Вы находитесь на странице: 1из 20

This article was downloaded by: [University of Nebraska, Lincoln]

On: 03 February 2015, At: 16:45


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Production Research


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tprs20

Genetic algorithm parameter optimisation using


Taguchi method for a flexible manufacturing system
scheduling problem
a a
Gke Candan & Harun Resit Yazgan
a
Industrial Engineering Department, Sakarya University, Sakarya, Turkey
Published online: 23 Jul 2014.

Click for updates

To cite this article: Gke Candan & Harun Resit Yazgan (2015) Genetic algorithm parameter optimisation using Taguchi
method for a flexible manufacturing system scheduling problem, International Journal of Production Research, 53:3, 897-915,
DOI: 10.1080/00207543.2014.939244

To link to this article: http://dx.doi.org/10.1080/00207543.2014.939244

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
International Journal of Production Research, 2015
Vol. 53, No. 3, 897915, http://dx.doi.org/10.1080/00207543.2014.939244

Genetic algorithm parameter optimisation using Taguchi method for a exible manufacturing
system scheduling problem
Gke Candan* and Harun Resit Yazgan

Industrial Engineering Department, Sakarya University, Sakarya, Turkey


(Received 27 January 2014; accepted 21 June 2014)

The exible manufacturing system (FMS) scheduling problem is one of the most difcult NP-hard combinatorial optimi-
sation problems. The exact solution of an FMS scheduling problem cannot be found within a reasonable amount of time,
even for small size problems. Therefore, a metaheuristic algorithm is required to solve such a problem. The objective of
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

this study is to develop a genetic algorithm (GA) approach to minimise makespan of the scheduling problem. A Taguchi
orthogonal array is proposed instead of a full factorial experimental design for determining the parameters of the GA.
The effects of the GA parameters on the minimum makespan values are determined and an analysis of variance is per-
formed to investigate signicance factors on the results.
Keywords: exible manufacturing system scheduling; genetic algorithm; Taguchi orthogonal arrays method

1. Introduction
A exible manufacturing system (FMS) is a highly automated machine cell consisting of a group of processing worksta-
tions (usually CNC machine tools), interconnected by an automated material handling and storage system and controlled
by a distributed computer system (Groover 2001). Another denition of the FMS is that it is a reprogrammable manu-
facturing system capable of producing a variety of products automatically (Chryssolouris 2006). In the global competi-
tive environment, developing technology in order to meet customer demands and expectations has brought exibility in
production. The FMS has been studied over the last 25 years. In the last decade, they became important elements in the
success of enterprises (Chan and Chan 2004). The FMS is a complex discrete event dynamic system and complete util-
isation of the available resources in a system is extremely important to optimise its productivity. The most important
objective of FMS scheduling is to increase the utilisation of resources and to reduce the idle time. The resource utilisa-
tion is improved by scheduling the set of tasks so as to reduce the makespan (Cmax) (Jyothi 2012).
One way of achieving high productivity in an FMS is to solve scheduling problems optimal or near optimal. The
scheduling problem can be considered more complex in an FMS than in a traditional manufacturing system (transfer
line, job shop, ow shop, etc.). In general, it is very difcult to determine optimal solutions for FMS scheduling prob-
lems (Lee and Kim 1999). Various methods are used to nd appropriate job schedules.
In this study, a metaheuristic genetic algorithm (GA) technique commonly used as a stochastic search method in recent
years was employed and an appropriate job schedule was obtained. The effects of the GAs factors affecting the scheduling
were determined using Taguchi experimental design technique. The objective of the problem was to minimise the makespan
(Cmax) of an FMS scheduling problem. Fourty two different FMS problem data-sets were generated to nd out parameters
of the proposed GA. A bench-mark problem was also solved to illustrate effectiveness of the proposed approach.
Although there may be some similarities with our paper and others from the literature in terms of nding minimum
makespan, but, there are signicant differences related with nding parameter optimisation on the GA parameter using
Taguchi orthogonal array method to reduce experiment numbers in the area of FMS scheduling. We believe that the
study presented in this article provides a new contribution to the FMS scheduling literature.
The paper is designed as follows. We started off with a literature review related to the FMS scheduling problem in
Section 2, then we presented some brief information on FMS scheduling and the fundamentals of a GA and the Taguchi
orthogonal array method were explained in Section 3. An application was presented to illustrate the effectiveness of the
proposed approach in Section 4. The conclusions were presented in Section 5. Discussion and further work was repre-
sented in Section 6.

*Corresponding author. Email: gcandan@sakarya.edu.tr

2014 Taylor & Francis


898 G. Candan and H.R. Yazgan

2. Literature review
In the literature, different algorithms such as integer programming, ant colony, tabu search, branch and bound and Petri
net have been used to solve the FMS scheduling problem. In this study, a GA was employed to minimise the completion
time of the scheduling problem. Therefore, our research aimed to broaden the heuristics, metaheuristics and particularly
genetics approaches which have been employed in the FMS scheduling problems.
A heuristic method based on some logical rules to reach an accurate solution. The main purpose of the method is to
nd out an acceptable good solution in a reasonable time period. This solution may not be the best (or optimal), but it
may be good enough for the problem. If a problem cannot be dened as mathematical programming forms or impossi-
ble to reach an optimum solution in a reasonable time constraints; the heuristic algorithm can be preferred. Some of
researches which are related with FMS scheduling problem and employed heuristic approaches are summarised as fol-
lows. Rachamadagu and Stecke (1993) classied the FMS scheduling problems and developed a heuristic method for
solving them. Kodeekha (2004) developed a heuristic method called the break and build method. Dodu, Onural, and
Cerit (2004) studied a scheduling problem that contains some processes and an assembly. Jansen, Mastrolilli, and
Oba (2005) proposed a two-phased method called linear time approximation scheme. Additionally, Gltekin (2006)
recommended a heuristic algorithm to solve scheduling problem under the FMS environment.
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

One of weakness of the developed heuristic algorithms is that the algorithms may not guaranties avoiding in a local
optimum values (minimum or maximum). In order to overcome the weakness of the heuristic algorithms, new heuristics
(metaheuristic-upper heuristic) were rst introduced by Glover (1986). A metaheuristic is a superimposed upper general
methodology that can be used as guiding strategies in designing underlying heuristics to solve specic optimisation prob-
lems. The new heuristic may provide a sufciently good solution to an optimisation problem, especially with incomplete
or imperfect information or limited computation capacity. The metaheuristic may make few assumptions about the optimi-
sation problem being solved, and so they may be usable for a variety of problems (Blum and Roli 2003). The metaheuristic
can include stochastic optimisation and often nd good solution with less computer effort than classic heuristics (Talbi
2009). Apart from the above-mentioned heuristic approaches related with the FMS scheduling, the metaheuristic
approaches have also been proposed by researchers. One of them was by Xing, Chen, and Yang (2009). They developed a
search method to nd an optimal or a near-optimal solution of their scheduling problem. Bagheri et al. (2010) suggested
an articial immune algorithm to minimise makespan for a exible job shop scheduling problem. Yazdani, Amiri, and
Zandieh (2010) used a parallel neighbourhood search method to minimise makespan in a exible job shop scheduling
problem. They claimed that their algorithm improved the search process ability. Li, Pan, and Liang (2010) developed a
hybrid algorithm called TSPCB and tried to minimise a maximum completion time. Hmida et al. (2010) minimised the
maximum completion time using a variant of the climbing discrepancy search algorithm. Moslehi and Mahnam (2011)
used hybrid particle swarm and local search algorithms to solve exible job shop scheduling problem with different release
times. Wang et al. (2012) used the BEDA algorithm to minimise the maximum completion time using the Taguchi experi-
mental design method to determine the appropriate number of parameters. Trkcan (1999) worked on a scheduling prob-
lem under due date constraints which was solved by a proposed approach based on a GA. Chryssolouris and Subramaniam
(2001) studied a dynamic job shop scheduling problem using GA and compared results with other dispatching rules. They
claimed that their proposed scheduling approach with GA was better than other rules. Ho and Tay (2004) studied the
scheduling problem using the GENACE method. They aimed to observe effecting selection and mutation operations of
chromosomes on the performance of the GA. Gen, Gao, and Lin (2009) and Zhang, Gao, and Shi (2011) proposed a multi-
stage genetic approach to minimise a maximum completion time for the FMS scheduling problem. In addition, to the
above studies, GAs based on fuzzy logic has also been considered to solve the scheduling problem. Deming Lei (2008,
2010) assumed processing time as fuzzy and aimed to minimise maximum fuzzy completion time using a GA. Further-
more, Zakaria and Petrovic (2012) and Deming Lei (2012) studied a fuzzy GA for the solution of a scheduling problem.
Mohammad Pour, Yadollahi, and Haghighat (2010) studied a FMS scheduling problem with a hybrid method that causes
Petri nets, GAs and tabu search to minimise makespan. They observed that the proposed algorithm requires a longer com-
putational time than the other two approaches because of the use of tabu search as the local search technique. Pezzella,
Morganti, and Ciaschetti (2008) solved a FMS scheduling problem using a GA. They solved the problem of assigning jobs
to machines with the help of a localisation approach and claimed that they achieved a better result than other approaches
described in the literature. Keung, Ip, and Chan (2001) used a GA to solve earliness/tardiness production scheduling and
planning problem in FMS. Chiang and Fu (2004) studied a sequencing problem; the objective was to maximise meeting
due-date rate. They applied a GA to reduce the computational time. Hsu et al. (2008) studied a cyclic scheduling problem
in an FMS and aimed to reduce the work in process (WIP) by using GA to obtaining the sequence of tasks for a FMS.
Another research was completed by Sharma, Singh, and Sharma (2012). They studied for determination of a schedule with
the objective of minimising Cmax using the GA with roulette wheel (RW) selection process in the FMS.
International Journal of Production Research 899

