4.58E þ 02 8.80E – 07
Location (x,y,z) 4.868, 7.502, 4.998, 7.484, 5.003, 7.496, 5.000, 7.499, 5.000, 7.499, 5.000, 7.500, 1.853, 5.793, 5.000, 7.499,
2.053 1.891 1.993 1.998 2.000 2.000 −5.842 2.000
Lw(db) 105.073 104.991 104.996 104.998 105.000 105.000 105.00 105.000

Table 7. Results in terms of accuracy with respect to various monitoring system for N ¼ 1.
Monitoring point Accuracy GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
M ¼ 18 Δr 0.6132 0.339 0.017 0.0024 0.0014 0.0014 2.00 0.08
ΔSWL 0.050 0.025 0.005 0.0086 0.00 0.00 0.00 0.00
M ¼ 36 Δr 0.2425 0.2030 0.005 0.0014 0.0009 0.0009 2.00 0.04
ΔSWL 0.219 0.004 0.002 0.00 0.00 0.00 0.00 0.00
M ¼ 66 Δr 0.1422 0.116 0.0086 0.0022 0.0009 0.00 9.79 0.00
ΔSWL 0.073 0.009 0.004 0.002 0.00 0.00 0.00 0.00

Table 8. Results for N ¼ 2 for fitness value and average location (x, y, z, Lw) (maximum NFE is taken as 8000).
Terms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
M ¼ 18 Best 9.52E – 01 3.49E – 01 8.26E – 03 3.59E – 03 2.33E – 05 2.41E– 07 1.95E þ 02 1.193E – 09
Worst 4.1808 9.82E – 01 1.70E – 01 1.09E – 01 2.26E – 03 4.50E – 07 1.82E þ 03 4.7503E– 07
Average 3.029 7.45E – 01 4.59E – 02 2.38E – 02 8.25E – 04 3.46E– 07 2.69E þ 02 5.530E – 08
SD 1.2687 4.22E – 01 7.09E – 02 4.81E – 02 1.24E – 03 1.47E – 07 2.23E þ 02 8.790E – 08
Location 1 5.062, 7.314, 5.009, 7.582, 5.019, 7.535, 5.005, 7.497, 5.00, 7.500, 5.00, 7.500, 8.048, 6.163, 4.999, 7.500,
1.993 1.388 1.974 1.992 1.999 1.999 4.552 2.000
Location 2 18.097, 8.13, 18.000, 7.585, 17.990, 7.510, 18.005, 7.504, 18.00, 7.501, 17.999, 7.500, 9.052, 7.249, 18.000, 7.500,
2.139 2.913 2.920 2.992 2.995 2.998 9.887 2.999
SWL 1 (db) 105.463 104.9076 105.0045 104.994 105.004 105.00 109.494 105.000
SWL 2 (db) 100.773 102.032 101.9307 101.990 101.991 101.999 69.035 102.000
M ¼ 36 Best 5.16E – 01 3.19E – 01 5.32E – 03 2.73E – 03 5.94E – 05 3.67E – 07 2.87E þ 02 2.473E 09
Worst 3.2979 6.01E – 01 3.49E – 02 4.04E – 02 8.79E – 04 5.38E – 07 1.07E þ 03 8.459E 08
Average 1.7994 4.52E – 01 3.19E – 02 1.09E – 02 5.08E – 04 4.77E – 07 6.45E þ 02 3.615E 08
SD 1.0106 3.50E – 01 1.31E – 02 1.66E – 02 4.15E – 04 1.55E – 07 2.59E þ 02 2.700E 08
Location 1 5.050, 7.467, 5.027, 7.497, 5.030, 7.505, 4.995, 7.504, 5.000, 7.499, 5.000, 7.500, 4.079, 7.520, 4.999, 7.500,
1.522 1.962 1.988 2.001 1.998 1.999 11.281 2.000
Location 2 17.914, 7.470, 18.000, 7.552, 17.993, 7.493, 17.996, 7.500, 17.999, 7.500, 17.999, 7.500, 2.535, 15.273, 18.000, 7.500,
2.626 2.975 2.932 2.993 2.998 3.000 7.887 2.999
SWL 1 (db) 104.894 105.023 104.963 104.996 105.000 105.000 113.632 105.000
SWL 2 (db) 101.514 101.998 101.9229 101.997 102.000 102.000 63.905 102.000
M ¼ 66 Best 2.47E – 01 1.92E – 01 4.21E – 03 1.97E – 03 5.31E – 05 1.14E – 06 4.35E þ 02 7.093E 09
Worst 5.0021 5.81E – 01 3.24E – 02 1.96E – 02 2.89E – 04 5.36E – 06 7.68E þ 02 2.083E 06
Average 3.3621 2.67E – 01 1.62E – 02 5.12E – 03 5.48E – 04 3.22E – 06 6.05E þ 02 2.633E 07
SD 1.0023 2.50E – 01 1.48E – 02 1.17E – 02 1.24E – 04 2.93E – 06 3.36E þ 02 6.430E 07
Location 1 4.859, 7.492, 5.022, 7.501, 5.007, 7.509, 4.995, 7.497, 4.999, 7.500, 5.000, 7.500, 1.377, 0.866, 4.999, 7.500,
2.123 1.944 1.995 2.000 2.002 2.000 31.266 2.000
Location 2 17.818, 7.414, 18.019, 7.472, 17.962, 7.493, 17.992, 7.499, 17.999, 7.500, 17.999, 7.499, 6.914, 7.500, 18.000, 7.5000,
2.969 2.913 2.994 2.981 2.997 2.999 4.083 2.999
SWL 1 (db) 105.289 104.961 105.004 105.005 105.001 105.000 65224.411 105.000
SWL 2 (db) 101.812 101.961 101.971 101.979 101.998 101.999 109.337 102.000
8 H. ZAHEER AND M. PANT

