Automation in Construction 12 (2003) 749 764

www.elsevier.com/locate/autcon

A hybrid CAD-based construction site layout planning system

using genetic algorithms
Hesham M. Osman, Maged E. Georgy *, Moheeb E. Ibrahim
Structural Engineering Department, Faculty of Engineering, Cairo University, Giza, Egypt
Accepted 26 June 2003

Abstract
The efficient layout planning of a construction site is a fundamental task to any project undertaking. In an attempt to enhance
the general practice of layout planning of construction sites, the paper introduces a novel approach for producing the sought
layouts. This approach integrates the highly sophisticated graphical capabilities of computer-aided design (CAD) platforms with
the robust search and optimization capabilities of genetic algorithms (GAs). In this context, GAs are utilized from within the
CAD environment to optimize the location of temporary facilities on site. The functional interaction between GAs and CAD
and the details of the GA-based layout optimization procedure are presented. A fully automated computer system is further
developed to demonstrate the practicality of the chosen approach. In order to evaluate the systems performance, a local
construction project with a 24,000m2 site is used. The automated system produced highly satisfactory results and showed
notable flexibility through its CAD-based input/output media.
D 2003 Elsevier B.V. All rights reserved.
Keywords: Site layout planning; Computer-aided design (CAD); Genetic algorithms

1. Introduction
Efficient layout planning of a construction site is
fundamental to any successful project undertaking.
This task usually consists of identifying the temporary
facilities needed to support construction operations,
determining their size and shape, and positioning
them in the unoccupied areas within the site boundaries. Examples of these temporary facilities include
offices and tool trailers, parking lots, warehouses,
batch plants, maintenance areas, fabrication yards or
buildings, staging areas and lay-down areas [1].
* Corresponding author. Tel.: +20-10-5848060.
E-mail address: mgeorgy@eng.cu.edu.eg (M.E. Georgy).
0926-5805/$ - see front matter D 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0926-5805(03)00058-X

The project manager or planner usually performs

the task of preparing the site layout based on his/her
own knowledge and expertise. Apparently, this could
result in layouts that differ significantly from one
person to another. To put this task into more perspective, researchers have introduced different approaches
to systematically plan the layout of construction sites
[1 9]. These approaches differ from one another in
the level of detail they provide and the extent to
which they yield a well round solution to the rather
complicated problem of layout planning of construction sites.
Site layout planning can generally be classified
according to two main aspects: (1) method of facility
assignment and (2) layout planning technique. With

750

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

regard to the method of facility assignment, or in other

words the manner in which temporary facilities are
assigned on site, two distinct methods are commonly
encountered. The two methods are called facility to
location assignment and facility to site assignment. The
method of facility to location assignment assigns a set
of predefined facilities to a set of predefined locations
on site such that (number of locations z number of
facilities). The method of facility to site assignment, on
the other hand, assigns a set of predefined facilities to
any unoccupied space available on site.
The method of facility to location assignment
frequently neglects one important issue, that of facility size. All locations are assumed to be able to fit all
facilities. This assumption is weakened by the fact
that there are usually substantial differences in size
among most construction site facilities. The method
of facility to site assignment is considered more
generic as it assumes that the planner has not yet
settled on the feasible locations for facility assignment. Nonetheless, during this type of assignment,
many spatial requirements must be satisfied simultaneously. This poses extra computational burden on
any automated site layout planning system that adopts
the later approach.

The second aspect, i.e., the layout planning technique, concerns the technique used in performing the
assignment process of temporary facilities. Many techniques have generally been utilized in the past to
perform the assignment process, ranging from purely
mathematical models to knowledge-based systems.
However, researchers have not reached a consensus
on or acknowledged a certain technique to be more
suitable than the others.
Mathematical techniques usually involve the identification of one or more goals that the sought layout
should strive to achieve. A widely used goal is the
minimization of transportation costs on site. These
goals are commonly interpreted to what mathematicians term objective functions. This objective
function is then optimized under problem-specific
constraints to produce the desired layout. Systems
utilizing knowledge-based techniques, in contrast,
provide rules that assist planners in layout planning
rather than perform the process based purely on a
specified optimization goal(s). Fig. 1 shows the
classification of some recent studies in construction
site layout planning based on the aforementioned
aspects of method of facility assignment and layout
planning technique.

