Professional Documents
Culture Documents
1 s2.0 S1568494617305422 Main PDF
1 s2.0 S1568494617305422 Main PDF
a r t i c l e i n f o a b s t r a c t
Article history: Multi-row facility layout problem (MRFLP) is a class of facility layout problems, which decides upon the
Received 18 December 2015 arrangement of facilities in some fixed numbers of rows in order to minimize material handling cost.
Received in revised form 3 August 2017 Nowadays, according to the new layout requirements, the facility layout problems (FLPs) have many
Accepted 3 September 2017
applications such as hospital layout, construction site layout planning and layout of logistics facilities.
Available online 8 September 2017
Therefore, we study an extended MRFLP, as a novel layout problem, with the following main assumptions:
1) the facilities are arranged in a two-dimensional area and without splitter rows, 2) multiple products
Keywords:
are available, 3) distance between each pair of facilities, due to inaccurate and flexible manufacturing
Facility layout
Multi-row facility layout problem (MRFLP)
processes and other limitations (such as WIPs, industrial instruments, transportation lines and etc.), is
Genetic algorithm considered as fuzzy number, and 4) the objective function is considered as minimizing the material han-
Lost opportunity cost dling and lost opportunity costs. To model these assumptions, a nonlinear mixed-integer programming
Fuzzy sets model with fuzzy constraints is presented and then converted to a linear mixed-integer programming
model. Since the developed model is an NP-hard problem, a genetic algorithm approach is suggested to
find the best solutions with a minimum cost function. Additionally, three different crossover methods
are compared in the proposed genetic algorithm and finally, a sensitivity analysis is performed to discuss
important parameters.
© 2017 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.asoc.2017.09.003
1568-4946/© 2017 Elsevier B.V. All rights reserved.
820 S. Safarzadeh, H. Koosha / Applied Soft Computing 61 (2017) 819–831
is lost opportunity cost related to waste spaces, which remains after solving methods. For example, Kothari and Ghosh [22] proposed
the facilities placement into the objective function of the proposed the Lin–Kernighan method, as a novel heuristic technique, for SRFLP
layout problem. Economically, any investment on an opportunity in which the lost opportunity cost had not been considered. In
causes losing other economic opportunities. For example, in layout another study for SRFLP, Amaral and Letchford [23] provided a poly-
design problems, the inappropriate layout of facilities in an area hedral approach based on the material handling cost. Moreover,
leads to losing some opportunities in that area such as locating a hybrid estimation of the distribution algorithm is proposed for
a new production line, making storage, renting the area and etc. SRFLP [24].
Therefore, considering the cost of lost opportunities with higher Researchers in the area of facility layout paid little attention
utilities is reasonable for a layout problem. Unfortunately, this issue to fuzzy layout environment [15,16,19]. For example, fuzzy tech-
has not been mentioned in the extant literature yet. niques could help to solve the unequal facility layout problems
(UA-FLPs) in a fuzzy random environment [17,20]. Moreover, qual-
1.1. Fuzzy approach in FLP itative and quantitative criteria were incorporated into FLP model
and then an integrated methodology, based on the synthetic value
Some important parameters such as demand or distance of fuzzy judgments and nonlinear programming, was provided [21].
between the facilities have been considered in the literature of the Some of the main parameters of layout problems such as demand
FLPs, usually, as a crisp number to calculate the handling cost and it (related to the handling cost), and closeness of departments (in the
has made the proposed models impractical. Because of some lim- multi-objective models) are considered as fuzzy numbers in these
itations in the real world such as required space for instruments, studies [6,20]. However, the investigated studies of FLPs have not
work in process, transportation facilities and etc., we cannot put the been dealt with the MRFLP in a fuzzy environment.
