Multi-objective optimization of time-cost-quality in construction projects

using genetic algorithm

Setenay Isikyildiz (Main Author)
Department of Civil Engineering, Istanbul University-Cerrahpasa
Istanbul, 34320 (Turkey)
[email protected]
https://orcid.org/0000-0002-0427-6122

Cemil Akcay (Corresponding Author)

Faculty of Architecture, Istanbul University
Istanbul, 34116 (Turkey)
[email protected]
https://orcid.org/0000-0002-8216-8688

Manuscript Code: 13854

Date of Acceptance/Reception: 24.11.2020/21.02.2020
DOI: 10.7764/RDLC.19.3.335

Abstract
In this study, it was aimed to develop a model that provides Time-Cost-Quality optimization in construction projects using the genetic algorithm
method. To fulfill that, Matlab codes for a multi-purpose genetic algorithm that performs Time-Cost-Quality optimization were developed and applied
to the problem, and the targeted success level was substantially achieved. This model is capable of performing and visualizing an advanced three-
dimensional time, cost, quality tradeoff analysis instead of the traditional two-dimensional time-cost tradeoff analysis. It was possible to select the
desired alternative under the project conditions obtaining Pareto images called a set of solutions rather than a single solution. In addition, one of the
main objectives of this study is to develop an effective model to assist the main contractors of the construction in selecting the most suitable
subcontractors.

Keywords: Time-cost-quality trade-off, Multi-objective optimization, Meta-heuristic, Genetic algorithm, Subcontractor selection.

Introduction

The main purpose of a project is to complete the planned works at the desired time, cost, and quality. Construction
projects consist of hundreds of activities that take place within a certain organization and model. These activities are
linked to each other in order. Each activity might have multiple alternatives for cost, time, and quality. The larger the
project, the more difficult it is to make a choice between these alternatives. Many different numbers of time, cost, and
quality options might come out. It becomes impossible to estimate the amount of time and resources allocated for the
project. For the projects of which are planned to satisfy minimum time, minimum cost, and maximum quality, no point
meets these three conditions. As the cost decreases, the duration of the project may increase or the quality may
decrease. It is required to increase the cost in order not to decrease the quality while the project duration shortens. It
is impossible to find a solution that meets all of these criteria consisting of minimum time, minimum cost, and maximum
quality. Therefore, it may be expected to concentrate the project manager's attention on one of these characteristics
depending on the project properties. Since there are hundreds of activities in a project and different time, cost and
quality options for each activity exist, it is difficult to make an appropriate choice. Utilizing algorithms, it becomes
possible to obtain the desired result from thousands of alternatives. The project progress can be regulated by
determining which characteristics of the projects are more important.

A significant amount of research has been done in the area of optimization for the use of construction resources.
Numerous studies have been conducted in the literature to minimize cost and time, which is one of the optimization
problems. In addition, project applications and benefits have been widely observed. However, few studies are available
in the literature on the problem of time-cost-quality optimization. Current research in the literature brought about a
series of optimization models developed using a variety of methods, including linear programming, integer
programming, dynamic programming, heuristic, and meta-heuristic algorithms. These models can be classified as
follows according to optimization objectives: i) Models that minimize project duration and/or improve resource
utilization and, ii. Models that minimize time and/or cost for construction projects using time-cost trade-off analysis
(Aminbakhsh et al., 2016; Easa, 1989; B. K. El-rayes, 2001; Geem, 2010; Gomar et al., 2002; Hegazy, 1999; Hegazy &
Wassef, 2001; Magalhães-mendes, 2017; Zalmai et al., 2019; Zhang & Li, 2010; Zheng et al., 2005).

335
Although these studies have contributed significantly to optimizing the use of construction resources, few studies are
focusing on multi-purpose models to optimize construction time, cost, and quality. A study in which a fuzzy logic system
was applied for different construction methods with different project duration, cost, and quality values exist (Iscioglu,
2011). Zhang et al worked on a combined optimization model in their research and applied this combined model which
is based on the time-cost and quality-time system [13]. Bingol and Polat applied particle swarm optimization for time,
cost, and quality optimization in subcontractor selection in their studies (Bingol & Polat, 2015). Similarly, Vedat and
Azim proposed a model that provides time-cost-quality optimization in construction projects using the Teaching
Learning Based Optimization method (Togan & Eirgash, 2019).

In recent years, literature has mostly focused on analyzing time and cost with little reported research on models for
optimizing construction time, cost, and quality together. Within the scope of this study, a multi-purpose and multi-
alternative model that makes time-cost-quality optimization using the genetic algorithm method which is one of the
meta-heuristic algorithms has been established. In order to measure the quality, the quality criteria, and weight ratios
for these criteria were determined with experts' opinions by a survey. Following this, Matlab codes of an algorithm
providing Pareto solutions inconsistent with the problem were written. In projects with a multi-purpose and multi-
alternative problem, a case for a time, cost, and quality optimization were solved by trying to use different population
values and the results were evaluated.

