You are on page 1of 10

Int J Adv Manuf Technol (2007) 35:541–550

DOI 10.1007/s00170-007-1048-2

ORIGINAL ARTICLE

Using ant colony optimization to solve hybrid flow shop


scheduling problems
Kemal Alaykýran & Orhan Engin & Alper Döyen

Received: 27 November 2006 / Accepted: 11 April 2007 / Published online: 24 May 2007
# Springer-Verlag London Limited 2007

Abstract In recent years, most researchers have focused on the end of this study, there will be a comparison of the
methods which mimic natural processes in problem solving. performance of the proposed method presented in this paper
These methods are most commonly termed “nature- and the branch and bound (B&B) method presented by
inspired” methods. Ant colony optimization (ACO) is a Neron et al. (Omega 29(6):501–511, 2001). The results
new and encouraging group of these algorithms. The ant show that the improved ACO method is an effective and
system (AS) is the first algorithm of ACO. In this study, an efficient method for solving HFS problems.
improved ACO method is used to solve hybrid flow shop
(HFS) problems. The n-job and k-stage HFS problem is one Keywords Ant colony optimization . Improved ant system .
of the general production scheduling problems. HFS Hybrid flow shop scheduling
problems are NP-hard when the objective is to minimize
the makespan [1]. This research deals with the criterion of
makespan minimization for HFS scheduling problems. The 1 Introduction
operating parameters of AS have an important role on the
quality of the solution. In order to achieve better results, a In a hybrid flow shop (HFS), machines are arranged into s
parameter optimization study is conducted in this paper. stages in series. In each stage k (k=1,..., s), there are mk
The improved ACO method is tested with benchmark identical machines in parallel. Job j (j=1,..., n) has to be
problems. The test problems are the same as those used by processed on any one machine at each stage. Job j has finite
Carlier and Neron (RAIRO-RO 34(1):1–25, 2000), Neron processing times at each stage (p1j, p2j,..., psj). Preemption
et al. (Omega 29(6):501–511, 2001), and Engin and Döyen is not allowed and each machine can process at most one
(Future Gener Comput Syst 20(6):1083–1095, 2004). At operation at a time. The objective is to minimize the
makespan. The hybrid flow shop problems have been
proved to be NP-hard [5] when the objective is to minimize
the makespan in the case of max(M(1), M(2))>1, where M(l)
K. Alaykýran is the number of machines at stage l.
Department of Industrial Engineering, Gazi University,
Ankara, Turkey
The branch and bound (B&B) method and heuristics are
e-mail: kalaykiran@yahoo.com the most often used methods to solve HFS problems
because of the NP-hard nature of the problem.
O. Engin (*) A previous study [6] which uses B&B as a solving
Department of Industrial Engineering, Faculty of Engineering,
Selçuk University (Alladdin Keykubat Kampüsü Selçuklu),
method shows that the minimum makespan is attained at
Konya 42079, Turkey least for one member of the set of schedules, called
e-mail: orhanengin@yahoo.com preferred schedules. Preferred schedules ensure less search-
ing effort. Later, a three-part B&B algorithm [7], which
A. Döyen
Department of Industrial Engineering, Boğaziçi University,
consists of lower bound (LB) calculation, branching, and
Istanbul, Turkey node elimination parts, has been presented. The LB values
e-mail: alper.doyen@boun.edu.tr found in the study have been improved by a genetic
542 Int J Adv Manuf Technol (2007) 35:541–550

Table 1 The ant colony optimization (ACO) studies in the literature

Problem Author(s) Year Algorithm

Traveling salesman problem Dorigo et al. [37] 1996 AS


Gambardella and Dorigo [38] 1995 Ant-Q
Dorigo and Gambardella [39] 1997 ACS
Stützle and Hoos [31] 1998 MMAS
Bullnheimer et al. [40] 1997 ASrank
Quadratic assignment Stützle and Hoos [31] 1998 MMAS-QAP
Maniezzo [41] 1998 ANTS-QAP
Maniezzo and Colorni [42] 1999 AS-QAP
Gambardella et al. [43] 1999 HAS-QAP
Vehicle routing Bullnheimer et al. [44] 1997 AS-VRP
Gambardella et al. [45] 1999 MACS-VRPTW
Adaptive ant net Schoonderwoerd et al. [46] 1996 ABC
White et al. [47] 2000 ASGA
Di Caro and Dorigo [48] 1998 AntNet-FS
Bonabeau et al. [49] 1998 ABC-smart ants
Non-adaptive ant net Di Caro and Dorigo [50] 1997 Ant Net
Subramanian et al. [51] 1997 ACS
Heusse et al. [52] 1998 CAF
van der Put and Rothkrantz [53] 1998 ABC-backward
Graphic painting Costa and Hertz [54] 1997 ANTCOL
Sequential ordering Gambardella and Dorigo [55] 1997 HAS-SOP
Discrete optimization Dorigo et al. [56] 1998 Ant
Job shop scheduling Colorni et al. [57] 1994 AS-JSP
Ventresca and Ombuki [58] 2004
Zhang et al. [59] 2004
Group shop scheduling Blum [60] 2002 Max-Min AS
Open shop scheduling Blum [61] 2005 Beam-ACO
Permutation flow shop scheduling Rajendran and Ziegler [62] 2004 Ant-Colony
Ying and Liao [35] 2004 Ant-Colony
Shyu et al. [34] 2004 Ant-Colony
Rajendran and Ziegler [63] 2005 Ant-Colony
Gajpal and Rajendran [64] 2006 Ant-Colony
Single-machine scheduling with total tardiness criteria den Besten et al. [65] 2003 ACO
Keyboard arrangement problem Eggers et al. [66] 2003 ACO
Maximal constraint ensuring problems Roli et al. [67] 2001 ACO
Resource-constrained project scheduling problem (RCPSP) Bautista and Pereira [68] 2002 ACO
Merkle et al. [69] 2002 RCPSP
Luo et al. [70] 2003 Ant-Colony
Layout problem Solimanpur et al. [71] 2004 Ant-Colony
McKendall and Shang [72] 2006 Hybrid ant systems

