Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Automation in Construction
a r t i c l e i n f o a b s t r a c t
Article history: A well-known practice to accelerate construction projects is to overlap the design phase activities. For a typical
Received 22 November 2014 construction project, a number of overlapping strategies exist during the design phase which all can result in
Received in revised form 13 August 2015 timesaving. However, the cost of these strategies varies significantly depending on the total rework and complex-
Accepted 14 August 2015
ity they generate. A favorable overlapping strategy is one that generates the required timesaving at the minimum
Available online 5 September 2015
cost. To find such a strategy, the question “Which activities have to be overlapped and to what extent to reduce
Keywords:
the project duration at the minimum cost?” should be answered. This research aimed at answering the question
Overlapping through generating an overlapping optimization algorithm. The algorithm works based on the principles of
Fast-tracking genetic algorithms (GAs). The algorithm explained in the paper is unique compared to previous algorithms
Optimization algorithms and frameworks available in the literature, as it can optimize multi-path networks and can handle all types of
Time–cost trade-off activity dependencies (i.e. finish-to-start, start-to-start, and finish-to-finish). It also takes both critical and
Rework non-critical activities into account and follows the critical path if the critical path changes or new critical paths
Construction management emerge. A computer tool was also developed to run, examine and validate the overlapping optimization algorithm.
This paper introduces the algorithm and the computer tool in detail and explains the results of their validation
through optimizing a real-world project schedule.
© 2015 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.autcon.2015.08.004
0926-5805/© 2015 Elsevier B.V. All rights reserved.
82 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
Table 1, which is adapted from Dehghan and Ruwanpura [1] and further
Construction–construction
completed, shows a brief comparative review of the existing literature.
The next sections provide further details.
X
activities. Loch and Terwiesch [3] and Terwiesch et al. [6] studied the
Design–construction
addition, they do not include the cost of rework in their model, which is
Design–design
very important as often the cost of rework is higher than the cost of
extra communication.
Information exchange and coordination are the focus of all research
performed by Prasad [5], Ha and Porteus [15], Nicoletti and Nicolo [4],
X
X
X
X
X
X
X
X
X
X
X
Loch and Terwiesch [3] and Terwiesch et al. [6]. Activity characteristics
Cascade of
overlapping
X
X
X
X
X
X
X
X
X
interdependent
X
X
X
X
X
construction activities.
Eppinger (1997)
Pena-Mora and Li [10], Bogus et al. [12], and Blacud et al. [14] is that
This Paper
they focus on one individual overlap and do not consider the overlap-
Table 1
ping in the context of a project schedule and with regard to other over-
laps. They do not provide any clue to which activities are better to be
R. Dehghan et al. / Automation in Construction 59 (2015) 81–95 83
overlapped and which activities are not; nor among those which are by considering all network paths, and Hussain and Chua [26] presented
overlapped, which activities can be more overlapped than others since an optimization model to overlap design and construction activities
they generate less risk and less cost. Roemer et al. [7] and Roemer and with minimal rework. The advantage of their effort over previous efforts
Ahmadi [8] have tried to provide answers to these questions. is that they consider multiple dependencies (multi-predecessors) when
optimizing the project schedule. However, they still do not address the
2.3. Overlapping time–cost trade-off cascade of overlaps, in which overlaps themselves overlap.
Dehghan and Ruwanpura [1,27] have introduced a model of trade-
Roemer et al. [7] and Roemer and Ahmadi [8] took a different off between overlapping and rework of design activities addressing
approach from Krishnan et al. [17] and tried to address the design how to handle both multi-predecessors and cascades of overlaps.
costs of overlapping by evaluating the trade-off between overlapping Studies conducted by Dehghan [21], Dehghan et al. [28,29] and
lead times and overlapping costs (Roemer and Ahmadi's research has Dehghan and Ruwanpura [1,27] focus on the costs and benefits of over-
a focus on product development, not construction projects). Roemer lapping and how to formulate them. Dehghan and Ruwanpura [1] have
and Ahmadi [8] could clearly explain the concept of generating rework the most recent study in the field. Their study is, to some extent, inspired
as a result of overlapping. Furthermore, their effort to optimize overlaps by the models developed by Roemer et al. [7], Roemer and Ahamdi [8]
in a full chain of activities distinguishes them from the majority of and Gerk and Qassim [18], in that all have proposed models optimizing
researchers who only investigated overlapping between two activities overlaps together, not separately. What differentiates Dehghan and
isolated from other activities. However, they limited their research Ruwanpura [1] is that they also demonstrated: 1) how to handle multi-
to only one chain of activities (i.e. one path). Therefore, they did not path networks through introducing a “multi-predecessor effect” and
study a network of activities with multiple paths and with multiple 2) how to handle concurrent overlapping through introducing the “cas-
predecessors to some activities. Another limitation is that their optimi- cade effect”. The results of their work are the starting point of this paper.