State of Art

Review on Optimization Algorithms

Optimization problems are generally considered as deterministic and stochastic in two main categories (Table 1). In the
deterministic approach, while solutions are made by classical methods taking into account the known parameters, the
variables that may occur in the design are not considered, whereas the variables in the stochastic approach are taken
into account.

Table 1. Categorization of optimization algorithms (Yang, 2010).

Deterministic Stochastic
Linear programming Heuristic
Non-linear programming Metaheuristic*
Gradient-based
Free of gradient
*
Population-based, Trajectory-based

The mathematical programming problem consists of optimizing a goal function under certain constraints. Heuristic
methods are algorithms based on natural phenomena to accomplish any purpose or goal, and these algorithms generally
provide good solutions but do not consider optimality. For this reason, the exact solution is not obtained because the
heuristic algorithms have the convergence feature, so a solution is obtained near the exact solution.

Metaheuristic methods aim to combine basic heuristic methods that will enable the search for solution space efficiently.
In this respect, these methods are general purpose heuristics designed to guide probing-specific intuitions to perform
searches in regions of high-quality solutions of search space (Dorigo & Stutzle, 2004). Top-heuristic algorithms; is a
decision mechanism that works on heuristic algorithms.

Meta-heuristic methods are the establishment of new approximate methods that combine basic heuristic methods
allowing to effectively scan the solution space (Dorigo & Stutzle, 2004). Meta-heuristics are general-purpose heuristic
methods created because of leading problem-specific heuristics to perform a search in areas of search space containing
high-quality solutions. Meta-heuristic algorithms are decision making systems that work on heuristic algorithms. In
other words, when it is assumed that three different methods can be applied for a problem and all of these methods
are heuristic algorithms that are advantageous in terms of different purposes, it is called a metaheuristic algorithm to
decide which one will be preferred. Numerous methods have been developed for the solution in various fields and these
methods have been revised according to special conditions of the problem. Thus, there is no need to make a selection
between multiple solution methods for a problem. Simply, a meta-heuristic algorithm makes a choice between existing
algorithms and applies the most successful one. This system which decides the most appropriate algorithm generally
works according to statistical information. In the literature, meta-heuristic techniques commonly used in solving
optimization problems are taboo search algorithm, ant colony optimization, genetic algorithm, artificial neural

336
networks, and particle swarm optimization. These methods vary by performance depending on the optimization
problem for which they find a solution.

Genetic algorithms are search and optimization tools used in project planning and control. The genetic algorithm helps
to identify optimal or near-optimal solutions for problems with large search space (Sorrentino, 2013). GA is also a widely
used algorithm for complex problems (Deb et al., 2002). The genetic algorithm (GA) can use discrete, continuous
relationships and linear and nonlinear functions can be implemented in it (Sivanandam, & Deepa, 2008). Different kinds
of algorithms can be merged with GA. The GA can solve various types of problems very easily and effectively because
complex calculations are not required. Thus, this study was used because of its simplicity and versatility. Due to its
simplicity and versatility, the genetic algorithm is convenient for this study.

Methodology

This study aims to provide a multi-purpose multi-alternative optimization model that supports minimizing construction
time and cost while maximizing quality. The developed optimization model used the Pareto optimality principles shown
(in Figure 1) to find optimum results within a wide range of possible solutions. The model produces not only one
optimum result but also an optimal set of solutions. Therefore, the most appropriate solution set will be available
according to the parameter that the manager gives priority.

Figure 1. Pareto optimal solution (K. El-rayes & Kandil, 2005).

Generally, the main contractors tend to have the subcontractors do some work items for the construction project. In
such cases, the selected subcontractors are responsible for the quality of the work. Accordingly, the success of a
construction project and, hence the success of the main contractor, depends highly upon the performance of the
subcontractors. Therefore, one of the major problems that the main contractor encounters is the selection of the most
appropriate subcontractors for each work packages. In most cases, the main contractors make this decision at the
beginning of the project and they have to evaluate the performance of the subcontractors in terms of time, cost, and
quality in the process of selection. Following this selection procedure, an optimal combination of subcontractors to carry
out the different workgroups in the project is selected. For the contractors, in general, it is difficult to select the most
appropriate combination that balances the time-cost-quality trade-off. One of the main objectives of this study is to
assist in the selection of the subcontractors which provide optimum time-cost-quality according to the parameter that
the manager gives priority by utilizing the experience data of the companies. In this study, the genetic algorithm method
was considered and applied to the problem.

Genetic Algorithm

A genetic algorithm is a meta-heuristic approach that enables the optimization of complex and difficult problems. This
procedure is based on the principle of survival of the strong through some operators mimicking nature (Goldberg, 1989).
Genetic algorithms are natural selection mechanics and natural genetics-based search algorithms. In order to build a
search algorithm with an innovative design, the survival of the best of the array structures is combined with random
information exchange. In each new generation, a new artificial population is created using the characteristics of the
most talented and surviving individuals and, occasionally, random changes can be done in genetic information for better
results. The problem is optimized by using previous information obtained from the population efficiently at new search
points. Genetic algorithm is a search and optimization tool used in project planning and control. It assists to identify
optimal or near optimum solutions for problems with large search space (Sorrentino, 2013).

