Studies in Systems, Decision and Control 480

Gebrail Bekdaş
Sinan Melih Nigdeli Editors

Hybrid
Metaheuristics
in Structural
Engineering
Including Machine Learning
Applications
Studies in Systems, Decision and Control

Volume 480

Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
The series “Studies in Systems, Decision and Control” (SSDC) covers both new
developments and advances, as well as the state of the art, in the various areas of
broadly perceived systems, decision making and control–quickly, up to date and
with a high quality. The intent is to cover the theory, applications, and perspectives
on the state of the art and future developments relevant to systems, decision
making, control, complex processes and related areas, as embedded in the fields of
engineering, computer science, physics, economics, social and life sciences, as well
as the paradigms and methodologies behind them. The series contains monographs,
textbooks, lecture notes and edited volumes in systems, decision making and
control spanning the areas of Cyber-Physical Systems, Autonomous Systems,
Sensor Networks, Control Systems, Energy Systems, Automotive Systems,
Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace
Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power
Systems, Robotics, Social Systems, Economic Systems and other. Of particular
value to both the contributors and the readership are the short publication timeframe
and the world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.
Indexed by SCOPUS, DBLP, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.
Gebrail Bekdaş · Sinan Melih Nigdeli
Editors

Hybrid Metaheuristics
in Structural Engineering
Including Machine Learning Applications
Editors
Gebrail Bekdaş Sinan Melih Nigdeli
Department of Civil Engineering Department of Civil Engineering
Istanbul University - Cerrahpaşa Istanbul University - Cerrahpaşa
Istanbul, Türkiye Istanbul, Türkiye

ISSN 2198-4182 ISSN 2198-4190 (electronic)

Studies in Systems, Decision and Control
ISBN 978-3-031-34727-6 ISBN 978-3-031-34728-3 (eBook)
https://doi.org/10.1007/978-3-031-34728-3

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2023

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

In structural engineering design, engineers search for a better design than before.
Also, reaching this design must be faster than before. In this case, to consider all
design constraints, it is needed to use iterative methods. As iterative methods, meta-
heuristics are a very useful tool. Instead of generating new metaheuristics, several
features of the existing ones may be combined to generate hybrid algorithms. Thus, an
improvement in convergence and best optimum results may be provided. In addition
to that, machine learning techniques can be used or combined with metaheuristics
as a hybrid method. In that case, prediction models can be developed to save time
without making an iterative process.
This book includes reviews and applications of hybrid metaheuristic algorithms
and machine learning used in structural engineering. It contains 14 chapters including
an overview and introduction. 14 of 13 chapters are presented in two parts, namely,
Part I: Hybrid Metaheuristics and Part II: Machine Learning. In Part I, a review and
seven structural engineering applications including reinforced concrete, truss struc-
tures, tuned mass damper, composite structures, and dam structures are given. In Part
II, two reviews and four machine learning applications about structural engineering
problems, reinforced concrete, and building information modeling are given.

Istanbul, Türkiye Gebrail Bekdaş

March 2023 Sinan Melih Nigdeli

v
Contents

Introduction and Overview: Hybrid Metaheuristics in Structural

Engineering—Including Machine Learning Applications . . . . . . . . . . . . . . 1
Gebrail Bekdaş and Sinan Melih Nigdeli

Hybrid Metaheuristics
The Development of Hybrid Metaheuristics in Structural
Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Aylin Ece Kayabekir, Sinan Melih Nigdeli, and Gebrail Bekdaş
Optimum Design of Reinforced Concrete Columns in Case of Fire . . . . . 35
Uğur Günay, Serdar Ulusoy, Gebrail Bekdaş, and Sinan Melih Nigdeli
Hybrid Social Network Search and Material Generation Algorithm
for Shape and Size Optimization of Truss Structures . . . . . . . . . . . . . . . . . . 49
M. Saraee, A. Jafari, D. Yazdani, M. Baghalzadeh Shishehgarkhaneh,
B. Nouhi, and S. Talatahari
Development of a Hybrid Algorithm for Optimum Design
of a Large-Scale Truss Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Melda Yücel, Gebrail Bekdaş, and Sinan Melih Nigdeli
Structural Control Systems and Tuned Mass Damper Optimization
by Using Jaya and Hybrid Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Muhammed Çoşut, Sinan Melih Nigdeli, and Gebrail Bekdaş
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete
Filled Steel Tubular Stub Columns Based on CO2 Emission . . . . . . . . . . . 111
Celal Cakiroglu, Kamrul Islam, and Gebrail Bekdaş
Optimum Design of Dam Structures Using Multi-objective Chaos
Game Optimization Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A. Jafari, M. Saraee, B. Nouhi, M. Baghalzadeh Shishehgarkhaneh,
and S. Talatahari

vii
viii Contents

Machine Learning
The State of Art in Machine Learning Applications in Civil
Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Yaren Aydin, Gebrail Bekdaş, Ümit Işıkdağ, and Sinan Melih Nigdeli
Machine Learning Application of Structural Engineering Problems . . . . 179
Ayla Ocak, Sinan Melih Nigdeli, Gebrail Bekdaş, and Ümit Işıkdağ
Modeling Civil Engineering Problems via Hybrid Versions
of Machine Learning and Metaheuristic Optimization Algorithms . . . . . 199
Vahdettin Demir, Esra Uray, and Serdar Carbas
Comparison of Multilayer Perceptron and Other Methods
for Prediction of Sustainable Optimum Design of Reinforced
Concrete Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Yaren Aydın, Gebrail Bekdaş, Sinan Melih Nigdeli, Ümit Işıkdağ,
and Zong Woo Geem
Artificial Intelligence and Deep Learning in Civil Engineering . . . . . . . . . 265
Ayla Ocak, Sinan Melih Nigdeli, Gebrail Bekdaş, and Ümit Işıkdağ
Deep Learning-Based Framework for Reconstruction
and Optimisation of Building Information Models Containing
Parametric Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Vincent J. L. Gan
Introduction and Overview: Hybrid
Metaheuristics in Structural
Engineering—Including Machine
Learning Applications

Gebrail Bekdaş and Sinan Melih Nigdeli

Abstract For the first chapter of the book entitled “Hybrid Metaheuristics in Struc-
tural Engineering—Including Machine Learning Applications”, an introduction and
overview chapter is given. This chapter includes the importance of optimization in
civil engineering as an introduction. Then, metaheuristics are defined as a general
frame. Then, machine learning is mentioned. Finally, an overview of the content of
the book was given.

Keywords Optimization · Metaheuristic algorithms · Hybrid algorithms ·

Structural engineering · Machine learning

1 Introduction

Important factors must be addressed during the design of an engineering structure.

These key factors are economics, safety, usage, and architecture. It is not enough to
consider one or just a few of these important factors to validate that a design is an
optimally designed object. A perfectly done engineering design must consider all of
these important aspects to meet the needs of individuals. In this case, an engineer is
under a heavy burden to think through all the problems to find the perfect balance
between these factors that have generally opposite effects on design objectives. Only
experienced engineers can handle all the elements with near perfection. To take
these factors into account, various measures should be established and these should
be formulated in the design. Since these factors are interrelated, design problems
are often non-linear. Because of that, various methods that are specific to the design
problem must be developed. Because of their non-linearity, these problems can be
approximately addressed using conventional methods. Due to the increasing demands

G. Bekdaş (B) · S. M. Nigdeli

Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_1
2 G. Bekdaş and S. M. Nigdeli

and the advantages of technology, we need a design that is beyond an approximate

solution.
A few nonlinearities can be handled with difficulty or approximation, or a classical
solution cannot be derived without the assumption of a few factors. In this case,
experience will have an important role in good design. For a precise and optimal
design, the evaluation of the design phases must be iteratively repeated and it can be
done by using an algorithm.
Civil engineering produces products that are directly effective on individuals. It is
one of the fundamental engineering disciplines. For that reason, civil engineers’ prod-
ucts must be economical, safe, and comfortable. In that case, optimization has become
a necessity. Civil engineering has many applications, as all designs that provide a
living environment fall within the scope of civil engineering. Optimization is needed
for all sub-branches of civil engineering including structural engineering, struc-
tural mechanics, building materials, building management, geotechnical engineering,
transportation engineering and hydraulics. As examples, the applications cover water
resources, superstructure and infrastructure projects, stabilization of landfills and soil
improvement, public transportation and resources, traffic in transportation, etc.
Civil engineering designs are affected by environmental conditions and the world
has different geographical areas with different resources and different natural disaster
risks. While the most important issue in the design of civil engineering applications
is safety, other issues are the preferability index.
Structural mechanics theory and experimental research for the safety of designs
have been formulated in various arrangements. Although the theory and experiments
are generally the same in the regulations, the standards are different in different
countries or there are different standard annexes in different countries due to the
material production quality, workmanship quality and specific natural disasters in the
region. Legislative differences are compounded by the demands of people and social
authorities, including resource opportunities in a region. In this case, the optimum
design for one region or person may not be the best acceptable design for another.
Therefore, these factors are considered specific design constraints and design variable
ranges.
Optimizing a mechanical engineering product and a civil engineering structure
can differ greatly in assigning optimum design results. Generally, using continuous
design variables is the best option in search of the best variables. A machine part or
an object can be produced in a factory with precise dimensional values. Precisely-
sized building construction cannot be achieved at a construction site. In this case,
discrete variables may be used for structural design or continuous values should be
rounded to the practical size. In this case, the exact optimum design can never be used
for structural design. Especially, the construction of reinforced concrete structures
involves pouring concrete on site and placing the fixed-sized reinforcements that are
available in local markets. This situation requires design variables to be assigned
specific values. Therefore, the reinforcing steel dimensions should be selected from
the dimensions defined by the designer during the optimization processes. In addition,
Introduction and Overview: Hybrid Metaheuristics in Structural … 3

profile dimensions for steel structures are produced according to the fixed dimen-
sions of local markets. The optimum design of structures includes the best possible
dimensions and design detail practical in construction.
In the regulation, design rules have to be defined as design constraints. These
constraints can only be calculated if the values of the design variables are known.
For that reason, minimization of an objective function is not possible using linear
mathematical methods and the optimization problem of finding the best vari-
ables is a nonlinear problem. Also for nonlinear methods, the existence of various
design constraints (stress limits, shear capacity, minimum and maximum limits, and
ductility conditions) and variables (cross-section size and amount of reinforcement
for different stress and section types) makes optimization problem to be complex.
In this case, numerical optimization methods using iterative analysis are suitable for
optimization.
Metaheuristic-based optimum design methodologies have been developed for civil
engineering problems. In general, these algorithms formulate a process, event or
behavior from social life or nature. Similarly, every process, event or behavior has
an ultimate purpose, such as optimization objectives in engineering problems.
As a first experimental idea, Galilei Galileo worked in 1665 to improve the shape
of beams as optimization [1]. Academically, early iterative computational efforts to
optimize a beam to minimize the weight of beams were made by Haug and Krimser
[2] using design constraints of stress and deflection. Venkaya [3] proposed a search
procedure that uses energy criteria to optimize structures including beams, frames
and trusses while fully considering the design constraints on the dimensions, stress
and displacement of the members under multiple loading conditions.
For the optimum design of reinforced concrete (RC) beams, Friel [4] developed
an optimum design formulation for the ratio of steel considering that the reinforced
concrete beam is subjected to moment only. Chou [5] optimized the depth and bending
reinforcement area of reinforced concrete T-beam sections using the Lagrange multi-
pliers method. Krishnamoorthy and Munro [6] proposed a linear programming model
that optimizes reinforced concrete frames for constraints about compatibility, limited
ductility, stability, and serviceability. Kirsch [7] developed a three-level iterative
methodology for cost optimization of a multi-span continuous RC beam. Laksh-
manan and Parameswaran [8] determined span-to-effective depth ratios to avoid
trial-and-error approaches in the optimum bending design of reinforced concrete
sections. Prakash, Agarwala and Singh [9] developed simple optimization tech-
niques for single and double-reinforced beams, T-beams and eccentrically loaded
columns. Hoit, Soeiro and Fagundo [10] used augmented Lagrange multipliers and
nonlinear programming techniques to minimize the weight of frames. Chakrabarty
[11] proposed a model using geometric programming and Newton–Raphson methods
for optimization with various constraints. Al-Salloum and Siddiqi [12] obtained
a closed-form solution for the area and depth according to the cost and strength
parameters by taking the derivatives of the augmented Lagrangian function. Chung
and Sun [13] used computational methods including sequential linear programming
and gradient projection method for reinforced concrete beams. Adamu et al. [14]
proposed a continuous-type optimality criteria method for RC beams.
4 G. Bekdaş and S. M. Nigdeli

2 Metaheuristics

The word metaheuristic is derived from the Greek word heuriskein, which means
to find or explore, and the word meta, which means upper, advanced, advanced,
etc. Many real-world applications have many more complex factors and parameters.
Also, the constraints that will affect the behavior and solution of the system, require a
specific approach generated via basic scientific thinking of optimum resources in any
field [15]. Additionally, the complexity of the problem directs us to iteratively search
for any possible solution or combination of solutions. Therefore, finding suitable
solutions for the problem within an acceptable duration becomes the main success.
Due to that, the uncertainty and irregular structure of metaheuristic approaches can
be solutions for various applications and problems that may need a long time to solve.
Generally, heuristic algorithms rely on local search procedures around the existing
solution and a better solution is iteratively tried to be found. This search is usually
terminated when the first local optimum is achieved. On the other hand, metaheuristic
methods combine some heuristic approaches and provide better solutions than the
previously found local optimums via local search [16].
The main components of any metaheuristic algorithm are concentration/
exploitation and diversification/discovery. Diversification is researching the entire
search area and producing various solutions Concentration means focusing on the
search (local search) in that region, using the knowledge of the chosen solution by
searching for the best available solutions in any region, and choosing the best candi-
date solutions. While randomization is beneficial in diversifying solutions, it avoids
being trap in the local optimum. The diversification phase enables the algorithm to
explore the search space more efficiently [17].
On the other hand, these methods are inspired by various things. These may be
natural features, special abilities of living things in their lives, some physical and
chemical events, evolution/genetic processes and natural events.
When we look at nature in general, it is understood that living things have charac-
teristics that are innate or developed for various purposes such as survival, nutrition,
and the continuation of species. It is possible to show many examples of these, such
as the fox making use of the earth’s magnetic field while hunting, the chameleon
changing color to hide from danger, the cuckoo using other birds’ nests for the
continuation of its species, the hedgehog stretching themselves and throwing their
thorns in danger. When all these processes are examined, it is seen that living things
change their defense or attack mechanisms in accordance with the conditions, and
they use a kind of heuristic optimization specific to the species that enables them to
use the limited opportunities they have in the most appropriate time and manner in
order to maintain their vital activities [18].
These intuitive optimization processes of living things in nature have attracted
the attention of researchers working on basic sciences and have produced various
algorithms that express these processes mathematically.
Introduction and Overview: Hybrid Metaheuristics in Structural … 5

Table 1 Trajectory-based
Algorithm Abbreviation Year Reference
metaheuristic algorithms
Simulated annealing SA 1983 [19]
Tabu search TS 1989 [20]
Variable neighborhood search VNS 1997 [21]
Guided local search GLS 1998 [22]
Iterative local search ILS 2003 [23]

Table 2 Evolutionary-based
Algorithm Abbreviation Year Reference
metaheuristic algorithms
Genetic algorithm GA 1992 [24]
Differential evaluation DE 1997 [25]
Clonal selection CSA 2000 [26]
algorithm
Harmony search HS 2001 [27]
Backtracking search BSA 2013 [28]
algorithm
Stochastic fractal SFF 2015 [29]
search
Across neighborhood ANS 2016 [30]
search

The oldest metaheuristics are trajectory-based ones. As given in Table 1, these

algorithms include simulated annealing inspired by the annealing process of metal
and tabu search using the human brain.
The mostly used metaheuristics are evolutionary-based that are generally inspired
by genetics. Also, an evolution process can be formulated. For example, harmony
search imitated from the music that is tried to be better for the audience. Common
evolutionary-based algorithms are listed in Table 2.
Most of the metaheuristic algorithms are nature inspired. As a subcategory, the
algorithm may use a different source in nature. As listed in Table 3, the subcategories
are bio-inspired, swarm-based, plant-based, physics/chemistry-based and human-
based.
Metaheuristics have different inspirations and all processes are formulated with
different formulations, but the phases of processes may show similar formulations. In
that case, problem-solving is more important than imitation. Thus, different features
of these algorithms can be combined to generate a new hybrid algorithm without
the need for a metaphor as an inspiration. The period that will be orientated on the
scientific problem will be the future period [92].
6 G. Bekdaş and S. M. Nigdeli

Table 3 Nature-based metaheuristic algorithms

Algorithm Abbreviation Subcategory Year Reference
Particle swarm PSO Swarm-based 1995 [31]
optimization
Artificial bee colony ABC Bio-inspired 2005 [32]
Ant colony optimization ACO Bio-inspired 2006 [33]
Glowworm swarm GSO Swarm-based 2006 [34]
optimization
Shuffled frog leaping SFLA Bio-inspired 2006 [35]
algorithm
Invasive weed IWO Plant-based 2006 [36]
optimization
Seeker optimization SOA Human-based 2006 [37]
algorithm
Imperialistic competitive ICA Human-based 2007 [38]
algorithm
Biogeography based BBO Human-based 2008 [39]
optimization
Firefly algorithm FA Bio-inspired 2008 [40]
Intelligent water drops IWD Swarm-based 2008 [41]
Monkey algorithm MA Bio-inspired 2008 [42]
Cuckoo search CS Bio-inspired 2009 [43]
Group search optimizer GSO Swarm-based 2009 [44]
Hunting search HS Swarm-based 2009 [45]
Chemical reaction CRO Physics/chemistry-based 2009 [46]
optimization
Bat algorithm BA Bio-inspired 2010 [47]
Charged system search CSS Physics/Chemistry-based 2010 [48]
Cuckoo optimization COA Bio-inspired 2011 [49]
algorithm
Teaching–learning-based TLBO Human-based 2011 [50]
optimization
Krill herd algorithm KHA Bio-inspired 2012 [51]
Migrating birds MBO Swarm-based 2012 [52]
optimization
Black hole algorithm BH Physics/chemistry-based 2013 [53]
Dolphin echolocation DE Bio-inspired 2013 [54]
Animal migration AMO Swarm-based 2013 [55]
optimization
SDA optimization SDA Bio-inspired 2014 [56]
algorithm
(continued)
Introduction and Overview: Hybrid Metaheuristics in Structural … 7

