You are on page 1of 10

Using an interactive schedule simulation platform to assess and improve

contingency management strategies


Pei Tang
a,
, Amlan Mukherjee
a
, Nilufer Onder
b
a
Department of Civil and Environmental Engineering, Michigan Technological University, Houghton, MI 49931, United States
b
Department of Computer Science, Michigan Technological University, Houghton, MI 49931, United States
a b s t r a c t a r t i c l e i n f o
Article history:
Accepted 27 July 2013
Available online 23 August 2013
Keywords:
Decision strategy
Construction contingency management
Interactive schedule simulation
Improvement
An understanding of the dynamic interactions between resources and the management decisions that control
themis critical to effectively manage contingencies during the execution phase of a project. Hence, the objective
of this paper is to study each resource decision within the context of sequences of decisions. This allows the con-
sideration of dynamic interactions between resources, and designs control responses that account for uncertain
outcomes and that minimize contingencies. It is hypothesizedthat the success of achieving project objectives and
priorities is dependent on understanding ways of developing coherent management decision sequences. For
given project priorities and outcomes, such decision sequences are dened as strategies. This paper proposes
and implements a simulation based method to assess and improve alternative decision strategies. The theoretical
contribution of this research is that it develops foundational simulation based methods that support the study of
construction decision-making as a dynamic control problem.
Published by Elsevier B.V.
1. Introduction
Managing uncertainties during the execution of a construction
project presents a difcult dynamic decision-making problem. External
events such as inclement weather, labor disturbances, or unexpected
rework profoundly affect the expected outcomes of a project. The
specic time in the schedule when such events occur signicantly inu-
ences the extent of the impact. Inaddition, suchevents and the resource
decisions that are made to manage them have cascading impacts on
the schedule that may reinforce existing delays, and/or set in motion
unexpected feedbacks. For example, a decision to crash a specic
activity using over-time work, may cause the crews to intersect with
other spatially co-located activities and increase congestion on site.
This may not have the desired result of reducing the project duration.
An understanding of these interactions between resources and the
management decisions that control thembecomes critical to effectively
manage contingencies during the execution phase of a project.
A key to study this decision-making problemis to examine decisions
within the context of their immediate impacts as well as how
they impact the broader scope of the project and the feedbacks they
generate. Existing research studied the impacts of decisions on specic
activity operations such as the tunneling [31], earth moving [17,11],
and concrete paving [12]. For example, the impacts of broken dozers
on the productivity or duration of earth moving were evaluated. How-
ever, from the perspective of construction process, the impacts of
earth moving on other activities such as concrete paving have hardly
been studied. A construction process is consisted of activities to com-
plete the project, whose sequences are determined by the project
schedule. To bridge the gap, this paper suggests studying the decision-
making problems pertaining to the construction process. The impact
of each resource decision on the construction process is studied within
the context of sequences of decisions. It is hypothesized that the success
of achieving project objectives and priorities is dependent on under-
standing ways of developing coherent management decisionsequences.
The sequence of decisions is dened as decision strategies given project
priorities.
To study decision strategies, two approaches can be considered:
(i) by establishing general performance trends from a signicant num-
ber of historical construction decision data sets and case studies;
and (ii) by using simulation platforms that allow the exploration
of what-if scenarios and multiple deviations fromthe as-planned sched-
ule. While the rst approach is rooted in direct observation, the second
one is based on the modeling of the construction process. The rst is
limited by the data collection process that may prove to be practically
difcult. Inaddition, the case studies include a limitednumber of project
outcomes, therefore allowing only partial observability of the success or
failure of a decision. In comparison, simulation environments allow a
decision-maker to test different management decisions and explore
large spaces of simulated futures for behavior investigation. Although
validating such models can prove to be difcult, analyzing these
Automation in Construction 35 (2013) 551560
Corresponding author. Tel.: +1 906 4871952; fax: +1 906 4872943.
E-mail address: ptang@mtu.edu (P. Tang).
0926-5805/$ see front matter. Published by Elsevier B.V.
http://dx.doi.org/10.1016/j.autcon.2013.07.005
Contents lists available at ScienceDirect
Automation in Construction
j our nal homepage: www. el sevi er . com/ l ocat e/ aut con
simulated futures can provide signicant support to decision-making
and planning. As the empirical method and the simulation method
complement and support each other, testing alternative strategies in a
simulation environment takes the rst step towards the coupling of
two methods.
Therefore, this paper proposes and implements a simulation based
method that can be used to assess and improve the effectiveness of
alternative decision strategies given project uncertainties and priorities.
The rst part of the paper introduces the simulation platform and
the underlying research method. The proposed analysis is intended to
support a priori analysis of projects. The second part illustrates the
application of the method to improve strategies for managing a high-
way construction project. The theoretical contribution of this research
is that it develops foundational simulation based methods that sup-
port the study of construction decision-making as a dynamic control
problem.
2. Background
At the operational level, contingencies are managed by identifying
uncertainties associated with variables that dene critical activities,
such as cycle times in earthwork. Discrete event simulations are
used to simulate the impacts of uncertainties on the operations
[17,11]. Dynamic Planning Methodology (DPM) [22] uses the concepts
and methods in System Dynamics (SD) while taking advantage of
stochastic schedule analysis methods. It analyzes complex relationships
between production reliability, schedule pressure, and uncertainty in
activity durations to study change management and fast-tracking of
design/build construction projects [20,21,19,13,14,18]. Such methods
studied decisions by comparing variables such as cost and duration
rather than analyzing their sensitivity to decisions and uncertainties at
different points in the project. To study the decision-making problems
pertaining to the construction process, the gaps between existing
research and the objective of this research are:
Existing research studied the impacts of decisions on activity opera-
tions, neglecting the impacts of one activity on others. The objective
of this researchis to study the impacts of decisions onthe construction
process.