337
Genetic algorithms have many advantages over traditional optimization algorithms and the most notable advantage is
the ability to cope with parallelism and complex optimization problem. Genetic algorithms can address various types of
optimization of the fitness function, such as stationary or non-stationary, linear or nonlinear, continuous or discrete.
Since the generations in a population act as independent agents, the population (or any subgroup) can explore the
search area in many ways at once. However, genetic algorithms have also some disadvantages. Selection of important
parameters such as formulation of the fitness function, the use of population size, mutation and crossover rates, and
selection criteria of the new population should be carefully determined. Any inappropriate selection might make it
difficult to combine the algorithm or produce meaningless results. Despite these disadvantages, genetic algorithms
remain one of the most widely used optimization algorithms in modern nonlinear optimization (Yang, 2010).

GA is a search technique used to find optimum or near-optimal solutions for optimization problems. GA searches for a
global optimum with a meiosis-based algorithm. The first population is randomly generated and new genes are
reproduced by crossover. Genetic differences are caused by mutation and inappropriate genes are eliminated with
natural selection procedures (Bettemir, 2011). Concepts for genetic algorithms and definitions regarding operators are
given in the Appendix. In a standard GA, candidate solutions (individuals) are expressed as binary vectors in equal size.
Initially, a group of these vectors is selected randomly to create a population of a certain size. Vectors called
chromosomes allow the development of new populations as a result of evolutionary mechanisms. The genes on the
chromosome represent a dimension of a vector with 'n' dimensions (Sari, 2008). The first step of GA is the production
of the initial population. The determination of population size is extremely important. Large populations require heavy-
load calculations and small populations have the critical risk of concealing the solution (Bettemir, 2011). The fitness
function generates an output from the input variables (chromosomes) in an array form. The fitness function can be a
mathematical function, an experiment or a game. The objective is to modify the output as desired by finding the most
appropriate values in the inputs. GA begins with the definition of a chromosome or a sequence of chromosomes or an
array consisting of variables to be optimized. If a chromosome has N variables in a form of P1, P2, …, PN, chromosome
elements are expressed as a line vector:

Chromosome= [P1, P2, P3, …, PN] (1)

For example, searching for the maximum altitude on a topographic map requires a fitness function with longitude (x)
and latitude (y) input variables. In this case, the chromosome with N = 2 can be identified as follows:

Chromosome = [x, y] (2)

Each chromosome created has a fitness value obtained from the fitness function f with variables P1, P2, …, PN.

Fitness value = f(chromosome) = f (P1, P2, …, PN) (3)

Crossover is a necessary process for genetic regeneration. The new genes are reproduced from randomly selected
genes. Pairs (parents) are determined by random numbers and the new two genes are reproduced with crossover
process. The crossover location is also determined by generating a random number. After the crossover, two new gene
combinations are derived from the two existing gene combinations in the population. A low crossover rate is vitally
important because sufficient genetic differences cannot be established and the convergence of the solution is reduced.
A very high crossover rate messes up genes excessively and it prevents the chromosomes carrying the well-matched
genes from reaching the global optimum. One-point, two-point and uniform crossover techniques are often used for
the crossover process. The mutation prevents the dominance of a particular gene with a high probability of survival.
Some well-matched chromosomes might be tripped to the local minimum. In this case, better solutions are overlooked.
The mutation operator prevents tripping to possible local optimum values. A high mutation rate can lead to disruption
of well-matched genes approximating to global optimum (Bettemir, 2011).

Natural selection is the final stage of the genetic algorithm cycle. Natural selection ensures that the size of the
population before the crossover process remains the same as the result of the process. It also improves the overall gene
quality of the population by eliminating individuals with low fitness values. On the other hand, low-fit chromosomes
can carry very important genes in their certain parts and some precautions are taken during natural selection to protect
these parts. In optimization algorithms, the stop criterion can be applied as i) the maximum number of iterations, ii) The
value of the objective function iii) The number of consecutive iterations in which the convergence or solution does not
develop, and iv) A specified time condition.

338
 Implementation of GA on a Construction Project and a Case Study

In this study, a genetic algorithm is applied to a multi-purpose optimization problem. The data of the problem examined
in Matlab code was obtained from a construction project. The problem investigated in this study consists of 15 main
activities. Since the Matlab code was for trial purposes, the problem was applied with only 15 main activities without
taking into account the substrates of the main activities. Table 2 presents data for activities of the problem dealt with.
These data include the activity ID, the name of the activity, the activities just before the main activity, and the different
time-cost-quality values up to five options for each activity. The duration and cost values of the five alternatives are
obtained from subcontractors' tenders of a real construction project. These subcontractors have previously worked on
different projects under the same contractor firm. The quality values of the alternatives are the average rates
determined and evaluated by the main contractor technical personnel according to the standard in Table 3. CPM
network and data belong to the problem are given in Figure 2.

Table 2. Problem application data for the case study (Isikyildiz, 2019).

Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5

ID Activity name IP*
T* C*103 Q* T* C*103 Q* T* C* 103 Q* T* C*103 Q* T* C* 103 Q*
Construction site
A - 1 - - - - - - - - - - - - - -
delivery
B Mobilization A 7 20 80 5 24 70 - - - - - - - - -
C Excavation B 15 300 80 17 285 95 14 320 85 12 330 70 - - -
D Foundation C 10 350 70 8 380 80 12 315 90 - - - - - -
E Concrete works D 88 1.150 70 86 1.200 60 93 1.000 80 82 1.500 90 - - -
Electic conduit in
F E 40 400 85 43 370 95 36 440 80 - - - - - -
Wall and Ceiling
G Wall works F 34 300 70 31 325 80 38 275 80 29 348 90 - - -
H Screed concrete J 15 120 80 17 100 90 12 145 70 - - - - - -
Heating
I G 30 320 75 33 300 80 26 352 70 29 347 90 - - -
installation
Sanitary
J G 20 390 90 18 410 80 22 370 70 25 330 60 16 459 50
installation
K Tile H 25 260 70 27 240 95 21 290 85 - - - - - -

Door and window

L G 15 300 70 12 340 80 10 400 60 18 280 80 19 275 95
works

M Plaster works I-L 55 190 80 57 170 70 49 275 90 59 150 60 - - -

N Ceiling floor M 33 270 75 29 360 80 35 280 70 38 250 60 27 400 90
O Satin paint K-N 15 70 80 17 70 90 12 100 70 18 65 50 - - -
IP*: Immediate Predecessor, T*: Time, C*: Cost, Q*: Quality.

Figure 2. CPM network for the case study (Isikyildiz, 2019).

Five alternatives are corresponding to the duration, cost, and quality value of each task for 15 activities in the sample
problem. When a random alternative is chosen, different times, cost, and quality components come out. Genetic
algorithms would be appropriate for such multi-purpose optimization problems. The fact that quality, duration, and cost
are equally important for time-cost-quality optimization problems prompted researchers to focus on multi-objective
optimization problems. Revisions in the project and factors affecting the progress of the work may reveal the possibility
of project completion in different time periods. It is not possible to obtain only one optimum solution for multi-purpose
optimization problems. Minimum cost and maximum quality should be determined for each time period. Similarly,
minimum time and minimum cost data are important for maximum quality values. The cost of the project is an
important parameter, as well.

339
In sample problem, it is not possible to conclude which option is essential when two options, which are i) 300 days for
the project completion time, 820000 USD for cost and 90% of average quality and ii) 200 days for the project completion
time, 820000 USD for cost and 82% of average quality, compared with each other. Similarly, it is not clear that which
option would be advantageous comparing two conditions, which are i) 250 days for the project completion time, 820000
USD for cost and 90% of average quality and ii) 300 days for the project completion time, 700000 USD for cost and 90%
of average quality. The project manager should evaluate the conditions and decide which criteria should be given
weight. In problem-solving, for hundreds of project completion time, different cost values and average quality values
were obtained. There is no case in which the cost and duration of the project have the lowest value and the average
quality has the highest value. Therefore, some solutions cannot dominate each other. Each of these solutions cannot
assert dominance over the others and, therefore, it is not possible to call one of these solutions the optimum solution.
The set of solutions includes non-dominated solutions which are also called Pareto front.

Quality Data of the Problem

In the multi-purpose optimization problem investigated in this study, the main objective is to find out the Pareto solution
contour, where the project activities were implemented, project duration and cost are minimum, and hence, the quality
is maximum. Each activity has a multi-alternative of duration and cost. Quality ratios also vary depending on these
alternatives. Quality is also effective in decision making, as the relationship between time and cost plays an effective
role in the decision-making process at the beginning of project implementation. Time, cost, and quality values in projects
vary depending on many factors. The evaluation criteria and weight ratios that are taken into consideration regarding
quality are determined by taking expert opinions with a questionnaire study. These quality assessment experts are;
project quality experts, project managers, and site chiefs who generally work on large-scale construction projects in
Istanbul. We aimed to select experts working on large-scale construction projects with a construction area of at least
20000m2 and above. We reached these experts with the help of the Istanbul Chamber of Civil Engineers. The results of
the survey are given in Table 3.

The case study project is a seven-storeyed (G+6) faculty building with a construction area of 26000m2, whose
construction is in progress in Istanbul, Turkey. Those who make quality assessments for subcontractors are the project
manager, quality control engineer, and site chief in charge of this project. In the application within the scope of the
study, the data in Table 3 were used for the weighted average concerning subcontractors' quality assessment by the
construction company.

Table 3. Quality criteria for the case study (Isikyildiz, 2019).

