You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/355782294

Construction facilities location selection for urban linear infrastructure


maintenance projects using uniform cost search method

Article  in  Soft Computing · October 2021


DOI: 10.1007/s00500-021-06408-7(

CITATIONS READS

0 47

3 authors:

Amr G. Mansour Mohamed S. Eid


Arab Academy for Science, Technology & Maritime Transport Arab Academy for Science, Technology & Maritime Transport
3 PUBLICATIONS   2 CITATIONS    40 PUBLICATIONS   241 CITATIONS   

SEE PROFILE SEE PROFILE

Emad Elbeltagi
Mansoura University
101 PUBLICATIONS   2,749 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Innovation in the Construction Industry View project

Construction Joint Ventures - Profit and Risk Sharing View project

All content following this page was uploaded by Amr G. Mansour on 30 October 2021.

The user has requested enhancement of the downloaded file.


Soft Computing
https://doi.org/10.1007/s00500-021-06408-7 (0123456789().,-volV)(0123456789().
,- volV)

OPTIMIZATION

Construction facilities location selection for urban linear infrastructure


maintenance projects using uniform cost search method
Amr G. Mansour1 • Mohamed S. Eid1 • Emad E. Elbeltagi2

Accepted: 10 October 2021


 The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021

Abstract
Linear infrastructure projects in congested urban cities require optimal site layout planning. Nevertheless, most of the
existing models are developed to address building construction projects with confined boundaries. This paper presents a
model for assigning construction facilities locations (CFL) to each work segment in linear infrastructure projects that are
characterized by frequent changes of construction site location as the work progresses. The model accounts for the resource
transportation cost, land renting cost, relocation cost of CFLs, as well as duration needed for each work segment. The
proposed model utilizes a tree search algorithm—Uniform Cost Search (UCS)—that guarantees a global optimal solution.
In addition, a pruning technique is utilized to decrease the tree size to reduce the needed computation time. Finally, to
account for the stochastic factors (i.e., transportation cost and duration needed per segment), the model is integrated into a
Monte Carlo simulation. The proposed model, with the pruning algorithm, is able to solve a 4 trillion-solution space
problem in under 3000 ms using an 8 GB Ram, 2 GHz machine. The proposed model can be integrated into the planning
and site layout decision making processes for linear infrastructure projects to reduce the overall site layout cost and
maintain adequate site conditions while reducing the interruptions in road utilization. Finally, the presented pruning
algorithm can be utilized in similar construction engineering problems to efficiently identify global optimal solutions.

Keywords Linear infrastructure projects  Tree search  Uniform cost search  Tree pruning

1 Introduction offices, mess rooms and maintenance rooms (Abotaleb


et al. 2016; Elbeltagi et al. 2001; El-Rayes and Khalafallah
Construction site layout requires adequate use of available 2005). In urban construction projects, site space is con-
limited site space to efficiently place and relocate tempo- sidered as critical as labor, money and material or any other
rary facilities (TFs) on site while respecting the work limited resources. This is due to the lack of free areas
interrelationships (Farmakis and Chassiakos 2018). TFs outside the construction site boundaries especially in con-
include storage areas for material, parking lots for equip- gested cities (Elbeltagi et al. 2001). Moreover, construction
ment, access roads, batch plants, fabrication shops, site site layout efficiency affects labors and equipment pro-
ductivities, project time and cost and site safety (Elbeltagi
et al. 2001; El-Rayes and Said 2009). Various studies have
& Mohamed S. Eid been carried out on construction site layout planning
meid@aast.edu
(CSLP) for building projects, but to the authors knowledge
Amr G. Mansour there is no significant models developed for CSLP for
amrgamal@aast.edu
linear infrastructure projects.
Emad E. Elbeltagi Urban cities linear infrastructure projects (i.e., road
eelbelta@mans.edu.eg
maintenance, pipe installation, etc.) require frequent
1
Construction and Building Engineering, Arab Academy for changes of site location as the work proceed (Eid et al.
Science, Technology, and Maritime – Sheraton Heliopolis, 2019; Moselhi and Hassanein 2003). For example, if the
Cairo 11799, Egypt work is divided into segments, the construction site neither
2
Department of Structural Engineering, Mansoura University, has an exact location nor boundaries, but dynamically
Mansoura 35516, Egypt