facilities in the optimal location. Furthermore, diversity of products Garey and Johnson [25] proved the high computational com-
forces the manufacturer into the flexible production methods and plexity of different layout design problems. Therefore, many
industrial machines that may lead to a partial change in the facility heuristics [26,27] and meta-heuristic algorithms have been pro-
layout. Thus, it is more realistic to consider the distance between posed in the previous literature to solve single and multi-row
each pair of facilities as fuzzy number as well as the recent studies layout design problems. For example, we can mention ant colony
in the FLPs [15–17]. By doing this, the existent inaccuracy of the optimization (ACO) algorithm [28–30], tabu search algorithm (TS)
distance is exerted in the model. [31–34], particle swarm optimization (PSO) algorithm and sim-
Fuzzy set theory which first introduced by Zadeh [15] explains ulated annealing (SA) [35–37] as some metaheuristics to solve
the inexact and ambiguous features of any measurable variable. SRFLP [38]. However, some researchers used exact optimization
In an uncertain environment, the fuzzy theory can help us to solve methods; for example, Solimanpur and Jafari [39] provided a two-
problems by considering inexact inputs for them and apply the spe- dimensional facility layout problem using a branch and bound
cial techniques, in accordance with the available limitations [18]. algorithm.
In this paper, we focus on an extension of MRFLP in a two- Nonetheless, genetic algorithm (GA) is the most popular
dimensional environment without splitter rows, to minimize lost approach used frequently in the literature of FLPs [40–45]. In most
opportunity cost and material handling cost of a multi-product sys- cases, it has been shown that GA concludes better solutions than
tem. In addition, we assume unequal departments and the fuzzy other metaheuristics methods in FLPs. The genetic algorithm, which
distance between each pair of facilities to improve the mathe- was introduced by Jon Holland, has been widely applied in solving
matical modeling of the extended problem. Thus far, however, no optimization problems due to its capabilities. For example, in one
research has been found surveying these two assumptions with of the first GA applications, an FLP is solved by a genetic algorithm
each other. Based on the problem definition, a linear mixed-integer with a heuristic crossover method [46]. In addition, Wu, Chu, Wang
programming model is proposed with fuzzy constraints. Since the and Yan [47] considered GA for cellular manufacturing design and
proposed model belongs to NP-hard class, an effective GA, as a pop- layout problem. On the other hand, Sadrzadeh [48] used GA with a
ular and useful algorithm in the FLPs’ literature, is developed to heuristic procedure to solve multi-line layout problem.
conclude the best layout solutions. In some papers, GA is proposed with novel features to improve
The rest of this research is organized as follows. Section 2 con- the efficiency of algorithm. Datta, Amaral, and Figueira [44] pro-
sists of a literature review of the SRFLP and MRFLP and their solution posed a permutation-based GA for single row facility layout
methods. Then, in Section 3, the mathematical foundations of the problem. Aiello, Scalia and Enea [43] studied a multi-objective GA
proposed model are explained. In Section 4, a genetic algorithm is (MOGA) for the facility layout problem based upon slicing struc-
studied as the solution method for the defined layout problem. Fol- ture encoding. In addition, Kiaa, Khaksar-Haghanib, Javadianc and
lowing, the numerical studies and its related computational results Tavakkoli-Moghaddam [42] solved a multi-floor FLP of a dynamic
are described in Section 5. Furthermore, sensitivity analysis and cellular manufacturing system by an efficient GA. Furthermore,
related discussions are performed in Section 6. Finally, Section 7 Pourvaziri and Naderi [40] exhibited a hybrid multi-population GA
presents conclusion, including the main understandings, limita- for dynamic facility layout problem. Table 1 presents an overview
tions, and future research. of the studied publications in this paper as follows.
Obviously, our proposed problem consists of multi-product
2. Literature review with fuzzy distances [15,17,20] in a two-dimensional area, is not
observed in the studied FLPs’ literature. Moreover, we contributed
In this section, we review the literature on SRFLP and MRFLP and to the studied cost function (lost opportunity cost) of the allocated
their proposed mathematical models, which establish major part spaces to the unequal departments, as a novel term, except for the
of the FLPs background, in the last decade. Moreover, we study the handling cost.
developed solving methods consisting of metaheuristics, heuris- In the next section, firstly, a novel nonlinear programming
tics, and exact techniques. Lastly, the summary of the reviewed model with fuzzy constraints is presented for the extended MRFLP
literature is presented in Table 1. along with the explained assumptions and notations. Then, the pro-
In most of the studied papers, the defined layout problems have posed mathematical model is converted to a linear mixed-integer
similar assumptions and we can observe more innovation in programming model.