Existing research set up decision variables before simulation started.
This research requires a consideration of the decision-maker's
responses to the contingency situations such as schedule delay,
budget overrun, or critical path changes during the simulation process.
Existing research simulated decisions independently. This research
requires a consideration of the interactions between decisions.
The limitations of existing research are caused by the selection of
research subjects and simulation platforms. First, the existing research
studies isolated decisions and their impacts on project operations. It
neglects the impacts of decisions on project process as well as the feed-
backs and interactions between decisions and resources. To address
this gap, this paper proposes to study decision strategies, the sequences
of decisions, to manage project processes. This allows the consideration
of dynamic interactions between resources, and designs control
responses that account for uncertain outcomes, and minimize contin-
gencies. Secondly, the use of non-interactive simulation platforms
does not allow a decision-maker to interactively monitor and control
the construction process. The activity cycle diagram based simulation
platforms such as AP
3
[27], (SimCon) [6], PICASSO [29,30,28], and
SDESA [15] are able to simulate both construction operations and pro-
cess. They treat each activity as an instantaneous time point, making it
difcult to express multiple activities across multiple tasks which over-
lap in time. Once the simulation starts, the decision-maker cannot stop
the simulation and make resource decisions to manage the project as
the project evolves. Instead, this paper proposes an interactive simula-
tion platformthat is based on the time interval method. The time inter-
val method simulates as-planned schedules on an interval basis. During
each interval, the project uncertainty is generated, decisions are
made, the activity productivity is calculated, and the work progress
is predicted.
The simulation platform is called Interactive Construction Decision
Making Aid (ICDMA). ICDMA [32] is based on the merger of two sep-
arate simulations, the Virtual Coach [25] and the work of TempOral
Network with Activities and Events (TONAE) [1,2]. The Virtual Coach
was developed from the formal models that drive the situational sim-
ulations. TONAE is the representation and reasoning about construc-
tion management information of activities, events, and constraints. It
can be used to generate contingency plans, such as weather events,
and automatically estimate contingencies by exploring combinatorial
future spaces resulting from constraint violations. TONAE was applied
successfully to study the construction of a steel frame ofce [2]. How-
ever, the case study was performed in a non-interactive mode, which
did not allow a decision-maker to interactively monitor and control
the construction process. Based on Virtual Coach and TONAE, the co-
authors developed ICDMA, which is capable of emulating the con-
struction process, simulating the impacts of disruptive events and de-
cisions on the project process, and predicting the extreme scenarios of
a project.
3. Methodology: iterative process to assess andimprove contingency
management strategies
Fig. 1 describes an iterative process to assess and improve contin-
gency management strategies at the pre-construction phase of a pro-
ject, given project uncertainties and different project priorities. The
simulation based method itself is platform independent. Starting by
creating a simulation project from contractors' project documents, the
method tests a set of initial strategies on a simulation platform. As
each strategy presents multiple trade-offs, the best strategy is usually
not a nominal strategy, but instead a hybrid strategy that emerges
through the analysis of different nominal strategies given the project
priorities. The new strategies can be continuously generated and im-
proved by repeating the process. The method is designed so that it
can generate satisfactory strategies from any set of initial strategies
by iteratively exploring and learning from competing strategies.
Starting with a list of suitably chosen strategies, a decision maker can
theoretically reduce the number of iterations. Historical records of
similar projects are not required but can be used as a validity check.
This paper applies the method using ICDMA, which consists of the fol-
lowing steps:
For a given project, a decision-maker rst establishes a simulation
project in ICDMA (using a resource loaded as-planned schedule and
an expected project environment), and develops a series of contin-
gency management strategies.
In the second step, the decision-maker implements each strategy to
manage the simulation project in ICDMA for a number of times in
the given project environment.
In the third step, the decision-maker compares and identies compet-
ing strategies which are efcient in achieving desired objectives such
as a low project cost or a short project duration. If no outstanding
strategy is identied, restart fromstep one to step three using a differ-
ent series of strategies.
In the fourth step, the decision-maker checks whether the out-
standing strategies meet the desired qualications. If they do,
these strategies are considered as satisfactory strategies. If not,
the decision-maker moves to the next step and investigates the
existence of general trends and patterns of outstanding strategies.
In the case of no patterns identied, restart from step one with dif-
ferent strategies.
In the fth step, the decision-maker generates new hybrid strategies
by learning from and combining the trends and patterns from the
fourth step.
552 P. Tang et al. / Automation in Construction 35 (2013) 551560
For continuous improvement, the decision-maker iterates the above
process with newly generated hybrid strategies until satisfactory
strategies are identied.
4. Simulation platform description
ICDMA is a general-purpose interactive simulation with the objec-
tive of exposing a decision-maker to rapidly unfolding events and the
pressure of decision making. The setup of a project in ICDMA requires
a resource loaded as-planned schedule and project environment
(Fig. 2). ICDMA emulates and advances the as-planned schedule on an
interval basis, which can be a day, week, or month unless otherwise
specied. The decision-maker takes on the role of a construction
manager to complete the project involving differing project priorities.
During each interval, the decision-maker is presented with random
external events that force the simulated project to deviate from its as-
planned schedule. The decision-maker has to respond to the events
by making resource related decisions. The consequences from the
decisions result in new scenarios for the decision-maker to respond.