It is also known that parameters of the GA affect solution quality. Unfortunately, there is no any specic rule which
provide a tuning of the GA parameters. To identify best parameter, the GA may take serious computational effort. Tagu-
chi experimental design technique can also be employed to reduce computational effort instead of inefcient trial and
error method. Cheng and Chang (2007) developed the GA with parameters design by the Taguchi for a ow shop
scheduling problem. They suggested that a better Cmax solution within a shorter time period could be found if these
parameters were employed as an initial solution. However, they did not illustrate any specic attempt to illustrate effec-
tiveness of their approach on any benchmark problem. The studies cited above are summarised in Table 1.
An objective of this study is to determine the parameters of a proposed GA while minimising the Cmax in the FMS
scheduling problem. To identify the parameters of the proposed GA, a Taguchi orthogonal array experimental design
approach has been chosen instead of a full factorial design. We are not aware of a study in the literature aiming to mini-
mise the Cmax. The FMS scheduling problems using a GA in which the parameters are determined using the Taguchi
experimental design. Therefore, the study presented in this article provides a new contribution to the FMS scheduling
literature.
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

3. The problem
In this section, complexity of the FMS scheduling problem in terms of difculties and its scope were discussed

3.1. Complexity of the FMS scheduling problem


A complexity class represents the set of all problems that can be solved within reasonable time. These classes of the
problem are is P and NP. P illustrate that a problem solution can be found in the polynomial time period. NP means that
set of all decision problems that can be solved by non-deterministic algorithm in polynomial time. NP-hard problems,
which are probably the efcient algorithms, do not exist. An algorithm performance needs two important elements to
solve a problem such as time and solution space. The time complexity is to require the number of steps to solve the
problem. The solution space consists of all possible solution which is required to evaluate to reach the best one. A prob-
lem is tractable (or easy) if there exists a polynomial time algorithms to solve it. A problem is intractable (or difcult)
if no polynomial time algorithms exists to solve the problem Talbi (2009).
In literature, several studies are showed us NP-hardness of the scheduling problems. Cheng and Wang proved ow
shop problem with machine and job numbers are NP-hard (2000). Sotskov and Shakhlevich (1995) solved ow shop
problems in order to minimise Cmax and proved that the ow shop problems with three machine and three job numbers
are NP-hard. Garey, Johnson, and Sethi (1976) showed that in their study, determining a shortest-length schedule in an
m-machine ow shop is NP-complete for M greater than equivalent to three. And determining a minimum mean-ow-
time schedule in an m-machine ow shop is NP-complete for every m greater than equivalent to two. Since there is
more than one machine on the shop oor, jobs have to be considered for sequencing on other exible machines as well.
This issue begets higher complexity for the FMS scheduling and makes its complexity rise to (n!)m and belongs to the
class of problems as NP-hard (Liu and MacCarthy 1997; Sankar, Ponnanbalam, and Rajendran 2003; Taghavifard,
Heydar, and Mousavi 2009). It requires an exponential time to be solved in optimality. Metaheuristic constitute an
important alternatives to solve the NP-hard problems.

3.2. Scope of the problem


Because of the NP-hardness of the FMS scheduling problem, a metaheuristic algorithm (i.e. GA) was chosen. Parame-
ters of the GA were identied by the Taguchi experimental design method instead of more costly full factorial design.
Table 1. Studies related to FMS scheduling problem.

Methods Studies

Heuristic algorithms Rachamadagu and Stecke (1993), Jansen, Mastrolilli, and Oba (2005), Kodeekha (2004), Dodu, Onural, and
Cerit (2004), Gltekin (2006)
Metaheuristic Xing, Chen, and Yang (2009), Bagheri et al. (2010), Yazdani, Amiri, and Zandieh (2010), Li, Pan, and Liang
algorithms (2010), Hmida et al. (2010), Moslehi and Mahnam (2011), Wang et al. (2012)
Genetic algorithms Keung, Ip, and Chan (2001), Trkcan (2002), Chiang and Fu (2004), Ho and Tay (2004), Hsu et al. (2008)
Gen, Gao, and Lin (2009), Zhang, Gao, and Shi (2011), Lei (2008, 2010, 2012), Zakaria and Petrovic (2012),
Cheng and Chang (2007), Mohammad Pour, Yadollahi, and Haghighat (2010), Chryssolouris and
Subramaniam (2001), Sharma, Singh, and Sharma (2012)
900 G. Candan and H.R. Yazgan

In order to complete experiments, the data-set were generated for different job and machine numbers. The processing
times of the data-set generated according to literature based on a uniform distribution U (1, 25) (Ceran 2006; Choua
and Lee 1999; Engin 2001). Forty-two different problem sets such as machines 2, 3, 5, 10, 20, 25, 30 and jobs 2, 3, 5,
10, 15, 20 were produced. Each problem set has ve operations. An example, processing times for 20 jobs, 5 operations
and 5 machines were given in Appendix 1. It is an NP-hard problem and there are (20!)5 solution space.

3.3. The proposed approach


In this study, a new approach based on a GA and Taguchi experimental design was proposed to determine a minimum
Cmax. The steps are summarised in Table 2.
The effectiveness of the proposed approach was illustrated on an example in section 5. The methods employed in
this study were explained briey in the following sections.

3.3.1. The genetic algorithm


Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

The GA was developed by John Holland and students in the 1970s. Apart from traditional optimisation methods, a code
format instead of a parameter set is employed in the GA. The GA works according to the rules of probability and
requires only an objective function. The algorithm searches a specic part of the solution rather than an entire solution.
Thus, a solution is expected to be attained in a much shorter time (Goldberg 1989). The GA can be applied to attaining
the solution of large problems which have a large number of factors affecting a solution space.
In the GA, the code of each point in the solution space with an array of binary bits is called a chromosome. The t-
ness value of each point has to be calculated. Instead of a single point, the GA maintains points as a set of a population.
For each generation, genetic operators such as crossover and mutation are employed to create a new population. The t-
ness value is expected to reach an improved value after repeating a number of generations.
The main parameters of the GA are population size, crossover rate, mutation rate, a crossover method and a deriva-
tion method. The performance of the algorithm depends on the selection of these parameters. Numerous studies have
been conducted to nd optimal control parameters, but the parameters used in general for all problems have not been
found (Altparmak, Dengiz, and Smith 2000). Brief descriptions of these parameters are given below.

