CHAPTER 2

LITERATURE REVIEW

In order to complete a construction project at maximum efficiency of time

and cost, judicious scheduling and allocation of available resources are required
(Clough et al. 2000). However, the widely utilized network scheduling techniques
such as Critical Path Method (CPM) and Precedence Diagram Method (PDM)
assume the unlimited supply of resources, and thereby they often produce
unintended high resource demands and undesirable resource fluctuations. Harris
(1978) indicated that there are several negative impacts of undesirable resource
fluctuation on project performance; as (1) it causes hiring and releasing of workers
in short-term basis which is troublesome, inefficient, and costly; (2) clerical costs
and unemployment compensation cost occur due to hiring and releasing of labors;
(3) it needs time to readjust to the working conditions of a particular job when
hiring new labors; (4) there is a possibility that once released workers might find
another job and not be available upon recall; (5) high-idle cost occurs when
employing high-cost resource units; and (6) it produces disruption in the learning
curve effects, and thereby decreases productivity (Stevens 1990). As such,
undesirable resource fluctuation affects on overall project performance including
productivity and cost. Moreover, high resource peak demand causes a congestion
problem on site with the activities that wait for the available resources.
Accordingly, it causes poor productivity, increased cost, and in the end, project
delays (Hendrickson 1989). In order to minimize those negative impacts on project
performance and resource utilization efficiency, there is a pressing need for
developing robust models that are capable of eliminating undesirable resource
fluctuation and peak demand.
1
 Multiple shifts are often used in construction projects to accelerate schedules
and meet project deadlines. This utilization of two or three 8-hour shifts per day is
reported to provide a number of advantages, including (1) enabling the number of
weekly work-hours to be almost doubled or tripled; (2) benefiting from the fact
that the premium cost of an evening or night shift is often lower than that of
overtime hours; and (3) minimizing overtime productivity losses that are often
encountered during overtime hours due to workers fatigue and site congestion
(Hanna et al. 2008). Despite these advantages, the utilization of multiple
construction shifts still suffers from a number of limitations including its negative
impacts on construction cost, productivity, and safety. The utilization of multiple
shifts often leads to additional costs that are required for shift premiums, nighttime
lighting, quality control, and safety measures. In addition, nighttime construction
disrupts the circadian rhythm of shift workers who often struggle with adapting to
new sleep cycles. As a result, the utilization of evening and/or night shifts is
reported to cause productivity losses due to workers fatigue, health disorders,
social life disruption, lower morale, and higher accident rates (Hanna et al. 2008;
Kogi 1985). For example, Folkard and Tucker (2003) found that the risk of injuries
and accidents on evening shifts is higher than day shifts, and they also found that
night shifts have the highest risk of injuries and accidents among the three shifts.
Staffing evening and/or night shifts is another problem that confronts construction
managers as most workers prefer to work on day shifts (Czeisler et al. 1982;
Guerin and DelPo 2007). Moreover, recent studies found that the utilization of
evening and night shifts causes higher rates of labor turnover and absenteeism
which leads to project delays and cost overruns (Gomar et al. 2002; Hanna et al.
2008; Knauth 2001). In order to minimize these negative impacts of utilizing
multiple shifts, the number of labor hours during the evening and/or night shifts
should be minimized whenever possible.
2
 Additionally, the utilization of multiple shifts needs to comply with labor
availability constraints. Construction managers and planners need additional labor
for the evening and/or night shifts in addition to those working in day shifts
because labor unions often restrict laborers to work no more than one shift per day.
In many projects, this required additional skilled construction labor is often not
available. The recent study by Construction User Round Table (CURT) reported
that 82% of responding companies experienced shortage of skilled workers on their
projects, and 78% of them indicated that the trend has worsened over the past years
(Srour et al. 2006). In order to comply with these labor availability constraints,
construction planners need to distribute and utilize the limited number of
construction labor among multiple shifts in the most efficient and effective manner
in order to maximize project performance.