Table 9. Results in terms of accuracy with respect to various monitoring system for N ¼ 2.
Monitoring point Accuracy GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
M ¼ 18 Δr1 0.1961 0.6175 0.0475 0.0098 0.0013 0.0013 07.348 0.0010
Δr2 1.0712 0.1216 0.0812 0.0102 0.0051 0.0022 31.571 0.0010
ΔSWL1 0.4630 0.0240 0.0045 0.0060 0.0040 0.0000 04.400 0.0000
ΔSWL2 1.2270 0.0320 0.0693 0.0100 0.0090 0.0010 32.965 0.0000
M ¼ 36 Δr1 0.4817 0.0467 0.0326 0.0064 0.0022 0.0013 13.313 0.0010
Δr2 0.3849 0.0576 0.0687 0.0081 0.0023 0.0000 24.507 0.0010
ΔSWL1 0.1060 0.0230 0.0370 0.0040 0.0000 0.0000 08.632 0.0000
ΔSWL2 0.4860 0.0020 0.0771 0.0030 0.0000 0.0000 38.095 0.0000
M ¼ 66 Δr1 0.1872 0.0326 0.0124 0.0058 0.0022 0.0000 34.113 0.0010
Δr2 0.2036 0.0687 0.0391 0.0206 0.0031 0.0017 13.155 0.0010
ΔSWL1 0.2890 0.0370 0.0040 0.0050 0.0010 0.0000 65120 0.0000
ΔSWL2 0.1880 0.0771 0.0290 0.0210 0.0020 0.0010 07.337 0.0000

Table 10. Results for N ¼ 4, M ¼ 18 for fitness value and average location (x, y, z, Lw) (maximum NFE is taken as 16,000).
Terms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
Best 3.6732 1.2384 6.30E – 03 1.63E – 02 4.98E – 03 2.59E – 03 1.49E þ 02 2.368E 04
Worst 9.6205 6.2758 1.23E – 01 6.86E – 02 4.42E – 02 4.40E – 03 1.65E þ 03 4.199E 03
Average 6.5412 3.7468 8.44E – 02 4.36E – 02 2.69E – 02 3.28E – 03 5.26E þ 02 1.844E – 03
SD 2.3371 1.8103 5.66E – 02 2.41E – 02 1.62E – 02 9.14E – 04 2.61E þ 02 1.690E – 03
Location 1 4.3981, 8.5726, 4.9574, 7.4564, 5.0037, 7.4108, 5.0108, 7.4903, 5.0127, 7.4935, 4.9915, 7.5022, 8.8390, 19.246, 5.0036, 7.5190,
2.8073 5.8901 2.0044 2.0256 2.0185 2.0057 33.565 2.0023
Location 2 13.760, 5.8091, 12.083, 6.2338, 13.2184, 6.9960, 15.9462, 9.0539, 14.6920, 8.7781, 15.3171, 6.8082, 5.971, 24.377, 14.999, 7.4999,
3.9663 2.2353 1.2755 2.0465 2.6707 2.5437 11.764 2.7854
Location 3 12.6596, 2.9983, 10.9777, 4.1540, 10.7802, 4.2532, 11.6018, 5.0947, 11.1942, 4.4855, 10.8099, 4.4708, 12.1400, 6.6190, 10.9980, 4.4999,
3.0431 2.1257 1.2098 1.7007 1.8463 2.3077 2.607 1.8996
Location 4 6.1252, 7.1045, 10.3942, 7.9252, 9.3822, 10.9584, 9.9377, 9.5283, 9.1411, 10.8591, 9.0021, 10.4919, 3.862, 4.968, 9.0000, 10.5012,
0.9510 2.6631 1.81411 1.7662 1.4937 1.4998 6.376 1.4999
Lw1(db) 106.171 104.6710 105.0358 105.0975 105.1032 104.8967 62211.5 104.907
Lw2(db) 92.538 87.9778 89.2685 89.4255 89.5164 89.8147 74.094 89.9900
Lw3(db) 97.022 99.7994 98.4747 99.7440 99.4269 100.6127 241349.6 100.053
Lw4(db) 88.9836 83.4439 84.9171 86.7369 85.7094 85.0798 110.180 84.9962

and standard deviation of fitness function, SaMDE gives the best is best among all the algorithms for all three monitoring points.
performance. Fmincon is also performing well by giving the The calculated SWL is 105 db that is equal to the actual SWL.
second best results, AIPDE is at third best position for one
equipment plant and 18 points monitoring system. For 36 points Feasibility
monitoring system, AIPDE gives best results in terms of all four From Table 14, it is observed that for N ¼ 1, all algorithms are
fitness values. For 66 points monitoring system, AIPDE is better giving feasible solution for M ¼ 18, 36, and 66 monitoring
among all in terms of best and average value, and second best points, except fmincon which gives feasible solution only for
for worst value and standard deviation. In terms of standard M ¼ 18 and M ¼ 36 monitoring point system. For M ¼ 66,
deviation and worst value, SaMDE gives best results. fmincon ceases to be feasible.

Accuracy of Average Location and Sound Power Level


Case II: For N ¼ 2
From Table 7, it can be seen that the average location obtained by
AIPDE for all three monitoring points are quite close to the actual From Table 8, it can be seen that for M ¼ 18, AIPDE gives
location (i.e., 5, 7.5, 2). Similarly, for sound power level, AIPDE best fitness in terms of best average and standard deviation.