Table 1
Overview of studied literature.
*Notes: HC—Handling Cost, LOC—Loss Opportunity Cost, MLFLP—Multi Line FLP, UA-FLP—Unequal Area FLP, MOGA—Multi-Objective GA, LKNS—Lin–Kernighan Neighborhood Search, FDFLP—Fuzzy Dynamic FLP, RC—Rearrangement
Cost, MOTS—Multi-Objective TS, MFLP—Multi Floor Layout Problem, PC—Purchase Cost, OC—Operating and Overhead Cost, DRLP—Double Row FLP, TAC—Total Area Consumed by the resulting layout, FRSRFLP—Fuzzy Robust
SRFLP, CLFLP—Closed Loop FLP, LS—Local Search, GRASP—Greedy Randomized Adaptive Search Procedure.
821
822 S. Safarzadeh, H. Koosha / Applied Soft Computing 61 (2017) 819–831
Table 2 Table 3
The description of the used indices and crisp parameters. The description of the used fuzzy parameters and decision variables.
3. Mathematical modeling
Where Eqs. (1) and (2) address the number of total machines. In
The related notations are illustrated in the next subsection. addition, Eqs. (3) and (4) refer to the material flow and handling
cost between each pair of facilities, respectively. Furthermore, we
3.1. Notations have two fuzzy parameters and three types of variables, which are
introduced in Table 3 as follows:
In this subsection, the used notations are divided into the In this paper, based on the fieldwork experiences and nature
indices, crisp and fuzzy parameters, and sequentially, the decision of fuzzy variables, we used the left trapezoidal numbers to show
variables are defined for the mathematical model. Furthermore, a the minimum distances as the best possible choice. Generally, the
comprehensive introduction to the fuzzy theory and fuzzy opti- left trapezoidal number denoted Al = (a1 , b1 , b2 , b2 ) that has sup-
mization is provided from the fuzzy literature [49,50]. Thus, at first, porting interval [a1 , b2 ] in theory [50]. Fig. 1 represents the general
we describe the indexes and crisp parameters in Table 2 as follows. form of the left trapezoidal number as follows.
S. Safarzadeh, H. Koosha / Applied Soft Computing 61 (2017) 819–831 823
MinimizeS0 X Y + cij fij (|xi − xj | + |yi − yj |) (7) upper bound of the objective which is obtained by placing upper
bound of right hand side values in the constraints and solving the
i=1 j=1 mathematical model. Finally, the new objective function is obtained
j=
/ i in Eq. (20) as follows:
j=
/ i
li lj w wj
xij+ − [Ri + Rj + (1 − Ri ) i + (1 − Rj ) ] + M(˛ij + ˇij ) − hij ≥ dhij − hij , ∀i, j (29)
2 2 2 2
li lj w wj
xij− − [Ri + Rj + (1 − Ri ) i + (1 − Rj ) ] + M(˛ij + (1 − ˇij )) − hij ≥ dhij − hij , ∀i, j (30)
2 2 2 2
xi − xj = xij+ − xij− , ∀i, j (31)
wi wj l lj
yij+ − [Ri + Rj + (1 − Ri ) i + (1 − Rj ) ] + M((1 − ˛ij ) + ij ) − vij ≥ dvij − vij , ∀i, j (32)
2 2 2 2
wi wj l lj
yij− − [Ri + Rj + (1 − Ri ) i + (1 − Rj ) ] + M((1 − ˛ij ) + (1 − ij )) − vij ≥ dvij − vij , ∀i, j (33)
2 2 2 2
yi − yj = yij+ − yij− , ∀i, j (34)
li w
xi − [Ri + (1 − Ri ) i ] ≥ 0, ∀i, j (35) most of the investigated studies generate the initial population
2 2
randomly (except Ref. [40,42]). Table 4 shows an overview of the
wi l
yi − [Ri + (1 − Ri ) i ] ≥ 0, ∀i, j (36) proposed GAs in the studied publications as follows.
