This article was downloaded by: [North Carolina State University]

On: 05 August 2013, At: 16:09

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

International Journal of Production Research

Publication details, including instructions for authors and subscription information:
[Link]

A multi-objective mixed-integer programming model

for a multi-floor facility layout
a a a
Jonathan Hathhorn , Esra Sisikoglu & Mustafa Y. Sir
a
Department of Industrial and Manufacturing Engineering , University of Missouri ,
Columbia , MO , 65211 , USA
Published online: 14 Mar 2013.

To cite this article: Jonathan Hathhorn , Esra Sisikoglu & Mustafa Y. Sir (2013) A multi-objective mixed-integer
programming model for a multi-floor facility layout, International Journal of Production Research, 51:14, 4223-4239, DOI:
10.1080/00207543.2012.753486

To link to this article: [Link]

PLEASE SCROLL DOWN FOR ARTICLE

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

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
[Link]/page/terms-and-conditions
 International Journal of Production Research, 2013
Vol. 51, No. 14, 4223–4239, [Link]

A multi-objective mixed-integer programming model for a multi-floor facility layout

Jonathan Hathhorn, Esra Sisikoglu* and Mustafa Y. Sir

Department of Industrial and Manufacturing Engineering, University of Missouri, Columbia MO 65211, USA
(Received 13 April 2012; final version received 17 November 2012)

We consider a multi-floor facility layout problem in which the overall length and width of the facility, the size and
location of each department, the number and the location of elevators and the number of floors in the facility are all
modelled as decision variables. We adapt a linear approximation scheme to represent the area of each department. We
consider two objective functions in our model, namely minimising material handling and facility building costs, and
Downloaded by [North Carolina State University] at 16:09 05 August 2013

propose a lexicographic ordering technique to handle multiple objectives. The numerical experiments show that the slack
used in the lexicographic ordering approach has a significant impact on the optimal solution. The experiments also show
that the material handling cost can be significantly reduced in a multi-floor facility compared with a single-floor facility.
Keywords: facility layout; mixed integer linear programming; multi-criteria decision making; optimisation; lexicographic
ordering; multi-floor facility layout

1. Introduction
Building a new facility involves investing a significant amount of resources in a project. Once construction of the
facility is complete, it is even harder and costlier to redesign or correct flaws and other issues. Therefore, the design
phase of such a project is of critical importance. The best way to place departments or other units such as elevators in a
facility has to be carefully determined. Depending on the available resources, the problem objectives and constraints
may change. For example, in the case of scarce land, the departments must be fit into an area of given width and length.
In other cases, however, the length and width of the facility might be part of the set of decision variables and one might
need to determine the size (length and width) of the facility so that it is large enough to efficiently place all the
departments and not waste any resources.
The specific facility layout problem we consider in this paper was inspired from a hospital design. The general
functions of hospitals are very much interrelated, however the relationships between functional units within and between
each general function are not fully understood by hospital management. Tzortzopoulos et al. (2008) comment that many
hospital processes tend to be organised around functional units. Many problems faced in healthcare service delivery
today are directly related to poor interactions between services and building design (Carr 2011). Therefore, managers
and designers must adopt an integrated and comprehensive approach when making building design decisions
(Tzortzopoulos et al. 2008).
Often, hospitals are designed as multi-floor, integrated facilities. Therefore, a primary hospital design issue is the
hospital layout – the layout of departments and other units (e.g., elevators) within the hospital. However, little research
has been done to determine when a multi-floor facility – as opposed to a single-floor facility – is a more efficient design
choice. A non-optimal hospital design results in congestion, longer travel time between functional units, and inefficient
space utilisation, leading to waste of valuable resources. Flow congestion and inefficient space utilisation increase
waiting and throughput times, sometimes resulting in a negative impact both on the quality of service delivered and the
capacity utilisation of the hospital, and always on the patients’ impression of the hospital’s efficiency and competence
(Vos, Groothuis, and van Merode 2007). Patient flow should be improved across the system as a whole rather than
within isolated units (Haraden and Resar 2004). Moreover, the location and size of space allocated to each functional
unit should be carefully determined in order to minimise very high building costs.
Motivated by the issues concerning the layout of hospitals, we model a multi-objective multi-floor facility layout
problem to minimise both the cost of flow between departments (referred to as the material handling cost in the
remainder of the paper) and also the total building cost. In our comprehensive model, the position, length and width of

*Corresponding author. Email: sisikoglue@[Link]

Ó 2013 Taylor & Francis
 4224 J. Hathhorn et al.

departments, the number of floors and size of the facility, the number and location of elevators are all decision variables.
We use an approximation to linearise the minimum area requirement constraints. We also use a lexicographic ordering
approach to prioritise the two objectives and solve two sub-problems in sequence. Moreover, we compare the effect of
having multiple floors on the material handling cost. The flowchart in Figure 1 summarises the different components of
our model and the order they are presented throughout the paper.
In the next section we present a literature review of the classical facility layout problem and multi-objective and
multi-floor design models. Section 3 presents a formulation to optimise a multiple floor facility layout that can be used
for hospital layout. Section 4 demonstrates our multi-floor facility layout optimisation formulation with an example
instance. We conclude with possible future research and extensions.

2. Literature review
The facility layout problem (Meller and Gau 1996, Singh and Sharma 2006) is concerned with solving the physical
organisational puzzle within a production system to minimise the material handling flow costs between departments.
Solutions arrange a discrete number of departments within the bounds of a facility. The most common objective is to
Downloaded by [North Carolina State University] at 16:09 05 August 2013

minimise the material handling cost – the sum of the distances between each department multiplied by the flow or
weight between those departments. There are many other important objectives such as maximising a closeness rating
measure between departments (Rosenblatt 1979) and minimising land and facility construction costs (Georgiadis et al.
1999). With any set of objectives, the layout must also satisfy numerous constraints. Constraints might include
departmental area requirements (Lacksonen 1997), departmental location restrictions, aisle network requirements (Wu
and Appleton 2002), or department overlapping restrictions. The facility layout problem formulation can also be applied
in many other applications, e.g. the arrangement of electronic components on a circuit board (Duman and Or 2007).
In developing solutions to the facility layout problem, researchers have utilised various modelling approaches,
formulations, and solution algorithms. Three of the most popular modelling approaches for the facility layout problem
are quadratic assignment (Lawler 1963), mixed-integer programming (Montreuil 1990), and the graph-theoretic approach
(Hassan and Hogg 1987).
The quadratic assignment problem (QAP) (Lawler 1963) is one of the most common facility layout approaches. The
objective is to minimise the material handling cost, which is equal to total flow times distance. The solution is typically
obtained by assigning a discrete set of equal-sized departments to a discrete number of locations in order to minimise
the flow and distance between departments. Finding solutions to the QAP is known to be NP-hard, so as the number of
departments to be assigned in the facility layout problem increases, the model complexity and solution time increase
enormously.
Montreuil (1990) extended the QAP for the continuous layout representation and proposed a mixed-integer
programming (MIP) formulation. This MIP formulation has the same objective as the QAP approach: to minimise the

