You are on page 1of 17

AUTCON-01810; No of Pages 17

Automation in Construction xxx (2014) xxxxxx

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Dynamic planning of construction activities using hybrid simulation


Hani Alzraiee , Tarek Zayed, Osama Moselhi
Department of Building, Civil, and Environmental Engineering, Concordia University, 1515 Ste.-Catherine, W., Montral, Quebec H3G 1M7, Canada

a r t i c l e

i n f o

Article history:
Received 7 February 2014
Received in revised form 24 August 2014
Accepted 26 August 2014
Available online xxxx
Keywords:
Traditional project planning method
Simulation
Construction
CPM

a b s t r a c t
Traditional planning methods such as CPM and PERT have been useful tools to manage construction projects' execution. However, models developed using these methods often fail to deliver realistic estimates of project duration, cost and productivity. Failure of traditional planning methods is attributed to the uncaptured causaleffect
relationships that exist among the project variables. This paper presents a new method that integrates DES and
SD models to address the operational and soft/strategic variables on a single computation platform. The expected
outcomes are realistic project schedule networks and enhanced understanding of the interactions of the project's
factors. Two cases from the construction sector are used to test and verify this method. Productivity and completion durations were monitored with/without the impact of factors, in which a signicant discrepancy has been
observed. The new method provided better understanding of the project behavior and contributed to overcome
limitations associated with traditional planning methods.
2014 Elsevier B.V. All rights reserved.

1. Introduction
A project schedule is developed by disintegrating the Work Breakdown Structure packages into activities, and then establishing a logical
relationship among those activities [1]. Such a model is expected to be
the tool for achieving a complex planning, executing and controlling
of the project. However, on many occasions, project schedule models
fail to provide a concise depiction of the project structure and its real behavior. Such failure is attributed to the complexity, dynamics, uncertainty and the heterogeneous nature of the construction environment
[24]. The interrelationships among the project variables and surrounding factors are in reality complex causaleffect relationships and not linear as traditional methods suggest [4]. Interactions among project
elements, whether internal among the elements themselves or external
due to the surrounding environment, are a source of challenges that can
hinder developing realistic and a representative planning models. The
other aspect of the problem is related to the changing behavior of the
project system over time (dynamics), e.g., impact of weather condition
on schedule and productivity, and consequence of overtime policy on
productivity. Those dynamics signicantly impact the execution of project plans and almost are neglected in traditional planning methods.
Consequently, Critical Path Method (CPM) based schedule baselines always experience high uncertainty in execution and require continuous
revisions and enhancements to capture the dynamics generated during

Corresponding author at: Department of Building, Civil, and Environmental


Engineering, Concordia University, Montreal, Quebec H3G 1M8, Canada. Tel.: + 1 514
848 2424x7091.
E-mail address: hani.alzraiee@gmail.com (H. Alzraiee).

the project's execution. In reality, the project is a system of interrelated


elements, in which each element is unique in nature, and interacts with
other element(s) to generate behavior. Current planning tools consider
planning from a static and fragmented perspective. The integrity of a
project's strategic and operational variables and their interconnectivity
has never been addressed by traditional methods. In contrast, traditional methods break the project system structure into subsystems and deal
with one aspect only (usually operational). Eventually, project plans
(e.g., schedules) developed using traditional methods result in models
that are discrete in nature and not representative of the system.
Since introducing CPM in the mid-50's, and the later evolution of the
Precedence Diagramming Method (PDM) and Program Evaluation and
Review Technique (PERT), traditional scheduling tools provide a useful
support although without high precision to decision makers. The shortcoming of traditional planning tools as discussed previously, can be
overcome by a more integrative computation environment, possibly
coupling with other techniques. The limitations associated with CPM
have been addressed by many researchers [5,6]. The strength of traditional methods lies in their ability to model details at the activity level
or entity (e.g., duration, resources, and cost); however, they neglect
the interrelationships and dynamics among these activities. While developing a CPM-based network, it is assumed that any surrounding factors such as weather, overtime, schedule pressure, rework cycle has
minor effects on the outcomes; in addition, the activities are considered
dynamically unrelated. This assumption has been proven a major pitfall
of CPM-based networks, and many projects are experiencing cost and
schedule overruns due to the simplied and linear approach of addressing project activity networks. In reality, the interrelationships among
project inuential factors are more complex than what have been suggested by traditional methods [711].

http://dx.doi.org/10.1016/j.autcon.2014.08.011
0926-5805/ 2014 Elsevier B.V. All rights reserved.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

In addition to neglecting a project's dynamics, traditional project


planning methods have shortcomings in addressing the uncertainty observed in estimating/computing costs and duration. For instance, the estimation of an activity's duration and cost is performed based on a
deterministic approach that eventually results in a deterministic number. However, in reality, durations/costs are uncertain and continuously
change based on the internal/external interactions of a project's elements. As such, those parameters are better represented using a probability distribution rather than a crisp number. This uncertainty has been
addressed by the PERT method, where three numbers are used to estimate most probable costs or durations. However, PERT tends to jump
to results considering a certain allowance for contingencies, and not addressing the reasons behind the cause of the contingencies. It is widely
believed that contingency allowances are added to account for uncertainty and dynamics effects. To address those concerns, simulation
tools mainly hybrid ones are considered a promising platform. Methods
such as Discrete Event Simulation (DES) and System Dynamics (SD) can
be utilized to account for uncertainty and dynamics. DES is effective in
analyzing the stochastic nature of parameters at the tactical level. However, DES falls short in modeling a project's holistic level, as well as the
interrelationships (feedback process) among its elements. On the other
hand, SD is a powerful tool in addressing DES pitfalls. SD is an excellent
tool in modeling feedback process in a project, which represents the interaction between project elements, as well as at the holistic level.
The objective of this paper is to present a planning and scheduling
method that addresses the uncertainty of estimates (e.g., costs and durations) as well as the project's dynamics, simultaneously. The proposed
method utilizes a CPM-based network built in a DES environment and
integrated with an SD model. It quanties the management policies, decisions, and soft variables surrounding a project, and then considers
their impacts on a schedule built using a traditional planning method.
The paper commences by presenting the background to state-of-theart in scheduling and simulation, in addition to the strength and weakness of each technique through a comparative study. The developed hybrid simulation method is presented in Section 4, followed by
implementation using two cases from the construction sector to test
the applicability of the proposed method. In conclusion, the paper summarizes the limitations of the planning method, future suggestion and
improvement.
2. Background
Modeling and simulation are powerful tools in representing a
system's real behavior in the virtual world [12]. A good simulation
model that is capable of mimicking reality should consider four major
requirements: 1) type of decision levels; 2) the nature of each variable;
3) complexity of system; and 4) the relationships between variables
[13]. Decision levels in construction projects are divided into two
parts: a) strategic and b) operational [14]. The strategic denition in
this paper is different from the denition pertaining to organizational
management. Strategic Level relates to achieving the project set of objectives within the project policy framework. This involves adjustment
of certain parameters like costs, resources, and time to meet a prior
set of goals [7]. On the other hand, the Operational Level is related the
necessary actions taken to meet the project goals set at the Strategic
Level. The Operational Level focuses on daily operational details required at the micro level of the project. Generally, construction projects
involve discrete and continuous variables. Those variables are related in
causaleffect relationships. The system behavior is mainly generated
based on the internal and external interactions of those variables within
the feedback process that result in a more complex but more representative system.
Several techniques and tools were developed to plan, schedule, and
control the execution of construction projects and assist management in
making informed decisions [15]. The scheduling techniques address the
Operational Level (activity), assuming no effects due to uncertainty,

dynamics, and interrelationships. Thus, traditional techniques describe


the project as top-to-bottom hierarchy through decomposition of project elements into the smallest acceptable level, where work packages
could be described easily by activities. Thereafter, costs, durations, and
resources are estimated, mainly from experience, as deterministic numbers. Then the project's job logic is demonstrated as a network of activities interconnected based on work sequence and logic. The apparent
purpose of this process is to develop a prototype that depicts the execution of the project in reality. One of the main concerns in such static and
linear philosophy of modeling a project's plan lies in the ability of the
restructured activities to behave based on assumptions considered at
the project's decomposition stage. Furthermore, management in reality
is dynamic and responsive to new changes to keep a project on track,
rather than adhering to the original plans. Those original plans are
targets, or baselines, of the management. When those targets are threatened by overrun, then management actions are triggered to streamline
a project toward its stated targets. Hence, traditional planning methods,
in reality, produce static baseline plans, which are implemented in a dynamic environment. Those dynamics are resultants from the causal
effect feedback loops that characterize any system with changing behavior over time. Researchers have addressed the limitations associated
with traditional planning methods, such as quantifying subjective factors to better estimate project completion duration [16], developing
project schedule in a DES environment to generate comprehensive information for project planning [17], combined DES and continuous simulation to account for factors surrounding the project and uncertainty
[8], and many other efforts in this eld. In general terms, the research
conducted in this regard is specic to certain limitations and can be categorized as special purpose models, e.g. weather impact, uncertainty in
duration estimates. In this paper, the authors present a generic method
and platform that is more capable of addressing the project dynamics,
subjective factors, and operational level (activity) than traditional planning methods.
The SD method was introduced by Forrester [18] as a method for
modeling and analyzing complex system behavior in industrial management. The method has been used solely or coupled with other
methods in different elds of social science where a holistic view and
feedback process are critical in understating the evolution of the system
behavior [1922]. The SD model aims to capture the feedback processes
responsible to the system behavior within a predened boundary.
When a project management team strives to close the gaps between
project actual performance and preset targets, such practice is an application of one principal of SD in project management and control [23].
The feedback process experienced in Case Study No.1 shown in Fig. 1
is considered to elaborate on the dynamics process as generated during
execution.
Productivity of engineers is inuenced by many factors such as high
schedule pressure, skill level, overtime, and rework. When project execution starts, the expected output is slow and takes the pattern of
ramping up in the rst 1/5 of project life. The ramping up continues
until normal productivity level is reached. The case study being analyzed had the completion duration underestimated; however, it was
still required to nish the project on time. Many factors played a role
at this stage. For instance high schedule pressure had caused a need
for increase drawing production beyond normal limits. When actual
productivity of a project falls behind the perceived required productivity, the anticipated completion date becomes invisible. Consequently,
management must adopt certain policies to reduce the adverse effects
of productivity loss. These policies can be overtime, hiring new workers,
extending the project completion duration to improve productivity, etc.
in order to attempt to nish the project on time. Another aspect that is
of concern is that not all work completed meets quality requirements. A
rework cycle during construction phase is inevitable, and initial error
correction may cause a secondary errors. The dynamics generated between these factors and other factors extends the project completion
duration and creates reinforcing and balancing loops. For instance, one

