Accepted Manuscript

An effective hybrid multi objective evolutionary algorithm for solving real time
event in flexible job shop scheduling problem

M.B.S. Sreekara Reddy, Ch. Ratnam, G. Rajyalakshmi, V.K. Manupati

PII: S0263-2241(17)30587-0
DOI: http://dx.doi.org/10.1016/j.measurement.2017.09.022
Reference: MEASUR 4972

To appear in: Measurement

Received Date: 14 May 2016

Revised Date: 27 April 2017
Accepted Date: 12 September 2017

Please cite this article as: M.B.S. Sreekara Reddy, Ch. Ratnam, G. Rajyalakshmi, V.K. Manupati, An effective
hybrid multi objective evolutionary algorithm for solving real time event in flexible job shop scheduling problem,
Measurement (2017), doi: http://dx.doi.org/10.1016/j.measurement.2017.09.022

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
An effective hybrid multi objective evolutionary algorithm for solving real time event in
flexible job shop scheduling problem

a
M B S Sreekara Reddy, bCh Ratnam, cRajyalakshmi.G and d*V.K.Manupati
a
Department of Mechanical Engineering, K L University, Green Fields, Vaddeswaram, Guntur district,
A.P,INDIA.
b
Dept. of Mechanical engineering, AUCE, Visakha Patnam, AP, India.
c
Department of Manufacturing Engineering, VIT University, Vellore, Tamil Nadu, INDIA.
d
Department of Mechanical Engineering, VIT University, Vellore, Tamil Nadu, INDIA.
Mobile: +91 9775627564
Email: [email protected], [email protected], [email protected], [email protected]
* Corresponding author: [email protected]

Abstract This paper addresses the multi-objective model for a flexible job shop
scheduling problem (FJSSP) to improve the system performance under the condition of
machines break down as a real time event. It is important to identify the relevant performance
measures to the mentioned problem for examining the system performance. Therefore,
minimization of make span and minimization of total machine load variation is considered as
two performance measures. Generally, it is very difficult to develop a mathematical model
for the real-time situations in FJSSP. Hence, in this paper we divided the research work into
two folds: Primarily, a mixed-integer non-linear programming (MINLP) model has been
developed to represent the above-mentioned multi-objectives that subjected to constraints
without considering machines break down. Secondarily, by incorporating the machines break
down as the real-time event the performance of the system is examined. Solving conflicting
objectives simultaneously for finding the optimal/near optimal solutions in a reasonable time
is a challenge. In this paper, we proposed a new evolutionary based multi-objective teacher
learning-based optimization algorithm (MOTLBO) to solve the above-mentioned complex
problem. Moreover, to improve the obtained solutions a local search technique has been
incorporated in the MOTLBO and comparisons has been made with existing multi-objective
particle swarm optimization (MOPSO) and conventional non-dominated sorting genetic
algorithm (CNSGA-II). Results found that the proposed multi-objective-based hybrid meta-
heuristic algorithm produced high-quality solutions as proved by the tests we performed over
a number of randomly generated test problems. Finally, comparisons also made with how the
machines break down can affect the proposed systems performance.

Keywords: Flexible job shop. Multi-objective evolutionary algorithm. NP-hard.

Optimization.
1. INTRODUCTION

Current manufacturing environment has to change its trend due to recent advancements in
information and communication technology, the requirement of customized products from
customers, and huge competition between the industries. To cater these mentioned
requirements, particularly in shop floor, it is necessary to choose effective manufacturing
strategies that need to integrate manufacturing functions such as process planning and
scheduling effectively and efficiently. In recent years, many researchers try to achieve the
feasible scheduling with extension version of classical job shop scheduling problem (JSSP)
i.e. FJSSP due to its wide variety of applications.

In JSSP, processing of each operation for a job can be only possible on one machine, but
in the case of FJSSP, each operation of a job can be processed on several machines for a
given set (Brucker and Schile, 1990). Therefore in FJSSP, it is very hard to identify the
assigned operations on a particular machine out of a set of competent machines and then
identify the sequence of operations is a complex task and it is much complex than JSSP.
Nouri, Driss and Ghédira (2016) generalized the classical job shop scheduling problem by
integrating the transportation times and many robots problem where a set of jobs are
processed by a set of alternative machines and transported by several robots. From the
literature, job shop scheduling problem is known as a NP-hard problem (Garey et al.,
1976) thus, it can be concluded that FJSSP as NP-hard combinatorial optimization
problems.

In general, FJSSPs have been solving by two different approaches i.e. hierarchical and
integrated approach (Zhang et al. 2011). In hierarchical/traditional approach, process
planning and scheduling have been conducted in a sequential manner. In other words, to
solve the problem with the hierarchical approach, divide the problem into sub-problems
and then solve it individually (Brandimarte, 1993). Javid and Hooshangi (2017) addressed
a ternary-integration scheduling problem which incorporated employee timetabling into
the schedules of machines and transportation systems in a job shop environment and also
proved that the effectiveness and efficiency of Anarchic Society Optimization (ASO) in
the integrated job-shop scheduling problem. The greatest difficulty with this approach is
there might be a chance of losing good quality solutions and obtaining a compromising
solution for both the objectives. Saygin and Kilc (1999) mentioned various difficulties
with the traditional/hierarchical approach and the obstacles to improve the productivity of
the manufacturing systems. Recently, researchers have been developed many methods and
approaches for integrating the manufacturing functions to improve the performance of the
system (Manupati et al., 2012). Khoukhi, Boukachour and Alaoui (2017) focused on the
FJSSP with machine unavailability constraints for proceeding preventive maintenance
activities. With the objective of minimizing the make-span, they proposed the formulations
of a mixed integer nonlinear program and a bi-level disjunctive/conjunctive graph.

Different approaches to solving the FJSSP were proposed by (Hurink et al., 1994;
Najid et al. 2002). Yang (2001) proposed the multistage genetic algorithm to solve the
FJSSP for improving the performance measures of the system. The complex problem of
FJSSP with the consideration of controllable processing times (CPT) was explored by (Lu
et al., 2017) with the main aim of minimizing the makespan and the total additional
resource consumption. Tay and Ho (2008) combined different dispatching rules as
composite rules for solving the multi-objective FJSSP by improving the objective
functions such as minimization of makespan, mean tardiness, and mean flow time. With
sensitivity analysis, the validation of the presented rules has been conducted and found the
proposed approach robustness. Wang et al. (2012) presented the bi-population based
estimation of distribution algorithm to solve the FJSSP. Here, with Taguchi method, the
parameters have been investigated and then identified the better ones. With these identified
parameters the algorithm performance has been examined and found the best performance.
Kaplanoğlu (2016) provided an object oriented (OO) approach for solving the multi
objective FJSSP by using a simulated annealing optimization algorithm. Over the past few
years, much research and study (Wu and Xia 2005; Zhang 2010; Bagheri and Zandieh
2001; Ulungu et al. 1999) are done by many researchers and they focused on different
hybrid evolutionary algorithms to solve the multi-objective FJSSP. Moreover, to our
knowledge, a real-time event such as machine breakdown in FJSSP is difficult to be
expressed and it has not been studied well.

In this paper, an effective hybrid meta-heuristic (HMOTLBO) algorithm is introduced

to solve the real time event in multi-objective FJSSP. To carry out the multi-objective FJSSP,
we have carried out minimization of make span, and minimization of total machines
workload variation as performance measures and then developed a mixed integer non-linear
programming (MINLP) based mathematical model to represent the FJSSP. Moreover, we
have introduced the machine breakdowns randomly in the system as a real-time event to
examine the proposed manufacturing system. Finally, with different instances the proposed
HMOTLBO algorithm is implemented and solved, comparisons is made with well-
established multi-objective evolutionary algorithms, i.e., CNSGA-II, MOPSO. Finally, with
results we proved that the proposed algorithm outperforms other compared algorithms and
also examine the system behaviour at real time situations.