123
A. G. Mansour et al.

move throughout the operations (Mansour et al. 2019; 3.1 Evaluating CSLP
Moselhi and Hassanein 2003). This makes the site layout
planning process different for linear infrastructure projects Based on the review of the existing body of knowledge,
in contrast to traditional building construction projects. most of the previous CSLP models have evaluated the
Infrastructure construction and maintenance work for CSLP based on two main factors: in-site resources trans-
urban crowded and congested cities do not have the luxury portation cost and TFs’ relocation cost (El-Rayes and Said
of free spaces surrounding the roads. Thus, a proper model 2009; Cheung et al. 2002; El-Rayes and Khalafallah 2005;
for selecting the locations of TFs and accounting for the RazaviAlavi and AbouRizk 2017; El-Rayes and Said 2009;
dynamic nature of the project is required. Such model Abotaleb et al. 2016; Farmakis and Chassiakos 2018;
should take into consideration the scarce available spaces, Mansour et al. 2019). Resources transportation cost is
transportation of crews, and the associated transportation, crucial in construction projects as almost all construction
relocation and renting cost (Mansour et al. 2019). The activities involve movement of equipment, labors and
aforementioned factors are stochastic in nature as their materials. Therefore, in-site resources transportation affects
values may vary depending on the road segments’ location the work flow, productivity of resources, project cost and
and traffic condition. Thus, the developed model needs to time (Cheung et al. 2002; El-Rayes and Khalafallah 2005;
account for such uncertainties. El-Rayes and Said 2009; Abotaleb et al. 2016; Mansour
et al. 2019). The resource transportation cost is calculated
between all construction facilities on site based on (1)
2 Goal and objectives traveled distance between facilities a and b (Da,b) and (2)
travel cost rate per unit distance between facilities a and
The goal of this research is to develop an optimization b (Ca,b) (El-Rayes and Khalafallah 2005).
model for the selection of the construction facilities loca- Da,b can be measured by three different methods: (1)
tions (CFL) for linear infrastructure projects in congested Cartesian distance, (2) shortest walk distance and (3)
urban cities. The model accounts for the planned progress rectangular distance (Cheung et al. 2002; Abotaleb et al.
of work throughout project segments. In addition, the 2016). The decision of which method to be used is based on
model finds a global optimal solution rather than relying on the actual movement of resources on site. Cartesian dis-
near optimal approaches. Finally, the model considers the tance is the simplest distance measurement technique and
impact of the stochastic factors of linear projects on the the most commonly used. It measures the shortest straight
optimal selection of each segment CFL. line distance between two facilities (Abotaleb et al. 2016).
The shortest walk distance is the minimum distance
between two facilities. The shortest walk is a more realistic
3 Literature review approach as it takes in consideration a specific path in order
to avoid collisions with obstacles (Abotaleb et al. 2016).
CSLP includes two main steps: (1) facility assignment and Rectangular distance is a summation of the horizontal and
(2) layout planning. Assigning facilities on site has two vertical movements of resources between facilities (xa-
types, facility-to-location assignment and facility-to-site - xb) ? (ya - yb) where xa, yb are the coordinates of the
assignment (Osman et al. 2003; Farmakis and Chassiakos center of facility a and xa, ya are those of facility b. This
2018). In the former type, the TFs are assigned to a pre- method is used when the physical sizes of the facilities
determined location. The number of these predetermined prevent the diagonal movement of resource between loca-
locations is greater than or equal the number of facilities tions (Cheung et al. 2002; Yahya and Saka 2014).
and the size of the largest facility can fit in the smallest The cost of relocating TFs throughout the project
location (Cheung et al. 2002; Osman et al. 2003). If the duration is another important factor in evaluating the fit-
number of the locations exceeds the number of TFs, a ness of the CSLP. TFs are divided into two categories
dummy facility with zero effect on the objective function is according to their mobility state. Stationary facilities (SFs)
added to facilitate the computational process (Cheung et al. locations are determined from the very beginning of the
2002; Farmakis and Chassiakos 2018). In facility-to-site project and do not change throughout the project stages.
assignment, any unoccupied space on site can be a poten- Tower cranes and batch plants are examples of such SFs.
tial location for the TFs (Osman et al. 2003). In the latter The relocation of such SFs is often infeasible due to the
approach the way of representing the facilities (facilities high relocation cost and significant time needed (El-Rayes
shapes) plays an important role in determining the opti- and Said 2009). On the other hand, movable facilities
mum location of the facilities, which makes it a more (MFs) can be relocated as work progresses or moving to a
dynamic approach (Abotaleb et al. 2016). new stage such as construction equipment parking

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

locations, materials storage areas and fabrication shops. the allocation of TFs on site and increases the site
A MF can be relocated when a new free space or location efficiency.
appeared on site, during the progress of work, which is The aforementioned approaches do not guarantee find-
more efficient from the facility’s original location. This ing an optimal solution (Farmakis and Chassiakos 2018;
change in location requires a relocation cost represented in Abotaleb et al. 2016). Meanwhile, tree search algorithms
demobilization of the facility from its old location, trans- guarantee optimality through searching through the prob-
porting to new location and reinstalling the facility in the lem’s solution space. Tree search algorithms solve a
new location (El-Rayes and Said 2009). problem by finding a sequence of actions that leads to an
optimal solution (Russell and Norvig 2010; Ibaraki 1975).
3.2 Site layout planning and optimization The problem is first coded into a tree of nodes (N) and
edges (e). Each N represents a state of the solution, while
CSLP is an NP-complete problem (Kaveh et al. 2016). each e depicts an action or transition from one state to the
Several techniques and approaches were utilized to assign next. All the possible states start from an initial point called
TFs on site, ranges from meta-heuristics and mathematical root node (No). Expanding No or any of the others states Ni
models to knowledge-based systems (Abotaleb et al. 2016; will lead to new child nodes depending on the branching
Farmakis and Chassiakos 2018; Patil and Joshi 2013). factor e (Russell and Norvig 2010; Felner 2011).
Meta-heuristics and mathematical models require an Breadth-first search, Depth-first search and uniform cost
identified goal (objective function) to be optimized under search (UCS) are the three main tree search algorithms
certain constraints to produce an optimum site layout. (Russell and Norvig 2010; Ibaraki 1975). The difference
Knowledge-based systems guide engineers to plan site among all three algorithms is the priority or strategy to
layout without getting involved into pure optimization expand the child node (Felner 2011). Breadth-first search
problems (Osman et al. 2003). algorithm expands the child nodes of the same level first
Meta-heuristics algorithms have been widely used by then it moves to the next level (Rahim et al. 2019). In each
researchers in solving CSLP problems as it shows high level, Breadth-first search gives priority of expansion to the
computational powers and produces near optimum solu- oldest created child node (first in first out queue, i.e., FIFO)
tions in an acceptable time (Farmakis and Chassiakos (Russell and Norvig 2010). Accordingly, the expansion
2018; Yahya and Saka 2014). Zouein et al. (2002) pre- continues using FIFO queue until a goal state is reached.
sented a model that utilized genetics algorithms (GAs) in Depth-first search algorithm gives priority to expand the
solving site layout problem with unequal-size and con- newest created child node (last in first out queue, i.e.,
strained facilities. Elbeltagi et al. (2004) using GAs intro- LIFO) (Russell and Norvig 2010). In this algorithm, the
duced a model that maximizes safety and increases expansion starts with the deepest child node of a certain
productivity in construction sites. El-Rayes and Khalafal- branch. The expansion continues in this branch until no
lah (2005) implemented GAs to decrease resources trans- more nodes are left for expansion in the branch. Then, it
portation cost along with maximizing site safety. Yahya moves to the second deepest node in another branch. Thus,
and Saka (2014) proposed a model that uses a multi-ob- the expansion backs up until a goal state is reached (Russell
jective artificial bee colony (MOABC) via Levy flights and Norvig 2010).
algorithm to determine the optimum site layout through The UCS method is the optimal algorithm when the cost
decreasing total handling cost between TFs and increasing of actions (e) are different. In contrast to the breadth-first
safety in site. RazaviAlavi and AbouRizk (2017) intro- search method, UCS does not expand the oldest created
duced an integration between GAs and simulation that child node (shallowest node). Instead, it expands the least
minimizes site layout construction cost. Farmakis and cost child node (Felner 2011). This is done by sorting the
Chassiakos (2018) developed a multi-objective dynamic frontier nodes from lowest to highest cost. The search
model to find global optimum site layout, (optimum site continues in that manner until a goal state is reached. Once
layout for all project phases), considering resources trans- a goal node is reached, then the path connecting this goal
portation cost between TFs and work places, construction node to the root node is selected as an optimal path (Russell
and relocation cost of TFs and safety of site. and Norvig 2010).
As for knowledge-based systems, Tommelein et al. The UCS method had been used as an optimization
(1992) presented an expert system (Sightplan) that com- technique to solve problems in different engineering fields
bines both construction managers experience in placing other than construction. Sun, et al. (2017) introduced a
TFs on site and the high computational powers of computer selection algorithm called active antenna group (AAG) that
in one model. Cheng and O’Connor (1996) introduced a involves the use of UCS algorithm as a base foundation for
knowledge-based system, called ArcSite, which automates the proposed model. AAG optimizes the selection of the
active antenna at each level of search in order to improve