Fig. 1. Recent studies in construction site layout planning.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

751

2. A computer-aided design (CAD)-based

approach for site layout planning

Alex [13] suggest that the use of GAs in optimization

is appropriate in the following circumstances:

CAD has been experiencing great advances since

the late 1980s. Its use became inevitable in many
engineering disciplines. However, the use of CAD
applications in the construction stage of a project
substantially lags behind the use of their counterparts in the design stage. Realizing their potential,
Mahoney and Tatum [10] reported the broad benefits of using CAD in managing/planning many
construction site operations. In particular, they suggested that CAD could be used to plan construction
site layouts, as adoption of such systems allows
easy and accurate visualization of the relationship
between the permanent structures and temporary
facilities on site.
Site layout planning is evidently graphical in
nature. Site boundaries, existing buildings on site,
obstacles and temporary site facilities all occupy space
and have distinct shapes. Thus, the need to represent
the relationship between all these entities in some sort
of graphical format can be quite advantageous. For
such reason, Cheng and OConnor [4] utilized a
geographical information system (GIS) to assist in
the graphical representation of the site layout problem.
However, full implementation of CAD-based site
layout planning systems, particularly those utilizing
mathematical techniques, has generally been limited
till our time.
The current study introduces a novel approach
for solving the site layout planning problem. It
utilizes genetic algorithms (GAs), as function optimizers, in determining the temporary facility locations according to the graphical information
depicted in a CAD environment. Based on the
classification presented earlier, this approach performs a facility to site assignment using a mathematical layout planning technique.
Simply, GAs are algorithms that encode a potential
solution to a specific problem on a simple chromosome like data structure and apply recombination
operators to these structures so as to improve the
solution while preserving all critical information [11].
Because of their characteristic of not utilizing gradient
information, GAs are highly applicable to problems
having non-differentiable functions, as well as functions with multiple local optima [12]. Al-Tabtabi and

1. Conventional statistical and mathematical methods

are inadequate.
2. The problem is very complex, because the possible
solution space is too large to analyze in finite time.
3. The additional information available to guide the
search is absent or not sufficient, so conventional
methods are not practical.
4. The solution to the problem can be encoded in the
form of strings and characters.
5. The problem is large and poorly understood.
6. There is an urgent need for near-optimal solutions
to use as starting points for conventional optimization methods.
Several of the aforementioned circumstances promote the use of GAs in solving the site layout planning
problem. Firstly, when modeling a large construction
site, the available solution space is immense. The size
of the solution space increases exponentially with the
number of temporary facilities to be assigned and the
available areas for facility placement. For illustrative
purposes, consider a certain sector of a construction
site with dimensions of 100  100 m. Consider also
that no obstacles or permanent structures are present in
this sector of the construction site and that four
temporary facilities (1  1 m) need to be assigned in
the given space.
Let l Number of locations available for
temporary facility assignment
based on a 1 m pitch=increment
r Number of temporary facilities needed to be
assigned
Excluding all geometrical constraints among the temporary facilities
li100  100 10; 000

Solution Space l Pr 10;000 P4 1  1016 :

Another fundamental reason that makes GAs suitable for solving the problem at hand is the fact that the
solution can be easily encoded in the form of strings.
This will be explained in detail in Section 4. Finally,

752

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

finding a comprehensive solution to the site layout

problem is not always as simple as minimizing an
objective function. Conditions on construction sites
involve far more constraints, variables and uncertainties than those taken into consideration in most mathematical approaches. Practically, the difference between optimum and near optimum solution is not
that significant, as even the optimum solution may
require slight enhancements dictated by unforeseen
site conditions.

3. The objective function

Researchers using mathematical techniques in site
layout planning have developed many forms to represent their optimization goal(s) or objective function(s). The pseudo models of these objective
function(s) are given in Table 1.
The objective function of several models given in
Table 1 takes the general form:
!
P1 X
P
X
Min:
3
Wi; j di; j
i1 ji1

Where P is the total number of permanent and

temporary facilities on site; i,j is a certain pair of
permanent and/or temporary facilities on site; di, j is
the distance between facilities i and j; Wi, j is a term
Table 1
Objective functions used in the literature
No.