Table 11. Results for N ¼ 4, M ¼ 36 for fitness value and average location (x, y, z, Lw) (maximum NFE is taken as 16,000).
Terms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
Best 2.9341 6.81E – 01 2.99E – 02 4.42E – 02 7.14E – 03 2.43E – 03 4.43E þ 02 1.13E 06
Worst 10.1561 9.6612 1.37E – 01 3.06E – 01 6.92E – 02 5.99E – 02 1.02E þ 03 4.21E 02
Average 5.2216 2.7831 7.25E – 02 1.75E – 01 2.01E – 02 3.54E – 03 7.40E þ 02 8.44E 03
SD 3.7193 3.8692 4.44E – 02 1.18E – 01 2.50E – 02 1.27E – 03 2.68E þ 02 1.88E 02
Location 1 4.8801, 7.6905, 4.98984, 7.4802, 5.0593, 7.5063, 5.0294, 7.5105, 5.021, 7.5032, 5.0119, 7.5009, 4.380, 15.637, 4.9222, 7.4950,
2.1642 2.1559 1.9898 1.9822 1.9926 1.9891 12.560 1.7818
Location 2 12.3068, 7.0569, 15.4057, 9.6496, 14.1917, 7.4964, 15.6924, 7.5314, 14.5671, 7.4385, 14.9036, 7.5234, 0.278, 0.289, 14.9660, 7.4974,
2.3102 2.4804 2.8253 2.6203 2.9170 2.9636 4.503 2.9610
Location 3 10.8041, 4.2885, 11.3516, 4.5426, 11.7340, 4.6199, 11.0816, 4.5059, 11.0076, 4.4826, 11.0203, 4.4979, 17.399, 0.0270, 11.0041, 4.4994,
2.10923 2.4242 2.2389 2.1711 1.9672 1.9986 3.022 1.9965
Location 4 7.3583, 8.7702, 8.1603, 9.9402, 9.8690, 9.3305, 9.5698, 10.0832, 9.0399, 10.8741, 9.2800, 10.2039, 12.283, 9.2760, 8.9836, 10.4996,
1.6398 2.8595 2.2471 2.0971 1.3827 1.5611 -5.321 1.5587
Lw1(db) 104.9545 105.0041 105.1157 105.0130 105.0273 105.0144 68201.73 104.9997
Lw2(db) 94.0873 83.6147 91.1662 87.6605 90.6714 90.0703 15232.59 89.9406
Lw3(db) 99.3495 100.2675 99.7938 100.166 99.8941 99.9521 118.200 99.9766
Lw4(db) 83.9269 85.0802 86.7600 84.2812 85.5619 84.7773 99405.51 85.9820
MATERIALS AND MANUFACTURING PROCESSES 9

Table 12. Results for N ¼ 4, M ¼ 66 for fitness value, and average location (x, y, z, Lw) (maximum NFE is taken as 16,000).
Terms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
Best 2.0926 1.4748 2.51E – 01 7.55E – 02 3.90E – 03 2.89E – 05 4.21E þ 02 1.23E – 06
Worst 8.9675 5.3499 1.770 4.26E – 01 1.86E – 01 7.94E – 05 8.01E þ 02 1.81E – 06
Average 4.8554 2.8142 7.08E – 01 2.46E – 01 1.59E – 02 5.00E – 05 6.22E þ 02 1.52E – 06
SD 2.6730 1.2121 6.14E – 01 1.61E – 01 1.02E – 02 2.96E – 05 3.67E þ 02 4.07E – 07
Location 1 5.539, 7.141, 5.5151, 7.1524, 5.0871, 7.5328, 5.0185, 7.5005, 4.9993, 7.5074, 5.000, 7.500, 5.662, 3.569, 4.9999, 7.5000,
2.848 1.6218 2.0558 2.0659 1.9622 1.999 8.410 1.9999
Location 2 12.803, 7.980, 13.2420, 7.1256, 15.1839, 7.6916, 15.1662, 7.8496, 14.9863, 7.4835, 15.0065, 7.5008, 8.673, 3.652, 15.0006, 7.5003,
1.693 2.6981 2.5485 2.9247 2.9094 3.0000 1.881 2.9994
Location 3 12.765, 5.2070, 11.3155, 3.8714, 11.0215, 4.5196, 11.0258, 4.5203, 10.9969, 4.5116, 10.9997, 4.5006, 4.393, 3.485, 10.999, 4.500,
0.3630 2.1151 2.1020 2.0422 2.0394 2.0004 1.397 2.000
Location 4 8.6122, 9.0140, 9.2613, 9.9447, 9.4465, 9.9749, 8.8579, 10.8743, 8.9709, 10.6891, 8.9958, 10.5027, 4.036, 10.715, 9.000, 10.500,
1.0051 2.6296 1.0336 1.4839 1.5195 1.5015 3.733 1.495
Lw1(db) 106.171 105.0547 104.8774 105.0694 104.9578 104.9992 109.307 105.000
Lw2(db) 92.538 90.6352 88.2168 89.0794 89.9017 89.9854 102.119 89.997
Lw3(db) 95.022 98.6262 100.1803 100.0588 99.9949 100.0020 104080 100.000
Lw4(db) 82.9836 83.0185 86.7906 83.6646 85.4482 85.0449 1844.26 84.993