2 2 We can observe in Table 4 that roulette wheel selection is a
li w popular selection method in the studied literature. In addition, the
xi + [Ri + (1 − Ri ) i ] ≤ X ∀i, j (37)
2 2 reproduction operators of GA consisting of PMX, N-point, and uni-
wi l form methods have more application than those of others. On the
yi + [Ri + (1 − Ri ) i ] ≤ Y , ∀i, j (38) other hand, the number of generations criterion is dominated stop
2 2
condition in the proposed GAs. Accordingly, we illustrate the GA
X ≤ X, (39)
elements in detail as follows.
Y ≤ Y, (40)
Table 5
DM* satisfaction
The variant size of MRFLP.
Stop condition
G. No. + Conv.*
G. No. + Conv.
Problem No. of machines No. of variables No. of constraints
* : Material handling cost; RWS: Roulette Wheel Selection; PMX: Partially Mapped Crossover; OX: Order Crossover; CX: Cycle Crossover; G. No.* : The Number of Generations; Conv: Convergence; DM: Decision Maker.
G. No.*
1 21 1074 1554
G. No.
G. No.
G. No.
G. No.
G. No.
G. No.
G. No.
G. No.
2 32 2515 3600
3 40 3943 5620
4 60 8913 12,630
Swap+ Heuristic
7 175 76,303 107,275
Uniform meth.
Uniform meth.
Uniform meth.
Group meth. 8 280 195,583 274,540
Swap meth.
Swap meth.
9 500 624,253 875,250
Mutation
Heuristic
Heuristic
Heuristic
10 2000 9,997,003 14,001,000
*
Heuristic + PMX* + OX + CX
PMX+2-Point+Uniform
PMX + N-Point meth.
Heuristic
Heuristic
Heuristic
Heuristic
Heuristic
PMX
the population.
Tournament meth.
Tournament meth.
Tournament meth.
The best solutions
RWS + Dynamic
RWS*+Random
Modified RWS
Heuristic
Heuristic
(max ϕ+1−ϕ)
1⁄ϕ
ϕ
ϕ
ϕ
ϕ
ϕ
that the studied problem has a high time complexity, even in the
JSSP-FSSP
UA-FLP
UA-FLP
UA-FLP
MLFLP*
MRFLP
MFLP*
SRFLP
DFLP
CMS
FMS
5. Computational results
Kiaa, Khaksar-Haghanib, Javadianc, and Tavakkoli-Moghaddam [42]
García-Hernández, Pierreval, Salas-Morera and Arauzo-Azofra [41]
values were obtained for the main parameters. In the next subsec-
Table 4
Table 6 Table 11
Dimensions and symbols of machines. The minimum fuzzy horizontal distance needed among machines (m).
Machine type Symbol Length (m) Width (m) Entrance Cutting Milling Drilling Welding
Table 7
The specification of products.
Table 12
Product Demand average (in each period) Processing Route (from left to right) The minimum fuzzy vertical distance needed among machines (m).
Entrance 0 3 1 5 2
Cutting 3 0 2 4 3
Milling 1 2 0 3 3
Drilling 5 4 3 0 5
Welding 2 3 3 5 0
Table 10
Product B handling cost among machines (currency per meter).