zation does not cover a condition in which two overlaps overlap Specifically, we have developed and computerized an overlapping
(concurrent overlapping or cascade of overlaps). optimization algorithm based on the model developed by Dehghan and
Similar research has been conducted by Gerk and Qassim [18] who Ruwanpura [1,27]. However, their model like any model is only a repre-
introduced a mixed-integer nonlinear programming model for the sentation of reality, clarifying the mechanism and characteristics of
acceleration of projects. They suggest that project acceleration is possi- design activity overlapping, but is unable to identify practically which
ble through three different techniques: activity crashing, activity activities have to be overlapped and to what extent to reduce the project
overlapping, and activity substitution. Gerk and Qassim [18] modeled duration at the minimum cost. For this purpose, an algorithm and a
this problem and developed an objective function aimed at minimizing practical tool are required to bridge the model to the real world applica-
the total cost of project acceleration, and a large number of constraint tions. The objective of this paper is to introduce such an algorithm and
functions. The details of network calculations and optimization compu- tool which are able to:
tations were not explained in Gerk and Qassim's paper, but the exam-
ples they have offered indicate that their model has similar limitations • Determine which activities have to be overlapped
to Roemer and Ahamadi's model [8]. • Determine the degree of overlapping
The most recent research, with similarities to the research by • Handle multi-path networks
Roemer and Ahamadi [8] and Gerk and Qassim [18], is conducted by • Handle cascade of overlaps
Berthaut et al. [19,20]. This research investigated the different feasible • Take into account all activities, critical and non-critical, and follow the
modes of overlapping, taking resource constraints into account. They critical path if the critical path changes or new critical paths emerge
showed the close interaction of overlapping modes and resource con- • Handle all types of activity dependencies including finish-to-start
straints. In their model, they have not considered the effect of parallel (FS), start-to-start (SS), finish-to-finish (FF), and start-to-finish (SF).
overlaps, but they have addressed the effect of multi-predecessor over-
laps. Their approach to a multi-predecessor effect in general differs from
what Dehghan [21] suggested, and Dehghan and Ruwanpura [1] later 3. Method: overlapping time–cost tradeoff
criticized it for lack of accuracy in evaluating the rework duration
resulting from multiple predecessors. Fig. 1 [1] shows the mechanism of overlapping two dependent activ-
Cho and Hastak [22] developed their time- and cost-optimized ities, in which the start of an activity depends on the finish of another
decision support model (TACTICS) based on fast-tracking methodology
and genetic algorithms. In their model, they tried to optimize overlap-
ping at the level of design and construction work packages, not activi-
ties. In addition, they further divided the construction work packages
into more specific work packages by considering the concept of space
zoning — availability of working space in the construction site — which
is a new approach. The major flaw in their study relates to the effect of
rework: central to the concept of activity overlapping is that overlapping
increases the risk of changes and rework, and rework negatively affects
both project cost and project duration. This is the fact that makes over-
lapping a difficult area for research. As Cho and Hastak [22] did not
consider rework resulted from overlapping (which can be drastically
high for construction activities), their model makes only a limited contri-
bution to both theory and practice.
Khoueiry et al. [23] and Srour et al. [24] presented overlapping
optimization models based on activity dependency information; that
is, the evaluation rate of upstream and sensitivity of downstream
activities. These research studies also assume that a project has only
one chain of activities.
Hazini et al. [25] explained a heuristic method to determine optimum
degree of activity accelerating and overlapping in schedule compression Fig. 1. The mechanism of activity overlapping [1].
84 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
activity and the second activity can only be started if the first activity is Lij Duration of the overlapped interval between predecessor
finished completely. This is because the successor needs the information activity i and successor activity j
generated by the predecessor (e.g.: the predecessor activity can be “pre- Pij The probability that a change happens for predecessor activity i
paring pump layout drawings” by the mechanical engineer and the suc- during its overlapping with successor activity j and the change
cessor activity can be “preparing pump foundation drawings” by the causes some rework for successor activity j
structural engineer). However, to compress the schedule, the successor Tij The extended duration added to successor activity j, as a
activity may be intentionally started before the completion of its prede- result of rework originating from the changes made by prede-
cessor. This becomes possible if the predecessor activity releases some cessor activity i, during its overlap with successor activity j
preliminary information before its completion to the successor activity. Rij The equivalent rework duration for successor activity j, as a
Therefore, the successor can start sooner, using the preliminary infor- result of its overlapping with predecessor activity i.
mation and making necessary assumptions and predictions. The two Based on the explanation of the mechanism of overlapping, both the
activities can proceed in parallel for a while and, during this period, probability, Pij, and amount of rework, Tij, are functions of the overlap-
some intermediate information may also be transferred until the prede- ping duration, Lij. Therefore, the equivalent rework, Rij, is also a function
cessor is completed; then, the predecessor will release its final informa- of Lij.
tion to the successor. At this point, it is likely that the final information is
different from the preliminary or intermediate information and therefore, Ri j ¼ f L i j ð3Þ
changes and adjustments must be made to the successor to make it com-
patible with the final information. The changes and adjustments will take
some additional coordination and work (rework) in form of extra person-
hours (i.e. extra cost and time), which means an increase in the duration 3.1. Overlapping time–cost trade-off objective function
of the successor activity compared to its normal duration [1].