algorithm (GA) [8]. Here, several heuristics have been study [11] has reduced the initial gap between upper and
introduced to compute an initial upper bound (UB), and lower bounds to half in a few minutes of running time.
then GA is used to improve that UB. Global LBs for Two polynomial bounded heuristic algorithms were
makespan (Cmax) minimization when it is not the case to developed by Gupta and Tunç [12] to find an acceptable
find optimal values have been proposed by Santos et al. [9]. solution for makespan minimization. It could solve prob-
In an effort to restrict the size of the search tree for a B&B lems with only two stages, where the first stage had only
method, a branching rule has been adopted, and 13 heuristic one machine and the second stage allowed multiple
methods have been used to generate an initial lower bound identical machines. Additionally, four heuristic algorithms
[10]. To increase the efficiency of B&B procedures, tests and have been presented to minimize makespan for the problem
time-bound adjustments based on energetic reasoning and with separable setup and removal times [12]. As a
global operations have been implemented [3]. A previous comparative study, three heuristic algorithms have been
Int J Adv Manuf Technol (2007) 35:541–550 543

Fig. 1 Flow chart for the im-


proved ant colony optimization START heuristic methods have been proposed for this problem. The
(ACO) local search methods, which are used to solve flow shop
Initialize the parameters and job scheduling, were adapted for HFS problems by
Set k = 0
Negenmann [17]. After comparison, it was found that the
variable-depth and tabu search methods performed better
For each operation find
the average of idle waiting
than the other algorithms. A tabu search algorithm for an
times HFS problem with limited buffer capacities [18] has been
proposed. The HFS problem is solved by a proposed
artificial immune system (AIS) [4] algorithm. The results
Get a probability value have been compared with the results found in [3], which
have used B&B for the problems. It is shown that AIS is an
efficient and effective problem solving technique for HFS
For each ant determine a
problems. In the last few years, HFS problem studies have
random beginning job been focused on studies with maintenance constraints [19,
20]. More works related to HFS problems can be found in
[10, 21–26].
Determine the next node In this study, we present an improved ant system (AS)
and probability value
algorithm which uses the same formula as classical AS, but
with a different starting solution procedure. The character-
istics of ant colony optimization (ACO) and the formula of
Update pheromone trails
k = k+1 AS are summarized in Sect. 2. In Sect. 3, the improved
algorithm is proposed. Also, the parameter optimizations
accomplished on HFS problems is shown. In Sect. 4, we
k < topt;
present an extensive computational study with the proposed
Yes algorithm. Experiments and the computational results are
No presented in Sect. 5. Finally, we conclude the paper with
some comments.
STOP

2 Ant colony optimization

presented and their results have been compared. The first Ant colony optimization (ACO) algorithms are multi-agent
two heuristic algorithms have applied the rules used by systems in which the behavior of each single agent, called
Campbell et al. [13] and the third algorithm was an an artificial ant, is inspired by the behavior of real ants
extension of a discussion presented by Gupta [14]. Another [27]. ACO algorithms are one of the most successful
comparative study [15] tests five different heuristics for examples of swarm intelligent systems [28], and have been
their performances of makespan and mean flow time applied to many classes of problems, ranging from the
criteria. The effects of problem characteristics and the classical traveling salesman problem to scheduling prob-
performance of heuristics have been examined by regres- lems. Ants are able to find good solutions to shortest path
sion analysis. The results have shown that structural problems between a food source and their home colony.
characteristics of the problems have a greater effect than ACO algorithms are inspired by the study of Charon and
the type of the employed heuristics on performance. Some Hudry [29] on a colony of Argentine ants to understand
of the researchers have studied a real-world problem that how ants can find the shortest path from their nest to the
has a particular structure (one machine in the first and third food source. The emergence of this shortest path selection
stages and two dedicated machines in stage two) [16]. Two behavior can be explained in terms of autocatalysis