123
A. G. Mansour et al.

the symbol error rate. Kuffner (2004) modified the UCS infrastructure projects through literature review; (2)
algorithm in order to increase the efficiency of optimal developing a mathematical model defining the objective
robot navigation planning on Euclidean-cost grid and function, constraints and variables of the proposed prob-
lattices. lem; (3) utilizing a well-established optimization technique
A comparison between the UCS and GAs, the most for selecting the optimum CFL for each road segment; (4)
commonly used meta-heuristic optimization approach, is running Monte Carlo simulations to capture the stochastic
presented in Table 1. The comparison shows the difference nature of the problem’s factors; and (5) verifying the
between the two algorithms in time and space complexities capabilities of the developed model on hypothetical and
and also the optimality of the provided solution. In GAs, real case studies.
the time taken by the algorithm to solve a problem of size
n is in a function of the population size and can be linear (O
(n)), up to O (n5) (Nopiah et al. 2010). As for solution 5 Model overview and assumptions
optimality, GAs provides a near optimal solution (Nopiah
et al. 2010; Farmakis and Chassiakos 2018; Yahya and In urban cities, constructing linear infrastructure or main-
Saka 2014; Abotaleb et al. 2016)). On the other hand, the taining existing ones is usually carried out during the low
problem size in UCS algorithm is determined by the traffic volume hours. This allows for the usage of the
function O (bd), where b is the branches of the tree and d is completed part of work during normal traffic volume hours.
the tree depth (Russell and Norvig 2010; Ibaraki 1975). In As such, equipment and any used material must be parked
contrast to GAs, UCS algorithm guarantees optimal solu- in a location other than the current project’s site. Unlike
tion (Russell and Norvig 2010; Ibaraki 1975). Accordingly, highways and rural areas, there is no adequate right of way
UCS provides a significantly better solution by guaran- in urban roads to be used as storage or parking areas, and
teeing the solution optimality, but at the expense of thus a remote location has to be determined. Such location
increasing the overall computational time. can be a parking lot, unused land, etc.
The proposed model presents such problem through
3.3 Knowledge gap dividing the linear construction or maintenance project into
segments ‘‘j’’ and assigning one of the available CFL ‘‘i’’
Previous models mentioned in the literature review address for each road segment j, as illustrated in Fig. 1. The length
building construction projects that has clear site boundaries of the segments ‘‘j’’ is determined based on the practi-
or certain construction area. The goal of these previous tioners need and the area of the CFL ‘‘i’’ must not be less
studies was to determine the optimal configuration for the than a predetermined area decided by the practitioners. All
location of TFs within a construction site area or bound- equipment, materials, caravans, or any other TFs that are
aries. However, the presented study is developed to solve used to serve the construction activities held in road seg-
site layout problems related to urban infrastructure projects ment j will be stored in the assigned CFL i. Accordingly, an
that are characterized by: (1) linearly distributed site optimization model must be developed in order to assign
locations along a huge length and (2) frequent mobilization the optimum CFL i to each construction segment j (Man-
of the TFs. As such, the goal of this study is to determine sour et al. 2019).
the optimal locations of the TFs considering the afore- Through developing the proposed model, the authors
mentioned special characteristics of these construction assumed that the construction work is held through the
sites. project segments in a consecutive sequence. For example,
construction work in segment j = 2 will start after finishing
construction works of segment j = 1. Thus, no overlapping
4 Methodology between segments allowed. Also, the authors assumed that
only one CFL can be assigned to a work segment. Finally,
To achieve the research objectives and find the optimum the authors focused on MF, and large SF were not
CFL for urban linear infrastructure projects, the authors addressed in the proposed model.
established a five-step research methodology: (1) identi-
fying the factors affecting the CFL selection for linear

Table 1 Comparison between


Genetic algorithms (GAs) UCS
UCS and GAs
Time and space complexity Up to O (n5) O (bd)
Optimality Provides near optimal solution Guarantee optimal solution