Pseudo model of the objective

function

Study
(year)

To minimize the frequency of trips

made by construction personnel
To minimize the total transportation
costs of resources between facilities
To minimize the cost of facility
construction and the interactive cost
between facilities
To minimize the total transportation
costs of resources between facilities
(presented through a system of
proximity weights associated with
an exponential scale)
To minimize the total transportation
costs of resources between facilities
and the total relocation costs
(presented through a system of
proximity weights and relocation weights)

[5]

2
3

[8,9]
[1]

[6]

[7]

Table 2
The six-value scale commonly used in industrial facility layout
planning
Desired relationship
between facilities

Proximity
weight

Absolutely necessary (A)

Especially important (E)
Important (I)
Ordinary closeness (O)
Unimportant (U)
Undesirable (X)

81
37
9
3
1
0

representing either the actual transportation cost per

unit distance between facilities i and j (taking into
consideration the number of trips made) or a relative
proximity weight that reflects the required closeness
between facilities i and j.
Using actual transportation costs to represent the
term Wi,j has the clear objective of minimizing the
total transportation costs between site facilities. The
objective is not as apparent when using the relative
proximity weight representation. However, obtaining
accurate values for the actual inter-facility transportation costs can become quite difficult, especially during
project planning stages. This limitation promotes the
use of proximity weights instead as they are generally
much easier for the site planner to provide.
Several scales have been adopted in engineering
applications to represent the proximity weights and
facilitate their verbal representation. One common
scale used in industrial facility layout planning is
shown in Table 2 [14].
Many scales can be used to represent the proximity
weights; the scale shown in Table 2 being an example.
The site planner can use any other preferred scale
appropriate to the case in hand. However, This study
will particularly adopt the scale presented in Table 2. It
is to be noted that whether accurate inter-facility
transportation costs are provided or relative proximity
weights are provided makes no difference whatsoever
with the functionality of the automated system presented in Section 4.

4. Automated CAD-based site layout planning

system
The automated CAD-based site layout planning
system integrates the highly sophisticated graphical

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

capabilities of CAD platforms with the robust search

and optimization capabilities of genetic algorithms for
producing the desired site layout(s). A schematic
diagram of the system is illustrated in Fig. 2. The
system is composed of three main components, i.e.,
input media, optimization engine and output media. In
order to perform its presumed function, the system
utilizes three main groups of data, namely, site geometrical data, temporary facility data and facility cost
data.

753

Part of the systems novelty lies in its utilization of

CAD capabilities as input/output media. The fact that
most construction companies have their project plans
and drawings in a CAD format greatly facilitates the
use of the system. This way the plans and drawings can
be input directly to the system. Genetic algorithms are
then employed to perform the optimization process
using the objective function described in Eq. (3).
Following the optimization process, the system produces the desired layout(s) in the form of a series of CAD

Fig. 2. Structure of the automated CAD-based site layout planning system.

754

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

drawings with all temporary facilities placed in their

proper (optimal) locations. Limited efforts can further
be made to integrate the automated site layout planning
system with other commonly known planning functions for a comprehensive project management system.
The automated system was fully implemented
using AutoCADk and greatly benefited from its
programmable features and integrated capabilities
with MS Visual Basick, with which the optimization
engine was developed.
4.1. Use of CAD capabilities
Performing the optimization process largely
depends on identifying the specifics of the input
CAD drawing, such as, site boundaries, permanent
facilities and obstacles. Accurately identifying the
available space on site for assigning the temporary
facilities is essential to yield a feasible solution. In this
context, CAD is utilized in performing two main
tasks: (1) space detection and (2) constraint satisfaction. The first task is performed only once prior to the
execution of the optimization process while the second task is continuously performed throughout the
optimization process.
4.1.1. Space detection
Space detection concerns the identification of unoccupied space available for assignment of temporary
facilities on site. It depends on the concept of space
discretization, shown in Fig. 3. In space discretiza-

tion, the 2-D space encompassed by the boundaries of

the construction site is divided into an orthogonal X
Y grid. This grid is then coded, each grid cell having a
unique (X,Y) coordinate. The required level of accuracy in facility assignment determines the increments
or pitch of the orthogonal grid. For example, if
facilities are required to be assigned within an accuracy of 2 m, space discretization should be performed
as illustrated on the left part of Fig. 3. After the grid
coding is complete, grid cells with unoccupied space
are identified after those with occupied space are
excluded from the entire set. It is worth mentioning
that the grid coding using the (X,Y) coordinates forms
the building block of the GA string coding, as will be
described in Section 4.2.
During space detection, CAD performs two main
steps: (1) identification of space encompassed by site
boundaries and (2) identification of fixed facilities and
obstacles.
4.1.1.1. Identification of space encompassed by site
boundaries.
(I) Using the rectilinear coordinates of the site boundarys vertices, CAD identifies a set of equations to
represent the edges of the site boundary:
a1 y b1 x c 1