Entrance 0 5 2 1 2
Cutting 5 0 2 2 5
Milling 2 2 0 1 3
Drilling 1 2 1 0 1
Welding 2 5 3 1 0
This section addresses a numerical example of the MRFLP with Fig. 5. The best-obtained layout for the numerical example.
two products and five machines as the production system. More-
over, the length and width values of the area were considered
50 × 50 , respectively. In each time period, the demand values of size = 300, iteration number = 300). Thus, the objective value is
products A and B must be satisfied. In addition, the lost oppor- obtained 0.6217 with the following arrangement, which is shown
tunity cost of each area unit is considered 10 units of currency. in Fig. 5:
The specifications of the required equipment to manufacture the As we can observe in Fig. 5, the facilities, with considering the
products and products are shown in Tables 6 and 7, respectively. existent constraints, have a reasonable and productive arrange-
Moreover, Table 8 shows the processing times of products at each ment. In addition, we show the trend of fitness value improvement
facility. Tables 9 and 10 present the handling cost for products A for the numerical example in Fig. 6. The main point of Fig. 6 is that
and B. Finally, Tables 11 and 12 denote the minimum horizontal the best and the mean fitness values have a dramatical ascendancy,
and vertical distances needed among facilities as fuzzy numbers. in the entire process of optimization. This shows that the proposed
According to the presented data in Table 7, Table 8 and available GA is efficient and convergent.
time in each time period (that is considered 70 min), the number of The next section investigates a comprehensive sensitivity anal-
required equipment for each product is calculated by Eq. (1), and ysis on the main parameters of the proposed GA and then, we define
the results are demonstrated in Table 13 as follows: three levels for each parameter and determine the optimized level
In the next step, fl and fu values (that were explained in by the Taguchi method. Finally, according to the result of parame-
subsection 3.4) obtain 10,000 and 14,000, respectively, by the opti- ter settings, we compare the obtained results of different crossover
mal GA, which set its parameters in subsection 6.1. (population operators and other GA parameters in detail.
828 S. Safarzadeh, H. Koosha / Applied Soft Computing 61 (2017) 819–831
Table 14
The GA parameter ranges and levels.
* beta is a GA parameter.
Table 15
Calibration process of GA.
A B C R1 R2 R3 R4 S. D* Mean
* Standard deviation.
In this section, the effects of changing the main parameters on
the solutions are addressed in detail. In the first step, the GA param-
eters are set by the Taguchi method. Next, the impact of crossover
observed in Table 15 in detail. In addition, Fig. 7 shows the main
operator on the output of the proposed GA is examined by many
effects for means of factor levels. It should be noted that each level
experiments in subsection 6.2. Finally, we conduct a comprehen-
has been run 10-times.
sive sensitivity analysis on the size of the population and number
It can be observed in Fig. 7, which the optimal factor levels of
of iterations, respectively.
crossover rate, mutation rate, and ˇ parameter are 0.75, 0.25 and
1, respectively. The results show that the proposed algorithm is
6.1. Parameter setting
sensitive to the mutation rate, and the ˇ parameter has insignificant
effect on the output. In the next subsection, we apply the optimal
Taguchi method is one of the most prevalent methods to analyze
GA to compare the crossover operators and the different size of
the output of experiments. It is used where the number of experi-
MRFLP practically.
mental factors or factor levels are numerous and take considerable
time to be optimized [54]. In this way, the experimental factors and
their levels are identified and then the Taguchi method is used to 6.2. Comparisons and numerical results
determine the optimal output level of each factor. Table 14 presents
the main parameters of the proposed GA and their ranges: We compare between the crossover operators with the optimal
Based on Table 14, the optimal level of each factor is obtained GA, in this subsection. Thus, the proposed GA was run 100-times
according to the numerical experiments, which are determined for 40 experiments. Furthermore, we apply uniform, 2-point and
by the Taguchi method with 9 experiments. The results can be position based crossover (PBX) methods in GA to determine the
Table 16
The experimental results for the proposed crossover operators.
Problem PBX Uniform cr. 2-P cr. Problem PBX Uniform cr. 2-P cr.