Table 3 (continued)
Algorithm Abbreviation Subcategory Year Reference
Artificial root foraging ARFA Plant-based 2014 [57]
algorithm
Chicken swarm CSO Bio-inspired 2014 [58]
optimization
Flower pollination FPA Plant-based 2014 [59]
algorithm
Radial movement RMO Swarm-based 2014 [60]
optimization
Spider monkey SMO Bio-inspired 2014 [61]
optimization
Elephant search algorithm ESA Bio-inspired 2015 [62]
Grey wolf optimizer GWO Bio-inspired 2015 [63]
Jaguar algorithm JA Bio-inspired 2015 [64]
Locust swarm algorithm LSA Swarm-based 2015 [65]
Moth-flame optimization MFO Bio-inspired 2015 [66]
Vortex search algorithm VSA Physics/chemistry-based 2015 [67]
Water wave optimization WWA Physics/chemistry-based 2015 [68]
Ant lion optimizer ALO Bio-inspired 2015 [69]
African buffalo ABO Swarm-based 2015 [70]
optimization
Lightning search algorithm LSA Physics/chemistry-based 2015 [71]
Crow search algorithm CSA Bio-inspired 2016 [72]
Electromagnetic field EFO Physics/Chemistry-based 2016 [73]
optimization
Joint operations algorithm JOA Swarm-based 2016 [74]
Lion optimization LOA Bio-inspired 2016 [75]
algorithm
Sine cosine algorithm SCA Physics/chemistry-based 2016 [76]
Virus colony search VCS Bio-inspired 2016 [77]
Whale optimization WOA Bio-inspired 2016 [78]
algorithm
Red deer algorithm RDA Bio-inspired 2016 [79]
Phototropic optimization POA Plant-based 2018 [80]
algorithm
Coyote optimization COA Swarm-based 2018 [81]
algorithm
Owl search algorithm OSA Bio-inspired 2018 [82]
Squirrel search algorithm SSA Bio-inspired 2018 [83]
(continued)
8 G. Bekdaş and S. M. Nigdeli

Table 3 (continued)
Algorithm Abbreviation Subcategory Year Reference
Social engineering SEO Human-based 2018 [84]
optimizer
Emperor penguin EPO Bio-inspired 2018 [85]
optimizer
Future search algorithm FSA Human-based 2019 [86]
Emperor penguins colony EPC Swarm-based 2019 [87]
Thermal exchange TEO Physics/chemistry-based 2019 [88]
optimization
Harris Hawks optimization HHO Bio-inspired 2019 [89]
Political optimizer PO Human-based 2020 [90]
Heap-based optimizer HBO Human-based 2020 [91]

3 Artificial Intelligence and Machine Learning

Machine learning, one of the sub-disciplines of artificial intelligence, is the process

of focusing on various methods and algorithms for providing learning activity by
computers. Here, in the machine learning process, when a sample data set is given
to any system, learning can be performed by generating information about the data
and the connections/relationships between the data. In this context, the concept of
machine learning can be defined as the development of systems that can learn, feel and
create a reaction/response for a situation by enabling machines, computers, software
and even robots to perform many activities that can only be done by a person with
real intelligence and mind.
Machine learning has become important in many working disciplines, especially
in recent years, and in this context, smart systems, software and tools have been
developed by using various learning techniques and algorithms. In this sense, the
machine learning technology in question can be used by researchers, designers,
engineers, etc. during the acquisition of some necessary outputs or the finding of
parameters/information. In addition to providing ease of use for anyone or anyone
interested in this field, it provides many advantages in terms of cost and time savings
and efficient expenditure of labor.
Thanks to the machine learning processes carried out, time and effort are saved by
leading the necessary activities in the scientific field as well as in many areas of daily
life. These methods have been used in several health services including determining
mental health [93–95], estimating the probability of occurrence of some diseases
[96–98], detecting cancerous cells in organisms [99–101], etc.
On the other hand, the engineering discipline is one of the fields where machine
learning applications are already being used and are becoming more widespread day
by day. In this context, estimating the water demand for existing users [102, 103],
energy planning to ensure efficiency [104, 105] and obtaining some mechanical
properties of structural materials [106–108] can be given as recent examples.
Introduction and Overview: Hybrid Metaheuristics in Structural … 9

In structural engineering, machine learning is generally used with metaheuristic

algorithms to generate a hybrid process. In this process, metaheuristics are employed
for the optimization and data of several design constants are generated. Then, this
data is used in machine learning. Thus, an artificial intelligence model that predicts
optimum results of a case of design constants without a rerun of the iterative process
is generated to save time. Examples of these applications are estimating the optimal
model parameters of structural elements [109–111] and estimation of optimized
mechanical parameters for tuned mass dampers (TMD) [112–114].

4 Overview of the Book Content

The book entitled “Hybrid Metaheuristics in Structural Engineering—Including

Machine Learning Applications” is organized into two parts. The first part is about
the optimization of structural engineering using hybrid metaheuristics. Chapter 2
summarised the recent developments in structural optimization using hybrid meta-
heuristics. Then, reinforced concrete columns are optimized via a hybrid algorithm
by considering the case of fire. Chapters 4 and 5 are about the optimization of truss
structures. As Chap. 6, a passive structural control that is tuned mass damper is
optimized via hybrid metaheuristics. A concrete-filled steel tubular (CFST) struc-
ture optimization is explained in Chap. 7. A multi-objective optimization of dam
structures is presented in Chap. 8.
The second part is related to prediction applications using machine learning.
As the first chapter of the second part, the current state of art in machine learning
in civil engineering is presented as Chap. 9. Then, several structural engineering
applications such as tubular column and I-beam optimization examples are presented
for the machine learning prediction model in Chap. 10. In Chap. 11, metaheuristic
optimization algorithms and machine learning techniques are hybridized to present
civil engineering problems. Chapter 12 presents a prediction application for the
eco-friendly optimum design of reinforced concrete columns. Then, a review of
artificial intelligence and deep learning is given in Chap. 13. Finally, a deep learning
application about building information modeling is generated in Chap. 14.

References

1. Galilei, G.: Dialogues Concerning Two New Sciences. Northwestern University Press,
Evanston, IL (originally published in 1665) (1950)
2. Haug, E.J., Kirmser, P.G.: Minimum weight design of beams with inequality constraints on
stress and deflection. J. Appl. Mech. 34(4), 999–1004 (1967)
3. Venkayya, V.B.: Design of optimum structures. Comput. Struct. 1(1–2), 265–309 (1971)
4. Friel, L.L.: Optimum singly reinforced concrete sections. J. Proc. 71(11), 556–558 (1974,
November)
10 G. Bekdaş and S. M. Nigdeli

5. Chou, T.: Optimum reinforced concrete T-beam sections. J. Struct. Div. 103(ASCE 13120)
(1977)
6. Krishnamoorthy, C.S., Munro, J.: Linear program for optimal design of reinforced concrete
frames. Proc. IABSE 3(1), 119–141 (1973)
7. Kirsch, U.: Multilevel optimal design of reinforced concrete structures. Eng. Optim. 6(4),
207–212 (1983)
8. Lakshmanan, N., Parameswaran, V.S.: Minimum weight design of reinforced concrete
sections for flexure. J. Inst. Eng. India. Civ. Eng. Div. 66(2), 92–98 (1985)
9. Prakash, A., Agarwala, S.K., Singh, K.K.: Optimum design of reinforced concrete sections.
Comput. Struct. 30(4), 1009–1011 (1988)
10. Hoit, M., Soeiro, A., Fagundo, F.: Probabilistic design and optimization of reinforced concrete
frames. Eng. Optim. 17(3), 229–235 (1991)
11. Chakrabarty, B.K.: Models for optimal design of reinforced concrete beams. Comput. Struct.
42(3), 447–451 (1992)
12. Al-Salloum, Y.A., Husainsiddiqi, G.: Cost-optimum design of reinforced concrete (RC)
beams. Struct. J. 91(6), 647–655 (1994)
13. Chung, T.T., Sun, T.C.: Weight optimization for flexural reinforced concrete beams with static
nonlinear response. Struct. Optim. 8(2–3), 174–180 (1994)
14. Adamu, A., Karihaloo, B.L., Rozvany, G.I.N.: Minimum cost design of reinforced concrete
beams using continuum-type optimality criteria. Struct. Optim. 7(1–2), 91–102 (1994)
15. Koziel, S., Yang, X.S. (eds.): Computational Optimization, Methods and Algorithms, vol.
356. Springer-Verlag, Heidelberg, Berlin (2011). ISBN: 978-3-642-20858-4
16. Onwubolu, G.C., Babu, B.V.: New Optimization Techniques in Engineering, vol. 141.
Springer-Verlag, Heidelberg, Berlin (2004). ISBN: 978-3-540-39930-8
17. Gandomi, A.H., Yang, X.S., Talatahari, S., Alavi, A.H. (eds.): Metaheuristic Algorithms in
Modeling and Optimization, Metaheuristic Applications in Structures and Infrastructures.
Elsevier, pp. 1–24 (2013). ISBN: 9780123983640
18. Bekdaş, G., Nigdeli, S.M., Yücel, M., Kayabekir, A.E.: Yapay Zeka Optimizasyon Algorit-
maları ve Mühendislik Uygulamaları. Seçkin, Ankara, Turkey (2021)
19. Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P.: Optimization by simulated annealing. Science
220(4598), 671–680 (1983)
20. Glover, F.: Tabu Search—Part I. ORSA J. Comput. 1(3), 190–206 (1989)
21. Mladenović, N., Hansen, P.: Variable neighborhood search. Comput. Oper. Res. 24(11), 1097–
1100 (1997)
22. Balas, E., Vazacopoulos, A.: Guided local search with shifting bottleneck for job shop
scheduling. Manag. Sci. 44(2), 262–275 (1998)
23. Lourenço, H.R., Martin, O.C., Stützle, T.: Iterated local search. In: Glover, F., Kochenberger,
G.A. (eds.) Handbook of Metaheuristics. International Series in Operations Research &
Management Science. Springer US, Boston, MA, pp. 320–353 (2003)
24. Holland, J.H.: Adaptation in natural and artificial systems: an introductory analysis with
applications to biology, control, and artificial intelligence, 1st edn. MIT Press, Cambridge,
Mass. (1992)
25. Storn, R.; Price, K.: Differential evolution—a simple and efficient heuristic for global
optimization over continuous spaces. J. Glob. Optim. 11(4), 341–359 (1997)
26. de Castro, L.N., Von Zuben, F.J.: Learning and optimization using the clonal selection
principle. IEEE Trans. Evol. Comput. 6(3), 239–251 (2002)
27. Geem, Z.W., Kim, J.H., Loganathan, G.V.: A new heuristic optimization algorithm: harmony
search. Simulation 76(2), 60–68 (2001)
28. Civicioglu, P.: Backtracking search optimization algorithm for numerical optimization
problems. Appl. Math. Comput. 219(15), 8121–8144 (2013)
29. Salimi, H.: Stochastic fractal search: a powerful metaheuristic algorithm. Knowl-Based Syst
75, 1–18 (2015)
30. Wu, G.: Across neighborhood search for numerical optimization. Information Sciences. Spec.
Issue Discov. Sci. 329, 597–618 (2016)
Introduction and Overview: Hybrid Metaheuristics in Structural … 11

31. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of ICNN’95—Inter-
national Conference on Neural Networks, vol. 4, pp. 1942–1948 (1995)
32. Karaboga, D., Basturk, B.: A powerful and efficient algorithm for numerical function
optimization: artificial bee colony (ABC) algorithm. J. Glob. Optim. 39(3), 459–471 (2007)
33. Dorigo, M., Birattari, M., Stutzle, T.: Ant colony optimization. IEEE Comput. Intell. Mag.
1(4), 28–39 (2006)
34. Krishnanand, K.N., Ghose, D.: Detection of multiple source locations using a glowworm
metaphor with applications to collective robotics. In: Proceedings 2005 IEEE Swarm
Intelligence Symposium, SIS 2005, pp. 84–91 (2005)
35. Eusuff, M., Lansey, K., Pasha, F.: Shuffled frog-leaping algorithm: a memetic meta-heuristic
for discrete optimization. Eng. Optim. 38(2), 129–154 (2006)
36. Mehrabian, A.R., Lucas, C.: A novel numerical optimization algorithm inspired from weed
colonization. Ecol. Inform. 1(4), 355–366 (2006)
37. Dai, C., Zhu, Y., Chen, W.: Seeker optimization algorithm. In: Wang, Y., Cheung, Y.-M., Liu,
H. (eds.) Computational Intelligence and Security. Lecture Notes in Computer Science, vol.
4456. Springer, Berlin, Heidelberg, pp. 167–176 (2007)
38. Atashpaz-Gargari, E., Lucas, C.: Imperialist competitive algorithm: an algorithm for opti-
mization inspired by imperialistic competition. IEEE Congr. Evol. Comput. 2007, 4661–4667
(2007)
39. Simon, D.: Biogeography-based optimization. IEEE Trans. Evol. Comput. 12(6), 702–713
(2008)
40. Yang, X.-S.: Firefly algorithms for multimodal optimization. In: Watanabe, O., Zeugmann,
T. (eds.) Stochastic Algorithms: Foundations and Applications. Lecture Notes in Computer
Science, vol. 5792. Springer, Berlin, Heidelberg, pp. 169–178 (2009)
41. Hosseini, H.S.: The intelligent water drops algorithm: a nature-inspired swarm-based
optimization algorithm. Int. J. Bio-Inspired Comput. 1(1/2), 71 (2009)
42. Zhao, R.Q., Tang, W.S.: Monkey algorithm for global numerical optimization. J. Uncertain
Syst. 2(3), 164–175 (2008)
43. Yang, X.-S., Deb, S.: Cuckoo search via Lévy flights. In: 2009 World Congress on Nature
Biologically Inspired Computing (NaBIC), pp. 210–214 (2009)
44. He, S., Wu, Q.H., Saunders, J.R.: Group search optimizer: an optimization algorithm inspired
by animal searching behavior. IEEE Trans. Evol. Comput. 13(5), 973–990 (2009)
45. Oftadeh, R., Mahjoob, M.J., Shariatpanahi, M.: A novel meta-heuristic optimization algorithm
inspired by group hunting of animals: hunting search. Comput. Math. Appl. 60(7), 2087–2098
(2010)
46. Lam, A.Y.S., Li, V.O.K.: Chemical-reaction-inspired metaheuristic for optimization. IEEE
Trans. Evol. Comput. 14(3), 381–399 (2010)
47. Yang, X.-S. A new metaheuristic bat-inspired algorithm. In: González, J.R., Pelta, D.A.; Cruz,
C., Terrazas, G. (eds.) Nature Inspired Cooperative Strategies for Optimization (NICSO 2010).
Studies in Computational Intelligence, vol. 284. Springer, Berlin, Heidelberg, pp. 65–74
(2010)
48. Kaveh, A., Talatahari, S.: A novel heuristic optimization method: charged system search. Acta
Mech. 213(3), 267–289 (2010)
49. Rajabioun, R.: Cuckoo optimization algorithm. Appl. Soft Comput. 11(8), 5508–5518 (2011)
50. Rao, R.V., Savsani, V.J., Vakharia, D.P.: Teaching–learning-based optimization: a novel
method for constrained mechanical design optimization problems. Comput.-Aided Des. 43(3),
303–315 (2011)
51. Gandomi, A.H., Alavi, A.H.: Krill herd: a new bio-inspired optimization algorithm. Commun.
Nonlinear Sci. Numer. Simul. 17(12), 4831–4845 (2012)
52. Duman, E., Uysal, M., Alkaya, A.F.: Migrating birds optimization: a new metaheuristic
approach and its performance on quadratic assignment problem. Inf. Sci. 217, 65–77 (2012)
53. Hatamlou, A.: Black hole: a new heuristic optimization approach for data clustering. Inf. Sci.
(Including Special Section on New Trends in Ambient Intelligence and Bio-inspired Systems)
222: 175–184 (2013)
12 G. Bekdaş and S. M. Nigdeli

54. Kaveh, A., Farhoudi, N.: A new optimization method: dolphin echolocation. Adv. Eng. Softw.
59, 53–70 (2013)
55. Li, X., Zhang, J., Yin, M.: Animal migration optimization: an optimization algorithm inspired
by animal migration behavior. Neural Comput. Appl. 24(7), 1867–1877 (2014)
56. Chandra, V.: Smell detection agent based optimization algorithm. J. Inst. Eng. India Ser. B
97(3), 431–436 (2014)
57. Ma, L., Hu, K., Zhu, Y., Chen, H., He, M.: A novel plant root foraging algorithm for image
segmentation problems. Math. Probl. Eng. 2014, 1–16 (2014)
58. Meng, X., Liu, Y., Gao, X., Zhang, H.: A new bio-inspired algorithm: chicken swarm opti-
mization. In: Tan, Y., Shi, Y., Coello, C.A. (eds.) Advances in Swarm Intelligence. Lecture
Notes in Computer Science, vol. 8794. Springer International Publishing, Cham, pp. 86–94
(2014)
59. Yang, X.-S.: Flower pollination algorithm for global optimization. In: Durand-Lose, J.,
Jonoska, N. (eds.) Unconventional Computation and Natural Computation. Lecture Notes
in Computer Science, vol. 7445. Springer, Berlin, Heidelberg. pp. 240–249 (2012)
60. Rahmani, R., Yusof, R.: A new simple, fast and efficient algorithm for global optimization
over continuous search-space problems: radial movement optimization. Appl. Math. Comput.
248, 287–300 (2014)
61. Bansal, J.C., Sharma, H., Jadon, S.S., Clerc, M.: Spider monkey optimization algorithm for
numerical optimization. Memetic Comput. 6(1), 31–47 (2014)
62. Deb, S., Fong, S., Tian, Z.: Elephant search algorithm for optimization problems. In: 2015
Tenth International Conference on Digital Information Management (ICDIM), pp. 249–255
(2015)
63. Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61
(2014)
64. Chen, C.-C., Tsai, Y.-C., Liu, I.-I., Lai, C.-C., Yeh, Y.-T., Kuo, S.-Y., Chou, Y.-H.: A
novel metaheuristic: jaguar algorithm with learning behavior. In: 2015 IEEE International
Conference on Systems, Man, and Cybernetics, pp. 1595–1600 (2015)
65. Cuevas, E., González, A., Zaldívar, D., Cisneros, M.P.: An optimisation algorithm based on
the behaviour of locust swarms. Int. J. Bio-Inspired Comput. 7(6), 402 (2015)
66. Mirjalili, S.: Moth-flame optimization algorithm: a novel nature-inspired heuristic paradigm.
Knowl.-Based Syst. 89, 228–249 (2015)
67. Doğan, B., Ölmez, T.: A new metaheuristic for numerical function optimization: vortex search
algorithm. Inf. Sci. 293, 125–145 (2015)
68. Zheng, Y.-J.: Water wave optimization: a new nature-inspired metaheuristic. Comput. Oper.
Res. 55, 1–11 (2015)
69. Mirjalili, S.: The ant lion optimizer. Adv. Eng. Softw. 83, 80–98 (2015)
70. Odili, J.B., Kahar, M.N.M., Anwar, S.: African buffalo optimization: a swarm-intelligence
technique. In: Procedia Computer Science. 2015 IEEE International Symposium on Robotics
and Intelligent Sensors (IEEE IRIS2015), vol. 76, pp. 443–448 (2015)
71. Shareef, H., Ibrahim, A.A., Mutlag, A.H.: Lightning search algorithm. Appl. Soft Comput.
36, 315–333 (2015)
72. Askarzadeh, A.: A novel metaheuristic method for solving constrained engineering optimiza-
tion problems: crow search algorithm. Comput. Struct. 169, 1–12 (2016)
73. Abedinpourshotorban, H., Mariyam Shamsuddin, S., Beheshti, Z., Jawawi, D.N.A.: Electro-
magnetic field optimization: a physics-inspired metaheuristic optimization algorithm. Swarm
Evol. Comput. 26, 8–22 (2016)
74. Sun, G., Zhao, R., Lan, Y.: Joint operations algorithm for large-scale global optimization.
Appl. Soft Comput. 38, 1025–1039 (2016)
75. Yazdani, M., Jolai, F.: Lion optimization algorithm (LOA): a nature-inspired metaheuristic
algorithm. J. Comput. Des. Eng. 3(1), 24–36 (2016)
76. Mirjalili, S.: SCA: a sine cosine algorithm for solving optimization problems. Knowl.-Based
Syst. 96, 120–133 (2016)
Introduction and Overview: Hybrid Metaheuristics in Structural … 13