a2 y b2 x c 2

a3 y b3 x c 3 . . .
(II)

an y bn x c n

Using a point inside the boundary, CAD

identifies the space encompassed by the edges

Fig. 3. Space discretization.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

755

of the site boundary based on a set of

inequalities:
a1 y z b1 x c 1 a2 y z b2 x c 2
a3 y z b3 x c 3 . . . an y z bn x c n
(III) By toggling through coordinates of all grid points
inside the Bounding Box of the site boundary,
CAD selects those grid points that simultaneously satisfy all linear inequalities to represent
the grid spaces encompassed by the site boundary. Any grid point that does not satisfy any
linear inequality is determined to lie outside the
site boundary.
4.1.1.2. Identification of fixed facilities and obstacles. Grid spaces inside the site boundary that are
occupied by permanent facilities and obstacles are
removed from those detected in 1. The remaining
grid spaces represent the solution space for assignment of temporary facilities.
4.1.1.3. Illustration. To illustrate the process of space
detection, consider the illustrative example shown in
Fig. 4. Following the identification of the site boundaries as a set of linear equations, the space encompassed
by the five edges of the site boundary can be identified
as the space that satisfies the following inequalities
simultaneously:
(1)
(2)
(3)
(4)
(5)
(6)

xz5
y V x + 55
y V  2x + 180
x V 75
y z 10
y z  x + 35

4.1.2. Constraint satisfaction

Geometrical constraints are vital in the layout
process. It is of utmost importance that temporary
facilities be placed: (1) inside the site boundaries
and (2) in such a manner that no overlap occurs
between any two temporary facilities or between
temporary facilities and permanent facilities. To
satisfy the geometrical constraints, two main modules are utilized, namely, CheckSite module and
CheckOverlap module. Both modules have been

Fig. 4. Illustrated example for space detection.

designed to deal with temporary facilities having

rectangular shapes.
4.1.2.1. CheckSite module. This module ensures
that any temporary facility: (1) lies inside the site
boundaries and (2) does not overlap with any permanent facility or site obstacle. This requires as input
four variables; the X-coordinate and Y-coordinate of
the bottom left corner of the temporary facility and its
dimensions in the X and Y directions. It provides a
Boolean true/false output. In its operation it toggles
through all grid coordinates occupied by the facility
and compares them with all available X and Y coordinates (Fig. 5).
If any (X,Y) of facility K Available(X,Y)
Then CheckSite = False
Else CheckSite = True.
4.1.2.2. CheckOverlap module. This module
ensures that any two temporary facilities do not
overlap with each other. Two sequential tasks are
performed in this module (Fig. 6):
(a) To make sure that the temporary facility being
checked does not occupy a space already being

756

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

Fig. 5. Functionality of the CheckSite module.

reserved for another temporary facility assigned

earlier on site.
If any (X,Y) of facility K Occupied(X,Y)
Then CheckOverlap = False
Else CheckOverlap = True.
(b) If no overlap occurs, i.e., CheckOverlap =
True, then the space is reserved for the temporary facility.
For b (X,Y) of facility, Occupied(X,Y) =
Facility(X,Y).
4.2. Use of genetic algorithms in site layout
optimization
Usually there are two main components of genetic algorithms that are problem dependant, i.e.,

the string coding and the objective function evaluation [12]. String coding refers to the process of
translating any solution into a unique string (similar
to the biological chromosome) prior to commencing
the genetic algorithm. The objective function evaluation is the process of deciphering the string back
to its problem-equivalent value and then checking
on the extent to which the problem objective is
achieved.
4.2.1. GA string coding
Most CAD platforms use rectilinear coordinate
systems in referencing entities, as depicted in
Section 4.1. The string encoding primarily depends on this rectilinear referencing of entities to
achieve the transformation from the graphical

Fig. 6. Functionality of the CheckOverlap module.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

757

N solutions, where n is a subset of N. The following

procedure outlines this enhancement:
1. Generate random initial pool of solutions
N.
2. Select best n solutions from initial pool
as first population.
3. Start GA.