This process continues until the completion of the simulated construc-
tion project.
4.1. Underlying algorithm
ICDMA provides a rich, state-based representation of construction
information and constraints that drives the construction management
domain. It is important to recognize that ICDMA is not a discrete event
simulation of a construction operation. Instead, it is a continuous time
advance simulation of the construction schedule comprising of an
emulator and a random event generator. The emulation engine [23]
enables the simulation to advance from one time point to the next
contiguous time point in its as-planned schedule, rather than advancing
from one event to the next event. The inference technique and random
event generator [24] generate and simulate the random events that
disrupt the schedule. The two techniques together allow a decision-
maker to stop in the middle of the simulation to enter different resource
decisions to manage the project. A third critical component of ICDMA,
the querying algorithm [2], enumerates Monte Carlo samples of the
combinatorial future spaces in dynamic risk construction scenarios as
situations evolve. It allows ICDMA to explore a number of future project
outcomes at any time point during the simulation with consideration of
the impacts of decisions made prior to that time point and the risks in
the future. For details about how each technique and algorithm works,
refer to the above papers.
4.2. Inputs in ICDMA
The resource loaded as-planned schedule provides a baseline to
complete the project, which requires the following information
(Fig. 2) to set up: (i) a list of activities and the estimated durations to
complete the project; (ii) material, labor, and equipment usages for
each activity; (iii) unit price of labor, equipment, and material;
(vi) logical relationship and constraints between activities to create
the as-planned schedule.
Project environment is dened as the expected disruptive events
specic to a project. Each event consists of the precondition,
postcondition, and probability. For example, the concrete testing fails
at a rate of 5%, and if that occurs, the concrete paving activity will stop
Fig. 1. Iterative process to assess and improve contingency management strategies.
553 P. Tang et al. / Automation in Construction 35 (2013) 551560
for one day to perform an extra test or tests. The pre-condition of
the event is the concrete paving activity being ready for operation. The
postcondition is that the concrete paving activity stops for one day.
The occurrence probability of the event is 5%. The precondition,
postcondition, and probability are set up by setting the values of
corresponding variables. In the example of concrete testing, the post-
condition is dened by assigning the variable peproductivity to zero
and duration to one for the concrete paving activity.
4.3. Outputs from ICDMA
During each simulation process, a large amount of structured
simulated as-built data (Fig. 2) is collected. They include:(i) project
outcomes, describing the cost and schedule condition of the project;
(ii) decision information, recording the material, labor, and equipment
allocation and re-allocation; (iii) a record of disruptive events; and
(vi) querying results, including a number of predicted project total
costs and completion dates during the simulation. The number of
predictions is determined prior to the beginning of the simulation.
5. Case study: highway reconstruction project
5.1. Project description and data collection
The case study involves a ten-mile highway concrete pavement re-
construction project in Southeast Michigan. This paper studied the re-
construction of the east bound section of a major interstate highway
in 2009. This section demonstrated the application of the proposed
method to improve strategies from a series of randomly developed
contingency management strategies. Existing project documents was
used to build the resource loaded as-planned schedule. And the record
of construction history was investigated to identify the actual project
environment.
5.1.1. As-planned information
The following information was collected to develop a resource
loaded as-planned schedule:
Fourteen major activities and associated durations to complete the
east bound section (Table 1). The information was acquired from the
progress schedule (MDOT Form 1130), a document submitted to the
Michigan Department of Transportation (MDOT) by the contractor
before the project started.
Material, labor, and equipment usages for each activity. The bidding
documents provided the material quantity take-offs. Labor and equip-
ment usage was determined by the material quantity and the labor
crew productivity in RS-Means, heavy construction cost data 2009
[26].
Material, equipment, and labor unit prices were obtained from the
RS-Means.
Actual spatial constraints between activities fromthe main contractor
(Table 1).
5.1.2. As-built information
In Michigan highway projects, MDOT requires the use of a software
called FieldManager
TM
on all their construction and rehabilitation
contracts. Inspectors (on behalf of MDOT) use the software to record
daily general site information, contractor personnel and equipment,
and postings of material quantities. The daily record fromFieldManager
TM
was analyzed to identify the disruptive events, their probabilities, and
their impacts on the project. Details about the information processing
can be found in the co-authors' previous work [4]. Bad weather, labor
crew problems (equipment failure or worker illness), and concrete
testing failure were found to be the most inuential events responsible
Fig. 2. Simulation set-up and data collection.
554 P. Tang et al. / Automation in Construction 35 (2013) 551560
for project delays (Table 2). The as-built disruptive event information
was used to simulate the actual project environment in ICDMA. In
addition, the record identied the as-built cost and completion date
as $20,277,970 and 129 working days for the case study project.
The as-built data can be compared against the simulated as-built
performance to verify the accuracy of ICDMA.
5.2. Step one: project set-up and strategy development
The as-planned information and disruptive event information were
input into the database to develop the simulation project. In addition,
another three variables were determined for this specic project:
(i) indirect cost was determined to be 25%of the direct cost; (ii) the sim-
ulation granularity was set to be one day interval, implying the
simulation advances the project schedule by one day at each simulation
step until the completion of the project; and (iii) 1000 complete future
queries every 15 days fromthe beginning of the project. For application
to real projects, the decision-maker candetermine appropriate values to
reect the reality. As a result, the project was scheduledto be completed
in 106 working days with an estimated cost of $28,111,603. The differ-
ences between the as-planned and the as-built costs were caused by
the use of different resources and resource unit prices. The as-built
cost used contractors' prices and the as-planned cost used RS-Means'
prices. As the expected and actual activity durations and the quantity
of work to be completed were available, it is assumed that the project
cost estimates from RS-Means still provide a useful benchmark for
strategy comparison. Fig. 3 shows the interface of the highway project
in ICDMA.