77. Li, M.D., Zhao, H., Weng, X.W., Han, T.: A novel nature-inspired algorithm for optimization:
virus colony search. Adv. Eng. Softw. 92, 65–88 (2016)
78. Mirjalili, S., Lewis, A.: The whale optimization algorithm. Adv. Eng. Softw. 95, 51–67 (2016)
79. Fathollahi-Fard, A.M., Hajiaghaei-Keshteli, M., Tavakkoli-Moghaddam, R.: Red deer algo-
rithm (RDA): a new nature-inspired meta-heuristic. Soft Comput. 24(19), 14637–14665
(2020)
80. Vinod, C.S.S., Anand, H.S.: Phototropic algorithm for global optimisation problems. Appl.
Intell. 51(8), 5965–5977 (2021)
81. Pierezan, J., Dos Santos Coelho, L.: Coyote optimization algorithm: a new metaheuristic for
global optimization problems. In: 2018 IEEE Congress on Evolutionary Computation (CEC),
pp. 1–8 (2018)
82. Jain, M., Maurya, S., Rani, A., Singh, V.: Owl search algorithm: a novel nature-inspired
heuristic paradigm for global optimization. In: Thampi, S.M., El-Alfy, E.-S.M., Mitra, S.,
Trajkovic, L. (eds.). J. Intell. Fuzzy Syst. 34(3), 1573–1582 (2018)
83. Jain, M., Singh, V., Rani, A.: A novel nature-inspired algorithm for optimization: squirrel
search algorithm. Swarm Evol. Comput. 44, 148–175 (2019)
84. Fathollahi-Fard, A.M., Hajiaghaei-Keshteli, M., Tavakkoli-Moghaddam, R.: The social
engineering optimizer (SEO). Eng. Appl. Artif. Intell. 72, 267–293 (2018)
85. Dhiman, G., Kumar, V.: Emperor penguin optimizer: a bio-inspired algorithm for engineering
problems. Knowl.-Based Syst. 159, 20–50 (2018)
86. Elsisi, M.: Future search algorithm for optimization. Evol. Intell. 12(1), 21–31 (2019)
87. Harifi, S., Khalilian, M., Mohammadzadeh, J., Ebrahimnejad, S.: Emperor penguins colony:
a new metaheuristic algorithm for optimization. Evol. Intell. 12(2), 211–226 (2019)
88. Kaveh, A., Dadras, A.: A novel meta-heuristic optimization algorithm: thermal exchange
optimization. Adv. Eng. Softw. 110, 69–84 (2017)
89. Heidari, A.A., Mirjalili, S., Faris, H., Aljarah, I., Mafarja, M., Chen, H.: Harris hawks
optimization: algorithm and applications. Futur. Gener. Comput. Syst. 97, 849–872 (2019)
90. Askari, Q., Younas, I., Saeed, M.: Political optimizer: a novel socio-inspired meta-heuristic
for global optimization. Knowl-Based Syst. 195, 105709 (2020)
91. Askari, Q., Saeed, M., Younas, I.: Heap-based optimizer inspired by corporate rank hierarchy
for global optimization. Expert Syst. Appl. 161, 113702 (2020)
92. Sörensen, K., Sevaux, M., Glover, F.: A history of metaheuristics. In: Handbook of Heuristics,
pp. 1–18 (2018)
93. Abou-Warda, H., Belal, N.A., El-Sonbaty, Y., Darwish, S.: A random forest model for mental
disorders diagnostic systems. In: International Conference on Advanced Intelligent Systems
and Informatics. Springer, Cham, pp. 670–680 (2016)
94. Bone, D., Lee, C.C., Chaspari, T., Gibson, J., Narayanan, S.: Signal processing and machine
learning for mental health research and clinical applications [perspectives]. IEEE Signal
Process. Mag. 34(5), 196–195 (2017)
95. Srividya, M., Mohanavalli, S., Bhalaji, N.: Behavioral modeling for mental health using
machine learning algorithms. J. Med. Syst. 42(5), 1–12 (2018)
96. Chen, M., Hao, Y., Hwang, K., Wang, L., Wang, L.: Disease prediction by machine learning
over big data from healthcare communities. IEEE Access 5, 8869–8879 (2017)
97. Sahoo, A.K., Pradhan, C., Das, H.: Performance evaluation of different machine learning
methods and deep-learning based convolutional neural network for health decision making.
In: Nature Inspired Computing for Data Science. Springer, Cham, pp. 201–212 (2020)
98. Koh, J.E.W., De Michele, S., Sudarshan, V.K., Jahmunah, V., Ciaccio, E.J., Ooi, C.P., Grurajan,
R., Grurajan, R., Oh, S.L., Lewis, S.K., Green, P.H., Bhagat, G., Acharya, U.R.: Automated
interpretation of biopsy images for the detection of celiac disease using a machine learning
approach. Comput. Methods Programs Biomed. 203, 106010 (2021)
99. Sawant, A., Bhandari, M., Yadav, R., Yele, R., Bendale, M.S.: Brain cancer detection from
mri: a machine learning approach (tensorflow). Brain 5(04) (2018)
100. Fujisawa, Y., Inoue, S., Nakamura, Y.: The possibility of deep learning-based, computer-aided
skin tumor classifiers. Front. Med. 6, 191 (2019)
14 G. Bekdaş and S. M. Nigdeli

101. Chand, S.: A comparative study of breast cancer tumor classification by classical machine
learning methods and deep learning method. Mach. Vis. Appl. 31(6), 1–10 (2020)
102. Shabani, S., Yousefi, P., Naser, G.: Support vector machines in urban water demand forecasting
using phase space reconstruction. Procedia Eng. 186, 537–543 (2017)
103. Lopez Farias, R., Puig, V., Rodriguez Rangel, H., Flores, J.J.: Multi-model prediction for
demand forecast in water distribution networks. Energies 11(3), 660 (2018)
104. Yücel, M., Namli, E.: Yapay zekâ modelleri ile betonarme yapilara ait enerji performans
siniflarinin tahmini. Uludağ Univer. J. Fac. Eng. 22(3), 325–346 (2018)
105. Yang, S., Wan, M.P., Chen, W., Ng, B.F., Dubey, S.: Model predictive control with adaptive
machine-learning-based model for building energy efficiency and comfort optimization. Appl.
Energy 271, 115147 (2020)
106. Chen, X.L., Fu, J.P., Yao, J.L., Gan, J.F.: Prediction of shear strength for squat RC walls using
a hybrid ANN–PSO model. Eng. Comput. 34(2), 367–383 (2018)
107. Hoang, N.D., Tran, X.L., Nguyen, H.: Predicting ultimate bond strength of corroded rein-
forcement and surrounding concrete using a metaheuristic optimized least squares support
vector regression model. Neural Comput. Appl. 32(11), 7289–7309 (2020)
108. Yucel, M., Namlı, E.: High performance concrete (HPC) compressive strength predic-
tion with advanced machine learning methods: combinations of machine learning algo-
rithms with bagging, rotation forest, and additive regression. In: Artificial Intelligence and
Machine Learning Applications in Civil, Mechanical, and Industrial Engineering. IGI Global,
pp. 118–140 (2020)
109. Yücel, M., Bekdaş, G., Nigdeli, S.M.: Prediction of optimum 3-bar truss model parameters
with an ANN model. In: International Conference on Harmony Search Algorithm. Springer,
Singapore, pp. 317–324 (2020)
110. Yücel, M., Nigdeli, S.M., Kayabekir, A.E., Bekdaş, G.: Optimization and artificial neural
network models for reinforced concrete members. In: Nature-Inspired Metaheuristic Algo-
rithms for Engineering Optimization Applications. Springer, Singapore, pp. 181–199 (2021)
111. Nigdeli, S.M., Yücel, M., Bekdaş, G.: A hybrid artificial intelligence model for design of
reinforced concrete columns. Neural Comput. Appl. 35(10), 7867–7875 (2023)
112. Yucel, M., Bekdaş, G., Nigdeli, S.M., Sevgen, S.: Estimation of optimum tuned mass damper
parameters via machine learning. J. Build. Eng. 26, 100847 (2019)
113. Lara-Valencia, L.A., Farbiarz-Farbiarz, Y., Valencia-González, Y.: Design of a tuned mass
damper inerter (TMDI) based on an exhaustive search optimization for structural control of
buildings under seismic excitations. Shock Vib. (2020)
114. Etedali, S., Bijaem, Z.K., Mollayi, N., Babaiyan, V.: Artificial intelligence-based prediction
models for optimal design of tuned mass dampers in damped structures subjected to different
excitations. Int. J. Struct. Stab. Dyn. 2150120 (2021)
Hybrid Metaheuristics
The Development of Hybrid
Metaheuristics in Structural Engineering

Aylin Ece Kayabekir, Sinan Melih Nigdeli, and Gebrail Bekdaş

Abstract In engineering designs, the variables in the problems are needed to define
by checking several constraints. In that case, the problem is a nonlinear one that needs
several iterations when the best suitable solution is wanted. To find the best solution,
several algorithms may be employed to iteratively search for the optimum solution.
These algorithms are inspired by happening or processes to provide different formu-
lations. As the current trend, multiple algorithms are combined to update efficient
features instead of using a metaphor. In this chapter, a review is presented for hybrid
metaheuristics in structural engineering applications.

Keywords Hybrid algorithms · Structural engineering · Optimization ·

Metaheuristics · Constraints

1 Introduction

At the start of life, people tried to solve problems to invent something. Then, they
need to consider additional factors to eliminate negative issues in the first design.
As time progress, engineers must make this design perfect and optimum. To find
the best balance between constraints and factors that are the design objectives, it is
needed to try several iterations and these iterations can be automatically done by
several algorithms. Metaheuristics have great importance in this solution.
According to Sörensen et al. [1], the development of metaheuristics can be given
in five periods. The first period was named the pre-theoretical period. Heuristics are
essentially used by the human brain, but it is not studied. In this period, humans
copied several previously solved problems to solve new ones that are essential to

A. E. Kayabekir
Department of Civil Engineering, Istanbul Gelişim University, Avcılar, Istanbul, Turkey
S. M. Nigdeli (B) · G. Bekdaş
Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 17

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_2
18 A. E. Kayabekir et al.

living and surviving. For that reason, the heuristic belongs to the human and these
algorithms can be mentioned as artificial intelligence.
After World War II, a new period called “the early period” starts and the principles
such as analogy, induction, and auxiliary problem are mentioned. As an analogy,
the problem was tried to be solved by another problem. As induction, the learned
techniques are used, and finally, the problem was transformed into smaller problems
as an auxiliary problem.
After 1980, the method-centric period started. In this period, several metaheuris-
tics such as Tabu Search [2], Genetic Algorithm [3, 4], and Simulated Annealing [5]
were developed.
After 2000, the framework gains more interest instead of solving the problem.
Thus, the number of metaheuristic methods showed a great increase. After the 80s, the
metaphor used in the development of the algorithm showed more interest. Opposite to
that, solving the problem is more scientifically important. For that reason, the future
will be the scientific period. In this period, more complex and challenging problems
will be needed to be solved. For that reason, great effort must be given to find the
exact method that is the best for the problem. This situation can be provided by
hybridizing the effective known features of existing algorithms instead of proposing
new ones that are similar to previous algorithms.
This chapter includes a brief definition of metaheuristic algorithms. Then, the
objectives and constraints in structural engineering are summarized. Then, the
applications using hybrid algorithms are reviewed in structural engineering.

2 Metaheuristic Algorithms and Optimization

In engineering design and optimization, the key method that is used are metaheuris-
tics. It provides a formalized generation of design variables to find the coded objective
functions that are penalized for constraint violation. In short, a design variable (x i )
is searched within defined minimum (x min ) and maximum (x max ) values by using
randomization. As the most basic formulation, Eq. (1) can be given as a linear search
with a random parameter shown as ε.

xi = xmin + ε(xmax − xmin ) (1)

Equation (1) is generally used in the definition of initial values and special features
of algorithms were used to update this solution. A general flowchart is shown in Fig. 1.
The Development of Hybrid Metaheuristics in Structural Engineering 19

Fig. 1 A general flowchart

for metaheuristic algorithms START

Define problem
constants, algorithm
parameters and ranges
of design variables

Randomly generate the initial solutions

Modify the existing solutions by using

the features of the algorithms

Are the YES

NO
stopping criteria STOP
provided?

3 The Structural Optimization Problems

The engineering optimization aims to find a set of design variables (x i ) that minimize
or maximize objective function ( f (x)) according to several constraints that may
be equality (h(x) = 0) or inequality (g(x) ≤ 0) functions. In this section, possible
objectives and constraints in structural engineering are mentioned.

3.1 Objective Function

Weight (W ) is one of the objective functions that is needed to minimize to maxi-

mize material amount or decrease the self-weight. Also, it may be efficient for
transportation or use in space.
The second objective is cost. It can be also directly related to weight, but composite
materials with different costs (especially in reinforced concrete structures) show
different situations since the costs of different materials are different, too.
Thirdly, the response such as deflection or stress of structural members can be
considered as objectives. The response of structures can be also considered as a
second objective or as a constraint.
20 A. E. Kayabekir et al.

The efficiency of the use of additional system in design can be used as an objec-
tive function. Especially, for structural control systems, the efficiency of response
reduction can be maximized to make a perfectly tuned control system.
Finally, environmental factors such as CO2 emission minimization can be used as
an objective to find a sustainable design.

3.2 Design Constraints

As the design constraint of structure, several rules are given in the design codes.
One of the factors in engineering design is not to exceed stress capacity. The stress
capacity must be checked for different sections and types including normal (axial)
stress and shear stress.
As seen in Fig. 2, the buckling load (N) must be checked for slender structural
systems.
For steel members, the fatigue of the member under cycled load must be checked
as a constraint.
Also, deflections must be limited and member sections may be wanted to be in a
specific range according to design codes.

Fig. 2 Buckling N
The Development of Hybrid Metaheuristics in Structural Engineering 21

4 Review of Structural Engineering Applications Using

Hybrid Algorithms

In 2008, a hybrid Harmony search algorithm using sequential quadratic programming

was presented by Fesanghary et al. [6] for the optimization of engineering problems.
The hybrid approach was implemented for various benchmark problems. According
to this, the proposed approach was capable to find quality results and was successful
in terms of required fitness function numbers.
Kaveh and Talatahari [7] proposed a hybrid algorithm combined harmony search
(HS) scheme, particle swarm optimizer, and ant colony optimization (ACO). The
analyzes done for truss and frame systems revealed that the proposed method is
effective to obtain optimal solutions to continuous or discrete optimization problems.
For the optimum design of fuzzy-PID controllers, the combination of PSO and
Evolutionary Programming (EP), was investigated by Chiou and Liu [8]. In this
study, it was shown that fuzzy-PID controllers using the PSO-EP method have
been successful in evaluating indices such as the rise time, settling time, peak time,
maximum overshoot, and Integral of Absolute Error.
Sandesh and Shankar [9] introduced a hybrid method based on PSO and Genetic
Algorithm (GA). The method runs for the identification of damages in a thin plate
due to multiple cracks. In this application, although a single PSO was to be fast and
hybrid method provided more accurate damage prediction.
In another study, PSO was hybridized with a simultaneous perturbation stochastic
approximation (SPSA) algorithm by Seyedpoor et al. [10] for structural optimization.
The benchmark examples on truss systems that were solved by various methods,
indicated that SPSA-PSO combination was successful in terms of quality of optimum
results and structural analyses.
In 2011, Plevris and Papadrakakis [11] investigated the improved PSO method
using the gradient-based quasi-Newton sequential quadratic programming (SQP)
method. The performance of the method was analyzed on the various truss system
and, it was concluded that the proposed method obtained better optimum solutions
according to other methods such as PSO, GA, SQP.
The Genetic algorithm was hybridized with Nelder Mead simplex was presented
by Rahami et al. [12] and then used to minimize the weight of truss structures. Anal-
ysis results showed that robust optimization results were obtained with the hybrid
algorithm.
Another truss optimization was done by Hadidi et al. [13]. In the optimization
process, PSO and Simulated Annealing (SA) combination (PSO-SA) was utilized to
prevent the stuck local minimum problem. The efficiency of the PSO-SA was tested
on several benchmark examples. This investigation demonstrated that PSO-SA was
efficient and provide computational advantages regarding standard PSO and several
heuristic algorithms.
Also, Lee et al. [14] used a PSO-SA hybrid algorithm for the Sliding mode control
of isolated bridges. In this investigation, the hybrid algorithm outperformed PSO as
well.
22 A. E. Kayabekir et al.