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

estimated
completion time

remaining time in +
schedule

project
deadline
+
+

time needed
-

days lost from


schedule
+
effect of morale

++
error generation
rate
+

+
schedule pressure
+
+

productivity loss
+

project actual
productivity -

perceived
completion rate

overtime

fatigue

drawings
quality
-

+
work completed
+
with error
+
work with
errors
+

productivity noise

+
new
workforce

work
remaining

+
+

project
scope

Fig. 1. Feedback processes in a project execution cycle (Case 1).

of the loops shown in Fig. 1 depicts what happens when project completion duration is underestimated. As can be seen at the top of the gure,
two external factors affect the schedule pressure, estimated completion time and remaining time of work already started. When the
schedule pressure rate increases to a level that impacts productivity
(e.g. greater than 1.5), different polices can be applied to overcome
this effect. This process can be overcome either be considering working
overtime or hiring new workers. Signicant overtime or hiring new
workers has impact on the quality of work. In both cases, more rework
and hence increasing scope of work to be completed and fatigue level
will also be at a higher level. This cycle of cause-and-effect will continue
to a point where another feedback loop must be enforced to counter
these adverse effects. For instance, training, incentives, and revising
the project completion date, could be solution to balance the negative
effect of schedule pressure on productivity. In the case study being analyzed, no such polices where considered to neutralize the adverse effects of the negative effect loops.
3. Planning methods comparison
In this section, the differences between core characteristics of the
traditional planning methods are presented. The purpose is to provide
a clear and concise focus on the main differences so that strength of
each method can be utilized in developing the proposed method. As
stated earlier, planning methods can be classied as methods that address the project process level, uncertainty, dynamics, and interaction.
There are many studies in literature that compare and contrast planning

and simulation methods [24,25]. We summarize these ndings in seven


main perspectives as shown in Table 1. The CPM-network and DES
methods are appropriate for modeling issues that are operational in
focus, reductionism in perspective, quantitative in nature, discrete in
change, and narrow in details. The only difference between the CPMnetwork and DES is the later addresses uncertainty. On the other side,
the SD method is appropriate for problems that are of strategic/context
focus, holistic in prospective, qualitative in nature, continuous in behavior, and broad in details [24]. The main strength of SD modeling method
is its ability to mitigate the limitations associated with CPM-network
and DES [25]. Thus, both SD and DES/CPM can be viewed as a complementary to one another and the limitations associated with each can
be overcome when both methods are integrated. From analysis of the
strengths and shortcomings of each method, an integrative platform
that builds on the strengths of each method will inevitably enhance
the practice of project planning and thereafter execution and control.
In contrast to CPM-networks and DES, SD modeling and analysis
offer a different perspective on understanding system behavior. This is
because an SD model developer rst understands the underlying inuences responsible on the outcomes, while a developer of CPM-networks
tries to anticipate to the outcomes without considering the underlying
inuences. In CPM, the outcomes are usually computed based on several
approximations that eventually neglect an important aspect of the
problem, such as, for example the surrounding environment. The need
to have a hybrid-planning platform is immense as results of the increasing complexity of construction projects and their planning became evident. Hybrid planning can be achieved in different ways based on the

Table 1
Comparison between planning methods.
Perspective

Traditional method CPM-network

DES method

SD method

Focus
Level of details
Behavior
Model type
Data type
State change
Complexity

Activity
High details
Linear
Interrelated but distinct packages
Quantitative
Discrete
Specic

Operation
High details
Stochastic
Interrelated but distinct packages
Quantitative
At discrete points in time
Narrow and focus on complexity and details

Holistic and feedbacks


Little details
Deterministic
Continuous ow
Qualitative
Continuous
Wider focus, general and abstract system

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

project management team needs. Rodrigues and Bowers [7] suggested


three approaches that can be utilized to develop a comprehensive
project-planning model:

Start

Build Hybrid simulation model


SD/DES

1- A more sophisticated network model including the feedback processes and detailed mechanisms for modeling activity durations
and costs in order to reect the underlying inuences,
2- A more detailed and phased SD model, and
3- Adopting lessons learned from SD models in the forms of set of rules
for use in planning and estimation.

Build new SD/DES model

Identify model hybrid


structure

The three aforementioned approaches attempt to utilize SD in addressing the two-decision Levels in projects (Operational and Strategic).
Attempts have been successful at strategic level modeling; however, at
the Operational Level, using SD has not been proved sound despite its
use in limited and small applications to model the process level. Whenever SD model has been used at the Operational Level, the model became too complex and involved tedious work. In addition, in cases
where SD modeling was used at the Operational Level, the modelers
broke down the whole project system into subsystems. Such practice
violates one of the fundamental principles of SD (holistic view). Therefore, it is more appropriate to use each simulation method in areas
where it shows strength, consistency, and simplicity; otherwise, the applicability of the method by the end-user becomes cumbersome and
inappropriate.

Define model boundary, function & role


for DES_SD system

Define inputs/outputs and origin


of data source
No

Are Inputs /outputs consistent with model


function and boundary

Yes
Define the interface variables in SD/DES
model
No

4. Methodology

Are selected points consistent with


model objectives

4.1. Integrating DES and SD

Yes

The systematic process of Integrating DES with SD is summarized in


Fig. 2. The method consists of: (1) developing DES and SD models,
(2) formalism to describe the models, (3) synchronizing DES and SD
models, and (4) execution platform (Executer).

Ready to formalize interface variables

End

4.2. Developing simulation models and identifying interface variables


Step 1 involves developing simulation models and selecting interface variables through which mapping between the variables of both
models will take place as shown in Fig. 3. The DES model will address
the project schedule network at the operational level while the SD
model will be concerned with quantifying the dynamics generated
from the interactions of the signicant factors. The interactions between
the two models will occur through the interface variables that receive
data from sender variables and deliver data to receiver variables. The
data exchange between the simulation models is unilateral. This
means the hybrid model structure that consists of DES and SD will
govern this process. For instance, in this paper, the purpose of
modeling Case No. 1 is to study the impact of the surrounding
factors and strategic variables on a CPM-network developed in a DES
Developing Hybrid Simulation Model

Fig. 3. Building DES and SD models for a hybrid simulation model.

simulation environment. Therefore, the data ows from the SD model


to the DES models through the interface variables. In Case No. 2, the purpose is to study the inuence of operational level variables on the global
SD model. Next step involves identifying the interface variables and the
interactions direction and then variables are described using the mathematical formalism.
4.3. Formalism of DES and SD models
Simulation modeling is not accomplished by directly writing out a
dynamic system structure, but indirectly, by using system specication
formalism. System specication formalism is a shorthand for specifying
a system [26]. In this paper, formalism is the method used to describe
the simulation modules to the Executer. For the purpose of the integration of DES and SD models, formalism must specify the following
properties:

Formalism of DES and SD

1.
2.
3.
4.

Synchronization of Simulation Clocks

Executer

Model type: whether it is DES or SD.


Input variable (receiver) and source of the input variable.
Output variable (sender) and source of the output variable.
Synchronization function that describes the simulation time management and the interfacing variables.
The aforementioned four requirements are summarized in Eq. (1).

Implementation
Input
Fig. 2. Hybrid simulation framework.

