You are on page 1of 15

International Journal of Production Research

ISSN: 0020-7543 (Print) 1366-588X (Online) Journal homepage: https://www.tandfonline.com/loi/tprs20

A mixed-integer programming formulation for the


double row layout of machines in manufacturing
systems

André R.S. Amaral

To cite this article: André R.S. Amaral (2019) A mixed-integer programming formulation for the
double row layout of machines in manufacturing systems, International Journal of Production
Research, 57:1, 34-47, DOI: 10.1080/00207543.2018.1457811

To link to this article: https://doi.org/10.1080/00207543.2018.1457811

Published online: 17 Apr 2018.

Submit your article to this journal

Article views: 464

View related articles

View Crossmark data

Citing articles: 17 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
International Journal of Production Research, 2019
Vol. 57, No. 1, 34–47, https://doi.org/10.1080/00207543.2018.1457811

A mixed-integer programming formulation for the double row layout of machines in


manufacturing systems
André R.S. Amarala∗
a Graduate School of Computer Science, Universidade Federal do Espírito Santo (UFES), Vitória, Brazil

(Received 15 May 2017; accepted 14 March 2018)

The Double Row Layout Problem (DRLP) is the problem of allocating a given set of machines on both sides of a straight line
corridor so as to minimise the total cost of transporting materials among machines. The DRLP occurs in several manufacturing
plants, particularly in semiconductor manufacturing. While it has a large practical importance, the problem is very difficult
to solve to optimality. In this paper, we construct a mixed-integer programming (MIP) formulation of the problem, which
favourably compares to a previously published MIP formulation. The new model is found to present similar performance to
another published MIP formulation, and it has the advantage of being more intuitive for handling qualitative inputs that may
be required in a layout refinement phase.
Keywords: facility design; integer programming; discrete optimisation; automated manufacturing systems; CAM

1. Introduction
Consider a manufacturing system, which consists of a set of machines and a material-handling device that transports materials
among machines. In order to increase the flexibility of the system, it is important to utilise an automated handling device,
such as an automated guided vehicle (AGV). The AGV is superior to the conventional conveyer with regard to utilisation,
cost and flexibility (Wang and Chang 2015). Due to these aspects, AGVs are largely employed in flexible manufacturing
systems (FMSs) and since AGVs have a better performance when moving in straight lines, they are particularly interesting
for FMSs that have machines arranged in straight rows (e.g. Tubaileh and Siam 2017).
It frequently occurs in practice that machines are to be arranged in a double row layout, i.e. a layout in which the machines
are placed on locations on both sides of a straight line corridor with their lengths parallel to the corridor. Thus, one has to
solve the Double Row Layout Problem (DRLP), which is how to assign the given set of machines to locations on either side
of the corridor so that the total cost of transporting materials among machines is minimised. The problem has been studied by
various authors (e.g. Heragu and Kusiak 1988; Chung and Tanchoco 2010; Murray, Smith, and Zhang 2013; Amaral 2013a).
Researchers such as Wang et al. (2015) and Zuo, Murray, and Smith (2014) stated that the DRLP has practical application
in various industries, particularly in semiconductor manufacturing. It should be noted that the semiconductor industry is of
tremendous economic relevance, with worldwide semiconductor industry sales in 2017 having reached US$ 412.2 billion
(Semiconductor Industry Association 2017). Studies reporting on the importance of the DRLP in the semiconductor industry
have appeared in the literature. For example, Yang and Peters (1997) pointed out that double row layouts are common designs
for the layout in semiconductor wafer fabrication facilities. Peters and Yang (1997) added that whereas there could be other
possible layout arrangement strategies, the semiconductor industry is reluctant to try them, because the currently adopted
double row configuration offers many advantages in the semiconductor environment.
Double row layouts have also been useful in state-of-the-art production lines for LCD manufacturing. For example,
Chung and Tanchoco (2010) described a practical example of the DRLP in an LCD fabrication line, in which seven toolsets
are located along a corridor and LCD glasses under fabrication are transferred in carriers by advanced automated material
handling systems.
A practical application of the DRLP often reported in the literature occurs in flexible manufacturing systems design
(Heragu and Kusiak 1988). In this regard, Kouvelis (1992) emphasised that machine layout is of large importance for the
effective utilisation of a flexible manufacturing system. Furthermore, Hassan (1994) pointed out that machine layout is an
important step in designing modern manufacturing facilities. This paper is focused on machine layout with a double row
configuration.

∗ Corresponding author. Email: amaral@inf.ufes.br

© 2018 Informa UK Limited, trading as Taylor & Francis Group


International Journal of Production Research 35

Figure 1. Representation of the distance di j and the clearance value ai j for a pair (i, j) of machines.

In machine layout, a minimum separation might be required between two adjacent machines, which is called a clearance.
In this regard, Heragu and Kusiak (1988) used the assumption that the clearance between each pair of machines is given. In
this case, data for a DRLP instance will consist of the number n of machines, the length i of each machine i, the amount of
flow ci j between each pair (i, j) of machines and the clearance ai j required between each pair (i, j) of machines.
In a feasible DRLP layout ϕ, let di j be the distance between the centres of the pair (i, j) of machines (see Figure 1).
Then, the DRLP is how to find a feasible layout ϕ that minimises

ci j di j .
1≤i< j≤n

The clearance between two machines varies according to which machine is adjacent to each other in a machine sequence;
thus, it is dependent on the sequence. However, as Chung and Tanchoco (2010) observed, in practical problems, the clearance
is dependent on the machine, because it is mainly used for maintenance; in this case, the clearance can be included in the
length of the machine. More specifically, suppose the required clearances are all equal to the same value, say a. Then,
the length of each machine can be increased by the value a after which, the problem can be treated as a problem without
clearances (see, e.g. Heragu and Kusiak 1988; Anjos and Vannelli 2008; Amaral 2009; Brunese and Tanchoco 2013).
We say that a DRLP instance has explicit clearances, when the input values (ai j )1≤i< j≤n are explicitly given in the
instance data. However, when the values (ai j )1≤i< j≤n do not appear in the instance data, because they have already been
included in the lengths of the machines, we say that the instance has implicit clearances. It should be noted that, if clearances
between machines are not all equal, implicit clearances may not be used because the ordering of machines will influence the
clearance values.
Here, in order to simplify the presentation, we assume that clearances are included in the lengths of the machines (implicit
clearances).
(a )
Chung and Tanchoco (2010) presented a Mixed Integer Programming (MIP) model for the DRLP, denoted here by MC Ti j ,
which contained the values ai j as parameters. By setting ai j = 0, for every machine pair (i, j) in that model, we obtain a
(0)
model called MC T that can be used with implicit clearances.
(a )
Zhang and Murray (2012) showed that Chung and Tanchoco’s original model MC Ti j (containing the input values ai j , not
all of which are zero) is not a correct formulation of the DRLP, because the clearance values will not be correctly observed in
(a ) (a ) (0)
MC Ti j . Although MC Ti j is not a correct DRLP formulation, it turns out that MC T is a valid model (because explicit clearance
values are omitted).
(0)
In this paper, a new MIP formulation of the DRLP is presented. The new model outperforms the model MC T of Chung
and Tanchoco (2010). In comparison with the model of Amaral (2013a), the new model presents a similar performance but
it can be more intuitive for handling qualitative input such as requiring that two specific machines meet some conditions of
a particular industrial environment.
This paper is organised as follows. In Section 2, we review literature concerning row layout problems, including the
DRLP. In Section 3, we present the new mixed-integer programming model for the DRLP. Computational experiments are
presented in Section 4, which is followed by the conclusions section.
36 A.R.S. Amaral