Once the project and uncertain environment were established, deci-
sion strategies were developed for testing and improvement. Each deci-
sion strategy consists of policies that determine the ordering, allocation,
and reallocation of space, material, labor and equipment resources.
These resource policies generate the sequences of decisions for different
project priorities. For example, in a project where timely completion is
critical, a nominal strategy will prioritize duration over cost. As a result,
a sequence of possible decisions will be made keeping an eye
on controlling project duration. Each resource allocation policy directly
affects activity productivity. For example, if the only paver on a site
breaks down, the productivity of paving activity will be reduced to
zero. Other unrelated activities' productivity remains unchanged. In
addition, a cost is inicted every time a resources policy is applied. To
understand howICDMAdeals with events such as disruptions and deci-
sions, please refer to the co-author's previous work [2325,2,7].
Realistic applications can start with strategies that reect the reality
of the project being simulated. However, for the sake of this discussion,
the critical take away is how the proposed method can continuously
improve strategies by learning from and rening the initial strategies.
For illustration purposes, this research considers three strategy cate-
gories: one control strategy, three passive management strategies,
and three aggressive management strategies. Passive management
is based on the ad-hoc response to unexpected events. Aggressive
management aims to anticipate the future and prepare contingency
plans that minimize adverse consequences of unexpected disrup-
tions to the project schedule. Strictly speaking, the control strategy
is a special case of passive strategies. It manages the schedule by tak-
ing the minimumnumber of actions to deal with interruptions and is
used as a baseline strategy to contrast with other strategies.
When applying the decision strategies, the decision-maker has to
decide what time to apply the resource policies. When using passive
strategies, the question is how much of a project delay can decision-
makers tolerate before they take action? It is hypothesized that the
appropriate decisiontime plays animportant role inmaximizing project
outcomes. This paper develops initial strategies by using the tolerance
of schedule delay (X) or acceleration (Y). Assuming that the tolerance
of schedule variance is 5% of the total project duration, the maximum
value of (X) and (Y) is 5 days (106 days 5% = 5 days) for this high-
way reconstruction project. The strategies are:
Control Strategy: No actions are taken when the project falls behind.
Resource allocation policies: (a) labor policy: no extra workers are
replaced in any case; (b) equipment policy: equipment is xed the
next day if it breaks down and never use extra equipment; (c) space
policy: critical activities are prioritized when allocating space on site.
Three passive strategies: dened as a Catch Up_X Strategy where X is
the tolerance of schedule delay (X = One, Three, Five). For Catch
Up_X Strategy, resource allocation policies are applied to catch up
with the schedule every time the project is X days behind the as-
planned schedule. When approaching the end of the project, Catch
Up Strategies reduce the tolerance of schedule delay to one day
behind for timely completion. Resource allocation policies: (a) labor
Table 1
Project information to build resource loaded as-planned schedule.
NO. Activity description Duration Precedence
activities
Spatial distance
between activities
1 Strip topsoil 10 days n/a
2 Remove concrete
pavement
30 days 1 0.5 to 1 miles between stripping
topsoil and concrete
pavement removal
3 Grade subbase 26 days 2 n/a
4 Install drainage 18 days 2,3 0.5 to 1 miles between installing
drainage and concrete pavement
removal; 1 to 3 miles between
installing drainage and grade
subbase
5 Place open graded
drainage course
(OGDC) mainline
18 days 2,3,4 1 mile between grade subbase
and placing OGDC mainline
6 Pave east bound
mainline
32 days 5 1 mile between placing OGDC
mainline and paving east bound
mainline
7 Place OGDC ramps
and gaps
8 days 4,5,6 3 to 5 miles between paving east
bound mainline and placing
OGDC ramps and gaps
8 Pave east bound
gaps and ramps
9 days 7 n/a
9 Place gravel
shoulder
4 days 4 2 to 3 miles between paving east
bound gaps and ramps and
placing gravel shoulder
10 Slope grading and
restoration east
bound
26 days 9 0.5 miles between placing gravel
shoulder and slope grading and
restoration east bound
11 Stripe to open
pavement east
bound
3 days 9 10 miles between placing gravel
shoulder and striping to open
pavement east bound
12 Relocate barrier
wall
10 days 10 10 miles between striping to
open pavement east bound and
relocating barrier wall
13 Re-stripe west
bound
3 days 12 n/a
14 All lanes open 1 days 12,13 n/a
Table 2
Project environment.
Disruptive
event
Precondition Postcondition Probability
Bad
weather
N/A Productivity reduces by half for each
activity on the bad weather day.
0.20
Equipment
failure or
worker
sick
N/A Random labor(s) is(are) sick or
equipment break(s) down when the
event occurs. The labor crew's
productivity reduces according to the
weight of the labor or equipment in the
labor crew.
0.12
Concrete
testing
failure
Paving
concrete is
being ready.
Productivity of paving activity reduces to
zero for one day waiting for clearance of
new test.
0.05
555 P. Tang et al. / Automation in Construction 35 (2013) 551560
policy: extra workers are hired and replaced in cases of illnesses and
project delay; (b) equipment policy: equipment is xed by the
mechanics immediately and extra equipment is used in case of
schedule delay; and (c) space policy: critical activities are prioritized
when allocating space on site.
