You are on page 1of 14

Discretized Cell Modeling for Optimal Facility Layout Plans

of Unequal and Irregular Facilities


C. Huang 1 and C. K. Wong, A.M.ASCE 2

Abstract: Facility layout design is an essential planning task to resolve potential spatial conflicts and overlapping during practical oper-
ations. A discretized cell optimization model is developed to optimize site space usages. Site areas and facilities are represented by small unit
cells to effectively model irregularities, and the availability of unit cells for use in a facility setup is modeled by binary-type variables. With
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

size and shape requirements through linear constraints, the site layout design can be formulated as a binary mixed integer-linear programming
(BMILP) problem to allocate different facilities onto different site available areas in optimal shapes and locations. Total material transpor-
tation costs across facilities are optimized subject to various design constraints ensuring safety, homogenous facility setups, physical size, and
orientation requirements. The proposed cell model is verified by comparing optimization results with results obtained by conventional point
notation methods. A standard branch-and-bound algorithm is applied to solve a global optimal solution, and the numerical example is opti-
mized for illustrating the very different optimized layout plan in terms of facility locations, positions of entry and exit points, shapes, and
orientations of different facilities. DOI: 10.1061/(ASCE)CO.1943-7862.0001206. © 2016 American Society of Civil Engineers.
Author keywords: Facility layout planning; Discretized cell modeling; Site layout optimization; Binary mixed integer-linear programming
problem; Project planning and design.

Introduction material transportation costs between any two related site facilities
at different locations based on their horizontal Manhattan distances.
Site layout planning is essential to construction management be- Activity links and material flow among pairs of site facilities at
cause of its direct impact on construction costs, production effi- different locations can be established for evaluating the objective
ciency, and site safety. Improper facility layouts decrease the cost function (Hegazy and Elbeltagi 1999; Easa and Hossain 2008;
overall work efficiency by requiring workers to travel further, in- Wong et al. 2010).
creasing both their travel time and length of their material trans- Effective construction site planning requires a proper designing
portation paths. Based on studies in the manufacturing industry, of a site layout and site facility location. Different site facilities
materials handling costs can be reduced by 20–60% if appropriate compete to occupy available site spaces at different spatial loca-
facility layout is adopted (Lam et al. 2005). Because of its impor- tions in construction sites. The conventional point modeling ap-
tance, site layout planning has been studied by many researchers proach is simple to use and also fast in computation but at the
using different optimization methods and modeling approaches that expense of modeling accuracy such as problems of irregularities in
have been developed for optimal site layout plans. For instance, shapes, facility orientations, and potential spatial conflicts in the
available site locations might be visually identified and available optimization process.
areas might be prespecified within site boundaries; these are re- To enhance these existing point notation models, the present
ferred to as fixed model inputs (Tam et al. 2001; Huang and Wong study aims to develop a discretized site layout optimization model
2015). Key decision variables are defined to allocate site facilities to allocate different facilities to adjacent linked unit cells for
onto available site locations using point notation and representa- homogenous facility setups, which form a mesh that represents an
tion. Regular site area and site facilities are modeled. Unequal area entire site area. The entire study area in a site is discretized and
modeling with detailed area size restrictions has also been devel- modeled by a cell matrix that consists of rows and columns such
oped to increase modeling accuracy (Hegazy and Elbeltagi 1999; that irregular site areas and site facilities with different physical
Easa and Hossain 2008). In these types of modeling, restrictions in shape requirements can be modeled; in this manner, the construc-
specific site locations, matching physical sizes among available site tion site area may be completely utilized, and flow paths for con-
areas and facilities, and fulfilling safety requirements are estab- struction materials may be optimized. The problem is challenging
lished in the form of mathematical constraints that can be integrated when the site is tight without much room and buffer areas for fa-
into an optimization framework. The total (generalized) cost is cility setups (i.e., total area of site ≈ total area of facilities). The
generally set as an objective function for optimization, including solution of the site layout plan is not a trivial one especially when
facilities that are unequal in size and irregular in shape are being
1
Lecturer, College of Architecture and Civil Engineering, Beijing Univ. allocated onto a site with irregular boundaries.
of Technology, 100 Pingleyuan, Chaoyang District, Beijing 100124, China.
2
Assistant Professor, Dept. of Architecture and Civil Engineering, City
Univ. of Hong Kong, Tat Chee Ave., Kowloon Tong, Hong Kong SAR
(corresponding author). E-mail: wongck@cityu.edu.hk Literature Review
Note. This manuscript was submitted on December 17, 2015; approved
on May 31, 2016; published online on July 19, 2016. Discussion period The latest publications in this area of study include research into
open until December 19, 2016; separate discussions must be submitted dynamic construction site layout and security planning (Li et al.
for individual papers. This paper is part of the Journal of Construction 2015) and site layout planning with safety considerations across
Engineering and Management, © ASCE, ISSN 0733-9364. multiple construction stages (Huang and Wong 2015). Both these

© ASCE 04016082-1 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


studies still apply the conventional point notation approach to the may be incorporated in their design framework (Sanad et al. 2008;
optimization of site layout plans. The conventional point modeling Elbeltagi et al. 2004).
approach is simple to use and also fast in computation but at the Furthermore, artificial intelligence (Tommelein et al. 1992), ge-
expense of the modeling accuracy. The problem of irregularities in netic algorithms (Mavridou and Pardalos 1997; Li and Love 1998,
sites and facilities is a critical one. When the irregularity problem 2000; Wong et al. 2010; Zouein et al. 2002; Hegazy and Elbeltagi
combined with the facility orientations and the potential spatial 1999; Elbeltagi et al. 2004), ant colonies (Corry and Kozan 2004;
conflict, the existing point modeling approach may not work prop- Lee 2012), tabu searches (Kulturel-Konak 2012), particle swarms
erly, and the problem formulation would be difficult. (Kulturel-Konak and Konak 2011), and annealed neural networks
Comprehensive reviews by Drira et al. (2007), Singh and (Yeh 1995) have been widely applied for different layout optimi-
Sharma (2006), Liggett (2000), Mavridou and Pardalos (1997), zation problems. However, a major weakness of applying prob-
and Meller and Gau (1996) cover the history of such studies; mod- abilistic optimization methods is that their overall design accuracy
eling methods for site spaces and facilities; deterministic, heuristic, is restricted because of the randomness in searching the solution
and probabilistic solution algorithms; chronological listing of re- spaces leading to suboptimal site layout planning. The advantage
searchers’ works; and commercialized computer design packages, of adopting a programming method is that such methods ensure
respectively. global optimal solutions in which a standard solver package is
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

Site layout planning with fixed locations and fixed shape readily available (Konak et al. 2006; Huang and Wong 2015;
designs is the most classical problem addressed (Li and Love Huang et al. 2010, 2011). To model different shapes and orienta-
1998; Wong et al. 2010; Huang et al. 2010, 2011; Easa and Hossain tions of irregular facilities during layout planning, the present study
2008). Other studies have proposed methods for designing three- proposes a new cell optimization model that is formulated as a
dimensional (3D) multilevel layout plans when vertical dimensions binary mixed integer-linear programming (BMILP) problem and
are modeled into the transportation path, for instance, lift transpor- solved by a standard branch-and-bound technique.
tation to different building floors or tower crane hooks may be
utilized (Huang et al. 2010, 2011; Meller and Bozer 1997). More Discretized Cell Model for Site Layout Planning and
advanced approaches include dynamic site layout planning that Optimization
deals with varying uses of space during different construction
stages (El-Rayes and Said 2009; Tommelein and Zouein 1993;
Zouein and Tommelein 1999). Model Assumptions
Unequal area modeling is another significant research area in 1. The site area is discretized into a uniform mesh composed of
the literature. Li and Love (2000) studied an unequal area site lay- small-sized, square unit cells.
out problem through the simple point modeling approach. Zhang 2. The regular (square and rectangle) or irregular (other than
and Wang (2008) applied particle swarm optimization method to square and rectangle) facilities of different shapes are divided
deal with an unequal area regular layout problem based on point into small parts (pieces), and each unit cell has a square shape.
notation approach again. Zouein et al. (2002) applied genetic algo- 3. The entry and exit point or interior of a facility is represented by
rithms to arrange facilities of unequal sizes but the potential over- one unit cell.
lapping problem was prevented. Corry and Kozan (2004) solved 4. The unit cell is the finest unit for analysis, and the cell size is
the similar problem using the ant colony method. Easa and Hossain sufficiently small to model any irregularity and shape.
(2008) developed a programming approach to improve the conven- 5. Materials inside facilities are evenly distributed over all assigned
tional point notation method by introducing facility length and unit cells and transportation of different types of materials in-
width requirements for optimization. The entire site area could be ternally inside all facilities will incur subtotal cost C 0 .
used without restriction by a given set of available site locations. 6. With the objective function, material transportation costs are
Constraints were developed to prevent overlapping effectively. optimized including costs induced externally from one facility
McKendall and Hakobyan (2010) developed heuristics for an un- to another facility, C, and costs induced internally inside a fa-
equal but regular facility layout problem. Material transportation cility from entry and exit point to other unit cells of the same
costs and facility setup costs were set as the objective for optimi- facility, C 0 .
zation. A three-stage solution algorithm encapsulating a tabu search 7. Once unit cells are assigned to setup facilities, sufficient buffer
was proposed to solve this heuristic model. Point notation was used areas exist and are reserved to satisfy all necessary safety re-
with mathematical constraints to prevent area overlapping. Similar quirements and to provide congestion-free material transporta-
work was conducted by Scholz et al. (2009), who combined the tion paths.
8. One facility should contain an entry and exit point and all inter-
slicing tree and tabu search for solving the unequal area facility
ior parts of that facility.
layout problem. Andayesh and Sadeghpour (2013) again applied
the point notation method to solve the unequal size by minimizing
total potential energy among facilities. Kulturel-Konak and Konak Cell Representation of a Construction Site
(2011) applied another particle swarm method to solve this problem Facilities of both regular and irregular shapes are discretized
and Ulutas and Kulturel-Konak (2013) applied genetic algorithms. into small square areas, referred to as unit cells in the proposed
Hegazy and Elbeltagi (1999) developed the EvoSite using a dis- modeling framework. Unit cells are uniformly arranged in a two-
cretized approach to model irregular sites and facilities. The great- dimensional (2D) rectangular mesh of M unit cells along the ver-
est common divisor (GCD) concept was introduced to fix the size tical direction and N unit cells along the horizontal direction, as
of a modeling unit. Genetic algorithms were applied to solve the shown in Fig. 1 in which an example trapezoidal-shaped site is
site layout problem. Elbeltagi and Hegazy (2001) replaced the ge- represented by a set of structured unit cells (small squares shown
netic algorithms by a hybrid artificial intelligence as the solution by dotted lines). The example site contains 11 × 11 unit cells
method for solving a similar problem. Different site layouts for dif- (M ¼ N ¼ 11 in size). Cell identification numbers, vertical and
ferent construction periods could be designed using the genetic horizontal positions, cell availability for facility setup, and gov-
algorithms (Elbeltagi et al. 2001). Safety and environmental factors erning constraints will be provided in the following sections.