2. Literature review
Row layout problems are very interesting from a practical point of view. Nevertheless, they are NP-Hard in general (e.g.
Drira, Pierreval, and Hajri-Gabouj 2007). Therefore, as the problem size increases, the required computational expenditure
to solve the problem increases considerably. The DRLP is a row layout problem in the same way as the Single Row Facility
Layout Problem (SRFLP). However, the DRLP differs from the SRFLP, because in the SRFLP all of the machines are placed
at the same side of a corridor. For the SRFLP, there are solution methods based on mixed-integer linear programming (MILP)
(e.g. Love and Wong 1976; Heragu and Kusiak 1991; Amaral 2006, 2008), 0-1 linear programming (e.g. Amaral 2009),
branch-and-cut (e.g. Amaral and Letchford 2013) and semi-definite programming (e.g. Anjos, Kennings, and Vannelli 2005;
Anjos and Vannelli 2008; Anjos and Yen 2009; Hungerländer and Rendl 2012).
The MIP model of the SRFLP given by Love and Wong (1976) contains a variable xi denoting the position of machine
n
i along the horizontal axis. In that model, a bound on xi is implemented by the constraint: i ≤ xi ≤ i=1 i . Brunese and
Tanchoco (2013) developed a new bound on xi as follows: sort the lengths of the machines in ascending order; then, place
the smallest machine length, say  , at one end of the ordering and the next smallest machine length, say  , at the opposite
end of the ordering. This bound is implemented as the constraint:
 +  
0 ≤ xi ≤ + k , (1 ≤ i ≤ n). (1)
2 
k >

They tested the efficacy of their developed constraint on a MIP formulation of the SRFLP.
Different heuristics have also been proposed for the SRFLP, which are based on tabu search (e.g. de Alvarenga, Negreiros-
Gomes, and Mestria 2000; Samarghandi and Eshghi 2010; Kothari and Ghosh 2013c), simulated annealing (e.g. Heragu
and Alfa 1992; de Alvarenga, Negreiros-Gomes, and Mestria 2000; Kothari and Ghosh 2013b; Palubeckis 2017), genetic
algorithms (e.g. Ozcelik 2012; Kothari and Ghosh 2013a), ant colony optimisation (e.g. Solimanpur, Vrat, and Shankar
2005, Guan and Lin 2016), scatter search (e.g. Kothari and Ghosh 2014), particle swarm optimisation (e.g. Samarghandi,
Taabayan, and Jahantigh 2010) and GRASP (e.g. Rubio-Sánchez et al. 2016). Other methods for the SRFLP are given in the
surveys by Kothari and Ghosh (2012) and Keller and Buscher (2015).
For the DRLP, a constructive heuristic was presented by Heragu and Kusiak (1988). They concluded that the solutions
obtained were satisfactory and the CPU time requirements were very low. Chung and Tanchoco (2010) presented a MIP
model of the DRLP intended to work with explicit clearances. In addition, based on existing constructive heuristics for the
SRFLP, Chung and Tanchoco (2010) developed five constructive heuristics for the DRLP: MinLCF (minimum location cost
first), MinFF (minimum flow first), MaxFF (maximum flow first), MinWF (minimum width first) and MaxWF (maximum
width first), while these heuristics differ in policies for selecting a machine to be assigned to some available location; they
share a similar structure, as follows: (i) a selection policy determines machine relative locations; (ii) Given these relative
locations, all integer variables in the MIP model are fixed, yielding a linear program and (iii) the linear program is solved to
obtain absolute locations of the machines. Zhang and Murray (2012) showed that the MIP model of Chung and Tanchoco
(2010) may allow layouts that do not respect the clearance requirements between some adjacent machines. The heuristics of
Chung and Tanchoco (2010) depend on outputs from this MIP model, which cast doubt upon the validity of their numerical
results. Zhang and Murray (2012) also provided a correction for Chung and Tanchoco’s model. The corrected model was
used in subsequent research involving explicit clearances.
Amaral (2013a) defined a polytope in which each vertex corresponds to a partition of a linear order on the set of n
machines into two linear orders. Some classes of inequalities were shown to be valid for this polytope. Then, a mixed-integer
programming formulation based on the polytope was given, which allowed the exact solution of DRLP instances with up to
12 machines. This work considered implicit clearances.
In order to solve the DRLP with asymmetric flows, Murray, Smith, and Zhang (2013) implemented the five construc-
tive heuristics of Chung and Tanchoco (2010) and three new ones. The new constructive heuristics, named minFFasym,
maxFFasym and maxFFmod, are based on Chung and Tanchoco’s MinFF and MaxFF heuristics and aim at accommodating
asymmetric material flow. They report that these eight constructive heuristics produce solutions of similar quality. However,
a large improvement can be obtained by a combination of a constructive heuristic and a simple local search heuristic.
Murray, Zuo, and Smith (2012) proposed an extended double row layout problem (EDRLP) in which the aisle may have a
nonzero width and two optimisation objectives (cost and layout area) are considered. The MIP model developed in previous
work (Chung and Tanchoco 2010; Zhang and Murray 2012) was extended for the EDRLP. They linearly combined the two
objectives to form a single objective and used CPLEX to solve the EDRLP with 10 machines.
The linear combination of the objectives creates challenges when these terms are of much different orders. In order to
treat this issue, the work of Zuo, Murray, and Smith (2014) builds a Pareto front of solutions pertaining to both objectives.
Their approach combines a multi-objective tabu search (MTS) with linear programming (LP). MTS finds a machine sequence
International Journal of Production Research 37

