You are on page 1of 19

Int. J. Operational Research, Vol. 38, No.

2, 2020 147

A review of job shop scheduling problems in


multi-factories

Imen Chaouch*
SOIE-COSMOS Laboratory,
and
Ecole Nationale des Sciences de l’Informatique,
Université de la Manouba, Tunisia
Email: imen.chaouch@ensi.rnu.tn
*Corresponding author

Olfa Belkahla Driss


SOIE-COSMOS Laboratory,
and
Ecole Supérieure de Commerce de Tunis,
Université de la Manouba,
Tunis, Tunisia
Email: olfa.belkahla@isg.rnu.tn

Khaled Ghedira
SOIE-COSMOS Laboratory,
Université de la Manouba, Tunisia
and
Institut Supérieur de Gestion de Tunis,
Université de Tunis, Tunisia
Email: khaled.ghedira@isg.rnu.tn

Abstract: The distributed job shop scheduling problem (DJSP) deals with the
assignment of jobs to factories geographically distributed and with determining
a good operation schedule of each factory. It is one of the well-known NP-hard
combinatorial optimisation problem to solve optimally. In the last two decades,
the problem has captured the interest of a number of researchers and therefore
various methods have been employed to study this problem. In this paper, we
first present an overview of pioneer studies conducted on solving distributed
job shop scheduling problems and a classification of the employed techniques
is given. Then, depth analysis of the outcome of existing literature is presented.

Keywords: distributed scheduling; job shop; flexible job shop; optimisation


method; survey.

Reference to this paper should be made as follows: Chaouch, I., Driss, O.B.
and Ghedira, K. (2020) ‘A review of job shop scheduling problems in
multi-factories’, Int. J. Operational Research, Vol. 38, No. 2, pp.147–165.

Copyright © 2020 Inderscience Enterprises Ltd.


148 I. Chaouch et al.

Biographical notes: Imen Chaouch received her Engineer degree in Industrial


Information Technology (IT) and Control System Engineering from the Institut
National des Sciences Appliquées et de Technologie (INSAT), University of
Carthage, Tunisia in 2014. She is currently a PhD candidate and a member of
SOIE-COSMOS Laboratory at the Ecole Nationale des Sciences de
l’Informatique, University of Manouba, Tunisia. Her research interests include
distributed manufacturing systems, optimisation, metaheuristics and
multi-agent systems.

Olfa Belkahla Driss received her BSc, MSc and PhD degrees in Computer
Science Applied to Management from the Institut Supérieur de Gestion de
Tunis, University of Tunis, Tunisia, in 1997, 2000 and 2006, respectively. She
is an Assistant Professor with the Department of Computer Science at the Ecole
Supérieure de Commerce de Tunis, University of Manouba, Tunisia from 2003,
where she coordinated the graduate degree program in Computer Science
Applied to Management from 2008 to 2014. She is actually a Supervisor of the
Research Master in Computational Intelligence and Decision Making Applied
to Management from 2010, and a member of SOIE-COSMOS Laboratory at the
Ecole Nationale des Sciences de l’Informatique, University of Manouba,
Tunisia. Her main research interests are in the field of industrial engineering,
transport and production logistics, scheduling, artificial intelligence,
multi-agent systems, constraint satisfaction problems, optimisation,
metaheuristics, hybrid algorithms.

Khaled Ghedira is a Professor at the Institut Supérieur de Gestion de Tunis,


University of Tunis, Tunisia, the Founding President of the Tunisian
Association of Artificial Intelligence (ATIA-Tunisia), the Founder and Director
of both the research unit URIASIS (1999 to 2011) and the SOIE Laboratory
(2011 to 2013). He is member of several international scientific committees and
is often invited as keynote speaker/visiting professor at national and
international level. He is/was also member of the think national committee for
higher education and member/president of several committees: evaluation of
higher education institutions, research projects reviewing, teachers recruiting,
LMD. He is currently a member of SOIE-COSMOS Laboratory at the Ecole
Nationale des Sciences de l’Informatique, University of Manouba, Tunisia. His
research areas include multi-agent systems, CSP, transport and production
logistics, mono and multiobjective optimisation, metaheuristics and security in
M/Egovernment.

1 Introduction

Manufacturing industry is facing multiple challenges: ensuring profitable growth,


improving productivity and reducing costs while responding quickly to customer
demands. To remain competitive and for closer proximity to the market, industrial
companies are increasingly merging to distributed ones and thus, the structure of their
shops changes from simple configurations to distributed ones. The study of production
systems is complex given the large number of integrated entities and their interactions.
As scientists, engineers, and managers, we always have to take decisions. These
decisions are becoming more complex with the evolution of the industrial environment
and market requirement. An effective decision making process consists mainly of four
steps (Talbi, 2009) summarised as follow:
A review of job shop scheduling problems in multi-factories 149

• formulation of the problem: different variables of the problem are identified


• modelisation of the problem: a mathematical model is built (optimisation equations
and constraints)
• finding a solution: solving procedure is executed
• implementation of the solution: the obtained solution is tested.
Following the complexity of the problem, it may be solved by an exact method or an
approximate method. Exact methods (branch and X, constraint programming, dynamic
programming, etc.) can be applied to small instances of difficult problems and guarantee
an optimal solution. In complex problems, utilisation of approximate methods is
inevitable. They provide a good quality solution in a reasonable time but there is no
guarantee of achieving the optimal solution. Simulated annealing, Tabu search, genetic
algorithm (GA) and ant colony optimisation are some examples of heuristics algorithms.
We can find distributed scheduling in different studies dealing with various shop types
such as distributed job shop (Naderi and Azab, 2014; Jia et al., 2007), among many
others, distributed flow shop (Komaki and Malakooti, 2017; Naderi and Ruiz, 2014; Rifai
et al., 2016; Bargaoui et al., 2016, 2017) among many others and distributed parallel
machines scheduling (Behnamian and Ghomi, 2013; Hatami et al., 2015).
The scope of this paper is to provide a consolidated survey of various techniques that
have been employed for resolution of the distributed job shop scheduling problem (DJSP)
and distributed and flexible job shop scheduling problem (DFJSP) since their appearance.
Numerous approaches have thus been investigated and these techniques are classified for
ease of analysis. The rest of the paper is organised as follows: The DJSP is defined in
Section 2. Related literature in the fields of DJSP and DFJSP are outlined in
Sections 3 and 4 respectively. Section 5 covers a detailed analysis of the literature.
Section 6 concludes the paper and gives directions for future research.