Table 13. Results in terms of accuracy with respect to various monitoring system for N ¼ 4.
Monitorpoint Accuracyterms GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE
M ¼ 18 Δr1 1.4712 3.8905 0.01220 0.0294 0.0233 0.01046 44.6699 0.01947
Δr2 2.3087 3.2706 2.5302 2.0541 1.3553 0.8873 21.0514 0.0460
Δr3 2.4693 0.3688 0.8565 0.8974 0.2480 0.3628 11.1937 0.0100
Δr4 4.4827 3.1505 0.6744 1.3763 0.3858 0.0083 10.9102 0.0012
ΔLw1 1.1710 0.0329 0.0358 0.0975 0.1032 0.1033 62316.5 0.0930
ΔLw2 2.5380 2.0223 0.7315 0.5745 0.4836 0.1853 15.9060 0.0100
ΔLw3 2.978 0.2006 1.5253 0.2560 0.5731 0.6127 241450 0.0530
ΔLw4 3.9836 1.5561 0.0829 1.7396 0.7094 0.0798 25.1800 0.0038
M ¼ 36 Δr1 0.2786 0.1574 0.0604 0.0359 0.0224 0.0161 16.6909 0.2317
Δr2 2.8152 2.2484 0.8269 0.7903 0.4451 0.1056 16.4632 0.0518
Δr3 0.3082 0.5526 0.7811 0.1896 0.0378 0.0204 29.1927 0.0080
Δr4 2.3889 1.6931 1.6373 0.9246 0.3940 0.4120 07.6682 0.0609
ΔLw1 0.0455 0.0041 0.1157 0.0130 0.0273 0.0144 68306.7 0.0003
ΔLw2 4.0873 6.3853 1.1662 2.3395 0.6714 0.0703 15322.6 0.0594
ΔLw3 0.6505 0.2675 0.2062 0.1660 0.1059 0.0479 18.2000 0.0234
ΔLw4 1.0731 0.0802 1.7600 0.7188 0.5619 0.2227 99490.5 0.9820
M ¼ 66 Δr1 1.0671 0.7274 0.1085 0.0684 0.0385 0.0009 7.5484 0.0001
Δr2 2.6010 1.8226 0.5238 0.3943 0.0931 0.0065 8.8691 0.0009
Δr3 2.5089 0.7126 0.1060 0.0534 0.0411 0.0007 17.670 0.0010
Δr4 1.6135 1.2855 0.8323 0.4006 0.1923 0.0052 25.444 0.0050
ΔLw1 1.1710 0.0547 0.1226 0.0694 0.0422 0.0008 4.3070 0.0000
ΔLw2 2.5380 0.6352 1.7832 0.9206 0.0983 0.0146 12.119 0.0030
ΔLw3 4.9780 1.3738 0.1803 0.0588 0.0051 0.0020 10418 0.0000
ΔLw4 2.0164 1.9815 1.7906 1.5355 0.4482 0.0449 1929.2 0.0070

However, in terms of worst fitness value, AIPDE gives the monitoring point systems through AIPDE. SaMDE, MRLDE,
second best results. For 36 and 66 monitoring points, AIPDE and DERL also give appropriate results. However, the results
performs well for all the fitness values and for standard devi- obtained by fmincon are not at par with the other algorithms.
ation. fmincon is performing worst for all monitoring points Similarly, AIPDE gives exact value of sound power level (105
in case of two equipment plants. and 102 db) for all 18, 36, and 66 monitoring systems.

Accuracy of Average Location and Sound Power Level Feasibility


For N ¼ 2 (Table 9), the average calculated location is closest to For N ¼ 2 (Table 14), all algorithms are giving feasible
the exact location, that is, (5, 7.5, 2) and (18, 7.5, 3) for all the solutions for M ¼ 18, 36, and 66 monitoring points except
for fmincon.
Table 14. Feasibility of different algorithms for various noise source and moni-
toring points. Table 15. Ranks of all algorithms using Friedman test.
Noise source MP GA DE TDE DERL MRLDE SaMDE Fmincon AIPDE N¼1 N¼2 N¼4
Algorithms 18 36 66 18 36 66 18 36 66 Average ranks
N¼1 18 Yes Yes Yes Yes Yes Yes Yes Yes
36 Yes Yes Yes Yes Yes Yes Yes Yes GA 7 7 7 7 7 7 7 7 7 7
66 Yes Yes Yes Yes Yes Yes No Yes DE 6 6 6 6 5 6 6 6 6 5.88
N¼2 18 Yes Yes Yes Yes Yes Yes No Yes TDE 5 4 5 5 6 5 4 5 5 4.88
36 Yes Yes Yes Yes Yes Yes No Yes DERL 4 5 4 4 4 4 5 4 4 4.22
66 Yes Yes Yes Yes Yes Yes No Yes MRLDE 1 1 3 2 3 3 3 3 3 2.44
N¼4 18 Yes Yes Yes Yes Yes Yes No Yes SAMDE 1 1 1 3 1 2 2 2 2 1.66
36 Yes Yes Yes Yes Yes Yes No Yes Fmincon 8 8 8 8 8 8 8 8 8 8
66 Yes Yes Yes Yes Yes Yes No Yes AIPDE 3 3 1 1 2 1 1 1 1 1.55
10 H. ZAHEER AND M. PANT

Case III: For N ¼ 4


Fitness Value
From Table 10, for 18 points monitoring system it can be
observed that AIPDE performs best in terms of best, worst, aver-
age value, and standard deviation. SaMDE, MRLDE, and DERL
also give good results. From Table 11, it can be seen that for 36
points monitoring system AIPDE gives best value for best and
worst value and second best for average and standard deviation.
Figure 6. Average rank bar graph of all algorithms.
For 66 points monitoring system (Table 12), AIPDE is best for all
fitness value, while fmincon gives the worst results.