3.3.1.1. Population size. The population consists of chromosomes. Each chromosome represents candidate solutions of
a problem. A chromosome tness value represents a solution value. The GA enables us to search for an improved solu-
tion by employing different chromosomes in the population. Therefore, the population size is a factor in determining the
quality of a solution. The greater the population size, the greater the potential of attaining an improved solution. In addi-
tion, increasing population size also implies an increasing number of evaluations. Although a small population size may
decrease the number of required iterations, the attained solution may become trapped in an undesired local optimum.
Therefore, a reasonable population size should be chosen (Sivanandam and Deepa 2008).

3.3.1.2. Crossover. Crossover is similar to the search space of the GA, but it provides the opportunity to reach unex-
plored regions. Swap is considered as a crossover operation. In this study, a swap operation is performed between two
chromosomes based on crossover ratios of 10, 50 and 90%. These crossover rates were chosen from the literature
(Aggoune, Mahdi, and Portman 2001; Kacem 2003).

3.3.1.3. Mutation rate. Mutation generates deterioration and provides a genetic variability in the chromosome structure.
Genetic mutation in some genes occurs randomly. Mutation operator provides a random diversity in the population
(Spears 1993). Mutation operators provide the ability to overcome a local optimum point solution. To mitigate or even

Table 2. The proposed approach steps.

Steps Activity

1 Generate the problem data-set


2 Determine parameters and their levels for the GA
3 Determine experimental design set for the Taguchi method and generate solutions according the Taguchi design set by the GA
4 Comparison between Taguchi L18 and the GA based on determined effective factors
5 Comparison benchmark problems
International Journal of Production Research 901

avoid trapping into the local optima, the mutation operator provides a mechanism to explore new solutions and
maintains the diversity of the population in GAs search (Tang and Tseng 2013). In this study, 0.5, 10 and 15% of the
mutation rates were chosen from the literature (Kacem 2003; Pezzella, Morganti, and Ciaschetti 2008).

3.3.1.4. Crossover method. The most important parameter of the GA is crossover method. There are several types of
this parameter, such as position based (PB), order based (OB) and partially-mapped (PMX) crossovers.
 Position-based crossover: In this method, several processes are chosen from the rst parent and placed in the
corresponding positions in the child string. The processes that are not taken from the rst parent are then taken
from the second parent to and placed in the corresponding positions in the child string in the order in which
they are positioned in the second parent (Miseviius and Kilda 2005). The result is a complete string with one
and only one copy of each process number as shown in Figure 1.
 Orde- based crossover: The basic feature of the OB is to preserve the relative order of the parents in the chro-
mosomes. Elements are selected from parent 1 and parent 2 and copied to the offspring. Missing elements are
taken from the other parent in order (Miseviius and Kilda 2005) as shown in Figure 2.
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

The basic feature of order-based crossovers (OB) is that they preserve the relative order of the alleles in chromo-
somes. So, a number of items (elements) are selected from one of the parents and copied to the offspring. The missing
elements (alleles) are taken from the other parent in order (Miseviius and Kilda 2005).
 Partially mapped crossover (PMX): The PMX builds an offspring by choosing a subsequence of a tour from
one parent to preserve the order and position of as many positions as possible from the other parent. Firstly, a
subsequence is delimited by two cut-off points. After this, the segments between the cut-off points are swapped
around. The extracts from the segments that are swapped form a series of mappings. In the example shown
below 1 is mapped to 8, 9 to 5, 8 to 7 and 6 to 3. Secondly, the positions for which there is no conict are
lled in. In the example, the positions 4 and 2 are in both chromosomes. Lastly, the mappings are used to ll
in the remaining positions (Michalewicz 1996) (Figure 3).

3.3.1.5. Derivation method. Different derivation methods are available in the selection of individuals that are to be used
in subsequent populations. The method used depends on the level of compliance of the population. The most commonly
used selection strategies are RW and tournament (T). The GA steps are given below:
 A job sequence is determined randomly and is considered as a starting solution.
 A tness value for each chromosome in the population is calculated.
 New chromosomes are generated using crossover and mutation operators under different rules and strategies.
 The tness values of the new chromosomes are calculated and compared with the values in the population. If
the tness value of a new chromosome is better than any of the values in the population, this new chromosome
is then kept and the chromosome with the worst value is deleted from the population.
 The generation process is terminated when the number of iterations reaches a certain threshold amount. The
minimum value of the solution is selected as a best solution. The GA steps are shown in Figure 4.

Figure 1. Position-based crossover.

Figure 2. Order-based crossover.


902 G. Candan and H.R. Yazgan
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Figure 3. Partially mapped crossover.

Determine a job
sequence randomly
to start

Calculate Fitness
value

Create new
chromosomes using
Crossover and
Mutation Methods

Calculate fitness
values of new
no chromosomes

Threshold number of
iterations reached?

yes

Choose minimum
value of the solution end
as the best solution

Figure 4. The GA steps.


International Journal of Production Research 903

Table 3. Taguchi experimental design steps.

Steps Activity

1 Selecting and evaluating interactions between factors


2 Determining levels of factors
3 Selecting the appropriate orthogonal array
4 Assign factors and their interactions to each column of the orthogonal array
5 Realisation of the experiments
6 Analysis of the results

3.3.2. Taguchi experimental design method


A classical experimental design method was rst developed in 1920 by Ronald Fischer to improve the efciency of
agricultural production (Yang and Tarng 1998). Although the method has been used in numerous disciplines since 1920,
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

in many cases it may not be practical to implement the method in the manufacturing sector because of cost and time
considerations. To overcome this problem, Genichi Taguchi developed a new experimental design method to increase
the efciency of implementation and evaluation of experiments. Its structure is more suitable for evaluating production
processes because the required number of experiments is reduced signicantly. The design of experiment using the
Taguchi method provides a simple, efcient and systematic approach to determine optimum conditions (Davidson,
Balasubramanian, and Tagore 2008; Du Plessis and De Villiers 2007).
Experimental results are converted to a signal/noise (S/N) ratio which means a ratio of an average standard deviation
(Taguchi, Chowdhury, and Wu 2005). This ratio can be calculated in three different ways, such as small value is good,
great value is good and nominal value is good. A larger S/N ratio indicates a better test result. So in experiments, a
level of the factor which has the highest ratio represents a better performance. The ratio allows controlling mean and
variance while at the same time an analysis of variance (ANOVA) is performed. In this way, the effects of factors can
be revealed statistically. S/N ratio can be calculated with Equation (1).
" #
1 X n
g 10  log10  y 2
(1)
n i1 i

n: the number of observations in the experiment and yi: makespan values (i = 1 to n).
The algorithms steps were illustrated in Table 3.
As a result of performing these steps, experimental parameters to obtain the optimum performance for process or
product were determined. Factors that inuence the quality can be estimated based on the experimental values and qual-
ity values can be obtained as a result of the specied optimal experimental parameters (Yang and Tarng 1998). Khaw,
Lim, and Lim (1995), mentioned that, Taguchi method offers orthogonal arrays as a mathematical tool that analyse the
smallest number of experiments that have a large number of parameters. Thus, that would reduce time and effort. With
the Taguchi experimental design, quality gains a new perspective. When quality improvement is formulated, costs
incurred as a result of variability can be minimised. The technique was applied in this study to reduce the number of
experiments. There were ve factors and one of them had two levels and the others had three levels. So we applied the
L18 (L18, one two-level and up to seven three-levels) orthogonal array to reduce the number of experiments. This meant
that only 18 experiments were required to reach a conclusion. To achieve accuracy experiments were repeated ve times
(i.e. 18 5 = 90) for each problem.