Best Avg. Time Best Avg. Time Best Avg. Time Best Avg. Time Best Avg. Time Best Avg. Time
1 0.1347 0.1345 262 0.5475 0.5085 290 0.1592 0.1561 195 21 0.2470 0.2469 316 0.0855 0.0854 342 0.3632 0.2916 286
2 0.3792 0.2775 272 0.1590 0.1572 303 0.4519 0.4519 202 22 0.3995 0.3987 300 0.3073 0.3072 295 0.2709 0.2706 250
3 0.6544 0.6364 276 0.3710 0.3593 321 0.2323 0.2323 178 23 0.0799 0.0706 373 0.2552 0.2530 338 0.1726 0.1725 312
4 0.3517 0.3493 269 0.1590 0.1572 304 0.4411 0.4411 235 24 0.4629 0.3650 298 0.0478 0.0463 268 0.6322 0.4178 289
5 0.1545 0.1499 301 0.3710 0.3593 328 0.2304 0.2304 229 25 0.0318 0.0317 317 0.5231 0.5211 350 0.3473 0.3193 298
6 0.2716 0.2716 273 0.5692 0.3904 255 0.0623 0.0608 201 26 0.1223 0.1090 305 0.4214 0.4015 268 0.4732 0.3407 339
7 0.2865 0.2549 257 0.0608 0.0446 299 0.1258 0.0831 189 27 0.2518 0.1375 374 0.2330 0.2132 312 0.2915 0.2882 254
8 0.2800 0.2799 292 0.1590 0.1572 309 0.4996 0.4244 216 28 0.4030 0.3840 325 0.2660 0.2659 293 0.4268 0.3916 249
9 0.1855 0.1852 270 0.3710 0.3593 317 0.4062 0.3865 225 29 0.3551 0.3462 330 0.1296 0.0459 296 0.2950 0.2946 262
10 0.5897 0.5799 264 0.5692 0.3904 247 0.3128 0.2837 235 30 0.1098 0.0352 285 0.3532 0.3531 316 0.0485 0.1030 268
11 0.6720 0.6052 310 0.0253 0.0240 411 0.2274 0.2088 244 31 0.5339 0.5325 313 0.1775 0.1611 305 0.2274 0.2088 252
12 0.1408 0.1398 258 0.1590 0.1572 348 0.1894 0.1893 254 32 0.3727 0.2947 371 0.3391 0.3390 276 0.0754 0.0753 230
13 0.3292 0.3198 331 0.3710 0.3593 367 0.1587 0.1221 329 33 0.3792 0.3572 343 0.4803 0.1962 354 0.2274 0.2088 249
14 0.3452 0.3451 312 0.5692 0.3904 286 0.1942 0.1753 248 34 0.2640 0.2635 312 0.3994 0.3963 313 0.0754 0.0753 229
15 0.5011 0.4955 313 0.3729 0.3478 339 0.6150 0.5963 268 35 0.4255 0.4255 334 0.1618 0.1412 355 0.0335 0.0334 222
16 0.2092 0.1997 322 0.2277 0.2259 333 0.5466 0.5454 266 36 0.6001 0.6001 272 0.1514 0.1148 376 0.3632 0.2916 240
17 0.4696 0.4695 302 0.5146 0.4919 304 0.0203 0.0201 294 37 0.1159 0.1143 336 0.1590 0.1572 342 0.2709 0.2706 255
18 0.0165 0.0034 293 0.4605 0.4530 338 0.2274 0.2088 270 38 0.4352 0.4351 308 0.4083 0.4007 340 0.1726 0.1725 278
19 0.1285 0.0994 359 0.4291 0.2282 318 0.0754 0.0753 287 39 0.4263 0.4226 313 0.4711 0.4559 282 0.6322 0.4178 258
20 0.2220 0.2057 334 0.4203 0.1238 333 0.3350 0.3340 264 40 0.3480 0.3478 361 0.6010 0.5651 295 0.3473 0.3193 274
Average: 0.3171 0.2980 309 0.3214 0.2776 317 0.2814 0.2547 253
Table 17
The experimental results of sensitivity analysis on the population and iteration size.
Problem n pop n iter Fitness value Problem n pop n iter Fitness value
best operator for the extended MRFLP. Table 16 shows the results Table 18
The coefficient and power of exponential trend lines.
as follows.