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

After work is done in segment j and progresses to seg-


ment j ? 1, the relocation of the temporary facilities might
be needed to decrease the overall cost. At this point the
relocation cost (Rlci,i*) becomes of great importance and
must be calculated. For the new segment j ? 1 there is a
choice whether to use the same CFL i used by the previous
segment j or use another CFL i*. Rlci,i* includes cost of
disassembling, transporting and reassembling of temporary
facilities from the current CFL i to the newly selected CFL
i*. Thus, the total relocation cost can be calculated using
Eq. (5).
Fig. 1 Dividing road into segments
X
I X
J
Tc ¼ xi;j Tcij  dj 8i; 8j ð3Þ
5.1 Model development i¼1 j¼1

Throughout the construction operation at segment j, it is


I X
X J
Rc ¼ xi;j Rci  dj 8i; 8j ð4Þ
required to assign a CFL i for storing materials, parking of i¼1 j¼1
equipment, etc. As such, the proposed model utilizes a
binary variable for the allocation of CFL i to the con- X
I X
I
Rlc ¼ Rlci;i xi;j  xi ;jþ1 8i; 8j; 8i 6¼ i ð5Þ
struction road segment j, as shown in Eq. (1). For example, i¼1 i
‘‘x1,2’’ means assigning CFL i = 1 to road segment j = 2.
 To this end, the model can evaluate a feasible solution
1 if CFLi is assigned to segement j for assigning CFLs to the different segments through
xi;j ¼ ð1Þ
0 otherwise Eq. (6). It should be noted that the objective function uti-
For every segment j there is a number of available CFL i lized in this model does not prioritize one cost over the
ready to be assigned. For instance, in Fig. 1 there are four others. This is due to lack of justification to carry out such
available CFLs (i = 1, 2, 3 and 4) that can be assigned to assumption. Nevertheless, the model can be modified to
segment j = 1. This gives four possible assignments for the provide weights if a practitioner sees it is of value to the
first segment. However, as previously discussed as an evolving project.
assumption, each segment j can use one and only one CFL. MinðTc þ Rc þ RlcÞ ð6Þ
As such the summation of xi,j over the same segment
should be equal to 1 as shown in Eq. (2).
X
I
6 Optimal location of TFs
xi;j ¼ 18j ð2Þ
i
6.1 Uniform cost search
where I is the maximum number of CFL (i).
Each assignment of CFL i to segment j affects three UCS is a tree search algorithm that expands nodes on the
different costs: (1) resource transportation cost from CFL least cost path till an end node is reached. The algorithm,
i to segment j (Tci,j); (2) CFL i renting cost (Rci); and (3) Fig. 2, starts by expanding the root node (No) and adding
the relocation cost of temporary facilities from current CFL all child nodes to unexpanded node frontier. The algorithm
i, to another i*. then selects the least cost child node from the frontier to be
Tci,j accounts for the daily transportation cost of crews, further expanded. This recursive approach is carried out till
equipment and materials from CFL i to segment j. Tci,j is a solution (end node) is reached. UCS guarantees a global
measured using shortest walk distance technique. This optimal solution as it searches all possible solutions of the
technique measures the actual moving distance of resour- problem (Russell and Norvig 2010; Felner 2011). UCS can
ces from CFL i to road segment j. The total Tc for a fea- adequately solve the presented model. At each segment, all
sible solution can be calculated via Eq. (3), which is the the possible assignments of CFLs can be explored as child
product of the variable xi,j by the daily transportation cost nodes. Thus, as the work progresses to the following seg-
(Tcij) and the duration of work in segment j (dj). The ments, the tree will be constructed, and optimal path can be
renting cost of CFL i (Rci) may differ from one CFL to the determined.
other depending on its location and the market renting
prices. The total Rc for a feasible solution is calculated in
the same manner as Tc through Eq. (4).

123
A. G. Mansour et al.

Fig. 2 UCS algorithm pseudo


Function Uniform Cost-Search returns a solution, or null
code
Frontier: a list containing nodes ready to be expanded
Explored: a list with expanded nodes
Solutions: a list of all possible solutions

Initialize the Frontier using the initial node to expand


Add child nodes to Frontier
Remove initial node from frontier and add it to Explored
Loop do
if Empty(Frontier)? then track all expanded nodes (in Explored) at level d (maximum
depth), get the corresponding least Path-Cost value, and nodes’ parents to provide solutions.
Add each solution in Solutions
else
Choose the lowest Path-Cost node in the frontier to expand
Add resulted child nodes to frontier
Remove the expanded node from frontier and added to explored
Return minimum cost in Solutions

6.2 Illustrative example Table 2 Transportation cost/day


j=1 j=2
($)
To present the optimization approach for the proposed i=1 200 300
problem, an illustrative hypothetical example, shown in i=2 400 150
Fig. 3, is utilized. Suppose a 400-m road needs to be
maintained. After scanning the surroundings of the road,
Table 3 Total relocating cost
two CFLs (i = 1, 2) were found available to be used with i* = 1 i* = 2
($)
renting cost (Rci) of $300/day and $400/day, respectively.
i=1 0 500
The road is divided into two equal segments (j = 1, 2). The
i=2 600 0
transportation cost of the round trips per day from each
CFL(i) to each segment (j) and total relocation cost from
old CFL i to new CFL i* are shown in Tables 2 and 3, the least cost node. At Fig. 4b, the unexpanded frontier
respectively. The duration of the construction work (dj) in includes N2, N3 and N4. Accordingly, N2 will be expanded
the road segments j = 1 and j = 2 are 5 and 7 days, as it is the least cost node. At Fig. 4c, Nodes N3, N4, N5 and
respectively. N6 are in the unexpanded frontier, and N3 will be expanded
Figure 4 shows a tree structure of applying UCS method as it is the least cost node. Figure 4d presents the solution
to optimize CFL selection of the road. In Fig. 4a, the root as N3 is an end node. The solution would be N1 and N3,
node (No) was expanded leaving the child nodes (N1 and representing x1,1 and x1,2, respectively.
N2) in the unexpanded frontier. N1 is thus expanded as it is It is noticeable that UCS has a time complexity of O(ij)
where i is the branching factors and j is the depth of the
tree. The presented example has ij solution space which is
four solutions. For a real-life problem with multiple
available CFLs and numerous segments, UCS will require
a significant long computation time to find an optimal
solution. This would hinder the developed model to capture
the stochastic parameters via Monte Carlo simulation as it
will take impractical long computation time. The authors
attempted to overcome this by pruning the tree while
maintaining the optimality of UCS approach.