Multiple shifts can also resolve resource conflicts among Network
scheduling techniques often produce impractical and costly schedules that require
significant fluctuations in the daily resource demands on site, as shown in Figure
2.1 (Harris 1978). These resource fluctuations cause additional construction costs,
including (1) idle costs for construction resources that cannot be released during
their non-productive time and need to be kept idle on site until they are needed at a
later time (Akpan 2000), as shown in alternative 1 of Figure 2.1; (2) release and
rehiring costs for construction resources that can be temporarily released during
low demand periods and rehired later when needed (Gomar et al. 2002), as shown
in alternative 2 of Figure 2.1; and (3) mobilization costs such as mobilization and
demobilization costs for construction equipment and clerical and training costs for
construction workers to gradually mobilize/hire all the required resources until the
peak demand level is reached (Karshenas and Haber 1990; Srour et al. 2006). In
order to minimize the aforementioned costs of resource fluctuations (i.e., idle,
release and rehiring, and/or mobilization costs), the level of resource fluctuations
3
can be reduced by shifting the early start date of selected construction activities, as
shown in Figure 2.2. This figure illustrates four alternative schedules and their
resource fluctuation costs for a simple example. The first alternative (see Figure
2.2(B)) represents the early project schedule for a resource that cannot be released
and rehired. This alternative causes 13 idle resource days, requires a peak demand
of 7 resources, and leads to a project duration of 8 days. The total resource
fluctuation costs (RFC) of this schedule is $2,650, which includes (a) $1,950 of
idle costs to retain non-productive construction resources on site during low
demand periods until they are needed later on in the project; (b) no release and
rehire costs; and (c) $700 of mobilization costs that are needed to gradually
increase the level of resources to satisfy the maximum demand (i.e. MaxR=7) for
the project and then gradually release them towards the end of project. The second
alternative represents the early project schedule for a resource that can be released
and rehired. The project duration of this is 8 days and its resource fluctuation cost
(RFC) is $1,600 which includes release and rehiring costs of $900 for 4 resources,
and mobilization costs of $700 to hire a peak demand of 7 resources. The third
alternative represents the impact of shifting the early start of activity B within its
float by 3 days (see Figure 2.2(D)), which causes 4 idle resource days, requires a
peak demand of 7 resources, and leads to a project duration of 8 days. The resource
fluctuation cost (RFC) of the third alternative ($1,300) is less than the first two
alternatives because it reduced the resource fluctuation in the schedule. In order to
achieve the most cost effective and efficient resource utilization on construction
projects, the impact of resource fluctuation and peak demand on resource idle
costs, release and rehiring costs, and mobilization costs should be directly
measured and minimized. The example in Figure 2.2(E) also illustrates that the
resource fluctuation and peak demands can often be minimized by allowing an
extension in the project duration. This fourth alternative shows the impact of
4
shifting the early start of activity B by 4 days, which requires zero idle resource
days, zero released and rehired resources, a peak demand of 6 resources, and
project duration of 9 days, as shown in Figure 2.2(E). Extending the project
duration by 1 day in this alternative enabled it to generate the least resource
fluctuation cost (RFC=$600) among the four alternatives because it produced the
minimum peak resource demand and mobilization costs while eliminating resource
idle costs and release and rehiring costs. As such, construction planners need to
analyze and optimize potential tradeoffs between minimizing resource fluctuation
costs and minimizing the project duration in order to identify and implement an
optimal project schedule and resource utilization plan for construction projects.
Critical path method (CPM) has been widely used for project planning and
controlling in construction industry. This method has the capabilities of showing
the precedence relationships among project activities, identifying project critical
paths, providing activity float times, and generating optimized resource allocation
(Okmen and Oztas 2008). However, this method has long been criticized by many
researchers due to its assumption that the durations of project activities are
deterministic and it is not changed by various uncertainties such as weather,
productivity, site conditions, and labor skill during construction period (Ang et al.
1975; Faniran et al. 1998; Jaafari 1984; Mo et al. 2008; Nasir et al. 2003; Okmen
and Oztas 2008). In real projects, these uncertainty factors always exist and can
affect on the duration of construction activities. Accordingly, this limitation of
critical path method (CPM) may lead to provide a construction planner with
incorrect critical paths and fail to complete the project by a target time (Jaafari
1984). Therefore, reliable forecast of project completion is essential for project
success, because it can provide the greater percentage of recovery in the early stage
of project and provide alternative decisions for cost escalation involving time/cost
tradeoffs (Ahuja and Nandakumar 1985). There is a pressing need for developing
5
an advanced project risk assessment model that is capable of forecasting the
project completion time and providing the reliable project performance.