Table 2 Parameter levels used


in the parameter optimization Parameter/level 1 2 3 4 5 6 7 8 9 10 11

α 0 1 2 3 4 5 6 7 8 9 10
β 0 1 2 3 4 5 6 7 8 9 10
ρ 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
544 Int J Adv Manuf Technol (2007) 35:541–550

Table 3 The best parameter set for the hybrid flow shop (HFS) (positive feedback) and differential path length, and it is
problem
made possible by an indirect form of communication,
α β ρ known as stigmergy [27]. It was found that the media used
to communicate among individuals’ information regarding
4 6 0.5 paths and used to decide where to go consists of pheromone

Table 4 Solutions to the HFS problems

Problem LB Best Cmax with improved AS % difference of improved AS Best Cmax with B&B % difference of B&B

j10c5a2 88 88 0 88 0
j10c5a3 117 117 0 117 0
j10c5a4 121 121 0 121 0
j10c5a5 122 124 1.6 122 0
j10c5a6 110 110 0 110 0
j10c5b1 130 131 0.7 130 0
j10c5b2 107 107 0 107 0
j10c5b3 109 109 0 109 0
j10c5b4 122 124 1.6 122 0
j10c5b5 153 153 0 153 0
j10c5b6 115 115 0 115 0
j10c5c1 68 68 0 68 0
j10c5c2 74 76 2 74 0
j10c5c3 71 72 1 71 0
j10c5c4 66 66 0 66 0
j10c5c5 78 78 0 78 0
j10c5c6 69 69 0 69 0
j10c10b1 163 163 0 163 0
j10c10b2 157 157 0 157 0
j10c10b3 169 169 0 169 0
j10c10b4 159 159 0 159 0
j10c10b5 165 165 0 165 0
j10c10b6 165 165 0 165 0
j10c10c1 113 118 4 127 12.4
j10c10c2 116 117 0.8 116 0
j10c10c3 98 108 10 133 35.7
j10c10c4 103 112 8 135 31.1
j10c10c5 121 126 4 145 19.8
j10c10c6 97 102 5 112 15.5
j15c5a1 178 178 0 178 0
j15c5a2 165 165 0 165 0
j15c5a3 130 132 1 130 0
j15c5a4 156 156 0 156 0
j15c5a5 164 166 1 164 0
j15c5a6 178 178 0 178 0
j15c5b1 170 170 0 170 0
j15c5b2 152 152 0 152 0
j15c5b3 157 157 0 157 0
j15c5b4 147 149 1 147 0
j15c5b5 166 166 0 166 0
j15c5b6 175 176 0.5 175 0
j15c5c1 85 85 0 85 0
j15c5c2 90 90 0 90 0
j15c5c3 87 87 0 87 0
j15c5c4 89 89 0 90 1.1
j15c5c5 73 73 0 84 15.1
j15c5c6 91 91 0 91 0
j15c5d1 167 167 0 167 0
Int J Adv Manuf Technol (2007) 35:541–550 545

Table 4 (continued)

Problem LB Best Cmax with improved AS % difference of improved AS Best Cmax with B&B % difference of B&B

j15c5d2 82 86 4 85 3.7
j15c5d3 77 83 7 96 24.7
j15c5d4 61 84 37 101 65.6
j15c5d5 67 80 19 97 44.8
j15c5d6 79 79 0 87 10.1
j15c10a1 236 236 0 236 0
j15c10a2 200 200 0 200 0
j15c10a3 198 198 0 198 0
j15c10a4 225 228 1 225 0
j15c10a5 182 182 0 183 0.5
j15c10a6 200 200 0 200 0
j15c10b1 222 222 0 222 0
j15c10b2 187 188 0.5 187 0
j15c10b3 222 224 0.9 222 0
j15c10b4 221 221 0 221 0

trails [30]. This system works as follows: a moving ant The first and most common algorithm of ACO is the ant
drops some pheromone on the ground in varying quantities, system (AS) algorithm, which was proposed by Dorigo et
related to the quality and quantity of the food source; in this al. [36]. The formula of AS is as follows:
way, the track becomes marked. The following ants can  a  b
detect this substance and they make their choice due to the t ij ðt Þ  hij
Pijk ðt Þ ¼ P ð1Þ
intensity of the pheromone trail. In ACO algorithms, this ½t il ðt Þa  ½hil b
natural process is used to solve combinatorial optimization l2Nik
problems. In recent years, so many papers have been
published about ACO algorithms and so many new and where:
robust algorithms have been developed. In the literature, there
t ij(t) The intensity of the pheromone trail on the path
is a number of studies on scheduling problems by ACO. Some
between i and j at time t
of these studies which may be considered as the milestones of
ηij The heuristic value 1/dij (known as visibility), where
ACO literature on scheduling problems include: Stützle and
dij is the physical distance between i and j
Hoos [31] dealing with permutation flow shop scheduling
α The relative importance of the pheromone trail
problems with the objective of minimizing the makespan;
β The relative importance of visibility
Merkle and Middendorf [32] considering the permutation
  k
scheduling problem; Chandrasekharan and Ziegler [33] Q L ðt Þ
considering the problem of scheduling in permutation flow Δt ij ðt Þ ¼
k
ð2Þ
0
shops by using ACO algorithms with the objective of
minimizing the sum of the total flow time of jobs; Shyu where:
et al. [34] considering the no-wait flow shop scheduling Lk(t) The length of the tour found by ant k
problem to minimize the total completion time with ACO; Q A constant
and Ying and Liao [35] considering the permutation flow
The pheromone updating rule [34] is as follows:
shop sequencing problem to minimize the makespan with
ACO. X
n
t ij ¼ ð1  pÞt ij þ Δt kij ð3Þ
k¼1