2 Problem description

Scheduling has been defined as “the art of assigning resources to tasks in order to insure
the termination of these tasks in a reasonable amount of time” (Dempster, 2012).
The job shop scheduling problem (JSP) is one of the most important and complex
problems in machine scheduling. A typical job shop can be stated as a set of n jobs,
which have to be processed on a set of m machines. Every job consists of a series of
operations in a predetermined order on machines. There are various constraints on both
jobs and machines. Each operation needs to be processed during an uninterrupted time of
a fixed processing period and a given machine. A job can be processed by at most one
machine at a time and a machine can process at most one job at a time. Furthermore,
there are no precedence constraints among the operations of different jobs. In addition, it
is assumed that a job does not visit the same machine twice. The aim is to determine the
operation sequences on the machines in order to optimise specified criteria such as:
• minimise the makespan
• minimise total tardiness
150 I. Chaouch et al.

• minimise the number of late jobs


• maximise production rate.
The most widely adopted criteria among literature is minimising the makespan which can
be set as the total time period it takes to complete all operations for all jobs.
To examine the performance of the most promising metaheuristics and compare their
efficiency, benchmark problems are used. For the JSP, several researchers propose sets of
benchmark problems including a number of instances with different dimensions and
configurations. An instance consists of the number of factories (f), the number of jobs (n),
the number of machines (m), and the processing time (pj,i) of each operation j of job i.
The most popular benchmark problems are summarised in Table 1.
Table 1 Benchmark problems of JSP

Problem Processing time Problem size Proposed by


ABZ [50, 100] ABZ 5 10 × 10 Adams et al. (1988)
[25, 100] ABZ 6 10 × 10
[11, 40] ABZ 7 to ABZ 9 20 × 15
FT [1, 10] FT 06 6×6 Fisher and
FT 10 10 × 10 Thompson (1963)
FT 20 20 × 5
SWV [1, 100] SWV 1 to SWV 5 20 × 10 Storer et al. (1993)
SWV 6 to SWV 10 20 × 15
SWV 11 to SWV 15 50 × 10 (hard)
SWV 16 to SWV 20 50 × 10 (easy)
TAI [1, 99] TAI 01 to TAI 10 15 × 15 Taillard (1993)
TAI 11 to TAI 20 20 × 15
TAI 21 to TAI 30 30 × 15
TAI 31 to TAI 40 30 × 20
TAI 41 to TAI 50 50 × 15
TAI 51 to TAI 60 50 × 20
TAI 61 to TAI 70 100 × 20
LA [5, 99] LA 01 to LA 05 10 × 5 Lawrence (1984)
LA 06 to LA 10 15 × 5
LA 11 to LA 15 20 × 5
LA 16 to LA 20 10 × 10
LA 21 to LA 25 15 × 10
LA 26 to LA 30 20 × 10
LA 31 to LA 35 30 × 10
LA 36 to LA 40 15 × 15

In a highly competitive economy, it is the quality of the service, that often makes the
difference between one company to another. Thus, companies are on a continuous quest
in order to respond to the requirements of a market in constant evolution and ensure full
satisfaction to their customers. Job shops are constantly facing more challenges and an
A review of job shop scheduling problems in multi-factories 151

increased need to reduce on both their costs and time-to-market (Naderi and Ruiz, 2010).
In this trend, the phenomenon of decentralisation appeared. In this context, DJSP can be
considered as an extension of the simple JSP. It can be treated as a set of f factories,
which are geographically distributed in different areas. Each factory contains m machines
on which a certain number of jobs have to be processed, cf. Figure 1. Distributed
scheduling problems in multi-factory production are much more complicated than
classical scheduling problems (Chung et al., 2009a) since two decisions have to be taken:
• allocation of jobs to suitable factories
• sequencing the operations on machines so that yield a feasible schedule aiming to
minimise a predefined performance criterion.
Therefore, the DJSP is more difficult than the classical JSP due to the consideration of
both factories selection and job scheduling. The JSP is strongly NP-hard (Garey et al.,
1976) while the DJSP is much more complicated. Hence, it can be concluded that DJSP
is also NP-hard. In distributed scheduling, makespan minimisation becomes the
minimisation of maximum makespan among all factories.

Figure 1 An outline of a typical distributed scheduling problem

3 Distributed job shop scheduling: literature review

Scheduling problems have become a popular issue for researchers and industrialists in the
last three decades, particularly the JSP since it is one of the most difficult tasks. Colorni
152 I. Chaouch et al.

et al. (1994), Dell’Amico and Trubian (1993), Adams et al. (1988) and Davis (1985) are
pioneer researches in the literature that dealt with the JSP.
Due to the trend of globalisation, the JSP has evolved from the classical JSP to the
Distributed one and becomes increasingly, one of the most important issues to rise.
Chaouch et al. (2017c) surveyed the literature focusing only on the DJS and presented
main papers dealing with the problem. In this paper, full study is presented aiming to
summarise the distributed job shop scheduling in both aspects: simple and flexible.
GA is the first technique used to solve the DJSP. It is a computational method mainly
proposed by Holland (1992), based on the mechanics of biological genetic evolution.
This method is commonly used to generate useful solutions to optimisation, using
technology inspired by the theory of evolution and biological processes, such as
mutation, selection, and crossover operators. The evolution of chromosomes is an
iterative process, usually starts from a population of randomly generated individuals.
These chromosomes will evolve until the predetermined stopping condition is reached.