(relative location of machines) and, then, linear programming is used to determine the absolute location of each machine on
its assigned row. The MTS-LP approach was reported to run quickly and to cover the breadth of the Pareto front well.
Wang et al. (2015) considered a dynamic double-row layout problem (DDRLP) in which material flows change over time
in different processing periods. The authors solved this problem using what they called an improved simulated annealing
(ISA) combined with mathematical programming. Zuo, Murray, and Smith (2016) studied a double-row layout problem with
shared clearances and presented an approach based on multi-objective tabu search and heuristic rules for this problem.
Some interesting row layout problems are related to the DRLP. For example, the corridor allocation problem (CAP)
(Amaral 2012) places rectangular rooms on both sides of a horizontal corridor respecting two conditions: (i) no space is
allowed between two adjacent rooms and (ii) the left wall of the left-most room on either side of the corridor should touch
a vertical line of reference. Usually, in both the CAP and the DRLP, the width of the corridor is assumed to be negligible,
but in the DRLP conditions (i) and (ii) do not have to be met. A layout of the CAP is defined by (π, t), where π is some
permutation of all rooms and the parameter t indicates that the first t rooms of π are placed in sequence at one side of the
corridor, and the next (n − t) rooms of π are placed in sequence at the other side of the corridor. As π and t vary, one obtains
all possible CAP layouts, and can choose one, which minimises material handling costs or which minimises the length of
the corridor; where it is necessary to have the corridor fit some limited floor space, one can choose a CAP layout with the
least material handling cost, and whose corridor length is not greater than a given value. A MIP formulation of the CAP
was proposed by Amaral (2012). Ghosh and Kothari (2012) presented a genetic algorithm with embedded local search and a
scatter search algorithm with path relinking. Ahonen, de Alvarenga, and Amaral (2014) developed tabu search and simulated
annealing algorithms. Kalita and Datta (2014) applied a permutation-based genetic algorithm (pGA) to handle the CAP as
an unconstrained bi-objective optimisation problem, in which both the total flow cost and the length of the corridor are to be
minimised.
Also related to the DRLP is the multirow layout problem (MRLP) in which machines should be arranged in several
rows in order to minimise the total transportation cost among machines. Heragu and Alfa (1992) considered the MRLP
with machines of equal-area. Gen, Ida, and Cheng (1995) formulated a fuzzy multirow machine layout problem where the
clearance between any two adjacent machines is given as a fuzzy set and solved the problem by genetic algorithms. Ficko,
Brezocnik, and Balic (2004) presented for the MRLP an approach that determines the number of rows and the sequence
of machines in each row by means of genetic algorithms. Amaral (2013b) investigated a version of the MRLP called the
k-parallel row ordering problem (k-PROP) where facilities are pre-assigned to k rows (k ≥ 2) and one wishes to find an
ordering of the facilities in each row so that some cost function is minimised. A MIP formulation of the 2-PROP was used to
obtain optimal solutions for problems of moderate size. Tubaileh and Siam (2017) studied the MRLP in which the length of
rows are limited by the vertical and horizontal dimensions of the facility. Machines are assumed to be rectangular in shape
with pick-up/drop-off points located at their centres. To solve the problem the authors propose two algorithms one based on
ant colony optimisation and the other based on simulated annealing.

3. New integer programming model of the DRLP


In this section, we propose a new mixed-integer programming model for the DRLP. We use the following notation:
• n: number of machines;
• N = {1, 2, . . ., n}: set of machines;
• R={lower row, upper row}: set of rows parallel to the x-axis;
• ci j : amount of flow between machines i and j (1 ≤ i < j ≤ n);
• i : length of machine i (i ∈ N ) (parallel to the x-axis).
n
An important assumption made in the DRLP is that the width of the corridor is negligible. Let L = i=1 i and assume that
the corridor is situated with its length along the x-axis on the interval [0, L]. Each machine has load/unload points and these
are located in the middle of the machine. Since the width of the corridor is not to be considered, the distance di j between
machines i and j is defined as the x-distance between their centres. We assume that n ≥ 3, otherwise the problem is trivial.
We also assume that clearances are included in the lengths of the machines.

3.1 Binary variables


When machines i and j are placed at the same row, we need the following variables to avoid overlap:

1, if machine j is placed to the right of machine i;
ti j =
0, if machine j is placed to the left of machine i;
38 A.R.S. Amaral

In order to determine whether or not machines i and j are placed at the same row, we need the following variables:

1, if machines i and j are both placed at the same row;
ui j =
0, otherwise.

3.2 Continuous Variables


We need the continuous variables:
xi : abscissa of the centre of machine i;
di j : x-distance between (the centres of) machines i and j .

3.3 Nonoverlapping constraints


The following constraints guarantee that no overlap occurs among machines placed at the same row:
 
i +  j
xj + ≤ xi + L(1 + ti j − u i j )
2
 
i +  j
xi + ≤ x j + L(2 − ti j − u i j ), (1 ≤ i < j ≤ n).
2
When machines i and j are both placed at the same row, u i j = 1, and the constraints state that machine i is either to the
left or to the right of machine j so that they do not overlap. On the other hand, when machines i and j are placed at different
rows, u i j = 0, and the constraints are satisfied.

3.4 Constraints for determining inter-machine distances


From the definition of xi and di j , we can write:

xi − x j , if xi > x j
di j = , (1 ≤ i < j ≤ n). (2)
x j − xi , otherwise
Since we have a minimisation problem, we may represent (2) as:
di j ≥ xi − x j ; di j ≥ x j − xi , (1 ≤ i < j ≤ n).
Note that the distance determination described in Equation (2) is used regardless of which row any two machines are
assigned to.

3.5 Constraints on u i j variables


In any DRLP feasible solution, if machines i and k are not in the same row, and machines j and k are not in the same row;
then machines i and j must be in the same row. This is ensured by following constraints:
u i j + u jk + u ik ≥ 1, (1 ≤ i < j < k ≤ n).

3.6 Bounds on variables


If machine i is in its leftmost position within the interval [0, L], its centre is at abscissa 2i , hence we have: xi ≥ 2i . Now,
suppose that machine i is farthest from the x-axis origin within the interval [0, L], then its centre is at abscissa: L − 2i . Thus:
i i
2 ≤ xi ≤ L − 2 .
Trivially, we have:
di j ≥ 0 (1 ≤ i < j ≤ n) (3)

3.7 Further considerations


The proposed model can be improved by considering the incorporation of some valid inequalities and of symmetry breaking
constraints as described in this section.
International Journal of Production Research 39

3.7.1 Strengthening lower bound on distance variables


 
i +  j
If machine i is placed at the same row as machine j, the distance between the centres of i and j cannot be less than ,
  2
 +
hence we can strengthen (3) to: di j ≥ i 2 j u i j

3.7.2 Valid inequalities on distance variables


By definition di j = |xi − x j |, then, for arbitrary i, j, k ∈ N , the distance variables satisfy inequalities of the form dik ≤
di j + d jk .