© ASCE 04016082-2 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


n=1 2 3 4 5 6 7 8 9 10 n=N=11
Apartment Block Apartment Block

m=1 Tree
Street Entrance
Tree
2 i=22
Available cells
3 for facility
assignment
s 4
s Apartment
t
t Block
r 5
r
e
Construction Site e Construction Site
e 6
e
t
t
7
Unavailable cells for
facility assignment
8
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

10
Street Entrance
i=111
n=M=11
Legend: Unit cell inside site boundary

Unit cell outside site boundary


Legend: Unit cell inside site boundary
Site boundary
Unit cell outside site boundary
Fig. 1. Cell representation of a site area
Site boundary

Fig. 2. Example unit cell representation of an irregular site


Vertical and Horizontal Positions of Unit Cells
In the proposed 2D modeling framework, the position of a unit cell
i is identified by its horizontal and vertical positions, H i and V i , as 
calculated by Eqs. (1) and (2), respectively. Mathematically, the 1 available for site facility assignment
ϖi ¼ ; ∀ i ∈ f1; Ig
symbol bc is used to round down a numerical figure (as calculated 0 unavailable for site facility assignment
by the mathematical function inside that symbol) to its nearest in-
teger. In Fig. 2, for example, a unit cell i ¼ 22 with N ¼ 11 and ð3Þ
ε ¼ 0.01, ði − εÞ=N, is 1.999; bði − εÞ=Nc converts that figure to
integral number 1. From Eqs. (1) and (2), the horizontal and vertical
positions of the cell i ¼ 22 can then be fixed at H22 ¼ 11 and Assignment of Unit Cells for Facility Setup
V 22 ¼ 2, respectively. This is similar for cell number i ¼ 111 with
H 111 ¼ 1 and V 111 ¼ 11 Existence of Entry and Exit Point for Each Facility
  Each facility is modeled to have one single entry and exit point,
ði − εÞ
Hi ¼ i − · N; ∀ i ∈ f1; Ig ð1Þ which occupies one full unit cell. Mathematically, to fix one and
N only one entry and exit point for facility k, a binary variable xi;k and
the constraint set in Eq. (4) are introduced. Binary variable xi;k
 
ði − εÞ equals 1 if the entry and exit point of facility k is located in unit
Vi ¼ þ 1; ∀ i ∈ f1; Ig ð2Þ cell i. Using the constraint set in Eq. (4), only one entry and exit
N
point for each facility will be allowed in one of the unit cells within
the mesh
Availability of Unit Cells for Facility Setup X
I
xi;k ¼ 1; ∀ k ∈ f1; Kg ð4Þ
Fig. 2 shows an example trapezoidal-shaped site; the site boundary i¼1
is outlined by solid lines. Unit cells (which are indicated by dotted
lines) fully inside the site boundary are considered to be available
for facility setup; on the other hand, those outside the boundary or Defining the Physical Size of a Facility Using the Number of
not fully inside the boundary will also be modeled but are unavail- Unit Cells
able for facility setup. Binary parameter ϖi is introduced in the Depending on the given physical size requirement for each facility,
constraint set in Eq. (3), where 1 indicates that the unit cell i is the total number of unit cells required for facility k is defined by Sk
available and 0 indicates that it is not. For example, unit cell i ¼ in the constraint set in Eq. (5). Variable Ak is the input size require-
111 in Fig. 2 is outside the site boundary; thus, ϖ111 ¼ 0. Mean- ment of facility k, and a2 represents the area of a unit cell where a is
while, unit cell i ¼ 22 is inside the boundary; thus, ϖ22 ¼ 1. Be- the side length. Mathematically, dAk =a2 e rounds up the result of
cause this parameter is based on user-specified input, it allows users Ak =a2 to its nearest integer, which is then the minimum required
to define the shape and boundary of any given site area number of unit cells for facility k

© ASCE 04016082-3 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


  n=1 2 3 4 5 n=N=6
Ak
Sk ¼ ; ∀ k ∈ f1; Kg ð5Þ
a2 j= j=
m=1 i-N-1
j=i-N
i-N+1
Assigning a Sufficient Number of Unit Cells to Setup an
Entire Facility 2 j=i-1 i j=i+1
A facility will be divided into two parts: the entry and exit point,
xi;k , and an interior space, yi;k . All parts of the interior space must 3
j=
j=i+N
j=
j=18
i+N-1 i+N+1
be allocated to different unit cells. In the constraint set in Eq. (6),
the sum of the binary variables xi;k and yi;k becomes the total num-
ber of unit cells assigned for facility k; this number should equal Sk 4 j=22 j=23 j=24
as given by the constraint set in Eq. (5), such that the exact physical
size of facility k is modeled 5 j=27 j=28 i=29 j=30

X
I
ðxi;k þ yi;k Þ ¼ Sk ; ∀ k ∈ f1; Kg ð6Þ m=M=6 j=34 j=35 j=36
i¼1
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

Facility Setup in Available Unit Cells Only


The constraint set in Eq. (7) is introduced to ensure that each unit Fig. 3. Modeling of typical unit cells
cell can only be assigned to the setup of a facility if that unit cell is
inside the prespecified site boundary given in Eq. (1) and also if that
unit cell is available—meaning, it is not currently occupied by other
facilities. For example, if the given binary parameter ϖi is 1— n=1 2 3 4 5 n=N=6
indicating that unit cell i is available inside the site boundary for
facility setup—then the unit cell is free for use in the setup of fa- m=1 i=1 j=i+1 j=i-1 i=N
cility k; thus, both xi;k and yi;k can be either 1 or 0. However, if ϖi
is 0, then unit cell i is either outside the site boundary or simply not j= j=
available for any facility setup; thus, the sum of xi;k and yi;k must be
2 j=i+N
i+N+1 i+N-1
j=i+N

forced to 0, using the constraint set in Eq. (7)