Three proactive strategies: dened as a Crash_Y Strategy where Y is
the tolerance of schedule acceleration (Y = One, Three, Five). The
decision-maker assesses future risks and applies resource allocation
policies to stay Y days ahead of the as-planned schedule. When
approaching the end of the project, the Crash_Y Strategy reduces the
desire of staying Y days ahead of the as-planned schedule to one day
ahead for timely completion. Resource allocation policies: (a) labor
policy: extra workers are hired and replaced in cases of illnesses or
to expedite the schedule; (b) equipment policy: equipment is xed
by the mechanics immediately and extra equipment is used to
expedite the schedule; and (c) space policy: critical activities are
prioritized when allocating space on site.
5.3. Step two: simulation experiment and validation
Each strategy was implemented for thirty ve runs to meet the
minimum requirement of statistical analysis. For the sake of uniformity
and given the experimental nature of this research, one of the authors
was the single decision-maker running all the simulations. The simulated
as-built data at each decision control point was collected. In the case of
no disruptive events, the simulated as-built total cost was $28,111,603
and the duration was 106 days, which veried that the simulation plat-
form can emulate the as-planned schedule. In addition, the simulated
as-built results were compared to the actual as-built results for valida-
tion. The actual as-built record showed that it took 129 working days
to complete the project with a cost of $20,277,970. In the simulation
experiment, the Control Strategy took an average of 127.91 days
(Fig. 4) to complete the job, which was comparable to the as-built dura-
tion (129 days). In addition, the Control Strategy, Catch Up Strategies,
and Crash Strategies all completed the project within 112% of the as-
built cost.
5.4. Steps three and four: strategy assessment and patterns identication
5.4.1. Strategy assessment regarding project cost and duration
A total of 245 instances of simulated as-built project costs and
completion dates were collected for the seven strategies. Fig. 4 shows
the average total cost and duration for each of the strategies. It is
observed that the Control Strategy completed the project at a higher
cost and longer durationthan others. Different strategies might produce
different costs. Now, the question is whether there is statistical evidence
to support the observation.
The one way ANalysis Of VAriance (ANOVA), an inferential statistical
test, was used to examine if the cost means were the same in the group.
It requires equal variances in the independent identically distributed
data [3]. The results are assumed to be independent and identically
distributed for each strategy when the number of experiments is large
because each simulation run was independent. The Levene's test was
used to test variance similarities, whose null hypothesis is that equal
variance exists. If this assumption turns out to be not applied, the
BrownForsythe test is used as an alternative version of F statistic. All
the tests were implemented in the Statistical Package for the Social
Sciences (SPSS). The test variables were 35 instances of simulated as-
built total costs from each strategy. The null hypothesis and alternative
hypothesis of ANOVA are:
Null hypothesis H
0
: (Control Strategy) = (Catch Up_Five
Strategy) = (Catch Up _Three Strategy)=(Catch Up_One Strat-
egy) = (Crash_Five Strategy) = (Crash_Three Strategy) =
(Crash_One Strategy), where represents the average project
cost produced by each strategy;
Alternative hypothesis H
a
: at least one of the means is different
from the rest;
The signicance level: = 0.05.
The Levene's test showed that there was not enough evidence
(0.509 N 0.05) to reject the null hypothesis (project's total costs of
seven strategies have equal variance). Therefore, as determined by
one-way ANOVA (F(6,238) = 34.807, p = 0.000), the null hypothesis
Fig. 3. Interface of simulated highway reconstruction project in ICDMA.
556 P. Tang et al. / Automation in Construction 35 (2013) 551560
was rejected, indicating that alternative strategies had different total
costs.
The next question was, which strategies outperformed others? The
Post Hoc test was performed in SPSS to identify the differences between
one group andthe rest. The test results showedthat the Control Strategy
had a different average total cost from other strategies, which afrmed
the initial observation and the results of ANOVA test. In addition, the
Post Hoc test also indicated that the Crash_Three Strategy and Catch
Up_Five Strategy outperformed other strategies in cost management
(the signicance value of the Crash_Three Strategy against the Catch
Up_Five Strategy was 0.122 N 0.05, and the values of the Crash_Three
Strategy against other strategies were less than 0.05).
5.4.2. Strategy assessment using the querying algorithm
Fromthe beginning of the project, the querying algorithmpredicted
1000 total costs and completion dates every 15 days through the entire
simulation process. The querying results can be used as a metric to
evaluate how project uncertainties evolve among different strategies.
The querying results were analyzed in boxplots by SPSS 16.0. Boxplots
[5] are an excellent tool for detecting and illustrating location and
variation changes between different groups of data. It identies the
inter-quartile range (the differences between the 25th and 75th
percentile), the median, and the extreme points (within 1.5 times
the inter-quartile range from the upper or lower quartile). Points
that are outside of the ends of the lines extending from the inter-
quartile are potential outliers. Figs. 5 and 6 present the distributions
of predicted total costs and completion dates at different time points
through the process of the project for the Catch Up_Five Strategy,
Crash_Three Strategy, and Control Strategy. The outliers were
not shown because they were such a small quantity. The x-axis
represents the simulation time. The y-axis represents the projected
total costs or durations. For example, for the Catch Up_Three Strategy
on Day 1 in Fig. 5, the project is predicted to be completed at an average
cost of 30.2 million dollars. There is a 25% chance that the predicted
costs are less than 28.8 million dollars (the 25th percentile), and a
25%chance that the predictedcosts are greater than30.5 million dollars
(the 75th percentile). Excluding the outliers, the maximum and
minimumpredicted total project costs are 31.5 and 28.9 million dollars
respectively. The graphs are interpreted qualitatively by analyzing the
shape and positional shift of boxplots as follows:
Fig. 4. Simulated as-built total costs and durations for seven strategies.