Table 16. Comparison of results (dsL ¼ drL ¼ 1.0; dsU ¼ drU ¼ 3.0) obtained using different methods.
Serial number dt ds,min ¼ dr,min ds,max ¼ dr,max dsopt opt
dr1 opt
dr2 opt
dr3 Nopt Minimum cost
Proposed method 2
1 6.0 1.0 3.0 2.9999 2.9990 – – 1 1.538
2 8.0 1.0 3.0 2.0903 2.9160 2.9937 – 2 2.113
3 8.5 1.0 3.0 2.9899 2.9703 2.6013 – 2 2.152
4 9.0 1.0 3.0 2.9999 3.000 3.000 – 2 2.320
5 9.5 1.0 3.0 2.0788 2.3423 3.0000 2.0788 3 2.102
6 10 1.0 3.0 2.0001 2.9999 2.9998 2.9889 3 2.760
Totally disturbed PSO (TDPSO)
1 6.0 1.0 3.0 2.9999 2.9990 – – 1 1.736
2 8.0 1.0 3.0 2.0924 2.9171 2.9893 – 2 2.277
3 8.5 1.0 3.0 2.9945 2.9852 2.5192 – 2 2.347
4 9.0 1.0 3.0 2.9990 3.000 2.9999 – 2 2.416
5 9.5 1.0 3.0 2.0788 2.3423 3.00 2.0788 3 2.750
6 10 1.0 3.0 2.1111 2.8212 2.0785 2.9880 3 2.819
Genetic algorithm (GA)
1 6.0 1.0 3.0 3.00 3.00 – – 1 1.940
2 8.0 1.0 3.0 2.99 2.91 2.10 – 2 2.480
3 8.5 1.0 3.0 3.00 2.74 2.76 – 2 2.550
4 9.0 1.0 3.0 3.00 3.00 3.00 – 2 2.621
5 9.5 1.0 3.0 2.99 2.19 2.22 2.10 3 2.952
6 10 1.0 3.0 2.99 2.64 2.10 2.27 3 3.022
Simulated annealing (SA)
1 6.0 1.0 3.0 3.00 3.00 – – 1 1.940
2 8.0 1.0 3.0 2.92 2.92 2.09 – 2 2.480
3 8.5 1.0 3.0 2.99 2.70 2.80 – 2 2.550
4 9.0 1.0 3.0 3.00 3.00 3.00 – 2 2.621
5 9.5 1.0 3.0 2.98 2.34 2.08 2.10 3 2.953
6 10 1.0 3.0 3.00 2.17 2.68 2.15 3 3.022
Ant colony optimization (ACO)
1 6.0 1.0 3.0 3.00 3.00 – – 1 1.940
2 8.0 1.0 3.0 2.985 2.705 2.095 – 2 2.480
3 8.5 1.0 3.0 3.00 2.515 3.00 – 2 2.551
4 9.0 1.0 3.0 3.00 3.00 3.00 – 2 2.621
5 9.5 1.0 3.0 3.00 2.180 2.16 2.16 3 2.950
6 10 1.0 3.0 3.00 2.095 2.175 2.73 3 3.019
Gupta, Batra, and Lal model
1 6.0 1.0 3.0 3.00 3.00 – – 1 1.940
2 8.0 1.0 3.0 3.00 2.10 2.90 – 2 2.481
3 8.5 1.0 3.0 3.00 2.50 3.00 – 2 2.551
4 9.0 1.0 3.0 3.00 3.00 3.00 – 2 2.621
5 9.5 1.0 3.0 2.90 1.90 2.80 1.90 3 3.005
6 10 1.0 3.0 3.00 2.10 2.80 2.10 3 3.022
Shin and Joo model
1 6.0 1.0 3.0 2.50 1.00 – – 3 2.385
2 8.0 1.0 3.0 2.33 1.00 – – 4 2.927
3 8.5 1.0 3.0 2.50 1.00 – – 4 2.996
4 9.0 1.0 3.0 2.67 1.00 – – 4 3.068
5 9.5 1.0 3.0 2.83 1.00 – – 4 3.134
6 10 1.0 3.0 3.00 1.00 – – 4 3.206
Fmincon
1 6.0 1.0 3.0 2.3523 2.5683 – – 1 4.5673
2 8.0 1.0 3.0 2.2320 2.3253 2.2540 – 2 7.347
3 8.5 1.0 3.0 2.6224 2.5436 2.6544 – 2 7.658
4 9.0 1.0 3.0 2.3213 3.3649 2.9879 – 2 8.465
5 9.5 1.0 3.0 2.2155 2.6854 3.6546 2.6546 3 8.765
6 10 1.0 3.0 2.1025 2.3513 2.3213 2.3543 3 8.884
MATERIALS AND MANUFACTURING PROCESSES 11

Accuracy of Average Location and Sound Power Level Table 18. Average rank using Friedman test.
From Table 13, it can be seen that the average calculated dt ¼ 6 dt ¼ 8 dt ¼ 8.5 dt ¼ 9 dt ¼ 9.5 dt ¼ 10 Average ranks
location is very near to the actual value when evaluated Proposed 1 1 1 1 1 1 1
through AIPDE for all monitoring points. Similarly, AIPDE TDPSO 2 2 2 2 2 2 2
GA 3 3 3 3 4 4 3.33
gives exact value of sound power level (105, 90, 100, and SA 3 3 3 3 5 4 3.5
85 db) for all 18, 36, and 66 monitoring systems. ACO 3 3 3 3 3 3 3
GBL 3 3 3 3 6 4 3.66
SJ 7 7 7 7 7 7 7
Feasibility Fmincon 8 8 8 8 8 8 8
For N ¼ 4 (Table 14), all algorithms give feasible solutions for
M ¼ 18, 36, and 66 monitoring points except fmincon.