4. An application
In this section; the proposed steps given in Table 2 are followed to illustrate the effectiveness of the approach. The
applications steps are as follows:
Step 1: Forty-two different problem data-set jobs (2, 3, 5, 10, 20, 25, 30) and machines (2, 3, 5, 10, 15, 20) were
generated to test the proposed approach. The processing times of the data-set generated based on a uniform distribution
U(1, 25). One of the data-set was given in detail in Appendix 1.
Step 2: In this step, the GA parameters were determined. The derivation method, population size, crossover method,
mutation rate and crossover rate were chosen as main parameters. These parameters and their value or type are com-
monly employed in the GA literatures (Lee and DiCesare 1994) and the parameters and their values or type are called
904 G. Candan and H.R. Yazgan

Table 4. Parameters (factors) and their levels.

Levels
Factors Level 1 Level 2 Level 3

Derivation method Roulette wheel (RW) Tournament (T)


Population size 50 100 150
Crossover method Position based (PB) Order based (OB) Partially-mapped (PMX)
Mutation rate 0,5 10 15
Crossover rate 10 50 90

factors and levels, respectively. The factors and their levels were summarised in Table 4. The number of generation
was constant and 1000 times for benchmark problem and the other problem sets (Pezzella, Morganti, and Ciaschetti
2008; Lin et al. 2012).
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Step 3: To investigate the effectiveness of these parameters, 162 (2 34) different experiments were required for a
full factorial search. In addition, the number of experiments would be repeated ve times to verify the accuracy of the
solutions. Therefore, the number of the experiments required for a problem set was 810 (162 5). In our case, there
were 42 different problem sets; the required number of experiments was therefore 42 810 = 34,020. The completion of
all these experiments requires considerable computational time. To reduce the number of experiments, the Taguchi
experimental design technique was chosen. In our problem, the derivation method (factor) had two levels but others fac-
tors such as population size, crossover method, mutation rate, crossover rate had three levels and interactions were not
assumed amongst the factors. Therefore, a L18 design set was as shown below was well suited for our investigation
(L18, one two-level and seven three-levels). Layout of orthogonal arrays for parameter design is seen on literature (Wu
and Wu 2000). Yih-fong (2006) designed parameters with L18 design for optimisation computerised numerical control
turning tool steels. Lin et al. (2009) adopted L18 orthogonal array to explore effects of magnetic force on electrical dis-
charge machine. Also, imek, , and Simsek (2013) employed L18 orthogonal array to determine optimum concrete
performance and Kvak (2014) used L18 orthogonal array to determine optimal machining parameters in milling
machine. Also, according to York Universitys (http://www.york.ac.uk/depts/maths/tables/orthogonal.htm) reference, one
two-level and up to seven three-level factors complies with L18 orthogonal array. The chosen L18 design set was shown
in Table 5.
A number of an experiment for a full factorial base on number of parameters and level of each parameters. A general
formulation can be dened as;

Table 5. L18 design set.

Column
Experiment
number 1 2 3 4 5 6 7 8

1 1 1 1 1 1 1 1 1
2 1 1 2 2 2 2 2 2
3 1 1 3 3 3 3 3 3
4 1 2 1 1 2 2 3 3
5 1 2 2 2 3 3 1 1
6 1 2 3 3 1 1 2 2
7 1 3 1 2 1 3 2 3
8 1 3 2 3 2 1 3 1
9 1 3 3 1 3 2 1 2
10 2 1 1 3 3 2 2 1
11 2 1 2 1 1 3 3 2
12 2 1 3 2 2 1 1 3
13 2 2 1 2 3 1 3 2
14 2 2 2 3 1 2 1 3
15 2 2 3 1 2 3 2 1
16 2 3 1 3 2 3 1 2
17 2 3 2 1 3 1 2 3
18 2 3 3 2 1 2 3 1
International Journal of Production Research 905

Table 6. Ex27 (20, 5) Cmax using the L18 orthogonal array.

Exp. Derivation Population Crossover Mutation rate Crossover rate Average Cmax Standard
no method size method (%) (%) (min) deviation

1 50 PB 0,5 10 183 2,61


2 50 OB 10 50 151 5,57
3 50 PMX 15 90 148 3,77
4 100 PB 0,5 50 163 4,52
5 RW 100 OB 10 90 151 2,61
6 100 PMX 15 10 146 2,90
7 150 PB 10 10 163 5,37
8 150 OB 15 50 149 5,57
9 150 PMX 0,5 90 145 2,00
10 50 PB 15 90 153 3,77
11 50 OB 0,5 10 176 4,52
12 50 PMX 10 50 164 4,52
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

13 100 PB 10 90 150 3,35


14 T 100 OB 15 10 163 5,57
15 100 PMX 0,5 50 173 5,57
16 150 PB 15 50 160 3,77
17 150 OB 0,5 90 163 4,52
18 150 PMX 10 10 168 5,57

Note: RW: roulette wheel, T: tournament, PB: position based, OB: order based, PMX : partially mapped.

n
P ai !
i1

ai: number of factor levels.


In this study, the proposed GA has ve parameters such derivation method (2 level), population size (3 level), cross-
over method (3 level), mutation rate (3 level), crossover rate (3 level).
The number of full factorial: (2!) (3!) ((3!) (3!)(3!) = 2592 times. The number is increased dramatically based
on level of the parameters.
However, in this study, Taguchi L18 experimental design is chosen, the expected number of experiment may be such
as 9 (L9), 18 (L18) or 36 (L36) based on level of parameters. Therefore, expected number of experiment was 18 times.
It was gained %99 times less experiment.
And choosing the Taguchi experimental design set (i.e. L18) instead of a full factorial design for each problem set,
results in the number of experiments being lower, 3780 (42 18 5), consequently only requiring a more reasonable
computational time. Therefore, the lower number of experiments and the consequent lower computational time required,
illustrated the rationale of why the Taguchi method was chosen in this study.
The average Cmax values were found using the GA for the problem given in Appendix 1. The results were given in
Table 6.
Effective factors and the magnitudes of the effectiveness were found using signal to noise ratios, (S/N), which were
chosen as the lowest value is the best. Equation 1 was used to calculate the S/N. The ANOVA was completed on the
ratios (S/N). The results were given in Table 7.

Table 7. Signal to noise ratios (S/N) and ANOVA results.

Signalnoise ratio
Factors
First level Second level Third level Sum of squares Factor effect (%) F value P value

Derivation method 43,8 44,25 0,9342 0,185,567 6,67 0,032


Population size 44,19 43,92 43,96 0,2528 0,050,216 0,90 0,443
Crossover method 44,17 44 43,89 0,2362 0,046,918 0,84 0,465
Mutation rate 44,44 43,95 43,68 1,7896 0,355,481 6,39 0,022
Crossover rate 44,39 44,07 43,61 1,8215 0,361,818 6,50 0,021
906 G. Candan and H.R. Yazgan

According to these results, the mutation rate and the crossover rate of p values were less than 0.05 within a 95%
condence interval. Therefore, these factors were chosen as being effective on minimisation of the Cmax. The levels of
the mutation and crossover rates were found using the S/N as 15 and 90%, respectively. Another important deduction
was related to ineffective factors such as derivation method, population size and crossover method. Any level of these
ineffective factors could be chosen as the levels of these ineffective factors do not produce any signicant differences
for the Cmax values. In total, we solved 42 problem sets with employing the GA in L18 problem. Detail problem date-
sets of 20 jobs and 5 machines were illustrated in Appendix 1. Way of nding results was illustrated step-by-step for
the problem. Remaining 41 problems were solved in same way. Summary of results are given at Table 8. An effective
factor for each problem set was identied using the ANOVA and was summarised and denoted with check mark in
Table 8.
According to the results in Table 8, the total percentage of the effective factors which were observed in the 42 prob-
lem sets, such as the crossover rate, the population size, the mutation rate, the crossover method and the derivation
method were 47.62, 23.81, 11.9, 11.9 and 4.77%, respectively.

Table 8. The most effective factors of 42 problem sets.


Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Exp. number (job, machine) Derivation method Population size Crossover method Mutation rate Crossover rate

Ex1 (2, 2) U
Ex2 (2, 3) U
Ex3 (2, 5) U
Ex4 (2, 10) U
Ex5 (2, 15) U
Ex6 (2, 20) U
Ex7 (3, 2) U
Ex8 (3, 3) U
Ex9 (3, 5)
Ex10 (3, 10) U
Ex11 (3, 15) U
Ex12 (3, 20) U
Ex13 (5, 2) U
Ex14 (5, 3) U
Ex15 (5, 5) U
Ex16 (5, 10) U
Ex17 (5, 15) U
Ex18 (5, 20) U
Ex19 (10, 2) U
Ex20 (10, 3) U
Ex21 (10, 5) U
Ex22 (10,10) U
Ex23 (10, 15) U
Ex24 (10, 20) U
Ex25 (20, 2) U
Ex26 (20, 3) U
Ex27 (20, 5) U
Ex28 (20, 10) U
Ex29 (20, 15) U
Ex30 (20, 20) U
Ex31 (25, 2) U
Ex32 (25, 3) U
Ex33 (25, 5) U U
Ex34 (25, 10) U
Ex35 (25, 15) U
Ex36 (25, 20) U
Ex37 (30, 2) U
Ex38 (30, 3) U
Ex39 (30, 5) U
Ex40 (30, 10) U
Ex41 (30, 15) U
Ex42 (30, 20) U
Total (%) 4,77 23,81 11,9 11,9 47,62
International Journal of Production Research 907

Minimum values of the Cmax could be chosen amongst the L18 values for each problem set. One of the weakness of
the Taguchi experimental design was that it did not search the whole solution space, therefore, the minimum value could
be a local minimum too, because the approach was not a full factorial design.
We want to illustrate the superiority of the effective factors used in the GA (approach 2) against the minimum Cmax
value used in the L18 experimental design (approach 1). C1max and C2max were found by approach 1 and approach 2,
respectively.

Table 9. The most effective combination of factor levels and results.

Exp. number Derivation Population Crossover Mutation rate Crossover rate C2max
(job, machine) method size method (%) (%) (min)

Ex1 (2, 2) RW 100 PB 15 90 33


Ex2 (2, 3) RW 150 PMX 15 50 28
Ex3 (2, 5) RW 100 PMX 10 90 19
Ex4 (2, 10) RW 150 PB 15 50 15
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Ex5 (2, 15) RW 100 OB 10 50 14


Ex6 (2, 20) RW 150 OB 10 90 18

Ex7 (3, 2) RW 100 PMX 15 90 43


Ex8 (3, 3) RW 100 PB 15 90 34
Ex9 (3, 5) T 150 PMX 15 50 23
Ex10 (3, 10) T 150 PMX 15 90 25
Ex11 (3, 15) RW 100 PB 15 90 24
Ex12 (3, 20) T 150 PMX 15 90 23

Ex13 (5, 2) RW 150 PB 10 90 66


Ex14 (5, 3) T 150 OB 15 90 56
Ex15 (5, 5) T 150 OB 15 90 32
Ex16 (5, 10) RW 150 PMX 10 90 32
Ex17 (5, 15) T 150 OB 15 90 31
Ex18 (5, 20) RW 100 OB 10 90 29

Ex19 (10, 2) T 100 OB 10 50 132


Ex20 (10, 3) RW 150 PB 15 90 108
Ex21 (10, 5) RW 150 OB 15 50 77
Ex22 (10, 10) RW 100 PMX 15 90 59
Ex23 (10, 15) RW 150 PB 15 90 48
Ex24 (10, 20) RW 150 PMX 15 90 44

Ex25 (20, 2) RW 100 OB 15 90 307


Ex26 (20, 3) T 150 PMX 50 90 216
Ex27 (20, 5) RW 100 PMX 15 90 142
Ex28 (20, 10) RW 150 PB 15 90 92
Ex29 (20, 15) RW 150 PB 15 90 77
Ex30 (20, 20) RW 150 PMX 15 90 64

Ex31 (25, 2) RW 150 PMX 15 50 439


Ex32 (25, 3) RW 100 PB 15 90 354
Ex33 (25, 5) T 150 PMX 15 90 189
Ex34 (25, 10) RW 100 PMX 15 50 114
Ex35 (25, 15) RW 100 PB 10 90 100
Ex36 (25, 20) RW 150 PMX 10 90 83

Ex37 (30, 2) T 150 PMX 10 90 522


Ex38 (30, 3) RW 150 PB 15 90 432
Ex39 (30, 5) T 150 PMX 15 90 257
Ex40 (30, 10) T 100 PB 15 50 137
Ex41 (30, 15) RW 150 PB 15 90 122
Ex42 (30, 20) RW 100 PMX 10 90 105
Most repetitive factor
Levels according to Mode RW 150 PMX 15 90

Note: RW: roulette wheel, T: tournament, PB: position based, OB: order based, PMX: partially mapped.
908 G. Candan and H.R. Yazgan

Step 4: The performances of two approaches were summarised in Table 9.


Approach 1: Determine and choose minimum C1max using the GA using parameters from the L18 experimental
design set.
Approach 2: Determine minimum C2max from the GA using effective factors as parameters.
In order to conclude these experimental investigations from Table 9, all these factor levels are analysed by calculat-
ing mode. The most effective levels of the GA factors were determined as; the crossover rate as 90%, the mutation rate
as 15%, population size as 150, the crossover method as PMX and the derivation method as RW. Then, we used these
factor levels on benchmark problem to proof our proposed method.
A hypothesis test was done to determine whether there was any difference between C1max and C2max or not?
H0 :l1 l2

H1 :l1 6 l2
Table 10 shows a comparison between two approaches and Table 11 shows the t-test results. According to the t-test
results, the p value was less than 0.05, so H0 was rejected. This meant that there was a signicant difference between
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Table 10. Comparison of two approaches (C1max vs. C2max).

Exp. number (job, machine) C1max (min) C2max (min)

Ex1 (2, 2) 33 33
Ex2 (2, 3) 28 28
Ex3 (2, 5) 22 19
Ex4 (2, 10) 15 15
Ex5 (2, 15) 14 14
Ex6 (2, 20) 18 18
Ex7 (3, 2) 44 43
Ex8 (3, 3) 34 34
Ex9 (3, 5) 23 23
Ex10 (3, 10) 25 25
Ex11 (3, 15) 24 24
Ex12 (3, 20) 24 23
Ex13 (5, 2) 66 66
Ex14 (5, 3) 57 56
Ex15 (5, 5) 32 32
Ex16 (5, 10) 32 32
Ex17 (5, 15) 31 31
Ex18 (5, 20) 29 29
Ex19 (10, 2) 136 132
Ex21 (10, 5) 77 77
Ex22 (10, 10) 60 59
Ex23 (10, 15) 48 48
Ex24 (10, 20) 44 44
Ex25 (20, 2) 308 307
Ex26 (20, 3) 216 216
Ex27 (20, 5) 145 142
Ex28 (20, 10) 93 92
Ex29 (20, 15) 77 77
Ex30 (20, 20) 68 64
Ex31 (25, 2) 442 439
Ex32 (25, 3) 356 354
Ex33 (25, 5) 189 189
Ex34 (25, 10) 114 114
Ex35 (25, 15) 101 100
Ex36 (25, 20) 83 83
Ex37 (30, 2) 524 522
Ex38 (30, 3) 432 432
Ex39 (30, 5) 260 257
Ex40 (30, 10) 137 137
Ex41 (30, 15) 122 122
Ex42 (30, 20) 109 105
International Journal of Production Research 909

Table 11. T test result.

N Mean St dev. SE mean

C1max 42 114,3 126,8 19,6


C2max 42 113,5 126,4 19,5
Difference 0,857 1,299 0,200

95% CI for mean difference, (0,452; 1262).