Figure 2 Procedure and iteration of the proposed GA

Source: Jia et al. (2002)


The DJSP was studied for the first time by Jia et al. (2002). The authors proposed a web
based system to enable production scheduling with the utilisation of the World Wide
Web technology, in order to facilitate collaboration between geographically distributed
plants. A GA approach was adopted, involving a once gene crossover and twice gene
mutation. To deal with the distributed scheduling problems, the genes in the GA must
comprise the two dominant factors, i.e., the selected factory for every job and the
operation processing sequence. Figure 2 illustrate the different steps of the proposed GA.
The proposed system allows to the manufacturing system to choose to use the scheduling
application locally, construct scheduling agents, or go to the web to make online
scheduling, according to every factory’s actual manufacturing situation and physical
constraints.
In their next paper, to solve the DJSP, Jia et al. (2003) presented a modified genetic
algorithm (MGA) in which two-step encoding method was used. The first one to encode
the factory candidates and the second one, to affect jobs and operations. To test the
performance of their MGA, the authors used benchmark instances (ten jobs/ten machines
and 20 jobs/five machines) proposed by Muth and Thompson (1963). In fact, MT1963
are instances for simple JS which are adapted to the DJS by distributing different jobs
into the factories.
Later, in Jia et al. (2007) refined their previous approach and proposed a GA
integrated with Gantt chart (GC) to derive the factory combination and schedule. The
A review of job shop scheduling problems in multi-factories 153

experimental results showed that the application of the GC is able to facilitate the
chromosome evaluation procedures and thus improve the computational performance
algorithm. In addition, the CPU time used for the GA-GC approach to solve the problem
is 10 s. This computational time is 50% shorter than the time consumed by using the GA
alone (15 s, without GC integration) developed by Jia et al. (2003) to solve the same
problem under a similar environment. In these three papers Jia et al. (2002, 2003, 2007)
proposed an encoding of chromosome, crossover mechanism and two mutation
mechanisms for the DJS. The proposed approach was applied to the classical
single-factory firstly, then to the DJSP. Numerical experiments were achieved and
promising results were obtained.

Figure 3 The general outline of GH1

Source: Naderi and Azab (2014)


Naderi and Azab (2014) have mathematically formulated the DJSP by two different
mixed integer linear programming models (MILP). The first model dealt with the
problem as a sequencing decision while the second one dealt it as a positional one. Since
the problem is NP-hard to solve, utilisation of heuristics was inevitable. Hence, three
well-known heuristics were first adapted to the problem; these are shortest processing
time first (SPT), longest processing time first (LPT) and longest remaining processing
time (LRPT). In addition, three greedy heuristics have been deployed (GH1, GH2 and
GH3). The algorithms are greedy since at each step, different alternatives are generated
and the best one is selected. In GH1, after assigning jobs to factories, the schedule is
built. Operations of each factory are iteratively inserted, one at a time, into a sequence to
have at the final a complete permutation of operations. Decisions of assignment of job to
factory and sequencing are sequentially taken (i.e., no interaction between the two
decisions). Figure 3 shows the general outline of GH1. Unlike GH1, job factory
assignment and sequencing are interactively determined in GH2 and the jobs are initially
sorted. Like GH2, GH3 interactively takes two decisions of job factory assignment and
sequencing. The main difference is the job factory selection rule. In GH2, each job is
154 I. Chaouch et al.

assigned to the factory with the lowest makespan. While in GH3, the job is assigned to
the factory resulting in the lowest makespan after sequencing the operations of the job.
That is, the assignments of jobs to all factories are tested and the best one is selected. The
performance of the two proposed mathematical models and six heuristics (SPT, LPT,
LRPT, GH1, GH2 and GH3) are evaluated and tested. Two sets of instances were used.
The first one, instances are generated by random processing times taken from a uniform
distribution between 1 and 99. The second set, the instances of Taillard benchmark for
job shops (Taillard, 1993) are being used. With regards to the obtained optimal solutions,
it is concluded that GH3 performs best with an average percentage deviation index (PDI)
of 1.21%. The second best heuristic is GH1 with an average PDI of 37.83%. Note that the
PDI can be calculated as follows:
Alg − Min
PDI = × 100
Max − Min
where
Alg makespan obtained by any of the algorithm
Min lowest makespan obtained for a given instance
Max highest makespan obtained for a given instance.
Finally, the redundancy mitigation mechanisms were applied and tested. In fact, the
proposed algorithms suffer from a serious shortcoming, which is redundancy. That is,
different permutations might represent the same schedule. For that reason, authors
developed the redundancy mitigation mechanism which allow recognisation and discard
of redundant permutations. The results showed that more than 80% reduction in
computational time with simple application of redundancy exclusion theorems.
Recently, Naderi and Azab (2015) have differently treated the same problem. First,
authors have mathematically formulated the DJSP by a MILP. And then, three different
versions of simulated annealing have been designed and implemented. The first one,
called SA, is implemented without any local search. The second one, called hybridised
simulated annealing (HSA), is hybridised with local search type 1 which assume that the
job number is first inserted into m random positions. Then, the position of one randomly
selected job number is shifted to a random position. Finally, the third version, called
greedy simulated annealing (GSA), employs the greedy local search two. Unlike the local
search type 1, the job number is added into permutation one by one. To insert the job
number, it is assigned to all the possible positions and the best position is chosen. The
performance of the proposed mathematical model and algorithms are evaluated and
tested. To do the experiments, three sets of instances were generated. The first set is for
parameter tuning, the second is for the experiment with small instances, and finally, the
last one is for the experiment with larger instances. For the first set, there are 20 different
combinations generated by random processing times taken from a uniform distribution
between 1 and 99. For the second set, benchmark instances generated by Naderi and
Azab (2014) were used including 24 small instances. And finally, for the third set,
80 instances of Taillard (1993) were tested with different levels of f (f = 2, 3, 4, 5),
summing up 320 large instances. The solutions proposed in this article obtained a
promising results and outperformed the other tested algorithms. Figure 4 shows the
A review of job shop scheduling problems in multi-factories 155

results of different experiments. The performance measure used in this research is


relative percentage deviation (RPD). It can be calculated as follows:
Alg − Min
RPD = × 100
Min
where
Alg makespan obtained by any of the algorithm
Min lowest makespan obtained for a given instance.
Based on the swarm intelligence, Chaouch et al. (2017d) applied a basic ant system (AS)
algorithm for the first time to solve the DJSP with makespan minimisation criterion.
Some initial results for small instances were encouraging and have shown the viability of
the approach. However, for larger instances, the algorithm gives a very poor solution
quality compared to state-of-the-art solutions. This is due to randomised character of the
AS algorithm which makes probabilistic decisions in the construction of the solution. In
fact, without using a local search, the basic AS can easily fall into a local optimum. In
their next paper, Chaouch et al. (2017a) have improved their solution and applied a
bio-inspired algorithm, namely the elitist ant system (EAS).