Statistical Analysis
Except for fmincon, which is visibly giving the worst results,
the performance of all the other algorithms look more or less
similar in nature, therefore a statistical comparison using the
nonparametric Friedman test is done to evaluate the average
rank of the algorithms and to indicate the significant perfor-
mance improvement of AIPDE over other variants. Fried-
man test can be used to detect the differences between the
performances of the algorithms over multiple problems.
The average rank of all equipment with 18, 36, and 66 moni- Figure 7. Average rank bar graph of all algorithms.
toring points is given in Table 15. From the table, it can be
seen that the rank obtained by Friedman test for AIPDE is
1.55 that is best among all. SaMDE with a rank of 1.66 can Bounds of depth of cut for finishing and roughing opera-
be considered as second best. Fmincon is showing worst tions are taken as 1 and 3 mm, respectively. The total opti-
rank, that is, 8. The performance of all the algorithms can mal production cost obtained by different methods, viz.
be seen more clearly with the help of Fig. 6, which gives TDPSO, GA, SA, ACO, Gupta et al., and Shin and Joo
the bar graph of average ranks for different algorithms. are reported in Table 16. The objective function, that is,
In this problem the performance measures used are: to minimize the cost for the proposed strategy is best
a) Minimum cost obtained in terms of different depths of cut. among all in terms of different depth of cut ranging from
b) Average performance improvement of the proposed 6 to 10. The comparative performance of all algorithms is
algorithm in comparison to other algorithms. shown in Table 19. From this table, it can be observed that
c) Statistical analysis to measure the performance of different the total production cost obtained by the proposed method
algorithms in terms of the obtained rank. Minimum cost is better in comparison with other reported methods for
is highlighted in boldface in Table 16. most values of depth of cut. The optimal results of the pro-
posed algorithm are shown in Table 17 in terms of depth of
Minimum Cost Obtained cut, feed rate, and cutting speed.
For this problem, the optimal results for minimizing the
total production cost and machining parameters are
reported by taking same variable bound of depth of cut Statistical Analysis
for finishing and roughing operations. Results are taken
In Table 18, average rank obtained for all the algorithms
by varying the value of depth of cut from 6 to 10 mm.
exploited for this problem through Friedman test are shown.
The rank for different depths of cut is calculated, and in the
last column the average rank is given. The calculated rank
Table 17. Optimal parameters for (dsL ¼ drL ¼ 1.0; dsU ¼ drU ¼ 3.0). using the proposed method is 1, implying the best perfor-
dt 6.0 8.0 8.5 9.0 9.5 10.0 mance of the proposed algorithm. Further, it is seen that
ds 3.0 2.0903 2.9899 2.9999 2.0788 2.0001
dr1 3.0 2.9160 2.9703 3 2.3423 2.9999
TDPSO obtains the second best rank, that is, 2. Ranks of all
dr2 – 2.9937 2.6013 3.0 3.0 2.9998
dr3 – – – – 2.0788 2.9889 Table 19. Percentage improvement of the proposed algorithm with other
fs 0.562430 0.8126 0.5507 0.5628 0.9 0.8354 variants.
fr1 0.562430 0.578 0.5654 0.5576 0.7842 0.6775
dt 1/2 1/3 1/4 1/5 1/6 1/7 1/8
fr2 – 0.6013 0.6990 0.5576 0.7813 0.9
fr3 – – – – 0.9 0.6947 6.0 11.40 20.72 20.72 20.72 20.72 35.51 66.32
Vs 123.04680 110.7863 123.2578 123.0233 110.7445 112.2473 8.0 7.20 14.79 14.79 14.79 14.83 27.81 71.23
Vr1 123.04680 122.1547 123.1475 123.3368 113.7355 122.3698 8.5 8.30 15.60 15.60 15.64 15.64 28.17 71.89
Vr2 – 123.1547 118.1440 123.3368 123.6968 112.4776 9.0 3.97 11.48 11.48 11.48 11.48 24.38 72.59
Vr3 – – – – 113.7447 123.4636 9.5 23.56 28.79 28.81 28.74 30.04 32.92 76.01
n 1 2 2 2 3 3 10 2.09 8.66 8.66 8.57 8.66 13.91 68.93
Minimum 1.5380 2.1130 2.1520 2.3200 2.1020 2.7600 Average % 9.42 16.67 16.68 16.66 16.89 27.11 71.16
cost ($) improvement
12 H. ZAHEER AND M. PANT