The primary purpose of resource leveling is to reduce peak requirements and
smooth out period-to-period fluctuations in resource assignment without changing
project time (Hegazy 1999b). A number of resource leveling models and
algorithms have been developed to reduce the level of fluctuations in resource
utilization and their negative impact on construction productivity and cost. These
models introduced and utilized a number of metrics to reduce resource fluctuations
including: (1) sum of squares method; (2) absolute difference between resource
consumption in consecutive time periods; (3) deviation between actual resource
usage and a desirable or a uniform resource usage; and (4) sum of squares of
resource changes. First, a number of research studies have utilized sum of square
method to calculate total sum of squares of daily resource demands in order to
reduce the level of resource fluctuation (Ahuja 1976; Bandelloni et al. 1994;
Burgess and Killebrew 1962; Harris 1978; Hegazy 1999b; Son and Skibniewski
1999) The objective of resource allocation is to resolve resource conflicts among
project activities for same labors and equipments by rescheduling activities while
keeping the unavoidable extension of the project duration to a minimum. Many
research studies have developed optimization model for a resource allocation
problem using various techniques including (1) heuristic methods (Ahuja 1976;
Bell and Han 1991; Boctor 1990; Sampson and Weiss 1993); (2) integer
programming (Talbot 1982); (3) linear programming (Mingozzi et al. 1998); (4)
dynamic programming (Gavish and Pirkul 1991); (5) branch and bound algorithm
(Brucker et al. 1998); (6) simulated annealing (Lee and Kim 1996); (7) genetic
algorithms (Chan et al. 1996; Chua et al. 1996; Hegazy 1999b; Kim and Ellis
2008; Lee and Kim 1996; Leu and Yang 1999; Senouci and Eldin 2004); and (8)
particle swarm optimization (Zhang et al. 2006b). Early attempts to solve resource
6
allocation problems using mathematical models such as linear programming,
integer programming, and dynamic programming often suffer from large scale
problems required in practical application due to a phenomenon called
“combinatorial explosion” (Chan et al. 1996). Heuristic methods based on priority-
rule are capable of overcoming the problem of combinatorial explosion. However,
these methods cannot guarantee optimal solution, although it provides good
feasible solutions. Therefore, a number of research studies have developed the
models using meta-heuristic search methods such as simulated annealing (SA) and
tabu search (TS), and naturally inspired algorithms such as genetic algorithms
(GA) and particle swarm optimization (PSO).
Lee and Kim (1996) have applied SA, TS, and GA for resource allocation
problem and compared their performance with each other. A new encoding scheme
which denotes priority of each activity was developed and used in those
algorithms. The analysis results showed that those algorithms outperform existing
heuristic methods such as the minimum slack method in searching for optimal
solution, and SA provides the best performance among three optimization
algorithms.
Chan et al. (1996) have presented GA-scheduler that searches for optimal
schedules that provide minimum project duration under the resource constraints
using genetic algorithm (GA). The objective function was designed to minimize
the deviation between the required resources and the available resources while
minimizing the extension of project duration by penalizing scheduling beyond
target date. This enables simultaneous optimization of resource leveling and
allocation as well as “due-date” problems. The application results showed that GA-
scheduler outperforms heuristic methods, and the computational time does not
grow exponentially for a larger problem.

7
 Chua et al. (1996) have developed co-evolving resource scheduling model
using genetic algorithm (GA) that is capable of generating optimal schedule that
minimizes project duration with respect to the resource availability profile that is
configured interactively with the schedule for minimum resource idling. The
resource contacts model was designed to subdivide the entire project into separate
resource contracts and control the resource hiring levels for each contract by GA.
The model is capable of simultaneously searching for optimum schedules and the
required resource profile that minimize project duration while providing minimum
resource idling. The case study illustrated that the model also could provide
schedules and corresponding resource available profiles that yield considerable
savings idle resources even in multi-resource problems.