3
X
K
ϖi ≥ ðxi;k þ yi;k Þ; ∀ i ∈ f1; Ig ð7Þ
k¼1
4
j= j=
Establishing Direct and Indirect Cell Linkages 5 j=25 j=26 i+N-1 i+N
One important requirement in the proposed cell modeling frame-
work is to maintain facility homogeneity without splitting a facility m=M=6 i=31 j=32 j=i-1 i=I
apart at different locations during unit cell assignment. To satisfy
this key requirement in the formulation, the cell model requires an-
other binary parameter, ωi;j ¼ 1, to denote the adjacency linkage
between any two unit cells, i and j, for being directly or indirectly Fig. 4. Modeling of unit cells at corners
linked. For any two directly linked adjacent unit cells, one of their
side edges is in common. For any two indirectly linked adjacent
unit cells, only one cell corner is in common. If unit cells i and
j have no cell corner in common, the cells are considered to be to different modeling constraints. In general, typical unit cells
not linked at all; thus, ωi;j ¼ 0. Fig. 3 gives a sample site made are governed by the constraint set in Eq. (8)
up of 36 unit cells. In that figure, unit cell i ¼ 29 is directly linked
to four adjacent unit cells (23, 28, 30, and 35) (with entire cell edge ωi;j ¼ 1; ∀i ¼ m · N þ n; 1 ≤ m ≤ M − 2; 2 ≤ n ≤ N − 1;
between two circled corners in common) and unit cell i ¼ 29 is
indirectly linked to other four unit cells (22, 24, 34, and 36) with ∀j ¼ fi − 1; i þ 1; i − N − 1; j − N; i − N þ 1; i þ N − 1;
only a corner (one circle) in common; thus, ω29,22 ; ω29,23 ; ω29,24 ; i þ N; i þ N þ 1g ð8Þ
ω29,28 ; ω29,30 ; ω29,34 ; ω29,35 ; and ω29,36 are set to 1. All other re-
maining unit cells have no linkage with unit cell i ¼ 29; thus, Unit cells in the four corners are governed by the constraint
all associated ω29;j ∀j ¼ f1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 11; 12; 13; 14; sets in Eqs. (9)–(12), respectively. For example, in Fig. 4, unit cell
15; 16; 17; 18; 19; 20; 21; 25; 26; 27; 29; 31; 32; 33g must be 0. i ¼ 31 is directly linked with unit cells j ¼ 25 and j ¼ 32 and
indirectly linked with unit cell j ¼ 26. Thus, ω31,25 ; ω31,26 , and
Classifying Typical, Corner, and Edge Cells in the Cell
ω31,32 are forced to be 1, as governed by the constraint set in
Modeling Framework
Eq. (11)
In Fig. 3, the 16 gray-shaded unit cells are classified as typical unit
cells. Each typical unit cell is always linked and surrounded by ωi;j ¼ 1; ∀i ¼ 1; ∀j ¼ fi þ 1; i þ N; i þ N − 1g ð9Þ
eight unit cells with four direct and another four indirect links
(e.g., typical unit cell i ¼ 22, 23, 27, 28, 29). Each unit cell located ωi;j ¼ 1; ∀i ¼ N; ∀j ¼ fi − 1; i þ N − 1; i þ Ng ð10Þ
in the four corners (e.g., the shaded cells in Fig. 4) is surrounded
and linked by three unit cells. Unit cells along the edges of a site ωi;j ¼ 1; ∀i ¼ ðM − 1Þ · N þ 1; ∀j ¼ fi − N; i − N þ 1; i þ 1g
boundary (e.g., the shaded cells in Fig. 5) are linked by five
unit cells. Having different properties, each unit cell is subject ð11Þ

© ASCE 04016082-4 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


n=1 2 3 4 5 n=N=6

m=1 i-1 i i+1 i-N-1 i-N

2 i+N-1 i+N i+N+1 i-1 i

3 i+N-1 i+N

4 i-N i-N+1

5 i i+1 i-N-1 i-N i-N+1

m=M=6 i+N i+N+1 i-1 i i+1


Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Modeling of unit cells along the edges Fig. 6. Indirectly linked unit cells

Fig. 6. Thus, more constraint sets are required to ensure that all
ωi;j ¼ 1; ∀i ¼ I; ∀j ¼ fi − N − 1; i − N; i − 1g ð12Þ
interior parts, together with the entry and exit point, are first as-
Unit cells along the four edges (e.g., the shaded cells in Fig. 5), signed to unit cells with direct links.
excluding unit cells in the corners, should always be linked to five For typical unit cells, the constraint sets in Eqs. (18) and (19) are
adjacent unit cells. For any given unit cell i along any of the four required. When xi;k ¼ 1 in the constraint set in Eq. (18), the sum of
edges, the constraint sets in Eqs. (13)–(16) are required to establish yi−N;k ; yi−1;k ; yiþ1;k , and yiþN;k must be at least 1, which means that
the correct linkages when the entry and exit point of facility k is located in unit cell i, at
least one of the four directly linked cells (i.e., cells i − N, i − 1,
ωi;j ¼ 1; 2 ≤ i ≤ N − 1; i þ 1, or i þ N) must be occupied by an interior part of facility
k. Similarly, when yi;k equals 1, the sum of the binary variables
∀j ¼ fi − 1; i þ 1; i þ N − 1; i þ N; i þ N þ 1g ð13Þ in the constraint set in Eq. (19) must be 1, thus ensuring that either
an interior part or the entry and exit point of the facility must be
ωi;j ¼ 1; ðM − 1Þ · N þ 2 ≤ i ≤ I − 1; assigned to one of the four directly linked unit cells. In Fig. 6, for
∀j ¼ fi − 1; i þ 1; i − N − 1; i − N; i − N þ 1g ð14Þ example, if unit cell i ¼ 29 is occupied by an interior part of the
facility, then the constraint sets in Eqs. (18) and (19) will force one
of the four unit cells 23, 28, 30, or 35 to be used to setup the same
ωi;j ¼ 1; ∀i ¼ m · N þ 1,1 ≤ m ≤ M − 1; facility
∀j ¼ fi − N; i − N þ 1; i þ 1; i þ N; i þ N þ 1g ð15Þ
yi−N;k þ yi−1;k þ yiþ1;k þ yiþN;k ≥ xi;k ; ∀i ¼ m · N þ n;
ωi;j ¼ 1; ∀i ¼ m · N; 2 ≤ m ≤ M − 1; 2 ≤ m ≤ M − 2; 2 ≤ n ≤ N − 1; 2 ≤ n ≤ N − 1;
∀j ¼ fi − N − 1; i − N; i − 1; i þ N − 1; i þ Ng ð16Þ ∀k ∈ f1; Kg ð18Þ

Facility Assignment to Directly and Indirectly


Linked Unit Cells xi−N;k þ xi−1;k þ xiþ1;k þ xiþN;k þ yi−N;k þ yi−1;k
For any two unit cells, i and j, without either direct or indirect links, þ yiþ1;k þ yiþN;k ≥ yi;k ; ∀i ¼ m · N þ n;
ωi;j ¼ 0 if the entry and exit point of facility k is assigned to unit
cell i—in which case, other interior parts of facility k should not be 2 ≤ m ≤ M − 2; 2 ≤ n ≤ N − 1; ∀k ∈ f1; Kg ð19Þ
assigned to unit cell j, according to the constraint set in Eq. (17),
Similarly, for unit cells in the different corners, the constraint
thus requiring yj;k ¼ 0. If ωi;j ¼ 1, then unit cell j is available for
sets in Eqs. (20)–(27) are provided
facility k setup because yj;k can be either 0 or 1 without violating
the constraint set in Eq. (17)
yiþ1;k þ yiþN;k ≥ xi;k ; ∀i ¼ 1; ∀k ∈ f1; Kg ð20Þ
1 − xi;k þ ωi;j ≥ yj;k ; ∀k ∈ f1; Kg; ∀i ∈ f1; Ig; ∀j ∈ f1; Ig ð17Þ
xiþ1;k þ xiþN;k þ yiþ1;k þ yiþN;k ≥ yi;k ; ∀i ¼ 1; ∀k ∈ f1; Kg ð21Þ
Shaping Facilities by Filling Directly and/or Indirectly
Linked Unit Cells
yi−1;k þ yiþN;k ≥ xi;k ; ∀i ¼ N; ∀k ∈ f1; Kg ð22Þ
With the constraint set in Eq. (17), the entry and exit point and an
interior part of a facility may be assigned onto either directly or
indirectly linked unit cells to match the physical size requirements. xi−1;k þ xiþN;k þ yi−1;k þ yiþN;k ≥ yi;k ; ∀i ¼ N; ∀k ∈ f1; Kg ð23Þ
However, indirect links between the entry and exit point and an
interior part of a facility in two different unit cells can possibly split
a single facility apart, as demonstrated by the six dotted cells in yiþ1;k þ yi−N;k ≥ xi;k ; ∀i ¼ ðM − 1Þ · N þ 1; ∀k ∈ f1; Kg ð24Þ