Comparing the obtained values of the three different crossover Seri n pop Coefficient Power Normalized power*
operators shows minor differences among the results that make 1 40 0.1575 0.0350 1.0000
selecting a specific operator as the best one for the studied problem 2 50 0.1397 0.0330 0.9429
difficult. The mean of objective values for PBX solutions is better 3 75 0.1322 0.0252 0.7200
than that of other methods. However, the uniform operator pro- 4 100 0.2239 0.0111 0.3171
5 150 0.2546 0.0085 0.2429
vides the best results. Comparing the average run-time shows that
6 200 0.2520 0.0070 0.2000
2-point crossover operator takes less run time compared with other 7 300 0.2654 0.0049 0.1400
ones. Moreover, another main point is that the minimum difference 8 500 0.2927 0.0040 0.1143
between the mean value and the best value can be observed in the * That is equal to Power/Power(Seri40).
PBX crossover method. In another study, the objective values are
evaluated by changing the population size (n pop) and the number
of iterations (n iter) at different levels. Table 17 summarizes the
dramatically. For better data analysis, the results are presented in
results as follows. It should be regarded that, each level has been
Fig. 8 as follows. Moreover, the coefficients and powers of exponen-
run 10-times.
tial trend lines for the total population series are shown in Table 18
Table 17 shows that with the significant increase in the popula-
to improve the conclusion.
tion size (n pop), the running time increases as well as the fitness
Table 18 determines that increasing the population size which is
function value. In addition, increasing the number of iterations not
shown with different series in Fig. 8 generally leads to less power.
only includes higher run-time but also improves the objective value
In other words, with the growth of population size, the required
830 S. Safarzadeh, H. Koosha / Applied Soft Computing 61 (2017) 819–831
Fig. 8. The fitness values change over time for population series.
[22] R. Kothari, D. Ghosh, Insertion based Lin–Kernighan heuristic for single row [38] H. Samarghandi, P. Taabayan, F.F. Jahantigh, A particle swarm optimization for
facility layout, Comput. Oper. Res. 40 (2013) 129–136. the single row facility layout problem, Comput. Ind. Eng. 58 (2010) 529–534.
[23] A.R. Amaral, A.N. Letchford, A polyhedral approach to the single row facility [39] M. Solimanpur, A. Jafari, Optimal solution for the two-dimensional facility
layout problem, Math. Programm. 141 (2013) 453–477. layout problem using a branch-and-bound algorithm, Comput. Ind. Eng. 55
[24] C. Ou-Yang, A. Utamima, Hybrid estimation of distribution algorithm for (2008) 606–619.
solving single row facility layout problem, Comput. Ind. Eng. 66 (2013) [40] H. Pourvaziri, B. Naderi, A hybrid multi-population genetic algorithm for the
95–103. dynamic facility layout problem, Appl. Soft Comput. 24 (2014) 457–469.
[25] M.R. Garey, D.S. Johnson, Computers and Intractability. A Guide to the Theory [41] L. García-Hernández, H. Pierreval, L. Salas-Morera, A. Arauzo-Azofra, Handling
of NP Completeness, WH Freeman, New York, 1979. qualitative aspects in unequal area facility layout problem: an interactive
[26] G. Palubeckis, Fast local search for single row facility layout, Eur. J. Oper. Res. genetic algorithm, Appl. Soft Comput. 13 (2013) 1718–1727.
246 (2015) 800–814. [42] R. Kiaa, F. Khaksar-Haghanib, N. Javadianc, R. Tavakkoli-Moghaddam, Solving
[27] M. Rubio-Sánchez, M. Gallego, F. Gortázar, A. Duarte, GRASP with path a multi-floor layout design model of a dynamic cellular manufacturing system
relinking for the single row facility layout problem, Knowl.-Based Syst. 106 by an efficient genetic algorithm, J. Manuf. Syst. 33 (2014) 218–232.
(2016) 1–13. [43] G. Aiello, G.L. Scalia, M. Enea, A multi objective genetic algorithm for the
[28] Y. Hani, L. Amodeo, F. Yalaoui, H. Chen, Ant colony optimization for solving an facility layout problem based upon slicing structure encoding, Expert Syst.
industrial layout problem, Eur. J. Oper. Res. 183 (2007) 633–642. Appl. 39 (2012) 10352–10358.