Leu and Yang (1999b) introduced new GA crossover and mutation
operators, UX3 and UM3 that overcome the limitation of traditional GA operators
for sequencing problems. GA-based resource constrained construction scheduling
(GARCS) model that incorporates those two new GA operators was developed to
generate optimal schedules that minimize project time under the resource
availability constraints. The model was extended to consider time-cost tradeoff as
well as resource constrained scheduling. In order to search for optimal time-cost
tradeoff under the resource availability constraints, the model was developed as
multi-objective GARCS that consists of six subsystems: activity duration
generation subsystem, activity order generation subsystem, project cost calculation
subsystem, project duration calculation subsystem, evaluation subsystem, and
selection subsystem. Application examples illustrated that the model has a
capability of searching for non-dominated solutions for the multi-objective
resource-constrained scheduling problems after a number of trials. Zhang et al.
(2006b) have developed the model using a particle swarm optimization (PSO) to
exploit both preemption and break for the resource constrained construction project
8
with the objective of minimizing project time. The model allows the break to take
place in off-working time and allows the interrupted activities not to resume soon
in the next working period, and accordingly, multiple activities can share the
constrained resources while the activities are interrupted. The example analysis
demonstrated that the preemptive scheduling under break and resource-constraints
(PSBRC) has a capability of minimizing project duration and illustrated that the
PSO-based methodology for the PSRBC is effective.
Kim and Ellis (2008) presented a permutation based elitist genetic algorithm
to optimize a large-scale resource constrained project scheduling. The algorithm
employed the elitist strategy to preserve the best individual solution for the next
generation in order to obtain an improved solution. New random number generator
was developed to examine precedence feasible individuals. The model employed a
serial schedule generation scheme proposed by Kelly (1963) for the permutation-
based decoding. In order to demonstrate the use of present model and its capability
of searching for optimal solutions in a large-scale project network, the project
networks with 30, 60, and 120 activities were analyzed by the present model. The
results of analysis illustrated that the present model can explore and exploit several
suboptimal solutions, which may include the optimal solution, and the proposed
algorithm is useful to solve the multiple resource-constrained project scheduling
problems. Time-Cost optimization (TCO) problem has been extensively examined
by a number of research studies. Various approaches have been proposed for
optimizing construction time and cost including (1) heuristic methods (Fondahl
1961; Moselhi 1993; Prager 1963; Siemens 1971); (2) mathematical programming
(Liu et al. 1995; Moussourakis and Haksever 2004); and (3) meta-heuristic
methods (El-Rayes and Kandil 2005; Feng et al. 1997; Hegazy 1999a; Hegazy and
Ersahin 2001; Jaskowski and Sobotka 2006; Leu and Yang 1999; Li et al. 1999; Li
and Love 1997; Senouci and Eldin 2004; Xiong and Kuang 2008; Zhang et al.
9
2006b; Zheng et al. 2004, 2005). Mathematical programming such as linear
programming is suitable for problems with linear time-cost relationships, but they
often fail to solve the problem with discrete time-cost relationships (Feng et al.
1997). Moreover, it requires a lot of computational efforts to solve a large scale
project network. Heuristic methods are able to overcome such limitation of a large
scale problem, but fail to guarantee optimal solutions. Therefore, many research
studies have focused on utilizing meta-heuristic methods in time-cost tradeoff
analysis to overcome the limitation of heuristic methods and mathematical
programming.
Liu et al. (1995) have developed optimization model using a hybrid method
that integrates linear and integer programming. Linear programming was used to
find lower bounds of the solutions, and then integer programming was used to
obtain the exact solution. In linear programming model, the time-cost curve of
each activity was identified by piecewise linear approximation. All the dominated
options were eliminated when establishing the convex hull. This convex hull was
then used as the set of the constraints (lower bounds) within each activity that
contributes to the linear model of the time-cost tradeoff problems. The integer
programming was then used to minimize total project cost with the constraints of
activity precedence and the selection of a single resource utilization option for each
activity. The hybrid model was developed using Microsoft Excel to provide a
construction planner with an efficient means of analyzing time-cost trade-off
decisions.
Feng et al. (1997) have presented the model using genetic algorithm (GA) to
search for optimal time-cost tradeoff. Pareto front approach was used to search for
non-dominated solutions that simultaneously minimize project time and cost.