Kaveh and Talatahari [15] combined PSO with the charged system search (CSS)
for the optimum design of structures. Three structures including containing a truss,
a frame, and a grillage system were optimized with a developed hybrid algorithm
(PSO-CSS). Generally, PSO-CSS found better solutions considering computational
costs according to CSS, PSO, and other advanced heuristic methods like GA, BB-BC,
evolution strategies (ESs), ACO, and harmony search (HS).
In 2013, multi-objective design problems including benchmark functions and a
welled beam were solved by Rachid and Rajae [16]. The optimization process was
carried out with the combination of the Normal Boundary Intersection approach
(NBI) and the Simulated Annealing Simultaneous Perturbation (SASP) method. It
resulted that, the NBI-SASP combination provided a well-distributed Pareto front
and was successful to solve real-world application problems.
Pholdee and Bureerat [17] carried out a multi-objective optimization process
via incorporation (RPBIL-DE) of differential evolution (DE) and population-based
incremental learning (RPBIL) for truss systems. RPBIL-DE was tested on several
examples solved by some other multi-objective evolutionary algorithms (MOEAs).
Results of the examinations showed that DE improved performance of RPBIL and
RPBIL-DE have better performance compared to other techniques.
For the size and geometry optimization of truss structures, Shojaee et al. [18]
proposed a method based on discrete PSO and the method of moving asymp-
totes (MMA). The analysis shows that the method has effectively accelerated the
convergence rate capability.
In addition to these studies, several hybrid methods have been developed to solve
truss problems such as HS-PSO [19], cellular automata-PSO [20], GA-phenotypical
probabilistic local search algorithm [21].
Amini and Ghaderi [22] suggested using HS with ACO to determine the
optimum location of dampers attached to structures. Numerical examples showed
that hybridized search algorithm improved the convergence rate.
Khajehzadeh et al. [23] integrated the firefly algorithm (FA) with sequen-
tial quadratic programming (SQP). The hybrid method was tested on the design
optimization of reinforced concrete (RC) foundations and it outperforms FA.
Glowworm swarm optimization (GSO) was combined with Simulated Annealing
(SA) by García-Segura et al. [24]. In the numerical example, the cost and CO2
emissions of a concrete I-beam were minimized, and the hybrid algorithm was more
efficient and provided more convergence rate than GSO.
Long et al. [25] proposed that the Pattern search method incorporate into the
cuckoo search (CS) algorithm to enhance approximation capability. The developed
method implemented several benchmark problems. According to the analysis results,
the proposed algorithm was effective, efficient, and robust in solving structural design
optimization problems.
A hybrid algorithm including the advantages of PSO and CSS was presented by
Talatahari et al. [26] for the seismic design optimization of steel frames. In the seismic
design process, four performance levels were considered to minimize structural costs.
Two building frameworks were optimized with the hybrid algorithm as well as GA,
The Development of Hybrid Metaheuristics in Structural Engineering 23

ACO, PSO, and PSACO. The optimum results showed that the presented hybrid
algorithm found a better seismic design.
Liu et al. [27] developed a hybrid procedure using finite element analysis, Neural
networks, GA, and the Monte-Carlo method to evaluate the reliability of cable-stayed
bridges. The method was tested on two numerical examples. According to this, the
hybrid procedure provided accuracy satisfactorily and was efficient in calculations.
A hybrid evolutionary approach combining gravitational search algorithm (GSA)
and pattern search (PS) was presented by Khajehzadeh et al. [28] for multi-objective
optimization of retaining walls. The hybrid approach is employed for two objective
functions i.e. total cost and amount of embedded CO2 emissions. The compara-
tive analysis indicates that the presented approach provided better computational
efficiency and robustness than standard GSA and some other methods.
For the optimum reinforced concrete (RC) design against seismic excitations,
the PSO algorithm using an intelligent regression model (IRM) was presented by
Gharehbaghi and Khatibinia [29]. The hybrid method was applied to 9- and 18-story
RC frames under earthquake loads for the minimum cost. It was concluded that
IRM was effective to predict average time-history responses (ATHR) of structure
and reduce the time of the optimization process.
Hadidi and Rafiee [30] introduced a hybrid algorithm combining Harmony Search
(HS) and Big Bang-Big Crunch (BB-BC) for the design optimization of semi-rigid
connection steel frames. The optimization process was to search optimum sections
and connections of beams and columns to minimize cost. Three benchmark opti-
mization problem was solved proposed approach. According to the results, it was
observed that the introduced method outperform classical BB-BC and HS methods.
An investigation was carried out by Akın and Aydoğdu [31] to calculate the
minimum weight of design steel space frames. In the optimization process, TLBO
and HS collaboration (hTLBO-HS) was utilized. Analysis results demonstrated that
hTLBO-HS was found better solution than standard HS.
In 2016, Babaei and Sanaei [32] developed a hybrid method using ACO and
GA for multi-objective optimization of braced frames. In the optimization process,
optimum cross sections of structural members and optimum topologies of the braces
were searched to provide minimum weight and minimum displacement. Analysis
results demonstrated that the developed method was capable to obtain optimum
topologies and sections.
Khajehzadeh [33] investigated a hybrid optimization method based on pattern
search with the gravitational search algorithm (GSA) for RC retaining walls. The
example done for minimum embedded CO2 emissions revealed that the hybrid
method performs better than GSA in terms of efficiency and robustness.
In another study conducted in 2016, Sheikholeslami et al. [34] solved the minimum
cost problem of the retaining wall. It has developed a higher-performance solution
method by combining the firefly algorithm with HS. For the reduction of computa-
tional effort, the upper bound strategy (UBS) was also included in the optimization
process. The proposed method was effective to find the minimum cost with a lower
number of analyses.
24 A. E. Kayabekir et al.

Kaveh and Shokohi [35] optimized castellated beams by hybridizing Colliding

Bodies Optimization (CBO) and PSO to obtain minimum cost. The hybrid method
developed in this study was applied to 3 different benchmark problems and its
performance was compared with existing methods. Hybridization of CBO and PSO
generally provided less cost.
Maheri et al. [36] proposed the PSO-GA hybrid technique to find the optimum
size and topology that provide the minimum weight of trusses. As a result of analyses
carried out on several benchmark problems, the PSO-GA hybrid method achieved
better solutions with better convergence compared to GA, PSO, and other hybrid
methods.
Hybrid HS using Global-best PSO search and neighborhood search was intro-
duced by Cheng et al. [37]. The method was tested on the optimum design of truss
structures. Analyzes results show that the proposed method generally was superior
to other optimization methods used for comparison in terms of optimum results and
convergence rate.
For a more effective, robust, and cheaper design of spatial structures, a hybrid
methodology consisting of genetic programming and local search methods was
proposed by Assimi et al. [38].
Hoseini Vaez and Fallah [39] investigated damage detection with its location and
its severity in thin plates. For that reason, a hybrid approach based on GA and PSO
was developed. In the optimization process, the objective function was formulated
considering natural frequencies and the mode shapes of the structure and minimized.
According to evaluations done on the three different plates for four damage scenarios,
it was observed that the proposed approach provides less error index in comparison
to GA and PSO.
A hybrid algorithm including modified PSO and modified GSA was developed
by Chutani and Singh [40] for the cost minimization of RC frames. Analysis results
revealed that the modified hybrid algorithm improved the performance of PSO and
GSA by overcoming the limitations of the algorithms.
Jiale [41] optimized structural sensor configuration for civil engineering problems.
The optimization process was carried out by simulated annealing-genetic algorithm
combination and the optimum number and position of the sensor were obtained.
In 2018, Nelder–Mead local search operator was incorporated into the genetic
programming algorithm by Assimi and Jamali [42] to find the optimum topology and
size of truss structures considering static and dynamic constraints. The proposed algo-
rithm was employed for some optimization problems with continuous and discrete
design variables. It was observed that the hybrid algorithm performed better generally
than other methods used in the optimization of the numerical examples.
Design optimization of large-scale dome truss structures was done by Kaveh and
Ilchi Ghazaan [43]. For that reason, the vibrating particle system algorithm, multi-
design variable configuration cascade optimization, and an upper-bound strategy
were combined. Comparative analyzes showed that the proposed hybrid combination
had powerful search capability and was suitable for structural engineering problems.
In 2019, a methodology using the plant growth simulation algorithm and particle
swarm optimization was developed by Jiang et al. [44] for structural optimization.
The Development of Hybrid Metaheuristics in Structural Engineering 25

The method tested on a pre-stressed spatial steel dome, was found to be effective and
applicable for structural optimization.
Huang et al. [45] suggested combining PSO and cuckoo search (CS) to determine
the damage to structures under noise and temperature environments. The numerical
analysis was done for the simply supported beam. The suggested method has superior
performance than PSO and CS algorithms.
Total potential optimization employing hybrid metaheuristics using fundamentals
of JA and phases of FPA and TLBO algorithms was investigated by Bekdaş et al.
[46]. In this study, four different hybrid methods were developed; i.e. JA using Lévy
flights, JA using Lévy flights and linear distribution, JA with consequent student
phase, and JA with probabilistic student phase (JA1SP). These hybrid methods were
also compared with DE, PSO, and HS considering large-scale truss analysis problems
(Fig. 3). Comparative analysis revealed that JA1SP provides better convergence to a
global optimum.
Jafari et al. [47] presented a hybrid optimization method that enhances the perfor-
mance of elephant herding optimization (EHO) via the cultural algorithm (CA). The
hybrid algorithm (EHOC) was employed for mathematical optimization problems
and to minimize the truss weight. Analysis results show that EHOC has efficient
convergence capability and provides better solutions in comparison to EHO and CA.
In the other study conducted by Jafari et al. [48], the PSO algorithm was improved
by using the cultural algorithm. Developed methods provide effective solutions and
convergence rates for the optimum design of truss structures.
For the cost minimization of the waffle slab, mouth-brooding fish (MBF) and
colliding bodies optimization (CBO) algorithms were hybridized by Shayegan et al.
[49]. Comparing the performance of the proposed method with MBF, CBO, harmony
search (HS), PSO, democratic particle swarm optimization (DPSO), charged system
search (CSS), and enhanced charged system search (ECSS), it was concluded that
the proposed method is effective to obtain an optimum solution.
Han et al. [50] presented a hybrid algorithm named (MSFWA) combining moth
search (MS) and fireworks algorithms (FWA). MSFWA was effective and provide
fast approximation in solving several benchmark functions and engineering design
problems.
García et al. [51] optimized cost and CO2 emissions for counterfort retaining
walls. In the optimization process, a method based on Cuckoo Search Algorithm
and k-Means Operator which is a machine learning algorithm was employed. This
application showed that k-Means Operator provided significant quality solutions and
the proposed method found the superior result according to HS.
A technique, a hybrid form of ant lion optimizer, and an improved Nelder–Mead
algorithm were suggested to detect structural damage by Chen and Yu [52]. For the
verification of the hybrid technique, benchmark functions, two numerical simulations
including a two-story rigid frame, and 31-bar planar truss structures were utilized and
experimental validation for the simply supported beam was carried out. According
to these, it was claimed that the proposed technique was successful to determine the
location and severity of the damage.
26 A. E. Kayabekir et al.

Fig. 3 Truss structures analyzed via hybrid algorithms [46]

The Development of Hybrid Metaheuristics in Structural Engineering 27

In addition to the optimization problems presented above, many research articles

have been done on the truss optimization problem. For this purpose, several hybrid
algorithms were proposed such as the invasive weed optimization-shuffled frog-
leaping algorithm [53], TLBO-HS [54, 55], PSO-GA [56] and firefly algorithm-
optimality criteria [57].
Kayabekir et al. [58] developed a hybrid procedure (HHS) combining the local
search phases of HS and the global search phase of the FPA to minimize total potential
energy for structural systems. In this study, HSS applied to solve plane stress members
(Fig. 4) and compared several metaheuristics such as s HS, FPA, TLBO, and JA. As
a result, HSS reached the best solutions quickly.
In the other study, four different hybrid algorithms using HS, TLBO, FPA, and JA
were proposed by Toklu et al. [59]. The hybrid algorithms were employed for the total
potential optimization of tunnel structures (Fig. 5). According to the analysis results,
it was observed that Total Potential Optimization via hybrid algorithms increased the
capacity of solving the problem.

Fig. 4 Plane-stress members [58]

28 A. E. Kayabekir et al.

Fig. 5 The tunnel system [59]

An adaptive version of HS was hybridized with JA to get rid of stuck-in local

optimum by Yücel et al. [60] for the optimum cost of RC retaining walls. A compar-
ative analysis conducted by several metaheuristic algorithms demonstrated that
adaptive-hybrid HS can achieve optimum solutions using fewer population numbers.
Kaveh et al. [61] combined Harris Hawks Optimizer (HHO) with the Imperialist
Competitive Algorithm (ICA) to enhance the exploration performance of HHO. The
performance of the developed hybrid algorithm was tested by several benchmark
optimization problems employing nature-inspired algorithms. These comparisons
showed that the hybrid algorithm is efficient in structural optimization.
Similarly, the HHO algorithm was improved by teaching the learning-based opti-
mization (TLBO) algorithm to get rid of the trap in the local optimum. Examinations
on benchmark functions and engineering optimization problems demonstrated to
increase in the global performance capability of the proposed method [62].
Three ANN (artificial neural network) models were developed by different opti-
mization algorithms; i.e. PSO, Archimedes optimization algorithm (AOA), and
sparrow search algorithm (SSA), to estimate the shear capacity of reinforced concrete
deep beams by Barkhordari et al. [63]. It has been observed that among hybrid models,
the PSO model can achieve more precise results by using a lower neuron and hidden
layer.
In the other study, Grey Wolf Optimization (GWO) improved via PSO was imple-
mented for static and dynamic crack identification of two-dimensional plates [64].
The comparative analysis demonstrated that the improved GWO (IGWO) outper-
forms than original GWO in terms of the accuracy of results. IGWO and GWO were
also used for the prediction of ANN parameters. In this application, IGWO provided
a better estimation of crack length as well.
The Development of Hybrid Metaheuristics in Structural Engineering 29

Firouzi et al. [65] conducted a study using optimization algorithms such as Harris
hawk optimization (HHO), electrostatic discharge algorithm (ESDA), pathfinder
algorithm (PFA), and Henry gas solubility optimization (HGSO), and hybrid
versions of the algorithms to detect cracks in cantilever beams. In the hybrid
version of algorithms, Nelder–Mead (NM) algorithm is included in the optimiza-
tion process. According to comparative analyzes, Hybrid algorithms outperformed
in the prediction of crack capability.
Kayabekir et al. [66] proposed metaheuristic-based hybrid design procedures to
optimize active tuned mass dampers (ATMDs) added to structures under earthquake
excitations (Fig. 6). For the generation of hybrid algorithms, four different meta-
heuristic algorithms; i.e. Harmony search (HS), Flower pollination (FPA), Teaching
Learning Based Optimization (TLBO), and Jaya (JA) are combined.
The slime mold algorithm (SMA) is hybridized with the sine cosine algorithm
to avoid getting stuck in the local search space by Örnek et al. [67]. The proposed
method was employed for various benchmark functions and optimization problems
such as the design of a cantilever beam, pressure vessel, 3-bar truss, and speed
reducer problems. As a result of these applications, the hybrid algorithm has superior
convergence than standard sine cosine and SMA algorithms.
Similarly, Chauhan et al. [68] also hybridized the SMA and AOA algorithms
against less internal memory and slow convergence. The hybrid method has better
performance than classical SMA and AOA on various real-world problems such as
the design of spring, pressure vessel, welded beam, and three-bar truss.
In 2023, Cao [69] presented a hybrid constraint-handling method considering the
Deb rule and the mapping strategy for the optimum size and shape of truss structures.
According to the results of the comparative analysis, it was seen that the Hybrid CHT
method is one of the effective methods that can be used in solving such problems.

5 Conclusion

As seen from the literature review about structural optimization using a hybrid meta-
heuristic algorithm, the subject is the latest trend in optimization. Hybrid algorithms
are effective in finding the optimum solution by eliminating back draws of existing
algorithms and adding new features of other algorithms.
In the future, hybridization will be more effective since the huge increase in the
number of metaheuristic algorithms. According to the no-free-lunch theorem, there
is no superior algorithm that can perfectly solve all problems. Also, the performance
of the algorithm may show differences according to design constant cases. For that
reason, humans will always need new ways to find the best solution for their current
interests.
30 A. E. Kayabekir et al.

Fig. 6 Structure with

ATMD [66]

References

1. Sörensen, K., Sevaux, M., Glover, F.: A history of metaheuristics. In: Handbook of Heuristics,
pp. 1–18 (2018)
2. Glover, F.: Future paths for integer programming and links to artificial intelligence. Comput.
Oper. Res. 13(5), 533–549 (1986)
The Development of Hybrid Metaheuristics in Structural Engineering 31

3. Holland, J.H.: Adaptation in Natural and Artificial Systems. University of Michigan Press, Ann
Arbor, Michigan (1975)
4. Goldberg, D.E., Samtani, M.P.: Engineering optimization via genetic algorithm. In: Proceedings
of Ninth Conference on Electronic Computation. ASCE, New York, NY, pp. 471–482 (1986)
5. Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P.: Optimization by simulated annealing. Science
220(4598), 671–680 (1983)
6. Fesanghary, M., Mahdavi, M., Minary-Jolandan, M., Alizadeh, Y.: Hybridizing harmony search
algorithm with sequential quadratic programming for engineering optimization problems.
Comput. Methods Appl. Mech. Eng. 197(33–40), 3080–3091 (2008)
7. Kaveh, A., Talatahari, S.: Hybrid algorithm of harmony search, particle swarm and ant colony
for structural design optimization. In: Harmony Search Algorithms for Structural Design
Optimization, 159–198 (2009)
8. Chiou, J.S., Liu, M.T.: Numerical simulation for fuzzy-PID controllers and helping EP
reproduction with PSO hybrid algorithm. Simul. Model. Pract. Theory 17(10), 1555–1565
(2009)
9. Sandesh, S., Shankar, K.: Application of a hybrid of particle swarm and genetic algorithm for
structural damage detection. Inverse Probl. Sci. Eng. 18(7), 997–1021 (2010)
10. Seyedpoor, S., Gholizadeh, S., Talebian, S.: An efficient structural optimisation algorithm using
a hybrid version of particle swarm optimisation with simultaneous perturbation stochastic
approximation. Civ. Eng. Environ. Syst. 27(4), 295–313 (2010)
11. Plevris, V., Papadrakakis, M.: A hybrid particle swarm—gradient algorithm for global structural
optimization. Comput.-Aided Civ. Infrastruct. Eng. 26(1), 48–68 (2011)
12. Rahami, H., Kaveh, A., Aslani, M., Asl, R.N.: A hybrid modified genetic-Nelder Mead simplex
algorithm for large-scale truss optimization. Int. J. Optim. Civ. Eng. 1(1), 29–46 (2011)
13. Hadidi, A., Kaveh, A., Farahmand Azar, B., Talatahari, S., Farahmandpour, C.: An efficient
hybrid algorithm based on particle swarm and simulated annealing for optimal design of space
trusses. Int. J. Optim. Civil. Eng. 1(3), 377–395 (2011)
14. Lee, T.Y., Chen, P.C., Juang, D.S.: Sliding mode control on isolated bridges with columns of
irregular heights using pole assignment and PSO-SA hybrid algorithm. Int. J. Struct. Stab. Dyn.
12(03), 1250014 (2012)
15. Kaveh, A., Talatahari, S.: A hybrid CSS and PSO algorithm for optimal design of structures.
Struct. Eng. Mech. 42(6), 783–797 (2012)
16. Rachid, E., Rajae, A.: New hybrid algorithm for multi-objective structural optimization.
In: Proceedings of 2013 International Conference on Industrial Engineering and Systems
Management (IESM). IEEE, pp. 1–5 (2013)
17. Pholdee, N., Bureerat, S.: Hybridisation of real-code population-based incremental learning
and differential evolution for multiobjective design of trusses. Inf. Sci. 223, 136–152 (2013)
18. Shojaee, S., Arjomand, M., Khatibinia, M.: A hybrid algorithm for sizing and layout optimiza-
tion of truss structures combining discrete PSO and convex approximation. Int. J. Optim. Civ.
Eng. 3(1), 57–83 (2013)
19. Kaveh, A., Nasr, E.A.: Engineering design optimization using a hybrid PSO and HS algorithm
(2013)
20. Gholizadeh, S.: Layout optimization of truss structures by hybridizing cellular automata and
particle swarm optimization. Comput. Struct. 125, 86–99 (2013)
21. Asl, R.N., Aslani, M., Panahi, M.S.: Sizing optimization of truss structures using a hybridized
genetic algorithm (2013). arXiv preprint arXiv:1306.1454
22. Amini, F., Ghaderi, P.: Hybridization of harmony search and ant colony optimization for optimal
locating of structural dampers. Appl. Soft Comput. 13(5), 2272–2280 (2013)
23. Khajehzadeh, M., Taha, M.R., Eslami, M.: A new hybrid firefly algorithm for foundation
optimization. Natl. Acad. Sci. Lett. 36, 279–288 (2013)
24. García-Segura, T., Yepes, V., Martí, J.V., Alcalá, J.: Optimization of concrete I-beams using a
new hybrid glowworm swarm algorithm. Lat. Am. J. Solids Struct. 11, 1190–1205 (2014)
25. Long, W., Zhang, W.Z., Huang, Y.F., Chen, Y.X.: A hybrid cuckoo search algorithm with
feasibility-based rule for constrained structural optimization. J. Cent. South Univ. 21(8), 3197–
3204 (2014)
32 A. E. Kayabekir et al.