Table 5 General performances of the two methods


where:
p The evaporation parameter (0<p<1)
Method % solved % difference
n The number of distributed ants and also the number of jobs
Improved AS 0.65 1.771
After consideration, many other algorithms have been
B&B 0.81 4.446
proposed by other researchers. These include the MAX-
546 Int J Adv Manuf Technol (2007) 35:541–550

Table 6 Percentage differences (from lower bounds, LBs) due to the which is used to solve the HFS problems. The improved AS
problem sizes
algorithm is presented in the next section.
Problem type 10×5 10×10 15×5 15×10
3 The improved AS algorithm
Percentage difference 0.406 2.65 2.9375 2.4

3.1 The proposed algorithm


MMIN Ant System (MMAS), Ant Colony System (ACS),
and ASrank. The main difference between these algorithms The improved AS algorithm uses the same formula as AS,
occurs in the pheromone updating mechanism. As seen in described in Sect. 2, but it uses a starting solution that
Eq. 3, in AS, all ants add pheromone, but in MMAS, only affects the probability function and, therefore, the quality of
the best or set a number of ants add pheromone to the trail. the solution. The first phase of the method introduced here
In addition, as seen in Eq. 2, in AS, there is no limit for is the weighting phase. The steps of weighting are as
adding pheromone, but in MMAS, there is a minimum and follows:
a maximum value for pheromone trail addition, which is Step 1 Setting initial conditions:
determined by the problem solver. ACS is the most similar
to AS, but the difference between these two algorithms can topt: the variable that shows the total operation
be seen in pheromone updating. number:
The ACO algorithm has been used to solve different i=from 1 to topt
combinatorial optimization problems in the last 15 years. j=from 1 to topt
The problem types and the ACO studies on that problem
are summarized in Table 1. For each (i, j) operation pairs, calculate the average
As seen in Table 1, ACO has been used in solving flow, of the idle machine waiting times when the pairs
job shop, and single-machine type scheduling problems in are consecutive
the past 10 years. Find the average of the calculated idle waiting
HFS scheduling combines the properties of flow shop times:
and parallel machine scheduling problems. And HFS is i=from 1 to topt
known as NP-hard in the strong sense, even if there is only j=from 1 to topt
one machine on the first stage and two machines on the
second stage [5]. Let ji be the average idle waiting time if operation i
HFS is more complicated and harder than flow shop precedes operation j
problems, and we aimed to investigate the performance of Let ij be the average idle waiting time if operation j
ACO algorithms for a more complex problem than classical precedes operation i
flow shops. For that reason, we solved various types of Assign 0.5*ji+0.5*ij value to (i−j) node:
HFS problems in this paper. To the best of our knowledge, i=from 1 to topt
this is the first attempt to solve the HFS and also the j=from 1 to topt
parallel machine scheduling problems by ACO algorithms
in the literature. In the study, the improved AS algorithm Get a probability value by dividing the value of
was improved by using a different heuristic value (ηij), each (i−j) node by the total value

Fig. 2 Percentage differences 3.50


(from lower bounds, LBs) due
Percentage Difference

3.00
to the problem sizes
2.50
2.00
Problem type
1.50
1.00
0.50
0.00
10x5 10x10 15x5 15x10