The remainder of this research work is organised as follows: Section 2 introduced the
problem description with basic assumptions and developed a multi-objective-based
mathematical model along with the constraints. In Section 3, CNSGA-II, MOPSO algorithms
are discussed and proposed hybrid MOTLBO is detailed for solving the proposed FJSSP
problem. Experiments with different problem instances are illustrated in Section 4. Section 5,
discussed obtained results of the proposed method. Finally, conclusions are drawn and future
work is suggested in Section 6.

2. Problem description

FJSSP is addressed in this research study that consists of a set of jobs j with k operations
where k varies between (1  k  n). Each operation kjm of jobs j is to be processed on one
machine m from the set of eligible machines Mkj. It is assumed that the processing times of
the operations to process on machines is known well in advance and all machines are
continuously available at time zero i.e. before scheduling of operations. Each machine can
only process one operation at a time. Thus the consecutive jobs can wait at buffers until its
preceding job can finish its process. Therefore, the buffers are considered as unlimited in
size. The setup time is considered and it is included in the processing time, and the
transportation time between two corresponding machines is assumed to be zero. In this paper,
we considered two performance measures such as minimization of makespan and workload
of machines with the intention of protecting machines from overuse. A mathematical model
has been developed with considered multi-objectives by assuming the real-time event as
machine breakdowns and repairs. In this study, it is assumed that the real time event occurs
randomly which follows an exponential distribution and the machines selected at the
beginning of the simulation can be termed as key machines. If the key machine fails its
process, then it is assumed that the real-time event occurs. Here, we consider the mean time
between failures (MTBF) and mean time to repair (MTTR) are two system parameters useful
for the analysis of the system effectiveness when machine breakdown exist (Zandeih and
Abidi, 2010). However, to meet the requirements from the real world production based
multi-objective problem is considered to be a new problem according to current
manufacturing scenario. From the literature it is clear that the FJSSP is NP-hard in nature, to
find compromising solution for the mentioned problem in a reasonable time is a challenge.
Thus an effective evolutionary algorithm based approach is proposed for enhancing the
performance of the system.

Assumptions:

(1) The considered machines before scheduling must be available at time zero.

(2) All jobs can be started at time zero.

(3) At a time each machine can process only one operation.

(4) Processing of operations on the machines should not be interrupted except when machine
breakdown.

(5) The sequence of operations of each job to further process is pre-specified.

(6) Neither release times nor due dates are specified.

(7) Job transportation time among machines is not considered.

Before describing the mathematical model, it is better to define some symbols and notations
are as follows:

Z The number of operations for each job as a set..

If z(j) = k, it defines that job j has k operations.

Cz(j)jk Completion time of Kth operation of job j on machine k

Trjk Processing time of rth operation of job j on machine k

Tk The sum of the time that machine k is doing process

Objective Functions:

F1: Min max C z ( j ) jk  ; z( j )  Z ; j  N ; k  M (1)

 n z( j) 
F 2 : Min max Tr    tljk  (2)
 j 1 l 1 

Subject to:
cljk  tljk  c( l 1) jk l  1, 2,3,..., z ( j ); (3)
j  1, 2,3,..., n; k  1, 2,..., m
cljk  tljk  clij ; l  1, 2,..., z ( j ); (4)
i  1, 2,..., n; j  1, 2,..., m
cljk  0; l  1, 2,..., z ( j ); j  1, 2,..., n; k  1, 2,..., m (5)

The above-mentioned objectives, i.e. minimization of make-span (F1) and

minimization of workload on the most loaded machine (F2) are given by Eqs (1) and (2).
Constraint (3) and (4) indicates processing constraints where the preceding constraints among
operations of the same job should follow and each machine should be available to other
operations if the concerned operations complete.

3. Multi-objective evolutionary algorithm

A comprehensive study on multi-objective optimization by meta-heuristics was first

discussed (Andrej 2001). Schaffer (1985) proposed the vector evaluated genetic algorithm
(VEGA) which is a modified version of the single-objective genetic algorithm. But, Coello
(1999) observed that the non-dominated solutions produced from VEGA are almost
concentrating on the middle region of the Pareto front, rather than spread in the Pareto fronts.
(Fonseca and Fleming 1995), proposed a method to overcome the mentioned difficulties, one
can find the detailed information of the recent multi-objective optimization for conflicting
objectives and how these algorithms can work effectively and efficiently can be found in
paper of (Zitzler et al. 2004). In our research work, we have used conventional multi-
objective-based non-dominated sorting genetic algorithm as a benchmark and recently
developed MOPSO algorithm. Later, with proposed hybrid MOTLBO both the algorithms is
compared. The detail description of the mentioned algorithms is as follows:
3.1 Non-dominated sorting genetic algorithm (NSGA-II)

The Non-dominated sorting genetic algorithm has been considered as one of the best
algorithms to solve multi-objective problems. It has operators such as non-dominated sorting
and crowding distance to improve the solution qualities by considering both the objectives
simultaneously. Goldberg (1989) and Srinivas and Deb, (1995) highlighted various
drawbacks of the non-dominated sorting genetic algorithm (NSGA) and suggested the
improved version of NSGA such as NSGA-II. In this work, we adopted NSGA-II to solve the
proposed problem and the details of the Schematic procedure for processing NSGA-II is
shown in Fig. 1.

<<< Insert Fig. 1 here>>>

3.2 Multi-objective particle swarm optimization algorithm (MOPSO)

In the area of manufacturing and production, particularly for FJSS based single objective
optimization problems, population-based particle swarm optimization (PSO) algorithm has
been used for obtaining an optimal solution. Its evolution lies from the concepts of social
behaviour of bird flocking and fish schooling. However, many researchers implemented this
algorithm in a wide range of applications. Kennedy and Eberhart (1995) briefed a variety of
applications that has been solved with particle swarm optimization algorithm. Xia and Wu
(2005) adopted a linear weighted summation approach in which aggregation of three
objective values has been simultaneously solved with hybrid PSO and simulated annealing
(SA) algorithm. Here, PSO strives for searching of good routing decisions, and SA algorithm
explore good sequencing decisions which leads to finding better optimal solutions. However,
to solve for multi-objective problems and to obtain best compromising solutions the existing
PSO algorithm has to be modified. Recently, a Pareto archive PSO was developed for the
multi-objective FJSSP to obtain a set of Pareto optimal solutions (Lei, 2009). G. Zhang, X.
Shao, P. Li and L. Gao (2009) developed an improved GA combined with tabu search (TS)
and also PSO combined with TS to solve the multi-objective FJSSP.

In this research work, we have incorporated non-dominated sorting and crowding distance
operators that may lead to high speed of convergence to find the multi-objective solutions.
The schematic procedure and the flowchart of MOPSO is shown as follows:

Step 1 Initialize swarms as random population

Step 2 Calculate the velocities for every particle.

Step 3 Set of leaders is initialized with solutions obtained from non-dominated sorting
operators and those are stored in the external archive.

Step 4 Quality measures such as velocity, position and cognitive learning factor are
considered for selecting the leaders in the archive.

Step 5 For each iteration update velocity and position of each particle in the swarm.

Step 6 Evaluate fitness function to sort pbest and gbest.

Step 7 Implemented non-dominated sorting and crowding distance operators to find the set
of optimal solutions spread across the optimal Pareto front.

Step 8 Update the solutions in the external archive to obtain optimal gbest

Step 9 Check the termination criteria. Otherwise, return to step 2.