Hybrid DES&SD model I; O; M; S

Where:
I: Set of inputs
O: Set of outputs

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx


SD variable state
update behaviour

M: Set of modules (DES or SD)


S: Synchronization process (time and interfacing)
The generic Eq. (1) provides a base for deriving Eq. (1.1) to represent
either DES or SD model to the simulation Executer. The proposed simulation Executer requires three sets and two functions to describe DES or
SD modules (m), as shown in Eq. (1.1).
m mt ; vall ; min ; mou ; T b

DES_SD
Hybrid State
after variables
interfacing

State
update

DES variable state


update behaviour

1:1

Where:

mt: Module type (SD or DES)


vall: Set of all interface variables in the module
min: Set of module input variables (receiver), described by Eq. (1.2).
mou: Set of module output variables (sender), described by Eq. (1.3).
Tb: Time point where the interfacing of variables occur

n
o
min m; vi ; ms ; opms ; md =mM; vi V M ; ms M; opms outportsall; md vall

1:2

Tb1

Tb2

Tb3

Tb4

Tb5

Tb6

Tb7

Simulation length L
Fig. 4. Synchronization of DES and SD models using time bucket.

points, rather than at the occurrence of events. Tb is set to be equal to the


SD model STEP TIME as shown in Eq. (2) and Fig. 4.
Tb hybrid model SD model STEP TIME

Where:
vi: Input variables to module m
ms: The module source from which input variables are imported
opms: The output port in ms from which the input variables are
imported to m module
md: Describes the variables in m that need to use the input variables
from ms. (md describes the variable input vi input port (ipm) in m, the
variables in m need to use vi and the time point in the simulation
clock where the interfacing of variables occur)
M: Describes the simulation model (DES and SD) in the hybrid environment (hybrid model)
VM: All variables in model M
outportsall: Set of all output ports in m
vall: All variables in m

mou

n
o

m; op; ov; =m M; op outport m ; ov vall

1:3

Where:
op: Output port in m
ov: Output variable given through op
4.3.0.1. Simulation clock synchronization. Time management or synchronization means the execution of events that occur in distributed simulation in a specied order that ensures repeated executions of a
simulation produces exactly the same results under the same inputs
[27]. The simulation clock of the DES method advances in a different
fashion than the SD method. In DES, the simulation time is driven by
the occurrence of events and subsequently system states are updated.
Events usually occur at unequal time intervals. In SD, the simulation
clock advances at equal time intervals and system states are updated
at the end of each time interval.
The proposed synchronization method utilizes the concept of Time
Bucket [28,29]. It divides the hybrid simulation time of length L into
equal time intervals called Time Buckets (Tb). At the end of these time intervals, interfacing of simulation models takes place. Tb should be small
enough to capture any signicant changes in system state and large
enough to discard unnecessary overhead computations. The rationale
behind using the Time Bucket is that SD updates states at equal and
pre-known time intervals while DES updates states at the occurrence
of events. Those events generally occur at unequal time intervals; therefore, it is easy to trace states of the simulation model at stipulated time

At the start of the simulation time length L, the proposed hybrid simulation method initializes the simulation clocks of DES and SD engines,
as well as, the variables data. Now, DES_SD code is positioned to advance the simulation clock to Tb1, where,
Tb Tb1 Tb2 :Tbn

At the end of Tb1, the interface of the variables and their respective
data exchange occurs between DES and SD as shown in Fig. 4. The
data ow direction between the two models is based on the structure
of the hybrid model through the interface variables. Upon completion
of this process, the hybrid DES_SD advances the simulation clock to
Tb2 and again at the end of the Tb2 interfacing of variables occur. The
events of the synchronization process that take place between Tb1
&Tb2 are illustrated in the algorithm shown in Fig. 5 [30]. The developed
algorithm works as follows. Initially, for the DES engine to start advancing the simulation time, a condition such as, the required resources and
entities should be available at the start of TB1. Now the simulation is in
position to start advancing at the beginning of TB1, if entities seize the
required resources, then all data of active resources and entities in the
simulation model are read and saved, otherwise, idle resources data
are read and saved. If the process involving the active resources has
not nished processing the entity at the end of TB1, then pause DES simulation clock advancement, save all data and perform DES and SD modules interfacing. Otherwise, eliminate saved resources and entity data
and return to re-allocate the next process and its entities, attributes
and resources. In the DES model, events having occurrence time less
than the TB1 nish their processes before the interfacing of variables
can take place, hence their data are not captured in the next earliest
scheduled interfacing, but their effects are propagated to the second
event. Therefore, in order to avoid events that start and nish before
the end of the TB, it is advised to set the SD TIME STEP less than or
equal to the lowest expected event time in the model.
After the interfacing is accomplished, all saved data at the end of TB1
(interface time point) of active or idle resources, are used by the DES engine for the next round of computations that begins by the commencement of TB2 and continue in the same sequence explained in TB1. The
time point between the end of TB1 and start of TB2 is the point where
the simulation clock resumes the progressing of model simulation.
The algorithm continues until the model reaches the initially set simulation time of length L and then terminates the simulation run.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx


Start
Tb+1
Resources and Entities are
available at start of Tb

Advance simulation clock

No

Do entities seize
resources

Idle resources

Yes

Active
resources

Read and save resources ID


Read and save resources idle state
Read and save queue length of Idle
resources
Read and save times

Read and save resources ID


Read and save entity ID
Read and save queue ID
Read and save queue length
Read and save start processing time
Read and save server time
Read and save served resources time

Yes
Is idle state of resource
changed before Tb end?

No

Yes
Is entity processing
completed before end of Tb?

No

Eliminate saved
resources data

Eliminate saved
resources and
entities data

Pauses DES simulation clock


advancement, save all data stated and
perform models interface

Pauses DES simulation clock


advancement, save all data stated
and perform models interface

Interface Completed
Resume simulation starting from all saved values
at end of Tb with considering the new values
resulted from the models interface

Clear all saved data

End

Fig. 5. Algorithm to control advancement of simulation clock.

4.4. Executer
The Executer is the VB code developed to integrate DES and SD
models through utilizing the components of the developed hybrid simulation method. It focuses on integrating and controlling the hybrid
simulation model. The Executer interacts with DES and SD simulation
engines to facilitate the integration process based on the developed
mathematical formalism, the synchronization protocol, and the discrete
simulation clock algorithm. The Executer is responsible for performing
the following tasks:
1. Providing the user interface layer that allows inputs and outputs as
required by the user.
2. Providing models data importexport management. The selected interface variables that are designated to receive or share their values
are specied in the Executer.
3. Implementing the DES simulation clock-advancing algorithm.
The aforementioned components of the simulation method are utilized and integrated to develop the hybrid simulation system shown
in Fig. 6. The DES and SD simulation models are developed using DES
(Stroboscope) and SD (Vensim) engines respectively, the lower part of
Fig. 6. The input information for the hybrid simulation model, such as
simulation run length, denition of the interface variables, hybrid
model structure etc. are entered through the interface layer. As soon
as the simulation engine advances the simulation time and reaches
the dened time step, synchronization of the interface variables takes

place based on the developed algorithm and the messages exchange sequence. The structure of the hybrid simulation model plays the main
role in executing and progressing the simulation model computation.
For instance, if the SD model is designed to act as a global model, and
certain variables within the SD model need to be computed by the operational level model (DES) and exported to the SD model, then the DES
engine will compute those variables, dump them into a spreadsheet
from where the SD model will import and receive those variables at
every time step. Finally, the outputs of the simulation model such as project completion duration, productivity, and cost are exported into PDF
format for end user analysis.
5. Implementation
The proposed hybrid simulation method is implemented in two
ways, using two different case studies.
1- In the rst case study, the purpose is to investigate the impact of the
surrounding factors on a CPM-network. This means, using the SD
model to quantify certain factors, then impact the CPM-network
with those factors to monitor the changes on duration and productivity, and,
2- In the second case study, the logic is inversed by using another case
study (earthmoving operations). The whole project is modeled using
the SD model except at the operational level where DES modeling is
used to quantify the operational variables. The productivity of each

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Interface Layer
Simulation data
Interface variable
Simulation length
Hybrid structure selection

Project performance
Total cost

DES_SD Executer

20 M
Start

15 M

Tb +1

Resources and Entities are


available at start of Tb

Start advance simulation clock

10 M

No
Do entities seize
resources

Excel
Application

Where:
I: Set of inputs
O: Set of outputs
M: Set of modules (DES or SD)
S: Synchronization process
(time and interfacing)
m = (m t, v all, m in, m ou, T b)

Executer

SD Model

DES Model

(1)

Active
resources

(2)
(3)

Idle resources

Read and save resources ID


Read and save entity ID
Read and save queue ID
Read and save queue length
Read and save start processing
time
Read and save server time
Read and save served resources
time

(4)

Yes
Is ideal state of resource
changed before Tb end ?

No

Yes
Is entity processing
completed before end of Tb ?

(5)
(6)
(7)

Eliminate saved
resources data

Keep saved
resources data

Productivity
Project Cost
Schedule

5M

25

50

75

100 125 150 175 200 225 250

Time (Hour)

Actual releassed production

Eliminate saved
resources and
entities data

No

450

(9)
(10)

Total cost : 1
600

Pauses DES simulation clock


advancement , save all data stated
and perform modules interface

(8)

Tb1

Read and save resources ID


Read and save resources idle
state
Read and save queue length of
Idle resources
Read and save times

Yes

Resume simulation starting from all


saved values at end of Tb

Clear all saved data

m3/Hour

Hybrid DES_SD = (I, O, M, S)

300

150

End

Formalization

Synchronization

Sim clock Adv. Algorithm

00

25

50

75

100 125 150 175 200 225 250

Time (Hour)
Actual releassed production : 1

Specify Hybrid Structure and Interface Variables

Strategic/Context layer

Operational Layer
Dumped

Loader Idel

Rocks

>0 , 1
1

Rocks to
Move

DES
Operational
level

>=38 , 38

38

Load

Haul
Triangular [4.9,5.6,5.7]

Uniform [2,2.2]

Dump
Uniform [1.4,1.6]

SD
Strategic
and context
level

>0 , 1

Truck Wt.