Fig. 7. CAD-based GA string encoding.

representation, i.e., (X,Y) coordinates, to the chromosome structure. The location of each facility is
referenced via its bottom left corner as shown in
Fig. 7.
Due to the specific nature of the optimization
problem at hand, a special-purpose GA optimization
engine has been particularly developed for this study
and integrated in the automated system. Therefore, the
automated system acts in a stand-alone fashion, as it
does not require the use of any commercial GA
software. A flowchart of the used GA optimization
process is detailed in Fig. 8.
4.2.2. Initialization of population
Any GA starts with an initial population of solutions. The number of initial solutions generated
influences the success of a GA in reaching its goal.
It is known that increasing the population size has the
following effects on the GA:
(1) Tremendously increases the time required for
generating a new population.
(2) Causes a very slow convergence rate.
(3) Causes the GA to reach more optimum solutions.
In order to assist the GA in its blind search, a slight
enhancement has been incorporated in the optimization process. Instead of working with a very large
population throughout the GA, the initial population is
selected as the best n solutions from an initial pool of

Thus, the GA benefits from the presence of a

large initial population that assists its random search
without paying the large computational penalty
posed by dealing with a large population at each
generation.
4.2.3. GA generations
The generation process adopted in this study is a
steady-state generation. Traditional GAs move from
generation (i) to generation (i + 1) via the generation
of a new population. Steady-state generation moves
from one generation to the other via the introduction
of new offspring to replace the worst solutions in the
population. In case the new offspring are not better
than the worst solutions, they are disregarded and
other offspring are chosen instead. Previous studies
[6] have successfully utilized steady-state generation
in the site layout optimization problem.
Elimination of the worst offspring means that as new
solutions are introduced, the population as a whole
would improve. It also means a chance for randomly
bred offspring to even outperform the best solution in
the population. Generation of new offspring involves
the three traditional genetic operators:
i. Replication: Traditional roulette wheel selection
is performed based on the fitness value for
individual solutions. Utilizing a steady-state population replication, two offspring are chosen to
replace the worst two solutions in the population.
ii. Crossover: The same roulette wheel selection
procedure is applied to select the parents that will
be crossed. Simple single-point crossover is used
so as to minimize the disruption of the schemata.
After the crossing, two checks are performed on
the offspring:
1. Constraint satisfaction: To verify the feasibility of the new solutions.

758

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

Fig. 8. GA optimization process flowchart.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

759

Fig. 9. Mutation operator flowchart.

2. Objective function improvement: To verify

that the new solutions are not worse that those
being replaced.
iii. Mutation: Mutation is used mainly to break the
stagnation in improvement by introducing new

genetic information into the population. It was

noticed during the testing of the automated site
layout planning system that performing the GA
without mutation led to near optimum solutions.
These near optimum solutions required slight

Fig. 10. Convergence of population parameters at different levels of mutation.

760

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

refinements to reach optima. These refinements

involved very small movements of one or more
temporary facilities in a specific direction. A
modified mutation operator is developed to attain
this function (Fig. 9). The mutation operator
randomly performs the following steps:
1. Chooses the temporary facility to be moved.
2. Chooses the movement heading (towards X- or
Y-axis).
3. Chooses the movement direction ( + ve or  ve
direction).
4. Applies a movement of one unit to the
temporary facility chosen in step 1 and in the
direction chosen in steps 2 and 3. Then, the
new solution is checked for constraint satisfaction and objective function improvement. If
violated, the mutation procedure is repeated.
4.2.4. Convergence condition
Generally, the GA generation process continues
until a convergence condition is reached. In the GA

of this study, convergence is reached when there is little

variation (D) within the population itself or, in other
words, there is a small difference between the maximum and minimum population values. Convergence
occurs when the following condition is satisfied:
D < Convergence Value
D

Max  Min
Max

4
5

Where Min: minimum solution in current population,

Max: maximum solution in current population, Convergence Value: user specified tolerance (usually 5
10%).
During experimentation with the GA, it was found
that the modified mutation operator assisted the GA to
reach more optimal solutions. In turn, its use caused
slower convergence. Fig. 10 shows how the maximum
and minimum population valuesat different levels
of mutationvary till the convergence condition is
reached. Although the modified mutation operator
causes a slower convergence, the optimal solution