3.3 Hybrid multi-objective teacher learning based optimization (MOTLBO)

In this paper, we developed a new kind of optimization approach to solve the proposed multi-
objective optimization problem called multi-objective-based teacher learning based algorithm
to obtain Pareto optimal solutions. Recent works on TLBO (Rao et al. 2011; Rao and Patel,
2012; Rao and Patel, 2013; Togan, 2012; Krishnan et al. 2011; Rao and Kalyankar, 2013),
proved that it has the capacity to solve wide variety of problems effectively and efficiently
for enhancing the systems performance. But, till date very limited applications with multi-
objective-based teacher learning optimization has been tested. In this research work, we
adopted the Pareto approach to the TLBO algorithm and then proposed a multi-objective
population based TLBO algorithm to solve the MOFJSSP. Moreover, we adopted a local
search technique to improve the convergence rate and quality of the solutions. The detailed
step-wise procedure of the proposed algorithm and its approach is described as follows:

Step 1 Initialization Stage: Initialize the population (learners), parameters of population size,
design variables (number of subjects offered to the learners) randomly with threshold
values, and termination criterion.

<<<<Insert table 1>>>>

 (1i , j )   min
j + rand ( i , j )  ( j
max
  min
j ) (6)
The initialization starts with randomly generated values, and it is represented with N rows
and K columns as the population size of the class and a total number of offered subjects. The
equation (6) represents the assigned randomly generated values for initial generation of ith
vector to the jth parameter. Where α represents the particular student i and rand (i, j)
represents uniformly distributed random variable ranging from 0 to 1. Equation (7) indicates
the parameters of the ith vector for the generation g.

<<<<Insert table 2>>>>

Z(gi )  [(gi ,1) , (gi ,2) ,...,(gi , j ) ,...,(gi , D ) ] (7)

  aig   fa( Z (i ) ) 
g

 g  (8)
  bi   fb( Z (i ) ) 
g

In this paper, we have carried out a multi-objective problem with a and b as two objectives
represented in Equation (8). The Table 1 represents an FJSSM problem of a manufacturing
industry which has ten jobs undergoing ten machining operations where each job can be
completed in any of the three sequences of operation. Table 2 is the initialized table which is
derived by substituting the values of Table 1 in Equation 6.

Step 2: Evaluate the knowledge of the learners as fitness values.

In any evolutionary algorithm premature convergence is avoided by one of a powerful
mechanism called as population sorting. In MOTLBO algorithm, we divide the population
into a number of groups and each groups having individual teachers. Thus population sorting
mechanism is provided by splitting the population into respective groups and adapting multi-
teacher concept. The multi-teacher concept selects the best learner of each group as a teacher
of its respective group and this teacher will be responsible for increasing the mean of the
group and trying to bring mean to its level. Once the group mean reaches the level of the
learner the entire group reaches the next group. The learner with least f(X) value is selected
as chief teacher and remaining teachers of the group is selected by the Equation 9.
Ts  f(Xb )  ri * f(Xb ) s = 2, 3, ........, N (9)

Step 3: Adapting teacher factor and assigning learners to the teacher

In TLBO, teacher factor used to be either 1 or 2 that is either students learns everything from
the teacher or nothing from the teacher, In real life situation the teacher-learner phenomenon
usually varies between 1 to 2 not either 1 or 2.So the earlier equation has to be modified in
such a way that we get near optimum value for teacher factor and this modified equation is
been represented in equation 10.
f (X k )
(TF ) s ,i = ( ) if Ts  0 ( 10 )
Ts
(TF )i = 1 if Ts = 0

Where f(Xk) is a result of any learner k associated with any group ‘s’ considering of i
iteration of all subjects and Ts is the teacher of the same group in the same iteration.
Assigning the learners is done by simple calculation that is shown below.
For k =1 to Population
If T1 ≤ f(Xk)< T2
Assign the learner f(Xk) to teacher 1 (i.e., T1).
Else If T2 ≤ f(Xk)< T3
Assign the learner f(Xk) to teacher 2 (i.e., T2)
.
.
.
Else If TN-1 ≤ f(Xk) < TN
Assign the learner f(Xk) to teacher N-1(i.e TN-1)
Else
Assign the learner f(Xk) to teacher TN .
End if
End For

The above algorithm illustrates how different students based on their performance can
be allotted to different teachers if the learning capacity for every student k i.e. f(Xk) is less
than that of the particular teacher who teaches the subject. This is followed by allotting that
student to that particular teacher for the purpose of his understanding and development. This
process continues till each and every student is allotted based on their performance in various
subjects under various teachers as per the requirement. Finally, the last student who is left is
well versed in all the subjects taught by teachers from 1 to N-1 and thus by default is assigned
the Nth teacher. For example if the student 1 is weak in a subject taught by the teacher 1 as
compared to the subject taught by teacher 2, the algorithm will allot the student under teacher
1.
The tables shown below represent teachers of the respective group and the grouping of the population.
The information of each table is mentioned below. Population group in a teaching phase for
minimization of make-span is presented in Table 3 and is derived from the entries of Table 2
which contains make-span values for each teacher and their calculated fitness value based on
fitness function f(x1). Table 4 presents the machine load population group in a teaching phase
derived from the entries in Table 2 containing the machine workload values for each teacher
and their calculated fitness value based on fitness function f(x2). Table 5 contains the groups
that are made within the jobs and their corresponding make-span values for each job and their
mean for each group. Similarly, Table 6 contains the groups that are made within the entries
of the machines for their respective workload values and their corresponding means for every
group.

<<<<Insert table 3>>>>

<<<<Insert table 4>>>>

<<<<Insert table 5>>>>

<<<<Insert table 6>>>>

Step 4 Teaching Phase:

In TLBO the student learners from the teacher of their respective classes but in real life, the
student not only learns from the class teacher but also learns from discussing with his
classmates and also from the tutorial teachers so there is a modification in the formula of to
calculate Xnew. The mean of the particular group is increased by the classroom teacher and
tutorial teacher. So the teacher phase formula consists of two parts which is shown in the
equation 11, 12, 13.
For s =1,2,....... till number of the group
For j = 1 to number of design variables
X 'jk  (X k  DM s , j ) + ri (X hj -X kj ) if f(X h ) < f(X k ) h  k (11)
or
X 'jk  (X k  DM s , j ) + ri (X kj -X hj ) if f(X k ) < f(X h ) h  k (12)
DM s , j  ri ( X sj  TF * Means , j ) (13)
X sj is the grade of the teacher
DM s , j is the difference mean
Means , j is the mean grade of the group

The f(Xnew) which has been generated for the each learner is compared with the f(X) value
and whichever values are lesser are retained and are carried to the next phase of MOTLBO.
The calculated DM values for the above problem are shown in Tables 7 and 8 and Xnew
values for the same problem is shown in Tables 9 and 10.

<<<<Insert table 7>>>>

<<<<Insert table 8>>>>

<<<<Insert table 9>>>>

<<<<Insert table 10>>>>

Step5: Learner phase

In learner phase of TLBO, the student learns from his peer and tries to improve his
performance, but in real life, the student learns from his peer not only from his peers but also
from improves his performance by self-motivation by improving his knowledge self-
studying. So in MOTLBO algorithm, there has been a modification in the search mechanism
so that the above condition is satisfied. The mathematical formula of learner phase of
MOTLBO is modified represented in the equation 13, 14 and 15.
For j = 1 to number of design variable
X 'jp,i  [ X pj ,i  ri (X pj ,i -X qj ,i )] + [ ri (X Sj ,i -E F *X pj ,i )], If f(X p ) < f(X q ) (13)
or
X 'jp,i  [ X pj ,i  ri (X qj ,i -X pj ,i )] + [ ri (X Sj ,i -E F *X pj ,i )], If f(X q ) < f(X p ) (14)
p  q and p,q,s  k,
X Sj is the grade of the teacher associated with group 's' in 'j' subject
EF = Round (1+ri ) (15)

EF is the exploring factor of the learner whose value can either be 1 or 2. The EF is the self-
motivation factor of the learner that means the student can learner can learn everything from
self-motivation or else nothing from it. The table 11 and 12 represents the final Xnew value of
learner phase of the first generation of the above problem. GANTT chart for scheduling of
jobs is presented in fig 2.
<<<<Insert table 11>>>>

<<<<Insert table 12>>>>

<<<< Insert Fig 2 >>>>>

Step 6 External Archive:

The main function of the external archive is to record of non-dominated vectors found along

the search process available in each generation. This algorithm uses a ε-dominance method to

store the data of each generation stored in the fixed sized external archeries to store best

solution that has been generated in that particular generation. The detailed study about ε-

dominance is explained in (Deb et al, 2005).