6.3 Tree pruning

Pruning a search tree is an important step to optimize the


computational efficiency and to avoid unnecessary time
and space complexities. Tree pruning techniques decrease
the tree size by removing redundant branches or sections of
Fig. 3 Hypothetical example

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

Fig. 4 UCS tree of hypothetical


example

the search tree. Such redundant branches should not lead to result in the same successors and the same solution but
an optimum solution, thus transforming a complete tree with different cost. For example, N4 and N6 are two iden-
into a smaller pruned one. Though tree pruning is not a tical nodes representing allocation of CFL 2 to road seg-
newly developed concept and various techniques are ment 2. Expanding N4 will result in a least cost path of N10,
available (Patel and Upadhyay 2012; Song and Ying 2015), and N22, representing a solution of x2,2 = 1, x2,3 = 1 and
the authors customized the generic tree pruning approach x2,4 = 1, respectively. The same solution but with different
to fit the problem at hand. Accordingly, this concept of tree cost is resulted from expanding N6 represented by nodes
pruning can be applied to the UCS tree to enhance the N6, N14 and N30. Thus, expanding the lower cost node of
model’s efficiency and decrease its processing time. the two identical ones (N4 and N6) will result in the lower
Consider two nodes N and N’, each of them represents cost solution. N6 has a lower cost than its identical node N4.
the same solution state xi,j (i.e., assigning the same CFL i to As such, N6 will be expanded and N4 will be excluded from
the same segment j). However, each node has a different the tree. Note that the cost of the path from N4 to N22 (i.e.,
parent node (i.e., xi,j-1 and x:i,j-1). Thus, though both $47–$22 = $25) is the same as N6 to N30 (i.e., $43–
parents are from the same segment j - 1, they are using $18 = $25). However, the cost from the No to N4 ($22) is
different CFLs. The cost from the root node (No) to N and higher than No to N6 ($18).
N’ are P(N) and P(N’), respectively. This process can be repeated at other identical nodes.
Since UCS always expands the least cost child node For example, N8, N10, N12 and N14. Accordingly, the pro-
when facing multiple children, UCS will always choose the posed tree pruning approach would significantly decrease
same successor for the same state xi,j, regardless of the the tree size as shown in Fig. 6. The redundant nodes (N4,
node. For example, since nodes N and N’ represent the N5, N8 and N13) and their branches that will not lead to an
same state, they will provide the same optimal path to its optimum solution were removed from the tree. This trim-
end node passing through the same child nodes. The cost of ming decreased the solution space from O (ij) 24 = 16
such path (without the cost P) is donated C(N) and C(N’). solutions to O (i2) 22 = 4 solutions. Therefore, at every
Since both paths from N and N’ to the end nodes are the iteration, there are at most i2 nodes to choose from.
same, C(N) should be equal to C(N’). To this end, the tree
can be pruned at node N’, if P(N’) C P(N), as the overall
cost of the solution from this node to the end node would 7 Model automation
be at best equal to the solution from node N.
Figure 5 presents an example of a complete tree of two The proposed model was coded in Java using NetBeans
CFLs and four segments, i.e., 16 solutions (24). According IDE 8.1 platform. The coded model requires the number of
to UCS method, expanding any identical nodes, which segments, available CFLs as well as the work duration for
represent the same allocation of CFL i to segment j, will each segment j, Tci,j between each CFL i and each segment

123
A. G. Mansour et al.

Fig. 5 Complete tree before pruning

Fig. 6 Pruned tree

j, Rci, of each CFL i, and all possible Rlc i,i* between the 8 Case study
different CFLs. The model automatically generates the
search tree and produces ij possible solutions. The model A real case study is used to present the model’s ability in
searches through all possible solutions using the UCS solving real linear infrastructure projects in congested
method and prune the tree as previously discussed. The urban cities. The proposed case study utilizes Salah Salem
outcome of the search is the least path cost representing the road, sketched in Fig. 7, which is one of the most crowded
optimum allocation of CFLs to different work segments. and vital roads in Cairo, Egypt. Salah Salem is an arterial
The authors used an 8 GB Ram, 2.00 GHz machine to road extending from Cairo International Airport to the east
verify the model performs as intended. The proposed bank of the Nile. It cuts through several neighborhoods and
model was able to solve the previous discussed illustrative important locations in Cairo including Heliopolis district,
hypothetical problems and resulted in finding the same the Citadel, and Old Cairo. Salah Salem road serves a high
optimum solution in under 1000 ms. daily traffic volume of private and public transportation
vehicles.
Road maintenance work in Salah Salem road is a perfect
demonstration for the effectiveness of the developed
model. Any construction or maintenance activity will block
the road due to the equipment parked and material storage.

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

On running the model without the proposed pruning


algorithm, the model solved a 5 million solution space
problem (11 segments and 4 CFLs only) in 8 days, which is
an impractical computational time. This is because the
model had to go through all 5 million possible solutions in
order to determine the optimum one. Therefore, the UCS
tree representing the proposed case study (seven-km in
Salah Salem road) must be pruned to decrease the com-
putational time.