© ASCE 04016082-5 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


xiþ1;k þ xi−N;k þ yiþ1;k þ yi−N;k ≥ yi;k ; n=1 2 3 4 5 n=N=6

∀i ¼ ðM − 1Þ · N þ 1; ∀k ∈ f1; Kg ð25Þ m=1

yi−N;k þ yi−1;k ≥ xi;k ; ∀i ¼ I; ∀k ∈ f1; Kg ð26Þ 2

xi−1;k þ xi−N;k þ yi−1;k þ yi−N;k ≥ yi;k ; ∀i ¼ I; ∀k ∈ f1; Kg ð27Þ 3

Unit cells along the edges, excluding unit cells at corners, are 4 i=24
always directly linked to three other unit cells. The constraint sets
in Eqs. (28)–(35) provide similar controls for those unit cells 5 VD24,33
yi−1;k þ yiþ1;k þ yiþN;k ≥ xi;k ; 2 ≤ i ≤ N − 1; ∀k ∈ f1; Kg ð28Þ
m=M=6 j=33

xi−1;k þ xiþ1;k þ xiþN;k þ yi−1;k þ yiþ1;k þ yiþN;k ≥ yi;k ; HD24,33


Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

2 ≤ i ≤ N − 1; ∀k ∈ f1; Kg ð29Þ H 24 − H 33

yi−N;k þ yiþ1;k þ yiþN;k ≥ xi;k ; Fig. 7. Separation between a pair of unit cells
∀i ¼ m · N þ 1; 1 ≤ m ≤ M − 2; ∀k ∈ f1; Kg ð30Þ

xi−N;k þ xiþ1;k þ xiþN;k þ yi−N;k þ yiþ1;k þ yiþN;k ≥ yi;k ; distance with unit cell length a, the true physical distance can then
be determined using Eqs. (36) and (37)
∀i ¼ m · N þ 1; 1 ≤ m ≤ M − 2; ∀k ∈ f1; Kg ð31Þ
HDi;j ¼ ðjHi − Hj j þ 1Þ · a; ∀i ≠ j ∈ f1; Ig ð36Þ
yi−N;k þ yi−1;k þ yiþN;k ≥ xi;k ;
VDi;j ¼ ðjV i − V j j þ 1Þ · a; ∀i ≠ j ∈ f1; Ig ð37Þ
∀i ¼ m · N; 2 ≤ m ≤ M − 1; ∀k ∈ f1; Kg ð32Þ

xi−N;k þ xi−1;k þ xiþN;k þ yi−N;k þ yi−1;k þ yiþN;k ≥ yi;k ; Facility Assignment in Linked Unit Cells Involving the
Entry and Exit Point
∀i ¼ m · N; 2 ≤ m ≤ M − 1; ∀k ∈ f1; Kg ð33Þ One new feature in the proposed optimization algorithm is the abil-
ity to assign all facilities in the site area using a discretized cell
yi−1;k þ yiþ1;k þ yi−N;k ≥ xi;k ; approach. Facilities of different shapes can then be modeled, thus
utilizing the best possible usage of the available site area. The entry
ðM − 1Þ · N þ 2 ≤ i ≤ I − 1; ∀k ∈ f1; Kg ð34Þ
and exit point of a facility, occupying one full unit cell, will be
assigned first; all interior parts of that facility will then be assigned
xi−1;k þ xiþ1;k þ xi−N;k þ yi−1;k þ yiþ1;k þ yi−N;k ≥ yi;k ; to adjacently linked unit cells until the physical size requirements
ðM − 1Þ · N þ 2 ≤ i ≤ I − 1; ∀k ∈ f1; Kg ð35Þ have been matched. To model the possible different orientations of
a facility, the length (along the vertical direction) and width (along
In short, the constraint sets in Eqs. (3)–(35) effectively avoid the horizontal direction) must be taken into consideration. To mea-
splitting facilities apart into different cell locations during the sure the separation between the entry and exit point and an interior
solution process. part of a facility, an auxiliary set of binary variables, λi;j;k , is in-
troduced to represent whether unit cells i and j are simultaneously
assigned for the setup of facility k. In the constraint set in Eq. (38),
Shaping of Facilities if both binary variables xi;k and yi;k equal 1, meaning that the entry
and exit point and an interior part of facility k are allocated, respec-
Separations between Pairs of Unit Cells in the Horizontal tively, in unit cells i and j, then λi;j;k must equal 1. Otherwise, if xi;k
and Vertical Directions or yi;k is 0, then λi;j;k must equal 0 according to the constraint sets
A unit cell is the finest unit used in the proposed 2D modeling in Eqs. (39) and (40)
framework. Thus, when measuring the horizontal and vertical sep- 2 − xi;k − yj;k ≥ 1 − λi;j;k ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð38Þ
arations that exist between a pair of unit cells, the number of unit
cells between them is a key unit of measurement. For example,
xi;k ≥ λi;j;k ; ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð39Þ
Fig. 7 graphically shows the separation that exists between two unit
cells, i and j; the number of unit cells between them, along the yj;k ≥ λi;j;k ; ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð40Þ
horizontal and vertical directions, are given by HDi;j and VDi;j ,
respectively. Recall that H i and V i are calculated by Eqs. (1)
and (2), respectively, and identify the horizontal and vertical posi- Facility Assignment in Linked Unit Cells, Excluding the
tions of unit cell i in the mesh system, and jH i − H j j and jV i − V j j Entry and Exit Point
calculate the absolute distance, in unit cells, between the centers of Another similar auxiliary binary variable, ηi;j;k , is defined and
unit cells i and j in the horizontal and vertical directions, respec- governed by the constraint set in Eq. (41) for unit cells i and j,
tively. Furthermore, recall that adding one unit cell to the calculated not involving the entry and exit point, to setup facility k. If both
distance is required to represent the actual distance (in unit cells) yi;k and yj;k equal 1, meaning that two interior parts of facility k
between any one pair of unit cells. By further multiplying the total are assigned to unit cells i and j, then ηi;j;k must equal 1. Otherwise,

© ASCE 04016082-6 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


ηi;j;k equals 0, which is restricted by the constraint sets in Eqs. (42) λi;j;k · HDi;j ≤ ϕk · W k þ ð1 − ϕk Þ · Lk ;
and (43)
∀k ∈ f1; Kg; ∀i ∈ f1; Ig; ∀j ∈ f1; Ig ð46Þ
2 − yi;k − yj;k ≥ 1 − ηi;j;k ; ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð41Þ
ηi;j;k · HDi;j ≤ ϕk · W k þ ð1 − ϕk Þ · Lk ;
yi;k ≥ ηi;j;k ; ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð42Þ ∀k ∈ f1; Kg; ∀i ∈ f1; Ig; ∀j ∈ f1; Ig ð47Þ

yj;k ≥ ηi;j;k ; ∀k ∈ f1; Kg; ∀i; j ∈ f1; Ig ð43Þ


Material Transportation Distance between Facilities
Facility Orientation Satisfying Length and Width The proposed cell model assumes that materials are delivered be-
Requirements tween the entry and exit points of different facilities. The distance
The proposed cell formulation is able to handle irregularities as between two assigned unit cells refers to the distance between their
well as differently shaped facilities and site boundaries. Generally, centers. In Eq. (48), jH i − Hj j and jV i − V j j are the horizontal and
the objective function of site layout planning is set to minimize the vertical distances between unit cells i and j, respectively. By multi-
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

total material transportation costs among different facilities (Huang plying the unit cell length, a, the actual transportation distance, Di;j
and Wong 2015). It is thus expected that facilities be closely packed (as indicated by a hidden line in Fig. 9), can be calculated
and that the orientation of any given facility be optimized for
achieving the best possible site layout shape in terms of unit cell Di;j ¼ ðjHi − Hj j þ jV i − V j jÞ · a; ∀i ≠ j ∈ f1; Ig ð48Þ
assignment. To ensure the best orientation and location of the fa-
cilities during the unit cell optimization process, a binary variable,
ϕk , is required to represent the orientation of facility k. In the con- Existence of Material Flows between Facilities
straint sets in Eqs. (44)–(47), when both λi;j;k and ηi;j;k are 1, then A set of binary variables, Fk;o;u , is defined to identify whether a
λi;j;k · VDi;j and ηi;j;k · VDi;j give the vertical distance between material type, u, is transported from a facility, k, to another facility,
unit cells i and j, while λi;j;k · HDi;j and ηi;j;k · HDi;j give the hori- o. In the constraint set in Eq. (49), a user-defined input, Qk;o;u ,
zontal distance between unit cells i and j (which here represent any specifies the quantity of material type u that is required from fa-
two parts of the given facility). In Fig. 8, for example, all facilities cility k to facility o. Parameter ε 0 is an arbitrary, large integral num-
that occupy five unit cells are different in their shape and orienta- ber. If Qk;o;u is 0, then ε 0 · Qk;o;u is also 0, which in turn forces
tion. For facilities k ¼ 1 and k ¼ 2, when binary variable ϕk ¼ 1 in Fk;o;u to be 0 as well; in other words, the respective material flow
the constraint sets in Eqs. (44) and (45), then Lk must be greater does not exist. Otherwise, ε 0 · Fk;o;u must be greater than Qk;o;u ;
than or equal to both ηi;j;k · VDi;j and λi;j;k · VDi;j ; when binary this then requires Fk;o;u to be 1 so that the existence of the material
variable ϕk ¼ 1 in the constraint sets in Eqs. (46) and (47), then flows may be denoted
W k must be greater than or equal to bothλi;j;k · HDi;j and
ηi;j;k · HDi;j . Therefore, when ϕk ¼ 1, the long edge of facilities ε 0 · Qk;o;u ≥ ε 0 · Fk;o;u ≥ Qk;o;u ; ∀k ≠ o ∈ f1; Kg; ∀u ∈ f1; Ug
k ¼ 1 and k ¼ 2 is parallel to the vertical direction and the short ð49Þ
edge is assigned along the horizontal direction. Otherwise, when
ϕk ¼ 0, the orientation of facility k ¼ 3 will be rotated until its long
edge is parallel to the horizontal direction Establishing Material Transportation Flows between
Facilities Externally
ηi;j;k · VDi;j ≤ ϕk · Lk þ ð1 − ϕk Þ · W k ;
In the proposed cell model, material movements inside a site area
∀k ∈ f1; Kg; ∀i ∈ f1; Ig; ∀j ∈ f1; Ig ð44Þ are referred to as the material flows between the entry and exit
points of different facilities. An auxiliary binary-type variable,
λi;j;k · VDi;j ≤ ϕk · Lk þ ð1 − ϕk Þ · W k ; δ i;j;k;o;u , is defined in Eq. (50) to represent the existence of the
material flows of material type u between the entry and exit points
∀k ∈ f1; Kg; ∀i ∈ f1; Ig; ∀j ∈ f1; Ig ð45Þ

n=1 2 3 4 5 n=N=6

n=1 2 3 4 5 n=N=6 m=1


m=1 i
2
k*
2 Facility k Vi − V j
3 Di , j
Facility 1
3 j
4 o*
φ1 = 1
4 Facility 2
φ2 = 1 5 Facility o
5 ·
Facility 3
φ3 = 0 m=M=6
m=M=6
Hi − H j