Criteria Weight ratio (%)

1 Quality standard certificates 5
2 Equipment (Technology - Innovation) 5
3 Technical personnel capacity 5
4 Starting the work in time 10
5 Finishing the work in time 10
6 Following the instructions 5
7 Observance of work safety regulations 5
8 The occurrence of work accident 20
9 The occurrence of near-miss events 5
10 Communication with stakeholders 5
11 Environmental conscience 5
12 Deficiency amount in receiving 5
13 Method and speed for solving the problems 5
14 Use of construction equipment 5
15 Material loss 5
Total 100

Subcontractors are evaluated by the contractor technical personnel in accordance with the criteria given in Table 3
regarding the work they perform in each project. As a result of the assessment, sub-contractor quality scores are
obtained and this scoring is used in the selection of sub-contractors for other works. Table 4 presents an example of a
quality assessment for a ceiling floor subcontractor. Similarly, quality scores were also obtained for other works. The
quality values of the alternatives for the problem dealt with in this study are given in Table 5.

340
 Table 4. Ceiling floor quality components for the case study (Isikyildiz, 2019).
Alternatives
Quality components of ceiling floor Point %
1 2 3 4 5
Point 85 75 95 85 95
1 Quality standard certificates 5
Percentage value 4.3 3.8 4.8 4.3 4.3
Point 95 80 95 75 90
2 Equipment (Technology- Innovation) 5
Percentage value 4.5 4 4.8 3.8 4.5
Point 80 90 90 70 95
3 Technical personnel capacity 5
Percentage value 4 4.5 4.5 3.5 4.8
Point 60 90 55 75 90
4 Starting the work in time 10
Percentage value 6 9 5.5 7.5 9
Point 75 75 60 60 85
5 Finishing the work in time 10
Percentage value 7.5 7.5 6 6 8.5
Point 75 80 70 55 85
6 Following the instructions 5
Percentage value 3.8 4 3.5 2.8 4.3
Point 75 85 65 55 95
7 Observance of work safety regulations 5
Percentage value 3.8 4.3 3.3 2.8 4.8
Point 55 80 60 55 80
8 The occurrence of work accident 20
Percentage value 11 16 12 11 16
Point 90 80 50 50 95
9 The occurrence of near-miss events 5
Percentage value 4.5 4 2.5 2.5 4.8
Point 90 80 70 50 95
10 Communication with stakeholders 5
Percentage value 4.5 4 4.3 2.5 4.3
Point 90 80 85 50 95
11 Environmental conscience 5
Percentage value 4.5 4 4.3 2.5 4.3
Point 90 80 80 60 95
12 Deficiency amount in receiving 5
Percentage value 4.5 4 4 3 4.3
Method and speed for solving the problems Point 90 70 70 50 95
13 5
Percentage value 4.5 3.5 3.5 2.5 4.3
Point 90 80 80 70 95
14 Use of construction equipment 5
Percentage value 4.5 4 4 3.5 4.3
Point 65 70 80 40 100
15 Material loss 5
Percentage value 3.3 3.5 4 2 5
Total quality value % 75 80 70 60 90

Table 5. Problem database for the case study (Isikyildiz, 2019).

Activity Alternative 1 Alternative 2 Alternative 3 Alternative 4 Alternative 5

Activity name
ID Quality %
A Construction site delivery - - - - -
B Mobilization 80 70 - - -
C Excavation 80 95 85 70 -
D Foundation 70 80 90 - -
E Concrete works 70 60 80 90 -
Electric conduit in Wall
F 85 95 80 - -
and Ceiling
G Wall works 70 80 80 90 -
H Screed concrete 80 90 70 - -
I Heating installation 75 80 70 90 -
J Sanitary installation 90 80 70 60 50
K Tile 70 95 85 - -
L Door and window works 70 80 60 80 95
M Plaster works 80 70 90 60 -
N Ceiling Floor 75 80 70 60 90
O Satin Paint 80 90 70 50 -

341
Solution of the problem

To solve the sample problem, the genetic algorithm code was improved and implemented in Matlab inconsistent with
a multi-purpose multi-alternative optimization problem. Figure 3 shows the flow chart of the project implementation.
The crossover rate is generally between 0.7–1 in the literature and the mutation rate is a range of 0.01- 0.1. In this study,
the crossover ratio was selected as 1 and the mutation ratio was used as 0.05 for the solution of the optimization
problem. The initial population for the genetic algorithm system was composed of 50, 100, 500, and 1000 individuals.

Figure 3. Flow chart of the project implementation. (Isikyildiz, 2019).

Required Data
 Planning network
 Alternative time-
cost-quality values
for activity
Data
Input

Crossover and mutation Creating and Updating Population

operations with Genetic Algorithm

Selecting elite individuals according to affinity Fitness value calculation for all
degree with respect to Pareto Cluster individuals

No
Creating 3-Dimensional Pareto Time
Cluster Check

If time is over Yes

Complete
the analysis

Calculation of Fitness Function

In this study, CPM back-and-forth calculations as fitness functions were generated with Matlab code for each
chromosome. Total project duration, total cost, and average quality information for the alternatives being composed of
all chromosomes were calculated. Provided that all activities are carried out, just before and after activities of the main
activities, and the critical path calculation is included in the fitness function.