Two points are noteworthy. First, overlapping has a maximum dura- Dehghan and Ruwanpura [1] also formulated the overlapping time–
tion beyond which further overlapping is actually impossible because cost trade-off (TCT) objective function which is central to solve the
no preliminary information of any kind can be produced by the prede- overlapping optimization problem as follows:
cessor activity. In the current research, this amount is called maximum
n X
X
overlapping or maximum allowable overlapping [1]. m C l f ðT−T t Þ if TNT t otherwise 0:0
C¼ P i j T i j W j þ Ei j þ : ð4Þ
Second, no rework occurs if the final, preliminary, and intermediate Be f ðT−T t Þ if TbT t otherwise 0:0
i¼1 j¼1
information are compatible. Rework is probable to take place. The prob-
ability of rework depends on several factors. The literature review
shows—and the interviews and focus groups endorse—that these factors In which:
are the type and complexity of overlapped activities, their relation with
other activities in the project schedule, and the amount of overlapping. C Cost (Benefit) of overlap
On the other hand, the amount (the duration) of additional work is a Pij The probability that a change happens for predecessor activity i
function of the overlapping duration, the strength of the successor activ- during its overlapping with successor activity j and the change
ity dependency upon the predecessor activity, and the intensity of causes some rework for successor activity j
nonconformity between final and preliminary information. The maxi- Tij The extended duration added to successor activity j, as a
mum rework may happen when the final information is significantly result of rework originating from the changes made by prede-
contradictory with the preliminary information, and the successor is cessor activity i, during its overlap with successor activity j
required to apply a major change. In such a situation, the worst scenario Wj Total daily wage for successor activity j, including daily
is that the successor must disregard all its progress during overlapping salaries and daily overheads
and start over. Therefore, the rework duration cannot be logically Eij Extra costs, other than daily wages and overheads, imposed
more than the overlapping duration [1,27]. on successor activity j or on other project areas (design,
Since the impact of the change on the predecessor activity i in Fig. 2 procurement, construction, etc.) because of the changes
is the extended (rework) period for successor activity j, the equivalent made by predecessor activity i during its overlapping with
rework is defined by multiplying the probability of rework and the successor activity j
duration of rework as below: Bef Daily benefits of project early finish
Clf Daily costs of project late finish
Ri j ¼ P i j T i j : ð1Þ T Project duration
Tt Project target duration.
Subject to:
Noting that:
0 b Pij ≤ 1 sets the limit for the rework probability
0≤P i j ≤1⇒0≤Ri j ≤T i j : ð2Þ Clf ≥ 0 ensures no reward for late finish
Bef ≤ 0 ensures no penalty for early finish
In which: Tt N 0 maintains the target duration not be negative or zero
Pij = 0 if i is not the direct or indirect predecessor to j
Lij b Lij,max ensures no overlapping beyond the maximum allowable.
In which:
The main objective of this paper is to indicate how overlapping This network encompasses activities with one, two and three prede-
optimization can actually be performed using the overlapping cessors (e.g. activity 2, activity 6 and activity 7 respectively). In addition,
time–cost trade-off objective function (Eq. (4)). This is inherently a activity durations have been defined in such a way that all activities may
multi-objective (bi-objective) optimization problem, as both time and become critical as a result of increasing overlapping between them. The
cost have to be minimized at the same time; while not independent, normal duration of this network when there is no overlapping between
they are intricately related. Due to the large number of possible overlaps activities is 110 days and its critical path is: 1 → 2 → 6 → 7.
and extremely large number of possible strategies, the optimization The objective is to compress the network by applying overlaps
process will be complex and extensive. As a result, a robust optimization between activities. This can be achieved in two different ways: The
technique or algorithm (for example, an evolutionary optimization al- first is when a finish time is given and the least expensive overlapping
gorithm) is required to deal with such a complex, multi-objective and strategy to meet the finish time is requested. The second is when an
multi-parametric problem. The objective of this paper is to introduce additional cost is given and the overlapping strategy with the shortest
and explain such an optimization algorithm and its computerization. duration is desired. For both cases, some additional information is
The computer tool works based on the overlapping optimization required. For instance, the maximum allowable overlapping between
algorithm, assessing various overlapping strategies and identifying the each pair of activities, the amount of rework generated by each overlap,
least expensive strategies. The tool is actually a cost evaluation module and the estimated cost of each overlap must all be known. This informa-
linked to a commercial project scheduling software (MS Project). This tion, as well as other necessary information about the sample case study,
computer tool is so user-friendly that any scheduler or cost controller were suggested by Dehghan et al. [21,28,29] and presented in Table 2.
can easily run it and modify the schedule accordingly. The computer The table shows activities, their durations, the predecessor(s) for
tool is unique and new, as so far no similar tools exist in industry or each activity, and the type of relation activities have with each other,
academia. It can optimize overlaps in large and complex project sched- before and after overlapping. In real-world practice, overlapping is
ules in fairly short processing times. It is able to handle multi-path applied mainly through finish-to-finish (FF) or start-to-start (SS) rela-
networks and all types of activity dependencies. The tool takes all activ- tionships with a positive lag, and sometimes through finish-to-start
ities, critical and non-critical, into account and follows the critical path if (FS) relationship with a negative lag. In the sample case study, a similar
the critical path changes or new critical paths emerge. The tool can also strategy is utilized. As shown in Table 2, some activities are overlapped
take into account resource limitations and schedule constraints. by means of FS relationships with negative lags (e.g. activities 1 and 2),
Our suggested algorithm uses the principles of genetic algorithms some of them are overlapped using SS relationships with positive lags
and is partially inspired by the algorithm developed by Hegazy [30] (e.g. activities 1 and 4), and the rest are overlapped through FF relation-
for activity crashing: both use the principles of genetic algorithms as ships with positive lags (e.g. activities 1 and 4). For lag times, Table 2
the optimization module; however, our algorithm is more complex also highlights the minimum allowable: for example, the minimum al-
than Hegazy's, as the nature of activity overlapping is more complicated lowable lag time for the FS relation between activity 1 and activity 3 is
than activity crashing. These complexities are related to the existence of −10. The amount for the SS relation between activity 2 and activity 5 is 8.