These non-dominated solutions were founded by plotting each solution in the
population and calculating the distance of each point in each generation from each
10
segment of the convex hull of the generation. The calculated distance is then used
to select the solutions in the population to reproduce the next generation solutions.
The optimal time-cost curve can be founded in the final population. An example
with 18 activities was analyzed by the model and it showed that the model was
able to generate 95% of the optimal points on the time-cost curve.
Hegazy (1999a) have implemented GA in the commercial project
management software, MS-project, using VBA language. The model was
developed to search for the least cost combination of construction methods for the
various tasks, considering deadline duration, late completion liquidated damages,
early completion incentive, and daily indirect cost. Moreover, the model combines
time-cost tradeoff with resource management procedure to schedule construction
projects under the resource availability constraints. This resource constrained
scheduling was performed by the feature originally included in the commercial
project management software. The results of case study illustrated that the model
could produce better solution than heuristic method. This development tool
provides a practical tool which can be used in practice.
Li et al. (1999) have presented machine learning and genetic algorithms
based system (MLGAS) that integrates a machine learning method with GA to
overcome the limitations of existing GA based systems for time-cost tradeoff
problems. The research study indicated that there are two limitations on existing
GA models: (1) the existing models requires the user to manually craft the time-
cost curves for formulating the objective functions; and (2) the systems only deal
with linear time-cost relationships. The Quadratic Time-Cost Curves Discovery
System (QTCD) was developed using machine learning method to automatically
generate the quadratic time-cost curves from historical data and measure the
credibility of each quadratic time-cost curve. The generated time-cost curves are
then used to formulate the objective function, and it was optimized by the GA. The
11
GA was improved by using multiple point crossover and variable mutation
probability. This prevents the GA operations from being trapped into local optima.
Case study showed that MLGAS outperforms the human expert in solving
nonlinear time-cost tradeoff problems and it could produce the better solutions than
the solutions produced by heuristic methods.
Zheng et al. (2004) have presented multi-objective optimization approach
that simultaneously optimize project time and cost. Modified adaptive weight
approach (MAWA) was developed to replace traditional fixed or random weights.
The objective functions for time and cost were integrated into a single objective for
simulation. This approach overcomes the limitation of single objective time-cost
tradeoff and enables GA to have greater searching pressure to resist the inherent
convergence power of GA. The adaptive weights have practical meaning in
representing the relative importance of each criterion for the project. These weights
enable GA to search a wider range against the objectives that have a relatively
small search space in previous generations. This approach can guarantee the
diversity of exploration and efficiency of exploitation of historical information.
Zheng et al. (2005) have developed GA-based multi-objective optimization
model that simultaneously minimizes time and cost. In order to overcome the
problem of genetic drift, the model utilized Pareto ranking, niche formation, and
adaptive mutation rate. The genetic drift occurs when GA converge to a single
peak due to the stochastic errors during processing. The model incorporated
modified adaptive weight approach (MAWA) to exert a search pressure in GA.
Pareto ranking mechanism overcomes the limitation of traditional proportional
selections. Niche formulation is useful for stabilizing multiple subpopulations that
arise along the Pareto-optimal front, and thereby it maintains diversity of
population. A computer prototype was developed as a macro in Microsoft Project
to evaluate the feasibility of solutions generated by GA. The medium sized project
12
was selected and fitted into the prototype to test the performance of model. The
results confirm that the developed model is reliable to generate optimum solutions.
El-Rayes and Kandil (2005a) have presented the multi-objective
optimization model for time-cost-quality tradeoff analysis. Multi-objective genetic
algorithm was used in the model to generate the optimal solutions for those
conflicting objectives. The decision variables were feasible resource utilization
options that include construction method, crew formation, and crew overtime
policy. The construction methods include the utilization of different type of
materials that have different strength. Each feasible resource utilization options
have various combinations of those available options, and it provides an expected
daily production rate, cost rate, and quality performance. The multi-objective
genetic algorithm was implemented in the model to search for optimal resource
utilization options for each activity that provides minimum project cost and time
while maximizing quality performance. The model also provided visualization of
tradeoffs among time, cost, and quality for the analyzed example.