the algorithms are shown in Fig. 7. Fmincon is having the In case of industrial optimization problems, improvement
worst rank (i.e., 8) among all the algorithms. in result even by a few significant digits can be of huge
importance and therefore it can be said that although all the
Percentage Improvement of the Proposed Algorithm in algorithms taken in this study, except for fmincon, gave good
Comparison with Other Algorithms results, the solutions obtained by DE variants are slightly
Table 19 shows a pair-wise comparison of proposed strategy better, and therefore the proposed schemes form an attractive
with other strategies and gives the percentage of the proposed alternative for dealing with such problems.
strategy in comparison with other variants. Percentage
improvement of the proposed algorithm with other algorithms
is shown in each column with respect to different depth of cuts Funding
indicated in each row. In the last row, the average percentage of
The reported study was partially supported by DST, through research
depth of cut of each algorithm is shown. It can be seen from project number INT/RFBR/P-164.
this table that except for TDPSO, the percentage improvement
of the proposed algorithm is more than 15% in comparison
with other algorithms going up to 71% improvement in case References
of fmincon.where 1, 2, 3, 4, 5, 6, 7, 8 are proposed method 2;
[1] Chakraborti, N.; Shekhara, A.; Singhala, A.; Chakraborty, S.;
TDPSO; GA; SA; ACO; Gupta, Batra, and Lal model; Shin Chowdhury, S.; Sripriya, R. Fluid flow in hydrocyclones optimized
and Joo model; and fmincon, respectively. through multi-objective genetic algorithms. Materials and
Manufacturing Processes 2008, 16, 1023–1046.
[2] Jourdan, L.; Schutze, O.; Legrand, T.; Talbi, E.-G.; Wojkiewicz, J.L.
Conclusion An analysis of the effect of multiple layers in the multi-objective
design of conducting polymer composites. Materials & Manufactur-
The present study focuses on finding the optimal solution to ing Process 2009, 24, 350–357.
two fundamental but challenging problems arising in indus- [3] Baraskar, S.S.; Banwait, S.S.; Laroiya, S.C. Multiobjective
tries through modified variants of DE. In the first problem, optimization of electrical discharge machining process using a
the objective is to identify the noise source in a multi-noise hybrid method. Materials & Manufacturing Process 2013, 28,
348–354.
plant and to reduce the sound power level. In the second prob- [4] Giri, B.K.; Pettersson, F.; Saxén, H.; Chakraborti, N. Genetic
lem, the objective is to optimize process parameters for a programming evolved through bi-objective genetic algorithms
multi-pass turning process so as to minimize the total cost. applied to a blast furnace. Materials and Manufacturing Processes
Some concluding remarks are as follows: 2013, 28 (7), 776–782.
1. Both the problems are tested for different levels of complex- [5] Farshid, Jafarian; Domenico, Umbrello; Saeid, Golpayegani; Zahra,
Darake. Experimental investigation to optimize tool life and surface
ities by increasing the number of monitoring points as 18, roughness in inconel 718 machining. Materials and Manufacturing
36, and 66 in the first case study, and by considering the Processes 2016, 31 (13), 1683–1691, doi:10.1080/10426914.2015.
number of passes as 1, 2, and 3 with depth of cut ranging 1090592
from 6 to 10 in the second case study. [6] Chockalingam, K.; Jawahar, N.; Praveen, J. Enhancement of
2. For both the problems, it is observed that the proposed var- anisotropic strength of fused deposited ABS parts by genetic algor-
ithm. Materials and Manufacturing Processes 2015. doi:10.1080/
iants of DE performed better than the other algorithms 10426914.2015.1127949
taken for comparison. [7] Nagaraju, S.; Vasantharaja, P.; Chandrasekhar, N.; Vasudevan, M.;
3. In the first example, it can be seen that for lesser number of Jayakumar, T. Optimization of welding process parameters for
variables all the algorithms gave more or less similar results; 9Cr-1Mo steel using RSM and GA. Materials and Manufacturing
however, AIPDE steadily gave a good performance even Processes 2016, 31 (3), 319–327, doi:10.1080/10426914.2015.
1025974
when the number of variables were increased. It can also [8] Simon, Klancnik; MiranBrezocnik; JozeBalic; Isak, Karabegovic.
be seen that the performance of fmincon deteriorated with Programming of CNC milling machines using particle swarm
the increase in the number of variables, indicating that for optimization. Materials and Manufacturing Processes 2013, 28 (7),
larger number of variables it may not be a suitable tech- 811–815, doi:10.1080/10426914.2012.718473
nique. Fmincon also suffered from feasibility issues when [9] Pawar, P.J.; Rao, R.V.; Davim, J.P. Multiobjective optimization of
grinding process parameters using particle swarm optimization
the number of points was increased. algorithm. Materials and Manufacturing Processes 2010, 25 (6),
4. For the second case study also, it can be seen that the results 424–431, doi:10.1080/10426910903124860
obtained by the proposed strategy are better than the results [10] Munish, Kumar Gupta; Sood, P.K.; Vishal, S. Sharma. Machining
obtained by other algorithms. Once again it is seen that the parameters optimization of titanium alloy using response surface
performance of fmincon deteriorates with the increase in methodology and particle swarm optimization under minimum-
quantity lubrication environment. Materials and Manufacturing
the number of variables and also it cannot meet the Processes 2016, 31 (13), 1671–1682, doi:10.1080/10426914.2015.
feasibility criteria. 1117632
5. Statistical comparison done through nonparametric Fried- [11] Yiming, Li; Yu-Hsiang, Tseng. Hybrid differential evolution and
man test also indicates the competitive performance of particle swarm optimization approach to surface-potential-based
the proposed algorithms for both the problems. Moreover, model parameter extraction for nanoscale MOSFETs. Materials
and Manufacturing Processes 2011, 26 (3), 388–397, doi:10.1080/
the improvement percentage calculated for the second 10426914.2010.526977
problem shows that the proposed method can provide [12] Chakraborti, N.; Das, S.; Jayakanth, R.; Pekoz, R.; Erkoç, Ş. Genetic
better results approximately by 10%. algorithms applied to Liþ ions contained in carbon nanotubes: An
MATERIALS AND MANUFACTURING PROCESSES 13