Return
Uniform [3.4,3.6]

DES Engine

SD Engine
Fig. 6. Architecture of the hybrid simulation system.

operation (excavation, hauling, and dumping) is inputted through


the interface variables into the SD model to get impacted by the project dynamics and subjective factors.
5.1. Case Study1: engineering drawings production by a design rm
The proposed planning method is implemented as described in the
methodology section. An SD model is developed to capture the effect
of the surrounding factors and dynamics such as limitation of skilled
labor, rework cycle, and schedule pressure (the paper focuses on the
idea itself rather than explaining the development of the SD model
structure). The SD model creates a dynamic framework that exhibits
the classic characteristics of a project's dynamics. Within this framework, a CPM-based network is developed in a DES environment to describe the job logic and compute the project's completion duration
considering uncertainty. The implementation of the case study in the
simulation environment is demonstrated in Fig. 7. The project scope is
decomposed into smaller units to develop the work breakdown structure (WBS), from which activities are identied. Each activity's duration
and cost are inputted as probability distributions. ProbSched is utilized
to develop the CPM network [31], as shown in Fig. 7. The ProSched is
a probabilistic scheduling package that uses Stroboscope as its engine

and Microsoft Visio as its Graphical User Interface. ProbSched allows


the denition of CPM networks where the cost and duration of each activity can be dened probabilistically. ProbSched produces graphical
output to indicate the criticality of each activity and statistics of the
early and late times and oats of each activity and the project. The second component of the implementation involves developing SD model
responsible on modeling the subjective factors. The SD model is developed using Vensim Software Package from Ventana Systems, Inc [32].
The Gantt chart shown in the gure is a sample output of the hybrid
model and represents the schedule network of activities after impact
by the SD model.
5.1.1. Data of Case Study1
A design rm was hired by a client to produce a variety of engineering drawings for expansion of Oil Sands existing facilities. The engineering drawings were: 1) civil (Task1); 2) mechanical and structural
(Task2); 3) electrical (Task3), and 4) piping (Task4). The designs were
prepared by different engineering departments within in the same
rm. A resources loaded schedule that depicts the logical sequence
among the four activities was developed using the traditional methods
and provided to the client. The main purpose of the schedule was to estimate the completion durations of the project, and track the progress of

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Gantt Chart of Simulated CPM


network after Impacted by SD model

Discrete Simulation Engine (CPM)


CPM or PDM

Project Policies and


information

Total Rework

Percentage of
Compacted Soil

Activities

Rework Process
Rate Duration

Error Generation
Rate

Spread Soil

Rework process
rate

Rework
Error Rate

Rework rate
<Impact of Fatigue
on Quality>

Avg, QP
Duration

Quality Check
Process Rate

Scope Task is Done C

Durations
Costs

Scope Task Start Flag C

Resource

Quality Process
Rate

Perceived
Quality Rate

Scope Start
Percentage of
Compaction

Scope Task is
Active C

Soil Compacted
and ready for
Quality Check

Soil to
Compact
Compaction
Rate

pread

Productivity Rate

Total Soil
Compacted and
Ready for Quality
Check

effect of schedule
pressure on
productivity

<Schedule
Pressure>

<Impact of Fatigue
on Productivity>

Activities Information

Final Work
Completed

Total Project Work of Soil,


Hauled, Dumped, Spread, and
Compacted

<Net Compaction
Rate>

Max Compaction Rate

Perceived
Rework Rate

Effect of Schedule
Pressure on
Productivity
Lookup

SD Simulation Engine (SD Model)


Fig. 7. Layout of the proposed planning system.

Table 2
Case study data.
Task name

Task scope in units

Triangular probability dist.


of task duration in weeks

Task1(T1)
Task2(T2)
Task3(T3)
Task4(T4)

40,000
50,000
18,000
100,000

20, 20.4, 20.9


20, 20.2,20.6
10, 11.7, 12
39.9, 40.5, 40.8

phenomenon is well known in construction work execution and is


depicted in Fig. 8.
5.1.2. Models development, result and discussion
The developed hybrid model consists of a CPM-based network
(Fig. 9) and SD model. The purpose of the CPM network is to capture
the operational level parameters such as durations and the logical sequence of the activities, while the SD model will capture the dynamics
generated from the model's variables due to internal and external interactions. The CPM network is compiled in DES engine (Stroboscope)
using a ProbSced add-on application. The activity start time is similar
to the activity start date in the conventional schedule network;
1

Performance Factor

completed drawings. Based on the CPM schedule, the project was expected to be completed in 70 weeks; however, the actual project completion duration exceeded this initial estimate by 40% (28 weeks).
Therefore, this was a base for a good case study to investigate, model,
and analyze. The data collections also involved understanding the circumstance surrounding the projects, such as maximum available manpower, overtimes, management policies, and schedule pressure level
as this helps in understanding the dynamics of the project system. The
rm used units system to quantify the efforts needed to nish a single
drawing (e.g., a drawing requires 2000 units of work to complete).
The productivity of individuals is measured by the number of drawings
completed and checked per week. The data of the four tasks is shown in
Table 2. Each Task duration was estimated using three points: optimistic, most likely, and pessimistic. The maximum available skilled workforce dedicated to this project was 160. It was recorded by the project
manager that maximum production per drawing is reached when 20%
of the drawing is complete and declined in the last 20%. This

20%

Percentage of drawing 80%


Completion
Fig. 8. Planned work prole.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Gross productivitytask normal productivitytask


effect of morale on productivitytask
effect offatigue on productivitytask
effect of schedule pressure on
productivitytask
Units : dwg=person=week
5

Task 1
Task 2
Task 3
Task 4

Undiscovered Reworktask gross completion ratetask


1work qualitytask
rework discoverytask; 0:

20
40
60
80
Simulation Time (weeks)
Fig. 9. Gantt chart of the tasks before simulation.

however, since the activities are bonded to the simulation time length,
we will refer to the start date as the start time as shown in column
(1) in Table 3. The durations for every activity are inputted into the
CPM-discrete simulation network as a triangular distribution. The
model ran for 500 cycles and the average duration for each Task were
computed as shown in column (2) in Table 3. The total completion duration as computed by the CPM network in the DES environment was
70.9 weeks. It was noticed that the difference between the deterministic and stochastic total project duration is minor, and this was attributed
to the minor differences in the three estimated durations of the
activities.
The next stage was developing an SD model to represent the
project's dynamics based on the causal loops diagram shown in Fig. 1.
These loops are rigorously researched and cited in literature by many
researchers [21,23]. The developed SD model is composed of four modules (workow, rework, quality, and labor demand). The workow
module describes the workow from execution to completion. A rework cycle module is added to account for work that does not pass
the quality standards and needs to be reworked. The scope of rework
is returned to the initial stock for further processing. The schedule pressure resulting from low productivity and increasing rework is captured
as well. The strategy of the rm in addressing mounting schedule pressure was by hiring likely additional workforce or considering overtime
for the current staff. The SD model structure was developed in such a
way to capture the effects of schedule pressure, fatigue, overtime,
and rework cycle on quality of work and project completion duration.
All of these interactions are captured within the causaleffect loops
that depicts the dynamics specied in the SD model framework
(Fig. 1). Samples of the equations used are shown in the equations numbered 4 to 9.
Work Completed Correctlytask
gross completion ratetask  work qualitytask; 0
Units : dwg

Rework discoverytask MAX0; MINUndiscovered Reworktask=


TIME STEP; UndiscoveredReworktask=
rework discovery timetask
SUMdownstream rework
discdownstream!; task
Un3its : dwg=week
7

Work remainingtask MAX0; TASK DEFINITIONtask


reported work completetask
Units : dwg

Reported work completetask Work Completed Correctlytask


Undiscovered Reworktask
Units : dwg
9

The SD model was simulated, and a sample of the results is shown in


Fig. 10. The project completion duration estimated by the discrete simulation model had been extended from 70.9 weeks to 92 weeks,
Fig. 10a. This represents an additional of 32% to the project planned
completion duration. In addition, it can be observed in the gure that
the start of Task 4 was delayed from the 30th week to the 48th week.
This delay is attributed to the interesting pattern observed in Fig. 10c
d during this period. In the gure, the quality standard of the executed
drawings was degraded in the period from the 25th week to the 48th
week. The poor quality evident is mainly due to the delays of starting
Task 4, which demanded more resources than was planned. Consequently, limited resources and signicant delay increased the schedule
pressure; therefore, a higher productivity rate was required from the
overstretched resources. Increasing the productivity rate beyond the allowable limit caused some drawings to be nished not per accepted
quality standards. Therefore, it can be observed in Fig. 10d that the number of the drawings that need to be reworked has increased to the maximum level between the period of the 30th week and the 45th week.