overlapping rework, heterogeneous information inside genes, multi- With this information, the maximum allowable overlapping can
predecessor and cascade effects [1], and modifications and relaxations easily be calculated. Any overlapping beyond the maximum allowable
required on network calculations. Therefore, the majority of this paper is considered to be technically impossible. Alternatively, any overlap-
explains and clarifies the different elements of the overlapping optimi- ping beyond the minimum lag time is technically impossible, because
zation algorithm. We use an illustrative sample network to explain the for each of FS, FF, and SS relations the minimum lag is equal to the max-
suggested algorithm step by step and with full details. This network imum overlap as per Eqs. (5), (6) and (7). The relation between the lag
was used by Dehghan et al. [21,28,29] to introduce the preliminary time and the overlapping duration are further detailed in Section 3.2
results of their research and was designed in a way that includes a vari- and particularly Eqs. (6), (7) and (8).
ety of network phenomena: various types of activity dependencies Overlapping rework durations as a function of overlapped amount
(FS, SS, FF), multi-predecessor overlaps, cascade overlaps, and change and overlapping cost values as a function of rework durations have
of critical paths. The intention of using the same network in this paper been also provided in Table 2. These functions are required to perform
is that the interested reader can better understand the algorithm and time–cost trade-off computations. The cost of overlapping and the
can implement it on similar problems. Later in the paper, however, we rework duration are functions of overlapping degree. In this sample
will optimize a real-world network to validate the suggested algorithm. network, various arbitrary types of functions have been used to present
the overlapping rework and overlapping cost. These functions are in
compliance with the data extracted from the design phase of industrial
3.2. Illustrative sample network construction projects as stated by Dehghan and Ruwanpura [1]. This
means: 1) The rework ranges are between 5 and 30% of overlapping
Fig. 3 shows a network including seven activities and nine finish-to- durations; 2) The cost values have a linear relation with rework dura-
start relationships. The original duration of each activity is also included. tions (therefore, cost functions are ultimately a function of overlapping
durations) [1].
In addition, the following values are given [21,28,29]:
4. Overlapping algorithm
Table 2
General attributes of the sample case study.
Activity Duration Predecessor Original relation original lag Relation after overlap Minimum allowable lag Overlapping rework function Overlapping cost function
Rij = f(Lij) Cij = g(Rij)
1 36 – – – – – – –
2 30 1 FS 0 FS −17 R12† = 0.2L12* C12‡ = 900R12
3 28 1 FS 0 FS −10 R13 = 0.1L13 C13 = 850R13
4 48 1 FS 0 SS 19 R14 = 0.15L14 C14 = 1050R14
5 22 2 FS 0 SS 8 R25 = 0.3L25 C25 = 800R25
6 24 2 FS 0 FS −20 R26 = 0.25L26 C26 = 1000R26
6 24 3 FS 0 FF 6 R36 = 0.2L36 C36 = 1000R36
7 20 4 FS 0 FF 8 R47 = 0.15L47 C47 = 950R47
7 20 5 FS 0 FS −15 R57 = 0.05L57 C57 = 950R57
7 20 6 FS 0 SS 10 R67 = 0.25L67 C67 = 950R67
two main variations, as the overlapping time–cost trade-off problem 4.2. Generating initial population
has two main variations. According to Dehghan and Ruwanpura [1],
the first variation is when the project target duration (Tt) is given, and An overlapping strategy is a collection of individual overlaps. Each
an overlapping strategy with the minimum cost (or maximum profit) overlapping strategy is presented by a chromosome and is a potential
is required. Therefore, the time–cost trade-off problem is to find over- solution to the problem of reducing the project duration to the targeted
lapping strategies that generate the minimum project costs. time. For the first step, a number of random chromosomes (overlapping
The second variation is when a project target cost Ct (maximum strategies) that meet the “time” criterion should be generated.
acceptable cost or minimum desirable profit) is given, and an overlap-
ping strategy resulting in minimum project duration is asked for. 4.2.1. Generating random chromosomes
In this case, the time–cost trade-off problem is finding overlapping Each chromosome represents an overlapping strategy and encom-
strategies that minimize project duration but do not exceed the passes several genes. Each gene represents a relation between two
predetermined project cost limit (Ct). The first variation is a more activities and each relation is a potential overlap. Like a real natural
common problem in the design and execution of construction projects. gene which holds the information required to build and maintain an
In this paper, the algorithm for the first variation is explained in organism's cells, the overlap genes hold the required information to
detail; the algorithm for the second variation is only slightly different. build the project network and evaluate its duration and cost. Each
Fig. 4 shows the flowchart of the suggested overlapping algorithm gene consists of eight pieces of information as follows:
for minimizing the project cost within the project target duration. A
part of the algorithm uses the principles of genetic algorithms (GAs) 1. Predecessor activity
optimization in order to solve overlapping time–cost trade-offs. The 2. Successor activity
algorithm encompasses extensive recursive calculations and numerous 3. Successor activity duration
trial and errors that require computer implementation. Two major 4. Type of relation between the predecessor and the successor
calculation modules exist in the algorithm. One module performs 5. Lag time
cost calculations and the other performs schedule computations and 6. Overlapping duration between activities
evaluates project time. The algorithm itself has two main steps. The 7. Rework duration for the successor activity
first step is to generate a collection of random potential solutions 8. Cost of overlapping as a result of rework.