Figure 4 The average RPD of the tested algorithms (see online version for colours)

Source: Naderi and Azab (2015)


The results have improved considerably and the maximum completion time was
decreased by approximately 50%. In their last paper, Chaouch et al. (2017c) developed a
modified ant colony optimisation algorithm (MACO). The basic principle of the MACO
is to improve the solution generated by the EAS developed in Chaouch et al. (2017a) by
applying a local search procedure in order to explore more search space and find better
result. The constructed solution generated by EAS is used as an initial solution S. Then S
is improved by the proposed method with new neighbourhood strategy. First of all, the
last scheduled job over all machines is determined and selected to make the first
improvement moves. The method consists of finding all inactive time intervals of
156 I. Chaouch et al.

machines and tries to insert the operations of the selected job in those intervals while
respecting DJSP constraints until no improvement is possible. To evaluate the proposed
algorithm, MACO was compared with both the AS developed in Chaouch et al. (2017d)
and EAS in Chaouch et al. (2017a). The results were very promising and show that the
MACO outperforms the EAS and the classic AS.

4 Distributed and flexible job shop scheduling

The flexible job shop problem (FJSP) is an extension of the classical JSP by allowing an
operation, taken from a set of N operations, to be performed by one machine out of a set
of machines (Mastrolilli and Gambardella, 2000). Brandimarte (1993), Kacem et al.
(2002), Xia and Wu (2005) and Nouiri et al. (2015) are among the important studies in
the literature that dealt with the FJSP. Like the DJSP, DFJSP has emerged and captivated
the attention of many researchers. A DFJSP consists of a set of f factories with m
machines and n jobs. Each job involves several operations that can be processed on a set
of machines. Therefore, a DFJS problem involves three decisions: job-to-factory
assignment, operation-to-machine assignment, and operation sequencing decision. In this
context, Chan et al. (2005, 2006a, 2006b) presented a sequence of works that treated and
discussed the problem. In Chan et al. (2005, 2006a) a genetic algorithm with dominant
genes (GADGs) was proposed to deal with flexible manufacturing system (FMS) in
which the chromosome representation (called S_CH) can explicitly model each of the
three DFJS decisions. GADG implemented the idea of adaptive strategy. A new
crossover mechanism, the dominated gene crossover, was presented to enhance the
performance of genetic search by recording and preserving the best genes, and to
eliminate the problem of determining the optimal crossover rate. To prevent saturation of
solution pool, a saturation operator was also introduced to monitor the similarity of the
chromosomes. The objective was to minimise the makespan. Some experiments have
been carried out. For the comparison purpose, five multi-factory models have been solved
by different well known optimisation approaches. Empirical results showed that the
proposed model (GADG) exhibited better performance than the other approaches,
especially when the problem sizes increase.
It is known that the efficient and effective production is based on the consistent
operation of machines. At a production level, considering machine maintenance is a key
factor, which allows having a feasible and most realistic scheduling. Namely,
maintenance is adopted to increase the reliability of the equipment by eliminating
unexpected breakdowns, minimising unscheduled downtimes, and reducing maintenance
costs (Alsyouf, 2007). Problems considering maintenance have attracted many
practitioners and researchers aiming to improve production quality and performance
efficiency (Graves and Lee, 1999; Qi et al., 1999; Al-Najjar, 2000; Riis et al., 1997).
In a distributed environment, Chan et al. (2006b) were generalised their previous
approach (GADG), to deal with distributed FMS scheduling problems subject to machine
maintenance constraints. The approach was applied to an environment with four plants
containing three machines each and aimed to minimise the makespan. The problem was
mathematically formulated and the optimisation performances of the proposed GADG
were compared with other existing approaches. The importance of considering
maintenance in distributed scheduling was demonstrated by two sets of simulation runs.
Set 1 did not consider maintenance, while Set 2 did. The comparison of the results
A review of job shop scheduling problems in multi-factories 157

demonstrated that considering maintenance during distributed scheduling can shorten the
makespan by improving the machine utilisation. The average makespan obtained by Set 2
is shorter. In addition, the machine utilisation is improved as well.
Chung et al. (2009a) proposed a MGA approach to deal with distributed shop models
with maintenance consideration, aiming to minimise the makespan. The modified GADG
approach was developed based on the model proposed by Chan et al. (2006b). It
improved the local searching ability of the original GADG approach by iteratively
solving the jobs allocation problem and scheduling of production with the maintenance
problem. In the second part of the paper, the authors studied the effect of the relation
between the maintenance repairing time and the machine on the performance of
distributed scheduling.
Aiming to keep the system’s reliability in a defined acceptable level, and minimise
the makespan of the jobs, Chung et al. (2009b) proposed a double tier GA approach for
multi-factory production network. The optimisation algorithm simultaneously scheduled
perfect and imperfect maintenance during the process of distributed scheduling (in
perfect maintenance the machine is considered as good as new after maintenance and in
imperfect maintenance the machine may only recover to a less-deteriorated condition
after maintenance). The optimisation reliability of the proposed approach was
demonstrated in three numerical examples, with the same parameters (two factories, three
machines). The first example demonstrated that the proposed approach is able to maintain
the reliability of the machines in a predefined acceptable level, which is the threshold of
the Hazard rate (H) which represents the reliability of the machine. The second example
demonstrated the relationship between the hazard rate and the makespan obtained. The
last example demonstrated the relationship between the age reduction factor of the
machines and the obtained makespan.
Always considering machine maintenance, Yadollahi and Rahmani (2009) have
proposed a memetic algorithm approach to solve the distributed FMS scheduling
problem. The objective of this algorithm was to create a time-cost trade-off solution
(i.e., production time and production costs). The experimental results showed that the
optimisation performance of the proposed algorithm was better than both (Rossi and
Boschi, 2009; Chan et al., 2006b) algorithms. This resulted from the utilisation of the
Tabu search as a local search algorithm and the classification of chromosomes due to
different purposes.
Based on the MGA proposed by Jia et al. (2003) and De Giovanni and Pezzella
(2010) proposed a new optimisation procedure combining GA and local search heuristics
to solve distributed scheduling problems in flexible manufacturing environments. Besides
traditional crossover and mutation operators, the improved genetic algorithm (IGA)
presented in this paper introduced a new decoding mechanism, able to handle alternative
job routings, and a refinement operator, which exploits a local search procedure to refine
the job routing and operations scheduling of most promising individuals. The proposed
approach has been compared with other algorithms (Jia et al., 2003; Chan et al., 2006b,
2006a) and evaluated with satisfactory results on a large set of distributed and flexible
scheduling problems. Thanks to the refinement operator, IGA is able to find the better
solution and to improve the average performance.
Khalid and Yusof (2013) proposed an improved immune algorithm (IIA) approach in
solving flexible manufacturing system in distributed system problem (FMSDS) subject to
machine maintenance aiming to minimise the makespan over all the factories. The
158 I. Chaouch et al.