342
Creating the Initial Population

In solutions developed using meta-heuristic algorithms, generally, the initial population is created with the results of
other heuristics or random selection. In this study, the initial population is randomly generated in the time-cost-quality
optimization problem. Chromosomes are defined as vectors and the population matrix is obtained. The cost, duration
and quality matrices corresponding to each gene in the random initial population were created separately in Matlab.
Matrices for time, cost and quality data, which are the alternative values of the project activities, were also created in
Matlab, separately. For the activities in the project, previous activities were determined and the fitness function was
applied in Matlab performing their back and forth calculations. For each possible project time period, initial population
values that would allow activities to choose different values from time, cost, and quality alternatives were applied by a
using genetic algorithm in Matlab. The emerging set of solutions was plotted by Matlab. For each project time period,
the detailed table of time, cost, and quality alternatives for the solutions was created by Matlab. A new CPM schedule
was created maintaining priority values for each project duration. After the initial population is formed, a project time
is formed with the alternative components of each of the chromosomes as well as the resulting Pareto image. Each
project has its unique total duration, total cost value, and average quality percentage.

It is possible to stop the analysis by checking whether it meets the stopping criterion. Convergence to a defined
objective function value, expiration of any given time, number of iterations, or number of iterations that do not produce
a better solution can be used as a stopping criterion. In this study, time was assigned as a stopping criterion. The process
is repeated throughout the whole iterations, and the resulting values may be optimum or near optimum. Even though
meta-heuristic methods cannot provide optimum results, they can guarantee near-optimum solutions for sufficient time
periods. The genetic algorithm aims to reach the optimum result by improving the solutions with each iteration or
transmitting the good solutions in the previous iteration to the next one by an elitist approach. Each generation
produces individuals in the same number of initial population and solutions are explored within this population. In this
study, data of the application problem was used in GA by developing a fitness function code. In addition, optional
solutions and Pareto images were created in Matlab changing the initial population values of the application problem.

Results and Discussion

In terms of implementation, the most important issues in the construction market are project duration and project cost.
In the scope of this study, quality values were also included in the analyzes of these two important factors. Project cost
and duration generally vary in inverse proportion to each other. The cost will increase as it will be required to increase
the number of teams in order to shorten the project duration. The possibility of a reduction in quality values with the
shortening of time exists, as well. On the other hand, the increasing quality will also cause an increase in costs. The
interaction of these three factors is quite complex. The purpose of the project is to minimize the duration and cost,
while the quality value is aimed to be at the highest point. Taking into consideration these three factors, it becomes
more difficult to find the optimum solution as the number of project activities and the time, cost and quality alternatives
increase. In the multi-purpose and multi-alternative optimization problem, it was aimed to decrease the project cost
and duration, and to increase quality values. In the scope of this study, it was aimed to obtain Pareto solutions with
different numbers of the initial population using the data of the application project and the results of the analyses were
evaluated. Running the GA algorithm properly, many optimal solutions have been accomplished. Since the case in this
study is a multipurpose optimization problem, it is expected that more than one optimal solution comes out. Figure 4
shows the 3D graphs of the analysis results with 50, 100, 500, and 1000 populations. It was observed that as the initial
population value increased, the number of solution alternatives also increased.

It was observed that, with the increase of population, repetition of solutions and time loss come out for analyses
including more than 1000 populations. In order to increase alternative solutions and to reach the most efficient results,
it was decided to keep on using, as an optimal choice, 1000 populations for the case problem including 15 activities and
5 alternatives. As can be seen in Figure 4, when the data was fed to the system with the initial population value of 50
and the genetic algorithm was run, 103 generations were created and 18 Pareto images were recorded. In addition, 105
generations were produced and 35 Pareto images were recorded for running of genetic algorithm with the initial
population value of 100. For the initial population value of 500, 102 generations were generated and 175 Pareto images
were recorded. Finally, when the data was input to the system with the initial population value of 1000 and the genetic
algorithm was operated, 91 generations were produced and 350pareto images were recorded. The results were
unexceptionally repeated when the initial population value is selected more than 1000. All Pareto results obtained from
the analysis are optimal solutions. They are incomparable and do not dominate each other. Selecting the most
appropriate solution depends entirely on the decision of the project manager. For example, if the project manager seeks
the average quality of the project to be over 80%, and decides to choose between the Pareto solutions that come out

343
of the analysis according to the second-degree cost priority, there are 86 Pareto solutions with a quality average of over
80%. The details for the best 20 Pareto solutions in terms of minimum cost and minimum time are given in Table 6.

Figure 4. Pareto and time-cost-quality relationships as a result of analysis with different populations. Source: Self-elaboration.

Solutions with 50 populations Solutions with 100 populations

Solutions with 500 populations Solutions with 1000 populations

Table 6. Pareto values of 50 population solutions of the application project. Source: Self-elaboration.