Multi-floor Facility Layout Design (Sec. 3)

• Defining Multiple Objectives (Sec. 3.2.1)
• Floor Constraints (Sec 3.2.2)
• Elevator Constraints (Sec. 3.2.3)
• Department Dimension Constraints (Sec 3.2.4)
• Non-overlapping Constraints (Sec. 3.2.5)
• Distance Constraints (Sec. 3.2.6)
• Facility Bounding Constraints (Sec. 3.2.7)

Approximation to Linearise Area Constraints (Sec. 3.3)

Balancing Multiple Objectives using Lexicographic

Ordering Approach (Sec. 3.4)

Numerical Results (Sec. 4)

• Symmetry Breaking Constraints (Sec. 4.2)

Figure 1. Flowchart showing the organisation of the paper.

 International Journal of Production Research 4225

material handling cost within a facility. However, unlike the QAP, which assigns a discrete number of equal-sized
departments to a set of discrete locations, the MIP approach attempts to position a discrete number of unequal-sized
departments into a continuous solution space (constrained by the overall facility size). Similar to the QAP, the MIP
formulation quickly becomes intractable as the number of departments increases (Konak et al. 2006). In addition, the
incorporation of practical considerations, such as establishing an aisle network or locations for pick-up and drop-off,
severely increase the complexity of the model.
The graph-theoretic approach varies significantly from the more common facility layout approaches, MIP and QAP
(Hassan and Hogg 1987). While the objective of MIP and QAP is to minimise the material handling cost within a
facility, the objective of the graph-theoretic approach is to maximise the weighted relationships between adjacent
departments. It is assumed that the desirability of locating each pair of departments adjacent to each other is known.
The graph-theoretic approach uses department relationships to construct a maximally weighted planar graph. Nodes in
the graph represent departments and arcs represent adjacency relationships between departments. The number of
adjacency relationships is the crucial constraint of this model: a department may be limited, for example, to six adjacent
departments to represent a certain manufacturing department. Once the graph is optimised it can be converted into a
block layout so that department shapes and sizes can be considered. Similar to MIP and QAP, tractability becomes an
Downloaded by [North Carolina State University] at 16:09 05 August 2013

issue as the number of departments increases, therefore several heuristics have been proposed to solve the facility layout
problem using the graph-theoretic approach (see, for example, Leung (1992)).
In addition, there have been numerous studies that use heuristic approaches such as genetic algorithms (Suresh, Vinod,
and Sahu 1995, Islier 1998, Mak, Wong, and Chan 1998, Rajasekharan, Peters, and Yang 1998, El-Baz 2004), simulated
annealing (Meller and Bozer 1996) and particle swarm optimisation (Jolai, Tavakkoli-Moghaddam, and Taghipour 2012)
for the facility layout problem, and tabu search (Abdinnour-Helm and Hadley 2000) and genetic algorithm-based heuristics
(Kochhar 1998, Lee, Roh, and Jeong 2005) for multi-floor facility layout problems. In this paper, we concentrate on
solving a mixed-integer programming model for the multi-floor and multi-objective facility layout problem by linearising
the area constraints and show the effect of using multiple floors on the quality of the solutions. The related literature on the
multi-floor facility layout problem and our contributions to the literature are described in the next section.

2.1 Multi-floor facility layout problem

The multi-floor facility layout problem is an extension to the single-floor facility layout problem in which a discrete
number of departments are assigned to a multiple-floor facility. The main objective of the multi-floor facility layout is
identical to the QAP for a single floor facility layout: minimise material handling cost by assigning departments to effec-
tive positions in the facility. However, the additional decision variables – such as the number of floors, the number of
elevators, and the location of elevators – add complexity to the problem. They also add additional constraints. For
example, multi-floor layouts can be infeasible if the total area of departments assigned to any floor is greater than the
total area available on the floor (Bozer, Meller, and Erlebacher 1994). Yet, multi-floor layouts may be necessary when
single-floor facilities are not appropriate. Buildings with multiple floors may be necessary for situations where the cost
of land is high, the amount of land is limited, or when a compact building allows for more efficient environmental or
operational control (Goetschalckx and Irohara 2007).
Some of the facility layout problem literature explores extensions to the QAP, MIP, and graph-theoretic approaches
with respect to multi-floor facility layout. Extensions vary in terms of assumptions of department sizes, facility size, and
elevators. Hahn and Krarup (2001) review the history of a QAP formulation applied to a multi-floor hospital facility.
Their model assumes a discrete number of equal-sized hospital departments in addition to an upper bound on the length,
width, and number of floors. Thus, along with the decision of where to effectively assign the hospital’s departments
between floors and on each floor, their model must decide the number of floors and the size of each floor. Later
heuristics, including simulated annealing, tabu search, and branch and bound, have been used to find solutions to layout
problems with larger numbers of departments (Drira, Pierreval, and Hajri-Gabouj 2007). Hahn and Krarup’s formulation
does not specify a flow path to the next floor via stairs or an elevator. Thus, the layout formulation does not encompass
all the issues within a multi-floor facility layout.
To address the case where department shapes are not equal, Bozer, Meller, and Erlebacher (1994) extend two single-
floor facility layout algorithms to create a multi-floor facility layout approach called MULTIPLE. Their research combines
two well-known algorithms: CRAFT and the space-filling curve. CRAFT (Buffa, Armour, and Vollmann 1964) begins with
an initial layout and performs exchanges between departments to reduce the layout cost. Since CRAFT is only capable of
exchanging departments adjacent to one another and equal in area, Bozer Meller, and Erlebacher combine it with the
space-filling curve (Sagan 1994) space in an attempt to allow flexible department exchanges and departmental area require-
ments during the improvement procedure. MULTIPLE is capable of restricting departments to certain floors and having
 4226 J. Hathhorn et al.