Fig. 8. Different facility orientations Fig. 9. Travel distance between facilities for material transportation

© ASCE 04016082-7 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


at unit cells i and j of any two related facilities, k and o, where Total Material Transportation Costs as the Objective
k ≠ o and i ≠ j. Parameter δ i;j;k;o;u is useful for maintaining Function
linearity while formulating relationships between different facili-
The total material transportation costs are defined as an objective
ties; it is governed by the constraint set in Eq. (50). Numerically,
function for optimization including interfacility transportation cost,
if xi;k ¼ 1, xj;o ¼ 1, and Fk;o;u ¼ 1, then a material flow must
C, for materials transported externally among facilities as given in
exist—that is, δ i;j;k;o;u ¼ 1. Otherwise, if xi;k , xj;o , or Fk;o;u equals
Eq. (63) and internal material transportation cost, C 0 , induced for
0, then δ i;j;k;o;u is forced to also be 0, according to the constraint sets
material movements from entry and exit points to all other interior
in Eqs.(51)–(53)
parts inside the same facility as given in Eq. (64)
3 − xi;k − xj;o − Fk;o;u ≥ 1 − δi;j;k;o;u ; X
I X
I X
K X
K X
U

∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig; ∀u ∈ f1; Ug ð50Þ C¼ δ i;j;k;o;u · ρk;o;u · Di;j · Qk;o;u ; ∀k ≠ o;∀i ≠ j
i¼1 j¼1 k¼1 o¼1 u¼1

ð63Þ
xi;k ≥ δi;j;k;o;u ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig; ∀u ∈ f1; Ug ð51Þ
X
I X
I X
K X
U X
K 
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

xj;o ≥ δ i;j;k;o;u ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig; ∀u ∈ f1; Ug ð52Þ C0 ¼ 2 · λi;j;o · Qk;o;u =So · ρo;u
0
· Di;j ;
i¼1 j¼1 o¼1 u¼1 k¼1

∀k ≠ o; ∀i ≠ j ð64Þ
Fk;o;u ≥ δi;j;k;o;u ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig; ∀u ∈ f1; Ug
ð53Þ An auxiliary binary variable δ i;j;k;o;u given in Eq. (63) denotes
which type of material, u, is transported between the entry and exit
points, k and o , of facilities k and o, which are located in unit
Minimum Safety Distance cells i and j, respectively. After multiplying the input parameters
Qk;o;u , ρk;o;u , and Di;j —which (respectively) represent the de-
To ensure that there exists a minimum safety distance separating manded quantity of material type u from facility k to facility o, the
any two facilities such as site staff office and storage for hazardous material transportation cost per unit of material u, and the travel
materials, two auxiliary binary-type variables, δ i;j;k;o
0 and δ i;j;k;o
00 , are distance between unit cells and j—the subtotal external material
defined and governed by Eqs. (54)–(62). When δ i;j;k;o 0
¼ 1, the en- transportation cost C can be obtained using Eq. (63). Eq. (64) is to
try and exit point of facility k and a part of facility o are respectively evaluate the subtotal cost C 0 for the internal material movements
allocated at unit cells i and j where k ≠ o and i ≠ j. Similarly, from the entry and exit point located at unit cell i to all other interior
δ i;j;k;o
00
¼ 1 indicates that a part of facility k and a part of facility parts of unit cells j of the same facility o. A numerical factor of 2 is
o are assigned respectively at unit cells i and j. An input parameter applied to realize the (round trip) traveling paths for (1) distributing
SDk;o specifying a minimum safety distance between facilities k the raw materials onto different unit cells within the facility from
and o is introduced. In the constraint set in Eq. (60), if δ i;j;k;o
0
equals the entry and exit point and (2) returning to the entry and exit point
1, the distance Di;j between unit cell i where the entry and exit from those interior unit cells after the production process. Param-
point of facility k locates and unit cell j where a part of facility eter λi;j;o is a set of binary variables specifying whether unit cell i is
o is located must be greater than the safety distance SDk;o . Eqs. (61) the entry and exit point andP unit cell j is an interior part of a facility
and (62) are similar constraint sets to govern δ i;j;k;o00
and δ i;j;k;o;u o (= 1 if yes and = 0 if no); Kk¼1 Qk;o;u represents the total quantity
fulfilling the respective safety distance requirements of material type u being transported from all other facilities k to
2 − xi;k − yj;o ≥ 1 − δ i;j;k;o
0
; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð54Þ enter facility o; So is a user input for the physical size of a facility
o; ρo;u
0
is a user-specified unit cost for internal material transporta-
tion inside a facility; and Di;j is the physical travel distance be-
xi;k ≥ δ i;j;k;o
0 ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð55Þ tween unit cell i and unit cell j. It is assumed that all materials
entering facility o from all other facilities k are evenly distributed
over all the unit cells assigned for a facility o
yj;o ≥ δi;j;k;o
0 ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð56Þ
Minimize C þ γC 0
Ω
2 − yi;k − yj;o ≥ 1 − δ i;j;k;o
00 ; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð57Þ
subject to constraint sets in Eqs. (1)–(62)

yi;k ≥ δ i;j;k;o
00
; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð58Þ Ω ∈ fδi;j;k;o;u ; Fk;o;u ; xi;k ; yi;k ; φk g

The problem of site layout planning is formulated as a standard


yj;o ≥ δi;j;k;o
00
; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð59Þ BMILP problem in order to minimize C þ γC 0 based on the key
binary variable set Ω ∈ fδi;j;k;o;u ; Fk;o;u ; xi;k ; yi;k ; ϕk g subject to
the linear constraint sets in Eqs. (1)–(62). A user-defined weighting
δ i;j;k;o
0
ðDi;j − SDk;o Þ ≥ 0; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð60Þ parameter γ is introduced in the objective function to weight and
differentiate the influences of the internal and external material
transportation costs in the total material transportation costs dur-
δ i;j;k;o
00 ðDi;j − SDk;o Þ ≥ 0; ∀k ≠ o ∈ f1; Kg; ∀i ≠ j ∈ f1; Ig ð61Þ
ing the optimization process. The present programming problem
can then be effectively solved by a standard branch-and-bound
δ i;j;k;o;u ðDi;j − SDk;o Þ ≥ 0; ∀k ≠ o ∈ f1; Kg; technique using a commercial solver package Gurobi with the
programming language Python to generate the optimal unit cell
∀i ≠ j ∈ f1; Ig; ∀u ∈ f1; Ug ð62Þ arrangements and the respective site layout plans.

© ASCE 04016082-8 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


Table 1. Facility Types and Their Physical Size Requirements
Physical size requirements
Dimensional requirement Requirement in cell unit
Facility Area Ak Length Lk Width W k Area a
Length Width
number (k) Facility ðm2 Þ (m) (m) (cells) (cells) (cells)
1 Main gate 45 7≥ 7≥ 1 1 1
2 Side gate 45 7≥ 7≥ 1 1 1
3 Batching plant 180 ≥13 ≥10 4 ≥2 ≥2
4 Steel storage yard 225 ≥7 ≥6 5 ≥1 ≥1
5 Formwork storage yard 225 ≥6 ≥6 5 ≥1 ≥1
6 Bending yard 180 ≥12 ≥10 4 ≥2 ≥2
7 Cement, sand, aggregate storage yard 225 ≥6 ≥6 5 ≥1 ≥1
8 Curing yard 225 ≥15 ≥12 5 ≥3 ≥2
9 Refuse dumping area 135 ≥7 ≥6 3 ≥1 ≥1
10 Casting yard 135 ≥12 ≥10 3 ≥2 ≥2
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

11 Lifting yard 180 ≥14 ≥11 4 ≥2 ≥2


a
Unit cell size is assumed to be 49 m2 and side length of unit cells is 7 m.