Xiong and Kuang (2008) have used ant colony optimization (ACO)
algorithm to solve time-cost tradeoff problem. The modified adaptive weight
approach (MAWA) proposed by Zheng et al. (2004) was incorporated in the model
to generate optimal time-cost tradeoff curve. ACO algorithm is inspired by ant
colony behavior, and it searches optimal solutions in combinatorial problems. Each
of m ants constructs one solution in every generation, and the heuristic information
and pheromone information are used to select an option to perform an activity.
MAWA utilizes information from the current set of solutions to generate an
adaptive weight for each objective, and thereby exerts a search pressure towards
the ideal point. The performance of developed model was compared with GA by
two test examples. The results showed that ACO-based multi-objective approach
provides an attractive alternative to solving construction time-cost optimization.
13
Optimizing resource leveling, resource allocation, and time-cost tradeoff analysis
are often interrelated with each other. Therefore, a number of research studies have
formulated the models that simultaneously optimize those conflicting objectives
(Hegazy 1999b; Hegazy and Ersahin 2001; Leu and Yang 1999; Senouci and Eldin
2004)
Hegazy (1999b) have developed the optimization model that can
simultaneously consider resource leveling and allocation. The model was
developed using the macro language of Microsoft project to provide a project
manager with an automated tool to improve the results of their familiar software.
MS-project allows user to specify the priority levels for activities, and thereby
resource conflicts can be resolved by the specified priority levels. These priority
levels for activities were set as genes in GA, and the activities were scheduled by
MS project based on the randomly generated priority levels by GA. Two resource
leveling metrics, Mx and My, were then calculated for the schedules to evaluate its
level of resource fluctuation. The calculated values of two metrics were used as
fitness functions in GA to optimize resource leveling. However, the resource-
leveling feature of MS project schedules the activities as early as possible based on
their specified priority levels. Accordingly, the model might not generate schedules
that might have better level of resource utilization than those produced by MS
project.
Hegazy and Ersahin (2001) have presented the spread sheet model that
simultaneously considers time-cost tradeoff analysis, resource allocation, resource
leveling, and cash flow management. The genetic algorithm was implemented in
Microsoft Excel to facilitate the development and the optimization. Those
objectives are aggregated in one objective function, total project cost, to be
minimized. Those objectives that violate the constraints were penalized using the
cost term in the objective function. The model has a capability of searching for
14
optimal schedule that minimizes the total project cost under time, cost, and
resource constraints simultaneously. The results of case studies confirmed the
consistency and good performance of the model.
Leu and Yang (1999) have developed the multi-criteria optimal model for
construction scheduling. The model integrates time-cost trade-off model, resource
allocation model, and resource leveling model. A genetic algorithm was used in the
model to search for optimal solutions. Two GA operators, UX3 and UM3, were
introduced and used in the model to enhance the performance of GA in searching
for optimal solutions. The study also developed a resource leveling index (RLI) as
an objective function to optimize multiple resources leveling. This objective
function uses the penalty value (P) to prevent the violation of precedence
relationships and resource overutilization. In order to search for non-dominated
solution, TOPSIS, a MADM method that was proposed by Hwang and Yoon
(1981) was used in the model. The performance of model was compared with
traditional heuristic and mathematical models by analyzing case studies. The
results showed that the GA has great advantages and performance in
simultaneously optimizing multi-objectives for constructions scheduling.
Senouci and Eldin (2004) have presented an augmented Lagrangian genetic
algorithm model for resource scheduling. The model considers all precedence
relationships, multiple crew strategies, total project cost minimization, and time-
cost tradeoff. The model was formulated to consider both resource leveling and
resource constrained scheduling. The quadratic penalty function was used to
transform the resource scheduling problem to an unconstrained one. The model
allows any linear or nonlinear functions for the presentation of cost-duration and
resource-duration relationships. The model is capable of searching for optimal
durations for each activity and construction schedules that minimize project cost
while satisfying the constraints such as resource constraint, activity relationship
15
constraint, and duration constraint. The application example showed that the
developed model was able to simultaneously optimize resource leveling and
allocation as well as total project cost.

16