Based on the number of objective functions of the problem archive space will be located in
such a way that the dimension of the space is equal to the number of the objective function.
Depending on the number of objective functions that is two, three or more than three the
dimensional area of the space is divided into square, cubic or hypercubic shapes. In a box if
there is only one solution the value is stored, if there is more than one solution in a particular
box the values which can dominate the remaining values is kept and remaining values will be
removed, then the box have particular values is checked with remaining values and try to find
weather the boxes possess the property of non-dominance or not. If the values present in the
boxes are dominated by the remaining values then such values will be removed. The problem
is solved under MOTLBO are solved by grid-based approach for achieving the values.
 <<<<Insert table 13>>>>

<<<<Insert table 14>>>>

Once the values that have been received after teacher phase and learner phase is continued for
a certain number of 100 to 150 iterate or until there is no significance improvement in the
values or the values keeps on increasing. At the end of the algorithm the values that have
been stored in the external archive is given as output at the end of the iteration.
The table 13 and 14 represents the output received from make-span and m/c workload
respectively. For each table the other objective function values have been calculated. These
two tables actually explains what happens in the external archive where each table is better in
the objective function it is calculated from and deviating away from the satisfying from other
objective function. Once these values are found out and are verified and the one which gets
dominated is eliminated keeping the other values in the box.

4. Experimental Evaluation

To solve the proposed problem with the presented model different test problem instances are
considered from Brandimarte (1993). For example, we have shown the 10- jobs, 10-machines
FJSSP case as an instance in Table 1. The presented problem is complex, multi-objective and
NP-hard in nature, therefore, multi-objective evolutionary algorithm approach is considered
as a root to solve the problem. Here, with proposed hybrid MOTLBO algorithm the presented
test instances are solved. Consequently the performance of the proposed algorithm is
compared with those of the obtained NSGA-II and MOPSO. In Table. 15 the proposed and
considered algorithms are their parameter values are compared.

<<< Insert Table. 15 here>>>

In this study, we consider machines break down as a real-time event and its effect on
system performance is examined. To understand the failure rate, we assume all the machines
have the same mean time to repair (MTTR) and mean time between failure (MTBF) where
Avg = MTTR/(MTBF+MTTR), it denotes the percentage of machines those are broken and
have failures. For example, if MTTR = 5 time units and MTBF = 45 time units then Avg = 0.5,
this indicates on an average 45 time units the machine is available and the machine breaks
down with the mean of 5 time units. The attributes and its values are shown in Table. 16.
 <<< Insert Table. 16 here>>>

Among all parameters of the presented algorithms, population size and termination
criteria are most common and important parameters to investigate. In this paper, we have
considered difference values for termination criteria ranging from 150 to 300 by fixing
population size as constant with 200. To solve the proposed problem, each instance is run for
ten times by set the termination criteria for each algorithm. The average of ten runs is taken
as best solutions for each instance of two different cases is shown in Tables 17 and 18. The
algorithms were coded in MATLAB software and the problem is executed on a computer
with Intel® Core™2 Duo CPU T7250 @2.00GHz, 1.99 GB of RAM.

5. Results and Discussion

In this paper, the proposed systems performance and its behaviour when real-time event
occurs is examined with considered conflicting objectives such as minimization of makespan
and machine load variation. In this research work, a multi-objective evolutionary algorithmic
approach is considered due to their nature to solve the conflicting objectives simultaneously
and help to increase the solutions diversity with better computational time. Since these
algorithms use Pareto optimal fronts to represent the optimal solutions, the set of solutions at
the last iteration is considered as the Pareto optimal solutions. Ten different scenarios from
Brandimarte (1993), has been taken to conduct the experiments with developed hybrid
MOTLBO algorithm. Subsequently, comparisons with two existing multi-objective-based
algorithms including conventional NSGA-II, MOPSO is made to find the proposed algorithm
effectiveness.
<<< Insert Table. 17 here>>>
<<< Insert Table. 18 here>>>

5.1 Experimentation results without considering machines breakdown

Table 4 and 5 indicates that most of the obtained solutions for considered objectives are
dominated by solutions obtained from other algorithms. Fig. 3a, 3b, 3c, and 3d shows the
non-dominated solutions of the tested algorithms for randomly chosen instances MK02,
MK04, MK06, and MK08 out of ten instances in which machines break down is not
considered. The X-axis in Fig.3a to 3d indicates make-span as a performance measure and Y-
axis indicates machine load variation as a performance measure and the generated Pareto
optimal curves for three different algorithms are shown with three different symbols. The
plots illustrate the convergence and divergence solutions of the three algorithms which depict
the performance of the system. The extreme curve indicates high convergence and diversity
with optimal Pareto solutions.

<<<Insert Fig. 3a here>>>

<<<Insert Fig. 3b here>>>

<<<Insert Fig. 3c here>>>

<<<Insert Fig. 3d here>>>

5.2 Experimental results with consideration of machines breakdown

In this section, the performance of the systems is examined when machines break down occur
and also how the proposed algorithm improves the performance measures is detailed. We use
similar instances and generated machines breakdown randomly. As indicated in Table 3, the
related parameters for machine breakdown such as MTBF and MTTR are assumed with
repair time between failures of machines as an exponential distribution. Fig. 4a, 4b, 4c, and
4d shows the non-dominated solutions of the tested algorithms for randomly chosen instances
MK02, MK04, MK06, and MK08 out of ten instances. X-axis and Y-axis in the below-
mentioned figures show the considered performance measures and Pareto curves show the
performance of the algorithms.

<<<Insert Fig. 4a here>>>

<<<Insert Fig. 4b here>>>

<<<Insert Fig. 4c here>>>

<<<Insert Fig. 4d here>>>

<<<Insert Table. 19 here>>>

From the results it is suggested that the proposed MOTLBO is effective compared to
other algorithms where it produces well spread non-dominated solutions for almost all the
instances. As indicated in Table 19, the scheduling efficiency is affected when machines
breakdown is taken into consideration and the proposed algorithm consistently improved the
performance of the system.

6. Conclusion and future work

In this paper, several contributions for literature review with the proposed FJSSP has been
made. Primarily, a multi-objective based FJSSP has been proposed with machines break
down as a real time event. Secondarily, we developed a mathematical model with objective
functions as minimization of total completion time and machine load variations to test the
system performance. The proposed problem is a multi-objective and combinatorial
optimization in nature where finding the compromising solution for the conflicting objectives
by using conventional optimization techniques is very hard. In this regard, to solve this
problem as a third contribution, we proposed a hybrid multi-objective based evolutionary
algorithm i.e., MOTLBO to find the optimal solutions. With different benchmark data sets
available from the literature the problem has been solved for two different scenarios i.e., with
and without machine breakdown with various instances. Computational experiments have
shown that the proposed MOTLBO algorithm outperforms the other algorithms and is able to
give useful near optimum results for the above mentioned scenarios. Further, with already
established MOPSO and CNSGA-II algorithms the proposed algorithm is compared and from
the results it has been proved that the proposed algorithm shows its best over other algorithms
in almost all the test problems. Moreover, the proposed algorithm eliminates the limitation of
the variations in the results shown by the other algorithms by giving close results for both the
cases. Different examinations are conducted to test the robustness of the proposed system
when the machines break down occurs.