Problem type
Int J Adv Manuf Technol (2007) 35:541–550 547

Table 7 Percentage differences (from LBs) due to the machine layout the pheromone trail (α), the relative importance of visibility
type
(β), and the evaporation parameter (ρ). In order to achieve
Layout type a b c d better results, a parameter optimization study is conducted in
this investigation for the α, β, and ρ parameters. The number
Percentage difference 0.27 0.24 1.93 11.17 of ants and the number of tours are assumed to be 50 and
100, respectively, as in the literature.
The α and β parameters are selected in the interval [1,
10]. Eleven different values are determined from 0 to 10 by
Step 2 Solving the problem: increasing by one at each level. Parameter ρ is selected in
the interval [0, 1] by increasing by 0.1 at each of the levels.
Determine the tour number
The implementation of parameter optimization is full
Determine the ant number
factorial, thus, not all but four of Carlier and Neron’s [2]
Determine the α parameter
HFS problems are considered. These are the j15c5b1,
Determine the β parameter
j15c5d1, j15c5a2, and j15c5c5 problems. The parameter
Determine the ρ parameter
levels are shown in Table 2.
For each ant, determine a random beginning job: For each parameter set, 25 trials were handled, therefore,
For each tour: for each of the problems, a total of 11*11*11*25=33,275
runs were made. The average tours and the shortest tour
For each ant: values were evaluated.
Determine the next node by using Eq. 1 after The best parameter set found after parameter optimiza-
calculating the ηij value from Step 1 tion is shown in Table 3. These parameters were used in the
Determine the probability value solving of all of the test problems.
Carry out the pheromone updating by using Eqs. 2
and 3
5 Solutions to the HFS problems
Stop
The summarized flow chart for the improved ACO In order to evaluate the success of the algorithm on HFS
method is given in Fig. 1. problems, it was been run on 63 different benchmark
problems taken from Carlier and Neron [2]. The problem
sizes vary from 10 jobs×5 stages to 15 jobs×10 stages. The
processing times of the jobs have a uniform distribution in
4 Parameter test on HFS problems the range (3, 20). Three characteristics that define a problem
are the number of jobs, the number of stages, and the number
The operating parameters of AS play an important role in the of identical machines at each stage. For example, the notation
quality of the solution. The parameters of ACO include the of j15c10b1 means a 15-job, 10-stage problem. The letters j
number of ants, number of tours, the relative importance of and c are abbreviations for job and stage, respectively. The

Fig. 3 Percentage differences 12.00


(from LBs) due to the machine
Percentage Differences

layout type 10.00

8.00

6.00 Problem types

4.00

2.00

0.00
a b c d

Machine Layouts
548 Int J Adv Manuf Technol (2007) 35:541–550

letter b defines the structure of the machine layout at the 6 Conclusions


stages. The last number 1 is the problem index for a specific
type. The meanings of the letters for machine layouts are In this study, a nature-inspired metaheuristic named ant
given below: colony optimization (ACO) was introduced and a new
method which might be used to improve the performance of
a There is one machine at the middle stage (bottleneck)
the first ACO algorithm, ant system (AS), was proposed.
and three machines at the other stages
Sixty-three different hybrid flow shop (HFS) problems [2]
b There is one machine at the first stage (bottleneck) and
were solved and satisfactory results were found with the
three machines at the other stages
improved AS method. For all of the problems, a 1.771%
c There are two machines at the middle stage (bottleneck)
difference from the lower bounds (LBs) was achieved,
and three machines at the other stages
which was much better than the performance of the branch
d There are three machines at each stage (no bottleneck)
and bound (B&B) method (which was 4.446%). Here, it
The same problems were also studied by Neron et al. [3]. was concluded that, due to the limited run time (1,600 s) for
The solutions found using the improved AS algorithm are both of the problems, the improved AS converges to the
compared to the solutions of the branch and bound (B&B) optimum much faster than the B&B method, but B&B
method given by Neron et al. [3]. Both of the algorithms reached the LB values for 81% of the problems, while the
are limited to a run time of 1,600 s. If an optimal solution improved AS did not perform as well (namely, 65%). Carlier
was not found within this time, the search was stopped and and Neron [2] classified the c and d types of 10 jobs×5
the best solution was accepted as the final schedule. The stages to 15 jobs×10 stages problems as hard problems. For
performance of the algorithm is evaluated according to hard problems, the improved AS outperformed the consid-
Eq. 4, which calculates the percentage difference of the ered B&B method. Improved AS solved 67% of the hard
solutions from the lower bounds. The lower bounds are problems, while B&B solved only 61% of them.
found in [3]: The inspiration of nature in problem solving seems to be
increasing its impact on researchers in future due to its
encouraging performance and adaptability to other prob-
Best Cmax value with improved AS  Lower Bound ðLBÞ
%Difference ¼ lems. Better results may be achieved by utilizing hybrid or
Lower Bound ðLBÞ
parallel applications and also be fine tuning the parameters
ð4Þ of these problem solving methods. The proposed method
may be used in hybrid with other metaheuristics, such as
In Table 4, the best Cmax values found by our improved genetic algorithms, simulated annealing, or tabu search.
AS algorithm and the B&B methods of Carlier and Neron Additionally, the results found here may be compared to
[2] are presented. Also, the percentage differences from the other metaheuristics. We have solved the same problems by
LBs for each of the two methods are shown. using artificial immune systems (AIS) [4]. The results
All of the results were calculated using a computer with an showed AIS to be more successful than B&B or improved
Intel Celeron 1500 microprocessor and 256 MB of RAM. AS by far.
The program was created using Microsoft Visual Basic 6.0.
In all of the solutions, the number of ants and the number of Acknowledgments The authors would like to thank Jacques Carlier
tours are assumed to be 5 and 2,000, respectively. and Emmanuel Neron for the benchmark problems, solution files, and
The average difference from the LBs, when all of the 63 all of their help, and are also grateful to the anonymous reviewers for
problems are considered, is calculated as 1.771% for the their helpful comments and suggestions.
improved AS. However, this statistic is calculated as
4.446% for the B&B method. The lower bound values are
achieved for 41 of the problems, while the lower bounds
are achieved for 50 of the problems with B&B. In Table 5, References
the percentage of problems (with respect to all 63
problems) which are solved to their LBs (%solved) and 1. Linn R, Zhang W (1999) Hybrid flow shop scheduling: a survey.
Comput Ind Eng 37(1):57–61
the average percentage differences (%difference) from the
2. Carlier J, Neron E (2000) An exact method for solving the
LBs for each method are shown. The percentage differences multiprocessor flow-shop. RAIRO-RO 34(1):1–25
(from LBs) due to the problem sizes are given in Table 6 3. Neron E, Baptiste P, Gupta JND (2001) Solving hybrid flow shop
and are represented in Fig. 2. problem using energetic reasoning and global operations. Omega
29(6):501–511
In Table 7, the values of the percentage differences (from
4. Engin O, Döyen A (2004) A new approach to solve hybrid flow
LBs) due to the machine layout types are given. Figure 3 shop scheduling problems by artificial immune system. Future
represents these findings graphically. Gener Comput Syst 20(6):1083–1095
Int J Adv Manuf Technol (2007) 35:541–550 549