Table 3
Simulation data inputs for the CPMDES schedule.
Task Name

Task start time


weeks(1)

DES Simulated average


duration in weeks (2)

Completion milestones
in Weeks (3)

Predecessor required
to start (4)

Productivity of engineer
drawing/week (5)

Task1
Task2
Task3
Task4

0
20
40
30

20.5
20.35
11.4
40.90

20.5
40.35
51.4
70.90
70.90

T1
T1,T2
T1,T2

20
25
15
25

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

10

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx


Run1

60,000

task is active[task]

Units

45,000

Units
TASK1

30,000
TASK2

15,000

4
12

TASK3

1234

25

50
Time (Week)

75

1
3

234

TASK4

10

234

34

20

30

34

34

40

34

a) Gantt chart of completed Tasks

50
60
Time (Week)

70

Cumulative Of Completed Work Units Per Task[TASK1] : Run1


Cumulative Of Completed Work Units Per Task[TASK2] : Run1
Cumulative Of Completed Work Units Per Task[TASK3] : Run1
Cumulative Of Completed Work Units Per Task[TASK4] : Run1

100

41

80

90

1
2

100

1
2

3
4

dwg
dwg
dwg
dwg

1
2

2
3

b) Cumulative of completed work units per Task


4,000

1
1

Units

1
1
2 34

0.9

2 34 1

12 3

12 3

12

1 23

1 23

12 3

3
3

23

3,000

0.8

Units

12 3

2,000

0.7

1,000

2
1

0.6
0

10

20

30

40

50
60
Time (Week)

70

80

90

100

123 4

2 34

10

20

c) Quality of executed drawings

40

41 23

1 23

50
60
Time (Week)

1 23

70

1 23

80

2
3

Units

4,500

4
123

90

2
3

75,000

100

2
3

2
3

4
1

41

3,000

123

d) Undiscovered rework

Units

30

2 3
3 41

34 1

100,000

6,000

2341

50,000

1,500
0

25,000
1

1
1

23 4

234 2341

10

20

34 1

341

30

23

40

3
412

4
4
4
4 12 3 1 2 3 1 2 3 1 2 3 1 2 3

50
60
Time (Week)

70

80

4
1 23

12

90

100

10

20

30

40

50
Time (Week)

60

70

80

90

100

f) Accumulated completed units

e) Productivity per Task

Fig. 10. Sample of the simulation model outputs.

This added extra scope of rework increased the initial scope of work;
consequently, increasing the schedule pressure, overtime demand,
and fatigue level. The project's execution exhausted 50% of its planned
duration while the actual productivity was not as perceived at the planning stage. These dynamics triggered the causaleffect loops that control the workforce, thus demanding additional workforce. However, at
this stage, the maximum available resources had been attained. The positive inuence of this loop was frozen, which caused other loops of negative inuence to occur, such as the schedule pressure loop to build up.
Higher schedule pressure, constrained by maximum resources triggered
the need for overtime as a policy to increase productivity, Fig. 10.e. Overtime is associated with labor fatigue (physically and mentally), and consequently, this increased the aws in the drawing production (Fig. 10d).
The accumulated impacts of those factors extended the project duration

to 92 week (32% higher than planned). The nal accumulation of the executed work is shown in Fig. 10f, which shows an s-curve behavior.
Based on these outcomes, the causaleffect loops of the model
should be reviewed and revised polices should be considered to address
the negative results observed in the outcomes. A remedy to the negative
impacts of certain factors in the model can be attained by tracing the
problematic loops and minimizing their effects as well as maximizing
the impact of the favorable loops.
5.2. Case Study2: earthmoving operations
5.2.1. Case study description
In the second case study, the earthmoving operations involved in a
dam construction were modeled and simulated using the proposed

Table 4
Scope of work.
Soil type & layers

Stage 1 m3

Stage 2 m3

Stage 3 m3

Loose density (t/m3)

Bank density (t/m3)

Load factor %

Total of soil m3

Rock
Granular
Moraine
Total

192,700
14,500
29,200
236,400

3,209,400
286,500
555,900
4,051,800

1,602,900
139,000
269,900
2,011,800

1.66
1.72
1.66
1.6

2.73
1.93
2.02
2.4

80
90
100
100

5,005,000
440,000
855,000
6,300,000

6.3 million m3
Excavation (river Bed)

1,038,000

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

11

Table 5
Fleet conguration and characteristics.
Hauled
Material

Hauler
Model

Loader
Model

Hauled Soils
(ton)

Load activity
Time Dist. (m)

Haul activity
Time Dist. (m)

Dump activity
Time Dist. (m)

Return activity Time


Distribution (m)

Rock
Moraine
Granular
River Bed Soil

777D
773D
769 C
777D

992G
990 SII
988F
375L

81.67
45.82
34.36
51.41

(3.94, 4.15, 4.57)


(3.01,3.2, 3.32)
(2.3, 2.42, 2.5)
(4.26, 4.48, 4.93)

(4.3, 4.53, 4.98)


(19.47, 20.5, 22.55)
(30.6, 32.34, 35.57)
(5.32, 5.6, 6.16)

(1.9, 2.2)
(1.6, 1.9)
(1.3, 1.5)
(1.6, 1.9)

(3.17, 3.34, 3.67)


(16.71, 17.59, 19.35)
(25.85, 26.51, 29.16)
(2.86, 3.01, 3.31)

dynamic planning [33]. The scope of the work had two main parts.
The rst was excavating and preparing the riverbed for the dam's
foundation, while the second was backlling three types of soils to
serve as the main dam structure in restraining the water. The scope of
work of excavation from riverbed was to remove, haul, and dump
1.038 million m3 of soil. Thereafter, backlling and compacting
6.3 million m3 of three soil types: 1) compacted moraine (clay) 2);
granular (sand and gravel); and 3) rock, as shown in Table 4. The backll
operations involved processes such as loading, hauling, dumping,
spreading, and compacting of the soil.
5.2.2. Fleet conguration and duration of operations
The eet conguration used to execute the project scope is shown in
Table 5. Information related to equipment was obtained from the
manufacturing specication manual of Caterpillar [34]. The eet of
equipment included three types of haulers (777D, 773D and 769C)
served by three types of loaders (992G, 990SII, and 988F), respectively.
The haulers' travel times under loaded and unloaded conditions corresponding to certain speeds were calculated by using manufacturer's
charts (RimpullSpeedGradeability and Brake Performance Charts),
total resistances, and road segment lengths. Duration times needed by
loaders to load a specic truck were calculated using loader specication charts and tables.
5.2.3. Elements of simulation model
As shown in Table 6, the case study is composed of elements related
to operational level and elements related to strategic level. The project
behavior mainly resulted from those elements and their interactions.
At the operational level, the excavation operation involved excavating,
loading, hauling, and dumping processes while the soil backlling operations involved loading, hauling, dumping, spreading, and compacting
processes. From a policy and strategic perspective, perceived productivity, weather, overtime, cut depth, road condition and others were critical elements. The elements, when modeled and simulated, are expected
to generate the real behavior of the operations in the virtual world. The
classication into operational/strategic and selection of simulation
method is performed based on the criteria presented in the background
section.
5.2.4. Developing DES models
Stroboscope [35] is a simulation language used to develop generalpurpose DES models. Processes such as excavating, loading, hauling,
dumping, spreading, and compacting are modeled using DES. Ten DES
simulation models were developed for excavation and backlling operations. The excavation DES model computes the Max Excavation Rate

and the Max Dumping Rate while the backlling DES models compute
the Max Dumping Rate, Max Spreading Rate, and Max compaction Rate.
Those ve variables are used as input in the SD global model, and later
in the synchronization implementation which will be called interface
variables. The outcomes of the ten discrete simulation results are
shown in Table 7.

5.2.5. Developing the SD model


5.2.5.1. Model boundary. An essential step in developing the SD model is
dening the model's boundary. This boundary involves selecting the
variables that generate the behavior of interest as specied by the
model's purpose. Variables in the model are classied as endogenous,
exogenous, and excluded. Endogenous variables are the main concern
of all model variables. They are variables in a causaleffect structure
whose values are determined by the states of other variables in the system. Exogenous variables are from outside of the model, and unexplained by the model's feedback structure. They are involved in a
causaleffect structure whose values are independent from the states
of other variables in the system. Variables categorized as excluded variables are cautiously not included in the structure of causaleffect feedbacks. The model's boundary of the developed SD model is summarized
in Fig. 11.

5.2.5.2. Stocks and ows diagram. Dynamic behavior in SD is raised due to


the principle of stock or level. As the name implies, stock represents a
variable state resulting from decisions. Stock is an accumulation characterizing the system state, and it generates information upon which decisions and actions are based and accumulated. Stock changes only
through ows, and creates delays in the model by accumulating the difference between inow to and outow from the stock. Stock: 1) has a
memory; 2) changes the time shape of ow; 3) decouples ow; and
4) creates delays. Finally, stock is modeled by the mathematical integration of the sum of the ows coming in to the stock and the ows
dispatched from the stock. On the other hand, ow represents actions
or variables that inuence the stock level or accumulation. Decoupling
the rate (ow) from the system, stock becomes the source of disequilibrium in system dynamics [36].
The SD model is composed of four modules: (1) workow module;
(2) schedule pressure module; (3) rainfall and road condition module;
and (4) cost module. Each module is responsible for modeling the behavior of the element that it represents. For instance, the workow
module is responsible for modeling the interactions of the variables
that describe the job-logic sequence at execution stage.