Pareto Activity IDs and the alternative order in which they belong
Time Cost (USD) Quality%
Numbers A B C D E F G H I J K L M N O
1 332 863000 83 1 1 2 3 3 2 4 2 4 1 2 5 3 3 2
2 328 874400 83 1 1 2 3 3 2 2 2 4 1 2 5 3 2 2
3 328 877600 83 1 1 2 3 3 2 4 2 2 1 2 5 3 5 2
4 326 879000 83 1 1 2 3 3 2 4 2 4 1 2 5 3 2 2
5 324 883000 83 1 1 2 3 3 2 4 2 4 3 2 5 3 5 2
6 323 886000 83 1 1 3 3 3 2 4 2 4 1 2 5 3 2 2
7 324 887000 84 1 1 2 3 3 2 4 2 4 1 2 5 3 5 3
8 321 888000 83 1 1 2 3 3 2 4 2 3 1 2 5 3 5 2
9 319 893000 83 1 1 2 3 3 2 4 2 4 1 2 5 3 5 2
10 321 894000 83 1 1 3 3 3 2 4 2 4 1 2 5 3 5 2
11 320 900000 83 1 1 2 2 3 2 4 2 4 1 2 5 3 5 2
12 317 901000 83 1 1 2 3 3 3 4 2 4 1 2 5 3 5 2
13 308 936600 83 1 1 4 3 4 2 4 2 4 1 2 5 3 5 2
14 317 974400 83 1 1 2 3 4 2 2 2 4 1 2 5 3 2 2
15 315 979000 84 1 1 2 3 4 2 4 2 4 1 2 5 3 2 2
16 313 983000 83 1 1 2 3 4 2 4 2 4 3 2 5 3 5 2
17 312 986000 83 1 1 3 3 4 2 4 2 4 1 2 5 3 2 2
18 313 987000 85 1 1 2 3 4 2 4 2 4 1 2 5 3 5 2
19 310 988000 83 1 1 2 3 4 2 4 2 3 1 2 5 3 5 2
20 308 993000 83 1 1 2 3 4 2 4 2 4 1 2 5 3 5 3

344
Activity options, which compose the project, belong to the best 20 Pareto images according to constraints defined by
the manager are presented in Table 6. If the manager aims to minimize the project cost and the duration of the project
is not critically important, he/she should choose the first solution with 83% average quality, 332 days duration, and
863000 USD cost. Accordingly, the main contractor should use the corresponding contractor planning in order to obtain
the first Pareto solutions as shown in Figure 4 (Hence, the sub-contractor array will have a form of A:1, B:1, C:2, D:3;
E:3; F:2; G:4; H:2; I:4; J:1; K:2; L:5; M:3; N:3; O:2. Likewise, the decision-maker can choose any of the Pareto optimum
solutions given to meet the priorities and objectives of the project.

Conclusions

A multi-purpose multi-alternative optimization model has been developed for three-dimensional time-cost-quality
tradeoff analysis, instead of the traditional two-dimensional time-cost analysis method. This model seeks optimal
resource utilization plans that minimize construction time and cost besides maximizing quality. In the case study of this
work, a time-cost-quality trade-off problem of a real construction project is examined and solved. In order to solve a
multi-purpose and multi-alternative optimization problem, a genetic algorithm consisting of evolutionary-based meta-
heuristics was used by writing computer code in Matlab. The effect of population size on Pareto value was also
investigated in GA applied to the problem of this study. The number of Pareto images increased with an increase in the
number of populations. Time constraint was created while the iteration number was not considered as a constraint. The
mutation value was used as 0.05 as in general GA applications. Therefore, the optimal results for the problem in the
application project could be obtained with 1000 populations. The number of populations varies according to the data
size and the constraints defined in each problem.

Comparing the results for the populations in the problem, the option with the shortest time period and minimum cost
exist in the application with 100 and 1000 populations. In terms of the cost, the minimum cost with the same value in
all populations and, therefore, the minimum duration for the project was reached. However, for the quality value, the
target has not been met for 50 populations and the quality value remained lower than that of the other populations.
Considering the quality, the maximum quality value was reached with 1000 populations. Taking all three criteria into
consideration, the most appropriate results for this problem were reached with 1000 populations. Huge numbers of
time, cost, and quality combinations, which is highly excessive for being created by the manager, were generated and
this study met, in a general manner, the objective. As a result of the investigations and analyses, it was observed that
there is no single solution to this problem. Obtaining a solution which provides minimum time and cost, and maximum
quality is not possible. Along with the Pareto front, the solution sets are created. The most accurate solution changes
according to the requests of the project manager. Within the set of solutions, the project manager is given the chance
to make a choice according to which time, cost, and quality values take priority or which criteria are prioritized for the
activities which can also be divided into sub-groups.

Choosing the right subcontractor for a specific work is also a crucial decision-making problem in the construction
industry. In this study, a genetic algorithm has been developed to assist the main contractors in the selection of the
most appropriate sub-contractors to carry out different work items considering the trade-off between time-cost-quality
in a construction project. The Pareto-optimal solutions obtained from the analysis showed that the developed GA
algorithm works properly and produces convincing results.

In future research, thousands of the solutions created on the Pareto front can be classified with a different method.
Activities can also be grouped in such a way that they can be evaluated in terms of different selection priorities. The
software can be developed and used actively in the construction industry to carry out this task.