TimePoints
Day121 Day106 Day91 Day76 Day61 Day46 Day 31 Day 16 Day 1
P
r
o
j
e
c
t
e
d
T
o
t
a
l
C
o
s
t
32
31
30
29
Control Strategy
Crash_Three Strategy
Catch Up_Five Strategy
Strategies
Fig. 5. Predicted completion costs through the construction process.
557 P. Tang et al. / Automation in Construction 35 (2013) 551560
A shift to the lower right position over time indicates improving
performance and that the most likely, minimum and maximum
completion costs are decreasing over time.
A shift to the upper right position over time indicates worsening
performance and that the most likely, minimum and maximum
completion costs are increasing over time.
An increase in the inter-quartile range over time indicates increasing
uncertainty in the project with the possibility of threshold events
that may govern the project outcome signicantly.
For example, in Fig. 5, the boxes of the Control Strategy moved in the
upper right direction, indicating ineffective management as the comple-
tion costs were predicted to increase over time. The Catch Up_Five
Strategy outperformed the Crash_Three Strategy in maintaining lower
predicted average cost in the rst 46 days. In Fig. 6, the shift of boxplots
to the upper right direction implied that for the Control Strategy, the
completion durations increased over time. The Crash_Three Strategy
maintained lower predicted average durations than the Catch Up_Five
Strategy over the entire simulation process. After Day 61, the inter-
quartile range reduced more for the Crash_Three Strategy than the
Catch Up_Five Strategy, implying that the Crash_Three Strategy had a
better control over duration deviations than the Catch Up_Five Strategy
when approaching the project completion.
5.5. Step ve: new strategy generation and evaluation
Assessment results from step three and four were summarized as:
(i) the Control Strategy had the worst performance in managing contin-
gency situations; (ii) generally, Crash Strategies outperformed Catch Up
Strategies in managing completion durations; (iii) the Crash_Three
Strategy and Catch Up_Five Strategy were the most cost efcient strate-
gies; (vi) the Catch Up_Five Strategy maintained lower predicted
average costs than the Crash_Three Strategy in the early half of the
construction phase; and (v) the Crash_Three Strategy had a better
control of duration variance than the Catch Up_Five Strategy in the
late half of the construction phase. In addition, differences between
three Crash Strategies (Fig. 4) indicated that an optimum Y might
exist for Crash Strategies to achieve the best cost management.
By combining the strategy assessment results, newhybrid strategies
are generated. For example, the Catch Up_Five Strategy can be used in
the early stage of the project to save interests because of the lower
expenditures in the early phase. If timely completion is critical, Crash
Strategies can be used, especially in the later construction phase. The
number of ways to combine assessment results is determined by the
decision-maker with regard to the specic objectives. In this study,
the satisfactory hybrid strategies were expected to be cost and duration
efcient while minimizing the deviations of predicted costs and
durations through the construction process. Therefore, a new Hybrid
Strategy was created with the expectation to inherit the advantages of
the Catch Up_One Strategy and Crash_Three Strategy. Another two
Crash Strategies were created to further test the hypothesis that an
optimum Y exists in the family of Crash Strategies to achieve better
project outcomes.
Hybrid Strategy: in the early half of the construction phase, imple-
ment Catch Up_Five Strategy; in the remaining construction phase,
use Crash_Three Strategy.
Crash_Y Strategy (Y = Two, Four), where Y is the tolerance of sched-
ule acceleration. The decision-maker assesses future risks and applies
resource allocation policies to stay Y days ahead of the as-planned
schedule. Whenapproaching the endof the project, the CrashStrategy
reduces the desire of staying Y days ahead of the as-planned schedule
to one day ahead for timely completion. Resource allocation policies:
(a) labor policy: extra workers are hired and replaced in cases of
illnesses or to expedite the schedule; (b) equipment policy: equip-
ment is xed by the mechanics immediately and extra equipment is
used to expedite the schedule; (c) space policy: critical activities are
prioritized when allocating space on site.
The two-sample t-test was used to examine if the means of simulated
as-built total costs were different between the new strategies and the
initial strategies. The Hybrid strategy, Crash_Two Strategy, and
Crash_Four Strategy were compared against the Crash_Three Strate-
gy, which had the lowest simulated as-built project cost. The null
hypothesis is that two strategies have the same mean cost. The alter-
native hypothesis is that the cost means of strategies are different.
The signicance level was set as = 0.05. The two-sample t-test
was performed assuming both equal and unequal variances. The
Levene's test was used to test the equal variances in data. All the
signicance values in the Levene's test were greater than 0.05,
indicating that there was not enough evidence to reject the null hy-
pothesis (equal variance exists). Therefore, the results of the t-test
that assumed equal variances were used.
When comparing the simulated as-built total costs of the
Crash_Three Strategy and Hybrid Strategy, the signicance value
of t-test was 0.056 (N0.05), thus not providing evidence to reject
the null hypothesis and indicating that the Crash_Three Strategy
TimePoints
Day121 Day106 Day91 Day76 Day61 Day46 Day 31 Day 16 Day 1
P
r
o
j
e
c
t
e
d
T
o
t
a
l
D
u
r
a
t
i
o
n
135
130
125
120
115
110
105
Control Strategy
Crash_Three Strategy
Catch Up_Five Strategy
Strategies
Fig. 6. Predicted completion dates through the construction process.