Fig. 11. Arrangement of permanent facilities and obstacles on site.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

Table 3
Temporary facilities and their dimensions
Temporary facility

Dimension
(m  m)

Storage area
2 Administrative caravans
2 Engineers caravans
Parking area
Toilets
Steel fabrication yard
Steel storage
Generators
Water tanks

42  20
10  3
10  3
12  4
10  3
10  18
10  12
22
Variable

reached in the case of pmutation = 5% was better than

the case when no mutation was used.

5. Case study
The selected case study is a major swimming pool
complex with all auxiliary buildings. The site, whose
area totals 24,000 m2, is part of the new Heliopolois
Sports Club located 30 km east of Cairo, Egypt. The

761

complex is comprised of six main permanent structures: (1) three main swimming pools (an Olympic,
diving and recreational pools), (2) a childrens pool, (3)
changing rooms and (4) a hardscape area. An underground basement occupies a large portion of the
Olympic and diving pools.
Before the start of the project, an access road was
set by the contractor and considered as an obstacle for
the temporary facility assignment. Arrangement of the
permanent facilities and obstacles within the site
boundary is shown in Fig. 11. These permanent
facilities and obstacles occupy nearly 15,000 m2 of
the site area, leaving 9000 m2 for assigning all needed
temporary facilities.
The temporary facilities needed to sustain the
construction operations and the dimensions of these
facilities are shown in Table 3. Water tanks were
neglected in the layout planning, as the contractor
assigns a tank inside the premises of each permanent
facility for water supply.
Based on input from the project manager, the interfacility proximity weight matrix was developed (Fig.
12). The values for the proximity weights follow the

Fig. 12. Proximity weight matrix for the case study.

762

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

Fig. 13. Existing layout of temporary facilities.

Fig. 14. Automated system assignment of temporary facilities.

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

six-value scale previously presented in Table 2. This

matrix was then transformed into its quantitative equivalents before being input to the automated system.
Using a convergence condition of (D < 5%) to
terminate the GA optimization process, the automated
system completed the temporary facility assignment in
53 min running on a Pentium-3 800 MHz processor.
The genetic parameters of Pmutation = 0.05, Pcrossover =
0.7 and population size = 250 were used in the process
and the convergence condition was reached after 2660
generations. The optimal solution had a representative
score of 36,669, in contrast to a representative score of
38,647 for the actual site layout planned by the project
management team.
The layout generated by the automated system
indicates that the system managed to properly recognize all spatial constraints imposed by the site geometrical characteristics and assigned all temporary
facilities accordingly. No major discrepancies were
found between the existing layout (Fig. 13) and that
produced by the automated system (Fig. 14), though
the automated system produced a more optimal solution. A close comparison of the two layouts shows
that the storage area and the steel fabrication/storage
yards were assigned in their exact positions. The fact
that extensive efforts were made by the project management team in preparing the site layout indicates
that the automated system could serve as a valuable
tool in site layout planning.
Although the system-generated layout scored
higher than the actual layout produced through accumulated planning experience, this does not necessarily
mean that the systems layout is superior in all
aspects. When the system-generated layout was
shown to the project manager, he indicated that some
temporary facilities, such as the administrative caravans, were placed in more favorable positions than
those in the actual layout. On the other hand, he
considered some other temporary facilities, such as
the engineers caravans and parking area, to be placed
in unfavorable positions in comparison to the actual
layout. This was due to some secondary objectives,
such as minimizing local congestions, as the tightly
packed assignment interferes with resource handling
and material flow. This confirms that the automated
system does not necessarily yield the best layout.
Minor adjustments may be required to the systemgenerated layout in order to attain other secondary

763

objectives not taken into consideration in the optimization process.