Numerical Example a discretized cell approach. In this section, a numerical example


containing (7 × 7 ¼ 49) unit cells is demonstrated to model the
To demonstrate the applicability of the proposed, discretized cell construction site.
model for designing facility locations, two numerical examples To verify the new discretized cell model, the proposed cell
will be given in which the proposed cell model will be used to model is first used in an attempt to replicate the optimization results
optimize facility locations without safety considerations containing obtained in previous studies. Modeling the 7 × 7 unit cells, a total
the constraint sets in Eqs. (1)–(53) and also with safety constraint of 49 unit cells are arranged in a square shape, as shown in Fig. 10,
sets comprising Eqs. (1)–(62). Both model runs will adopt the in which only 11 selected unit cells i ∈ f5; 9; 21; 23; 24; 29; 32; 35;
total material transportation cost in Eqs. (63) and (64) as the ob- 44; 45; 48g are activated as available site areas, thus duplicating the
jective function for optimization. Problem settings used in the point notation settings as found in Wong et al. (2010). The other
examples have been used in previous research based on point no- remaining 38 unit cells are assumed to be occupied (these are
tations; however, those works did not give any practical consider- shown as shaded) for the cell model verification. Every facility is
ation to the shapes of site and facilities or facility orientations while assumed to occupy only one full unit cell. The optimized facility
assigning facility locations. The example site layout has already locations resulting from the cell model are exactly the same as those
been optimized using genetic algorithms and integer programming resulting from Wong et al. (2010) in Fig. 11.
(Wong et al. 2010). Material flows among different facilities in- Having successfully demonstrated that the proposed cell model
clude completed precast unit (CPU); aggregate, sand, cement is able to replicate the same optimization results as those optimized
(ASC); reinforcement (R); and formwork (F). Eleven facilities by a conventional point modeling framework, more complicated
are modeled with the physical size requirements listed in Table 1. problem conditions will be modeled by relaxing all available unit
The unit transportation costs, ρk;o;u and ρo;u0
, for different types cells for facility assignment, facility sizes, irregular site boundaries,
of raw materials are 4.0 for R (u ¼ 1), 5.0 for ASC (u ¼ 2), 8.0 for irregular shapes, and site orientation as new design optimization
F (u ¼ 3), and 8.5 for CPU (u ¼ 4). The given quantities of differ- parameters. To implement the new cell optimization model, the
ent types of materials demanded from different facilities are given study site area is discretized and represented by structured unit
in Table 2. cells. The availability of the unit cells can be specified by the binary
Conventional models require users to specify all available site parameter ϖi, where ϖi ¼ 1 indicates that unit cell i is inside the
locations as input. A key enhancement provided by the present site boundary and available for facility assignment and ϖi ¼ 0 in-
study is the ability to model a site area and various facilities using dicates that unit cell i is either outside the site boundary or not
available for facility assignment. In Fig. 12, the gray-shaded unit
cells marked with 0 are not available for facility assignment; other
Table 2. Quantity of Demanded Material Flows in the Study Site unit cells, which are marked with 1, are available.
With the physical dimensions and respective length and width
From facility (k) To facility (o) Material flow (Qk;o;u )
constraints given in Table 1, optimization of the cell model can be
1 4 30 (R) conducted based on the problem formulation given in the previous
1 7 20 (ASC) section. In general, to determine the size of unit cells, the approach
2 4 20 (R)
adopted by Hegazy and Elbeltagi (1999), whose study set the size
2 7 15 (ASC)
3 10 35 (ASC) of each cell using the greatest common divisor (g.c.d.) among all
4 6 50 (R) modeled facilities, is followed. In this case study, Columns 1, 3, 4,
5 10 48 (F) and 5 in Table 1 are input dimensional requirements. The physical
6 10 50 (R) area required for each facility ranged from 45 to 225 m2 . The great-
7 3 35 (ASC) est common divisor among all sizes of facilities is 45 m2 in the case
8 11 48 (CPU) study. Because unit cells are modeled in square shape in the pro-
10 8 48 (CPU) posed modeling framework (Assumption 1), the nearest size of a
11 1 28 (CPU) unit cell would be 7 × 7 m ¼ 49 m2 (45 ~ m2 ) and the remaining
11 2 20 (CPU) 2
(4 m ¼ 49 − 45) is assumed to reserve for material transportation

© ASCE 04016082-9 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


i=5
11
0 0 1 1 1 1 0
i=9
1 0 1 1 1 1 1 1
i=21
0 1 1 1 1 1 1
9
i=23 i=24
0 1 1 1 1 1 1
2 3
i=29 i=32 i=35
1 1 1 1 1 1 1
10 4 8
0 1 1 1 1 1 1
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

i=44 i=45 i=48


0 1 1 1 1 1 1
6 5 7

Unavailable Available
0 Space 1 Space
Available 1 Available Location
Location Number
Fig. 12. Unit cell availability for facility setup
i Cell Number

Fig. 10. Discretized cell modeling framework

4 4 4 6

(9) 11 1* 4 4* 6* 6

(1) 1 11 8 8 10 6 5

(7) 9 11* 11 8* 10* 5* 5

(11) 2 (8) 3 2* 11 8 8 10 5 5

(2) 10 (10) 4 (3) 8 7 7* 3 3* 3 9

7 7 7 3 9* 9

(4) 6 (6) 5 (5) 7 1 1

1* The entry/exit 1*
Facility 1
of facility 1
1 1
Available Location
1
Number (1) Facility Number
Fig. 13. Optimal site layout without safety constraints by the cell
Fig. 11. Cell representation of the optimal site layout model

path (Assumption 7). With these, Columns 6–8 in Table 1 can be approach. In total, five different shapes exist in the optimized site
prepared to serve as the user’s input to the cell model as the physical layout plan (L, T, I, U, and F) for different facilities. For example,
size requirements in terms of unit cell numbers for implementing Facilities 8 (curing yard), 9 (refuse dumping area), and 10 (casting
the cell model. yard) are in U, L, and I shapes, respectively. Facilities 3 (batching
The optimized facility locations and layouts are diagrammati- yard), 6 (bending yard), and 11 (lifting yard) each occupy four unit
cally given in Fig. 13, presenting all unit cell usages for the 11 types cells are all in T shape but different orientations. The minimized
of facilities. Observe that—in terms of facility locations, positions subtotal external material transportation cost C is 36,183.0 cost
of the entry and exit points, and the shapes and orientations of units and the minimized subtotal internal material transportation
different facilities—the optimized site layout plan result has greater cost γC 0 is 1,651.76 cost units and their details are listed in Tables 3
solution complexity than does the conventional point modeling and 4, respectively. In Table 3, Column 8 gives the subtotal of

© ASCE 04016082-10 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


external material transportation cost that is the product of the internal material transportation cost inside all facilities, γC 0 , are
material flow quantities (Column 5), unit transportation costs (Col- given in Table 4.
umn 6), and transportation distances between pairs of facilities al- The proposed model is solved by the commercial Gurobi soft-
located in different unit cells (Column 7); the sum of the subtotal ware (academic version) on a desktop computer (Lenovo, China)
cost is 36,183.0 cost units. Similar cost evaluation details for with an i5-2400 3.10 GHz CPU and 4 GB RAM. A total of 912,464

Table 3. Material Flows and Subtotal of External Material Transportation Cost


From To
Material flow Unit Subtotal
Facility Unit cell Facility Unit cell quantity transportation Transportation transportation Construction
(k) (i) (o) (j) (Qk;o;u ) cost (ρk;o;u ) distance (Di;j ) cost material (u)a
1 9 4 11 30 4.0 14 1,680 R
1 9 7 38 20 5.0 35 3,500 ASC
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

2 29 4 11 20 4.0 42 3,360 R
2 29 7 38 15 5.0 21 1,575 ASC
3 40 10 26 35 5.0 14 2,450 ASC
4 11 6 12 50 4.0 7 1,400 R
5 27 10 26 48 8.0 7 2,688 F
6 12 10 26 50 4.0 14 2,800 R
7 38 3 40 35 5.0 14 2,450 ASC
8 25 11 23 48 8.5 14 5,712 CPU
10 26 8 25 48 8.5 7 2,856 CPU
11 23 1 9 28 8.5 14 3,332 CPU
11 23 2 29 20 8.5 14 2,380 CPU
Note: Optimized subtotal cost for external material transportation is C ¼ 36,183.0.
a
u ¼ 1 for R, u ¼ 2 for ASC, u ¼ 3 for F, and u ¼ 4 for CPU.