fixed floors. However, it assumes that the floor dimensions, number of floors, and number and positions of elevators are
known a priori. Furthermore, the department locations and sizes are approximated by a set of discrete, equal-sized squares.
Depending on the application, these assumptions may be too strong. However, they do result in a much simpler computa-
tion. Even for relatively large problems, the computation time with MULTIPLE is short. Lee, Han, and Roh (2003) present
the multi-floor facility layout problem with inner walls and passages as a combination of a MIP model and a
graph-theoretic model. Such a model can be used for the re-layout process of an old building or the layout of a new build-
ing in which inner walls that outline existing rooms and hallways are present and fixed. Given the initial layout skeleton of
the facility, an optimal layout is pursued by allocating departments within the layout skeleton with respect to multiple
objective functions, including minimising the material handling cost and maximising the inter-departmental adjacencies.
Lee Han, and Roh made several assumptions to improve the tractability of their multi-objective MIP model. Some of the
primary assumptions include inter-departmental material flow values, inter-departmental relationship values, the number of
floors, the number of elevators, the number, positions and sizes of elevators, and the boundary shape for each floor. How-
ever, even with these assumptions, the model quickly becomes intractable as the number of departments increases. Lee
Han, and Roh propose a genetic algorithm to solve their model and apply it to a multi-deck compartment ship layout
design problem.
Downloaded by [North Carolina State University] at 16:09 05 August 2013

Goetschalckx and Irohara (2007) also developed a mixed-integer programming model for the multi-floor facility
layout problem. In particular, Goetschalckx and Irohara attempt to create efficient formulations for multi-floor facility
layout problems with elevators. In their model, both the number and positions of elevators are decision variables.
Goetschalckx and Irohara investigate two cases: one in which full-service elevators stop at all floors and another in
which some full-service elevators can stop at all floors in addition to less expensive elevators that do not provide service
to all floors. For each case, the assumed parameters include the number of floors, the maximum number of elevators,
the length and width of each department, and the material flow and cost between departments. The model assumes
lower and upper bounds for the length and width of each floor. This allows each floor to be a different size as long as
the length and width of each floor is equal to or shorter than the floor directly below it. With these additional decision
variables, the formulation results in a large mixed-integer model that is very difficult and time-consuming to solve
optimally. Goetschalckx and Irohara alter the formulations to eliminate symmetry and utilise valid lower bounds on
distances in order to reduce computation times.

Table 1. Comparison of FLP models.

iou
ara

org
roh

ge
pa
dI
h

Pa
up

n
Ro

an

t io
rar

nd
kx
nd

ul a
dK

is a
alc
na
E

orm
IPL

ch
an

atz
Ha
T

LT

ets
AF

rF
hn

tsi
e,
MU
CR

Ou
Go
Ha

Pa
Le

Discrete Department Locations X X X

Continuous Department Locations X X X X
Fixed Equal Department Shape X X
Fixed Unequal Department Shape X X
Variable Discrete Department Shape X
Variable Continuous Department Shape X X
Single Floor X
Fixed Number of Floors X X X X
Variable Number of Floors X X
No Elevators X X X
Fixed Number of Elevators X X
Variable Number of Elevators X X
No Elevators X X X
Fixed Elevator Location X X
Variable Continuous Elevator Location X X
Fixed Facility Size X X X X X
Variable Continuous Facility Size X X
 International Journal of Production Research 4227

Patsiatzis and Papageorgiou (2002) propose a multi-floor process plant model with a comprehensive objective
function to minimise the total plant layout cost by determining the number of floors, land area, equipment–floor alloca-
tion and detailed layout of each floor. The total plant layout cost includes the basic material handling cost function that
is used by all facility layout problems (the product of inter-departmental flows, cost of travelling between departments,
and distances between each department) and the additional total fixed cost of all floors (fixed cost per floor multiplied
by the total number of floors), the total area-dependent construction cost (area-dependent floor construction cost multi-
plied by the number of floors and the area of each floor), and the total cost of land for the facility layout (area of facility
multiplied by the cost of the land). Since the decision variables include the number of floors, the allocation of
departments, and the length and width of the facility (or area), the formulation results in a mixed-integer, nonlinear
program. patsiatzis linearise the formulation by providing a set of possible facility areas.
The model we propose (see Section 3) borrows much from the formulations presented by Patsiatzis and Papageorgiou
(2002) and Goetschalckx and Irohara (2007). Similar to Goetschalckx and Irohara, our formulation includes the number
and position of elevators as decision variables. Also similar to Goetschalckx and Irohara, elevators in our formulation are
assumed to have zero area. Other key characteristics of our formulation are modelled after Patsiatzis and Papageorgiou’s
model of the process plant. Table 1 compares the flexibility of our model with the other models reviewed above and
Downloaded by [North Carolina State University] at 16:09 05 August 2013

shows that our model is one of the most comprehensive models proposed in the literature. In addition, we use a linearisa-
tion scheme to handle area constraints and propose a lexicographic ordering technique to handle multiple objectives,
namely minimising material handling and facility building costs. Through numerical experiments, we study the effect of
several parameters on the solution, including the effect of having multiple floors on the material handling cost.

3. Multi-floor facility layout model

An optimal facility layout which includes multiple floors is practical when land is scarce, when a compact facility
allows for more efficient control, or when renovating an already existing multi-floor facility. Among the many possible
objectives, two are particularly important for a multi-floor facility layout: (1) minimising the material handling cost and
(2) minimising the building cost. Finding an optimal facility layout so as to minimise these costs requires the following
decisions: the number of floors, the land area required by the facility, the number of elevators within the facility, the
length and width of each department, and the department and elevator allocation to each floor and placement of the
departments and elevator within each floor.
We make several simplifying but relatively mild assumptions to improve the tractability of the formulation.

• Facility floors have a rectangular shape.

• Each facility floor has an identical length, width, and height.
• Vertical travel between floors can only occur through elevators. All elevators are vertically bidirectional.
• Elevator capacity is not considered and elevators are assumed to have an area of zero units.
• All departments have a rectangular shape and their height is equal to the floor height.
• All department area lower bounds are specified as parameters.
• A department cannot be split among multiple floors.
• Elevators cannot be located within a department; however, elevators may be located on the boundary of
departments.
• Movement between departments (and elevators) is modelled as rectilinear centroid-to-centroid movement.
Movement is not explicitly modelled through an aisles network and the movement aisles do not consume any
area.
• Material flow volumes between departments are given.

We first define the notation below, which is followed by the multi-objective MIP for the multi-floor facility layout
problem.

3.1 Notation
In this paper, we will use indices i and j to represent departments, k to represent floors and e to represent elevators.
Below we list several parameters used in the model, which are followed by definitions of decision variables.
 4228 J. Hathhorn et al.