(shown as “Generating the initial population” in Fig. 4) and the sec-
Fig. 5 shows a random chromosome from the sample network with
ond step is to evolve the random solutions to more favorable solu-
the above eight items. The chromosome has 9 genes and each gene has
tions (shown as “Generating the offspring population” in Fig. 4).
8 pieces of information. Item 1 indicates which activity is the predecessor
The following subsections further clarify the flowchart (algorithm)
and item 2 indicates which activity is the successor. Item 3 shows the
of Fig. 4.
original duration of the successor activity and item 4 determines what
the relation between the two activities will be if they overlap. These
four pieces of information are extracted from the original project
4.1. Input data
network—i.e. the network at its normal condition. The fifth piece of infor-
mation is lag time. The lag time is generated quite randomly because the
From the required input data to start the algorithm, activity dura-
chromosomes of the initial population must be random. Then the over-
tions and activity dependencies are primary network variables required
lapping duration (item 6) is calculated using the lag time, the predecessor
to build the project network. Minimum allowable lag times, rework
activity duration and the successor activity duration. Calculating over-
duration functions (or overlapping rework durations), and change
lapping duration depends on the type of dependency between the
transfer ratios are complementary network variables. Overlapping cost
two activities. For example, the overlapping duration of Gene no. 4 is
functions (or cost values for different degrees of overlapping), project
30 − 11 = 19 because for an SS relation:
daily benefits for early completion (Bef) and project daily losses for
late completion (Clf) are cost variables that, along with the targeted
project duration (Tt), are used to calculate the total costs/benefits Overlapping duration ¼ predecessor duration–SS lag: ð5Þ
of overlapping. GA parameters are genetic algorithm variables and
their variations can change the efficiency of the calculations. Main
For an FS relation the overlapping duration is calculated as fol-
GA parameters are the desired initial population size, the desired
lows:
number of offspring reproductions, and the mutation-to-crossover
ratio. These parameters will be further clarified through the next
sections. Overlapping duration ¼ –FS lag: ð6Þ
R. Dehghan et al. / Automation in Construction 59 (2015) 81–95 87
And for an FF relation: integer numbers are the preferred way of showing activity durations
in this research, the developed algorithm can deal with any type of
Overlapping duration ¼ successor duration− FF lag: ð7Þ numbers, whether integer or real. As it is normal practice to express
duration in terms of “day” in construction projects, this time unit is
The overlapping duration can be any amount between zero used all throughout this paper; since one day is small enough for a
(no overlapping) and the maximum allowable overlapping. While construction project, half-days or hours are not used. Therefore, the
88 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
Fig. 5. A random chromosome. The above rework corrections are applied on the chromosome and
result in the chromosome shown in Fig. 6. With these corrections, the
associated costs will also be changed. Details on how to calculate the
durations can be expressed as integer numbers presenting the number total cost of overlaps will be explained in the next sections.
of days. For example, the overlapping duration between activity 1 and
activity 2 can be any integer number between 0 and 8. 4.2.3. Cascade effect
When the amount of overlap is available, its associated rework and The rework durations may still need corrections if any cascades of
cost for the successor activity can be obtained through rework and overlaps exist. The sample network has the potential to generate some;
cost functions (Table 2). For example, the overlapping value of gene in fact, any paths with three or more activities may have a cascade of over-
#4 is 19. According to Table 2, the rework function for this gene is laps. The overlapping algorithm identifies such paths and checks whether
R25 = 0.3L25, which means that the rework for activity 5 should be any cascades of overlaps exist. Potential paths that may generate cascades
6 days: of overlaps can be identified from the chromosome:
20 − 202 Þ 20 ¼ 7:25≈7:
Rework for activity 6 ¼ ð5þ3 53
Fig. 8. The sample network with the earliest start and finish times. Fig. 10. The chromosome associated with the network in Fig. 9.
90 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
of timesaving. The total amount of cost (or benefit) of chromosomes can [30]. Chromosomes generated through crossover or mutation have to
be calculated using the main objective function (Eq. (4)). Individual undergo the same process as initial random chromosomes.
costs of overlapping are obtained from the genes (Fig. 10) and totalled.
Also, the cost/benefit of a project late or early finish time is calculated 4.3.2. Network (schedule) calculations, time check, and cost calculation
and added to overlapping costs. The associated calculations with an offspring chromosome including
For the chromosome of Fig. 10, the total cost is calculated by sum- network calculations, time check, and cost calculations are essentially
ming up the cost of rework for all 7 activities. Cost of rework for activity the same as such calculations for a parent chromosome. Therefore,
1 is zero as it does not have any rework (this cost is not reflected in the what was described in previous sections are also applicable for offspring
chromosome). Cost of rework for activities 2, 3, 4, 5, 6, and 7 is 0, 0, 2100, chromosomes.
4800, 6000 and 2850 respectively. Therefore:
4.3.3. Cost check
Overlapping cost ¼ 0 þ 0 þ 0 þ 2100 þ 4800 þ 6000 þ 2850 ¼ 15750:
When an offspring is generated by crossover or mutation and its cost
is calculated, a fitness evaluation should take place to ensure the
Of which, $6000 is the cost value of genes 5 and 6. Since both genes
offspring chromosome is acceptable. This is performed by comparing
include the rework cost for activity 6, only one of them is taken into
the cost (benefit) of the offspring chromosome with other chromo-
account to calculate the total cost. The same is true for $2850 which is
somes; if it is better (lower cost or higher benefit) than the worst chro-
the cost value of genes 7, 8 and 9. Since all of them are the rework cost
mosome in the population, the offspring replaces the old chromosome.
for activity 7, only one of them should be taken into account to calculate
Otherwise, the offspring is ignored.
the total cost.