8.1 Testing optimality pre and post tree pruning

In order to test the completeness of the model while


applying the pruning algorithm, a comparison of the
model’s results before and after pruning is presented in
Fig. 7 Salah Salem road—case study Table 8. The table also presents the computation time. The
authors tested the model in incremental steps to better
As such the proposed model would be of great help in present the pruning time reduction ability and its impact on
determining the optimum CFL for each road segment as optimality. Table 8 presents three cases with 4 CFLs
work progresses. In addition, the model helps in decreasing serving 7, 10 and 11 construction segments, respectively.
the congestion of traffic made by the construction works This increased the solution size from 16,000 to 5,000,000
through avoiding lane rental strategies (Herbsman and possible CFL assignments. As per Table 8, it can be
Glagola 1998) that would decrease the traffic flow. noticed that without pruning the tree, UCS solves the
To demonstrate the model’s capabilities, a seven-km problem in a significant long time, 8 days. The proposed
section in the most crowded part of Salah Salem road is pruning algorithm, with O(i2), drastically decreased the
used. The road is divided into 14 segments where each computation time to less than 1000 ms. More importantly,
segment is 500 m, as shown in Fig. 7. It should be noted the proposed pruning algorithm did not impact the optimal
that using a 500-m length segment is only an assumption solution of the UCS, as it provided the same total cost and
and practitioners can adjust it based on their needs. CFL assignments per construction segment.
Changing the segments’ length will only impact the num- After proving the significantly higher computational
ber of segments j which will affect the solution space and power of the model using the proposed pruning algorithm,
processing time. After scanning the surroundings of the the model was tested on the whole 14 segments with 8
road, it turned out that there are eight eligible CFLs CFLs serving these segments. Figure 8 shows the model’s
available to be used. Table 4 shows the area of each CFL output regarding to the total cost and the optimal assign-
and the daily Rci. The transportation cost of the round trips ment of CFLs to each of the 14 segments in the 7 km of
per day from each CFL (i) to each segment (j), total relo- Salah Salem road. The model succeeded in solving the
cation cost from old CFL i to new CFL i*, and the duration problem of 4 trillion (814) solution space in under 3000 ms.
of the construction work (dj) in each of the road segments This indicates the ability of the model in solving large
are shown in Tables 5, 6 and 7, respectively. solution space problems in significant short time.

Table 4 CFL areas and daily renting cost (Rci) 8.2 Monte Carlo simulation for stochastic factors
CFL i Area (m2) Rci ($/day)
As previously discussed, all the mentioned factors that
CFL 1 440 1100
affect the model (i.e., resources transportation cost, land
CFL 2 360 900 renting cost, duration of construction work and facilities
CFL 3 500 1250 relocation cost) are stochastic in their nature. This is due to
CFL 4 560 1400 the uncertainties affecting those factors. For example, the
CFL 5 600 1500 duration of construction work depends on the weather,
CFL 6 630 1575 efficiency of the equipment and labors. Also, the resources
CFL 7 700 1750 transportation costs varies depending on the traffic condi-
CFL 8 800 2000 tions. Land renting cost may differ depending on the
changes in the market prices. However, not all factors have
high diversity. Therefore, the authors selected the most

123
A. G. Mansour et al.

Table 5 Transportation cost (Tci,i) ($/day)


j=1 j=2 j=3 j=4 j=5 j=6 j=7 j=8 j=9 j = 10 j = 11 j = 12 j = 13 j = 14

i=1 39 189 324 459 594 756 891 1080 1242 1404 1593 1728 1836 1890
i=2 351 486 621 756 891 1026 1215 1296 1431 1566 1701 1836 1971 2106
i=3 122 38 216 351 459 594 729 918 1080 1242 1404 1539 1647 1728
i=4 243 54 81 216 351 486 621 770 891 1053 1215 1323 1431 1512
i=5 324 189 378 243 351 811 676 703 946 1351 1216 1757 1622 1486
i=6 324 189 54 122 257 392 527 662 797 932 1068 1203 1338 1473
i=7 892 757 622 486 351 216 54 189 324 459 595 730 865 1000
i=8 1216 1081 946 811 676 541 405 270 135 270 405 541 676 811

Table 6 Total relocating cost


i* = 1 i* = 2 i* = 3 i* = 4 i* = 5 i* = 6 i* = 7 i* = 8
(Rlci,i*) ($)
i=1 – 110 75 150 175 185 525 650
i=2 243 – 189 135 153 300 500 600
i=3 68 176 – 68 107 110 447 572
i=4 135 190 54 – 35 40 370 495
i=5 180 153 107 35 – 200 385 450
i=6 192 300 110 40 220 – 340 500
i=7 530 500 450 375 390 360 – 160
i=8 650 600 572 500 450 500 160 –

Table 7 Construction work duration (dj) per segment remained deterministic as they did not have significant data
Segments j Duration (days) dj Segments j Duration (days) dj
variation.
To address the stochastic nature of the problem, a Monte
j=1 5 j=8 8 Carlo simulation is integrated with the tree search model.
j=2 7 j=9 7 Monte Carlo simulation is defined as an algorithm that runs
j=3 10 j = 10 6 the model several times with variable input values (i.e.,
j=4 6 j = 11 8 transportation cost or work duration) to capture the
j=5 5 j = 12 7 stochastic nature of the process (Avlijas 2019; Inyim and
j=6 6 j = 13 6 Zhu 2014). The value of each input is chosen randomly for
j=7 7 j = 14 7 each trial or iteration from a pre-defined probability dis-
tribution of the variable (Tokdemir et al. 2019; Avlijas
2019). Therefore, the outcome of the proposed model
would be the most likely assignment of CFLs for each
diverse factors based on their variance (resources trans- construction segment that achieves the optimal value for
portation cost and duration of construction work) to be the the objective function in Eq. (6).
stochastic factors of the model while the other factors

Table 8 Comparison between before trimming and after trimming processing time
Solution space Optimum solution Optimum total cost ($) Processing time
Before trimming After trimming

16,000 x1,1; x1,2; x1,3; x1,4; x1,5; x3,6; x3,7 71,774 6s Under 1000 ms
1,000,000 x1,1; x1,2; x1,3; x1,4; x1,5; x3,6; x3,7; x3,8; x4,9; x4,10 119,927 7.5 h Under 1000 ms
5,000,000 x1,1; x1,2; x1,3; x1,4; x1,5; x3,6; x3,7; x3,8; x4,9; x4,10 x4,11 140,847 8 days Under 1000 ms

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

it CFL 7 for segments 7–14. This provided a total cost of


$186,815.