3.1.1 Parameters

N number of departments
M maximum number of floors
EMAX , E MIN maximum and minimum number of elevators, respectively
L; W maximum length and width of the facility, respectively
Fij flow between departments i and j (assumed to be fixed and given)
CijH ; CijV horizontal and vertical transportation cost per unit flow and per unit distance between departments i and j, respectively
Ai lower bound of the area for department i
Si lower bound of the length of any side for department i
H floor height
CL ; CW cost of making the facility one unit greater in length and width, respectively
CF cost of adding a floor to the facility
CE cost of adding an elevator to the facility

3.1.2 Binary and integer variables

Downloaded by [North Carolina State University] at 16:09 05 August 2013

1; if traffic between departments i and j travels through elevator e;
vEije =
0; otherwise:


1; if department i is assigned to floor k;
vik =
0; otherwise:


1; if departments i and j are assingned to the same floor;
zij =
0; otherwise:
8
< 1; if departments i and j do not overlap along the x-axis
tijX = and department i has x-coordinate closer to the origin;
:
0; otherwise:
8
< 1 if departments i and j do not overlap along the y-axis
tijY = and department i has y-coordinate closer to the origin;
:
0; otherwise:
8
< 1; if department i does not overlap elevator e along the x-axis
ie ; qie
qXL XR = and is either to the left ðXLÞ or to the right ðXRÞ of elevator e;
:
0; otherwise:

8
ie ; qie
qYB < 1;
YA
if department i does not overlap elevator e along the y-axis
= and is either below ðYBÞ or above ðYAÞ elevator e;
:
0; otherwise:

1; if elevator e is utilised in the solution;
pe =
0; otherwise:

u number of floors used ðinteger variableÞ

r number of elevators used ðinteger variableÞ

3.1.3 Continuous variables

xi ; yi x- and y-coordinates for the centroid of department i, respectively

e ; ee
eX Y x- and y-coordinates of elevator e, respectively
bL ; bW length and width of the facility along x- and y-axes, respectively
li ; w i length and width of department i along x- and y-axes, respectively
dijH horizontal rectilinear distance between the centroids of departments i and j
dijV vertical distance between the centroids of departments i and j
dieE horizontal rectilinear distance between department i’s centroid and elevator e
 International Journal of Production Research 4229

3.2 Multi-objective MIP formulation for the multi-floor facility layout problem
3.2.1 Objective functions
We consider two objective functions:

X
N 1 X
N
Minimise Fij ðCijH dijH þ CijV dijV Þ ðmaterial handling costÞ; ð1Þ
i¼1 j¼iþ1

Minimise C L bL þ C W bW þ C F u þ C E r ðbuilding costÞ: ð2Þ

Objective function (1) is the material handling cost, which is equal to the sum of the flow-weighted transportation
costs between each pair of departments. The transportation cost between each set of departments is equal to the sum of
the horizontal distance multiplied by the horizontal transportation cost and the vertical distance multiplied by the vertical
transportation cost. Objective function (2) represents the total building cost of the facility. The first and second terms in
Downloaded by [North Carolina State University] at 16:09 05 August 2013

objective (2) represent the costs of making the building greater in length and width, respectively. The third term repre-
sents the cost of adding floors to the facility. The fourth term represents the cost of adding elevators to the facility. Note
that the facility costs can readily be converted into a monetary value, whereas it is not as straightforward to estimate
values of parameters related to the material handling cost.

3.2.2 Floor constraints

X
M
vik ¼ 1; 8i ¼ 1; . . . ; N ; ð3Þ
k¼1

zij  vik þ vjk  1; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; k ¼ 1; . . . ; M; ð4Þ

zij  1  vik þ vjk ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; k ¼ 1; . . . ; M; ð5Þ

zij  1 þ vik  vjk ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; k ¼ 1; . . . ; M; ð6Þ

u  kvik ; 8i ¼ 1; . . . ; N ; k ¼ 1; . . . ; M ; ð7Þ

1  u  M: ð8Þ

Constraints (3) ensure that each department is assigned to only one floor. Constraints (4), (5), and (6) constrain the
variable zij . If two departments i and j are allocated to the same floor (vik ¼ vjk ), then zij will be equal to one by
constraint (4), while constraints (5) and (6) remain inactive. Conversely, if departments i and j are allocated to different
floors, then constraint (4) is inactive and constraints (5) and (6) ensure that the value of zij is zero. Constraints (7) and
(8) constrain the number of floors used, which must be greater than or equal to the floor number that each department
is assigned to, less than or equal to the maximum number of floors specified, and at least one.

3.2.3 Elevator constraints

EX
MAX

r¼ pe ; ð9Þ
e¼1

E MIN  r  EMAX ; ð10Þ

 4230 J. Hathhorn et al.

EX
MAX

vEije ¼ 1  zij ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð11Þ

e¼1

pe  vEije ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; e ¼ 1; . . . ; E MAX : ð12Þ

Constraint (9) defines the total number of elevators used. Constraint (10) requires the total number of elevators to be
between predefined lower and upper bounds. Each pair of departments that are not assigned to the same floor are
required by constraints (11) to route their flow through an elevator. All elevators that have flows routed through them
must be considered active by constraints (12).

3.2.4 Department dimension constraints

Downloaded by [North Carolina State University] at 16:09 05 August 2013

Si  l i  bL ; Si  w i  bW ; 8i ¼ 1; . . . ; N ; ð13Þ

li wi  Ai ; 8i ¼ 1; . . . ; N : ð14Þ

Constraints (13) require each department’s length and width to be between the specified minimum side length for
that department and the maximum length and width of the facility. Constraints (14) require that each department’s area
be greater than the minimum department area specified for each department. Constraints (14) in their current form are
nonlinear. We describe an approximation to linearise these area constraints in Section 3.3.

3.2.5 Non-overlapping constraints

tijX þ tjiX þ tijY þ tjiY ¼ zij ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; i – j; ð15Þ

li þ lj
xj  x i   Lð1  tijX Þ; 8i ¼ 1; . . . ; N ; j ¼ 1; . . . ; N ; i – j; ð16Þ
2

wi þ wj
yj  y i   W ð1  tijY Þ; 8i ¼ 1; . . . ; N ; j ¼ 1; . . . ; N ; i – j; ð17Þ
2

ie þ qie þ qie þ qie ¼ 1;

qXL 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð18Þ
XR YB YA

li
eXe  xi   Lð1  qXL
ie Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; E MAX ; ð19Þ
2

li
xi  eXe   Lð1  qXR
ie Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð20Þ
2

wi
eYe  yi   W ð1  qYB
ie Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; E MAX ; ð21Þ
2

wi
yi  eYe   W ð1  qYA
ie Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX : ð22Þ
2
 International Journal of Production Research 4231

The above constraints are needed to prevent departments and elevators from occupying the same physical location.
Constraints (15) force one of the two constraints (16) or (17) to be active if two departments are located on the same
floor. Constraints (16) prevent departments from overlapping on the x-axis, while constraints (18) prevent departments
from overlapping on the y-axis. Therefore, departments on the same floor may be located next to one another but will
not overlap.
Similar to preventing departments from occupying the same physical location, it is necessary to prevent elevators
from occupying the same location at a department. Although our model assumes elevators do not occupy any space, it
restricts their placement to only department boundaries, not to department interiors. Constraints (18) through (22)
prevent an elevator from overlapping departments in the same manner as departments are prevented from overlapping.
Constraints (18) force one of the four constraints (19), (20), (21), or (22) to be active. Constraints (19) and (20) prevent
the elevator from overlapping each department on the x-axis, while constraints (21) and (22) prevent the elevator from
overlapping each department on the y-axis.