The timesaving benefit for early completion is calculated as follows:
4.3.4. Offspring population
Timesaving benefit ¼ 1000 dollars=day ð105–89Þ days ¼ 16000: The process of generating offspring chromosomes should be con-
tinued until a number of optimum chromosomes that best fit the objec-
tive function emerge. Miscellaneous termination criteria exist to stop
Finally, the total net cost is:
offspring generation. In the current algorithm, the number of genera-
Total cost ¼ 15750–16000 ¼ −250: tions determines when the process should be stopped. The final off-
spring population includes a range of best solutions. At this point, the
first variation of the overlapping optimization algorithm ends.
Therefore, the chromosome represents an overlapping strategy
which results in $250 net benefit.
4.4. Overlapping algorithm: second variation
4.2.7. Initial population
The generated chromosome, along with the project total duration The second variation, the overlapping algorithm for minimizing the
and project total cost it generates, is recorded. The process of generating project duration within the project targeted cost, is only slightly different
random chromosomes is repeated until a suitable number of chromo- from the first variation. As shown in Fig. 4, the “time check” in the “gener-
somes, perhaps 100, is generated. These chromosomes form the initial ating the initial population” step of the first variation is changed to a “cost
population of solutions. The population encompasses completely check” in the second variation. The checking ensures that the generated
random overlapping strategies with various time and cost attributes. chromosome meets the cost constraint imposed. The rest of the flow-
In the next steps, the initial population is used to reproduce offspring charts are similar to each other. However, the time check and cost
population with targeted cost and time attributes. check in the “generating the offspring population” step are performed in
different ways for different objectives. In the first variation, the time
4.3. Generating offspring population check is to ensure that the imposed time constraint is met and the cost
check is to verify if the generated chromosome is better than the worst
The initial population of solutions must evolve to generate better chromosome in the population. In the second variation, the cost check is
solutions. This is possible by exchanging the stronger solutions' genes to ensure that the imposed cost constraint is met and the time check is
with each other (marriage or crossover) to generate new solutions to make sure the generated chromosome is shorter in duration than the
(offspring genes). New solutions are evaluated and, if they are better longest duration chromosome in the population.
than the weakest solutions in the population, will replace them. This Both variations of the algorithm are able to handle both continuous
process of generating better and better solutions is repeated until a and discrete values for activity durations, lag times, rework durations
satisfactory population is generated. The fittest members of the popula- and costs. Therefore, both can perform continuous and discrete
tion are the best solutions. optimization.
For several reasons Excel was selected as the GA algorithm platform: Microsoft Project applications as well as its flexibility to incorporate
first, it is a suitable platform for applying a wide variety of variables changes and perform sensitivity analyses.
and conducting complex and iterative computations; second, it gives In this section, the computer implementation of the first variation of
flexibility to the researcher to easily change the variables and review overlapping optimization algorithm is described. The second variation is
and analyze the results; third, it is quite compatible with MSP and a quite similar to the first. The required input data according to Fig. 11
flow of information can easily take place between them. is entered into the Excel template. From this, activity durations and
Visual Basic for Applications (VBA) has been used for automating the activity dependencies are primary network variables and are exported
calculations and coding and programming the simulation model. The to MSP to build the network. Minimum allowable lag times and over-
advantage of VBA is its compatibility with both Microsoft Excel and lapping rework functions are complementary network variables also
exported to MSP from Excel. Overlapping cost functions, project daily
benefits for early completion and project daily losses for late completion
Table 3
Results of the experiment. are cost variables and will be used by Excel to calculate the total costs/
benefits of overlapping. Project target duration is a constraint defined
No. Population Max. benefit Duration Calculation Min. benefit
by the user. The GA parameters are genetic algorithm variables and
($) (days) time (sec.) ($)
their variations can change the efficiency of the calculations.
1 Initial 4700 93 22 −5600 During optimization process, activity durations, activity dependen-
2 1st CO–MU 5400 93 31 −200
3 2nd CO–MU 5400 99 28 850
cies and suggested lag times are sent from Excel to MSP, which performs
4 3rd CO–MU 5400 99 29 1850 network calculations. Project duration and new lag times are then
5 4th CO–MU 5650 94 28 2650 returned to Excel. If project duration (T) is more than project target
6 5th CO–MU 6500 95 25 3050 duration (Tt), then the chromosome does not meet the time constraint
7 6th CO–MU 6500 95 27 3400
and the above steps should be repeated to generate a new chromosome.
8 7th CO–MU 6500 95 25 3750
9 8th CO–MU 6500 95 23 3850 However, if project duration (T) is less than project target duration (Tt),
10 9th CO–MU 6500 95 24 4350 the Excel template uses new lags to form a modified chromosome and
11 10th CO–MU 7700 93 24 4400 recalculate the total cost. This process is repeated until a desired num-
12 11th CO–MU 7700 93 25 4600 ber of first generation solutions (perhaps 100) is produced.