performance of the IIA has been tested on several instances. Four datasets were
considered from Chan et al. (2006a, 2006b, 2008) and Fisher and Thompson (1963). The
results showed that the IIA required less iteration numbers, therefore, it promoted
solution diversity and faster solution evaluation due to a simplified solution
representation scheme.
Ziaee (2014) investigated the DFJP with the objective of minimising the makespan.
The main purpose was to produce feasible schedules on reasonable time. This approach
was motivated by the idea of developing a constructive heuristic that considers
simultaneously many factors affecting the solution quality and intelligently balances their
effects in the process of schedule generation. This algorithm had a simple structure, easy
implementation, and required a few computational effort, which made it preferable over
other more complex and time-consuming approaches.
Liu et al. (2014) introduced an IGA to solve DFJS and proposed a refined encoding
operator that integrates probability concepts into a real-parameter encoding method. It
can be used to solve either FJS or DFJS problems, even more multi-variable scheduling
problems. The proposed refined GA was evaluated with satisfactory results. First, a
classical DFJS problem was adopted to show the effectiveness of the algorithm, and then,
the algorithm was used to solve a real-world case (100 and 200 sets of work orders of a
middle size fastener manufacturer in Taiwan). The results indicated that the proposed
algorithm could effectively solve the conflicts caused by the encoding problem of GAs.
However, many factors influenced the final results of scheduling such as uncertainties
and rescheduling because of disruptions.
Lu et al. (2015) proposed a GA GA_ JS for solving DFJS. In this paper, the algorithm
was developed by proposing a new and concise chromosome representation, which
modeled a three dimensional scheduling solution by a one-dimensional scheme.
According to experimental results, GA_ JS appeared to outperform IGA, developed by
De Giovanni and Pezzella (2010). Yet, their performance differences become smaller
when they increased the number of cells. In two-cell, three-cell and four-cell
experiments, the difference of percentage gap is 2.3%, 1.1% and 0.2% respectively.
Recently, Chang and Liu (2017) proposed a hybrid genetic algorithm (HGA) for
solving the DFJS scheduling problem. The proposed approach used the Taguchi method
to optimise the parameters of a GA. Furthermore, a novel encoding mechanism was
employed and various crossover and mutation operators were adopted for increasing the
probability of finding the optimal solution and diversity of chromosomes and for refining
a makespan solution. The roulette method was adopted for encoding flexible
manufacturing units (FMUs) and selecting a machine to be incorporated into the initial
chromosomes. Results demonstrated that this approach could produce a satisfactory
scheduling outcome.
Wu et al. (2017) proposed GA with a new chromosome representation scheme to
solve the DFJSP. Several GAs have been developed using different chromosome
representations. The proposed GAOP is compared against three prior GAs: IGA by
De Giovanni and Pezzella (2010), GA_CL by Chang and Liu (2017) and GA_JS by Lu
et al. (2015). 31 DFJS instances are used in the experiments and results reveal that
GA_OP outperforms these three benchmark algorithms in terms of solution quality.
A review of job shop scheduling problems in multi-factories 159

5 Analysis of the literature and critics

The first systematic approach to scheduling problems was undertaken in the mid-1950s.
Since then, thousands of papers on different scheduling problems have appeared in the
literature (Allahverdi et al., 2008). DJSP is a fertile field for future research and literature
still lacks works on this area. In this review, we have surveyed almost all studies dealing
with the DJSP and DFJSP.

Figure 5 Evolution of the number of papers per year on DJSP, DFJSP

As we can see from Figure 5, published articles in the field of DJSP was not constant
during the past 15 years. Almost all articles considered the minimising of the makespan
as objective function, except for Jia et al. (2003) which have considered multi-objective
scheduling (makespan, total tardiness).
We have seen that several techniques have been employed to solve the DJSP, but the
use of approximate methods was dominant, especially GA approaches. This can be
explained by the fact that the DJSP is NP-hard to solve which makes inevitable the use of
approximate methods. In the other hand, DJSP is a real world problem and in this case,
approximate methods are more applicable than exact ones. All the studied papers
consider the case of homogeneous network factories, which is an abstraction of a real
world manufacturing problem because it is rare and difficult to have identical factories or
cells in the real world. Table 2 summarises all the studies we have reviewed in this paper
on solving DJSP and DFJSP and provides a clear classification of them in term of year of
publication, objective function, employed techniques for resolution and scheduling type
(dynamic or static).
Distributed scheduling problems are one of the most critical issues for manufacturing
systems. As mentioned earlier, operations research analysts and engineers have tried to
find solutions to this problem with varied degrees of success. In what follows, we will try
to treat the studies already presented in this paper, focusing on their shortcomings and
loopholes.
160

Table 2

Static vs. dynamic Method


Problem Papers Proposed Objective function
Static Dynamic Exact Approx. Others
I. Chaouch et al.