T-test of mean difference = 0 (vs not = 0): T-value = 4,28; p-value = 0000.

the C1max and C2max values. Additionally, the average of the C2max values was better than the average of the C1max val-
ues so these effective factors could be used in further investigations. Now, we reached a point where we compared the
proposed approach and a benchmark problem from the literature. The factors levels extracted from Table 9 were
employed as a parameter set of the proposed GA parameters. In the next step, the details of the comparison were given.
Step 5: In this step, we compared the proposed GA with studies from the literature on a benchmark problem. The
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

problem data in Table 12 (from Lee and DiCesares 1994) were employed to determine the minimum Cmax value using
the proposed GA. The results were summarised in Table 13.
According to the results, the proposed GA produced a better result than the others. The solution of the problem is
illustrated in Figure 5.
One of the important objectives of this study was to identify parameters and their value for the GA in FMS environ-
ment. Forty-two different problems were generated for this investigation. The Taguchi experimental design set (i.e. L18
in this study) was employed to nd out effective parameters level (values) base on 18 experimental tests. This is the
big advantage of the approach compare to full factorial design approach. The GA run with found effective factors value
to solve same date-set for each problem. It was seen that the GA which runs with effective factors value, produced bet-
ter results or same (but not worse) than the GA run different parameter setting in L18. Of course, we did not compare
the results with full factorial design set because of bigger number of test (i.e. 20 jobs, 5 machines, number of full facto-
rial test was 2592. If it was repeated 5 times, 5 2592 = 12,960). We could jump another discussion whether could be
found general parameters which could be used different problem set for the GA parameters tuning. To nd out answer
of this problem, we completed a statistical analysis to identify most repeated parameters (mode) among 42 experiments

Table 12. A benchmark problem data-set (Taguchi, Chowdhury, and Wu 2005).

Job Operation Machine Processing time Job Operation Machine Processing time

1 1 1 7 4 1 2 9
1 3 4 1 3 5
2 2 3 2 1 6
3 1 3 2 3 2
3 3 6 3 2 7
4 1 2 3 3 12
4 2 4 4 1 9
2 1 1 8 4 2 6
1 2 12 4 3 3
2 3 4 5 1 1 10
3 1 7 1 3 15
3 2 14 2 2 7
4 1 8 2 3 14
4 3 4 3 1 5
3 1 1 10 3 2 8
1 2 15 4 1 4
1 3 8 4 2 6
2 2 2 4 3 8
2 3 6
3 1 2
3 3 4
4 1 6
4 2 3
910 G. Candan and H.R. Yazgan

Table 13. A comparison among the proposed GA and others from the literature.

No Authors Cmax (operation times)

1 Lee and DiCesare (1994) 439


2 Kumar, Tiwari, and Shankar (2003) 420
3 Chan, Chung, and Chan (2005) 360
4 Leung et al. (2010) 380
5 Imran Ali Chaudhry (2012) 360
The proposed GA 350

Machine
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

M1 J51 J21 J53 J23 J13 J14

M2 J12 J52 J32 J43 J54 J34

M3
J11 J31 J41 J42 J22 J33 J44 J24
Cmax

0 40 70 100 120 170 180 190 200 230 260 270 300 320 330340 350

Figure 5. A Gantt from the proposed GAs result.

in Table 8. It was concluded that GA parameters could be chosen from statistical mode values. The found general
parameters were then tested on a benchmark problem. The results illustrate that the proposed approach produce better
approximation than the benchmark problem.

5. Conclusion
In this study, we aimed to dene an approach which provides a way to nd GA parameters instead of using a
trial-and-error method. Instead of full factorial experimental design (in our case resulting in 34,020 times) to determine
the best parameter combination of the GA, the Taguchi experimental design technique was employed, resulting in fewer
iterations, 3780. The main parameters such as derivation method, population size, crossover method, mutation rate and
crossover rate were chosen for tuning of the GA. The Taguchi L18 experimental design set was employed to determine
the Cmax value for a 42 problem set. The ANOVA was performed for each problem set to determine effective factors.
The percentages of effective factors amongst the 42 experiments were ordered to determine the best effective factors.
Finally, the most effective factors and their levels were employed in the proposed GA to compare against a bench-
mark problem from the literature. The results showed that the proposed GA algorithm produces better result than the
others.
One of the difculties of employing the GA approach is, however, that there is no specic way or method for tuning
the parameters. The best parameters may be found using trial-and-error methods or accepted values used by other
researchers. In this study, an approach was proposed to identify the parameters of a GA by employing the Taguchi
experimental design technique. The proposed approach can decrease the number of trials signicantly and with high
probability produce the best parameter set within a reasonable timeframe. Our proposed approach which aims to identify
GA parameters using the Taguchi experimental design method is one of the rst such attempts in the FMS environment.
Through Taguchi method, most effective parameters of processs can be found directly. According to parameters and
their levels number, number of experiment and proposed orthogonal array can be changed but all the time it is less than
full factorial designs experiment number. Taguchi method can decrease the experiment cost and time.
International Journal of Production Research 911

6. Discussion and further work


In many of engineering studies, there are lots of alternatives and realisation of all these alternative combinations are time
consuming. Full factorial investigation of any problem cannot be the best method in terms of cost and time limitation.
However, the proposed Taguchi experimental design set can be alternative for further investigation. Although the
method doesnt scan all solution space, optimum solution points may not be captured. But, it will reveal signicant
reductions on cost and time criteria.
It concluded that the GA parameter tuning problem can be achieved by the proposed approach instead of full facto-
rial design investigation attempt. In addition, most effective factors and their levels of the GA can be investigated to
solve scheduling problem among reasonable date-set. This may help us to reach some general tuning of parameters. Of
course, it can be a tool for tuning of GA parameters to other design problems.
Makespan is used as performance measure in this study. In addition to this study new performance measures such as
fuzzy tardiness will be preferred.

References
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Aggoune, R., H. A. Mahdi, and M. Portman. 2001. Genetic Algorithm for the Flow Shop Scheduling Problem with Availability
Constraints. IEEE International Conference on Systems, Man, and Cybernetics 4: 25462551.
Altparmak, F., B. Dengiz, and A. Smith. 2000. An Evolutionary Approach for Reliability Optimization in Fixed Topology Computer
Networks. Transactions on Operational Research 12 (12): 5775.
Bagheri, A., M. Zandieh, I. Mahdavi, and M. Yazdani. 2010. An Articial Immune Algorithm for the Flexible Job-shop Scheduling
Problem. Future Generation Computer Systems 26 (4): 533541.
Blum, C., and A. Roli. 2003. Metaheuristics in Combinatorial Optimization. ACM Computing Surveys 35 (3): 268308.
Ceran, G. 2006. Solving Hybrid Flow Shop Scheduling Problems by Using Data Mining and Genetic Algorithm. Masters thesis,
Selcuk University.
Chan, F. T. S., and H. K. Chan. 2004. A Comprehensive Survey and Future Trend of Simulation Study on FMS Scheduling.
Journal of Intelligent Manufacturing 15: 87102.
Chan, F. T. S., S. Chung, and P. Chan. 2005. An Introduction of Dominant Genes in Genetic Algorithm for Scheduling of FMS. In
Proceedings of the IEEE International Symposium on Intelligent Control. Limassol, Cyprus.
Chaudhry, I. 2012. A Genetic Algorithm Approach for Process Planning and Scheduling in Job Shop Environment. In Proceedings
of the World Congress on Engineering, Vol. III. London: WCE.
Cheng, B. W., and C. L. Chang. 2007. A Study on Flow Shop Scheduling Problem Combining Taguchi Experimental Design and
Genetic Algorithm. Expert Systems with Applications 32: 415421.
Cheng, E., and G. Wang. 2000. Single Machine Scheduling with Learning Effect Considerations. Annals of Operations Research
98: 273290.
Chiang, T. C., and L. C. Fu. 2004. Solving the FMS Scheduling Problem by Critical Ratio-based Heuristics and the Genetic
Algorithm. IEEE International Conference on Robotics and Automation 3: 31313336.
Choua, Fuh-Der, and Ching-En Lee. 1999. Two-machine Flowshop Scheduling with Bicriteria Problem. Computers & Industrial
Engineering 36 (3): 549564.
Chryssolouris, G. 2006. Manufacturing Systems: Theory and Practice. 2nd ed. New York: Springer-Verlag.
Chryssolouris, G., and V. Subramaniam. 2001. Dynamic Scheduling of Manufacturing Job Shops Using Genetic Algorithms.
Journal of Intelligent Manufacturing 12 (3): 281293.
Davidson, M. J., K. Balasubramanian, and G. Tagore. 2008. Experimental Investigation on Flow-forming of AA6061 Alloy a
Taguchi Approach. Journal of Materials Processing Technology 200: 283287.
Dodu, N., . Onural, and B. Cerit. 2004. A Study of Optimal Scheduling of Flexible Manufacturing Systems. Nigde University
Journal of Engineering Sciences 8 (1): 3139.
Du Plessis, B. J., and G. H. De Villiers. 2007. The Application of the Taguchi Method in the Evaluation of Mechanical Flotation in
Waste Activated Sludge Thickening. Resources, Conservation and Recycling 50: 202210.
Engin, O. 2001. To Increase the Performance of Flow-shop Scheduling Problems Solving with Genetic Algorithms: A Parameter
Optimization. PhD dissertation, Istanbul Technical University.
Garey, M. R., D. S. Johnson, and R. Sethi. 1976. The Complexity of Flowshop and Jobshop Scheduling. Mathematics of
Operations Research 1 (2): 117129.
Gen, M., J. Gao, and L. Lin. 2009. Multistage-based Genetic Algorithm for Flexible Job-shop Scheduling Problem. Studies in
Computational Intelligence 187: 183196. doi:10.1007/978-3-540-95978-6_13.
Glover, F. 1986. Future Paths for Integer Programming and Links to Articial Intelligence. Computers & Operational Research 13
(5): 533549.
Goldberg, D. 1989. Genetic Algorithms in Search, Optimization and Machine Learning. Boston, MA: Addison-Wesley.
912 G. Candan and H.R. Yazgan