3.7.3 Valid inequalities on u i j variables


The u i j variables define a partition of the machine set {1, 2, . . . , n}, thus the following inequalities (see, Grötschel and
Wakabayashi 1990) are satisfied by the u i j variables:

−u i j + u jk + u ik ≤ 1, +u i j − u jk + u ik ≤ 1,
+u i j + u jk − u ik ≤ 1, (1 ≤ i < j < k ≤ n).

3.7.4 Symmetry-breaking constraints


Consider two specific machines, say i∗, j∗; since there could be two symmetric optimal solutions (one with xi∗ < x j∗ and
the other with x j∗ < xi∗ ), we can try to eliminate one of these symmetric solutions by imposing: xi∗ ≤ x j∗ .
Note that if xi∗ = x j∗ in an optimal solution, then the constraint xi∗ ≤ x j∗ is redundant/ noneffective. Thus, we wish to
heuristically choose i∗ and j∗ so that xi∗ = x j∗ in an optimal solution. To this intent, define the total flow of machine i as
the sum of the amount of flow between machine i and all other machines. Intuitively, if the machine i∗ is the machine with
the largest total flow, then i∗ would be surrounded by machines whose flows to i∗ are large, while a machine j∗ with the
least flow to i∗ would be farther from i∗ in an optimal solution. If there are more than one such j∗, we prefer the one which
has the largest length, because if it is near i∗, it prevents (to a lengthier extent) the other machines from getting close to i∗.
Define Machine i∗ as

i∗ = arg max { ci j }
i∈N
j∈N , j=i

and break ties choosing the i∗ with the least index.


Now, define Machine j∗ as

j∗ = arg min {ci∗, j }


j∈N , j=i∗

and break ties choosing the j∗ with the largest length.

3.8 Proposed formulation


Our proposed formulation of the DRLP is given by:
Model M2:
Minimise

n−1 
n
ci j di j (4)
i=1 j=i+1

Subject to:

di j ≥ xi − x j,
di j ≥ x j − xi , (1 ≤ i < j ≤ n). (5)
 
i +  j
xj + ≤ xi + L(1 + ti j − u i j ),
2
40 A.R.S. Amaral

Table 1. Comparison of MIP formulations of the DRLP.

Number of variables
Model Number of constraints* Binary Continuous

(0) 5 n(n − 1) + 3n
MC T 2 n(n − 1) + 2n n(n − 1) + 2n
M1 5 n(n − 1) + n(n − 1)(n − 2) + 1 n(n − 1) 1 n(n − 1) + 2n
2 2
M2 5 n(n − 1) + 7 n(n − 1)(n − 2) + 1 n(n − 1) 1 n(n − 1) + n
2 6 2

*Excluding non-negativity constraints and bounds on variables.


 
i +  j
xi + ≤ x j + L(2 − ti j − u i j ),
2
(1 ≤ i < j ≤ n). (6)
 
i +  j
di j ≥ u i j , (1 ≤ i < j ≤ n). (7)
2

xi∗ ≤ x j∗ , i∗ = arg max{ ci j }, (8)
i∈N
j∈N , j=i
j∗ = arg min {ci∗, j }.
j∈N , j=i∗
+di j − d jk −dik ≤ 0, (9)
−di j + d jk − dik ≤ 0, (10)
−di j − d jk + dik ≤ 0, ( 1 ≤ i < j < k ≤ n). (11)
u i j + u jk + u ik ≥ 1, (12)
−u i j + u jk + u ik ≤ 1, (13)
+u i j − u jk + u ik ≤ 1, (14)
+u i j + u jk − u ik ≤ 1, (1 ≤ i < j < k ≤ n). (15)
i i
≤ xi ≤ L − , (1 ≤ i ≤ n). (16)
2 2
ti j ∈ {0, 1}, (1 ≤ i < j ≤ n). (17)
ui j ∈ {0, 1}, (1 ≤ i < j ≤ n). (18)

(0)
Table 1 compares the number of variables and constraints for the new model M2 and previous models: model MC T of
Chung and Tanchoco (2010) and model M1 of Amaral (2013a). Note that the new model (M2) has a smaller number of binary
(0)
variables and a smaller number of continuous variables than MC T . It has a larger number of constraints, most of which are
(0)
valid inequalities. Such inequalities play an important role for its improved performance relative to the model MC T .
For both M1 and M2, the number of constraints is O(n ) and they have the same number of binary variables. The new
3

model M2 has the least number of continuous variables among the three models.
To close this section, it is interesting to note that, sometimes, after an initial layout has been found based on quantitative
concerns, a phase of layout refinement motivated by qualitative considerations is started. In comparison with M1, the new
model M2 can be more intuitive for handling qualitative input. For example, in order to consider an alternative layout in
which two specific machines, say 3 and 7, are located at the same row, it is easy to add to M2 a constraint: u 3,7 = 1. With
M1 this constraint is α3,7 + α7,3 = 1. As another example, in order to have machine 2 to the left of machine 5 if they turn
out to be at the same row, one can add to M2 a constraint: t2,5 = 1. With model M1 this constraint may not be as intuitive:
α5,2 = 0.

4. Computational experiments
The new MIP model (denoted by M2), the MIP model MC(0)T , and the model of Amaral (2013a) (denoted by M1) were input
to the CPLEX 12.7.1.0 solver. The CPLEX solver reports the optimality gap, which is a measure of its progress on the way
to proving solution optimality (the optimality gap is the difference between the value of the incumbent solution and the best
MIP bound, divided by the best MIP bound). When the CPLEX solver terminates with a proved optimal solution, then the
International Journal of Production Research 41
(0)
Table 2. Performance of the proposed model M2 and Chung and Tanchoco’s model MC T .

(0)
Model MC T Model M2
Problem Optimal TC T GAP TM2 GAP
data Name n value (sec) % (sec) %

S9 9 1,179.0 19.69 0 7.30 0


Simmons (1969) S9H 9 2,293.0 * 9.27 82.13 0
S10 S10 1,351.0 108.55 0 23.93 0
S11 S11 3,424.5 * 20.81 450.84 0

*Aborted after exceeding an imposed time limit of 10 min.

Table 3. Performance of the proposed model M2 and model M1.