Table 6
Summary of the simulation model elements.
Operations

Operational level

Strategic/context level

Excavation

Excavation, Loading, Hauling, dumping, and Return

Cut depth, schedule pressure, road condition, operator skill,


soil type, job efciency, weather condition, and overtime.
Schedule pressure, road condition, overtime, rework cycle,
soil type, operator skill, weather condition, and overtime.
SD

Rock11, Granular12, Moraine13, Rock21, Granular22,


Moraine23, Rock31, Granular32, Moraine33
Simulation Method
1

Loading, Hauling, Dumping, Return, Spreading, and


Compacting
DES & traditional method

Rock 11 means, rst layer & rst stage (refer to Table 4).

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

12

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Table 7
DES model outputs.

Dump
(m3/h)

Spread
(m3/h)

Compact
(m3/h)

Process

Excavation

Rock 11

Granular 12

Moraine 13

Rock 21

Granular 22

Moraine 23

Rock 31

Granular 32

Moraine 33

Scope of work (m3)

1,038,000

192,700

14,500

29,200

3,209,400

286,500

555,900

1,602,900

139,000

269,900

Haulers

10

10

10

10

Loaders

Bulldozers

Compactors

Max
Min
Average
St. Deviation
Max
Min
Average
St. Deviation
Max
Min
Average
St. Deviation
Duration (hour)

1421.33*/1367.39**
1222.65*/1200.56**
1320*/1284.52**
34.40*/28.27**

808.33

1462.43
1323.11
1393.21
35.20
1462.56
1323.05
1393.21
35.16
1464.98
1321.41
1393.21
36.17
138.44

216.87
157.30
187.21
15.37
216.94
157.18
187.43
15.275
218.26
155.15
187.43
16.25
77

347.88
260.10
304.55
22.19
343.95
264.07
304.55
20.32
341.79
266.14
304.55
19.32
96.79

1462.90
1323.08
1393
35.36
1456.75
1329.15
1393.21
32.16
1458.77
1327.09
1393.21
33.67
2303.70

221.28
158.10
190
16.88
221.65
158.26
190.32
16.67
221.66
158.06
190.68
16.63
1054.72

349.82
262.01
306.29
22.06
349.05
262.18
306.29
22.06
351.71
260.42
306.29
23.06
1814.44

1466.84
1319.26
1393.08
37.25
1462.94
1323.15
1393.08
35.25
1466.72
1319.79
1393.08
37.25
1151.85

223.70
156.76
190
17.88
221.81
158.47
190.23
16.88
221.83
158.11
190
16.88
730.42

424.90
337.55
381.49
22.13
434.98
327
381.49
27.13
430
331.00
381.49
25.13
707.63

Total backll duration = 8504 h. By considering 50% overlapping, duration = 4620 h. * Excavation rate of riverbed, ** Dumping rate of Excavation of riverbed.

The workow module is composed of ve structures: 1) loading


dumping; 2) spreading; 3) compaction; 4) rework; and 5) released productivity as shown in Fig. 12. Those structures describe the real work execution sequence. The SD model is initialized at Soil to Haul stock. The
initial value of this stock is the total scope of work (6.3 million m3),
modeled mathematically, using the subscript control feature in Vensim
to distinguish the different soil types. Then the scope is processed at the
Dumping Rate ow that represents the impacted eet dumping productivity. The Max Dumping Rate variable shown in Fig. 12 is the interface variable that receives and sends values from DES to SD. The
exported DES variable (sender) represents the ideal rate. This variable
is impacted by the causaleffects loops in the SD model to deliver near
actual rates of production. The treated rate ends up in Net Dumping
Rate (receiver variable). After processed by Dumping Rate, the scope
is accumulated in Soil Dumped stock. Now, soil is ready for the next
stage, which is the spreading and thereafter compaction. The same procedures are followed for Net Spreading Rate and Net Compaction Rate
variables. The quality of compacted soil must be checked based on the

5.2.6. Identifying the interface variables


The interface variables are the interface points between the DES and
SD simulation models. Those variables are responsible to ensure that

Impact of
scope change

Excluded

Environmental
Impcat

Exogenous

Project
deadline

Safety

design standards before the nal release. Thus, the compacted soil is
stocked at Soil Compacted and Ready for Quality Check stock. The soil
that passes the compaction test is processed through the Productivity
Rate ow, and the faulty compacted soil is passed to the Rework
stock for further re-work to assure the required quality. The summation
of the ow's Productivity Rate and Rework Rate represents the actual
released work productivity. The Gantt chart of workow execution is
shown in Fig. 13. As observed in the chart, operations are scheduled
with 50% overlapping between scopes. The excavation scope is
not shown in this chart because excavation operation involved only
excavation and haulingdumping while the backlling involved
haulingdumping, spreading, compaction, and quality check.
Thus, the structure of the SD for excavation scope should be different
from the SD model of the backlling scope.

Operator skill
Secondary
error

Endogenous

Learning
effect
Overtime
impact
scope

Actual productivity
Project progress
Overtime
Weather
Schedule pressure
Quality
Forecasted
Error rate

Cash flow constraints

production
Project progress
Road condition
Grade of cut
Depth Cut
Workforce

Weather
impact

Equipment
maintenance

Soil type

Information
flow

Workforce shortage

Fig. 11. SD model boundary.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Rock11
Granular12
Moraine13
Rock21
Granular22
Moraine23
Rock31
Granular32
Moraine33

13

points that act to receive DES input from sender variables and deliver
them to receiver variables in the SD model.
6. Results and analysis

1550

3100
Time (hr)

4650

Due to the harsh cold weather at the dam construction location, the
project execution was planned in three phases. Each phase starts on
April and ends by November. In the project-planning phase, several scenarios can be considered to execute the project. The simulation tools
play an essential role in this context to show the discrepancy between
the different scenarios and assist the manager in making informed decisions. To test the developed hybrid simulation model, three scenarios
were considered as shown in Table 8. Each scenario was inputted to
the model and simulated for 6200 h. The initial normal project duration
was calculated by adding up the durations of the nine scopes computed
by DES models resulting in a total of 8504 h. In the planning stage, and
due to structural stability of the backll layers, it was considered that
every two successive activities were overlapping by 50% (the successive

6200

Fig. 12. Gantt chart of soil backll scope.

hybrid structure functions based on the selected data mapping between


the models. Referring to the SD model, Max Dumping Rate, Max Spreading Rate, and Max Compaction Rate shown in the red triangular shapes
(Fig. 13) are the interface variables. Those variables are the interface

Compaction and Rework


Cycle
Loading-Dumping
Operation

Rework Process
Rate Duration
Total Rework

Spreading Operation
Error Generation
Rate

Rework process
rate

Perceived
Rework Rate

Scope ID
Scope Size

percentage of
Dumped Soil

Percentage of
Spread Soil

Rework

Percentage of
Compacted Soil

Start Scope Task


Percentag Spreading

Error Rate

Spread Soil

Total Scope

Rework rate
<Impact of Fatigue
on Quality>

Avg, QP
Duration

Quality Check
Process Rate

Scope Task is Done C


Scope Task is
Done S

Start Task Flag


Scope Task is Done D
Start Scope Task
Scope Task is
Percentage
Active D
Dumping

Scope Task is
Active C

Dumping
Rate

Spreading
Rate

<Net Spreading
Rate>

<Net Dumping
Rate>

Max Compaction Rate

LHD_DES Model Input Zone

Spreading_DES Model Input Zone

<Total Soil Compacted


and Ready for Quality
Check>

Final Work
Completed

Total Project Work of Soil,


Hauled, Dumped, Spread, and
Compacted

Total Soil
Compacted and
Ready for Quality
Check

Max Spreading Rate

Perceived
Quality Rate

Productivity Rate

effect of schedule
pressure on
productivity

<Net Compaction
Rate>

Total Soil Dumped

Max Dumping Rate

Soil Compacted
and ready for
Quality Check
Compaction
Rate

Total Soil Spread

Average time

Scope Start
Percentage of
Compaction

Soil to
Compact

Soil Spread

Quality Process
Rate

Scope Task Start Flag C

Dumped Soil
Scope Task is
Active S

Soil to
Spread

Soil Dumped

Soil to Haul

Scope Task
Flag S

<Impact of Fatigue
on Productivity>

Compaction_DES Model Input Zone

Effect of Schedule
Pressure on
<Schedule Productivity
Lookup
Pressure>

<Total Scope>
Final Project
Completion
Duration

<Total Rework>
Total Work Not
Done

Schedule Index used to adjust


released productivity to be
same as the planned

Time Required
Forecasted
Productivity to
Complete

<Total Scope>

Project is Done

Time needed to
reached planned
productivity

Perceived
Productivity

Restart Fraction
Project was Done
Fraction Completed

Project Deadline
Schedule Pressure
Actual Final
Released
Productivity

<TIME STEP>
Time Available
Overtimne
Lookup

<Total Scope>

<Total Work Completed>

<Time>

Overtime Fraction

<Total Work Completed>


Predicted
Overtime

Impact of Fatigue
on Productivity

Eff. Fatigue on
productivity Lookup

Impact of Fatigue
on Quality

Effect of Fatigue on
Quality Lookup

Fig. 13. Work ow, schedule pressure and overtime feedback loops.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

14

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Table 8
Implementation scenarios.
Scenario 1 (Hybrid base case)

Scenario 2

Scenario 3

-Project planned duration = 4620 h


-Error in compaction quality = 0%
-No weather impact
-No schedule pressure impact
-No overtime and fatigue impact
-No depth of cut impact
-No adverse road condition

-Project planned duration = 4620 h.