3.2.6 Distance constraints

Downloaded by [North Carolina State University] at 16:09 05 August 2013

X
M
dijV  H kðvik  vjk Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð23Þ
k¼1

X
M
dijV  H kðvjk  vik Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð24Þ
k¼1

dijH  ðxi  xj Þ þ ðyi  yj Þ  ð1  zij ÞðL þ W Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð25Þ

dijH  ðxj  xi Þ þ ðyj  yi Þ  ð1  zij ÞðL þ W Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð26Þ

dijH  ðxi  xj Þ þ ðyj  yi Þ  ð1  zij ÞðL þ W Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð27Þ

dijH  ðxj  xi Þ þ ðyi  yj Þ  ð1  zij ÞðL þ W Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; ð28Þ

dieE  ðxi  eXe Þ þ ðyi  eYe Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð29Þ

dieE  ðeXe  xi Þ þ ðeYe  yi Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð30Þ

dieE  ðxi  eXe Þ þ ðeYe  yi Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð31Þ

dieE  ðeXe  xi Þ þ ðyi  eYe Þ; 8i ¼ 1; . . . ; N ; e ¼ 1; . . . ; EMAX ; ð32Þ

dijH  ðdieE þ djeE Þ  2ðzij ÞðL þ W Þ  2ð1  vEije ÞðL þ W Þ; 8i ¼ 1; . . . ; N  1; j ¼ i þ 1; . . . ; N ; e ¼ 1; . . . ; E MAX :

ð33Þ

Constraints (23) and (24) define the vertical distances between two departments i and j as the height of each floor
multiplied by the number of floors separating the two departments. When the departments are located on the same floor
(i.e. vik ¼ vjk ), the value of the vertical distance is zero.
 4232 J. Hathhorn et al.

Since two departments i and j can either occupy the same floor or different floors, and an elevator must only be used
if departments i and j occupy different floors, there are two constraint sets for the horizontal rectilinear distances. For
the first case, when departments i and j are assigned to the same floor (zij ¼ 1), an elevator is not utilised; thus, the
horizontal distance between the two departments is the rectilinear distance between the centroids of the two departments
i and j. Constraints (25) through (28), which define the rectilinear distance between two departments, are active in this
case but inactive when departments i and j are assigned to different floors (zij ¼ 0). For the second case, when depart-
ments i and j are assigned to different floors, the horizontal distance between the two departments is dependent on the
rectilinear distance between the elevator and each department. The distance between department i and each elevator is
obtained from constraints (29) through (32). Constraint (33) sets the vertical distance between the two departments equal
to the sum of the rectilinear distance between each department and the elevator determined by constraints (29) through
(32). This constraint is only active when departments i and j are assigned to different floors (zij ¼ 0) and only when
evaluating the elevator used to transport flow between the two departments (vEije ¼ 1). Accordingly, the horizontal
rectilinear distance for any two departments i and j is either (1) the rectilinear distance between the centroids of the two
departments, if the departments are assigned to the same floor, or (2) the total rectilinear distance from department i to
the elevator and from the elevator to department j when the two departments are assigned to different floors.
Downloaded by [North Carolina State University] at 16:09 05 August 2013

3.2.7 Facility bounding constraints

1 1
xi  li ; yi  wi ; 8i ¼ 1; . . . ; N ; ð34Þ
2 2

1 1
bL  x i þ li ; bW  y i þ w i ; 8i ¼ 1; . . . ; N ; ð35Þ
2 2

0  eXe  bL ; 0  eYe  bW ; 8e ¼ 1; . . . ; E MAX ; ð36Þ

0  bL  L; 0  bW  W : ð37Þ

Constraints (34) require each department to be located within the lower bounds of the facility. Constraints (35)
require each department to be located within the upper bounds of the facility. The elevators are similarly bounded by
constraints (36). Constraints (37) constrain the facility to the non-negative region less than the upper bounds specified
as parameters.

3.2.8 Additional constraints

In addition to the constraints described in Sections 3.2.2 through 3.2.7, our model includes several non-negativity and
integrality constraints, which are omitted here for space reasons.

3.3 Approximation to linearise area constraints

Constraint (14) requires the area of each department to be at least as great as the minimum area parameter specified for
that department; however, the area calculation is not linear. The feasible region for a department’s width and length
subject to the minimum area requirement is shown Figure 2(a). The horizontal axis of this graph represents the length
of department i (li ) and the vertical axis represents its width (wi ). The region shaded red in the figure represents all
feasible length–width combinations. As can be seen from the figure, the boundary of the feasible region (red curve) is
nonlinear due to constraint (14). The length–width combinations below the red curve are infeasible because the resulting
area will be less than the minimum area required. To linearise these constraints, we approximate the curve with a piece-
wise linear function using the two lines shown in Figure 2(a). Note that a department’s length or width must be greater
than the minimum side length parameter Si. Also, due to the objective to minimise the building cost, it is unlikely that a
side will exceed the value of Ai =Si since this would result in a department area that is larger than the minimum require-
ments. Within the region ½Si ; Ai =Si Þ, bounding each side length, the linear constraints are reasonably accurate approxima-
tions of the nonlinear constraint. The linear approximation constraints are given as follows:
 International Journal of Production Research 4233

Figure 2. Approximation to linearise the area constraint.

pffiffiffiffi
Downloaded by [North Carolina State University] at 16:09 05 August 2013

pffiffiffiffi Si  Ai pffiffiffiffi
wi  Ai  pffiffiffiffi ðli  Ai Þ; 8i ¼ 1; . . . ; N ; ð38Þ
Ai =Si  Ai

pffiffiffiffi
pffiffiffiffi Ai =Si  Ai pffiffiffiffi
wi  Ai  pffiffiffiffi ðli  Ai Þ; 8i ¼ 1; . . . ; N : ð39Þ
Si  Ai

When solving the model, we substitute for the nonlinear department area constraint (14) the linear approximation
constraints (38) and (39). Note that it is very easy to improve the accuracy of this approximation by simply using addi-
tional lines to linearise the curve. Each additional line would, however, add one more constraint for each department.