9 Discussion

Through this paper, the authors presented a model that


guarantees optimal CFL assignment for each segment in an
urban linear infrastructure project through a tree search
approach. The problem domain is trivial in case of con-
struction projects in the suburbs or rural areas where ade-
quate shoulder lanes and open fields that can be used to
store equipment and other temporary facilities. Neverthe-
less, for densely populated urban cities, such problem may
hinder the success of the linear project and may impact the
Fig. 8 Case study model’s solution
proper utilization of the existing infrastructure. To this end,
the authors argue that the decision makers should integrate
The authors utilized the aforementioned case study to
such approach in planning the construction and mainte-
present the integration of Monte Carlo simulation into the
nance of linear infrastructure projects in urban cities.
tree search model. First, a probability distribution was
Decision makers should account for multiple factors
assigned for each probabilistic factor of the four mentioned
when utilizing the proposed model. First, the number and
factors (Tci,j; Rci,j; Rlci,i*; and dj) based on log-normal
size of segments impact the model output. The smaller the
distribution (Avlijas 2019; Wall 1997; Yang 2005). The
segment sizes (thus, larger number of segments), the more
probability distributions were based on real data and esti-
accurate the optimal CFL assignment would be. This would
mates based on the most likely scenarios. The authors then
require significantly longer time for data entry and simu-
selected the most diverse factors based on their variance
lation runs. Also, accurate data collection for representing
while the other factors remained deterministic as they did
the stochastic factors highly impacts the accuracy of the
not have significant data variation. The proposed search
total cost and CFL assignments.
model with the pruning algorithm was then utilized to run
More importantly, the proposed tree pruning algorithm
multiple simulation based on the stochastic inputs of the
provides a significant contribution to the body of knowl-
various factors.
edge. The pruning algorithm converts the searchable tree
To produce statistically accurate and reliable results, the
size from O(ij) to O(i2). As such, the presented approach
minimum number of required simulation runs to achieve a
can be utilized to find global optimal solution in reasonable
95% degree of confidence and maximum error band of
time without the need for meta-heuristic near optimal
± $100 was calculated using Eq. (7) (Hahn 1972). Thus,
techniques. Nevertheless, modelers should notice that the
the minimum statistically significant sample size in the
proposed problem is 20,000 simulation run.
" #
Zð1þcÞ  r 2
2
nmin ¼ ð7Þ
E

where nmin is the minimum sample size, c is the degree of


confidence, E is the maximum allowable error and r is an
initial estimated value for standard deviation.
To this end, the integration of the proposed tree search
approach into a Monte Carlo simulation generated a pow-
erful tool capable of finding the global optimum solution
for stochastic problems. The model was able to find the
solution with a 95% degree of confidence for the afore-
mentioned case study in under 60,000 ms. According to the
simulation results given in Fig. 9, the optimum solution for
the problem is to assign CFL 1 to segments 1–5, then
transport the TFs to CFL 3 for segment 6, and finally move Fig. 9 Optimal CFL assignment for case study via Monte Carlo
simulation

123
A. G. Mansour et al.

tree pruning algorithm does not impact the optimality of resources transportation routes or paths as direct routes
UCS due the properties of the problem domain. As this is only and (4) the model does not reflect its impact to traffic
an assignment problem with multiple similar states reached congestion and the society. To address these limitations,
through various paths, the pruning algorithm can be uti- the model should be expanded in the future work to address
lized with no impact on the UCS optimality. This might not working multiple segments at the same time, considering a
be true for all problems. Other applications for the UCS tradeoff between maximizing site safety and minimization
with the proposed pruning algorithm should be investi- of total cost, using agent based modeling for resources
gated, for example minimum travel distance between two transportation routes and, finally, introducing a new per-
nodes in a network graph, or optimal resource allocation in formance metric to measure and quantify the model’s
a scheduling problem. impact on decreasing traffic interruptions and user’s cost.

Funding Not applicable.


10 Conclusion and future work
Availability of data and material All the available data that support
A CFL selection model was developed to assist planners the findings of this study are available from the corresponding author
and decision makers in placing temporary facilities on site upon reasonable request.
for urban linear infrastructure projects in congested cities.
Code availability The developed code is available from the corre-
The developed model seeks global optimum CFL for each sponding authors upon reasonable request.
road segment throughout the work operations. The model
was also designed to consider and capture the stochastic
factors and uncertainties to practically comply with the Declarations
nature of linear infrastructure projects. The model used
Conflict of interest The authors do not have any competing interest in
UCS method as an optimization technique and utilized the presented results.
Monte Carlo simulation to produce a global optimum
solution with a 95% degree of confidence. The CFL
assignment is an exponential problem (ij). Therefore, a tree References
pruning algorithm was utilized to convert the problem into
a polynomial problem (i2). Thus, the presented tree pruning Abotaleb I, Nassar K, Hosny O (2016) Layout optimization of
construction site facilities with dynamic freeform geometric
algorithm significantly enhanced the model computational
representations. Autom Constr 66:15–28
time, with no impact on the UCS optimality, allowing it to Avlijas G (2019) Examining the value of Monte Carlo simulation for
solve larger solution space problems. As such, the proposed project time management. J Sustain Bus Manag Solut Emerg
model has the potential to impact the success of the Econ 24(1):11–21
Cheng MY, O’Connor JT (1996) ArcSite: enhanced GIS for
infrastructure projects in urban cities through applying it at
construction site layout. J Constr Eng Manag 122:329–336
the planning phase. Cheung S-O, Tong TK-L, Tam C-M (2002) Site pre-cast yard layout
Optimal CFL per segment was evaluated through min- arrangement through genetic algorithms. Autom Constr
imizing resources transportation cost, CFL renting cost and 11:35–46
Eid M, Elbeltagi E, El-Adaway IH (2019) Multi-objective simulta-
facilities’ relocation cost. A real case study was carried out
neous optimization for linear projects scheduling. In: The 2019
to present the model’s ability in solving real linear infras- ASCE international conference on computing in civil engineer-
tructure projects in congested urban cities. Thus, the ing, Atlanta, Georgia, pp 561–568
model’s results illustrated its powerful capabilities in Elbeltagi E, Hegazy T, Hosny AH, Eldosouky A (2001) Schedule-
dependent evolution of site layout planning. Constr Manag Econ
finding the global optimal solution of a large solution space
19:689–697
problem (4 trillion possible solutions) in under 3000 ms Elbeltagi E, Hegazy T, Eldosouky A (2004) Dynamic layout of
using an 8 GB Ram, 2.00 GHz machine. The developed construction temporary facilities considering safety. J Constr
model capabilities have a significant potential in planning Eng Manag 130:534–541
El-Rayes K, Khalafallah A (2005) Trade-off between safety and cost
the construction and maintenance of linear infrastructure
in planning construction site layouts. J Constr Eng Manag
projects in urban cities. In addition, the presented UCS 131:1186–1195
algorithm along with the tree pruning algorithm could be of El-Rayes K, Said H (2009) Dynamic site layout planning using
great help in other engineering problems. However, this approximate dynamic programming. J Comput Civ Eng
23:119–127
needs further investigations in future work along with other
Farmakis PM, Chassiakos AP (2018) Genetic algorithm optimization
limitations in the current model. These limitations include for dynamic construction site layout planning. Organ Technol
(1) preventing the allocation of more than one CFLs to Manag Constr Int J 10:1655–1664
different road segments at the same time, (2) considering Felner A (2011) Position paper: Dijkstra’s algorithm versus uniform
cost search or a case against Dijkstra’s algorithm. In:
only minimization of total cost, (3) approximating