The future research work may include incorporation of more real-time events to make
the considered system as more practical FJJSP. The proposed approach may be applied to
improve the Pareto optimal fronts for generating best compromising non-dominated optimal
solutions for many objectives based FJSSP.

References

1. Brucker, P., Schile, R. (1990). Job shop scheduling with multi-purpose machines. Journal
of Computing, 45, 369–75.
2. Garey, MR., Johnson, DS., Sethi, R. (1976). The complexity of flow shop and job shop
scheduling. Mathematics of Operations Research, 1, 117–29.

3. Zhang, G., Gao, L., Shi, Y. (2011). An effective genetic algorithm for the flexible job shop
scheduling problem. Expert Systems with Applications, 38, 3563–73.

4. Brandimarte, P. (1993). Routing and scheduling in a flexible job shop by Tabu search
Annals. Journal of Operations Research, 41, 157–83.

5. Hurink, J., Jurisch, B., Thole, M. (1994). Tabu search for the job shop scheduling
problemwith multi-purpose machines. Journal of OR Spectrum, 15, 205–15.

6. Najid, NM., Dauzere-Peres, S., Zaidat, A. (2002). A modified simulated annealing

methodfor flexible job shop scheduling problem. IEEE international conference on
system, man and cybernetics, vol. 5.

7. Yang, JB. (2001). GA-based discrete dynamic programming approach for scheduling in
FMS environments. IEEE Transactions on Systems, Man and Cybernetics, Part B, 31,
824–35.

8. Tay, JX., Ho, NB. (2008). Evolving dispatching rules using genetic programming for
solving multi-objective flexible job shop problems. Journal of Computers &
Industrial Engineering, 54, 453–73.

9. Wang, L., Wang, S., Xu, Y., Zhou, G., Liu, M. (2011). A bi-population based estimation
ofdistribution algorithm for the flexible job-shop scheduling problem. Journal of
Computers & Industrial Engineering, 64, 917–26.

10. Zhang, Q., Manier, H., Manier, MA. (2011). A genetic algorithm with tabu search
procedure for flexible job shop scheduling with transportation constraints and
bounded processing times. Journal of Computer & Operations Research, 39, 1713–
23.

11. Xia, W., Wu, Z. (2005). An effective hybrid optimization approach for multi-objective
flexible job shop scheduling problems. Journal of Computer & Industrial
Engineering, 48, 409–25.

12. Zhang, GH., Shao, XY., Li, PG., Gao, L. (2010). A genetic algorithm and tabu search for
multiobjective flexible job shop scheduling problems. IEEE International Conference
on Computing. Journal of Control and Industrial Engineering, 1, 251–4.
13. Bagheri, A., Zandieh, M. (2011). Bi-criteria flexible job shop scheduling with sequence-
dependent setup times—variable neighbourhood search approach. Journal of
Manufacturing Systems, 30, 8–15.

14. Ulungu, EL., Teghem, J., Fortemps, PH., Tuyttens, D. (1999). MOSA method: a tool for
solving MOCO problems. Journal of Multi-Criteria Decision Analysis, 8, 221–36.

15. Kennedy, J., and Eberhart, RC. (2001). Swarm Intelligence. San Mateo, CA, Morgan
Kaufmann.

16. Xia, W., Wu, Z. (2005). An effective hybrid optimization approach for multi objective
flexible job-shop scheduling problems. Journal of Computers & Industrial
Engineering, 48, 409–425.

17. Zhang, G., Shao, X., Li, P., Gao, L. (2009). An effective hybrid particle swarm
optimization algorithm for multi-objective flexible job-shop scheduling problem.
Journal of Computers & Industrial Engineering, 56, 1309–1318.

18. Lei, D. (2008). Pareto archive particle swarm optimization for multi-objective fuzzy job
shop scheduling problems. International Journal of Advance Manufacturing
Technology, 37, 157–165.

19. Rao, RV., Savsani, V., Vakharia, D. (2011). Teaching–learning-based optimization: a

novel method for constrained mechanical design optimization problems. Journal of
Computer Aided Design, 43, 303–315.

20. Rao, RV., Patel, V. (2012). Multi-objective optimization of heat exchangers using a
modified teaching–learning-based optimization algorithm. Journal of Applied
Mathematical Model, 37, 1147–1162.

21. Rao, RV., Patel, V. (2013). Multi-objective optimization of two stage thermoelectric
cooler using a modified teaching–learning-based optimization algorithm. Journal of
Engineering Applied Artificial Intelligence, 26, 430–445.

22. Toğan, V. (2012). Design of planar steel frames using teaching–learning based
optimization. Journal of Engineering. Structures, 34, 225–232.

23. Krishnanand, KR., Panigrahi, BK., Rout, PK., Mohapatra, A. (2011). Application of
multiobjective teaching–learning-based algorithm to an economic load dispatch
problem with incommensurable objectives, in: B.K. Panigrahi et al. (Eds.), Swarm,
Evolutionary, and Memetic Computing, Springer Verlag, Berlin, 697–705.
24. Rao, RV., Kalyankar, V. (2013). Parameter optimization of modern machining processes
using teaching–learning-based optimization algorithm. Journal of Engineering
Applied Artificial Intelligence, 26, 524–531.

25. Goldberg, D. E., Korb, B., Deb, K. (1989). Messy genetic algorithms: Motivation,
analysis, and first results. Journal of Complex systems, 3(5), 493-530.

26. Deb, K., & Beyer, HG. (2001). Self-adaptive genetic algorithms with simulated binary
crossover. Journal of Evolutionary Computation, 9(2), 197-221.

27. Khoukhi, F., Boukachour, J., Alaoui, A. (2017). The “Dual-Ants Colony”: A novel hybrid
approach for the flexible job shop scheduling problem with preventive maintenance.
Computers & Industrial Engineering, 106(5), 236-255.

28. Lu, C., Li, X., Gao, L., Liao, W., Yi, J. (2017). An effective multi-objective discrete
virus optimization algorithm for the flexible job shop scheduling problem with
controllable processing times. Computers & Industrial Engineering, 104(8), 156-
174.

29. Javid, A., Hooshangi, P. (2017). Integrating employee timetabling with scheduling of
machines and transporters in a job shop environment: A mathematical formulation
and an Anarchic Society Optimization algorithm. Computers & Operations
Research, 84(3), 73-91.

30. Nouri, H., Driss, B., Ghédira, K. (2016). Simultaneous scheduling of machines and
transport robots in flexible job shop environment using hybrid metaheuristics based
on clustered holonic multiagent model. Computers & Industrial Engineering, 102,
488-501.