26. Talatahari, S., Hosseini, A., Mirghaderi, S.R., Rezazadeh, F.: Optimum performance-based
seismic design using a hybrid optimization algorithm. Math. Probl. Eng. 2014, 1–8 (2014)
27. Liu, Y., Lu, N.W., Wang, Q.Y.: Reliability assessment of longspan cable-stayed bridges based
on hybrid algorithm. J. Highw. Transp. Res. Dev. 31(7), 72–79 (2014)
28. Khajehzadeh, M., Taha, M.R., Eslami, M.: Multi-objective optimisation of retaining walls
using hybrid adaptive gravitational search algorithm. Civ. Eng. Environ. Syst. 31(3), 229–242
(2014)
29. Gharehbaghi, S., Khatibinia, M.: Optimal seismic design of reinforced concrete structures
under time-history earthquake loads using an intelligent hybrid algorithm. Earthq. Eng. Eng.
Vib. 14, 97–109 (2015)
30. Hadidi, A., Rafiee, A.: A new hybrid algorithm for simultaneous size and semi-rigid connection
type optimization of steel frames. Int. J. Steel Struct. 15, 89–102 (2015)
31. Akın, A., Aydoğdu, İ: Optimum design of steel space frames by hybrid teaching-learning based
optimization and harmony search algorithms. Int. J. Struct. Constr. Eng. 9(7), 1367–1374 (2015)
32. Babaei, M., Sanaei, E.: Multi-objective optimal design of braced frames using hybrid genetic
and ant colony optimization. Front. Struct. Civ. Eng. 10, 472–480 (2016)
33. Khajehzadeh, M.: A new hybrid algorithm for CO2 emissions optimization of retaining walls.
Int. J. Adv. Mech. Civ. Eng 3(1), 7–11 (2016)
34. Sheikholeslami, R., Khalili, B.G., Sadollah, A., Kim, J.: Optimization of reinforced concrete
retaining walls via hybrid firefly algorithm with upper bound strategy. KSCE J. Civ. Eng. 20,
2428–2438 (2016)
35. Kaveh, A., Shokohi, F.: A hybrid optimization algorithm for the optimal design of laterally-
supported castellated beams. Sci. Iran. 23(2), 508–519 (2016)
36. Maheri, M.R., Askarian, M., Shojaee, S.: Size and topology optimization of trusses using
hybrid genetic-particle swarm algorithms. Iran. J. Sci. Technol., Trans. Civ. Eng. 40, 179–193
(2016)
37. Cheng, M.Y., Prayogo, D., Wu, Y.W., Lukito, M.M.: A hybrid harmony search algorithm for
discrete sizing optimization of truss structure. Autom. Constr. 69, 21–33 (2016)
38. Assimi, H., Jamali, A., Nariman Zadeh, N.: Sizing and topology optimization of spatial truss
structures using hybrid algorithm of genetic programing and Nelder–Mead. Modares Mech.
Eng. 17(6), 32–40 (2017)
39. Hoseini Vaez, S.R., Fallah, N.: Damage detection of thin plates using GA-PSO algorithm based
on modal data. Arab. J. Sci. Eng. 42, 1251–1263 (2017)
40. Chutani, S., Singh, J.: Optimal design of RC frames using a modified hybrid PSOGSA
algorithm. Arch. Civ. Eng. 63(4), 123–134 (2017)
41. Jiale, T.: Optimal allocation of structural sensor in civil engineering based on simulated
annealing genetic algorithm
42. Assimi, H., Jamali, A.: A hybrid algorithm coupling genetic programming and Nelder–Mead
for topology and size optimization of trusses with static and dynamic constraints. Expert Syst.
Appl. 95, 127–141 (2018)
43. Kaveh, A., Ilchi Ghazaan, M.: A new hybrid meta-heuristic algorithm for optimal design of
large-scale dome structures. Eng. Optim. 50(2), 235–252 (2018)
44. Jiang, Z., Lin, Q., Shi, K., Pan, W.: A novel PGSA-PSO hybrid algorithm for structural
optimization. Eng. Comput. 37(1), 144–160 (2019)
45. Huang, M., Cheng, S., Zhang, H., Gul, M., Lu, H.: Structural damage identification under
temperature variations based on PSO-CS hybrid algorithm. Int. J. Struct. Stab. Dyn. 19(11),
1950139 (2019)
46. Bekdaş, G., Kayabekir, A.E., Nigdeli, S.M., Toklu, Y.C.: Advanced energy based analyses of
trusses employing hybrid metaheuristics. Struct. Des. Tall Spec. Build. 28(9), 1–19, e1609
(2019)
47. Jafari, M., Salajegheh, E., Salajegheh, J.: An efficient hybrid of elephant herding optimization
and cultural algorithm for optimal design of trusses. Eng. Comput. 35, 781–801 (2019)
48. Jafari, M., Salajegheh, E., Salajegheh, J.: Optimal design of truss structures using a hybrid
method based on particle swarm optimizer and cultural algorithm. In: Structures, vol. 32.
Elsevier, pp. 391–405 (2021)
The Development of Hybrid Metaheuristics in Structural Engineering 33

49. Shayegan, D.S., Lork, A., Hashemi, S.A.H.: A new hybrid algorithm for cost optimization of
waffle slab. Slovak J. Civ. Eng. 28(3), 40–46 (2020)
50. Han, X., Yue, L., Dong, Y., Xu, Q., Xie, G., Xu, X.: Efficient hybrid algorithm based on moth
search and fireworks algorithm for solving numerical and constrained engineering optimization
problems. J. Supercomput. 76, 9404–9429 (2020)
51. García, J., Yepes, V., Martí, J.V.: A hybrid k-means cuckoo search algorithm applied to the
counterfort retaining walls problem. Mathematics 8(4), 555 (2020)
52. Chen, C., Yu, L.: A hybrid ant lion optimizer with improved Nelder–Mead algorithm for
structural damage detection by improving weighted trace lasso regularization. Adv. Struct.
Eng. 23(3), 468–484 (2020)
53. Kaveh, A., Talatahari, S., Khodadadi, N.: Hybrid invasive weed optimization-shuffled frog-
leaping algorithm for optimal design of truss structures. Iran. J. Sci. Technol., Trans. Civ. Eng.
44, 405–420 (2020)
54. Talatahari, S., Goodarzimehr, V.: A discrete hybrid teaching-learning-based optimization
algorithm for optimization of space trusses. J. Struct. Eng. Geo-Tech. 10(1), 55–72 (2020)
55. Talatahari, S., Goodarzimehr, V., Taghizadieh, N.: Hybrid teaching-learning-based optimiza-
tion and harmony search for optimum design of space trusses. J. Optim. Ind. Eng. 13(1),
177–194 (2020)
56. Omidinasab, F., Goodarzimehr, V.: A hybrid particle swarm optimization and genetic algorithm
for truss structures with discrete variables. J. Appl. Comput. Mech. 6(3), 593–604 (2020)
57. Gebremedhen, H.S., Woldemichael, D.E., Hashim, F.M.: A firefly algorithm based hybrid
method for structural topology optimization. Adv. Model. Simul. Eng. Sci. 7, 1–20 (2020)
58. Kayabekir, A.E., Toklu, Y.C., Bekdaş, G., Nigdeli, S.M., Yücel, M., Geem, Z.W.: A novel hybrid
harmony search approach for analysis of plane stress systems via total potential optimization.
Appl. Sci. 10(7), 2301 (2020)
59. Toklu, Y.C., Bekdaş, G., Kayabekir, A.E., Nigdeli, S.M., Yücel, M.: Total potential optimization
using hybrid metaheuristics: a tunnel problem solved via plane stress members. In: Nigdeli,
et al. (eds.) Advances in Structural Engineering—Optimization Emerging Trends in Structural
Optimization. Springer, pp. 181–198 (2020)
60. Yücel, M., Kayabekir, A.E., Bekdaş, G., Nigdeli, S.M., Kim, S., Geem, Z.W.: Adaptive-hybrid
harmony search algorithm for multi-constrained optimum eco-design of reinforced concrete
retaining walls. Sustainability 13(4), 1639 (2021)
61. Kaveh, A., Rahmani, P., Eslamlou, A.D.: An efficient hybrid approach based on Harris Hawks
optimization and imperialist competitive algorithm for structural optimization. Eng. Comput.
1–29 (2021)
62. Kundu, T., Garg, H.: A hybrid ITLHHO algorithm for numerical and engineering optimization
problems. Int. J. Intell. Syst. 37(7), 3900–3980 (2022)
63. Barkhordari, M.S., Feng, D.C., Tehranizadeh, M.: Efficiency of hybrid algorithms for esti-
mating the shear strength of deep reinforced concrete beams. Period. Polytech. Civ. Eng. 66(2),
398–410 (2022)
64. Al Thobiani, F., Khatir, S., Benaissa, B., Ghandourah, E., Mirjalili, S., Wahab, M.A.: A hybrid
PSO and grey wolf optimization algorithm for static and dynamic crack identification. Theoret.
Appl. Fract. Mech. 118, 103213 (2022)
65. Firouzi, B., Abbasi, A., Sendur, P.: Improvement of the computational efficiency of meta-
heuristic algorithms for the crack detection of cantilever beams using hybrid methods. Eng.
Optim. 54(7), 1236–1257 (2022)
66. Kayabekir, A.E., Nigdeli, S.M., Bekdaş, G.: A hybrid metaheuristic method for optimization
of active tuned mass dampers. Comput.-Aided Civ. Infrastruct. Eng. 37(8), 1027–1043 (2022)
67. Örnek, B.N., Aydemir, S.B., Düzenli, T., Özak, B.: A novel version of slime mould algorithm
for global optimization and real world engineering problems: Enhanced slime mould algorithm.
Math. Comput. Simul. 198, 253–288 (2022)
34 A. E. Kayabekir et al.

68. Chauhan, S., Vashishtha, G., Kumar, A.: A symbiosis of arithmetic optimizer with slime mould
algorithm for improving global optimization and conventional design problem. J. Supercomput.
78(5), 6234–6274 (2022)
69. Cao, H., Sun, W., Chen, Y., Kong, F., Feng, L.: Sizing and shape optimization of truss employing
a hybrid constraint-handling technique and manta ray foraging optimization. Expert Syst. Appl.
213, 118999 (2023)
Optimum Design of Reinforced Concrete
Columns in Case of Fire

Uğur Günay, Serdar Ulusoy, Gebrail Bekdaş, and Sinan Melih Nigdeli

Abstract When designing load-bearing reinforced concrete components, external

factors such as fire should be taken into account in addition to horizontal and vertical
forces. Parameters such as cross-section of the reinforced concrete components,
concrete cover, material properties and quantity of the concrete steel affect signif-
icantly the fire resistance of the reinforced concrete section exposed to fire. In this
study, optimal cross-sections are obtained for 30, 60 and 90 fire duration taking into
account the different cross-sections and concrete cover of the reinforced concrete
column. The method for calculating the optimal sections of reinforced concrete
columns is the modified metaheuristic algorithm, which is a combination of Flower
Pollination Algorithm (FPA) with Jaya Algorithm (JA). The behavior of reinforced
concrete column in case of fire is investigated according to EN-1992-1-2 (Eurocode
2: Design of reinforced concrete structures-Part 1–2: General rules—Structural fire
design). The Eurocode offers 3 different methods for the fire design of concrete
structures (simplified, advanced calculation methods and the tabulated date). In this
study, optimal results are obtained in the tabulated date section. Thus, some limi-
tations in EN-1992-1-2 such as minimum reinforcement cross-sectional area and
column buckling length have been overcome.

Keywords Reinforced concrete column · Fire design · Metaheuristic algorithm ·

Optimization · Eurocode 2

U. Günay · S. Ulusoy
Department of Civil Engineering, Turkish-German University, 34820 Beykoz, Istanbul, Turkey
e-mail: [email protected]
S. Ulusoy
e-mail: [email protected]
G. Bekdaş · S. M. Nigdeli (B)
Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 35

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_3
36 U. Günay et al.

1 Introduction

Fire is a rapid combustion event that occurs when three main factors come together,
referred to as the fire triangle in the literature. The three main factors in this event
accompanied by flames are combustible material, oxygen and heat source. At the
end of this combustion process, carbon monoxide (CO), which can be fatal for many
living beings, is formed as a combustion product of oxygen and carbon in burning
materials. A further fatal effect of this phenomenon can be seen in the damage
it causes to the failure of buildings. In recent years, especially in large cities, the
incidence and damages caused by this effect have been increasing with the growing
density of building stock per unit area. In the 10 year period between 2011 and 2020,
over 13 million fire incidents were recorded across the United States, resulting in
over 30,000 fatalities [1]. As the statistics show, the importance of the design and
protection of buildings against fire is getting vital.
The strength properties of concrete and structural steel materials, which are the
two main components of reinforced concrete structural elements, have an important
connection with temperature change. In particular, it is undeniable that a sudden
effect such as fire, which rises above 1000 °C (Fig. 1) in short periods, leads to
rapid heating of the materials themselves depending on their thermal properties,
thus causing sharp changes in their behavior [2]. In a study on the fire resistance of
concrete specimens with different compressive strengths, it was observed that the
specimens lost more than half of their compressive strength at 600 °C and more than
75% of their tensile strength at room temperature (20 °C) [3]. On the other hand,
in another study on the behavior of the other main element, structural steel, at high
temperatures, a strength loss exceeding 80% was observed in the strength of 12 mm
diameter structural steel with a yield strength of 426 MPa at normal temperatures
(Fig. 2) [4]. It should be noted here that this temperature does not reach all points
of the column at the same time and this loss of strength is not in the whole element.
Attention should be paid to spalling, which is one of the most important factors that
cause this spread to accelerate.
Resilience capacity in the context of a fire disaster reflects the ability to prevent a
fire from occurring and to limit the initial damage if a fire does occur. When protection
methods are considered in order to limit this damage, two main headings, active and
passive protection, emerge. While active protection requires an external intervention,
passive protection can be seen as measures taken in the design and manufacture
of building elements. Systems such as audible warning systems, smoke detection
systems, and automatic fire sprinklers that are already in the building before the fire
and are activated during the fire can be given as examples of active protection systems.
The design of building elements, fire compartments, fire walls, paints that increase
resistance to heat and thermal barrier building elements made of polypropylene fiber
can be examples of passive protection.
In addition to protecting the structural steel in reinforced concrete elements against
corrosion, the concrete cover also protects against events such as fire, which cause
high temperatures and greatly reduce the ductility of steel. Concrete, which plays a
Optimum Design of Reinforced Concrete Columns in Case of Fire 37

Fig. 1 ISO 834 standard fire curve (ISO, 1975)

Fig. 2 Stress–strain graph

of structural steel at elevated
temperatures [4]

major role in resistance against pressure, which is one of the main tasks of elements
such as columns, causes loss of cross-section as a result of pouring. This can lead to
an increase in stress and result in the collapse of the structure. Many factors such as
aggregate type and quantity, concrete cover and water content affect the spalling. The
water and moisture in the concrete tend to expand with increasing temperatures and
move to a volume larger than the volume of the voids inside. This leads to explosive
spalling, which is the most important type of spalling [5]. For this reason, the spread
38 U. Günay et al.

of heat is accelerating and protection methods to reduce this spread are becoming
very important.
The studies and the codes implemented as a result of these studies prioritize the
geometry of the concrete in order to design the structural elements resistant to a
certain fire duration. In addition to obtaining the required cross-section for design,
it is also economically very important to determine the optimum value of the size of
this cross-section.
When the numerical calculation of fire resistance of reinforced concrete columns
was checked through an experimental study, the strength time prediction was quite
acceptable, but the displacement prediction could not be made accurately because
the analytical models of the material behavior of concrete and steel were inadequate
[6]. A study investigating the restraint of thermal expansion of a reinforced concrete
column shows that the fully restrained columns are slightly beneficial for the fire
performance of the column as the surrounding structural elements can transfer part of
the load to other columns as they normally would [7]. A study of 80 column specimens
with different reinforcement ratios and geometry showed that reinforcement ratio
and cross-section size are very important parameters in fire resistance [8]. Another
study with columns exposed to 400–800 °C showed that the distance between the
reinforcement and the edge of the concrete was at least 20 mm and the cross-sectional
separation was at least 100 mm, which made a sharp difference in increasing the
fire resistance, and it was said that the use of stirrups also provided a significant
increase [9]. In a study on high-strength concrete (HSC) columns, it was shown
that the fire performance was longer than that of normal-strength columns and a
concrete cover of at least 30 mm was recommended for the fire resistance of HSC
[10]. In a study investigating the effect of Fibre-Reinforced Polymer (FRP) wrapped
concrete columns on fire resistance, a total of 6 columns of 3.81 m length and 0.4 m
diameter, two of which were wrapped with FRP, were exposed to a standard fire
[11]. The study showed that the wrapped columns achieved a 5 h ULC-S101 fire
endurance rating. In a study that investigate the effect of reinforcement with only
fibers and not FRP, with 99 normal and HSC columns, explosive spalling, which is
the most important spalling type, was observed in the first 45 min in all elements
and it was seen that the use of polypropylene fiber was instrumental in reducing
spalling up to 22% [12]. A study investigating the relationship between the cross-
sectional shape of the element and fire resistance was conducted with 9 different
cross-sectional columns, 3 of which were L, 3 of which were T and 3 of which
were +type, and the highest strength was found in the T type cross-section, while
the lowest strength was found in the +type with the highest surface area [13]. In
the behavior of concrete-filled steel slender columns under the effect of 30–120 min
fire, the columns with a height/diameter ratio of 20, 30 and 40 obtained significantly
higher strength than the columns with a ratio of 20. As the slenderness increased,
a significant increase in the decrease in strength was observed [14]. In a study on
the fire resistance of reinforced concrete columns produced with recycled aggregate
(RA), the strength of the elements produced from normal concrete under the same
axial load was lower than that produced with RA [15]. The reason for this is that the
penetration of heat is more difficult in elements produced with RA. It investigated
Optimum Design of Reinforced Concrete Columns in Case of Fire 39

the fire resistance of columns with 50, 100 and 150 mm stirrup spacing and found
that as the stirrup spacing increased, significant decreases in strength were observed
due to the decrease in load-carrying capacity and the possibility of spalling [16].
However, the early crack formation was observed in columns with frequent stirrups.
In a study investigating the fire effect of reinforced concrete columns with seismic
damage history, 6 columns were exposed to 120 min of fire. As a result of the
research, it was observed that the fire resistance of the columns with seismic damage
history decreased dramatically and collapse occurred after 120 min of fire, while
the columns in the control group did not experience any collapse [17]. In another
study that aims to investigate the relationship between eccentricity and fire resistance,
loaded reinforced concrete columns with eccentricities of 20 and 40 mm. Increasing
the eccentricity by 100% resulted in a 200% increase in spalling and a 43% decrease
in strength [18].
There are many studies in the literature about the optimization of reinforced
concrete structural components or structures. The optimum diameter of the rein-
forced concrete column, the optimum dimensions of the reinforced footing, the
minimum structural cost of the reinforced concrete retaining wall are determined
using metaheuristic algorithms [19–22]. Different metaheuristic Algorithms such
as Flower Pollination Algorithm (FPA), Jaya algorithm (JA), Teaching–Learning-
Based optimization (TLBO), Harmony search (HS) algorithm and modified Harmony
search algorithm are used to calculate the optimum cross-section of T-shaped or rect-
angular reinforced concrete beam, a post-tensioned reinforced concrete cylindrical
wall, reinforced concrete cantilever soldier piles and reinforced concrete deep beams
considering Carbon emission [23–28].