Table 4. Material Flows and Subtotal of Internal Material Transportation Cost


Entry and exit point Facility parts
Material quantity in Unit transportation Transportation Subtotal
Facility Unit cell Unit cell each
P facility part cost inside distance transportation Material
(o) (i) (j) ( Kk¼1 Qk;o;u =So ) 0 )
facility (ρo;u (Di;j ) cost type (u)a
3 40 39 8.75 5.0 7 306.25 ASC
41 7 306.25
47 7 306.25
4 11 3 10.0 4.0 14 560.00 R
4 7 280.00
5 14 560.00
10 7 280.00
5 27 21 9.6 8.0 14 1,075.20 F
28 7 537.60
34 7 537.60
35 14 1,075.20
6 12 6 12.5 4.0 14 700.00 R
13 7 350.00
20 14 700.00
7 38 37 7.0 5.0 7 245.00 ASC
44 14 490.00
45 7 245.00
46 14 490.00
8 25 17 9.6 8.5 14 1,142.40 CPU
18 7 571.20
31 14 1,142.40
32 7 571.20
10 26 19 16.0 8.5 7 952.00 CPU
33 7 952.00
11 23 23 12.0 8.5 7 714.00 CPU
16 7 714.00
24 7 714.00
16,517.55
Note: Optimized subtotal cost for internal material transportation is ¼ 0.05 × 2 × 16,517.55 ¼ 1,651.76γC0 .
a
u ¼ 1 for R, u ¼ 2 for ASC, u ¼ 3 for F, and u ¼ 4 for CPU.

© ASCE 04016082-11 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


11 facilities. The entire site layout and shapes of some facilities will
be changed as well. While fulfilling the new safety requirements,
4 4 11 11
the objective function value that is the total material transportation
cost, C þ γC 0 , would be increased by 12.93% from 37,834.76 to
1* 4* 8 11 * 11 42,726.36 cost units. Tables 5 and 6 provide the calculation details
for the subtotal material transportation costs C 0 and γC 0 with safety
4 4 8 8* 8 5 constraints, respectively.

6 6* 8 10 * 5* 5 Conclusions

Facility layout design is an essential planning step in order to re-


2* 6 6 10 10 5 5
solve potential spatial conflicts while setting different facilities onto
a given site location during the operational stage. Conventionally,
7 7* 3 3* 3 9 available spaces and facilities have been modeled and represented
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

by simplified point notations without any consideration given to


the requirements presented by actual facility shapes and orienta-
7 7 7 3 9* 9
tions. The proposed cell optimization model utilizes small unit cells
to discretize any given site area in a way that facilitates represen-
1 1 tations of irregular site spaces and facilities.
1* The entry/exit 1* Having verified the cell model’s ability to replicate an
of facility 1 Facility 1 optimal site layout that was previously determined by conventional
1 1
point modeling, the cell model was then applied to more compli-
cated requirements involving different facility sizes and orienta-
Fig. 14. Optimal site layout with safety constraints by the cell model tions, irregular site boundaries, and irregular facility shapes.
More realistic design constraints regarding size, orientation, and
shape were added in the form of linear-type mathematical con-
straints. The cell model was then formulated as a BMILP, and a
binary variables and 1,376,584 linear constraints have been built standard branch-and-bound technique was applied to obtain the
for the cell model in the site layout optimization problem. The total global optimal solution. A numerical example consisting of 11 fa-
computation time required to reach the global optimal solution is cilities to be allocated in a 7 × 7 unit cell system was demonstrated.
approximately 7.81 h (~28,120 s). A total of 912,464 binary variables and 1,376,584 linear con-
For the cell model with the additional safety constraint sets in straints were developed; the solution time was approximately
Eqs. (54)–(62) added, safety distance SD1,11 ¼ 10 m is specified to 7.8 h.
separate the lifting yard (Facility 11) and the entrance gate (Facility 1) Results indicate that the cell model is capable of optimizing very
and another safety distance SD2,11 ¼ 30 m is set to separate the different site arrangements in order to setup different facilities at
lifting yard (Facility 11) and the exit gate (Facility 2). With these various locations. Because different shapes and orientations may
constraint sets, the lifting yard (Facility 11) can no longer be allo- be used when allocating facilities, the available site space can
cated next to the two entrance and exit gates of the site but is forced be better utilized. Future work will likely enhance the cell model
to be set up at a new location (top right-hand corner of the site) as such that it may be used to plan site layouts for multiple construc-
shown in Fig. 14 presenting a new optimized site layout plan for all tion stages at a time.

Table 5. Material Flows and Subtotal of External Material Transportation Cost with Safety Constraints
From To
Material flow Unit Transportation Subtotal
Facility Unit cell Facility Unit cell quantity transportation distance transportation Construction
(k) (i) (o) (j) (Qk;o;u ) cost (ρk;o;u ) (Di;j ) cost material ua
1 9 4 10 30 4.0 7 840 R
1 9 7 38 20 5.0 35 3,500 ASC
2 29 4 10 20 4.0 35 2,800 R
2 29 7 38 15 5.0 21 1,575 ASC
3 40 10 26 35 5.0 14 2,450 ASC
4 10 6 24 50 4.0 14 2,800 R
5 27 10 26 48 8.0 7 2,688 F
6 24 10 26 50 4.0 14 2,800 R
7 38 3 40 35 5.0 14 2,450 ASC
8 19 11 12 48 8.5 7 2,856 CPU
10 26 8 19 48 8.5 7 2,856 CPU
11 12 1 9 28 8.5 21 4,998 CPU
11 12 2 29 20 8.5 49 8,330 CPU
Note: Optimized subtotal cost for external material transportation cost is C ¼ 40,943.0.
a
u ¼ 1 for R, u ¼ 2 for ASCE, u ¼ 3 for F, and u ¼ 4 for CPU.

© ASCE 04016082-12 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


Table 6. Material Flows and Subtotal of Internal Material Transportation Cost with Safety Constraints
Entry and exit point Facility parts
Material quantity in Unit transportation Transportation Subtotal
Facility Unit cell Unit cell each
P facility part cost inside distance transportation Material
(o) (i) (j) ( Kk¼1 Qk;o;u =So ) 0 )
facility (ρo;u (Di;j ) cost type ua
3 40 39 8.75 5.0 7 306.25 ASC
41 7 306.25
47 7 306.25
4 10 3 10 4.0 7 280.00 R
4 14 560.00
16 14 560.00
17 7 280.00
5 27 21 9.6 8.0 14 1,075.20 F
28 7 537.60
34 7 537.60
35 14 1,075.20
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

6 24 23 12.5 4.0 7 350.00 R


30 14 700.00
31 7 350.00
7 38 37 7.0 5.0 7 245.00 ASC
44 14 490.00
45 7 245.00
46 14 490.00
8 19 11 9.6 8.5 14 1,142.40 CPU
18 7 571.20
20 7 571.20
25 14 1,142.40
10 26 32 16 8.5 14 1,904.00 CPU
33 7 952.0
11 12 5 12 8.5 7 714.00 CPU
6 14 1,428.00
13 7 714.00
17,833.55
Note: Optimized subtotal cost for internal material transportation is ¼ 0.05 × 2 × 17,833.55 ¼ 1,783.36γC 0 .
a
u ¼ 1 for R, u ¼ 2 for ASC, u ¼ 3 for F, and u ¼ 4 for CPU.

Acknowledgments M = total number of rows in a cell mesh for site


representation;
This work was supported by the City University of Hong Kong’s m = row number of a cell mesh for site representation;
Strategic Research Grant under Grant No. CityU 7004370 and N = total number of columns in a cell mesh for site
Natural Science Foundation of China under Grant No. NSFC representation;
5150081022. n = column number of a cell mesh for site representation;
Qk;o;u = quantity of demand flow of material type u between
facilities k and o;
Notation
SDk;o = minimal distance between facilities k and o;
The following symbols are used in this paper: Sk = number of unit cells required to setup facility k;
a = unit length of a square unit cell; U = total number of material types;
Ak = actual size (area) of facility k; u = material types;
C = subtotal of external material transportation cost among V i = vertical position of a unit cell i in a cell mesh;
facilities in a site; VDi;j = vertical separation between two unit cells i and j;
C 0 = subtotal of internal material transportation cost inside W k = physical width of facility k;
facilities; xi;k = binary decision variable where xi;k ¼ 1 means the entry
Di;j = rectangular distance between unit cells i and j; and exit point of facility;
Fk;o;u = binary variable where Fk;o;u ¼ 1 means that a material yi;k = binary variable where yi;k ¼ 1 means that facility k
type u is delivered from facility k to another facility o, other than the entry and exit point is setup in a unit cell i,
and Fk;o;u ¼ 0 if no such material flow exists; and yi;k ¼ 0 if not;
Hi = horizontal position of a unit cell i in a cell mesh; γ = weighting parameter to specify the relative importance
HDi;j = horizontal separation between two unit cells i and j; between external (outside facilities) and internal (inside
I = total number of cells in a cell mesh; facilities) material transportation costs in the
i; j = cell numbers in a cell mesh; optimization process;
K = total number of facility types; δ i;j;k;o;u = auxiliary binary variable where δi;j;k;o;u ¼ 1 means that
k = setup in a unit cell i, where xi;k ¼ 0 if not; a material type u is delivered from facility k to o through
k; o = facility types; entry and exit points located in respective unit cells i and
Lk = physical length of facility k; j, and δ i;j;k;o;u ¼ 0 if not;