32. Kaplanoğlu, V. (2016). An object-oriented approach for multi-objective flexible job

shop scheduling problem. Expert Systems with Applications, 45, 71-84.
 Start

Combine population (Select N

Generate initial population (P) of size
individuals, Rank population)
(N)

Evaluate objective function with Non-domination sorting oncombined

processing time and due date population

Calculate Pareto front and crowding

distance
Selection

Selection: Based on its rank and

Select parent solutions for mating based
crowding distance
on tournament selection

NO Stopping
Apply crossover and mutation for
criteria
selected chromosomes (offspring)
YES

Report final population and

Evaluate objective function for new stop
offspring solutions (O)

Fig. 1 Flow Chart of conventional NSGA-II

 Fig. 2 GANTT chart for scheduling of jobs after Iteration 1

120 Makespan vs workload on machines: MK02

H-TLBO
100 MOPSO
NSGA II
Workload on machines

80

60

40

20

0
0 20 40 60 80 100 120 140
Makespan
 Fig. 3a. Pareto optimal curves of Instance 2 for three algorithms

Makespan vs workload on machines: MK04

700
H-TLBO
600
MOPSO
Workload on machines

NSGA II
500

400

300

200

100

0
100 120 140 160 180 200 220 240
Makespan

Fig. 3b. Pareto optimal curves of Instance 4 for three algorithms

600 Makespan vs workload on machines: MK06

H-TLBO
500 MOPSO
Workload on machines

NSGA II
400

300

200

100

0
0 20 40 60Makespan80 100 120 140
 Fig. 3c. Pareto optimal curves of Instance 6 for three algorithms

Makespan vs workload on machines: MK08

1000 H-TLBO
900 MOPSO
NSGA II
Workload on machines

800
700
600
500
400
300
200
100
0
450 470 490 510 530 550 570 590 610 630
Makespan

Fig. 3d. Pareto optimal curves of Instance 8 for three algorithms

Makespan vs workload on machines: MK02

350

300 H-TLBO
Workload on machines

MOPSO
250 NSGA II
200

150

100

50

0
0 20 40 60 80 100 120 140
Makespan

Fig. 4a. Pareto optimal curves of Instance 2 for three algorithms with consideration of
machines break down
 1200 Makespan vs workload on machines: MK04
H-TLBO
1000
MOPSO
Workload on machines

NSGA II
800

600

400

200

0
50 70 90 110 130 150 170 190 210
Makespan

Fig. 4b. Pareto optimal curves of Instance 4 for three algorithms with consideration of
machines break down

1200 Makespan vs workload on machines: MK06

1000 H-TLBO
MOPSO
Workload on machines

800 NSGA II

600

400

200

0
50 70 90 110 130 150 170 190 210
Makespan

Fig. 4c. Pareto optimal curves of Instance 6 for three algorithms with consideration of
machines break down
 2000
Makespan vs workload on machines: MK08

1800 H-TLBO
MOPSO
Workload on machines

1600 NSGA II

1400

1200

1000

800

600
540 560 580 600 620 640 660 680
Makespan

Fig. 4d. Pareto optimal curves of Instance 8 for three algorithms with consideration of
machines break down

Table 1. Input data of the FJSSP of 10 by 10 instance (Brandimarte, 1993).

M/C M/C M/C M/C M/C M/C M/C M/C M/C M/C
Jobs/Machines
1 2 3 4 5 6 7 8 9 10

Oper1 1 4 6 9 3 5 2 8 9 5

Oper 2 4 1 1 3 4 8 10 4 11 4
J1
Oper 3 3 2 5 1 5 6 9 5 10 3

Oper 1 2 10 4 5 9 8 4 15 8 4

Oper 2 4 8 7 1 9 6 1 10 7 1
J2
Oper 3 6 11 2 7 5 3 5 14 9 2

Oper 1 8 8 8 9 4 3 5 3 8 1

Oper 2 9 3 6 1 2 6 4 1 7 2
J3
Oper 3 7 1 8 5 4 9 1 2 3 4

Oper 1 5 10 6 4 9 5 1 7 1 6

Oper 2 4 2 3 8 7 4 6 9 8 4
J4
Oper 3 7 3 12 1 6 5 8 3 5 2

Oper 1 7 10 4 5 6 3 5 15 2 6

Oper 2 5 6 3 9 8 2 8 6 1 7
J5
Oper 3 6 1 4 1 10 4 3 11 13 9

Oper 1 8 9 10 8 4 2 7 8 3 10
 J6 Oper 2 7 3 12 5 4 3 6 9 2 15

Oper 3 4 7 3 6 3 4 1 5 1 11

Oper 1 1 7 8 3 4 9 4 13 10 7

Oper 2 3 8 1 2 3 6 11 2 13 3
J7
Oper 3 5 4 2 1 2 1 8 14 5 7

Oper 1 5 7 11 3 2 9 8 5 12 8

Oper 2 8 3 10 7 5 13 4 6 8 4
J8
Oper 3 6 2 13 5 4 3 5 7 9 5

Oper 1 3 9 1 3 8 1 6 7 5 4

Oper 2 4 6 2 5 7 3 1 9 6 7
J9
Oper 3 8 5 4 8 6 1 2 3 10 12

Oper 1 4 3 1 6 7 1 2 6 20 6

Oper 2 3 1 8 1 9 4 1 4 17 15
J10
Oper 3 9 2 4 2 3 5 2 4 10 23

Table2: Initialization of the data variables

M/c M/c M/c M/c M/c M/c M/c

1 M/c 2 3 4 5 6 7 M/c 8 M/c 9 10 f(x1)
J1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38 49.47
J2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80 70.65
J3 3.30 9.59 5.63 4.39 6.69 7.77 6.66 9.38 8.19 2.06 114.77
J4 4.93 8.80 4.54 3.82 5.45 4.77 4.62 12.68 7.64 1.77 174.87
J5 5.95 9.73 2.21 2.15 4.58 5.47 2.81 13.55 8.90 1.88 234.01
J6 7.88 10.27 7.74 8.19 4.57 5.90 4.63 7.54 7.85 1.45 290.82
J7 8.71 2.49 7.42 2.23 2.07 7.16 3.39 2.21 4.50 3.43 358.81
J8 8.12 7.31 6.71 1.78 6.13 5.70 2.87 5.84 6.48 5.33 404.33
J9 5.77 3.66 6.81 7.29 6.18 6.97 3.16 2.55 3.62 5.22 460.46
J10 4.56 6.03 4.34 1.47 7.34 4.72 5.25 5.95 3.33 5.20 511.69
Mean 5.46 6.79 5.27 4.41 5.17 6.05 4.44 7.48 7.04 3.35
f(x2) 4.50 8.72 3.35 6.88 2.08 0.89 2.34 14.22 5.54 2.10

Table 3: Population group in a teaching phase for minimization of makespan

Teacher 1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38 49.47
Teacher 2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80 70.65
Teacher 3 3.30 9.59 5.63 4.39 6.69 7.77 6.66 9.38 8.19 2.06 114.77
 Teacher 4 5.95 9.73 2.21 2.15 4.58 5.47 2.81 13.55 8.90 1.88 234.01
Teacher 5 8.12 7.31 6.71 1.78 6.13 5.70 2.87 5.84 6.48 5.33 404.33

Table 4: Population group in teaching phase for Minimization of m/c workload

Teacher 1 Teacher 2 Teacher 3 Teacher 4 Teacher 5
5.93 4.14 4.75 9.57 1.92
6.07 4.52 2.58 10.35 8.10
7.77 6.69 5.63 8.19 9.59
4.77 5.45 4.54 7.64 8.80
5.47 4.58 2.21 8.90 9.73
5.90 4.57 7.74 7.85 10.27
7.16 2.07 7.42 4.50 2.49
5.70 6.13 6.71 6.48 7.31
6.97 6.18 6.81 3.62 3.66
4.72 7.34 4.34 3.33 6.03
0.89 2.08 3.35 5.54 8.72
 Table 5: Assignment of groups for Minimization of makespan
J1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38
Group 1
Mean 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38
J2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80
Group 2
Mean 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80
J3 3.30 9.59 5.63 4.39 6.69 7.77 6.66 9.38 8.19 2.06
Group 3 J4 4.93 8.80 4.54 3.82 5.45 4.77 4.62 12.68 7.64 1.77
Mean 4.11 9.20 5.08 4.10 6.07 6.27 5.64 11.03 7.91 1.92
J5 5.95 9.73 2.21 2.15 4.58 5.47 2.81 13.55 8.90 1.88
J6 7.88 10.27 7.74 8.19 4.57 5.90 4.63 7.54 7.85 1.45
Group 4
J7 8.71 2.49 7.42 2.23 2.07 7.16 3.39 2.21 4.50 3.43
Mean 7.51 7.49 5.79 4.19 3.74 6.18 3.61 7.77 7.08 2.25
J8 8.12 7.31 6.71 1.78 6.13 5.70 2.87 5.84 6.48 5.33
J9 5.77 3.66 6.81 7.29 6.18 6.97 3.16 2.55 3.62 5.22
Group 5
J10 4.56 6.03 4.34 1.47 7.34 4.72 5.25 5.95 3.33 5.20
Mean 6.15 5.67 5.95 3.51 6.55 5.79 3.76 4.78 4.48 5.25