Distributed job Jia et al. (2002, 2003, 2007) Genetic algorithm Multiple criteria  – –  –
shop Naderi and Azab (2014) MILP + heuristic algorithms Minimise makespan  –   –
Naderi and Azab (2015) MILP + simulated annealing Minimise makespan  –  
Chaouch et al. (2017c) Survey – – – –
Chaouch et al. (2017d) Ant system algorithm Minimise makespan  – 
Chaouch et al. (2017a) Elitist ant system algorithm Minimise makespan  – 
Chaouch et al. (2017b) Modified ant colony Minimise makespan  –  
Distributed Chan et al. (2005, 2006a, 2006b) Genetic algorithm with dominant genes Minimise makespan  – –  –
and flexible Chung et al. (2009a, 2009b) A modified genetic algorithm approach Minimise makespan –  –  –
job shop
Yadollahi and Rahmani (2009) Memetic algorithms approach Minimise makespan  – –  –
De Giovanni and Pezzella (2010) Genetic algorithm, local search heuristic Minimise makespan  – –  –
Khalid and Yusof (2013) Immune algorithm Minimise makespan  – –  
Ziaee (2014) Heuristic algorithm Minimise makespan  – –  –
Chang and Liu (2017) Hybrid genetic algorithm-Taguchi method Minimise makespan  – –  
Classification of multi-factory scheduling problem papers

Liu et al. (2014) Genetic algorithm Minimise makespan  – –  –


Lu et al. (2015) Genetic algorithm Minimise makespan  – –  –
Wu et al. (2017) Genetic algorithm Minimise makespan  – –  –
A review of job shop scheduling problems in multi-factories 161

Jia et al. (2002, 2003, 2007) centred their study on the traditional problem (one factory)
rather than its distributed version (multi-factories). In Jia et al. (2003) a long
experimental study was carried out, taken the case of one factory, six machines and six
jobs. The makespan obtained was 55 time units, equal to previous researches.
Furthermore, two other popular benchmarks for scheduling problems proposed by Fisher
and Thompson (1963) were also employed to test the performance of their MGA. The
makespan obtained was 972 time units for ten jobs/ten machines and 1,207 time units for
20 jobs/five machines. These results are worse than some results already achieved by
other researchers such as Nakano and Yamada (1991), Fang et al. (1993) and Storer et al.
(1993). The second example proposed by the authors includes only three factories, four
machines and six jobs.
In these three papers, Jia et al. (2002, 2003, 2007) proposed a GA, which suffers from
many drawbacks. This algorithm almost ignored the affectation of jobs to suitable
factory. It assumed that all the individuals in a generation have the same job factory
assignment and from a generation to another, all the job factory assignment is changed
altogether.
Chaouch et al. (2017a, 2017c, 2017d) proposed different ant-based algorithms to
solve the DJSP that have shown promising results. But this approach uses a static
assignment of jobs to factories : once the job is assigned to a factory it can not be moved
to another which can limit the optimization of the of scheduling.
Regarding the DFJSP, Chan et al. (2005, 2006a) tested their solution on also three
factories, each one with four machines. This study considers only perfect maintenance,
meaning that after maintenance, the machine will be considered as good as new.
However, in reality, this may not be the case. After maintenance, the machine may only
recover to a less-deteriorated condition. It is commonly known as imperfect maintenance.
This point has been treated by Chung et al. (2009b) and Yadollahi and Rahmani (2009)
who proposed a reliable algorithm to schedule simultaneously perfect and imperfect
maintenance. The weaknesses of their approach consist also on the small size of the
tested problem (only two factories and three machines were considered).
The memetic algorithm proposed by Yadollahi and Rahmani (2009) needs more time
to run, compared to the two other algorithms SMGA (Rossi and Boschi, 2009) and
GADG (Chan et al., 2006b), and this is because of using Tabu search as a technique of
local search in the proposed algorithm.
One of the pioneer studies that had a significant impact in the field of DFJSP, was
De Giovanni and Pezzella (2010). However, the algorithm is efficient only for small
sized problems. In fact, in most of the tested instances, there is a relevant improvement of
the makespan when considering two FMUs instead of one, while the difference is not
always appreciable when a third or a fourth FMU is considered. The recent study
established by Chang and Liu (2017) is clearly worse than that of previous studies
regarding the computation time. This can be justified by the use of the Taguchi method,
which required a considerable length of time to optimise the GA parameters.
The other researches were limited to small instances. They conducted their
experiments on examples of only 2, 3 and 4 FMUs: Khalid and Yusof (2013) on 2 FMUs,
Ziaee (2014) on 4 FMUs, Liu et al. (2014) on 3 FMUs and Lu et al. (2015) on 4 FMUs.
162 I. Chaouch et al.

6 Conclusions and research opportunities

Nowadays, companies have to enlarge their production and distribution network due to
the globalisation and market unpredictability. It is essential to study the impact of this
strategy on the production system and especially on scheduling which represents the first
function to be influenced by such organisation.
This review has dealt with the JSP in multi-factory environment. We investigate and
summarise almost all previous researches dealing with the problem. The thematic area of
distributed scheduling is very promising for research; however, we notice that there is a
gap in the literature dealing with this issue. Studies are still focusing on single-
manufacturing workshop, namely one factory or facility. In addition, most of the authors
considered the makespan as principal criteria to be minimised and used mainly
approximate algorithms as a method of resolution.
According to the analysis of literature, future researches should be more focused on
distributed scheduling and various methods of resolution should be proposed.
Hybridisation of algorithms has not been exploited in the distributed job shop scheduling,
it can be a new area of research. It is interesting to explore larger instances in terms of
number of factories since the existing works considered five factories as maximum
number. Another aspect would be interesting to study, is the case where factories are not
identical since it is rarely the case in the reality. In fact, machines could have different
technologies from one factory to another. The simple DJSP with machine maintenance
has not been raised until today, only the flexible problem has been studied in the
literature. It could be a good area to investigate for future research. And finally, it will be
interesting to extend the problem and study the impact of the transportation time from
factory to customers since it is a distributed problem.

References
Adams, J., Balas, E. and Zawack, D. (1988) ‘The shifting bottleneck procedure for job shop
scheduling’, Management Science, Vol. 34, No. 3, pp.391–401.
Al-Najjar, B. (2000) ‘Impact of integrated vibration-based maintenance on plant LCC: a case
study’, Quality Reliability and Maintenance (QRM).
Allahverdi, A., Ng, C., Cheng, T.E. and Kovalyov, M.Y. (2008) ‘A survey of scheduling problems
with setup times or costs’, European Journal of Operational Research, Vol. 187, No. 3,
pp.985–1032.
Alsyouf, I. (2007) ‘The role of maintenance in improving companies’ productivity and
profitability’, International Journal of Production Economics, Vol. 105, No. 1, pp.70–78.
Bargaoui, H., Belkahla Driss, O. and Ghedira, K. (2016) ‘Minimizing makespan in multifactory
flow shop problem using a chemical reaction metaheuristic’, in 2016 IEEE Congress on
Evolutionary Computation (CEC), IEEE, pp.2919–2926.
Bargaoui, H., Belkahla Driss, O. and Ghedira, K. (2017) ‘A novel chemical reaction optimization
for the distributed permutation flowshop scheduling problem with makespan criterion’,
Computers & Industrial Engineering, Vol. 111, pp.239–250.
Behnamian, J. and Ghomi, S.F. (2013) ‘The heterogeneous multi-factory production network
scheduling with adaptive communication policy and parallel machine’, Information Sciences,
Vol. 219, pp.181–196.
Brandimarte, P. (1993) ‘Routing and scheduling in a flexible job shop by Tabu search’, Annals of
Operations Research, Vol. 41, No. 3, pp.157–183.
A review of job shop scheduling problems in multi-factories 163