2 Design of Reinforced Concrete Column According

to Eurocode 2

In EN 1992-1-2 [29], many calculation methods related to fire design and resistance of
reinforced concrete structures are proposed. Among these, the advanced calculation
method uses Fourier differential equations and numerical methods such as finite
element and finite volume to calculate the heat transfer in structural elements. As
a result of these calculations, the temperature value at any point of the structural
element at any point in time, for example the point where the reinforcement is located,
can be calculated. These values for specific sections are already available in Annex
A of EN 1992-1-2. The effect of this thermal change in sections on the mechanical
response is also calculated. In another method, the simplified calculation method, a
faster calculation can be made using the temperature profiles available in Annex A.
Apart from these two methods, a tabulated data section is presented. Table 5.2
in Eurocode 2 with certain cross-sections is available in this section. In a study, 82
fire tests were performed on columns using the database from which Table 5.2 was
created and Formula (5.7) in Eurocode 2 was derived from regression analysis of the
40 U. Günay et al.

results of these tests [30]. Therefore, this formula is presented in the EC under the
tabulated data section.
The table method can be used to obtain the minimum cross-section and concrete
cover values of the columns at certain loading conditions (μfi = 0.2, μfi = 0.5 and
μfi = 0.7) for fire durations from R30 to R240. Linear interpolation can be applied
for values not shown in the table. These values depend on the exposure of one or
more surfaces of the element to fire (Table 1). In order to use Table 5.2, columns
must have the following values:
• Effective length of the column in fire condition: l0,fi ≤ 3 m,
• 1st degree eccentricity value in case of fire: e = M 0Ed,fi /N 0Ed,fi ≤ emax ;
recommended emax value = 0.15h,
• Reinforcement area ratio: As < 0.04Ac .

must be met. Here, the effective length l0,fi of the column under fire effect is considered
equal to the length l in the non-fire condition. l is the distance between the axes
where the columns meet the beams; if the floor where the column is located is the
intermediate floor, the l0,fi value is taken as 0.5l, and if it is the upper floor, it is
taken between 0.5l and 0.7l. In the same way, the eccentricity value in the fire-free
condition can also be used in the calculation in case of fire.
The main determining factor in the table, μfi , the load utilization factor in the fire
condition, is the ratio of the axial load rating N Ed.fi of the column in the fire condition
to the axial load rating N Rd in the normal condition. N Rd can be calculated according
to 1992-1-1.
/
μfi = NEd.fi NRd (1)

Table 1 Minimum column dimensions and axis distances for columns with rectangular or circular
section
Standard fire resistance Minimum dimensions (mm)
Column width bmin /axis distance a of the main bars
Column exposed on more than one side Exposed on one side
μfi = 0.2 μfi = 0.5 μfi = 0.7 μfi = 0.7
R 30 200/25 200/25 200/32 155/25
300/27
R 60 200/25 200/36 250/46 155/25
300/31 350/40
R 90 200/31 300/45 350/53 155/25
300/25 400/38 450/40a
R 120 250/40 350/45a 350/57a 175/35
350/35 450/40a 450/51a
R 180 350/45a 350/63a 450/70a 230/55
R 240 350/61a 450/75a – 295/70
a Minimum 8 bars
Optimum Design of Reinforced Concrete Columns in Case of Fire 41

EC states that the reduction factor ηfi can be used instead of μfi . The reason for
this is to assume that the design force N Ed is equal to the axial load resistance N Rd , in
other words, to assume that the column is performing at full capacity and to design
safely. EC states that ηfi can be taken as 0.7 for simplification but gives the following
equations for detailed calculation. In the following equations, the live load value is
Qk,1 and the fixed load value is Gk , γ Q,1 = 1.5 and γ G = 1.35 are taken as safety
coefficients.

ηfi = NEd.fi /NEd (2)

NEd,fi = G k + ψfi Q k,1 (3)

NEd = γG G k + γQ,1 Q k,1 (4)

It is possible to make a rapid design with Table 5.2 in Eurocode 2 (Table 1).
However, the fire resistance duration of most of the sections already designed cannot
be clearly determined because they are not included in Table 5.2. Equation (5.7) in
Eurocode 2 can give the fire resistance time of a given section by taking into account
the effect of the load level Rηfi , the effect of the distance between axes Ra , the effect
of the column length Rl , the effect of the cross-section size Rb and the effect of the
reinforcement Rn . These effects are calculated with the formulas given below.

Rηfi = 83[1.00 − μfi ((1 + ω))/((0.85/αcc ) + ω)] (5)

Ra = 1.60(a − 30) (6)

( )
Rl = 9.60 5 − lo,fi (7)

Rb = 0.09b' (8)

Rn = 0 (n = 4) or Rn = 12 (n > 4) (9)

With the combination of the above effects, Eq. (5.7) is determined as follows.
(( ) )1.8
R = 120 Rμfi + Ra + Rl + Rb + Rn /120 (10)

R gives the fire resistance time of the column element and the inputs in the above
equations are subject to the following constraints and formulas.
• A is the distance between axes of longitudinal reinforcement; 25 mm ≤ a ≤
80 mm,
42 U. Günay et al.

• l o,fi is the effective length of the column in case of fire; 2 m ≤ lo,fi ≤ 6 m (l o,fi less
than 2 m shall be taken as 2).
• b' for rectangular sections = Ac /(b + h), for circular columns = Φ column ;
200 mm ≤ b' ≤ 450 mm; h ≤ 1.5b,

• ω, mechanical reinforcement ratio at normal temperature conditions = (Asfyd )/

(Acfcd ),
• α cc coefficient for long-term effects on compressive strength; α cc = 0.85

3 Hybrid Metaheuristic Algorithm

Flower Pollination Algorithm (FPA), which is inspired by the properties of flowers

while considering changes in color and smell, have been developed by Yang [31].
The creation of local pollination and global pollination situations in the algorithm
ensures that optimal results of the engineering problems are achieved. In this study,
Flower Pollination Algorithm is used to compute the optimum cross-section of the
reinforced concrete column in case of fire. The phases of the optimization process
can be summarized as follows:
• First Step: Identify the design variable of the reinforced concrete column in case
fire (shown in Table 2)
• Second Step: Produce randomly the candidate cross-sections of the reinforced
column and the associated total weight
• Third Step: Give a start for optimization process with FPA
• Fourth step: Replace present candidate cross-sections using local or global
pollination taking into account the cross-sections with the lowest total weight
• Fifth Step: Carry on process in step 4 for 10,000 iterations
• Last Step: Complete the optimization process.

Table 2 Design variables of reinforced concrete in case of fire

Explanation Symbol Unit Min. value Max. value
Width b mm 200 450
Height h mm 200 450
Compressive strength of concrete f ck MPa 16 50
Yield strength of steel f yk MPa 420 500
Clear cover cnom mm 30 45
Specific gravity of reinforced concrete Γ kg/m3 2300 2800
Buckling length in fire l fi m 2 6
Degree of utilization in case of fire μfi – 0.2 0.7
Fire duration T min 30 90
Optimum Design of Reinforced Concrete Columns in Case of Fire 43

A simple algorithm called Jaya [32], which has no specific check parameters, are
developed by Rao after the effectiveness of the TLBO [33] in optimization problems.
Jaya algorithm aims to be near to the best solution by avoiding the convergence to
the worst solution.
The mathematical expression of local, global pollination and Jaya algorithm are
given in Eqs. (11), (12) and (13), respectively. g* is the best present cross-section,
w* is the present worst cross-section, xit+1 is the newly produced cross-section, xit is
the present cross-section, r 1 and r 2 are two random numbers between 0 and 1, L is
a Lévy distribution, ε is the linear distribution, x tj and xkt are two randomly chosen
cross-sections.
( )
xit+1 = xit + ε x tj − xkt (11)

( )
xit+1 = xit + L xit − g ∗ (12)

( ) ( )
xit+1 = xit + r1 g ∗ − abs(xit ) − r2 w ∗ − abs(xit ) (13)

A hybrid algorithm is proposed to find the optimum cross-section of the reinforced

concrete columns by integrating a part of Jaya Algorithms to FPA. The code of these
generated algorithms is written in Matlab [34]. The global pollination equation is
removed from FPA’s codes and the Jaya algorithm equation is added in its place. The
Flowchart diagram of the hybrid algorithm is given in Fig. 3.

4 Numerical Examples

Optimum results of the reinforced concrete column with different cases are obtained,
which are shown in Table 3 as a numerical example. In the first case, the duration of
the fire is accepted as 90 min. In this case, the clear cover is 40 mm and the number
of reinforcement bars is over 4. Also, this situation corresponds to both cross-section
types, square and circular. Height and width values for a square section are equal to
the diameter for a circular section. In the second and third cases, the fire duration of
the examined cross-section is 60 min and the number of reinforcement bars is again
over 4. Both case 2 and case 3 shows the situation resulting from the use of different
clear covers. In the second case, the clear cover is 35 mm, while in the third case,
40 mm is taken. In the following cases, the number of reinforcement bars is taken as
4. These cases are investigated under 30, 60 and 90 min fire exposure with a clear
cover between 30 and 45 mm.
The optimum cross-section values of the reinforced concrete column according to
the hybrid metaheuristic algorithm for different cases are presented in Figs. 4 and 5.
In Case 1, the optimum height and width of the column vary between 200 and 442 mm
according to different loading conditions factors and buckling length. In case 1, no
optimum cross-section is calculated if the loading condition factor (μfi ) is greater
44 U. Günay et al.

Fig. 3 Flowchart diagram of hybrid algorithm

than 0,6 and the buckling length (l0,fi ) is greater than 4 m. Optimum cross-section
values decrease by rising the number of bars in the cross-section. The importance
of the clear cover is visible from the comparison of cases 5 and 6. For example, the
optimum width and height for a buckling length of 3 m and a load capacity of 0.7
Optimum Design of Reinforced Concrete Columns in Case of Fire 45

Table 3 Cases for reinforced concrete columns in case of fire

Cases Corner bars Clear cover (mm) Fire duration (min)
Case 1 40 90

Case 2 35 60

Case 3 30 60

Case 4 40 90

Case 5 35 60

Case 6 30 60

Case 7 30 30

are increased from 329 to 418 mm. In case 7, the optimum cross-section is available
for all loading condition factors and buckling length.
46 U. Günay et al.

Fig. 4 Optimum results of reinforced column cross-section under cases 1, 2 and 3

Fig. 5 Optimum results of reinforced column cross-section under cases 4, 5, 6 and 7

Optimum Design of Reinforced Concrete Columns in Case of Fire 47

5 Results and Conclusion

In this study, a hybrid metaheuristic algorithm is proposed for the optimum design of
the reinforced concrete column in case of fire according to Eurocode 2. The results
of this study are as follows:
• Although the design of column elements in case of fire according to Eurocode 2 is
simple using Table 5.2, it is not possible to obtain economical cross-sections. The
values in the table are prepared for α cc = 1,0. This also results in higher values
for E d,fi . The optimum cross-section for the reinforced column under different
fire conditions is calculated by using Eq. (5.7) in Eurocode 2 and metaheuristic
algorithms.
• Equation (5.7) in Eurocode is permissible to design rectangular columns with lo,fi
< 6. This value is limited to 6 m in Table 5.2. The effects of the buckling length,
loading condition factor, clear cover, fire time and the number of reinforcements
bar on the optimum cross-section are shown through numerical examples.
• If the hybrid algorithm and the FPA algorithm are compared, both algorithms give
the same results. However, the hybrid algorithm achieves the optimal result in a
shorter time. This shortens the calculation time. This situation turns out to be an
important factor when investigating the fire situation of larger three-dimensional
structures.

References

1. United States Fire Administration: Fire statistics. http://usfa.fema.gov/statistics/. Accessed 6

Apr 2023
2. ISO: Fire resistance tests-elements of building construction. International Standard ISO 834,
Geneva (1975)
3. Park, G.K., Yim, H.J.: Evaluation of fire-damaged concrete: an experimental analysis based on
destructive and nondestructive methods. Int. J. Concr. Struct. Mater. 11(3), 447–457 (2017)
4. Guo, Z., Shi, X.: Experiment and Calculation of Reinforced Concrete at Elevated Temperatures.
Elsevier, pp. 56–66 (2011)
5. Kahanji, C., Ali, F., Nadjai, A.: Explosive spalling of ultra-high performance fibre reinforced
concrete beams under fire. J. Struct. Fire Eng. 7(4), 328–348 (2016)
6. Haksever, A., Anderberg, Y., Haksever, A., Anderberg, Y.: Comparison between measured and
computed structural response of some reinforced concrete columns in fire. Fire Saf. J. 4(4),
293–297 (1982)
7. Lie, T.T., Lin, T.D.: Influence of restraint on fire performance of reinforced concrete columns.
Fire Saf. Sci. 1, 291–300 (1986)
8. Lie, T.T.: Fire resistance of reinforced concrete columns: a parametric study. J. Fire Prot. Eng.
1(4), 121–129 (1989)
9. Lin, T.D., Zwiers, R.I., Burg, R.G., Lie, T.T., McGrath, R.J.: Fire Resistance of Reinforced
Concrete Columns (No. RD101B) (1992)
10. Kodur, V.K.R., Sultan, M.A.: Structural Behaviour of High Strength Concrete Columns
Exposed to Fire (1998)
11. Bisby, L.A.: Fire Behaviour of Fibre-Reinforced Polymer (FRP) Reinforced or Confined
Concrete, p. 2520. Queen’s University, Kingston (Kanada) (2003)
48 U. Günay et al.

12. Ali, F., Nadjai, A., Silcock, G., Abu-Tair, A.: Outcomes of a major research on fire resistance
of concrete columns. Fire Saf. J. 39(6), 433–445 (2004)
13. Xu, Y.Y., Wu, B.: Fire resistance of reinforced concrete columns with L-, T-, and +-shaped
cross-sections. Fire Saf. J. 44(6), 869–880 (2009)
14. Romero, M.L., Moliner, V., Espinos, A., Ibañez, C., Hospitaler, A.: Fire behavior of axially
loaded slender high strength concrete-filled tubular columns. J. Constr. Steel Res. 67(12),
1953–1965 (2011)
15. Dong, H., Cao, W., Bian, J., Zhang, J.: The fire resistance performance of recycled aggregate
concrete columns with different concrete compressive strengths. Materials 7(12), 7843–7860
(2014)
16. Ser˛ega, S.: Effect of transverse reinforcement spacing on fire resistance of high strength concrete
columns. Fire Saf. J. 71, 150–161 (2015)
17. Wu, B., Xiong, W., Liu, F.: Fire behaviours of concrete columns with prior seismic damage.
Mag. Concr. Res. 69(7), 365–378 (2017)
18. Buch, S.H., Sharma, U.K.: Fire resistance of eccentrically loaded reinforced concrete columns.
Fire Technol. 55, 1517–1552 (2019)
19. Nigdeli, S.M., Bekdas, G., Kım, S., Geem, Z.W.: A novel harmony search based optimization
of reinforced concrete biaxially loaded columns. Struct. Eng. Mech. 54(6), 1097–1109 (2015)
20. Bekdaş, G., Cakiroglu, C., Kim, S., Geem, Z.W.: Optimization and predictive modeling of
reinforced concrete circular columns. Materials 15(19) (2022)
21. Bekdaş, G., Cakiroglu, C., Kim, S., Geem, Z.W.: Optimal dimensioning of retaining walls
using explainable ensemble learning algorithms. Materials 15(14) (2022)
22. Nigdeli, S.M., Bekdaş, G., Yang, X.: Metaheuristic optimization of reinforced concrete
footings. KSCE J. Civ. Eng. 22(11), 4555–4563 (2018)
23. Yucel, M., Bekdaş, G., Işıkdağ, Ü., Apak, S.: Optimum modelling for rectangular shape rein-
forced concrete (RC) beam with the aim of minimum CO2 emission. J. Environ. Prot. Ecol.
22(5), 1992–2002 (2021)
24. Yucel, M., Nigdeli, S.M., Bekdaş, G.: Generation of sustainable models with multi-objective
optimum design of reinforced concrete (RC) structures. Structures 40, 223–236 (2022)
25. Bekdaş, G., Yucel, M., Nigdeli, S.M.: Generation of eco-friendly design for post-tensioned
axially symmetric reinforced concrete cylindrical walls by minimizing of CO2 emission. Struct.
Des. Tall Spec. Build. 31(13) (2022)
26. Bekdaş, G., Nigdeli, S.M., Kim, S., Geem, Z.W.: Modified harmony search algorithm-based
optimization for eco-friendly reinforced concrete frames. Sustainability 14(6), 3361 (2022)
27. Arama, Z.A., Kayabekir, A.E., Bekdas, G., Geem, Z.W.: CO2 and cost optimization of rein-
forced concrete cantilever soldier piles: a parametric study with harmony search algorithm.
Sustainability 12 (2020)
28. Cakiroglu, C., Islam, K., Bekdaş, G., Apak, S.: Cost and CO2 emission-based optimisation of
reinforced concrete deep beams using Jaya algorithm. J. Environ. Prot. Ecol. 23(6), 2420–2429
(2022)
29. Eurocode 2: Design of Concrete Structures, Part 1–2: General Rules–Structural Fire Design,
EN 1992-1-2 (2004)
30. Franssen, J.M.: Design of concrete columns based on EC2 tabulated data–a critical review. In:
1rst int. Structures in Fire Workshop (2000)
31. Yang, X.S., Karamanoglu, M., He, X.: Flower pollination algorithm: a novel approach for
multiobjective optimization. Eng. Optim. 46(9), 1222–1237 (2014)
32. Rao, R.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7(1), 19–34 (2016)
33. Rao, R.V., Savsani, V.J., Vakharia, D.P.: Teaching–learning-based optimization: a novel method
for constrained mechanical design optimization problems. Comput. Aided Des. 43(3), 303–315
(2011)
34. MathWorks Inc.: MATLAB R2015a.Natick, MA, USA (2020)
Hybrid Social Network Search
and Material Generation Algorithm
for Shape and Size Optimization of Truss
Structures

M. Saraee, A. Jafari, D. Yazdani, M. Baghalzadeh Shishehgarkhaneh,

B. Nouhi, and S. Talatahari

Abstract Metaheuristic algorithms (MH) are widely used in engineering applica-

tions due to their global search capabilities and independence from gradient infor-
mation. Truss structures are commonly used in structural design, and the best solu-
tion typically involves minimizing the weight while ensuring adequate strength and
stiffness. To achieve this goal, several hybrid MH algorithms have been proposed
by combining the strengths of two or more algorithms. In this paper, we propose
a hybrid social network search (SNS) and material generation algorithm (MGA)
for truss structure optimization. The main levels as parallel and series levels are
defined for the hybrid algorithm. The proposed algorithm is evaluated using several
benchmark truss structures, and the results demonstrate its superiority over other
state-of-the-art MH algorithms.