© ASCE 04016082-13 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1


δi;j;k;o
0
= auxiliary binary variable where δi;j;k;o
0
¼ 1 means that Kulturel-Konak, S. (2012). “A linear programming embedded probabilistic
entry and exit point of facility k and a part of facility o tabu search for the unequal-area facility layout problem with flexible
are located in respective unit cells i and j, and δ i;j;k;o
0
¼ bays.” Eur. J. Oper. Res., 223(3), 614–625.
0 if not; Kulturel-Konak, S., and Konak, A. (2011). “A new relaxed flexible
bay structure representation and particle swarm optimization for the un-
δi;j;k;o
00 = auxiliary binary variable where δ i;j;k;o
00 ¼ 1 means that a
equal area facility layout problem.” Eng. Optim., 43(12), 1263–1287.
part of facility k and a part of facility o are located in Lam, K. C., Tang, C. M., and Lee, W. C. (2005). “Application of the
respective unit cells i and j, and δi;j;k;o
00 ¼ 0 if not; entropy technique and genetic algorithms to construction site layout plan-
ε = arbitrary small number; ning of medium-size projects.” Constr. Manage. Econ., 23(2), 127–145.
ε 0 = arbitrary large number; Lee, H. Y. (2012). “Integrating simulation and ant colony optimization
ρk;o;u = unit cost of transporting material type u between to improve the service facility layout in a station.” J. Comput. Civ. Eng.,
facilities k and o; 10.1061/(ASCE)CP.1943-5487.0000128, 259–269.
ρo;u
0 = unit cost of transporting material type u inside Li, H., and Love, P. E. D. (1998). “Site-level facility layout using genetic
facility o; algorithms.” J. Comput. Civ. Eng., 10.1061/(ASCE)0887-3801(1998)
ϕk = binary variable where ϕk ¼ 1 indicates that the long side 12:4(227), 227–231.
of facility k is assigned along the vertical direction, and Li, H., and Love, P. E. D. (2000). “Genetic search for solving construction
Downloaded from ascelibrary.org by Budi Atmoko on 05/09/17. Copyright ASCE. For personal use only; all rights reserved.

site-level unequal-area facility layout problems.” Autom. Constr., 9(2),


ϕk ¼ 0 indicates that the long side is assigned along the
217–226.
horizontal direction; Li, Z., Shen, W., Xu, J., and Lev, B. (2015). “Bilevel and multi-objective
ωi;j = binary parameter where ωi;j ¼ 1 means that a unit cell j dynamic construction site layout and security planning.” Autom.
is located around another unit cell i, and ωi;j ¼ 0 if not; Constr., 57, 1–16.
and Liggett, R. (2000). “Automated facilities layout: Past, present and future.”
ϖi = binary parameter where ϖi ¼ 1 means that a unit cell i Autom. Constr., 9(2), 197–215.
is available for facility setup, and ϖi ¼ 0 if not Mavridou, T. D., and Pardalos, R. M. (1997). “Simulated annealing and
available. genetic algorithms for the facility layout problem: A survey.” Comput.
Optim. Appl., 7(1), 111–126.
McKendall, A. R., and Hakobyan, A. (2010). “Heuristics for the dynamic
facility layout problem with unequal-area departments.” Eur. J. Oper.
References
Res., 201(1), 171–182.
Andayesh, M., and Sadeghpour, F. (2013). “Dynamic site layout planning Meller, R. D., and Bozer, Y. A. (1997). “Alternative approaches to solve
through minimization of total potential energy.” Autom. Constr., 31, the multi-floor facility layout problem.” J. Manuf. Syst., 16(3), 192–203.
92–102. Meller, R. D., and Gau, K. Y. (1996). “Facility layout problem: Recent and
Corry, P., and Kozan, E. (2004). “Ant colony optimisation for machine emerging trends and perspectives.” J. Manuf. Syst., 15(5), 351–366.
layout problems.” Comput. Optim. Appl., 28(3), 287–310. Sanad, H. M., Ammar, M. A., and Ibrahim, M. E. (2008). “Optimal
Drira, A., Pierreval, H., and Hajri-Gabouj, S. (2007). “Facility layout prob- construction site layout considering safety and environmental aspects.”
lems: A survey.” Annu. Rev. Control, 31(2), 255–267. J. Civ. Eng. Manage., 10.1061/(ASCE)0733-9364(2008)134:7(536),
Easa, S. M., and Hossain, K. M. A. (2008). “New mathematical optimiza- 536–544.
tion model for construction site layout.” J. Civ. Eng. Manage., 10.1061/ Scholz, D., Petrick, A., and Domschke, W. (2009). “STaTS: A slicing tree
(ASCE)0733-9364(2008)134:8(653), 653–662. and tabu search based heuristic for the unequal area facility layout prob-
Elbeltagi, E., and Hegazy, T. (2001). “A hybrid AL-based system lem.” Eur. J. Oper. Res., 197(1), 166–178.
for site layout planning in construction.” Comput. Aided Civ. Inf., Singh, S., and Sharma, R. (2006). “A review of different approaches to the
16(2), 79–93. facility layout problems.” Int. J. Adv. Manuf. Technol., 30(5–6), 425–433.
Elbeltagi, E., Hegazy, T., and Eldosouky, A. (2004). “Dynamic layout Tam, C. M., Tong, T. K. L., and Chan, W. K. W. (2001). “Genetic algorithm
of construction temporary facilities considering safety.” J. Civ. Eng. for optimizing supply locations around tower crane.” J. Civ. Eng.
Manage., 10.1061/(ASCE)0733-9364(2004)130:4(534), 534–541. Manage., 10.1061/(ASCE)0733-9364(2001)127:4(315), 315–321.
Elbeltagi, E., Hegazy, T., and Hosny, A. H. (2001). “Schedule-dependent Tommelein, I. D., Levitt, R. E., and Hayes-Roth, B. (1992). “Site-layout
evolution of site layout planning.” Constr. Manage. Econ., 19(7), modeling: How can artificial intelligence help?” J. Civ. Eng. Manage.,
689–697. 10.1061/(ASCE)0733-9364(2001)127:4(315), 315–321.
El-Rayes, K., and Said, H. (2009). “Dynamic site layout planning using Tommelein, I. D., and Zouein, P. P. (1993). “Interactive dynamic layout
approximate dynamic programming.” J. Comput. Civ. Eng., 10.1061/ planning.” J. Civ. Eng. Manage., 10.1061/(ASCE)0733-9364(1993)
(ASCE)0887-3801(2009)23:2(119), 119–127. 119:2(266), 266–287.
Gurobi version 5.5 [Computer software]. Gurobi Optimization, Houston, Ulutas, B. H., and Kulturel-Konak, S. (2013). “Assessing hypermutation
TX. operators of a clonal selection algorithm for the unequal area facility
Hegazy, T., and Elbeltagi, E. (1999). “EvoSite: Evolution-based model for layout problem.” Eng. Optim., 45(3), 375–395.
site layout planning.” J. Comput. Civ. Eng., 10.1061/(ASCE)0887-3801 Wong, C. K., Fung, I. W. H., and Tam, C. M. (2010). “Comparing mixed-
(1999)13:3(198), 198–206. integer programming and genetic algorithm for construction site facility
Huang, C., and Wong, C. K. (2015). “Optimisation of site layout planning layout planning.” J. Civ. Eng. Manage., 10.1061/(ASCE)CO.1943
for multiple construction stages with safety considerations and require- -7862.0000214, 1116–1128.
ments.” Autom. Constr., 53, 58–68. Yeh, I. C. (1995). “Construction-site layout using annealed neural-network.”
Huang, C., Wong, C. K., and Tam, C. M. (2010). “Optimization of material J. Comput. Civ. Eng., 10.1061/(ASCE)0887-3801(1995)9:3(201), 201–208.
hoisting operations and storage locations in multi-storey building Zhang, H., and Wang, J. Y. (2008). “Particle swarm optimization for con-
construction by mixed-integer programming.” Autom. Constr., 19(5), struction site unequal-area layout.” J. Constr. Eng. Manage., 10.1061/
656–663. (ASCE)0733-9364(2008)134:9(739), 739–748.
Huang, C., Wong, C. K., and Tam, C. M. (2011). “Optimization of tower Zouein, P. P., Harmanani, H., and Hajar, A. (2002). “Genetic algorithm
crane and material supply location operations and storage locations for solving site layout problem with unequal size and constrained
in multi-storey building construction by mixed-integer programming.” facilities.” J. Comput. Civ. Eng., 10.1061/(ASCE)0887-3801(2002)
Autom. Constr., 20(5), 571–580. 16:2(143), 143–151.
Konak, A., Kulturel-Konak, S., Norman, B. A., and Smith, A. E. (2006). “A Zouein, P. P., and Tommelein, I. D. (1999). “Dynamic layout planning
new integer programming formulation for facility layout design using using a hybrid incremental solution method.” J. Civ. Eng. Manage.,
flexible bays.” Oper. Res. Lett., 34(6), 660–672. 10.1061/(ASCE)0733-9364(1999)125:6(400), 400–408.

© ASCE 04016082-14 J. Constr. Eng. Manage.

J. Constr. Eng. Manage., 2017, 143(1): -1--1

You might also like