Table 6: Assignment of groups for Minimization of m/c workload

Group 1 Group 2 Group 3 Group 4 Group 5
Mean Mean Mean Mean Mean
5.93 5.93 4.14 3.38 3.57 3.70 4.75 2.71 3.73 9.57 8.91 9.24 1.92 4.59 3.26
6.07 6.07 4.52 3.80 7.47 5.26 2.58 2.67 2.62 10.35 3.91 7.13 8.10 10.51 9.31
7.77 7.77 6.69 2.06 6.66 5.14 5.63 3.30 4.46 8.19 4.39 6.29 9.59 9.38 9.48
4.77 4.77 5.45 1.77 4.62 3.95 4.54 4.93 4.73 7.64 3.82 5.73 8.80 12.68 10.74
5.47 5.47 4.58 1.88 2.81 3.09 2.21 5.95 4.08 8.90 2.15 5.53 9.73 13.55 11.64
5.90 5.90 4.57 1.45 4.63 3.55 7.74 7.88 7.81 7.85 8.19 8.02 10.27 7.54 8.90
7.16 7.16 2.07 3.43 3.39 2.96 7.42 8.71 8.06 4.50 2.23 3.36 2.49 2.21 2.35
5.70 5.70 6.13 5.33 2.87 4.78 6.71 8.12 7.41 6.48 1.78 4.13 7.31 5.84 6.57
6.97 6.97 6.18 5.22 3.16 4.85 6.81 5.77 6.29 3.62 7.29 5.46 3.66 2.55 3.11
4.72 4.72 7.34 5.20 5.25 5.93 4.34 4.56 4.45 3.33 1.47 2.40 6.03 5.95 5.99
 Table7: Distance Matrix in a teaching phase for minimization of makespan

Group M.D

1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38

2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80

3 -1.76 -3.66 -1.08 -1.06 -1.14 -0.79 -0.80 -6.39 -1.91 -0.29

4 -0.47 0.25 -1.02 -1.87 -0.05 -0.83 -1.13 2.80 0.05 -0.39

5 0.65 0.76 -0.05 -0.73 -0.65 -0.40 -0.53 0.08 0.87 -0.34

Table8: Distance Matrix in a teaching phase for minimization of m/c workload

M.D
1 2 3 4 5
5.93 -0.01 -0.17 -1.79 -3.01
6.07 -1.15 -0.66 0.31 -6.39
7.77 0.37 -0.27 0.21 -4.59
4.77 0.87 -0.56 0.51 -1.81
5.47 0.19 -0.14 1.65 -4.75
5.90 0.02 -0.16 -1.85 -3.40
7.16 -1.21 -0.68 0.14 -0.67
5.70 0.25 -2.45 1.18 -0.02
6.97 0.35 -0.61 -3.02 -0.39
4.72 0.27 -0.89 0.05 -0.69
 Table9: Xnew values generated after learner phase for make span
X-New Final
M/c 1 M/c 2 M/c 3 M/c 4 M/c 5 M/c 6 M/c 7 M/c 8 M/c 9 M/c 10 f(x)
J1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38 49.47
J2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80 70.65
J3 1.53 5.93 4.55 3.32 5.54 6.98 5.87 2.99 6.28 1.77 111.21
J4 1.63 5.70 3.63 3.32 5.47 6.91 5.75 5.03 5.85 1.63 155.85
J5 5.49 9.98 1.19 0.28 4.53 4.64 1.68 16.35 8.95 1.49 200.63
J6 5.63 10.50 6.23 2.55 4.52 4.66 2.22 11.60 8.87 1.43 255.15
J7 7.19 4.85 6.24 0.30 2.37 4.68 1.78 14.11 5.75 2.29 314.21
J8 8.77 8.07 6.66 1.05 5.47 5.29 2.34 5.91 7.34 4.99 366.48
J9 8.41 5.37 6.68 3.38 5.51 5.91 2.43 4.89 5.87 4.94 422.32
J10 7.42 7.76 6.45 0.82 5.61 5.15 4.58 5.96 5.95 4.87 475.65

Table10: Xnew values generated after learner phase for machine load

X-New Final
M/c 1 M/c 2 M/c 3 M/c 4 M/c 5 M/c 6 M/c 7 M/c 8 M/c 9 M/c 10
J1 2.72 1.92 4.57 7.43 4.14 5.93 3.91 4.59 9.57 3.38
J2 1.94 8.10 1.92 7.77 4.52 6.07 5.94 10.51 10.35 3.80
J3 3.56 9.59 5.37 8.21 6.69 7.77 7.04 9.38 8.19 2.06
J4 4.24 8.80 3.98 5.89 5.45 4.77 6.03 12.68 7.64 1.77
J5 5.80 9.73 2.08 4.47 4.58 5.47 4.45 13.55 8.90 1.88
J6 7.65 10.27 7.58 6.14 4.57 5.90 4.65 7.54 7.85 1.45
J7 7.36 2.49 6.74 3.37 2.07 7.16 2.06 2.21 4.50 3.43
J8 4.40 7.31 4.26 3.99 6.13 5.70 5.99 5.84 6.48 5.33
J9 5.67 3.66 6.20 1.51 6.18 6.97 6.40 2.55 3.62 5.22
J10 3.55 6.03 3.45 3.30 7.34 4.72 6.65 5.95 3.33 5.20
Mean 4.69 6.79 4.61 5.21 5.17 6.05 5.31 7.48 7.04 3.35
f(x2) 3.24 8.72 3.19 4.45 2.08 0.89 2.10 14.22 5.54 2.10
 Table11: Optimal make span values after Iteration 1

X-New Final
M/c 1 M/c 2 M/c 3 M/c 4 M/c 5 M/c 6 M/c 7 M/c 8 M/c 9 M/c 10 f(x)
J1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38 49.47
J2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80 70.65
J3 1.50 5.08 6.23 3.27 6.50 1.09 1.25 7.20 2.58 2.03 111.48
J4 3.19 8.99 3.99 3.96 6.52 2.69 6.24 2.52 2.94 1.44 147.60
J5 3.38 9.33 1.64 1.58 1.13 2.65 2.00 17.99 4.01 1.85 190.47
J6 5.79 15.38 6.02 3.47 6.66 4.78 2.80 10.18 5.31 1.74 235.93
J7 4.66 6.75 2.50 1.82 3.00 5.42 2.76 8.89 6.06 1.81 298.12
J8 8.47 9.87 2.10 1.12 5.64 5.34 2.42 4.90 4.74 5.18 345.17
J9 7.21 6.47 6.83 4.36 5.46 5.81 1.23 1.78 6.34 4.70 394.49
J10 7.16 2.15 6.57 1.45 6.04 5.69 1.62 3.97 3.48 5.10 445.08