3.4 Balancing multiple objectives

As described in Section 3.2.1, we consider two objective functions: (1) minimising the material handling cost, which is
equal to the sum of the flow-weighted transportation costs between each pair of departments, and (2) the total building
cost of the facility. It is not straightforward to combine these two objectives into a single objective. The terms for the
facility building costs could be in a monetary unit, while the the material handling cost, which is given in terms of the
inter-departmental flows, is primarily an efficiency indicator that is difficult to meaningfully transform into a monetary
unit. This makes balancing the weights on these two (possibly conflicting) objectives difficult.
Placing too much weight on the inter-departmental flows (i.e. the material handling cost) results in an efficient layout
in terms of inter-departmental flows, but the department areas may not efficiently utilise the facility’s space. For exam-
ple, all departments might be on one floor or – depending on the vertical transportation cost parameters – a floor might
be created for each department (with lots of unutilised space). On the other hand, too much weight placed on the facility
building cost results in the departments being arranged to take up the smallest area possible with little regard to the
efficiency of inter-departmental flows. While any of these solutions are Pareto optimal, a balance between the two
objectives can be obtained by combining the two objectives into a single one with careful selection of weights for each
objective. One disadvantage of such a weighted sum approach is that it requires defining artificial weights, which may
not be meaningful or easy to interpret.
Instead of using a standard weighted sum approach, we employ the method of lexicographic ordering (LO) (Romero
2001, Jee, McShan, and Fraass 2007) to balance multiple objectives in a more meaningful manner. In this approach, a
sub-problem considering a single objective at a time is solved (Jee, McShan, and Fraass 2007). The sequence of these
sub-problems is based on the objective priorities (also called lexicographic ordering) determined by the decision maker.
Each objective’s optimal value is then used as a benchmark to constrain the optimality of the solution with respect to
that objective. Since a separate optimisation is solved for each objective, LO can naturally handle multiple objectives
having different unit scales.
In the case of two different objectives, LO first solves a sub-problem that only considers the first objective. The
solution is then incorporated into the final problem in the form of a constraint. This constraint requires that the solution
of the final problem, which only includes the second objective, be within some range of optimality with respect to the
first objective.
 4234 J. Hathhorn et al.

In the numerical experiments, we chose the objective of minimising the building cost of the facility to be optimised
first. This is an intuitive choice since we primarily want a facility size just large enough to house all departments.
However, during the second optimisation, the model is given some slack with regard to the building cost, so that the
departments can be moved around to optimise inter-departmental flows. Suppose, for example, that the first sub-problem
results in a facility that costs $10 million to build. This value can be used to limit the final problem to determine a
facility layout that does not cost more than, say, $11 million to build. In other words, the solution must optimise the
building cost to be within 10% (or any percentage value of our choice) of optimal for this objective.

4. Numerical results
To illustrate the model, we perform several numerical experiments in this section. We generated data for model
parameters for seven departments, randomly. The units of length are meters and the units related to building costs are in
dollars. The minimum area of each department was set to be a randomly generated integer between 700 and 5000. The
minimum side length parameter for each department was generated similarly with the upper bounds set to the square
root of the area for that department and the lower bounds set to half the value of the upper bounds. This ensures that
Downloaded by [North Carolina State University] at 16:09 05 August 2013

each department has a feasible minimum side length and that the aspect ratio of each department is no greater than 4:1.
Table 2 lists values of minimum area and side length parameters.
The maximum length and width of the facility parameters were determined by the sum of each department’s
minimum area divided by that department’s minimum side length. This is an upper bound for the possible length of the
facility: it is the length of the facility if all the departments were rectangular and lined up lengthwise, and therefore is
not a constraining value. The floor height was set to four. The cost of a meter of facility length or width was set to
$60,000. The cost of floors and elevators was set to $5,000,000 and $300,000, respectively.
The flow between two departments was set to be a randomly generated integer between 0 and 100 to create a
relatively large variance between the flows between different departments. The vertical and horizontal transportation
costs per unit distance matrices were set to be a random integer between 10 and 20.
We used Python 2.6.5 to interact with the Gurobi Optimizer version 3.0 (Gurobi Optimization 2011) to solve our
model. The Gurobi Optimizer is a state-of-the-art solver for mixed-integer linear programming and designed to utilise

Table 2. Values for department size parameters used in numerical experiments.

Parameter Dep.1 Dep.2 Dep.3 Dep.4 Dep.5 Dep.6 Dep.7

Min side length (Si ) 24 63 42 52 33 30 36

Min area (Ai ) 797 4180 3001 4194 3669 1015 4185

Figure 3. Example layout of the model obtained using the LO method.

 International Journal of Production Research 4235

multiple processing cores. All computations were executed on a personal desktop computer running the Microsoft Vista
32-bit operating system on an Intel Core 2 Quad Q6600 CPU, with four physical cores at 2.40 GHz, and 4 GB of
RAM.

4.1 Sample facility layout

Using the parameters defined as above, and setting the maximum number of floors to two and the maximum number of
elevators to one, we solved the model in two stages as explained in Section 3.4. Figure 3(a) shows the solution of the
sub-problem, which ignores the cost of material handling and minimises the building cost. In this solution departments
are compacted into the least building cost formation possible, in this case on a single floor. This resulted in a building
cost of $22,590,000. Next, the optimal value obtained in the first stage is used as a constraint with 1% slack (i.e. the
building cost is constrained to be less than 1.01 times the optimal building cost from the first subproblem) and the
problem is solved minimising the material handling cost. The second-stage solution is given in Figure 3(b).
This layout uses a single floor. The facility cost of this layout is $22,816,000, while the material handling cost is
1,093,717 (recall that the material handling cost is not a monetary unit but primarily a metric representing how efficient
Downloaded by [North Carolina State University] at 16:09 05 August 2013

materials and people flow between departments). For comparison, we solved the second optimisation model using
different slacks. The resulting layouts for 2.5% and 13% slack are shown in Figures 4 and 5, respectively. Both layouts
have two floors and a single elevator. The location of the elevator is represented by a blue dot. Compared with the first
sub-problem and 1% slack layouts, the layout with 2.5% slack results in a higher building cost of $23,155,000 and

Figure 4. Example layout using a slack of 2.5%.

Figure 5. Example layout using a slack of 13%.

 4236 J. Hathhorn et al.

Table 3. The minimum required area and the actual area of departments in solutions with various slack parameters.