Model M1 Model M2
Problem Optimal TM1 GAP TM2 GAP
data n value (sec) % (sec) %

9 1,179.0 7.96 0 7.30 0


Simmons (1969) 9H 2,293.0 104.73 0 82.13 0
10 1,351.0 32.70 0 23.93 0
11 3,424.5 325.36 0 450.84 0

solver reports an optimality gap of zero. However, if the solver is aborted before optimality has been proven, the solver
reports a gap greater than zero (in this case, a large gap indicates that the solver was far to proving optimality at the point it
was aborted. Moreover, a large gap admits the possibility that the best integer solution found by the solver, if any, might be
far from optimality).
As already mentioned, it is assumed that, for all instances tested here, clearances have already been included in the lengths
of the machines. When a t-test was applied, a p-value of less than 0.05 was regarded to indicate a statistically significant
difference.

4.1 Computational tests on small instances


In this Section, the four largest instances introduced by Simmons (1969), which have n ≤ 11 are considered. In addition,
20 random instances (10 instances for each n ∈ {9, 10}) are tested. All instances are available from the author. The tests
pertaining Section 4.1 were run on an Intel® Core™ i35005U CPU 2GHz with 8 GB of RAM with the Windows 10 operating
system.

4.1.1 Experiments with instances of Simmons (1969)


(0)
The computational performance of M2 is compared with that of MC T in Table 2 on the Simmons’ instances. For each problem
instance, the first three columns of Table 2 present: the reference to the problem instance data, the number n of machines
and the optimal value obtained for the instance. The next four columns display for each model: the amount of computational
time spent (in seconds) and the CPLEX optimality gap attained. A time limit of 10 min was imposed, after which CPLEX is
to be aborted.
As it can be seen from Table 2, with the new model M2 each problem is solved within the specified time limit of 10 min.
The instance with n ≤ 10 is solved to optimality in less than 2 min and with n = 11 in less than 8 min. However, with Model
(0)
MC T the time limit is exceeded for the instance S9H (n = 9) and for the instance S11 (n = 11), giving large gaps of 9.27
and 20.81%, respectively.
A comparison of M2 with M1 is made in Table 3. With both models M1 and M2, the Simmons’ instances are solved
within the time limit of 10 min. In addition, with M2 faster solutions than M1 are obtained for the instances with n ≤ 10. In
the next section, Section M1 and M2 are tested on 20 random instances with n ∈ {9, 10} and in Section 4.2.3, M1 and M2
are further tested on larger instances.
42 A.R.S. Amaral

Table 4. Comparison of Model M2 with Model M1 on 20 instances: 10 instance for each with n ∈ {9, 10}.

Model M1 Model M2
Optimal TM1 TM2
Instance n value (sec) (sec)

Small.09-1 2393.0 86.92 75.19


Small.09-2 2266.0 78.28 151.59
Small.09-3 2241.0 104.43 202.69
Small.09-4 2231.5 125.06 103.80
Small.09-5 9 3980.0 119.63 111.63
Small.09-6 4108.0 149.51 309.95
Small.09-7 4089.0 164.77 261.84
Small.09-8 4154.0 156.23 144.62
Small.09-9 935.0 46.40 36.02
Small.09-10 1073.5 64. 05 46.14

Small.10-1 1385.0 34.89 33.40


Small.10-2 1437.0 49.56 38.78
Small.10-3 1452.5 40.88 50.10
Small.10-4 1313.5 35.15 44.46
Small.10-5 10 722.5 21.46 25.45
Small.10-6 792.0 37.72 57.21
Small.10-7 607.5 20.13 40.94
Small.10-8 529.0 11.36 17.39
Small.10-9 929.5 33.83 32.25
Small.10-10 828.0 50.93 42.37

4.1.2 Experiments with 20 instances with n ∈ {9, 10}


A comparison of run times of models M2 and M1 on 20 instances with n ∈ {9, 10} is presented in Table 4. A t-test was
applied to analyze whether the differences in performance with the models M2 and M1 were significant. The null hypothesis
is that the average run times with models M2 and M1 are the same. A p-value of 0.085 was obtained, which indicates that
there are no statistically significant difference between their average run times at the 0.05 level.

4.2 Computational tests on larger instances


In this section, four larger instances with n ∈ {12, 13} given by Amaral (2012), plus 14 random instances with n ∈ {11, 12, 13}
are considered. All of these instances are available from the author. The tests reported in Section 4.2 were run on an Intel®
Core™ i7-3770 CPU 3.40 GHZ with 8 GB of RAM with the Windows 8 operating system.

4.2.1 Verifying the effectiveness of the valid inequalities


To illustrate the effectiveness of the raised valid inequalities, Model M2 is compared with three of its variants subject to a
time limit of four hours on nine of the larger instances. These variants are obtained by removing valid inequalities from M2.
Inequalities (9)–(11) are relative to distance variables, while inequalities (13)–(15) involve u i j variables. Thus, the following
variants are considered:
• M2 – {(9)–(11)} – {(13)–(15)}
• M2 – {(13)–(15)}
• M2 – {(9)–(11)}
The results are presented in Table 5. The variant denoted by M2 – {(9)–(11)} – {(13)–(15)} is the worst performing one as
none of the instances are solved within the four-hour time limit and for one instance memory problems occur. The second
worst variant is M2 – {(13)–(15)}, with which no instances are solved within the time limit. With the variant M2 – {(9)–(11)}
all instances are solved but one (13a).
These results indicate that Inequalities (13)–(15) play an important role for the effectiveness of M2. Moreover, having
Inequalities (13)–(15) and Inequalities (9)–(11) allowed all of the instances to be solved within the time limit.
International Journal of Production Research 43

Table 5. Comparison of Model M2 with its variants.

M2– {(9)–(11)} M2 M2
– {(13)–(15)} – {(13)–(15)} – {(9)–(11)} M2
Time GAP Time GAP Time GAP Time GAP
Instance (sec) % (sec) % (sec) % (sec) %

11a * 39.86 * 29.32 504.88 0 343.90 0


11b * 35.57 * 35.62 6,418.56 0 277.52 0
11c * 42.07 * 38.33 591.91 0 344.92 0
12a * 35.92 * 32.51 859.04 0 374.63 0
12b ** 34.95 * 28.44 636.42 0 403.92 0
12c * 31.57 * 33.63 847.58 0 411.28 0
13a * 39.71 * 38.63 * 13.23 12,481.17 0
13b * 35.49 * 32.03 4,790.48 0 10,935.62 0
13c * 40.55 * 36.26 3,318.73 0 1,450.49 0

*Aborted after exceeding an imposed time limit of 4 h.


**Aborted due to running out of memory after 12,864.11 s.


Table 6. Comparison of Model M2 with Model M2.