558 P. Tang et al. / Automation in Construction 35 (2013) 551560
($29,269,459) and Hybrid Strategy ($29,420,561) had a comparable
amount of total project costs. Comparing the average accumulated
as-built cost during the simulation process, Fig. 7 shows that
the Hybrid Strategy inherited the properties of the Catch Up_Five
Strategy in maintaining lower costs in the rst half of the construc-
tion. After that, the Hybrid Strategy inherited the property of early
project completion from the Crash_Three Strategy. In addition,
the t-test showed that the Crash_Two Strategy and Crash_Three
Strategy had the same as-built total cost (0.475 N 0.05), but the
Crash_Four Strategy and Crash_Three Strategy had a different as-
built total cost (0.035 b 0.05). Therefore, aggressive strategies that
attempt to stay two or three days ahead of the as-planned schedule
were identied as the most cost efcient strategies (Fig. 8). The
evaluation of three new strategies conrmed that the proposed
method generated better strategies from the original seven strate-
gies. For continuous improvement, the analysis process can be iter-
ated to generate new better strategies using the Hybrid Strategy
and Crash_Two Strategy.
6. Discussion
This paper proposes a simulation based method that can be used to
assess and improve the effectiveness of alternative decision strategies
given project uncertainties and priorities. In the case study, the method
was implemented to design better strategies for managing a highway
construction project. By testing and analyzing seven initial contingency
management strategies, new improved decision strategies were gener-
ated on the criteria of low project costs, early project completion dates,
and robust risk control. The hypothesis is validated that appropriately
designed decision strategies can better manage the projects. When
planning complex projects with limited historical information to
depend on, the proposed simulation and analytical methods can be
used for speculative exploration of various contingency management
strategies by using information no more than project documents, as
illustrated by the case study.
The reliability and validity of the results from using the illustrated
method are worth discussing. Given the speculative nature of the
proposed simulation method, inductive validation methods that strive
to minimize the difference between model outcomes and system
performance do not apply. At any given time, the proposed method
simulates a distribution of project outcomes. Any parameter character-
izing this distribution cannot be fairly compared with an actual project
realization, as the latter is a single instance from the actual distribution
of outcomes that cannot be observed [16]. In addition, as complex
construction projects are non-prototypical and subject to unique
circumstances, it is likely that the general best practice strategies
across all projects cannot be inductively identied.
A plausible approach lies in the project specic solution that is based
in Popperian falsication. The proposed method is such an approach
that generates satisfactory strategies based on how a project responds
to different strategies. As new evidence becomes available, the strate-
gies are further conditioned or falsied, leading to an iterative continu-
ous improvement process. This strikes a balance between employing
experimental simulation approaches and longitudinal empirical obser-
vation. Dreyfus [8] establishes that for complex non-prototypical
domains such as construction this speculative approach is more
valid in developing an understanding of how overarching principles
and policies dene system behavior. In addition, Genova [10] defends
the need and importance of such speculative research, arguing
that validation in such cases requires that the proposed methods be
(i) illustrated to stakeholders who are involved and willing to
use them, and (ii) reproducible and open to improvement by other
academic researchers.
Although the results are fundamentally project specic, the proposed
method provides a general way to generate, test, and improve alternative
strategies at the pre-construction phase of a project. By involving the
decision-maker into the interactive decision-making process, the paper
generated better contingency management practices using the analytical
methods with ICDMA. This current laboratory experimental research will
lay the foundation for future work that intends to directly involve stake-
holders. There are various limitations to this research at this current
point: (i) a knowledge of the relationship between resource policies and
activity's productivity needs to be established; (ii) strategy assessment
metrics, strategy assessment outcomes, criteria to determine satisfactory
strategies, and strategy re-generation demand more standardization
and less arbitrariness; and (iii) a disruptive event database for different
types of construction projects needs to be built.
7. Conclusion
As a departure fromexisting research that studies isolated decisions,
this paper studies decision strategies to manage construction process.
A method has been designed to generate satisfactory strategies at the
pre-construction phase from any series of nominal decision strategies.
This is a supplement to the traditional contingency management [9]
which budgets contingency money to address emergencies. As each
project is unique, the goal of such a decision-making process is to
further the comprehension of howthe project and the specic problem
at hand behaves and to improve the decision strategies through an
iterative continuous improvement process, rather than merely
predicting the exact project outcomes. Practically speaking, the
Fig. 7. Accumulated as-built costs through the construction process.
559 P. Tang et al. / Automation in Construction 35 (2013) 551560
proposed method lays the foundation to design robust construction
schedules accompanied by effective management strategies that can
be used to control the project during its execution.
Acknowledgment
This work is supported by the Michigan Department of Transporta-
tion under Contract No. #080741 and the National Science Foundation
under Grant No. SES-0624118. Any opinions, ndings and conclusions
or recommendations expressed in this material are those of the authors
and do not necessarily reect views of the Michigan Department of
Transportation and the National Science Foundation.
References
[1] G. Anderson, N. Onder, A. Mukherjee, Expecting the unexpected: representing,
reasoning about, and assessing construction project contingencies, Simulation
Conference, 2007 Winter, 2007, pp. 20412050.
[2] G.R. Anderson, A. Mukherjee, N. Onder, Traversing and querying constraint driven
temporal networks toestimateconstructioncontingencies, AutomationinConstruction
18 (6) (2009) 798813.
[3] G.G.C. Casella, R.L. Berger, Statistical Inference, 002 edition Duxbury Press, 2001.