123
Construction facilities location selection for urban linear infrastructure maintenance projects using…

Proceedings, the fourth international symposium on combinato- Patil AD, Joshi DA (2013) A review paper on construction site layout
rial search (SoCS-2011), pp 47–51 planning. Int J Innov Eng Technol 3:4
Hahn G (1972) Sample sizes for Monte Carlo simulation. IEEE Trans Rahim R, Dijaya R, Multazam MT, Daengs A. Sudrajat D (2019)
Syst Man Cybern 2(5):678–680 Puzzle game solving with breadth first search algorithm. In: 4th
Herbsman ZJ, Glagola CR (1998) Lane rental-innovative way to annual applied science and engineering conference
reduce road conrtuction time. J Constr Eng Manag RazaviAlavi S, AbouRizk S (2017) Site layout and construction plan
124(5):411–417 optimization using an integrated genetic algorithm simulation
Ibaraki T (1975) Theoretical comparisons of search strategies in framework. J Comput Civ Eng 31:04017011
branch-and-bound algorithms. Int J Comput Inf Sci Russell J, Norvig P (2010) Artificial intelligence: a modern approach,
5(4):315–344 3rd edn. Pearson Education Inc, Upper Saddle River, p 07458
Inyim P, Zhu Y (2014) Application of Monte Carlo simulation and Song YY, Ying LU (2015) Decision tree methods: applications for
optimization to multi-objective analysis of sustainable building classification and prediction. Shanghai Arch Psychiatry
designs. In: The 2014 international conference on computing in 27(2):130–135
civil and building engineering, Orlando, Florida, pp 2009–2016 Sun Y, Wang J, He L, Song J (2017) Iterative uniform-cost search of
Kaveh A, Khanzadi M, Alipour M, Moghadd MR (2016) Construc- active antenna group selection for generalised spatial modula-
tion site layout planning using two new meta-heuristic algo- tion. In: Proceedings of the IEEE international conference on
rithms. Iran J Sci Technol Trans Civ Eng 40(4):263–275 communications (ICC), pp 1–6, May 2017
Kuffner J (2004) Efficient optimal search of uniform-cost grids and Tokdemir OB, Erol H, Dikmen I (2019) Delay risk assessment of
lattices. In: Proceedings of 2004 IEEE/RSJ international con- repetitive construction projects using line-of-balance scheduling
ference on intelligent robots and systems, Sendai, Japan, and monte carlo simulation. J Constr Eng Manag 145(2):1–12
pp 1946–1951 Tommelein ID, Levitt RE, Hayes Roth B (1992) SightPlan model for
Mansour A, Eid M, Elbeltagi E (2019) Optimal construction facilities site layout. J Constr Eng Manag 118:749–766
location selection for linear infrastructure projects. In: The 2019 Wall D (1997) Distributions and correlations in Monte Carlo
ASCE international conference on computing in civil engineer- simulation. Constr Manag Econ 15:241–258
ing, Atlanta, Georgia, pp 482–490 Yahya M, Saka MP (2014) Construction site layout planning using
Moselhi O, Hassanein A (2003) Optimized scheduling of linear multi-objective artificial bee colony algorithm with Levy flights.
projects. J Constr Eng Manag 129(6):664–673 Autom Constr 38:14–29
Nopiah ZM, Khairir MI, Abdullah S, Baharin MN, Arifin A (2010) Yang IT (2005) Simulation-based estimation for correlated cost
Time complexity analysis of the genetic algorithm clustering elements. Int J Project Manag 23:275–282
method. In: Proceedings of the 9th WSEAS international Zouein PP, Harmanani H, Hajar A (2002) Genetic algorithm for
conference on signal processing, robotics and automation, solving site layout problem with unequal-size and constrained
ISPRA 10, pp 171–176 facilities. J Comput Civ Eng 16:143–151
Osman HM, Georgy ME, Ibrahim ME (2003) A hybrid CAD-based
construction site layout planning system using genetic algo- Publisher’s Note Springer Nature remains neutral with regard to
rithms. Autom Constr 12:749–764 jurisdictional claims in published maps and institutional affiliations.
Patel N, Upadhyay S (2012) Study of various decision tree pruning
methods with their empirical comparison in WEKA. Int J
Comput Appl 60:20–25

123

View publication stats

You might also like