Model M2 Model M2
Optimal TM2
TM2
Instance n value (sec) (sec)

11a 5,559.0 506.4 343.90


11b 3,655.5 212.8 277.52
11c 11 3,832.5 271.0 344.92
11d 906.5 67.2 69.26
11e 578.0 52.7 63.51
11f 825.5 39.0 31.64

12a 1,493.0 390.8 374.63


12b 1,606.5 221.0 403.92
12c 12 2,012.5 435.0 411.28
12d 1,107.0 103.8 115.37
12e 1,066.0 230.4 120.45
12f 997.5 223.6 150.35

13a 2,456.5 7,395.3 12,481.17


13b 2,864.0 4,403.3 10,935.62
13c 13 4,136.0 1,697.9 1,450.49
13d 6,164.5 3,163.8 3,380.06
13e 6,502.5 7,138.3 9,144.51
13f 7,699.5 13,774.4 11,920.70

4.2.2 Verifying the effect of using a different symmetry-breaking constraint


Here, we verify the effect of using a different symmetry-breaking constraint on the performance of M2. Let us consider a
simpler symmetry-breaking constraint, obtained by just selecting two arbitrary machines, like 2 and 1 in Constraint (8), rather
than carefully selecting the two machines. Denote by M2 the model M2 with the simpler symmetry-breaking constraint in
place of the more elaborated one.
Table 6 shows a comparison of M2 and M2.
A t-test is applied to the results in Table 6. The null hypothesis is that the average run time for M2 and for M2 are
equal. The p-value obtained is 0.188. Because p-value > 0.05, we do not reject the null hypothesis. Therefore, there is no
statistically significant difference in average run time between M2 and M2. This result raises a question of whether it is
44 A.R.S. Amaral

Table 7. Comparison of Model M2 with Model M1.

Model M1 Model M2
Optimal TM1 TM2
Instance n value (sec) (sec)

11a 5,559.0 126.20 343.90


11b 3,655.5 145.65 277.52
11c 11 3,832.5 153.47 344.92
11d 906.5 101.13 69.26
11e 578.0 17.24 63.51
11f 825.5 20.02 31.64

12a 1,493.0 155.70 374.63


12b 1,606.5 140.37 403.92
12c 12 2,012.5 198.88 411.28
12d 1,107.0 55.63 115.37
12e 1,066.0 92.45 120.45
12f 997.5 167.69 150.35

13a 2,456.5 4,968.92 12,481.17


13b 2,864.0 953.51 10,935.62
13c 13 4,136.0 568.88 1,450.49
13d 6,164.5 1,325.64 3,380.06
13e 6,502.5 1,636.78 9,144.51
13f 7,699.5 15,199.36 11,920.70

Table 8. Comparison of Model

M2 with Model M1.

Model M1 Model

M2
Optimal TM1 T
M2
Instance n value (sec) (sec)

11a 5,559.0 126.20 314.89


11b 3,655.5 145.65 373.91
11c 11 3,832.5 153.47 258.58
11d 906.5 101.13 57.14
11e 578.0 17.24 36.03
11f 825.5 20.02 35.50

12a 1,493.0 155.70 154.83


12b 1,606.5 140.37 136.24
12c 12 2,012.5 198.88 145.91
12d 1,107.0 55.63 54.72
12e 1,066.0 92.445 419.48
12f 997.5 167.686 131.81

13a 2,456.5 4,968.92 4,678.89


13b 2,864.0 953.51 3,339.22
13c 13 4,136.0 568.88 751.70
13d 6,164.5 1,325.64 4,238.00
13e 6,502.5 1,636.78 2,942.58
13f 7,699.5 15,199.36 8,402.15
International Journal of Production Research 45

possible to develop a method to choose two machines in Constraint (8) so as to improve average run time relative to Model
M2.

4.2.3 Comparison of the proposed model with Model M1


Table 7 shows a comparison of run times obtained with models M2 and M1. It is seen that, for most of the tested instances,
run time with Model M2 lags behind that with M1. However, M2 performed 31 s. faster for one instance with n = 11;
17 s. faster for one instance with n = 12; and 3278 s. faster for one instance with n = 13. Therefore, a t-test was applied
to investigate statistical significance of the hypothesis that their average run times differ. The p-value obtained was 0.085,
which is greater than 0.05, and then there is no significant difference between average run times of the two models.
As an additional test, denote by
M2 the model M2 with Constraint (1) in place of Constraint (16), i.e. we replace the
current bound on variable xi with the bound of Brunese and Tanchoco (2013). Let us consider a comparison of run times
obtained with models
M2 and M1. The results are presented in Table 8. A t-test was applied to determine whether a significant
difference may exist between the average run times of
M2 and M1. The obtained p-value = 0.957 > 0.05. Therefore, no
significant difference can be claimed.

5. Conclusions
We considered the double row layout problem (DRLP), which is NP-hard. A mixed-integer programming (MIP) model was
derived for the problem, including valid inequalities and a symmetry-breaking constraint. The computational results showed
that the new model outperforms the MIP model of Chung and Tanchoco (2010). In addition, the new model was found to
present similar performance to the model of Amaral (2013a), as no statistically significant difference in performance exists
between these two models. Both models can be solved optimally for instances up to n = 13 machines. The new model has
the advantage of being more intuitive for handling qualitative inputs.
Future research should investigate how to increase the size of problems that can be efficiently solved to optimality. The
development of meta-heuristics for the problem is also an interesting topic for future research.

Acknowledgements
I would like to express my appreciation to the reviewers for their useful comments.

Disclosure statement
No potential conflict of interest was reported by the author.

Funding
This work was supported by CAPES [grant number 99999.002643/2015-04].

References

Ahonen, H., A. G. de Alvarenga, and A. R. S. Amaral. 2014. “Simulated Annealing and Tabu Search Approaches for the Corridor Allocation
Problem.” European Journal of Operational Research 232 (1): 221–233. https://ideas.repec.org/a/eee/ejores/v232y2014i1p221-233.
html
Amaral, André R. S. 2006. “On the Exact Solution of a Facility Layout Problem.” European Journal of Operational Research 173 (2): 508–
518. ISSN 0377-2217. http://www.sciencedirect.com/science/article/B6VCT-4FTWJK5-2/2/f4b62b39fd1bdd3e3419093dbe333638
Amaral, André R. S. 2008. “An Exact Approach to the One-dimensional Facility Layout Problem.” Operations Research 56 (4): 1026–1033.
http://or.journal.informs.org/cgi/content/abstract/56/4/1026
Amaral, André R. S. 2009. “A New Lower Bound for the Single Row Facility Layout Problem.” Discrete Applied Mathematics 157 (1): 183–
190. ISSN 0166-218X. http://www.sciencedirect.com/science/article/B6TYW-4T24FJY-2/2/e315136eb851a49eb5bc769a901bd0d7
Amaral, André R. S. 2012. “The Corridor Allocation Problem.” Computers & Operations Research 39 (12): 3325–3330. ISSN 0305-0548.
http://www.sciencedirect.com/science/article/pii/S0305054812000962
Amaral, André R. S. 2013a. “Optimal Solutions for the Double Row Layout Problem.” Optimization Letters 7: 407–413. ISSN 1862-4472.
doi:10.1007/s11590-011-0426-8.
Amaral, André R. S. 2013b. “A Parallel Ordering Problem in Facilities Layout.” Computers & Operations Research 40 (12): 2930–2939.
ISSN 0305-0548. http://www.sciencedirect.com/science/article/pii/S0305054813001767
46 A.R.S. Amaral