[29] J. Guan, G. Lin, Hybridizing variable neighborhood search with ant colony [44] D. Datta, A.R.S. Amaral, J.R. Figueira, Single row facility layout problem using a
optimization for solving the single row facility layout problem, Eur. J. Oper. permutation-based genetic algorithm, Eur. J. Oper. Res. 213 (2011) 388–394.
Res. 248 (2016) 899–909. [45] M. Ficko, M. Brezocnik, J. Balic, Designing the layout of single-and
[30] M. Solimanpur, P. Vrat, R. Shankar, An ant algorithm for the single row layout multiple-rows flexible manufacturing system by genetic algorithms, J. Mater.
problem in flexible manufacturing systems, Comput. Oper. Res. 32 (2005) Process. Technol. 157 (2004) 150–158.
583–598. [46] K.L. Mak, Y.S. Wong, F.T.S. Ghan, A genetic algorithm for facility layout
[31] R. Kothari, D. Ghosh, Tabu search for the single row facility layout problem problem, Comput. Integrat. Manuf. Syst. 11 (1998) 113–127.
using exhaustive 2-opt and insertion neighborhoods, Eur. J. Oper. Res. 224 [47] X. Wu, C.-H. Chu, Y. Wang, W. Yan, A genetic algorithm for cellular
(2013) 93–100. manufacturing design and layout, Eur. J. Oper. Res. 181 (2007) 156–167.
[32] H. Samarghandi, K. Eshghi, An efficient tabu algorithm for the single row [48] A. Sadrzadeh, A genetic algorithm with the heuristic procedure to solve the
facility layout problem, Eur. J. Oper. Res. 205 (2010) 98–105. multi-line layout problem, Comput. Indu. Eng. 62 (2012) 1055–1064.
[33] X. Zuo, C.C. Murray, A.E. Smith, Solving an extended double row layout [49] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic-Theory and Applications,
problem using multi-objective tabu search and linear programming, IEEE Prentice Hall PTR, Upper Saddle River, New Jersey, 1995.
Trans. Autom. Sci. Eng. 11 (2014) 1122–1132. [50] G. Bojadziev, M. Bojadziev, Fuzzy logic for business, finance, and
[34] N. Lenin, M. Siva Kumar, D. Ravindran, M.N. Islam, A tabu search for management, World Sci. 23 (2007) 9–25.
multi-objective single row facility layout problem, J. Adv. Manuf. Syst. 13 [51] M. Jiménez, M. Arenas, A. Bilbao, M.V. Rodrı, Linear programming with fuzzy
(2014) 17–40. parameters: an interactive method resolution, Eur. J. Oper. Res. 177 (2007)
[35] S. Wang, X. Zuo, X. Liu, X. Zhao, J. Li, Solving dynamic double row layout 1599–1609.
problem via combining simulated annealing and mathematical programming, [52] H. Maleki, M. Tata, M. Mashinchi, Linear programming with fuzzy variables,
Appl. Soft Comput. 37 (2015) 303–310. Fuzzy Sets Syst. 109 (2000) 21–33.
[36] G. Palubeckis, Fast simulated annealing for single-row equidistant facility [53] J.H. Holland, Adaptation in Natural and Artificial Systems: An Introductory
layout, Appl. Math. Comput. 263 (2015) 287–301. Analysis with Applications to Biology, Control, and Artificial Intelligence, U
[37] S. Niroomand, N. Mirzaei, R. Şahin, B. Vizvári, A new exact formulation and Michigan Press, 1975.
simulated annealing algorithm for one-Sided closed loop layout, J. Comput. [54] K. Tatebayashi, Computer aided engineering combined with Taguchi
Theor. Nanosci. 12 (2015) 3817–3826. methods, Ann. Meet. Japan Soc. Mechan. Eng. 8 (2005) 224–225.