Chan, F., Chung, S. and Chan, L. (2008) ‘An introduction of dominant genes in genetic algorithm
for FMS’, International Journal of Production Research, Vol. 46, No. 16, pp.4369–4389.
Chan, F., Chung, S. and Chan, P. (2006a) ‘Application of genetic algorithms with dominant genes
in a distributed scheduling problem in flexible manufacturing systems’, International Journal
of Production Research, Vol. 44, No. 3, pp.523–543.
Chan, F.T., Chung, S., Chan, L., Finke, G. and Tiwari, M. (2006b) ‘Solving distributed FMS
scheduling problems subject to maintenance: genetic algorithms approach’, Robotics and
Computer-Integrated Manufacturing, Vol. 22, No. 5, pp.493–504.
Chan, F.T., Chung, S.H. and Chan, P. (2005) ‘Anadaptive genetic algorithm with dominated genes
for distributed scheduling problems’, Expert Systems with Applications, Vol. 29, No. 2,
pp.364–371.
Chang, H-C. and Liu, T-K. (2017) ‘Optimisation of distributed manufacturing flexible job shop
scheduling by using hybrid genetic algorithms’, Journal of Intelligent Manufacturing,
pp.1–14.
Chaouch, I., Belkahla Driss, O. and Ghedira, K. (2017a) ‘Elitist ant system for the distributed job
shop scheduling problem’, in Advances in Artificial Intelligence: From Theory to Practice –
30th International Conference on Industrial Engineering and Other Applications of Applied
Intelligent Systems, IEA/AIE 2017, Arras, France, 27–30 June, Proceedings, Part I,
pp.112–117.
Chaouch, I., Belkahla Driss, O. and Ghedira, K. (2017b) ‘A modified ant colony optimization
algorithm for the distributed job shop scheduling problem’, Procedia Computer Science,
Vol. 112, pp.296–305.
Chaouch, I., Belkahla Driss, O. and Ghedira, K. (2017c) ‘A survey of optimization techniques for
distributed job shop scheduling problems in multi-factories’, in Computer Science Online
Conference, pp.369–378, Springer.
Chaouch, I., Belkahla Driss, O. and Ghedira, K. (2017d) ‘Weaknesses of ant system for the
distributed job shop scheduling problem’, in ICEIS 2017 – Proceedings of the 19th
International Conference on Enterprise Information Systems, Porto, Portugal, 26–29 April,
Vol. 1, pp.574–581.
Chung, S.H., Chan, F.T. and Chan, H.K. (2009a) ‘A modified genetic algorithm approach for
scheduling of perfect maintenance in distributed production scheduling’, Engineering
Applications of Artificial Intelligence, Vol. 22, No. 7, pp.1005–1014.
Chung, S.H., Lau, H.C., Ho, G.T. and Ip, W. (2009b) ‘Optimization of system reliability in
multi-factory production networks by maintenance approach’, Expert Systems with
Applications, Vol. 36, No. 6, pp.10188–10196.
Colorni, A., Dorigo, M., Maniezzo, V. and Trubian, M. (1994) ‘Ant system for job shop
scheduling’, Belgian Journal of Operations Research, Statistics and Computer Science, Vol.
34, No. 1, pp.39–53.
Davis, L. (1985) ‘Job shop scheduling with genetic algorithms’, in Proceedings of an International
Conference on Genetic Algorithms and their Applications, Vol. 140, Carnegie-Mellon
University Pittsburgh, PA.
De Giovanni, L. and Pezzella, F. (2010) ‘An improved genetic algorithm for the distributed and
flexible job shop scheduling problem’, European Journal of Operational Research, Vol. 200,
No. 2, pp.395–408.
Dell’Amico, M. and Trubian, M. (1993) ‘Applying Tabu search to the job shop scheduling
problem’, Annals of Operations Research, Vol. 41, No. 3, pp.231–252.
Dempster, M.A. (2012) Deterministic and Stochastic Scheduling: Proceedings of the NATO
Advanced Study and Research Institute on Theoretical Approaches to Scheduling Problems
held in Durham, England, 6–17 July 1981, Vol. 84, Springer Science & Business Media.
Fang, H-L., Ross, P. and Corne, D. (1993) A Promising Genetic Algorithm Approach to Job Shop
Scheduling, Rescheduling, And Open-Shop Scheduling Problems, University of Edinburgh,
Department of Artificial Intelligence.
164 I. Chaouch et al.