-Error in compaction quality = 5%
-Weather impact
-Schedule pressure impact
-Overtime and fatigue impact
-No depth of cut impact
-No adverse road condition

-Project planned duration = 6000 h.


-Error in compaction quality = 5%.
-Weather impact
-Schedule pressure impact
-Overtime and fatigue impact
-No depth of cut impact
-No adverse road condition

the hauling roads and the capability of the eet to function in an optimum manner. The impact of precipitation rate on earthmoving projects
had been investigated thoroughly by El-Rayes & Moselhi [37]. The study
presented a quantication method of days and productivity losses due
to the amount of the rainfall. The outcome of the study has been utilized
in this research to predict days lost due to monthly precipitation
amount. The average monthly precipitation amount for the site location
(Sept Ile, Quebec) of the dam has been acquired from the National
Climate Data and Information Archive-Canada. The data represents the
average monthly precipitation amount in millimeter for 29 years
(19712002).
Schedule pressure: perceived maximum project productivity per
time unit (management policy factor) was calculated based on dividing
the project overall scope of work by the total planned project completion duration. Accounting for weather impact and other factors has increased the actual project duration beyond the planned duration. The
increase in the project duration is calculated at hourly step-time and is
divided by the remaining time from the original planned project duration. The resulted ratio is called schedule pressure ratio, which affects
work quality and productivity. From schedule pressure, the anticipated

activity starts as soon as 50% of the precedent activity is completed).


Based on that, the project completion duration was calculated as
4620 h (refer to Table 7).
Scenario (1) is concerned with generating a hybrid base case that
demonstrates the behavior in ideal situations. The planned project completion duration was set to 4620 h without considering inuences of
context variables and polices. The stochastic value of variables computed by DES models were input into the SD model through interface variable as shown in Fig. 14 and the model was simulated for 6200 h. The
base hybrid model structure and assumption considered for scenario
(1) should results in outputs (project duration and productivity) similar
to these computed using DES models since no surrounding factors are
considered. The hybrid base case (scenario 1) resulted in 4620 h project
completion duration, which is similar to the DES models computation.
In scenario (2), factors such as weather impact, quality level, schedule
pressure, and overtime were fed to the model to monitor their impact
on the simulation model's outcome. The factors considered in the
model are explained next.
Weather impact: weather and specically precipitation has great
impact on earthmoving projects. Rainwater has an adverse effect on

Compaction Rate

Dumping Rate
2,000

2,000

9 operations output (Table 7)

m3/h

m3/h

1,000

0
0

620 1240 1860 2480 3100 3720 4340 4960 5580 6200

620 1240 1860 2480 3100 3720 4340 4960 5580 6200

Time (hr)

Time (hr)
Spreading Rate

Productivity Rate

2,000

2,000

m3/h

m3/h
1,000

1,000

0
0

0
0

620 1240 1860 2480 3100 3720 4340 4960 5580 6200

620

1240 1860 2480 3100 3720 4340 4960 5580 6200

Time (hr)
dumping rate[ROCK11] : Run1
dumping rate[GRANULAR12] : Run1
dumping rate[MORAINE13] : Run1
dumping rate[ROCK21] : Run1
dumping rate[GRANULAR22] : Run1
dumping rate[MORAINE23] : Run1
dumping rate[ROCK31] : Run1
7
dumping rate[GRANULAR32] : Run1
dumping rate[MORAINE33] : Run1

Time (hr)

1
2

7
8
9

6
7

8
9

5
6

7
8

4
5

6
7

3
4

5
6

7
8

1
2

3
4

5
6

1
2

3
4

5
6

1
2

3
4

1
2

3
4

1
2

8
9

m3/hr
m3/hr
m3/hr
m3/hr
m3/hr
m3/hr
m3/hr
m3/hr
m3/hr

Fig. 14. DES model variables used by SD model.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

Scenario 2

Scenario 2

Scenario 3

0.5

25

0.4

20

0.3
0.2
0.1

Fractions

Scenario 1

5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0

15

Scenario 1

Scenario 3
1.5

15
10

0.5

5
0

0
0
345
690
1035
1380
1725
2070
2415
2760
3105
3450
3795
4140
4485
4830
5175
5520
5865

0
0
415
830
1245
1660
2075
2490
2905
3320
3735
4150
4565
4980
5395
5810

Fraction

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

Time (hr)

Time (hr)

b) Overtime required

a) Schedule Pressure
1
1

0.9

0.9

0.8

0.8
0

620 1240 1860 2480 3100 3720 4340 4960 5580 6200

620 1240 1860 2480 3100 3720 4340 4960 5580 6200

Time (hr)

Time (hr)

c) Impact of Fatigue on Productivity

d) Impact of Schedule Pressure on Productivity

Fig. 15. Schedule pressure and overtime.

overtime needed to nish the project is also computed on an hourly


basis. Overtime also affects productivity and its impact was quantied
based on a reference model from Hanna et al. [38]. The impact of the
schedule pressure on the three scenarios is shown in Fig. 15. The schedule pressure for scenario (1) as shown in Fig. 15.a is zero, while for scenario (2) the schedule pressure increases after 1900 h. In scenario (3),
where project planned duration was relaxed from 4620 h to 6000 h,
the schedule pressure was maximum at fraction of 0.4. Corresponding
to the increase or decrease in schedule pressure, the demand on overtime was increased or was decreased respectively, Fig. 15.b. The schedule pressure and overtime have adverse effects on productivity and this
can be noticed in Fig. 15.c&d.
Depth of the cut: one of the important factors that affect the excavator productivity is depth of the cut of the excavated soil. The equipment
operator faces difculty in lling the bucket of excavator in one pass
when the depth of the cut increases beyond certain depths [39]. As
the work progresses and the depth of the cut increases, the excavator
productivity will be affected.
Road surface condition: rutted and soft roads that have higher
rolling resistance may affect the hauling duration, and consequently

affect the eet productivity [38]. The scenario that represents the adverse site and access road conditions considers good road conditions
for the rst 16 h of the work execution, then the road surface gets deteriorated slowly until scheduled maintenance at start time 80 h.
Based on the above discussion, scenario (2) has resulted in a project
duration of 6020 h, which represents a 30.30% increase from the original
planned (4620 h). The productivity of the eet has dropped from an average of 1400 m3/h to an average of 1000 m3/h ( 28.5%). In scenario
(3), the planned project duration has been set to be 6000 h, and the
model computed the project duration as 5865 h. Scenario (3) has
been subject to the same conditions as in scenario (2) except that the
project duration was relaxed, thus decreasing the adverse impacts of
the schedule pressure and the overtime on productivity and quality
(Fig. 15.c&d). As a result, scenario (3) nished the project earlier than
in scenario (2). Fig. 16 summarizes the differences between the three
scenarios in terms of near actual productivity and accumulated work
performed. These three scenarios demonstrate importance of project
dynamics and their impacts on outcomes.
Follow-up reviews and an investigation of real project data conrmed the obtained results. The model successfully predicted what
Total Project Work of Soil, Hauled, Dumped, Spread, and Compacted

Actual Final Released Productivity


8M

2,000
1,500

6M

1,000

3 2

4M

2
2M

500
0

620

1240 1860 2480 3100 3720 4340 4960 5580


Time (hr)

Actual Final Released Productivity : Scenario 1


Actual Final Released Productivity : Scenario 2
Actual Final Released Productivity : Scenario 3

6200
m3/hr
m3/hr
m3/hr

620

1240

1860

2480

3100 3720
Time (hr)

4340

4960

5580

6200

"Total Project Work of Soil, Hauled, Dumped, Spread, and Compacted" : Scenario 1m3
"Total Project Work of Soil, Hauled, Dumped, Spread, and Compacted" : Scenario 2m3
"Total Project Work of Soil, Hauled, Dumped, Spread, and Compacted" : Scenario 3m3

Fig. 16. Productivity comparison between the three scenarios.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