[4] D. Cass, P. Tang, A. Mukherjee, Managing construction operations to minimize
greenhouse gas emissions, The 2nd International Conference on Transportation
Construction Management, February 7-10, Orlando, FL, USA, 2011.
[5] Chambers John, William Cleveland, B. K., P. Tukey, Graphical Methods for Data
Analysis. Wadsworth, 1983.
[6] N. Chehayeb, S. AbouRizk, Simulation-based scheduling with continuous activity
relationships, Journal of Construction Engineering and Management 124 (2) (1998)
107115.
[7] C.S. Dossick, A. Mukherjee, E.M. Rojas, C. Tebo, Developing construction manage-
ment events in situational simulations, Computer-aided Civil and Infrastructure
Engineering 25 (3) (2010) 205217.
[8] S. Dreyfus, Richard bellman on the birth of dynamic programming, Operations
Research 50 (1) (2002) 4851.
[9] D. Ford, Achieving multiple project objectives through contingency management,
Journal of Construction Engineering and Management 128 (1) (2002) 3039.
[10] Gonzalo Genova, Is computer science truly scientic, Communication of the ACM 53
(7) (2010) 3739.
[11] D. Hajjar, S. AbouRizk, Unied modeling methodology for construction simulation,
Journal of Construction Engineering and Management 128 (2) (2002) 174185.
[12] M. Hassan, S. Gruber, Simulation of concrete paving operations on interstate-74,
Journal of Construction Engineering and Management 134 (1) (2008) 29.
[13] S. Lee, F. Pena-Mora, Understanding and managing iterative error in change cycle,
Systems Dynamics Review 23 (1) (2007) 5360.
[14] S.H. Lee, F. Pena-Mora, M. Park, Dynamic planning and control methodology for
strategic and operational construction project management, Automation in
Construction 15 (1) (2006) 8497.
[15] M. Lu, Simplied discrete-event simulation approach for construction simulation,
Journal of Construction Engineering and Management 129 (5) (2003) 537546.
[16] J. Martinez, Methodology for conducting discrete-event simulation studies in
construction engineering and management, Journal of Construction Engineering
Management 136 (2010), (Special Issue: Research Methodologies in Construction
Engineering and Management 316).
[17] J. Martinez, P. Ioannou, General-purpose system for effective construction simula-
tion, Journal of Construction Engineering and Management 125 (4) (1999) 265276.
[18] I.A. Motawa, C.J. Anumba, S. Lee, F. Pena-Mora, An integrated system for change
management in construction, Automation in Construction 16 (2007) 368377.
[19] M. Park, F. Pena-Mora, Dynamic change management for construction: introducing
the change cycle into model-based project management, Systems Dynamics Review
19 (3) (2003) 213242.
[20] F. Pena-Mora, M. Li, Dynamic planning and control methodology for design/build
fast-trackconstructionprojects, Journal of ConstructionEngineering andManagement
127 (1) (2001) 117.
[21] F. Pena-Mora, M. Park, Dynamic planning for fast-tracking building construction
projects, Journal of Construction Engineering and Management 127 (6) (2001)
445456.
[22] F. Pena-Mora, M. Park, S. Lee, M. Li, and M. Fulenwider, (Issued March 25, 2008).
Dynamic planning method and system. LLP le MIT-086AUS, US Patent No.
7,349,863.
[23] E. Rojas, A. Mukherjee, Modeling the construction management process to support
situational simulations, Journal of Computing in Civil Engineering 17 (4) (2003)
273280.
[24] E. Rojas, A. Mukherjee, Interval temporal logic in general-purpose situational
simulations, Journal of Computing in Civil Engineering 19 (1) (2005) 8393.
[25] E. Rojas, A. Mukherjee, Multi-agent framework for general-purpose situational
simulations in the construction management domain, Journal of Computing in
Civil Engineering 20 (3) (2006) 165176.
[26] R. B. Information, Rs means cost works, http://www.meanscostworks.com 2009,
(Accessed April, 2011).
[27] A. Sawhney, S. AbouRizk, Ap3-advanced project planning paradigmfor construction,
Simulation Conference Proceedings, 1994. Winter, 1994, pp. 11531158.
[28] A. Sawhney, S. AbouRizk, D. Halpin, Construction project simulation using CYCLONE,
Canadian Journal of Civil Engineering 25 (1) (1998) 1625.
[29] B. Senior, Late-time computation for task chains using discrete-event simulation,
Journal of Construction Engineering and Management 121 (4) (1995) 397403.
[30] B. Senior, D. Halpin, Simplied simulation system for construction projects, Journal
of Construction Engineering and Management 124 (1) (1998) 7281.
[31] A. Touran, T. Asai, Simulation of tunneling operations, Journal of Construction
Engineering and Management 113 (4) (1987) 554568.
[32] M.T. Watkins, Developing Top-down and Bottom-up Simulated Environments for
Decision-making Research in Construction Project Management, 2008.
29.70
Total Cost (Million $)
Total Cost Duration
Completion Time (Days)
29.65
29.60
29.55
29.50
29.45
29.40
29.35
29.30
29.25
29.20
Crash_One Strategy Crash_Rwo Strategy Crash_Three Strategy Crash_Four Strategy Crash_Five Strategy
140.00
135.00
130.00
125.00
120.00
115.00
110.00
105.00
100.00
Fig. 8. Comparison of simulated as-built total costs for crash strategies.
560 P. Tang et al. / Automation in Construction 35 (2013) 551560

You might also like