Dept Min area Area (sub-opt) Area (1% slack) Area (2% slack) Area (13% slack)

1 797 797 792.5 792.0 799.0

2 4180 4200.9 4200.9 4182.9 4200.9
3 3001 2991.0 2982.0 2992.1 2982.0
4 4194 4211.1 4218.8 4226.6 4160.0
5 3669 3663.0 3885.8 3663.0 3663.0
6 1015 1046.6 1027.1 1015.0 1046.6
7 4185 4241.4 4355.6 4306.3 4176.0
Downloaded by [North Carolina State University] at 16:09 05 August 2013

Figure 6. The flow cost for single and multiple floors as slack is increased.

inefficient use of space, but a lower material handling cost of 986,397. Similarly, increasing the slack to 13% further
increases the building cost ($25,527,000) with large unoccupied space, but significantly reduces the material handling
cost (794,220). In fact, flow cost cannot be improved for slack more than or equal to 13%, as shown in Figure 6 (more
details in Section 4.4).
Table 3 shows the minimum required area and the actual area of a department in the solution. The approximate area
constraints (38) and (39) are designed to ensure that the area of a department is always greater than the minimum
requirement. However, as can be seen from this table, some of the departments are slightly smaller than the minimum
requirements. This difference is due to roundoff errors and several optimality, feasibility, and integrability tolerances
used in the Gurobi solver. The largest violation of the minimum area requirement we observed in our experiments was
0.8%, therefore it can be assumed that the potential error is negligible.
One drawback with the model’s current facility cost structure is the use of a parameter for the cost of adding an
additional unit of length or width to the facility. This is very unrealistic and strongly favors a square facility. This could
be improved by including a variable that approximates the area of the facility, and a single parameter for the construc-
tion costs per square foot – a much more common metric. Furthermore, since all floors are assumed to be of equal size,
the cost of an additional floor could be more accurately modelled with a fixed cost and a variable cost proportionate to
the square footage of the facility.

4.2 Symmetry-breaking constraints

We tested the effect of adding a set of symmetry-breaking constraints on the overall computation time. Goetschalckx
and Irohara (2007) demonstrated these constraints in a similar model and found them to effectively reduce the
computation time for some cases. These constraints are referred to as the ‘position q method’ (PQM):
 International Journal of Production Research 4237

Table 4. Computation time in seconds for various scenario sizes.

Without PQM With PQM

Computation Time
Departments Floors Slack Part 1 Part 2 Total Part 1 Part 2 Total

5 Single 3.0% 0.31 0.47 0.78 0.08 0.25 0.33

6.0% 0.38 0.69 0.27 0.35
Multiple 3.0% 0.61 0.8 1.41 0.11 0.23 0.34
6.0% 0.78 1.39 0.67 0.78
6 Single 3.0% 2.41 2.34 4.75 1.33 0.75 2.08
6.0% 1.88 4.29 0.78 2.11
Multiple 3.0% 5.06 20.01 25.07 2.58 2.09 4.67
6.0% 11.2 16.26 3.83 6.41
7 Single 3.0% 404.86 223.22 628.08 63.45 54.5 117.95
6.0% 82.98 487.84 15 78.45
Multiple 3.0% 98.64 77.39 176.03 29.67 17.39 47.06
6.0% 951.22 1049.9 39.09 68.76
Downloaded by [North Carolina State University] at 16:09 05 August 2013

xq  0:5bl ; ð40Þ

yq  0:5bw ; ð41Þ

vqM ¼ 0: ð42Þ

These constraints do not change the model, but attempt to reduce the computation time by eliminating duplicate
feasible solutions. Constraints (40) and (41) restrict department q’s centroid to the left-bottom quarter of the facility,
where department q is the department with the largest sum of flows with other departments. This eliminates duplicate
symmetric solutions about the x- and y-axes. Constraint (42) restricts department q from being located on the highest
floor M. This final constraint will eliminate the symmetry for the floors and will only be included when the maximum
number of floors is greater than one.

4.3 Computation times

We solved models for scenarios with five, six, and seven departments. For each number of departments, the maximum
number of floors and the maximum number of elevators were varied between two sets of values. The first restricts the
facility to a single floor and no elevators to represent the case of a single-floor facility. The second set allows two floors
and one elevator, a reasonable maximum number for a facility of only seven departments. Finally, the value of the slack
parameter is set to 3% and 6%. The resulting computation times are shown in Table 4.
In every case, PQM significantly reduced the computation time. Increasing the number of departments unsurprisingly
increases the computation time. Allowing multiple floors and an elevator generally increased the computation time, but
unexpectedly decreased it in some cases. The slack percentage appears to have a significant effect on computation time,
but no correlation was apparent.

4.4 Flow cost

We also investigated the effects of increasing the slack and allowing multiple floors on the resulting layout’s flow cost.
The flow cost is shown in Figure 6 for the previous set of scenarios.
Recall that the slack parameter represents a percentage of the optimal facility building cost when the objective of
minimising the material handling cost is removed from the objective function. As the slack is increased, the second opti-
misation model is allowed to ‘spend’ more on building the facility to decrease the cost of the inter-departmental flows.
As can be seen from Figure 6, when the number of floors is limited to one, the flows can only be marginally improved.
Allowing a second floor quickly reduces flow costs when the slack is increased. In this case, increasing the slack
beyond 13 and 16% no longer reduces flow costs for the multiple- and single-floor facilities, respectively.
 4238 J. Hathhorn et al.

5. Conclusions
In this paper, we develop a comprehensive model for a multi-objective multi-floor facility layout problem to minimise
both the material handling cost (i.e. the cost of flow between the departments) and also the total facility building cost.
Our facility layout model accounts for many design decisions (i.e. the position, length and width of departments, the
number of floors and size of the facility, the number and location of elevators) – more than any other model reviewed.
We use an approximation to linearise the minimum area requirement constraint for each department. We also use a
lexicographic ordering approach to prioritise the two objectives and solve two sub-problems in sequence.
In our numerical experiments, we observed that the computation time increases dramatically as the number of
departments increases. Adding PQM constraints seems to consistently reduce computation times, however investigation
of other methods to reduce the computation time and therefore solve the problem for a large number of departments is
left for future work. One simple approach we will investigate is to cluster the departments into ‘super-departments’ (see,
for example, Tam and Li (1991)) based on inter-departmental flows and solve the facility layout problem for these
super-departments. The super-departments are then decomposed by solving several sub-problems for each cluster.
Parameters are the most important factor in determining whether a multi-floor facility is more efficient than a single-
floor facility. Before the size and number of floors in the facility can be effectively modelled as decision variables, the
Downloaded by [North Carolina State University] at 16:09 05 August 2013

model’s facility building cost structure needs to be improved to better represent realistic costs of construction. Accurate
construction cost data must be input in the model for a truly optimal facility layout to be generated.