Groover, M. P. 2001. Automation Production Systems and Computer ntegrated Manufacturing. 2nd ed. Upper Saddle River, NJ:
Prentice Hall.
Gltekin, H. 2006. Scheduling Flexible Robotic Manufacturing Cells. PhD thesis, Bilkent University.
Hmida, A., M. Haouari, M. Huguet, and P. Lopez. 2010. Discrepancy Search for the Flexible Job Shop Scheduling Problem.
Computers & Operations Research 37 (12): 21922201.
Ho, N., and J. Tay. 2004. GENACE: An Efcient Cultural Algorithm for Solving the Flexible Job Shop Problem. Evolutionary
Computation CEC Congress 2: 17591766. doi:10.1109/CEC.2004.1331108.
Hsu, T., O. Korbaa, R. Dupas, and G. Goncalves. 2008. Cyclic Scheduling for FMS: Modeling and Evolutionary Solving
Approach. European Journal of Operational Research 191: 464484.
Jansen, K., M. Mastrolilli, and R. Oba. 2005. Approximation Algorithms for Flexible Job Shop Problems. International Journal of
Foundations of Computer Science 16: 361379.
Jyothi, S. D. 2012. Scheduling Flexible Manufacturing System Using Petri-nets and Genetic Algorithm. Project Report Indian Institute
of Space Science and Technology, Thiruvananthapuram.
Kacem, I. 2003. Genetic Algorithm for the Flexible Job Shop Scheduling Problem. IEEE International Conference on Systems,
Man, and Cybernetics 4: 34643469.
Keung, K. W., W. H. Ip, and C. Y. Chan. 2001. An Enhanced MPS Solution for FMS using GAs. Integrated Manufacturing
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Systems 12 (5): 351359.


Khaw, J. F. C., B. S. Lim, and L. E. N. Lim. 1995. Optimal Design of Neural Networks Using the Taguchi Method. Neurocomput-
ing 7: 225245.
Kvak, T. 2014. Optimization of Surface Roughness and Flank Wear Using the Taguchi Method in Milling of Hadeld Steel with
PVD and CVD Coated Inserts. Measurement 50: 1928.
Kodeekha, E. 2004. A New Method of FMS Scheduling Using Optimization and Simulation. In Proceedings of the 16th European
Simulation Symposium, Budapest, 2934.
Kumar, R., M. Tiwari, and R. Shankar. 2003. Scheduling of Flexible Manufacturing Systems: An Ant Colony Optimization
Approach. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture 217:
14431453.
Lei, D. 2008. Solving Fuzzy Flexible Job Shop Scheduling Problems Using Genetic Algorithm. In International Conference on
Machine Learning and Cybernetics, Kunming.
Lei, D. 2010. A Genetic Algorithm for Flexible Job Shop Scheduling with Fuzzy Processing Time. International Journal of
Production Research 48 (10): 29953013.
Lei, D. 2012. Co-evolutionary Genetic Algorithm for Fuzzy Flexible Job Shop Scheduling. Applied Soft Computing 12 (8):
22372245.
Lee, D., and F. DiCesare. 1994. Scheduling Flexible Manufacturing Systems Using Petri Nets and Heuristic Search. IEEE
Transactons on Robotcs and Automaton 10 (2): 123132.
Lee, D., and Y. Kim. 1999. Scheduling Algorithms for Flexible Manufacturing Systems with Partially Grouped Machines. Journal
of Manufacturing Systems 18 (4): 301309.
Leung, C. W., T. N. Wong, K. L. Mak, and R. Y. K. Fung. 2010. Integrated Process Planning and Scheduling by an Agent-based
Ant Colony Optimization. Computers & Industrial Engineering 59: 166180.
Li, J., Q. Pan, and Y. Liang. 2010. An Effective Hybrid Tabu Search Algorithm for Multi-objective Flexible Job Shop Scheduling
Problems. Computers & Industrial Engineering 59 (4): 647662.
Lin, H., C. Su, C. Wang, B. Chang, and R. Juang. 2012. Parameter Optimization of Continuous Sputtering Process Based on
Taguchi Methods, Neural Networks, Desirability Function, and Genetic Algorithms. Expert Systems with Applications 39:
1291812925.
Lin, Y., Y. Chen, D. Wang, and H. Lee. 2009. Optimization of Machining Parameters in Magnetic Force Assisted EDM Based on
Taguchi Method. Journal of Materials Processing Technology 209: 33743383.
Liu, J. A., and B. L. MacCarthy. 1997. A Global MILP Model for FMS Scheduling. European Journal of Operational Research
100 (3): 441453.
Michalewicz, Z. 1996. Genetic Algorithms and Data Structures Evolution Programs. 3rd revised and extended ed, 216220. Berlin
Heidelberg: Springer-Verlag.
Miseviius, A., and B. Kilda. 2005. Comparison of Crossover Operators for the Quadratic Assignment Problem. Information
Technology and Control 34 (2): 109119.
Mohammad Pour, T., M. Yadollahi, and A. T. Haghighat. 2010. HPA-PN: A New Algorithm for Scheduling FMS using
Combinational Genetic Algorithm and Timed Petri Net. Second Int Conf Comput Modeling Simul. 3: 1518.
Moslehi, G., and M. Mahnam. 2011. A Pareto Approach to Multi-objective Flexible Job-Shop Scheduling Problem Using Particle
Swarm Optimization and Local Search. International Journal of Production Economics 129 (1): 1422.
Pezzella, F., G. Morganti, and G. Ciaschetti. 2008. A Genetic Algorithm for the Flexible Job-shop Scheduling Problem. Computers
& Operations Research 35: 32023212.
Rachamadagu, R., and K. Stecke 1993. Classication and Review of FMS Scheduling Procedures. Working paper 711. Division of
Research School of Business Administration, The University of Michigan.
International Journal of Production Research 913