5. Gupta JND (1988) Two-stage, hybrid flowshop scheduling 30. Bevilacqua A (2002) A methodological approach to parallel
problem. J Oper Res Soc 39(4):359–364 simulated annealing on an SMP system. J Parallel Distrib Comput
6. Arthanari TS, Ramamurthy KG (1971) An extension of two 62(10):1548–1570
machines sequencing problem. Opsearch 8(1):10–22 31. Stützle T, Hoos HH (1998) Improvements on the ANT System:
7. Brah SA, Hunsucker JL (1991) Branch and bound algorithm for the introducing the MAX-MIN Ant System. In: RF Albrecht, GD
flow shop with multiple processors. Eur J Oper Res 51(1):88–99 Smith, NC Steele (Eds) Proceedings of the 3rd International
8. Portmann M-C, Vignier A, Dardilhac D, Dezalay D (1998) Conference on Artificial Neural Networks and Genetic Algorithms
Branch and bound crossed with GA to solve hybrid flowshops. (ICANNGA’97), Norwich, UK, April 1997, pp 245–249
Eur J Oper Res 107(2):389–400 32. Merkle D, Middendorf M (2001) A New approach to solve
9. Santos DL, Hunsucker JL, Deal DE (1995) Global lower bounds permutation scheduling problems with ant colony optimization.
for flow shops with multiple processors. Eur J Oper Res 80 Applications of evolutionary computing. In: Proceedings of
(1):112–120 EvoWorkshops 2001, Como, Italy, April 2001, pp 484–494
10. Gupta JND, Hariri AMA, Potts CN (1994) Scheduling a two-stage 33. Chandrasekharan R, Ziegler H (2005) Two ant-colony algorithms
hybrid flow shop with parallel machines at the first stage. Ann for minimizing total flowtime in permutation flowshops. Comput
Oper Res 69(0):171–191 Ind Eng 48(4):789–797
11. Moursli O, Pochet YA (2000) A branch-and-bound algorithm for 34. Shyu SJ, Lin BMT, Yin PY (2004) Application of ant colony
the hybrid flowshop. Int J Prod Econ 64(1):113–125 optimization for no-wait flowshop scheduling problem to mini-
12. Gupta JND, Tunç EA (1994) Scheduling a two-stage hybrid mize the total completion time. Comput Ind Eng 47(2–3):181–193
flowshop with separable setup and removal times. Eur J Oper Res 35. Ying K-C, Liao C-J (2004) An ant colony system for permutation
77(3):415–428 flow shop sequencing. Comput Oper Res 31(5):791–801
13. Campbell HG, Dudek RA, Smith ML (1970) A heuristic 36. Dorigo M, Maniezzo V, Colorni A (1991) Positive feedback as a
algorithm for the n-job, m-machine sequencing problem. Manage search strategy. Technical report 91–016, Dip. Elettronica,
Sci 16(10):B630–B637 Politecnico di Milano, Italy
14. Gupta JND (1971) A functional heuristic algorithm for the flow 37. Dorigo M, Maniezzo V, Colorni A (1996) Ant system: optimiza-
shop scheduling problem. Oper Res Q 22(1):39–47 tion by a colony of cooperating agents. IEEE Trans Syst Man
15. Brah SA, Loo LL (1999) Heuristics for scheduling in a flow shop Cybern B 26(1):29–41
with multiple processors. Eur J Oper Res 113(1):113–122 38. Gambardella LM, Dorigo M (1995) Ant-Q: a reinforcement
16. Riane F, Artiba A, Elmaghraby SE (1998) A hybrid three-stage learning approach to the traveling salesman problem. In: Proceed-
flowshop problem: efficient heuristics to minimize makespan. Eur ings of the 12th International Conference on Machine Learning
J Oper Res 109(2):321–329 (ML’96), Tahoe City, California, July 1995, pp 252–260
17. Negenmann EG (2001) Local search algorithms for the multipro- 39. Dorigo M, Gambardella LM (1997) Ant colonies for the traveling
cessor flow shop scheduling problem. Eur J Oper Res 128(1): salesman problem. Biosystems 43(1):73–81
147–158 40. Bullnheimer B, Hartl RF, Strauss C (1997) A new rank-based