Amaral, André R. S., and Adam N. Letchford. 2013. “A Polyhedral Approach to the Single Row Facility Layout Problem.” Mathematical
Programming 141 (1–2): 453–477. ISSN 0025-5610. doi:10.1007/s10107-012-0533-z.
Anjos, Miguel F., Andrew Kennings, and Anthony Vannelli. 2005. “A Semidefinite Optimization Approach for the Single-row
Layout Problem with Unequal Dimensions.” Discrete Optimization 2 (2): 113–122. http://www.sciencedirect.com/science/article/
B7GWV-4GH49WS-1/2/15a4a681fd6869d01a17eeeba8863588
Anjos, Miguel F., andAnthony Vannelli. 2008. “Computing Globally Optimal Solutions for Single-row Layout Problems Using Semidefinite
Programming and Cutting Planes.” INFORMS Journal on Computing 20 (4): 611–617. ISSN 1055-6788. http://joc.journal.informs.
org/cgi/content/abstract/20/4/611
Anjos, Miguel F., and Ginger Yen. 2009. “Provably Near-optimal Solutions for Very Large Single-row Facility Layout Problems.”
Optimization Methods and Software 24 (4): 805–817. ISSN 1572-5286.
Brunese, P. A., and J. M. A. Tanchoco. 2013. “On Implied Within-building Constraints for Machine Layout.” International Journal of
Production Research 51 (6): 1937–1952. doi:10.1080/00207543.2012.721571.
Chung, Jaewoo, and J. M. A. Tanchoco. 2010. “The Double Row Layout Problem.” International Journal of Production Research 48
(3):709–727. ISSN 0020-7543. http://www.informaworld.com/10.1080/00207540802192126
de Alvarenga, Arlindo G., Francisco J. Negreiros-Gomes, and Mario Mestria. 2000. “Metaheuristic Methods for a Class of the Facility
Layout Problem.” Journal of Intelligent Manufacturing 11: 421–430. ISSN 0956-5515. doi:10.1023/A:1008982420344.
Drira, Amine, Henri Pierreval, and Sonia Hajri-Gabouj. 2007. “Facility Layout Problems: A Survey.” Annual Reviews in Control 31 (2):
255–267. ISSN 1367-5788. http://www.sciencedirect.com/science/article/pii/S1367578807000417
Ficko, M., M. Brezocnik, and J. Balic. 2004. “Designing the Layout of Single- and Multiple-rows Flexible Manufacturing System by
Genetic Algorithms.” Journal of Materials Processing Technology 157–158 :150–158. ISSN 0924-0136. http://www.sciencedirect.
com/science/article/pii/S0924013604010544. Achievements in Mechanical and Materials Engineering Conference.
Gen, Mitsuo, Kenichi Ida, and Chunhui Cheng. 1995. “Multirow Machine Layout Problem in Fuzzy Environment Using Genetic
Algorithms.” Computers & Industrial Engineering 29 (1): 519–523. ISSN 0360-8352. http://www.sciencedirect.com/science/article/
pii/036083529500127M
Ghosh, Diptesh, and Ravi Kothari. 2012. “Population Heuristics for the Corridor Allocation Problem.” IIMA Working Papers WP2012-
09-02. Indian Institute of Management Ahmedabad, Research and Publication Department. http://EconPapers.repec.org/RePEc:iim:
iimawp:11453
Grötschel, M., and Y. Wakabayashi. 1990. “Facets of the Clique Partitioning Polytope.” Mathematical Programming 47: 367–387. ISSN
0025-5610.
Guan, Jian, and Geng Lin. 2016. “Hybridizing Variable Neighborhood Search with Ant Colony Optimization for Solving the Single Row
Facility Layout Problem.” European Journal of Operational Research 248 (3): 899–909. ISSN 0377-2217. http://www.sciencedirect.
com/science/article/pii/S0377221715007316
Hassan, M. M. D. 1994. “Machine Layout Problem in Modern Manufacturing Facilities.” International Journal of Production Research
32 (11): 2559–2584. doi:10.1080/00207549408957084.
Heragu, Sunderesh S., and Attahiru Sule Alfa. 1992. “Experimental Analysis of Simulated Annealing Based Algorithms for the Layout
Problem.” European Journal of Operational Research 57 (2): 190–202. ISSN 0377-2217. http://www.sciencedirect.com/science/
article/pii/0377221792900428
Heragu, Sunderesh S., and Andrew Kusiak. 1988. “Machine Layout Problem in Flexible Manufacturing Systems.” Operations Research
36 (2): 258–268. http://or.journal.informs.org/cgi/content/abstract/36/2/258
Heragu, Sunderesh S., and Andrew Kusiak. 1991. “Efficient Models for the Facility Layout Problem.” European Journal
of Operational Research 53 (1): 1–13. ISSN 0377-2217. http://www.sciencedirect.com/science/article/B6VCT-48M3FS1-46/2/
1df1fc95473f7f71734cdae03e1d490e
Hungerländer, Philipp, and Franz Rendl. 2012. “A Computational Study and Survey of Methods for the Single-row Facility Layout
Problem.” Computational Optimization and Applications. 55 (1): 1–20. ISSN 0926-6003. doi:10.1007/s10589-012-9505-8.
Kalita, Zahnupriya, and Dilip Datta. 2014. “Solving the Bi-objective Corridor Allocation Problem Using a Permutation-based Genetic
Algorithm.” Computers & Operations Research 52: 123–134. ISSN 0305-0548. http://www.sciencedirect.com/science/article/pii/
S0305054814001919
Keller, Birgit, and Udo Buscher. 2015. “Single Row Layout Models.” European Journal of Operational Research 245 (3): 629–644. ISSN
0377-2217. http://www.sciencedirect.com/science/article/pii/S037722171500209X
Kothari, Ravi, and Diptesh Ghosh. 2012. “The Single Row Facility Layout Problem: State of the Art.” OPSEARCH 49: 442–462. ISSN
0030-3887. doi:10.1007/s12597-012-0091-4.
Kothari, Ravi, and Diptesh Ghosh. 2013a. “An Efficient Genetic Algorithm for Single Row Facility Layout.” Optimization Letters 9 (2):
679–690. ISSN 1862-4472. doi:10.1007/s11590-012-0605-2.
Kothari, Ravi, and Diptesh Ghosh. 2013b. “Insertion Based Lin-Kernighan Heuristic for Single Row Facility Layout.” Computers &
Operations Research 40 (1): 129–136. ISSN 0377-2217. http://www.sciencedirect.com/science/article/pii/S0305054812001219
Kothari, Ravi, and Diptesh Ghosh. 2013c. “Tabu Search for the Single Row Facility Layout Problem Using Exhaustive 2-opt and
Insertion Neighborhoods.” European Journal of Operational Research 224 (1): 93–100. ISSN 1862-4472. http://www.sciencedirect.
com/science/article/pii/S0377221712005942
Kothari, Ravi, and Diptesh Ghosh. 2014. “A Scatter Search Algorithm for the Single Row Facility Layout Problem.” Journal of Heuristics
20 (2): 125–142. ISSN 1572-9397. doi:10.1007/s10732-013-9234-x.
International Journal of Production Research 47