Fisher, H. and Thompson, G.L. (1963) ‘Probabilistic learning combinations of local job shop
scheduling rules’, Industrial Scheduling, Vol. 3, pp.225–251.
Garey, M.R., Johnson, D.S. and Sethi, R. (1976) ‘The complexity of flow shop and job shop
scheduling’, Mathematics of Operations Research, Vol. 1, No. 2, pp.117–129.
Graves, G.H. and Lee, C-Y. (1999) ‘Scheduling maintenance and semi-resumable jobs on a single
machine’, Naval Research Logistics (NRL), Vol. 46, No. 7, pp.845–863.
Hatami, S., Ruiz, R. and Andres-Romano, C. (2015) ‘Heuristics for a distributed parallel machine
assembly scheduling problem with eligibility constraints’, In Industrial Engineering and
Systems Management (IESM), 2015 International Conference on, IEEE, pp.145–153.
Holland, J.H. (1992) Adaptation in Natural and Artificial Systems: An Introductory analysis with
Applications to Biology, Control, and Artificial Intelligence, MIT Press, Cambridge, MA,
USA.
Jia, H., Fuh, J.Y., Nee, A.Y. and Zhang, Y. (2002) ‘Web-based multi-functional scheduling system
for a distributed manufacturing environment’, Concurrent Engineering, Vol. 10, No. 1,
pp.27–39.
Jia, H., Fuh, J.Y., Nee, A.Y. and Zhang, Y. (2007) ‘Integration of genetic algorithm and Gantt chart
for job shop scheduling in distributed manufacturing systems’, Computers & Industrial
Engineering, Vol. 53, No. 2, pp.313–320.
Jia, H., Nee, A.Y., Fuh, J.Y. and Zhang, Y. (2003) ‘A modified genetic algorithm for distributed
scheduling problems’, Journal of Intelligent Manufacturing, Vol. 14, Nos. 3/4, pp.351–362.
Kacem, I., Hammadi, S. and Borne, P. (2002) ‘Approach by localization and multiobjective
evolutionary optimization for flexible job shop scheduling problems’, IEEE Transactions on
Systems, Man, and Cybernetics, Part C (Applications and Reviews), Vol. 32, No. 1, pp.1–13.
Khalid, M.N.A. and Yusof, U.K. (2013) ‘An improved immune algorithms for solving flexible
manufacturing system distributed production scheduling problem subjects to machine
maintenance’, in Proceedings of the International Conference on Applied Mathematics and
Computational Methods, Vol. 1, pp.132–139.
Komaki, M. and Malakooti, B. (2017) ‘General variable neighborhood search algorithm to
minimize makespan of the distributed no-wait flow shop scheduling problem’, Production
Engineering, Vol. 11, No. 3, pp.1–15.
Lawrence, S. (1984) Resource Constrained Project Scheduling: An Experimental Investigation of
Heuristic Scheduling Techniques (Supplement), Graduate School of Industrial Administration.
Liu, T-K., Chen, Y-P. and Chou, J-H. (2014) ‘Solving distributed and flexible job shop scheduling
problems for a real-world fastener manufacturer’, Access, IEEE, Vol. 2, pp.1598–1606.
Lu, P-H., Wu, M-C., Tan, H., Peng, Y-H. and Chen, C-F. (2015) ‘A genetic algorithm embedded
with a concise chromosome representation for distributed and flexible job shop scheduling
problems’, Journal of Intelligent Manufacturing, pp.1–16.
Mastrolilli, M. and Gambardella, L.M. (2000) ‘Effective neighbourhood functions for the flexible
job shop problem’, Journal of Scheduling, Vol. 3, No. 1, pp.3–20.
Muth, J.F. and Thompson, G.L. (1963) Industrial Scheduling, Prentice-Hall, Englewood Cliffs, N.J.
Naderi, B. and Azab, A. (2014) ‘Modeling and heuristics for scheduling of distributed job shops’,
Expert Systems with Applications, Vol. 41, No. 17, pp.7754–7763.
Naderi, B. and Azab, A. (2015) ‘An improved model and novel simulated annealing for distributed
job shop problems’, The International Journal of Advanced Manufacturing Technology,
Vol. 239, No. 2, pp.1–11.
Naderi, B. and Ruiz, R. (2010) ‘The distributed permutation flow shop scheduling problem’,
Computers & Operations Research, Vol. 37, No. 4, pp.754–768.
Naderi, B. and Ruiz, R. (2014) ‘A scatter search algorithm for the distributed permutation flow
shop scheduling problem’, European Journal of Operational Research, Vol. 239, No. 2,
pp.323–334.
A review of job shop scheduling problems in multi-factories 165

Nakano, R. and Yamada, T. (1991) ‘Conventional genetic algorithm for job shop problems’, in
Proceedings of the 4th International Conference on Genetic Algorithms, San Diego, CA,
USA, July, pp.474–479.
Nouiri, M., Bekrar, A., Jemai, A., Niar, S. and Ammari, A.C. (2015) ‘An effective and distributed
particle swarm optimization algorithm for flexible job shop scheduling problem’, Journal of
Intelligent Manufacturing, pp.1–13.
Qi, X., Chen, T. and Tu, F. (1999) ‘Scheduling the maintenance on a single machine’, Journal of
the Operational Research Society, Vol. 50, No. 10, pp.1071–1078.
Rifai, A.P., Nguyen, H-T. and Dawal, S.Z.M. (2016) ‘Multi-objective adaptive large neighborhood
search for distributed reentrant permutation flow shop scheduling’, Applied Soft Computing,
Vol. 40, Supplement C, pp.42–57.
Riis, J.O., Luxhøj, J.T. and Thorsteinsson, U. (1997) ‘A situational maintenance model’,
International Journal of Quality & Reliability Management, Vol. 14, No. 4, pp.349–366.
Rossi, A. and Boschi, E. (2009) ‘A hybrid heuristic to solve the parallel machines job shop
scheduling problem’, Advances in Engineering Software, Vol. 40, No. 2, pp.118–127.
Storer, R.H., Wu, S.D. and Park, I. (1993) ‘Genetic algorithms in problem space for sequencing
problems’, in Operations Research in Production Planning and Control, pp.584–597,
Springer.
Taillard, E. (1993) ‘Benchmarks for basic scheduling problems’, European Journal of Operational
Research, Vol. 64, No. 2, pp.278–285.
Talbi, E-G. (2009) Metaheuristics: from Design to Implementation, Vol. 74, John Wiley & Sons.
Wu, M-C., Lin, C-S., Lin, C-H. and Chen, C-F. (2017) ‘Effects of different chromosome
representations in developing genetic algorithms to solve DFJS scheduling problems’,
Computers & Operations Research, Vol. 80, Supplement C, pp.101–112.
Xia, W. and Wu, Z. (2005) ‘An effective hybrid optimization approach for multi-objective flexible
job shop scheduling problems’, Computers & Industrial Engineering, Vol. 48, No. 2,
pp.409–425.
Yadollahi, M. and Rahmani, A.M. (2009) ‘Solving distributed flexible manufacturing systems
scheduling problems subject to maintenance: memetic algorithms approach’, in IEEE Ninth
International Conference on Computer and Information Technology, IEEE, Vol. 1, pp.36–41.
Ziaee, M. (2014) ‘A heuristic algorithm for the distributed and flexible job shop scheduling
problem’, The Journal of Supercomputing, Vol. 67, No. 1, pp.69–83.

You might also like