Keywords Social network search · Material generation algorithm · Truss

structure · Structural optimization

M. Saraee · A. Jafari · S. Talatahari (B)

Department of Civil Engineering, University of Tabriz, Tabriz, Iran
e-mail: [email protected]
D. Yazdani
Department of Computer Engineering, Mashhad Branch, Azad University, Mashhad, Iran
M. B. Shishehgarkhaneh
Department of Civil Engineering, Monash University, Clayton, VIC 3800, Australia
B. Nouhi · S. Talatahari
Faculty of Engineering & IT, University of Technology Sydney, Ultimo, NSW 2007, Australia

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 49

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_4
50 M. Saraee et al.

1 Introduction

Structural optimization problems have emerged as one of the most active subfields
of structural engineering. Optimization entails minimizing or maximizing a func-
tion and monitoring the system’s performance. Designers created a plethora of
knowledge-based models to get a model with better qualities. Metaheuristic (MH)
algorithms are searching strategies that employ higher-level methodologies to
perform search operations to find optimum results. These approaches have earned
a lot of interest due to their tremendous potential for modeling engineering issues
in instances where traditional techniques have failed to provide a solution. They are
more capable of global search than traditional optimization algorithms and do not
need gradient information [1]. Some of the prominent and novel MH algorithms
are Genetic Algorithms (GAs) [2], Particle Swarm Optimization (PSO) [3], Chaos
Game Optimization (CGO) [4, 5], Energy Valley Optimizer (EVO) [6], Charged
System Search (CSS) [7–10], Crystal Structure Algorithm (CryStAl) [11, 12], Bat
Algorithm (BA) [13], Grey Wolf Optimizer (GWO) [14], Whale Optimization Algo-
rithm (WOA) [15], Nutcracker Optimization Algorithm (NOA) [16], Cuckoo Search
(CS) [17], Fire Hawk Optimizer (FHO) [18, 19], Plant Competition Optimization
(PCO) [20], Salp Swarm Algorithm (SSA) [21], Atomic Orbital Search (AOS) [22,
23], Jellyfish Search (JS) [24], Krill Herd (KH) [25], Black Widow Optimization
(BWO) [26], Special Relativity Search (SRS) [27], and Fusion–Fission Optimiza-
tion (FuFiO) [28]. Furthermore, the hybrid version of MH algorithms—which are
two or more algorithms that work in tandem and complement one another for creating
a beneficial synergy [29]—has been developed and proposed for different purposes
[30–37].
A structural optimization issue involves the structure system’s weight minimiza-
tion by using various parameters, like layout and size, as the decision variables.
Cross-sectional areas and nodal coordinates of the structure are used as the deciding
factors in size and layout optimization, respectively [38]. Truss structures are made
of stiff elements that are pin-connected at the joints and solely provide axial forces.
Truss structures are made up of rigid beams that are joined together with pins and
only exert axial forces. With this simple shape, trusses could be deemed as a group
of connected three-dimensional segments, where each element has exactly two ends
and each joint can fit any number of beams [39]. The best solution in truss optimiza-
tion is typically understood to be achieving the desired structure while using the least
amount of material and the best structural configuration to sustain the design loads
and, thereby making the design effective regarding weight and strength/stiffness.
Within a predetermined structural typology, this also indirectly lowers construction
costs [40]. Generally speaking, to optimize truss structures, the topology is often
changed beginning with a ground structure, which is the collection of all feasible
member placements between the truss nodes, and the design factors include the sizes
of the member cross sections [41].
Over the last few years, MH algorithms have been employed in truss optimization
problems. Kazemzadeh Azad and Aminbakhsh [42] employed one of the novel MH
Hybrid Social Network Search and Material Generation Algorithm … 51

algorithms called guided stochastic search (GSS) in dealing with large-scale steel
truss structure optimization problems. Awad [43] used a political optimizer (PO)
algorithm for seven different truss structure optimization problems; According to the
results, the PO algorithm surpasses various cutting-edge optimization approaches in
small- or medium-sized structural systems regarding algorithmic stability, optimized
weight, and convergence rates. Li et al. [44] proposed an improved version of the
chicken swarm optimization (CSO) algorithm for ameliorating the design effective-
ness of the truss optimization. Azizi et al. [5] employed the chaos game optimization
(CGO) algorithm in dealing with truss structures’ shape and size optimization, in
which they evaluated the capability of the CGO algorithm on five different large-
scale truss structures. Pierezan et al. [45] employed the modified version of the
coyote optimization algorithm (MCOA) for truss structure optimization problems
considering discrete variables. Khodadadi and Mirjalili [46] employed the general-
ized normal distribution optimization (GNDO) approach for the weight optimization
of truss structures. Lemonge et al. [47] utilized the improved differential evolution
(IDE) algorithm for spatial and plane truss optimization problems. Jawad et al. [48]
employed the artificial bee colony (ABC) algorithm in dealing with members’ size
and layout optimization of truss structures. However, some research works have
employed the hybrid version of MH algorithms in truss optimization problems. Liu
and Xia [49] suggested a novel hybrid intelligent genetic algorithm (HIGA) for the
sake of improving the truss optimization problems’ efficiency. Furthermore, for truss
structures’ weight minimization, Yücel et al. [50] proposed a hybrid optimization
algorithm regarding miscellaneous MH algorithms, consisting of flower pollination
algorithm (FPA), teaching–learning-based optimization (TLBO), and Jaya algorithm
(JA). Omidinasab and Goodarzimehr [51] proposed a novel hybrid version of the GA
and PSO algorithm in coping with truss structure optimization problems considering
discrete design variables. Kaveh and Talatahar [52] developed a hybrid version of
the ant colony and PSO algorithms for truss optimization problems. Table 1 summa-
rizes some recent hybrid algorithms applied for solving truss structure optimization
problems.
In the current research work, the optimization of truss structures is deemed using
a new hybrid algorithm based on Social Network Search (SNS) algorithm and Mate-
rial Generation Algorithm (MGA). SNS was proposed by Talatahari et al. [68] and
it is inspired by users’ efforts to increase their reputation on social networks by
modeling their emotions, such as Imitation, Conversation, Disputation, and Innova-
tion, in the actual world while expressing their thoughts. Furthermore, MGA was
proposed by Talatahari et al. [69]. Several advanced and fundamental features of
material chemistry, including the arrangement of chemical compounds and chemical
processes in the production of new materials, are employed as inspiration for the
MGA. The 37-bar, 52-bar, and 72-bar truss structures—three of the field’s bench-
mark problems—are regarded as design examples for purposes of computation. The
main purpose of this research is to assess the capability of the proposed algorithm
in dealing with the optimization of complex truss structures and compare its results
with other well-known metaheuristic oprimization algorithms.
52 M. Saraee et al.

Table 1 List of some of the hybrid algorithms in truss structure optimization problems
No References Year Algorithms Optimization purposes
1 [53] 2023 PSO, SQP, and FE Optimum design of truss structures
2 [54] 2023 CHT and MRFO Truss structures’ size and shape
optimization
3 [55] 2023 DE and Jaya Size optimization of truss structures
4 [56] 2022 ICA and BBO Optimum design of spatial truss
structures
5 [57] 2022 TLBO and CSS Minimization of truss structures’
weight
6 [58] 2022 PSO and GA Minimization of truss structures’
weight
7 [59] 2022 AOA and DE Optimum design of truss structures
considering frequency constraints
8 [60] 2022 DE and SOS Size, shape, and topology optimization
of truss structures
9 [61] 2021 DE and SOS Size and shape optimization of truss
structures considering frequency
constraints
10 [62] 2021 PSO and CA Optimum design of truss structures
11 [63] 2021 FPA and DE Size optimization of truss structures
12 [64] 2021 HS and Jaya Minimization of large-scale truss
structures’ weight
13 [65] 2020 IWO and SFLA Optimum design of truss structures
14 [66] 2020 TLBO and HS Optimum design of space trusses
15 [67] 2019 EHO and CA Minimization of truss structures’
weight
PSO: particle swarm optimization; SQP: sequential quadratic programming algorithm; FE: finite
element method; ICA: Imperialist Competitive Algorithm; BBO: biogeography-based optimization;
TLBO: teaching–learning-based optimization; CSS: charged system search; GA: genetic algorithm;
CHT: constraint-handling technique; MRFO: manta ray foraging optimization; AOA: arithmetic
optimization algorithm; DE: differential evolution; SOS: symbiotic organisms search; CA: cultural
algorithm; FPA: flower pollination algorithm; EHO: elephant herding optimization; HS: harmony
search; IWO: invasive weed optimization; SFLA: shuffled frog-leaping algorithm

The rest of this research work is structured as follows: the statement of optimum
design of structures and a summary of the SNS and MGA algorithms are represented
in Sect. 2. Section 3 explains the new hybrid approach. The problem statement and
numerical investigations are presented in Sects. 4 and 5. Finally, the core findings of
this research are provided in the last section.
Hybrid Social Network Search and Material Generation Algorithm … 53

2 Utilized Methods

2.1 Social Network Search

The human species is a sociable species that strives to interact with one another
continuously. Social networks are virtual tools that were developed for this purpose as
a result of technological advancements. The suggested SNS algorithm replicates the
interactive attitude of social network users in order to increase their fame. The user’s
perspective may be influenced by other perspectives including Imitation, Conversa-
tion, Disputation, and Innovation. Imitation implies that the opinions of other users
are appealing since users often attempt to emulate one another while expressing their
thoughts. Conversation indicates that people may converse and use other viewpoints.
In the Disputation, people may debate and discuss their viewpoints with a group of
other users. Finally, Innovation reveals that individuals occasionally discuss topics
related to their novel ideas and experiences on social networks. It is possible to
represent imitation mood mathematically as:

X i (new) = X j + rand(−1, 1) × R (1)

R = rand(0, 1) × r (2)

r = X j − Xi (3)

where, X j shows jth user’s view’s (position) vector; X i elucidates ith user’s view’s
vector; rand(0, 1) and rand(−1, 1) represent vectors in intervals [0, 1] and [−1, 1]
randomly. R represents the shock radius, which expresses the effect of the jth user
and is measured as a multiple of r . . In the second mood (Conversation), users get
an understanding of events from various points of view, and ultimately, because of
the variety in viewpoints, they may create a new understanding of the issue using the
following equation:

X i (new) = X k + R (4)

R = rand(0, 1) × D (5)

( ) ( )
D = sign f i − f j × X j − X i (6)

where, X k illustrates the problem’s vector because that was randomly selected to
be discussed; R is chat’s impact, which is based on differences in viewpoints and
signifies a shift in their opinions about the problem (X k ); D elucidates the difference
among users’ viewpoints, and there are no input parameters for this calculation of
view differences; rand(0, 1) represents a vector in the range [0, 1] randomly; in a
54 M. Saraee et al.

conversation, X j clarifies the vector of the perspective of a randomly selected user,

while X i represents the vector of the opinion
( of) the ith user. Furthermore, sign
demonstrates the sign function and sign f i − f j compares f i and f j to identify
the movement direction of X k . However, the third mood (Disputation) describes a
scenario in which users explain and defend their perspectives on situations to other
individuals. In this mood, a subset of users is selected at random to participate as a
commenter or group members, and their perspective is then used to inform the new
viewpoint in dispute. These features are described mathematically as follows:

X i(new) = X i + rand(0, 1) × (M − AF × X i ) (7)

∑ Nr
Xt
M= t
(8)
Nr

AF = 1 + r ound(rand) (9)

where, X i shows the ith user’s view vector; a random vector in the range [0, 1] is
returned by rand(0, 1). M demonstrates the opinions’ mean of group members or
commenters. The Admission Factor, or AF, is a randomly generated number that
may either be 1 or 2. It represents the users’ tendency to insist on their opinions while
speaking with others. rand elucidates a number in the range [0, 1] randomly, while
r ound(.) shows a function that rounds its input to the closest integer value. The size
of the group of commentators, Nr , is a random integer between 1 and Nuser , where
Nuser shows the network users’ number or Network size. Finally, in the last mood
(Innovation), users sometimes share the results of their opinions and experiences.
Each of a subject’s distinctive characteristics may have an impact on how well the
issue is understood. As a consequence, by altering the perception of just one of them,
the subject’s overall meaning will also be altered, leading to the development of a
fresh perspective. These features are described mathematically as follows:

xid(new) = t × x dj + (1 − t) × n dnew (10)

n dnew = lbd + rand 1 × (ubd − lbd ) (11)

t = rand 2 (12)

where, D shows the variables’ number in the issue, and d is the dth variable that is
chosen from the range [1, D] randomly. There are two random numbers in the range
[0, 1] called rand 1 and rand 2 . The dth variable’s minimum and maximum values are
lbd and ubd , respectively. The novel concept about the dth dimension of the issue is
represented by n dnew . x dj shows the current notion offered by another user concerning
the dth variable, and the ith user wants to modify it due to a novel thought (n dnew ).
The novel perspective of the dth dimension will eventually be formed as xi(new) d
.
Hybrid Social Network Search and Material Generation Algorithm … 55

d
Meanwhile, change in one dimension (xi(new) ) affects the whole notion and might be
deemed a new perspective to convey. This procedure could be represented as follows:
[ ]
X i(new) = x1 , x2 , x3 . . . .xid(new) . . . .x D (13)

2.2 Material Generation Algorithm

Material is a concoction of several components made up of the substances that

make up the cosmos and have mass and volume. The procedure of generating novel
materials involves the capacity of several components to combine to produce novel
materials with greater functionality and increased energy levels. To better a mate-
rial’s functionality and regulate its unique qualities, materials are created on an
atomic, nano, micro, or macro scale. Furthermore, chemical reactions and chemical
mixtures are used to modify materials chemically. Hence, compounds, reactions, and
stability—three fundamental ideas in material chemistry—were taken into account
in developing the MGA. A variety of materials (Mat) made up of various periodic
table elements (PTEs) are identified by MGA. In this method, a variety of materials
(Mat n ) that include certain components designated as decision variables are taken
j
into consideration as solution candidates ( P T E i ). These two features are presented
mathematically as follows:
⎡ ⎤ ⎡ j
⎤
Mat 1 P T E 11 P T E 21 · · · P T E 1 · · · P T E d1
⎢ Mat ⎥ ⎢ P T E 1 P T E 2 · · · P T E j · · · P T E d ⎥
⎢ 2⎥ ⎢ 2 ⎥
⎢ . ⎥ ⎢ 2 2
...... . . ..
2
⎥ {
⎢ . ⎥ ⎢ ⎥
⎢ . ⎥ ⎢ ... . . ⎥ i = 1, 2, . . . , n.
Mat = ⎢ ⎥=⎢ ⎥,
⎢ Mat i ⎥ ⎢ P T E i1 P T E i2 · · · P T E i · · · P T E id ⎥ j = 1, 2, . . . , d.
j
⎢ . ⎥ ⎢ ⎥
⎢ . ⎥ ⎢ ...... . . .. ⎥
⎣ . ⎦ ⎣ ... . . ⎦
Mat n PT En PT En · · · PT En · · · PT En
1 2 j d

(14)

Each material (solution candidate) has d elements (decision variables), and n

j
materials are thought of as solution candidates. Furthermore, P T E i is set at random
in the first phase of the optimization process, while the decision variable bound-
aries are established according to the issue under consideration. The PTEs’ starting
positions in the search space are determined randomly as follows:
{
( )
j j j j i = 1, 2, . . . , n.
P T E i (0) = P T E i,min + U ni f (0, 1). P T E i,max − P T E i,min ,
j = 1, 2, . . . , d.
(15)
56 M. Saraee et al.

where, the initial value of the jth element in the ith material is determined by
j j j
P T E i (0); P T E i,min and P T E i,max are the jth decision variable of the ith solution
candidate’s minimum and maximum permissible values, respectively; U ni f (0, 1)
elucidates an integer within the range [0, 1] randomly. The probability theory is used
to simulate the processes of losing, gaining, or sharing electrons for the chosen PTEs.
To do this, a chemical molecule, which is regarded as a new PTE, is configured as
follows using a continuous probability distribution for each PTE:

P T E knew = P T E rr21 ± e− , k = 1, 2, . . . , d. (16)

where, the probabilistic component represented by e− is used to model the process

of losing, gaining, or sharing electrons, and is expressed as a normal Gaussian distri-
bution in the mathematical model. The novel material is denoted as P T E knew . r1 and
r2 are random integers uniformly distributed in the intervals [1, n] and [1, d], respec-
tively. P T E rr21 represents a randomly selected PTE from the material Mat. A new
material, denoted as Mat new1 , is generated using the newly generated PTEs and is
added as a new solution candidate to the original material list (Mat) in the following
manner:
[ ]
Mat new1 = P T E 1new P T E 2new · · · P T E knew · · · P T E dnew , k = 1, 2, . . . , d. (17)

Regarding the randomly picked starting element (P T E rr21 ), the probability of

choosing a new element (P T E knew ) is stated as follows:

( ) 1 −(x−μ)2
f P T E knew |μ, σ 2 = √ .e 2σ 2 , k = 1, 2, . . . , d. (18)
2π σ 2

where, μ represents the distribution related to the randomly chosen PTE’s (P T E rr21 )
mean, median, or expectation; the natural logarithm’s natural base, also known as the
Naperian base, is e; σ and σ 2 are the standard deviation and variance, respectively.
Finally, since various materials might engage in the reactions to varying degrees,
a participation factor (p) is also computed for each material. This characteristic is
presented mathematically as follows:
∑l
m=1 ( p m .Mat m j )
Mat new2 = ∑l , j = 1, 2, . . . , l. (19)
m=1 ( p m j )

where, Mat new2 shows the new material formed by the chemical reaction idea. Mat m
represents the mth randomly picked material from the original Mat; pm is the normal
Gaussian distribution for the mth material participation factor.
Hybrid Social Network Search and Material Generation Algorithm … 57

3 Hybrid Method

Hybrid algorithms are created when two or more algorithms are combined in order
to create a new algorithm that has the best features of each algorithm. By combining
two or more algorithms, those weaknesses can be mitigated, resulting in a stronger
and more efficient algorithm that can solve complex problems faster and more accu-
rately. Additionally, hybrid algorithms are also capable of scaling more effectively,
meaning they can be used for more complex problems. Therefore, hybrid algo-
rithms are highly valuable when it comes to solving problems such as finding the
optimal structural design, where accuracy and speed are crucial. We present a novel
hybrid algorithm that combines social network optimization and Material Generation
Algorithm, which is called SNS-MGA in this chapter.
The SNS-MGA algorithm is defined in accordance with the parallel-series
concept. In other words, the SNS-MGA combines the best features of both serial
and parallel systems. The hybrid algorithm considers the following levels:
Level (1) Parallel Level: The two algorithms are executed separately and in
parallel in the first step. During this step, half of all iterations considered for the
hybrid algorithm are considered.
Level (2) Series Level: In the second step, the two algorithms are combined. We
use the results of the first level as the initial solution to the algorithms at this level.
Using the following equations, we are able to obtain new solutions based on the
concepts of both algorithms:

X i(new) = X j + rand(−1, 1) × rand(0, 1) × (X j − X i ) (20)

( ) ( )
X i(new) = X k + rand(0, 1) × sign f i − f j × X j − X i (21)
(∑ )
Nr
Xt
X i(new) = X i + rand(0, 1) × t
− (1 + r ound(rand)) × X i (22)
Nr

d
xi(new) = rand 2 × x dj + (1 − rand 2 ) × {lbd + rand 1 × (ubd − lbd )} (23)

xi(new) = xrr12 ± e− (24)

∑l
m=1 ( p m .Mat m j )
X new = ∑l (25)
m=1 ( p m j )

A half-number of iterations is reached at the end of this step. Figure 1, sumurize

the levels of the hybrid algorihm.
58 M. Saraee et al.