18. Wardono B, Fathi YA (2004) Tabu search algorithm for the multi- version of the ant system: a computational study. Technical report
stage parallel machine problem with limited buffer capacities. Eur POM-03/97, Institute of Management Science, University of
J Oper Res 155(2):380–401 Vienna, Austria
19. Allaoui H, Artiba A (2004) Integrating simulation and optimiza- 41. Maniezzo V (1998) Exact and approximate nondeterministic tree-
tion to schedule a hybrid flow shop with maintenance constraints. search procedures for the quadratic assignment problem.
Comput Ind Eng 47(4):431–450 INFORMS J Comput 11(4):358–369
20. Allaoui H, Artiba A (2006) Scheduling two-stage hybrid flow 42. Maniezzo V, Colorni A (1999) The ant system applied to the
shop with availability constraints. Comput Ind Eng 33(5): quadratic assignment problem. IEEE Trans Knowl Data Eng 11
1399–1419 (5):769–778
21. Sahni SK (1976) Algorithms for scheduling independent tasks. 43. Gambardella LM, Taillard ED, Dorigo M (1999) Ant colonies
J ACM 23(1):116–127 for the quadratic assignment problem. J Oper Res Soc 50(2):
22. Narasimhan SL, Panwalker SS (1984) Scheduling in a two-stage 167–176
manufacturing process. Int J Prod Res 22(4):555–564 44. Bullnheimer B, Hartl RF, Strauss C (1997) Applying the ant
23. Dessouky MM, Dessouky MI, Verma S (1998) Flowshop system to the vehicle routing problem. In: Proceedings of the 2nd
scheduling with identical jobs and uniform parallel machines. International Conference on Metaheuristics (MIC’97), Sophia
Eur J Oper Res 109(3):620–631 Antipolis, France, July 1997
24. Verma S, Dessouky MM (1999) Multistage hybrid flowshop 45. Gambardella LM, Taillard E, Agazzi G (1999) MACS-VRPTW: a
scheduling with identical jobs and uniform parallel machines. multiple ant colony system for vehicle routing problems with time
J Sched 2(3):135–150 windows. Technical report, IDSIA, pp 1–17
25. Schuurman P, Woeginger GJ (2000) A polynomial time approx- 46. Schoonderwoerd R, Holland OE, Bruten JL, Rothkrantz LJM
imation scheme for the two-stage multiprocessor flow shop (1996) Ant-based load balancing in telecommunications networks.
problem. Theor Comp Sci 237(1–2):105–122 Adapt Behav 5(2):169–207
26. Ohno K, Jin ZH, Ito T, Elmaghraby SE (2002) Scheduling hybrid 47. White T, Pagurek B, Oppacher F (2000) ASGA: improving the ant
flowshops in printed circuit board assembly lines. Prod Oper system by integration with genetic algorithms. Systems and
Manag 11(2):216–230 Computer Engineering Department, Carleton University Press
27. Choi I-C, Kim S-I, Kim H-S ( 2003) A genetic algorithm with a 48. Di Caro G, Dorigo M (1998) Extending AntNet for best-effort
mixed region search for the asymmetric traveling salesman quality-of-service routing. In: Proceedings of the 1st International
problem. Comput Oper Res 30(5):773–786 Workshop on Ant Colony Optimization (ANT’98), From Ant
28. Ulder NLJ, Aarts EHL, Bandelt H-J, van Laarhoven PJM, Pesch E Colonies to Artificial Ants, Brussels, Belgium, October 1998
(1990) Genetic local search algorithms for the travelling salesman 49. Bonabeau E, Henaux F, Guérin S, Snyers D, Kuntz P, Theraulaz G
problem. Lect Notes Comput Sci 496:109–119 (1998) Routing in telecommunication networks with “Smart” ant-
29. Charon I, Hudry O (2000) Application of the noising method to like agents. In: Proceedings of the 2nd International Workshop on
the travelling salesman problem. Eur J Oper Res 125(2):266–277 Intelligent Agents for Telecommunication Applications (IATA’98),
550 Int J Adv Manuf Technol (2007) 35:541–550