13 12th CO–MU 7700 93 22 4650
14 13th CO–MU 7700 93 23 4750
In the next step, the initial population should evolve to generate
15 14th CO–MU 7700 93 24 5350 better solutions. For this purpose, Excel selects two chromosomes ran-
16 15th CO–MU 8700 93 23 5400 domly as parent chromosomes and exchanges their genetic information
17 16th CO–MU 8700 93 22 5900 to reproduce an offspring chromosome. Once in a while Excel performs
18 17th CO–MU 8700 93 20 6350
mutation as well, according to a mutation to crossover ratio. The off-
19 18th CO–MU 8700 93 19 6350
20 19th CO–MU 8700 93 19 6500 spring chromosome is exported to MSP to perform network calculations
21 20th CO–MU 8700 93 19 6500 and determine project duration, modify lag times, and return the infor-
22 30th CO–MU 8700 93 206 6600 mation to Excel. Again, if project duration (T) is less than project target
23 40th CO–MU 8700 93 198 6700 duration (Tt), the Excel template uses the new lags to form a modified
24 50th CO–MU 8700 93 200 6700
offspring chromosome and recalculate the total cost. If the total cost is
92 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
less than the total cost of a chromosome in the population with the
highest total cost, the old chromosome is replaced by the new chromo-
some. The above process—crossover-mutation and cost recalculation in
Excel and network calculations in MSP—is repeated until a satisfactory
number of offspring chromosomes are reproduced (perhaps 1000).
The final population is a collection of low-cost chromosomes. The
chromosome with the lowest total cost is the best solution. However,
other chromosomes in the vicinity are alternative solutions depending
on project conditions.
In this section, the results of optimizing the sample case study are
presented. For the experiment, a personal computer (PC) with the Fig. 12. The best solution.
following attributes was used:
chromosome in the populations, on the other hand, was improved from
– Manufacturer: DELL $5400 in the 15th generation to $6500 benefit in the 50th generation.
– Model: Studio XPS 9100 Therefore, even though the best solution remains unchanged, later gener-
– Processor: Intel(R) Core(TM) i7 CPU 960 @ 3.2 GHz ations encompass better solutions than earlier generations.
– Installed memory (RAM): 12.0 GB Fig. 13 shows some of the results achieved during the experiment. To
– System type: 64-bit Operating System keep the figure readable, only the results of the initial population and
– Windows: Windows 7 Professional. the 1st, 10th and 20th crossover–mutation populations are shown.
As shown in the figure, costly solutions happen more with longer
project durations such as 101 or 102 days. However, beneficial solutions
The mutation rate, crossover rate and the population size are the key
happen more with shorter project durations such as 93 or 94 days.
parameters of a GA. For obtaining good performance, these parameters
Therefore, for the current case maximum benefits happen at lowest pro-
have to be “tuned” based on results obtained. Typically, a very small
ject durations and the optimization process improves the results from
mutation rate may lead to stagnation in local optima. A too-high mutation
costly long durations to beneficial short durations.
rate may lead to loss of good solutions and an essentially random search.
The total optimization calculation time to reach the best solution,
No general theory exists to determine good rates. In this case study,
which was first obtained in the 15th population, is the sum of previous
several trials were performed and the following GA parameters were
calculation times of the previous populations. Therefore, the total opti-
found to be suitable:
mization calculation time is 403 s, less than 7 min.
GA parameters:
Table 4
List of activities to be overlapped.
Predecessor activity Original duration Successor activity Original duration Rework functions
The section used had 40 activities including the piping and structural For better understanding, a part of the project schedule which
design. Two full sessions were spent educating the experts what was includes the above activities and their dependencies with each other
meant by “an overlapping rework function”, and four further sessions is schematically shown in Fig. 14.
were needed to review all 40 activities, identify those pairs that could
be overlapped, and develop their overlapping rework functions. With
the design of the project just finished and construction under progress, 6.2. Cost functions
the company was not able to provide more extensive support to the
research. Many other researchers have reported similar difficulties in Cost functions were generated by obtaining the direct and indirect
their research, and developing large numbers of overlapping rework costs of performing the successor activities from the company. The
and cost functions remains a hotly-discussed topic in the field of activity cost values (and functions) are reflected in Table 5.
overlapping and concurrent engineering [7,13,14]. The company also provided the project parameters as follows:
– Daily benefit for each day of early completion (Bef) = $1000 per day
6.1. Rework functions – Daily loss for each day of late completion (Clf) = $1000 per day
– Project targeted completion date (Tt) = 730 days.
The project schedule was reviewed and four pairs of activities which
could be overlapped were identified. Other activities were either not
able to be overlapped or were so short in duration (less than 3 days) With the rework functions, cost functions, and other project para-
that their overlapping could not generate a tangible timesaving. The meters available, the optimization was performed. The results are
equivalent rework durations as a function of overlapping duration shown in Table 6.
were also determined. The experts were comfortable and confident According to Table 6, an initial population with 100 chromosomes
with generating rough estimates about the most likely “equivalent was generated. The processing time was 57 s. The best chromosome
rework”. These four pairs of activities and their associated rework func- resulted in a project maximum benefit of $9600 and a project duration
tions are listed in Table 4. of 713 days. The worst chromosome in the population generated
Fig. 14. Activities to be overlapped and their dependencies with each other.
94 R. Dehghan et al. / Automation in Construction 59 (2015) 81–95
[13] S.M. Bogus, K.R. Molenaar, J.E. Diekmann, Strategies for overlapping dependent [25] K. Hazini, R. Dehghan, J.Y. Ruwanpura, A heuristic method to determine optimum
design activities, Construction Management and Economics, August 2006 (24) degree of activity accelerating and overlapping in schedule compression, Can.
(2006) 829–837, http://dx.doi.org/10.1080/01446190600658529. J. Civ. Eng. 40 (4) (2013) 382–391, http://dx.doi.org/10.1139/cjce-2012-0380.