16

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx

really happened in execution. The earthmoving operations of the dam


construction were completed in three years (3 years x7.5 months X
30 days x10-hour work). This makes the total actual project completion
duration approximately 6300 h. The hybrid simulation model predicted
the total project duration as 6020 h. The difference between the two gures was 4.44%. Therefore, it can be said that the hybrid simulation
model predicted the project completion duration and productivity
with high accuracy.
7. Conclusion
Traditional planning methods are based on deconstructing the project into individual activities, and then establishing the job logic
among those activities. This results in addressing the construction problem in fragmented and linear fashion. The purpose of any project model,
whether it is an SD model or a CPM network, is to strive to deliver an unbiased model that captures the likely behavior of a project during the
execution phase, and produces realistic estimates of productivity and
project completion duration. Generally, construction planers address
uctuating productivity by using contingency allowances to estimate
the most likely project completion duration. However, contingency estimates are subjective, and mainly rely on personal judgment. The previous research indicated that SD modeling is well suited to address the
dynamic nature of the interrelated project parameters at the strategic
level, while traditional methods are well suited for modeling the tactical
aspects of these parameters. Therefore, this paper has addressed those
concerns by presenting an innovative method that integrates the
CPM-network developed in DES simulation environment with the SD
model. The implementation infrastructure uses a discrete simulation
engine, CPM network, and an SD simulation engine.
The proposed method has been illustrated using two real cases from
the construction industry. A signicant difference in the outcomes between the static and dynamic models has been observed. The project
completion duration has thus noted to increase signicantly from the
planned duration by considering the effects of the dynamic impact on
the project execution.
Although the proposed method provides practical steps and tools to
develop a dynamic schedule, yet there are certain limitations and possible future enhancements to this hybrid system. For instance, the proposed method is based on the unidirectional interactions of variables
in DES and SD models. Bidirectional interactions of variables in simulation is has not been explored extensively yet. It is believed that simultaneous interaction between the hybrid simulation modules would be of
great value in better understanding of project behaviors, as well as producing estimates that are more accurate. In addition, different hybrid
model structures need to be tested and investigated to conrm the
method's applicability and accuracy.
SD is not fully utilized in construction modeling despite the potential
benets that it might bring to traditional planning methods. The coupling of traditional methods with SD is expected to provide a better understanding of project mechanisms resulting in more realistic schedule
and cost estimates. Traditional modeling techniques supply detailed information while the SD quanties the impacts of management policies
and strategies on project execution. The proposed hybrid method is expected to enhance the current practices in project planning and modeling, thus potentially beneting owners and clients worldwide. Better
accounting of uncertainty and dynamics experienced during actual
real-time projects, is the main feature of the proposed hybrid modeling
method.
Acknowledgment
The authors would like to acknowledge the nancial support of the
Natural Sciences and Engineering Research Council of Canada (NSERC)
(Grant no. RGPIN/4430-2010) for the present research.

References

[1] D.W. Halpin, L.S. Riggs, Planning and Analysis of Construction Operations, Wiley,
New York, 1992.
[2] A.P.C. Chan, D. Scott, A.P.L. Chan, Factors affecting the success of a construction project, J. Constr. Eng. Manage. 130 (1) (2004) 153155.
[3] S. Ogunlana, H. Li, F. Sukhera, System dynamics approach to exploring performance
enhancement in a construction organization, J. Constr. Eng. Manage. 129 (5) (2003)
528536.
[4] S. AbouRizk, S. Hague, An overview of the COSYE environment for construction simulation, Proc. of the 2009 Winter, Austin, TX, 2009, pp. 26242634.
[5] K. McCrimmon, C. Rayvec, An analytical study of the PERT assumptions, Oper. Res.
12 (1) (1964) 1637.
[6] A. Pritsker, C.E. Sigal, R.D.J. Harnmesfahr, SLAM II Network Models for Decision Support, Prentice Hall, Inc., Englewood Cliffs, N.J., 1989
[7] A. Rodrigues, J. Bowers, System dynamics in project management: a comparative analysis with traditional methods, Syst. Dyn. Rev. 12 (2) (1996)
121139.
[8] S. Lee, F. Pea-Mora, M. Park, Dynamic planning and control methodology for strategic and operational construction project management, Autom. Constr. 15 (1)
(2006) 8497.
[9] L. Rabelo, M. Helal, A. Jones, H. Min, Enterprise simulation: a hybrid system approach, Int. J. Comput. Integr. Manuf. 18 (6) (2005) 498508.
[10] S. AbouRizk, R. Wales, Combined discrete-event/continuous simulation for project
planning, J. Constr. Eng. Manag. 123 (1) (1997) 1120 (Online publication date:
1-Mar-1997).
[11] A. Laufer, R. Tucker, Is construction project planning really doing its job? A critical
examination of focus, role and process, Constr. Manage. Econ. 5 (3) (December
1987) 243 (serial online, Available from: Business Source Complete, Ipswich, MA.
Accessed May 10, 2014).
[12] D. Halpin, H. Jen, J. Kim, A construction process simulation web service, Simulation
Conference, 2003, Proceedings of the 2003 Winter Simulation Conference, New
Orleans, LA, 2, 2003, pp. 15031509.
[13] H. Alzraiee, O. Moselhi, T. Zayed, A hybrid framework for modeling construction operations using discrete event simulation and system dynamics, Proceedings of Construction Research Congress- ASCE- West Lafayette, IN, USA, 2012,
pp. 10631073.
[14] J. Lyneis, K. Cooper, S. Els, Strategic management of complex projects: a case study
using system dynamics, Syst. Dyn. Rev. 17 (3) (2001) 237260.
[15] Edward W. Davis, Project scheduling under resource constraintsHistorical review
and categorization of procedures, AIIE Trans. 5 (4) (1973) 297313.
[16] AbouRizk, Sawhney, Subjective and interactive duration estimation, Can. J. Civ. Eng.
20 (1993) 457470.
[17] H. Zhang, C.M. Tam, J.J. Shi, Simulation-based methodology for project scheduling,
Constr. Manage. Econ. 20 (8) (2002) 667678.
[18] J. Forrester, Industrial Dynamics, Productivity Press, Cambridge, MA, 1961.
[19] T. Abdel-Hamid, S. Madnick, Software Project Dynamics: An Integrated Approach,
Prentice-Hall, Englewood Cliffs, NJ, 1991.
[20] Y. Lee, M. Cho, S. Kim, Y. Kim, Supply chain simulation with discrete-continuous
combined modeling, Comput. Ind. Eng. 43 (12) (2002) 375392.
[21] M. Park, F. Pea-Mora, Dynamic change management for construction: introducing
the change cycle into model-based project management, Syst. Dyn. Rev. 19 (3)
(2003) 213242.
[22] K. Chahal, T. Eldabi, A Generic Framework for Hybrid Simulation in Healthcare,
2008. 14691477 (Miami, FL).
[23] J. Lyneis, D. Ford, System dynamics applied to project management: a survey,
assessment, and directions for future research, Syst. Dyn. Rev. 23 (2/3)
(2007) 157189.
[24] S.C. Brailsford, N.A. Hilton, A Comparison of Discrete Event Simulation and System
Dynamics for Modelling Health Care Systems, 2001.
[25] J. Morecroft, S. Robinson, Explaining puzzling dynamics: comparing the use of system dynamics and discrete-event simulation, Proceedings of the 23rd International
Conference of the System Dynamics Society, Boston, USA, 2005.
[26] B.P. Zeigler, H. Praehofer, T.G. Kim, Theory of Modeling and Simulation: Integrating Discrete Event and Continuous Complex Dynamic Systems, Academic
press, 2000.
[27] R. Fujimoto, Parallel simulation: distributed simulation systems, Proceedings of the
2003 Winter Simulation Conference: Driving Innovation, Dec 710, New Orleans,
LA, 2003, pp. 124134.
[28] J. Steinman, Multi-node test bed: a distributed emulation of space communications
for the strategic defense system, Pittsburgh 21 (3) (1990) 11111115.
[29] S. Fujii, H. Tsunoda, A. Ogita, Y. Kidani, Distributed simulation model for computer
integrated manufacturing, Proceedings of the 1994 Winter Simulation Conference,
Dec 58, Orlando, FL, 1994, pp. 946953.
[30] H. Alzraiee, Hybrid simulation for construction operations, (Ph.D. Dissertation)
Building, Civil and Environmental Dept., Concordia University, Montreal, QC,
Canada, 2013.
[31] P.G. Ioannou, J.C. Martinez, Project Scheduling Using State-based Probabilistic Decision Networks Proceedings, Winter Simulation Conference, Washington DC, USA,
1998, pp. 12871294.
[32] Ventana Systems Inc, Vensim, Massachusetts, Harvard, 2013.
[33] M. Marzouk, O. Moselhi, Object-oriented simulation model for earthmoving operations, J. Constr. Eng. Manage. 129 (3) (2003) 173182.
[34] FPC User's ManualVersion 3.0, Caterpillar Inc., Peoria, Ill., 2010

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

H. Alzraiee et al. / Automation in Construction xxx (2014) xxxxxx


[35] J. Martinez, P. Ioannou, R. Carr, State and resource based construction process simulation, Proceedings, ASCE First Congress on Computing in Civil Engineering, Washington, D.C., 1994, pp. 177184.
[36] J. Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex
World, McGraw-Hill, New York, USA, 2000.
[37] K. El-Rayes, O. Moselhi, Impact of rainfall on the productivity of highway construction, J. Constr. Eng. Manage. 127 (2) (2001) 125131.

17

[38] A.S. Hanna, C.S. Taylor, K.T. Sullivan, Impact of extended overtime on construction
labor productivity, J. Constr. Eng. Manage. 131 (6) (2005) 734739.
[39] G. Kannan, A methodology for the development of a production experience database for earthmoving operations using automated data collection, (PhD Thesis) Virginia Polytechnic Institute, 1999.

Please cite this article as: H. Alzraiee, et al., Dynamic planning of construction activities using hybrid simulation, Automation in Construction
(2014), http://dx.doi.org/10.1016/j.autcon.2014.08.011

You might also like