Paris, France, July 1998, Lectures Notes in Computer Science, vol 62. Rajendran C, Ziegler H (2004) Ant-colony algorithms for
1437 permutation flowshop scheduling to minimize makespan/total
50. Di Caro G, Dorigo M (1997) AntNet: a mobile agents approach to flow time of jobs. Eur J Oper Res 155(2):426–438
adaptive routing. Technical report IRIDIA/97-12, IRIDIA, Uni- 63. Rajendran C, Ziegler H (2005) Two ant-colony algorithms for
versite Libre de Bruxelles, Belgium minimizing total flowtime in permutation flowshops. Comput Ind
51. Subramanian D, Druschel P, Chen J (1997) Ants and reinforcement Eng 48(4):789–797
learning: a case study in routing in dynamic networks. In: 64. Gajpal Y, Rajendran C (2006) An ant colony optimization
Proceedings of the 15th International Joint Conference on Artificial algorithm for minimizing the completion-time variance of jobs
Intelligence (IJCAI’97), Nagoya, Japan, August 1997, pp 832–838 in flow shops. Int J Prod Econ 101(2):259–273
52. Heusse M, Snyers D, Guérin S, Kuntz P (1998) Adaptive agent- 65. den Beste M, Stützle T, Dorigo M (2000) Ant colony optimization
driven routing and load balancing in communication networks. for the total weighted tardiness problem. In: Proceedings of the
Adv Complex Syst 1(2–3):237–254 6th International Conference on Parallel Problem Solving from
53. van der Put R, Rothkrantz L (1998) Routing in packet switched Nature (PPSN 2000), Paris, France, September 2000
networks using agents. Simulat Pract Theory (in press) 66. Eggers J, Feillet D, Kehl S, Wagner MO, Yannou B (2003)
54. Costa D, Hertz A (1997) Ants can colour graphs. J Oper Res Soc Optimization of the keyboard arrangement problem using an Ant
48(3):295–305 Colony algorithm. Eur J Oper Res 148(3):672–686
55. Gambardella LM, Dorigo M (1997) HAS-SOP: hybrid ant system 67. Roli A, Blum C, Dorigo M (2001) ACO for maximal constraint
for the sequential ordering problem. Technical report IDSIA-11– satisfaction problems. In: Proceedings of the 4th Metaheuristics
97, IDSIA, Lugano, Switzerland International Conference (MIC 2001), Porto, Portugal, July 2001,
56. Dorigo M, Di Caro G, Gambardella LM (1998) Ant algorithms for pp 187–191
discrete optimization, Technical report, IDSIA, Lugano, Switzerland 68. Bautista J, Pereira J (2002) Ant colonies for RCPS problem. In:
57. Colorni A, Dorigo M, Maniezzo V, Trubian M (1994) Ant system Proceedings of the 5th Catalonian Conference on AI (CCIA
for job-shop scheduling. Belg J Oper Res 34(1):39–53 2002), Topics in Artificial Intelligence, Castellón, Spain, October
58. Ventresca M, Ombuki B (2004) Ant colony optimization for job 2002, Lecture Notes in Artificial Intelligence, vol 2504, pp
shop scheduling problem. In: Proceedings of the 8th International 257–268
Conference on Artificial Intelligence and Soft Computing (ASC 69. Merkle D, Middendorf M, Schmeck, H (2002) Ant colony
2004), Marbella, Spain, September 2004, CDROM 451–152 optimization for resource-constrained project scheduling. IEEE
59. Zhang H, Gen M, Fujimura S, Kim KW (2004) Hybrid ant colony Trans Evol Comput 6(4):333–346
optimization for job shop scheduling problem. In: Proceedings of 70. Luo S, Wang C, Wang J (2003) Ant colony optimization for
the 20th Fuzzy System Symposium, Kitakyushu, Japan, June resource-constrained project scheduling with generalized prece-
2004, pp 304–305 dence relations. In: Proceedings of the 15th IEEE International
60. Blum C (2002) ACO applied to group shop scheduling: a case Conference on Tools with Artificial Intelligence (ICTAI 2003),
study on intensification and diversification. In: Dorigo M, Di Caro Sacramento, California, November 2003, pp 284–288
G, Sampels M (Eds) Proceedings of ANTS 2002, the 3rd 71. Solimanpur M, Vrat P, Shankar R (2004) Ant colony optimization
International Workshop on Ant Algorithms, Brussels, Belgium, algorithm to the inter-cell layout problem in cellular manufactur-
September 2002, Lecture Notes in Computer Science 2463/2002 ing. Eur J Oper Res 157(3):592–606
61. Blum C (2005) Beam-ACO—hybridizing ant colony optimization 72. McKendall AR Jr, Shang J (2006) Hybrid ant systems for the
with beam search: an application to open shop scheduling. dynamic facility layout problem. Comput Oper Res 33(3):
Comput Oper Res 32(6):1565–1591 790–803

You might also like