References

Abdinnour-Helm, S., and S. Hadley. 2000. “Tabu Search Based Heuristics for Multi-Floor Facility Layout.” International Journal of
Production Research 38 (2): 365–383.
Bozer, Y. A., R. D. Meller, and S. J. Erlebacher. 1994. “An Improvement-Type Layout Algorithm for Single and Multiple-Floor Facilities.”
Management Science 40 (7): 918–932.
Buffa, E., G. Armour, and T. Vollmann. 1964. Allocating Facilities with CRAFT. Harvard University.
Carr, R. 2011. Hospital. Available from: [Link] [Accessed 11 April 2012].
Drira, A., H. Pierreval, and S. Hajri-Gabouj. 2007. “Facility Layout Problems: A Survey.” Annual Reviews in Control 31: 255–267.
Duman, E., and I. Or. 2007. “The Quadratic Assignment Problem in the Context of the Printed Circuit Board Assembly Process.”
Computers & Operations Research 34 (1): 163–179.
El-Baz, M. 2004. “A Genetic Algorithm for Facility Layout Problems of Different Manufacturing Environments.” Computers &
Industrial Engineering 47 (2/3): 233–246.
Georgiadis, M., et al. 1999. “A General Mathematical Programming Approach for Process Plant Layout.” Computers & Chemical
Engineering 23 (7): 823–840.
Goetschalckx, M., and Irohara, T. 2007. “Efficient Formulations for the Multi-Floor Facility Layout Problem with Elevators.”
Available from: [Link] [Accessed 11 April 2012].
Gurobi Optimization I. 2011. Gurobi Optimizer Reference Manual. Available from: [Link] [Accessed 11 April 2012].
Hahn, P. M., and J. Krarup. 2001. “A Hospital Facility Layout Problem Finally Solved.” Journal of Intelligent Manufacturing 12:
487–496.
Haraden, C., and R. Resar. 2004. “Patient Flow in Hospitals: Understanding and Controlling it Better.” Frontiers in Health Service
Management 20: 3–15.
Hassan, M., and G. Hogg. 1987. “A Review of Graph Theory Application to the Facilities Layout Problem.” Omega 15 (4): 291–300.
Islier, A. A. 1998. “A Genetic Algorithm Approach for Multiple Criteria Facility Layout Design.” International Journal of Production
Research 36 (6): 1549–1569.
Jee, K. W., D. L. McShan, and B. A. Fraass. 2007. “Lexicographic Ordering: Intuitive Multicriteria Optimization for IMRT.” Physics
in Medicine and Biology 52: 1845–1861.
Jolai, F., R. Tavakkoli-Moghaddam, and M. Taghipour. 2012. “A Multi-Objective Particle Swarm Optimisation Algorithm for Unequal
Sized Dynamic Facility Layout Problem with Pickup/Drop-off Locations.” International Journal of Production Research 50
(15): 4279–4293.
Kochhar, J. S. 1998. “MULTI-HOPE: A Tool for Multiple Floor Layout Problems.” International Journal of Production Research 36
(12): 3421–3435.
Konak, A. et al. 2006. “A New Mixed Integer Programming Formulation for Facility Layout Design Using Flexible Bays.” Opera-
tions Research Letters 34 (6): 660–672.
Lacksonen, T. 1997. “Preprocessing for Static and Dynamic Facility Layout Problems.” International Journal of Production Research
35 (4): 1095–1106.
Lawler, E. 1963. “The Quadratic Assignment Problem.” Management Science, 586–599.
Lee, K., M. Roh, and H. Jeong. 2005. “An Improved Genetic Algorithm for Multi-Floor Facility Layout Problems Having Inner
Structure Walls and Passages.” Computers & Operations Research 32 (4): 879–899.
 International Journal of Production Research 4239

Lee, K. Y., S. N. Han, and M. I. Roh. 2003. “An Improved Genetic Algorithm for Facility Layout Problems Having Inner Structure
Walls and Passages.” Computers and Operations Research 30: 117–138.
Leung, J. 1992. “A New Graph-Theoretic Heuristic for Facility Layout.” Management Science 38: 594–605.
Mak, K. L., Y. S. Wong, and F. T. S. Chan. 1998. “A Genetic Algorithm for Facility Layout Problems.” Computer Integrated Manu-
facturing Systems 11 (1/2): 113–127.
Meller, R., and Y. Bozer. 1996. “A New Simulated Annealing Algorithm for the Facility Layout Problem.” International Journal of
Production Research 34 (6): 1675–1692.
Meller, R. D., and K. Y. Gau. 1996. “The Facility Layout Problem: Recent and Emerging Trends and Perspectives.” Journal of
Manufacturing Systems 15 (5): 351–366.
Montreuil, B. 1990. “A Modeling Framework for Integrating Layout Design and Flow Network Design.” In: Proceedings of the
Material Handling Research Colloquium, Vol. 90, 43–58.
Patsiatzis, D. I., and L. G. Papageorgiou. 2002. “Optimal Multi-Floor Process Plant Layout.” Computers and Chemical Engineering
26: 575–583.
Rajasekharan, M., B. Peters, and T. Yang. 1998. “A Genetic Algorithm for Facility Layout Design in Flexible Manufacturing
Systems.” International Journal of Production Research 36 (1): 95–110.
Romero, C. 2001. “Extended Lexicographic Goal Programming: A Unifying Approach.” Omega 29: 63–71.
Rosenblatt, M. 1979. “The Facilities Layout Problem: A Multi-Goal Approach.” International Journal of Production Research 17 (4):
Downloaded by [North Carolina State University] at 16:09 05 August 2013

323–332.
Sagan, H. 1994. Space-Filling Curves, Vol. 2. New York: Springer.
Singh, S. P., and R. R. K. Sharma. 2006. “A Review of Different Approaches to the Facility Layout Problems.” The International
Journal of Advanced Manufacturing Technologies 30 (5): 425–433.
Suresh, G., V. Vinod, and S. Sahu. 1995. “A Genetic Algorithm for Facility Layout.” International Journal of Production Research
33 (12): 3411–3423.
Tam, K., and S. Li. 1991. “A Hierarchical Approach to the Facility Layout Problem.” International Journal of Production Research
29: 165–184.
Tzortzopoulos, P. et al. 2008. “Design for Operational Efficiency – Linking Building and Service Design in Healthcare Environments.” In:
Proceedings: ELAGEC-II Encuentro Latino-Americano De Gestion Y Economia De La Construccion, Santiago, Chile, January,
24–25.
Vos, L., S. Groothuis, and G. G. van Merode. 2007. “Evaluating Hospital Design from an Operations Management Perspective.”
Health Care Management Science 10: 357–364.
Wu, Y., and E. Appleton. 2002. “The Optimisation of Block Layout and Aisle Structure by a Genetic Algorithm.” Computers &
Industrial Engineering 41 (4): 371–387.