[14] N.A. Blacud, S.M. Bogus, J.E. Diekmann, K.R. Molenaar, Sensitivity of construction ac- [26] M.A. Hossain, D.K.H. Chua, Overlapping design and construction activities and an
tivities under design uncertainty, ASCE J. Constr. Eng. Manag. 135 (3) (2009) optimization approach to minimize rework, Int. J. Proj. Manag. 32 (6)
199–206, http://dx.doi.org/10.1061/(ASCE)0733-9364(2009)135:3(199). (2013)http://dx.doi.org/10.1016/j.ijproman.2013.10.019.
[15] A.Y. Ha, E.L. Porteus, Optimal timing of reviews in concurrent design for manufactur- [27] R. Dehghan, J.Y. Ruwanpura, The mechanism of design activity overlapping in
ability, Manag. Sci. 41 (9) (1995) 1431–1447, http://dx.doi.org/10.1287/mnsc.41.9. construction projects and the time–cost trade-off function, The 12th East Asia-Pacific
1431. Conference on Structuralq Engineering and Construction (EASEC-12), January 26–28,
[16] V. Krishnan, S.D. Eppinger, D.E. Whitney, Accelerating product development by the 2011, Hong Kong, 2011http://dx.doi.org/10.1016/j.proeng.2011.07.246.
exchange of preliminary product design information, ASME J. Mech. Des. 117 (1995) [28] R. Dehghan, K. Hazini, J.Y. Ruwanpura, Optimum activity overlapping using genetic
491–498, http://dx.doi.org/10.1115/1.2826709. algorithms, The Sixth International Structural Engineering and Construction Confer-
[17] V. Krishnan, S.D. Eppinger, D.E. Whitney, A model based framework to overlap product ence (ISEC-6), June 21–26, 2011, Zürich, Switzerland, 2011http://dx.doi.org/10.
development activities, Manag. Sci. 43 (4) (1997) 437–451, http://dx.doi.org/10.1287/ 3850/978-981-08-7920-4_S1-CP19-cd.
mnsc.43.4.437. [29] R. Dehghan, J.Y. Ruwanpura, K. Hazini, A new model to minimize project costs
[18] J.E.V. Gerk, R.Y. Qassim, Project acceleration via activity crashing, overlapping, and through optimum activity overlapping, Proceedings of the Pacific Association of
substitution, IEEE Trans. Eng. Manag. 55 (4) (2008) 590–601, http://dx.doi.org/10. Quantity Surveyors (PAQS) 15th Annual Congress, July 23–26, 2011, Colombo, Sri
1109/TEM.2008.927786. Lanka, 2011.
[19] F. Berthaut, L. Greze, R. Pellerin, N. Perrier, A. Hajji, Optimal resource-constraint [30] T. Hegazy, Optimization of construction time–cost trade-off analysis using genetic
project scheduling with overlapping modes, CIRRELT-2011-092011. algorithms, Can. J. Civ. Eng. 26 (1999) 685–697, http://dx.doi.org/10.1139/cjce-26-
[20] F. Berthaut, R. Pellerin, N. Perrier, A. Hajji, Time–cost trade-offs in resource- 6-685.
constraint project scheduling problems with overlapping modes, CIRRELT-2011- [31] D.E. Goldberg, Genetic algorithms in search, optimization and machine learning,
102011. Addison-Wesley Publishing Co., Reading, Mass, 1989.
[21] R. Dehghan, A new model, algorithm and computer tool to optimize overlapping of [32] J. Zhou, P.E.D. Love, X. Wang, K.L. Teo, Z. Irani, A review of methods and algorithms
design activities in construction projectsPh.D. Dissertation Department of Civil Engi- for optimizing construction scheduling, J. Oper. Res. Soc. 2013 (64) (2013)
neering, Schulich School of Engineering, University of Calgary, AB, Canada, 2011. 1091–1105, http://dx.doi.org/10.1057/jors.2012.174.
[22] A.K. Cho, M. Hastak, Time and cost-optimized decision support model for fast-track [33] D.X.M. Zheng, S.T. Ng, M.M. Kumaraswamy, Applying a genetic algorithm-based
projects, ASCE J. Construction Engineering and Management. 139 (1) (2013) multiobjective approach for time–cost optimization, ASCE Journal of Construction
90–101, http://dx.doi.org/10.1061/(ASCE)CO.1943-7862.0000570. Engineering and Management 130 (2) (2004) 168–176, http://dx.doi.org/10.1061/
[23] Y. Khoueiry, I. Srour, A. Yassine, An optimization-based model for maximizing (ASCE)0733-9364(2004)130:2(168).
the benefits of fast-track construction activities, J. Oper. Res. Soc. 64 (8) (2013) [34] C.W. Feng, L. Liu, S.A. Burns, Using genetic algorithms to solve construction time–
1137–1146, http://dx.doi.org/10.1057/jors.2013.30. cost trade-off problems, ASCE Journal of Construction Engineering and Management
[24] I.M. Srour, M.A.U. Abdul-Malak, A.A. Yassine, M. Ramadan, A methodology for 123 (3) (1997) 184–189, http://dx.doi.org/10.1061/(ASCE)0887-3801(1997)11:
scheduling overlapped design activities based on dependency information, Autom. 3(184).
Constr. 29 (2013) 1–11, http://dx.doi.org/10.1016/j.autcon.2012.08.001.