6. Summary and conclusions

The paper presents an automated hybrid system for
layout planning of construction sites. The automated
system benefits from the optimization capabilities of
GAs in performing the task of temporary facility
assignment. A special-purpose GA, that uses a
steady-state generation with a modified mutation
operator, was developed in order to suit the site layout
planning problem. An important aspect worth of
consideration in GA systems is the accuracy of
solution reached through the optimization process.
The GA itself does not assure optimum solutions,
but may yield near optimum solutions. The solution
reached is very sensitive to the GA parameters used
such as population size, pmutation and pcrossover.
Due to the evident graphical nature of the problem,
the GA was integrated with a widely known CAD
platform, i.e., AutoCADk. The CAD environment is
utilized in space detection of the site layout and in the
satisfaction of geometrical constraints dictated by the
facility assignment problem. The fact that geometrical
constraints are modeled using CAD and not through
traditional mathematical formulations adds to the
flexibility of constraint representation.
The optimization goal taken into consideration in
the current study is to minimize the total transportation costs (or its proximity weight equivalents) between facilities. This is not necessarily the only goal
the site layout should strive to achieve. For example,
the layout should decrease site congestions and provide for safe working environment. Unfortunately,
some of these goals are very difficult to gauge and
the formulation of a comprehensive mathematical
model that takes into consideration all these goals
would be a very complicated task. However, some of
these goals can be dealt with in a different manner.
Supplementary modules that utilize artificial intelligence (AI) techniques, particularly knowledge-based
systems (KBS), could be developed in the future to
account for the less tangible aspects of site congestion
precautions and safety regulations.
In conclusion, the use of GA in site layout planning
can provide the planner with a good initial layout to

764

H.M. Osman et al. / Automation in Construction 12 (2003) 749764

start with and modify according to the other secondary

objectives. By minimizing the objective function the
GA accomplishes the complex task of assigning
temporary facilities in positions consistent with their
respective proximity needs, and at the same time
abiding to spatial constraints imposed by the site
geometry. The fact that the presented GA optimization
engine is incorporated with a CAD-based input/output
media greatly increases its practicality for use. Possible future expansion of the automated system to
account for aspects such as safety regulations would
substantially improve its capabilities in solving the
site layout planning problem.

[5]
[6]

[7]

[8]

[9]

[10]

References
[1] I.-C. Yeh, Construction-site layout using annealed neural network, J. Comput. Civ. Eng., ASCE 9 (1995) 201 208.
[2] I.D. Tommelien, R.E. Levit, B. Hayes-Roth, SitePlan model
for site layout, J. Constr. Eng. Manag., ASCE 118 (1992)
749 766.
[3] I.D. Tommelien, P.P. Zouein, Interactive dynamic layout planning, J. Constr. Eng. Manag., ASCE 119 (1993) 266 287.
[4] M.Y. Cheng, J.T. OConnor, ArcSite: enhanced GIS for con-

[11]

[12]
[13]

[14]

struction site layout, J. Constr. Eng. Manag., ASCE 122

(1996) 329 336.
H. Li, P.E. Love, Site level facilities using genetic algorithms,
J. Comput. Civ. Eng., ASCE 12 (1998) 227 231.
T. Hegazy, E. Elbeltagi, EvoSite: evolution based model for
site layout planning, J. Comput. Civ. Eng., ASCE 13 (1999)
198 206.
P.P. Zouein, I.D. Tommelien, Dynamic layout planning using
a hybrid incremental solution method, J. Constr. Eng. Manag.,
ASCE 125 (1999) 400 408.
C.M. Tam, T.K. Tong, W.K. Chan, Genetic algorithm for optimizing supply locations around tower crane, J. Constr. Eng.
Manag., ASCE 127 (2001) 315 321.
S.-O. Cheung, T.K.-L. Tong, C.-M. Tam, Site pre-cast yard
layout arrangement through genetic algorithms, Autom.
Constr. 11 (2002) 35 46.
J.J. Mahoney, C.B. Tatum, Construction site applications of
CAD, J. Constr. Eng. Manag., ASCE 120 (1994) 617 631.
W.T. Chan, D.K. Chaua, G. Kannan, Construction resource
scheduling with genetic algorithms, J. Constr. Eng. Manag.,
ASCE 122 (1996) 125 132.
D. Whitley, A genetic algorithm tutorial, Tech. Rep. CS-93103, Colorado State University, Fort Collins, CO, 1993.
H. Al-Tabtabi, A. Alex, An evolutionary approach to the capital budgeting of construction projects, Cost Eng., AACE 40
(1998) 28 34.
R.G. Askin, C.R. Standrige, Modeling and Analysis of Manufacturing Systems, Wiley, New York, 1993.