Sankar, S. S., S. G. Ponnanbalam, and C. Rajendran. 2003. A Multiobjective Genetic Algorithm for Scheduling a Flexible
Manufacturing System. The International Journal of Advanced Manufacturing Technology 22: 229236.
Sharma, D., V. Singh, and C. Sharma. 2012. GA Based Scheduling of FMS Using Roulette Wheel Selection Process. Advances in
Intelligent and Soft Computing, AISC 131 (2): 931940.
Simsek, B., Y. T. , and E. H. Simsek. 2013. A TOPSIS-based Taguchi Optimization to Determine Optimal Mixture Proportions of
the High Strength Self-compacting concrete. Chemometrics and Intelligent Laboratory Systems 125: 1832.
Sivanandam, S., and S. Deepa. 2008. Introduction to Genetic Algorithms. Berlin Heidelberg: Springer. doi:10.1007/978-3-540-73190-0.
Sotskov, Yu. N., and N. V. Shakhlevich. 1995. NP-hardness of Shop-scheduling Problems with Three Jobs. Discrete Applied
Mathematics 59: 237266.
Spears, W. M. 1993. Crossover or Mutation? In Foundations of Genetic Algorithms 2, edited by L. D. Whitley, 221238. San
Mateo, CA: Morgan Kaufmann.
Taghavifard, M. T. A., M. B. Heydar, and S. S. Mousavi. 2009. A Genetic Algorithm for Scheduling Flexible Manufacturing Cells.
Journal of Applied Sciences 9 (1): 97104.
Taguchi, G., S. Chowdhury, and Y. Wu. 2005. Taguchis Quality Engineering Handbook. Hoboken, NJ: Wiley.
Talbi, El-Ghazali. 2009. Metaheuristics. New York: Wiley.
Tang, Ping-Hung, and Ming-Hseng Tseng. 2013. Adaptive Directed Mutation for Real-coded Genetic Algorithms. Applied Soft
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

Computing 13: 600614.


Trkcan, A. 1999. Solving of Job Loading, Scheduling and Management of Cutter Problems Together on Flexible Manufacturing
Systems. In National Congress on Operations Research and Industrial Engineering. Ankara: METU.
Wang, L., S. Wang, Y. Xu, G. Zhou, and M. Liu. 2012. A Bi-population Based Estimation of Distribution Algorithm for the Flexible
Job-shop Scheduling Problem. Computers & Industrial Engineering 62 (4): 917926.
Wu, Y., and A. Wu. 2000. Taguchi Methods for Robust Design. New York: The American Society of Mechanical Engineers.
Xing, L., Y. Chen, and K. Yang. 2009. An Efcient Search Method for Multi-Objective Flexible Job Shop Scheduling Problems.
Journal of Intelligent Manufacturing 20: 283293. doi:10.1007/s10845-008-0216-z.
Yang, W. H., and Y. S. Tarng. 1998. Design Optimization of Cutting Parameters for Turning Operations Based on Taguchi Method.
Journal of Materials Processing Technology 84: 122129.
Yazdani, M., M. Amiri, and M. Zandieh. 2010. Flexible Job-shop Scheduling with Parallel Variable Neighborhood Search
Algorithm. Expert Systems with Applications 37 (1): 678687.
Yih-fong, T. 2006. Parameter Design Optimisation of Computerised Numerical Control Turning Tool Steels for High Dimensional
Precision and Accuracy. Materials and Design 27: 665675.
Zakaria, Z., and S. Petrovic. 2012. Genetic Algorithms for Match-up Rescheduling of the Flexible Manufacturing Systems.
Computers & Industrial Engineering 62 (2): 670686.
Zhang, G., L. Gao, and Y. Shi. 2011. An Effective Genetic Algorithm for the Flexible Job-shop Scheduling Problem. Expert
Systems with Applications 38 (4): 35633573.
914 G. Candan and H.R. Yazgan

Appendx 1. Processing times (minute) of 20 jobs, 5 machines (with 5 operations)

Machines
Jobs/operations M1 M2 M3 M4 M5

J1/Op1 5 3 9 4 7
J1/Op2 7 6 2 8 4
J1/Op3 5 8 4 2 8
J1/Op4 8 5 10 9 7
J1/Op5 6 4 8 5 4

J2/Op1 2 7 5 9 2
J2/Op2 5 5 9 6 5
J2/Op3 2 6 5 6 2
J2/Op4 9 10 11 9 9
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

J2/Op5 5 6 4 6 5

J3/Op1 9 5 8 6 11
J3/Op2 6 1 12 9 3
J3/Op3 6 8 4 1 5
J3/Op4 9 3 5 9 9
J3/Op5 6 4 8 5 6

J4/Op1 5 6 11 7 5
J4/Op2 6 10 9 5 9
J4/Op3 4 6 16 6 5
J4/Op4 9 7 9 10 11
J4/Op5 7 5 5 6 4

J5/Op1 13 8 9 7 14
J5/Op2 8 3 7 5 2
J5/Op3 6 11 5 6 4
J5/Op4 7 10 6 10 6
J5/Op5 5 5 4 6 6

J6/Op1 6 13 5 9 11
J6/Op2 10 3 3 5 5
J6/Op3 5 9 8 2 6
J6/Op4 6 1 3 9 4
J6/Op5 5 6 4 5 5

J7/Op1 5 8 6 11 9
J7/Op2 1 12 9 3 4
J7/Op3 8 4 1 5 11
J7/Op4 3 5 9 9 10
J7/Op5 4 8 5 6 6

J8/Op1 4 8 4 2 8
J8/Op2 6 3 10 9 7
J8/Op3 6 4 8 13 4
J8/Op4 14 5 12 11 7
J8/Op5 2 1 2 8 4

J9/Op1 11 7 5 9 2
J9/Op2 9 5 9 6 5
J9/Op3 16 6 5 6 2
J9/Op4 9 10 11 9 9
J9/Op5 5 6 4 6 5

J10/Op1 12 6 8 9 2
J10/Op2 7 5 9 6 5

(Continued)
International Journal of Production Research 915

Appendix 1. (Continued)

Machines
Jobs/operations M1 M2 M3 M4 M5

J10/Op3 25 6 5 6 2
J10/Op4 9 10 11 9 9
J10/Op5 15 6 4 6 5

J11/Op1 6 11 9 2 4
J11/Op2 9 3 4 8 10
J11/Op3 1 5 11 6 8
J11/Op4 9 9 10 9 4
J11/Op5 5 6 6 14 2

J12/Op1 6 4 10 9 7
J12/Op2 6 5 8 13 4
Downloaded by [University of Nebraska, Lincoln] at 16:45 03 February 2015

J12/Op3 2 8 10 6 9
J12/Op4 9 7 8 7 4
J12/Op5 5 6 6 14 2

J13/Op1 6 4 2 8 10
J13/Op2 4 10 9 7 8
J13/Op3 6 8 9 2 12
J13/Op4 6 4 2 8 10
J13/Op5 5 2 9 4 12

J14/Op1 7 17 9 5 8
J14/Op2 5 12 6 1 12
J14/Op3 6 10 6 8 4
J14/Op4 10 8 9 3 5
J14/Op5 6 6 6 4 8

J15/Op1 6 8 9 2 9
J15/Op2 5 9 6 5 6
J15/Op3 6 5 6 2 6
J15/Op4 10 11 9 9 9
J15/Op5 6 4 6 5 6

J16/Op1 17 9 5 8 6
J16/Op2 12 6 1 12 9
J16/Op3 10 6 8 4 1
J16/Op4 8 9 3 5 9
J16/Op5 6 6 4 8 5

J17/Op1 8 6 11 9 2
J17/Op2 12 9 3 4 8
J17/Op3 4 1 5 11 6
J17/Op4 5 9 9 10 9
J17/Op5 8 5 6 6 14

J18/Op1 12 6 8 9 2
J18/Op2 7 5 9 6 5
J18/Op3 25 6 5 6 2
J18/Op4 9 10 11 9 9
J18/Op5 15 6 4 6 5

J19/Op1 12 6 8 9 2
J19/Op2 7 5 9 6 5
J19/Op3 25 6 5 6 2
J19/Op4 9 10 11 9 9
J19/Op5 15 6 4 6 5

J20/Op1 6 5 6 2 6
J20/Op2 10 11 9 9 9
J20/Op3 6 4 6 5 6
J20/Op4 7 5 9 6 5
J20/Op5 25 6 5 6 2

Вам также может понравиться