Table12: Optimal machine load values after Iteration 1

X-New Final
M/c 1 M/c 2 M/c 3 M/c 4 M/c 5 M/c 6 M/c 7 M/c 8 M/c 9 M/c 10
J1 2.72 1.92 4.57 7.43 4.14 5.93 3.91 4.59 9.57 3.38
J2 1.94 8.10 1.92 7.77 4.52 6.07 5.94 10.51 10.35 3.80
J3 3.56 9.59 5.37 8.21 6.69 7.77 7.04 9.38 8.19 2.06
J4 4.24 8.80 3.98 5.89 5.45 4.77 6.03 12.68 7.64 1.77
J5 5.80 9.73 2.08 4.47 4.58 5.47 4.45 13.55 8.90 1.88
J6 7.65 10.27 7.58 6.14 4.57 5.90 4.65 7.54 7.85 1.45
J7 7.36 2.49 6.74 3.37 2.07 7.16 2.06 2.21 4.50 3.43
J8 4.40 7.31 4.26 3.99 6.13 5.70 5.99 5.84 6.48 5.33
J9 5.67 3.66 6.20 1.51 6.18 6.97 6.40 2.55 3.62 5.22
J10 3.55 6.03 3.45 3.30 7.34 4.72 6.65 5.95 3.33 5.20
Mean 4.69 6.79 4.61 5.21 5.17 6.05 5.31 7.48 7.04 3.35
f(x2) 3.24 8.72 3.19 4.45 2.08 0.89 2.10 14.22 5.54 2.10
Table13: Archived population for the Iteration 1 values of make span from the m/c workload
values

X-New Final of Make span

M/c M/c M/c M/c M/c M/c M/c
M/c 2 M/c 8 M/c 9 f(x)
1 3 4 5 6 7 10
J1 2.71 1.92 4.75 8.91 4.14 5.93 3.57 4.59 9.57 3.38 49.47
J2 2.67 8.10 2.58 3.91 4.52 6.07 7.47 10.51 10.35 3.80 70.65
J3 1.50 5.08 6.23 3.27 6.50 1.09 1.25 7.20 2.58 2.03 111.48
J4 3.19 8.99 3.99 3.96 6.52 2.69 6.24 2.52 2.94 1.44 147.60
J5 3.38 9.33 1.64 1.58 1.13 2.65 2.00 17.99 4.01 1.85 190.47
J6 5.79 15.38 6.02 3.47 6.66 4.78 2.80 10.18 5.31 1.74 235.93
J7 4.66 6.75 2.50 1.82 3.00 5.42 2.76 8.89 6.06 1.81 298.12
J8 8.47 9.87 2.10 1.12 5.64 5.34 2.42 4.90 4.74 5.18 345.17
J9 7.21 6.47 6.83 4.36 5.46 5.81 1.23 1.78 6.34 4.70 394.49
J10 7.16 2.15 6.57 1.45 6.04 5.69 1.62 3.97 3.48 5.10 445.08
Mean 4.67 7.40 4.32 3.39 4.96 4.55 3.13 7.25 5.54 3.10
f(x2) 5.01 14.08 3.66 4.67 2.89 2.76 4.01 21.28 6.31 2.05

Table14: Archived population for the Iteration 1 values of make work load from the m/c
makespan values

X-New Final Of Workload

M/c M/c M/c M/c M/c M/c M/c M/c
M/c 2 M/c 8 f(x)
1 3 4 5 6 7 9 10
J1 2.72 1.92 4.57 7.43 4.14 5.93 3.91 4.56 2.64 3.38 41.20
J2 1.94 8.10 1.92 7.77 4.52 6.07 5.94 12.38 8.93 3.80 71.00
J3 3.56 9.59 5.37 8.21 6.69 7.77 7.04 9.33 0.74 2.06 107.72
J4 4.24 8.80 3.98 5.89 5.45 4.77 6.03 4.55 2.91 1.77 164.69
J5 5.80 9.73 2.08 4.47 4.58 5.47 4.45 13.43 5.09 1.88 213.20
J6 7.65 10.27 7.58 6.14 4.57 5.90 4.65 5.32 2.17 1.45 269.73
J7 7.36 2.49 6.74 3.37 2.07 7.16 2.06 2.25 1.49 3.43 327.40
J8 4.40 7.31 4.26 3.99 6.13 5.70 5.99 4.08 3.06 5.33 367.72
J9 5.67 3.66 6.20 1.51 6.18 6.97 6.40 2.52 0.72 5.22 417.85
J10 3.55 6.03 3.45 3.30 7.34 4.72 6.65 5.99 3.31 5.20 462.89
Mean 4.69 6.79 4.61 5.21 5.17 6.05 5.31 6.44 3.11 3.35
f(x2) 3.24 8.72 3.19 4.45 2.08 0.89 2.10 13.96 5.29 2.10
 Table 15. Input parameters values of the proposed algorithms

Input parameters of the considered Parameters value

algorithms
CNSGA-II
Population size 200
Total number of Iterations 150 to 300
Probability of cross over (Pc) 0.65 to 0.9
Probability of mutation (Pm) 0.01 to 0.1
MOPSO
Cognitive factor (c1) 0.5 to 2
Social factor (c2) 0.5 to 2
Swarm size (N)or Population Size 200
Number of iterations (K)or Total 150 to 300
number of generations
MOTLBO
Population size 200
Total number of generations 150 to 300
Teacher factor 1 or 2
Probability of cross over (Pc) 0.65 to 0.9
Probability of mutation (Pm) 0.001 to 0.1

Table. 16 Experimental conditions of different attributes with their values

Type Attributes Values

Shop Floor Size 6-15

Machine breakdown level 0.025
MTTR 5
MTBF, MTTR Exponential Distribution
Key machine Maximum 1 with random
Performance measures Minimization of Makespan

Minimization of work load on

most loaded machine
 Table 17. Makespan and total machine load variation of different without machine break
down scenarios of different algorithms

Brandimarte benchmark Instances Makespan Total Machine load variation

Scenarios
number number Proposed MOPSO NSGA-II Proposed MOPSO NSGA-II
of jobs of MOTLBO algorithm
machines LB algorithm

MK01 10 6 36 40 40 40 44 49 52
MK02 10 6 24 24 28 29 52 67 74
MK03 15 8 204 287 300 321 432 474 488

MK04 15 8 48 97 106 112 200 225 246

MK05 15 4 168 186 203 205 413 448 471

MK06 10 15 33 58 65 72 176 187 192

MK07 20 5 133 140 152 167 298 324 345
MK08 20 10 523 530 545 556 786 792 812
MK09 20 10 299 315 340 358 564 574 589

MK10 20 15 165 212 228 254 310 345 356

 Table 18. Makespan and total machine load variation of different instances with machine
break down scenarios of different algorithms.

Scenarios number number Proposed MOPSO NSGA-II Proposed MOPSO NSGA-II

of jobs of LB MOTLBO algorithm
machines algorithm

MK01 10 6 36 56 62 69
144 154 175

MK02 10 6 24 39 48 52
85 98 146

MK03 15 8 204 332 374 397

651 724 906

MK04 15 8 48 123 136 148

322 386 554

MK05 15 4 168 227 265 271

964 1211 1456

MK06 10 15 33 83 94 112
310 417 621

MK07 20 5 133 172 246 263

677 748 873

MK08 20 10 523 574 623 631

1024 1347 1546

MK09 20 10 299 384 392 427

816 920 1144

MK10 20 15 165 242 275 280

624 754 837
Table 19. Comparison of results with machines break down and without machines break
down with different algorithms

Without machines break down With machines break down

Brandimarte benchmark
Instances Total Machine Total Machine
Makespan Makespan
load variation load variation

num Prop Prop Prop

nu
ber osed Prop osed osed
Scen mb NS NS NS NS
of MOT MO osed MO MOT MO MOT MO
arios er L GA GA GA GA
mac LBO PSO algor PSO LBO PSO LBO PSO
of B -II -II -II -II
hine algor ithm algor algor
jobs
s ithm ithm ithm

MK0 2 14
10 6 24 28 29 52 67 74 39 48 52 85 98
2 4 6

MK0 4 11 24 14 55
15 8 97 106 200 225 123 136 322 386
4 8 2 6 8 4

MK0 3 19 11 62
10 15 58 65 72 176 187 83 94 310 417
6 3 2 2 1

15
5
MK0 55 81 63 134 46
20 10 2 530 545 786 792 574 623 1024
8 6 2 1 7
3