References

Aminbakhsh, S., Sönmez, R., & Bilir, M. (2016). Tamsayılı Doğrusal Programlama Yöntemiyle Kesikli Zaman-Maliyet Ödünleşim Probleminin Optimal
Pareto Çözümü [Optimal pareto solution of discrete time-cost trade-off problem with ınteger linear programming method]. 4th Project and
Construction Management Congress, 281–292.

Bettemir, O. H. (2011). Experimental design for genetic algorithm simulated annealing for time-cost trade-off problems. International Journal of
Engineering & Applied Sciences (IJEAS), 3(1), 15–26.

Bingol, B. N., & Polat, G. (2015). Time-cost-quality trade-off model for subcontractor selection using discrete particle swarm optimization algorithm.
In Thirty-First Annual Conference, September, 13–22.

Dorigo, M., & Stutzle, T. (2004). Ant Colony Optimization. Massachusetts Institute of Technology.

Easa, S. M. (1989). Resource leveling in construction by optimization. Journal of Constructıon Engıneerıng and Management, 115(2), 302–316.

345
El-rayes, B. K. (2001). optimum planning of highway construction A+B bidding method. Journal of Computing in Civil Engineering, 127(August), 261–
269.

El-rayes, K., & Kandil, A. (2005). Time-cost-quality trade-off analysis for highway construction. Journal of Construction Engineering and Management,
131(4), 477–486.

Geem, Z. W. (2010). Multiobjective optimization of time-cost trade-off using harmony search. Journal of Computing in Civil Engineering, 136(June),
711–716.

Goldberg, D. E. (1989). Genetic algorithms in search, optimization, and machine learning (2nd ed.). Addison-Wesley Publishing Company, Inc.

Gomar, J. E., Haas, C. T., & Morton, D. P. (2002). Assignment and allocation optimization of partially multiskilled workforce. Journal of Constructıon
Engıneerıng and Management, 128(April), 103–109.

Hegazy, T. (1999). Optimization of resource allocation and leveling using genetic algorithms. Journal of Constructıon Engıneerıng and Management,
125(June), 167–175.

Hegazy, T., & Wassef, N. (2001). Cost optimization in projects with repetitive nonserial activities. Journal of Construction Engineering and
Management, 127(June), 183–191.

Iscioglu, Y. (2011). Time – cost – quality optimization in a residential project by using fuzzy logic decision – making method. In Doctoral dissertation,
ITU Institute of Science.

Isikyildiz, S. (2019). Time-cost-quality optimization with a meta-heuristic approach in construction projects. (Master’s thesis). Retrieved from Turkey
Council of Higher Education Thesis Center Database. (Accession No. 578147).

Magalhães-mendes, J. (2017). Multiobjective optimization of road construction project time-cost-quality trade-off using genetic algorithms. In
Congress on Numerical Methods in Engineering CMN2017, 3-5 July.

Sari, T. (2008). Metasezgisel yöntemlerle proje çizelgeleme optimizasyonu [Project scheduling optimization with metaheuristic methods]. Marmara
University (Master's thesis). Retrieved from Turkey Council of Higher Education Thesis Center Database. (Accession No. 226628).

Sivanandam, D., & Deepa, S. N. (2008). Introduction to genetic algorithms. Springer Berlin Heidelberg New York.

Sorrentino, M. (2013). Genetic algorithms for construction time-cost-quality trade-off : a road project case study. Construction Management, 4(2),
163–176.

Togan, V., & Eirgash, M. A. (2019). Time-cost trade-off optimization of construction projects using teaching-learning based optimization. KSCE Journal
of Civil Engineering, 23(1), 10–20. https://doi.org/10.1007/s12205-018-1670-6

Yang, X.-S. (2010). Engineering optimization: an ıntroduction with metaheuristic applications. (1st ed.). John Wiley & Sons, Inc.

Zalmai, M. L., Akcay, C., & Manisali, E. (2019). Time-cost optimization using harmony search algorithm in construction projects. Journal of Construction,
226–237. https://doi.org/10.7764/RDLC.18.2.226

Zhang, H., & Li, H. (2010). Multi-objective particle swarm optimization for construction time-cost trade-off problems. Construction Management and
Economics, 6193. https://doi.org/10.1080/01446190903406170

Zheng, D. X. M., Ng, S. T., & Kumaraswamy, M. M. (2005). Applying Pareto ranking and niche formation to genetic algorithm-based multiobjective
time – cost optimization. Journal of Constructıon Engıneerıng and Management, 131(1), 81–91.

Appendix: Concepts of genetic algorithms and definitions of operators

Gene: Each of the parameters of a solution. They are hereditary units that characterize the organism.
Allele: The value of genes.
Locus: The position of the gene on the chromosome.
Chromosome: The sequence formed by the combination of more than one gene is each individual showing possible
solutions.
Population: A collection of chromosomes showing possible solutions to a problem.
Suitability value: The evaluation of each chromosome using the suitability function (often the objective function) is the
performance score obtained.
Selection: Selection of the most successful solutions in a population for reproduction with a selection algorithm based
on the suitability value.
Crossover: Randomly mixed parts of two different solutions to derive new solutions.
Mutation: It is the alteration of the gene structure of a random candidate solution.

346