Kouvelis, Panagiotis. 1992. “Design and Planning Problems in Flexible Manufacturing Systems: A Critical Review.” Journal of Intelligent
Manufacturing 3 (2): 75–99. ISSN 1572-8145. doi:10.1007/BF01474748.
Love, Robert F., and Jsun Y. Wong. 1976. “On Solving a One-dimensional Space Allocation Problem with Integer Programming.” INFOR
14 (2): 139–143. ISSN 0315-5986.
Murray, C. C., X. Q. Zuo, and A. E. Smith. 2012. “An Extended Double Row Layout Problem.” In Proceedings of the 12th International
Material Handling Research Colloquium, 535–550. France: Gardanne.
Murray, Chase C., Alice E. Smith, and Zeqiang Zhang. 2013. “An Efficient Local Search Heuristic for the Double Row Layout Problem with
Asymmetric Material Flow.” International Journal of Production Research 51 (20): 6129–6139. doi:10.1080/00207543.2013.803168.
Ozcelik, Feristah. 2012. “A Hybrid Genetic Algorithm for the Single Row Layout Problem.” International Journal of Production Research
50 (20): 5872–5886. doi:10.1080/00207543.2011.636386.
Palubeckis, Gintaras. 2017. “Single Row Facility Layout Using Multi-start Simulated Annealing.” Computers & Industrial Engineering
103: 1–16. ISSN 0360-8352. http://www.sciencedirect.com/science/article/pii/S0360835216303667
Peters, B. A., and Taho Yang. 1997. “Integrated Facility Layout and Material Handling System Design in Semiconductor Fabrication
Facilities.” IEEE Transactions on Semiconductor Manufacturing 10 (3): 360–369. ISSN 0894-6507.
Rubio-Sánchez, Manuel, Micael Gallego, Francisco Gortázar, and Abraham Duarte. 2016. “GRASP with Path Relinking for the Single Row
Facility Layout Problem.” Knowledge-Based Systems 106: 1–13. ISSN 0950-7051. http://www.sciencedirect.com/science/article/pii/
S0950705116301204
Samarghandi, Hamed, and Kourosh Eshghi. 2010. “An Efficient Tabu Algorithm for the Single Row Facility Layout Problem.”
European Journal of Operational Research 205 (1): 98–105. ISSN 0377-2217. http://www.sciencedirect.com/science/article/pii/
S0377221709009072
Samarghandi, Hamed, Pouria Taabayan, and Farzad Firouzi Jahantigh. 2010. “A Particle Swarm Optimization for the Single Row Facility
Layout Problem.” Computers & Industrial Engineering 58 (4): 529–534. ISSN 0360-8352. http://www.sciencedirect.com/science/
article/pii/S0360835209003027
Semiconductor Industry Association. 2017. Global Sales Report. Technical Report, Feb.
Simmons, Donald M. 1969. “One-dimensional Space Allocation: An Ordering Algorithm.” Operations Research 17 (5): 812–826. http://
or.journal.informs.org/cgi/content/abstract/17/5/812
Solimanpur, M., Prem Vrat, and Ravi Shankar. 2005. “An Ant Algorithm for the Single Row Layout Problem in Flexible Manufacturing
Systems.” Computers & Operations Research 32 (3): 583–598. ISSN 0305-0548. http://www.sciencedirect.com/science/article/
B6VC5-49W33JC-1/2/24bc8615d8cb398a5c2c5421d046d3cb
Tubaileh, Allan, and Jamal Siam. 2017. “Single and Multi-row Layout Design for Flexible Manufacturing Systems.” International Journal
of Computer Integrated Manufacturing 30 (12): 1316–1330. doi:10.1080/0951192X.2017.1314013.
Wang, Hsiao-Fan, and Ching-Min Chang. 2015. “Facility Layout for an Automated Guided Vehicle System.” Procedia Computer Science
55: 52–61. http://www.sciencedirect.com/science/article/pii/S1877050915014829
Wang, Shengli, Xingquan Zuo, Xueqing Liu, Xinchao Zhao, and Jianqiang Li. 2015. “Solving Dynamic Double Row Layout Problem via
Combining Simulated Annealing and Mathematical Programming.” Applied Soft Computing 37: 303–310. ISSN 1568-4946. http://
www.sciencedirect.com/science/article/pii/S1568494615005244
Yang, Taho, and Brett A. Peters. 1997. “A Spine Layout Design Method for Semiconductor Fabrication Facilities Containing
Automated Material-handling Systems.” International Journal of Operations & Production Management 17 (5): 490–501.
doi:10.1108/01443579710167212.
Zhang, Zeqiang, and Chase C. Murray. 2012. “A Corrected Formulation for the Double Row Layout Problem.” International Journal of
Production Research 50 (15): 4220–4223. http://www.tandfonline.com/doi/abs/10.1080/00207543.2011.603371
Zuo, X. Q., C. C. Murray, and A. E. Smith. 2016. “Sharing Clearances to Improve Machine Layout.” International Journal of Production
Research 54 (14): 4272–4285. ISSN 1545-5955. doi:10.1080/00207543.2016.1142134.
Zuo, X., C. C. Murray, and A. E. Smith. 2014. “Solving an Extended Double Row Layout Problem Using Multiobjective Tabu Search and
Linear Programming.” IEEE Transactions on Automation Science and Engineering 11 (4): 1122–1132.

You might also like