investigation using particle swarm optimization and differential [29] Yildiz, A.R. Hybrid Taguchi-differential evolution algorithm for
evolution along with molecular dynamics. Materials and optimization of multi-pass turning operations. Applied Soft
Manufacturing Processes 2007, 22 (5), 562–569, doi:10.1080/ Computing 2012.
10426910701319605 [30] Yildiz, Ali R. Hybrid Taguchi-differential evolution algorithm for
[13] Rao, R.V.; Pawar, P.J.; Davim, J.P. Parameter optimization of optimization of multi-pass turning operations. Applied Soft
ultrasonic machining process using nontraditional optimization Computing 2013, 13 (3), 1433–1439.
algorithms. Materials and Manufacturing Processes 2010, 25 (10), [31] Lu, Chao; et al. Energy-efficient multi-pass turning operation using
1120–1130, doi:10.1080/10426914.2010.489788 multi-objective backtracking search algorithm. Journal of Cleaner
[14] Mitra, K. Genetic algorithms in polymeric material production, Production 2016.
design, processing and other applications: A review. International [32] Storn, R.; Price, K. Differential evolution—A simple and efficient
Materials Reviews 2008, 53 (5), 275–297. heuristic for global optimization over continuous spaces. Journal
[15] Mittal, Prateek; KedarKulkarni; KishalayMitra. A novel hybrid opti- of Global Optimization 1997, 11 (4), 341–359.
mization methodology to optimize the total number and placement [33] Brest, J.; Greiner, S.; Boskovic, B.; Mernik, M.; Zumer, V. Self-
of wind turbines. Renewable Energy 2016, 86, 133–147. adapting control parameters in differential evolution: A comparative
[16] Mittal, Prateek; KedarKulkarni; KishalayMitra. Multi-objective opti- study on numerical benchmark problems. IEEE Transactions on
mization of energy generation and noise propagation: A hybrid Evolutionary Computation 2006, 10 (6), 646–657.
approach. In Indian Control Conference (ICC). IEEE, 2016 [34] Storn, R. On the usage of differential evolution for function optimiza-
[17] Munjal, M.L. Acoustics of Ducts and Mufflers with Application to tion. In Fuzzy Information Processing Society. NAFIPS, 1996. Biennial
Exhaust and Ventilation System Design. John Wiley & Sons: Conference of the North American, IEEE, 1996, June, pp. 519–523.
New York, 1987. [35] Parouha, Raghav Prasad; KedarNath, Das. A memory based
[18] Lord, H.; Gatley, W.S.; Evensen, H.A. Noise Control for Engineers. differential evolution algorithm for unconstrained optimization.
McGraw-Hill: New York, 1985. Applied Soft Computing 2016, 38, 501–517.
[19] Yeh, L.J.; Chang, Y.C.; Chiu, M.C. Optimization of allocation and [36] Zaheer, Hira; et al. A Novel mutation strategy for differential
noise reduction on multi-noises system by using genetic algorithm. evolution. Problem Solving and Uncertainty Modeling through
Noise Vibration Worldwide 2004, 35 (4), 11–8. Optimization and Soft Computing Applications 2016, 20.
[20] Lan, T.S.; Chiu, M.C. Identification of noise sources in factory’s [37] Zaheer, Hira; et al. A Portfolio analysis of ten national banks
sound field by using genetic algorithm. Applied Acoustics 2008, through differential evolution. In Proceedings of Fifth International
69, 733–750. Conference on Soft Computing for Problem Solving. Springer:
[21] Kumar, Pravesh; Millie, Pant. Noisy source recognition in multi Singapore, 2016.
noise plants by differential evolution. Swarm Intelligence (SIS), [38] Das, S.; Suganthan, P.N. Differential evolution: A survey of the
2013 IEEE Symposium on IEEE, 2013. state-of-the-art. IEEE Transactions on Evolutionary Computation
[22] Sharma, Tarun Kumar; Millie, Pant. Identification of noise in multi 2011, 15 (1), 4–31.
noise plant using enhanced version of shuffled frog leaping algor- [39] Deb, Kalyanmoy. An efficient constraint handling method for
ithm. International Journal of System Assurance Engineering and genetic algorithms. Computer Methods in Applied Mechanics and
Management 1–9. Engineering 2000, 186 (2), 311–338.
[23] Gopalakrishnan, B.; Al-Khayyal, F. Machine parameter selection for [40] Chauhan, Pinkey; Millie, Pant; Kusum, Deep. Parameter optimiza-
turning with constraints: an analytical approach based on geometric tion of multi-pass turning using chaotic PSO. International Journal
programming. International Journal Production Research 1991, of Machine Learning and Cybernetics 2015, 6 (2), 319–337.
29 (9), 1897–1908. [41] Shin, Y.C.; Joo, Y.S. Optimization of machining conditions with
[24] Prasad, A.V.S.; Rao, R.K.; Rao, V.K.S. Optimal selection of process practical constraints. International Journal Production Research
parameter for turning operation in CAPP system. International 1992, 30 (12), 2907–2919.
Journal Production Research 1997, 35 (6), 1495–1522. [42] Fan, H.Y.; Lampinen, J. A trigonometric mutation operation to
[25] Hui, Y.V.; Leung, L.C.; Linn, R. Optimal machining conditions with differential evolution. Journal of Global Optimization 2003, 27,
cost of quality and tool maintenance for turning. International 105–129.
Journal of Production Research 2001, 39 (4), 647–665. [43] Kaelo, P.; Ali, M.M. A numerical study of some modified differential
[26] Saravanan, R.; Ashokan, P.; Sachithanandam, M. Comparative evolution algorithms. European Journal of Operational Research
analysis of conventional and non-conventional optimization tech- 2006, 169, 1176–1184.
nique for CNC-turning process. International Journal of Advance [44] Kumar, Pravesh; Millie, Pant. Enhanced mutation strategy for differ-
Manufacturing Technology 2001, 17, 471–476. ential evolution. 2012 IEEE Congress on Evolutionary Computation,
[27] Sharma, V.S.; Dhiman, S.; Sehgal, R.; Sharma, S.K. Estimation of IEEE, 2012.
cutting forces and surface roughness for hard turning using neural [45] Hwang, Chii-Ruey. Simulated annealing: Theory and appli-
networks. International Journal of Advance Manufacturing Tech- cations. Acta Applicandae Mathematicae 1988, 12 (1), 10–11.
nology 2008, 19 (4), 473–483. [46] Dorigo, Marco; Mauro, Birattari; Thomas, Stutzle. Ant colony optimi-
[28] Chandrasekaran, M.; Muralidhar, M.; Krishna, C.; Dixit, U. zation. IEEE Computational Intelligence Magazine 2006, 1 (4), 28–39.
Application of soft computing techniques in machining performance [47] Gupta, R.; Batra, J.L.; Lal, G.K. Determination of optimal subdivision
prediction and optimization: A literature review. International Journal of depth of cut in multipass turning with constraints. International
of Advanced Manufacturing Technology 2010, 46 (5), 445–464. Journal of Production Research 1995, 33 (9), 2555–2565.

You might also like