Combination
Define problem SNS MGA of SNS and
MGA

Parallel Step Series Step

Fig. 1 The levels of the hybrid algorithm

4 Problem Statement

In this part, the generic formulation of structural design optimization issues is

described, along with a weight minimization approach that takes frequency design
constraints into account. The structure’s total weight is calculated for the objective
function. In size optimization problems, the design variables are the cross-sectional
areas of structural components, while in shape optimization problems, the design vari-
ables are the nodal coordinates of structures. These characteristics can be expressed
mathematically as follows:
∑e
W eight(A, X ) = ρi L i (xi )Ai , i = 1, 2, . . . , n (26)
i=1

ω j ≥ ω∗j , j = 1, 2, . . . , p (27)

ωk ≤ ωk∗ , k = 1, 2, . . . , p (28)

up
Al ≤ Al , l = 1, 2, . . . , n (29)

xmlow ≤ xm ≤ xmup , m = 1, 2, . . . , r (30)

where, A is a vector with a total number of structural components that includes

the design variables for the structural elements’ cross-sectional areas. The nodal
coordinate design variables are represented by the vector X , where r shows the total
number of nodes in the structure. ρi and L i elucidate the materials’ density and
structural components’ length, respectively; ωk and ω j show the structure’s kth and
jth natural frequency including the overall number of p frequencies. Structure’s kth
and jth natural frequency’s upper and lower bounds are represented by ωk∗ and ω∗j ,
up
respectively. Al demonstrates the lth member’s (Al ) cross-sectional area’s upper and
up
variables for the mth node’s (xm ) nodal coordinates are xmlow and xm , respectively.
Because structural design optimization is a constraint optimization issue, the
optimization process should be carried out using a constraint handling approach.
Hybrid Social Network Search and Material Generation Algorithm … 59

Consequently, this study employs the penalty constraint handling technique with the
following penalty function:

f penalt y (A) = (1 + ε1 .ϑ)ε2 × W eight(A, X ) (31)

∑q
ϑ= max{0, gi (A, X )} (32)
i=1

where, q shows the design constraints’ overall number; ε1 and ε2 stand for the control
values used to calculate the penalty value throughout the optimization procedure;
gi (A) is the ith design constraint.

5 Design Examples

In the current research work, 5 different truss structures, including 37, 52, and 72-bar
trusses, are considered for optimization purposes. Table 2 represents their physical
characteristics, while their descriptions are described as follows.

5.1 Describing the Examples

5.1.1 37-Bar Truss Structure

In the first design example, a truss structure with 20 nodes and 37 structural members
is taken into consideration as a procedure of size and shape optimization simulta-
neously. While a total of 19 design variables are chosen for the size and shape
optimization of the structure, the constraint restrictions of 20, 40, and 60 Hz for the
first three natural frequencies of the structure are considered. Figure 2a is a graphical
depiction of this structure.

Table 2 Features of the truss

37-bar 52-bar 72-bar
structures
Elasticity modulus (N/ 2.1 × 1011 2.1 × 1011 6.8 × 1010
m2 )
Steels’ density (kg/ 7800 7800 2770
m3 )
Lower bound (m2 ) 0.0001 0.0001 0.645 × 10−4
Upper bound (m2 ) 0.001 0.001 20 × 10−4
Added mass (kg) 10 50 2770
60 M. Saraee et al.

(a)

(b)

(c)

Fig. 2 Graphical view of the selected structure

Hybrid Social Network Search and Material Generation Algorithm … 61

5.1.2 52-Bar Truss Structure

This design example, which includes 52 structural elements and 21 nodes, is the
second shape and size optimization issue in this work. Five shape and eight size
design factors are taken into account concurrently while the optimization methods
are being performed. While taking into account the constraint restrictions of 15.961
and 28.648 Hz for the structure’s first two natural frequencies, all of the free nodes
are allowed to move with a maximum tolerance of ±2 (m). Figure 2b is a graphical
depiction of this structure.

5.1.3 72-Bar Truss Structure

This truss structure, which has 72 members and 20 nodes with constraint constraints
of 4 and 6 Hz for the first and third natural frequencies of the structure, is the second
size optimization problem in this study. Figure 2c is a graphical depiction of this
structure.

5.2 Numerical Results

In this sub-section, we provide a comprehensive overview of the optimum results of

the selected truss design problems. We also present the best results and convergence
histories of the optimization procedures, based on 30 independent runs conducted
for statistical validity. To ensure that our findings are reliable, we compare the results
of our proposed algorithm with those of other metaheuristic approaches found in the
literature.

5.2.1 37-Bar Truss Structure

Table 3 presents the best results obtained by the SNS-MGA algorithm for the 37-
bar truss design problem, based on 30 independent optimization runs conducted
for comparative purposes. The table also includes results from other optimization
methods in the literature, and statistical results are provided to ensure a fair compar-
ison. According to the results obtained, the new algorithm achieves an optimal weight
of 359.917 kg for the 37-bar truss structure, which is very close to the weight achieved
by other methods. However, the new algorithm is significantly more computation-
ally efficient than other algorithms, and it requires only 3000 analyses to arrive at the
final result. The new algorithm also provides a mean weight of 361.25 kg across 30
independent runs, with a standard deviation of 1.68, demonstrating its reliability. The
convergence history of the hybrid algorithm for the 37-bar truss problem is depicted
in Fig. 3. The figure indicate that the algorithm was able to converge to the optimum
design very fast. This improved efficiency was achieved by the algorithm’s ability to
62 M. Saraee et al.

quickly identify the optimal locations for the truss bars, as well as the optimal lengths
for the bars. Consequently, the algorithm was able to identify the best solution for
the structure quickly, while also reducing the number of analyses required to reach
the final outcome.

5.2.2 52-Bar Truss Structure

Table 4 presents the best results obtained by the new algorithm and other meta-
heuristic optimization methods in the literature for the 52-bar truss design problem.
The results indicate that the hybrid algorithm provides the best optimum weight of
193.93 kg with a mean wieght of 197.68 kg. These results demonstrate the effec-
tiveness of the proposed algorithm in addressing complex truss design problems and
meeting multiple design constraints, including simultaneous shape and size optimiza-
tion. In Fig. 4, the convergence history of the SNS-MGA is shown for the 52-bar
truss design problem. The convergence history demonstrates that the algorithm is
able to quickly find the optimal solution to the problem. The results show that the
algorithm is a promising approach for solving truss design problems. Furthermore,
the results exhibit the effectiveness of the new algorithm in achieving the desired
optimization results in a faster and more efficient manner.

5.2.3 72-Bar Truss Structure

SNS-MGA’s convergence history is illustrated in Fig. 5 as it tackles a 72-bar truss

design example. A convergence history of the best optimization runs is presented in
the figure. The figure shows that the optimization process converges to a near-optimal
solution after only a few iterations. The optimization process is efficient, converging
quickly to the optimal design, and the solution is stable, meaning that it remains
close to the optimal value despite changes in the design parameters. As a result
of the speed of the algorithm for detecting the optimal space, the new algorithm
is able to reach an acceptable result after only 500 analyses. This highlights the
effectiveness of the new algorithm and ensures that the design can be optimized
rapidly and effectively without compromising accuracy. In Table 5, the best results
are presented for various metaheuristic approaches, including SNS-MGA, in solving
the 72-bar truss problem. In addition to the optimal design variables, the table also
provides the corresponding statistical results. The results show that SNS-MGA was
able to achieve the best results among the different approaches tested. Additionally,
the new algorithm results in a mean weight of 325.26 kg across 30 independent
runs. The results obtained using the SNS-MGA algorithm are the best among all
other approaches, highlighting its robustness and efficiency. The statistical results
demonstrate that SNS-MGA was able to achieve better results in terms of mean
weight, compared to other approaches. These results also indicate that SNS-MGA
has a higher degree of reliability and accuracy, which makes it the reliable choice
for optimal design.
Table 3 Final results for the 37-bar truss problem
Design Wang et al. Wei et al. Gomes Miguel Farshchin Goodarzimeher Tejani et al. This study
variables [70] [71] [72] [73] et al. [74] et al. [75] [76]
Y3 , Y19 (m) 1.2086 1.1998 0.9637 0.8415 0.9392 0.9830 0.9598 1.0037 0.9598 1.0001
Y5 , Y17 (m) 1.5788 1.6553 1.3978 1.2409 1.3270 1.3803 1.3289 1.3707 1.3867 1.3638
Y7 , Y15 (m) 1.6719 1.9652 1.5929 1.4464 1.5063 1.5645 1.5273 1.5559 1.5698 1.5353
Y9 , Y13 (m) 1.7703 2.0737 1.8812 1.5334 1.6086 1.6871 1.6727 1.6897 1.6687 1.6633
Y11 (m) 1.8502 2.3050 2.0856 1.5971 1.6679 1.7590 1.7509 1.7942 1.7203 1.7346
A1 , A27 3.2508 2.8932 2.6797 3.2031 2.9838 2.9913 2.9219 2.8228 2.9038 2.9773
(cm2 )
A2 , A26 1.2364 1.1201 1.1568 1.1107 1.1098 1.0005 1.0007 1.0019 1.0163 1.0023
(cm2 )
A3 , A24 1.0000 1.0000 2.3476 1.1871 1.0091 1.0042 1.0005 1.0023 1.0033 1.0057
(cm2 )
A4 , A 25 2.5386 1.3655 1.7182 3.3281 2.5955 2.5958 2.6633 2.5221 3.1940 2.6010
(cm2 )
A5 , A23 1.3714 1.5962 1.2751 1.4057 1.2610 1.2139 1.2387 1.1640 1.0109 1.0001
(cm2 )
A6 , A21 1.3681 1.2642 1.4819 1.0883 1.1975 1.1423 1.2030 1.2097 1.5877 1.2736
Hybrid Social Network Search and Material Generation Algorithm …

(cm2 )
A7 , A22 2.4290 1.8254 4.6850 2.1881 2.4264 2.3170 2.4843 2.4757 2.4104 2.5629
(cm2 )
A8 , A20 1.6522 2.0009 1.1246 1.2223 1.3588 1.5100 1.3706 1.3791 1.3864 1.4176
(cm2 )
A9 , A18 1.8257 1.9526 2.1214 1.7033 1.4771 1.5172 1.4618 1.5124 1.6276 1.5216
(cm2 )
(continued)
63
Table 3 (continued)
64

Design Wang et al. Wei et al. Gomes Miguel Farshchin Goodarzimeher Tejani et al. This study
variables [70] [71] [72] [73] et al. [74] et al. [75] [76]
A10 , A19 2.3022 1.9705 3.8600 3.1885 2.5648 2.2722 2.4432 2.5330 2.3594 2.4711
(cm2 )
A11 , A17 1.3103 1.8294 2.9817 1.0100 1.1295 1.2112 1.2758 1.2715 1.0293 1.1810
(cm2 )
A12 , A15 1.4067 1.2358 1.2021 1.4074 1.3199 1.2739 1.3491 1.2929 1.3721 1.2790
(cm2 )
A13 , A16 2.1896 1.4049 1.2563 2.8499 2.9217 2.4934 2.3831 2.3443 2.0673 2.4191
(cm2 )
A14 (cm2 ) 1.0000 1.0000 3.3276 1.0269 1.0004 1.0000 1.0000 1.0003 1.0000 1.0001
Best (kg) 366.50 368.84 377.20 361.50 360.05 359.96 359.88 359.92 360.865 359.917
Mean NA NA 381.20 362.04 360.37 360.83 360.23 361.05 364.852 361.25
SD NA 9.03 4.26 0.52 0.26 0.49 0.47 1.24 2.97 1.68
NFEs NA NA 12,500 20,000 5,000 12,000 12,000 3,000 4,000 3,000
M. Saraee et al.
Hybrid Social Network Search and Material Generation Algorithm … 65

Fig. 3 Convergance history for the 37-bar truss obtained by the hybrid algorithm

Table 4 Final results for the 52-bar truss problem

Design Lin Wei Gomes Miguel Farshchin Tejani Goodarzimeher This
variables et al. et al. [72] [78] et al. [74] et al. [76] et al. [75] study
[77] [71]
Z A (m) 4.3201 5.8851 5.5344 6.4332 5.9531 5.7624 5.9140 5.9033
X B (m) 1.3153 1.7623 2.0885 2.2208 2.2908 2.3239 2.2207 2.1243
Z B (m) 4.1740 4.4091 3.9283 3.9202 3.7037 3.7379 3.7092 3.7405
X F (m) 2.9169 3.4406 4.0255 4.0296 3.9660 3.9842 3.9375 3.9048
Z F (m) 3.2676 3.1874 2.4575 2.5200 2.5001 2.5121 2.5002 2.5013
A1 –A4 (cm2 ) 1.00 1.0000 0.3696 1.0050 1.0002 1.0988 1.0001 1.0095
A5 –A8 (cm2 ) 1.33 2.1417 4.1912 1.3823 1.0962 1.0031 1.1864 1.2488
A9 –A16 (cm2 ) 1.58 1.4858 1.5123 1.2295 1.2252 1.1956 1.2602 1.2636
A17 –A20 (cm2 ) 1.00 1.4018 1.5620 1.2662 1.4555 1.4563 1.4384 1.4369
A21 –A28 (cm2 ) 1.71 1.9110 1.9154 1.4478 1.4172 1.3773 1.4006 1.3796
A29 –A36 (cm2 ) 1.54 1.0109 1.1315 1.0000 1.0003 1.0055 1.0008 1.0019
A37 –A44 (cm2 ) 2.65 1.4693 1.8233 1.5728 1.6204 1.7397 1.5530 1.5893
A45 –A52 (cm2 ) 2.87 2.1411 1.0904 1.4153 1.3296 1.3084 1.3979 1.3859
Best (kg) 298.00 236.04 228.38 197.53 193.18 195.49 193.342 193.629
Mean NA NA 234.30 212.80 197.87 214.66 198.875 197.68
SD NA 37.462 5.22 17.98 5.79 14.14 4.21 4.95
NFEs NA NA 11,270 10,000 15,000 4,000 3,000 3,000
66 M. Saraee et al.

Fig. 4 Convergance history for the 52-bar truss obtained by the hybrid algorithm

Fig. 5 Convergance history for the 72-bar truss obtained by the hybrid algorithm
Hybrid Social Network Search and Material Generation Algorithm … 67

Table 5 Final results for the 72-bar truss problem

Design Miguel [78] Farshchin et al. Tejani et al. Goodarzimeher This study
variables [74] [76] et al. [75]
A1 –A4 (cm2 ) 3.3411 3.4188 3.6957 3.3434 3.5862
A5 –A12 (cm2 ) 7.7587 7.9263 7.1779 7.9841 7.8450
A13 –A16 (cm2 ) 0.6450 0.6450 0.6450 0.6450 0.6450
A17 –A18 (cm2 ) 0.6450 0.6450 0.6569 0.6450 0.6452
A19 –A22 (cm2 ) 9.0202 8.0143 7.7017 7.9805 7.7377
A23 –A30 (cm2 ) 8.2567 7.9603 7.9509 7.8631 7.9276
A31 – A34 (cm2 ) 0.6450 0.6450 0.6450 0.6450 0.6452
A35 – A36 (cm2 ) 0.6450 0.6450 0.6450 0.6450 0.6451
A37 – A40 (cm2 ) 12.045 12.7903 12.3994 12.6510 12.6649
A41 – A48 (cm2 ) 8.0401 8.1013 8.6121 7.7268 7.9628
A49 – A52 (cm2 ) 0.6450 0.6450 0.6450 0.6450 0.6463
A53 – A54 (cm2 ) 0.6450 0.6473 0.6450 0.6450 0.6450
A55 – A58 (cm2 ) 17.380 17.4615 17.4827 17.2382 17.2305
A59 – A66 (cm2 ) 8.0561 8.1304 8.1502 8.1914 8.0091
A67 – A70 (cm2 ) 0.6450 0.6450 0.6740 0.6450 0.6463
A71 – A72 (cm2 ) 0.6450 0.6451 0.6550 0.6450 0.6450
Best (kg) 327.691 327.575 325.558 324.348 324.209
Mean 329.890 327.693 331.122 325.569 325.26
SD 2.59 0.1250 4.227 0.9503 1.231
NFEs 10,000 15,000 4,000 2,500 3,000

6 Conclusion

This paper presented a novel hybrid algorithm, combining the Social Network
Search and Material Generation Algorithm, for optimization of truss structures. The
proposed hybrid SNS and MGA (SNS-MGA) demonstrated remarkable performance
in dealing with various truss optimization problems, with the results showing its
superiority over other existing optimization techniques. The SNS-MGA algorithm
leverages the strengths of both Social Network Search and Material Generation Algo-
rithm to efficiently explore the solution space and converge to the optimal solution.
The algorithm ensures the satisfaction of constraints, such as the natural frequencies
of truss structures, by incorporating these constraints into the optimization process.
To assess the capability of the proposed Hyb SNS-MGA for truss structures’
shape and size optimization, three benchmark problems with varying numbers of bars
(37, 52, and 72) were considered. The optimization procedure considered frequency
constraints as limits, which are commonly encountered in truss design. Multiple
optimization runs were performed to ensure the statistical significance of the results,
and a comparative analysis with other algorithms in the literature was conducted.
68 M. Saraee et al.

The outcomes show that the SNS-MGA algorithm can find optimum designs for the
truss structures.

References

1. Kaveh, A., Talatahari, S.: Particle swarm optimizer, ant colony strategy and harmony search
scheme hybridized for optimization of truss structures. Comput. Struct. 87(5), 267–283 (2009)
2. Holland, J.H.: Genetic algorithms. Sci. Am. 267(1), 66–73 (1992)
3. Kennedy, J., Eberhart, R.: Particle swarm optimization. In Proceedings of ICNN’95-
International Conference on Neural Networks. IEEE (1995)
4. Talatahari, S., Azizi, M.: Chaos game optimization: a novel metaheuristic algorithm. Artif.
Intell. Rev. 54(2), 917–1004 (2021)
5. Azizi, M., et al.: Shape and size optimization of truss structures by Chaos game optimization
considering frequency constraints. J. Adv. Res. (2022)
6. Azizi, M., et al.: Energy valley optimizer: a novel metaheuristic algorithm for global and
engineering optimization. Sci. Rep. 13(1), 1–23 (2023)
7. Kaveh, A., Talatahari, S.: A novel heuristic optimization method: charged system search. Acta
Mech. 213(3), 267–289 (2010)
8. Kaveh, A., Talatahari, S.: Charged system search for optimal design of frame structures. Appl.
Soft Comput. 12(1), 382–393 (2012)
9. Kaveh, A., Talatahari, S.: Optimal design of skeletal structures via the charged system search
algorithm. Struct. Multidiscip. Optim. 41(6), 893–911 (2010)
10. Talatahari, S., et al.: Optimization of large-scale frame structures using fuzzy adaptive quantum
inspired charged system search. Int. J. Steel Struct. 22(3), 686–707 (2022)
11. Talatahari, S., et al.: Crystal structure algorithm (CryStAl): a metaheuristic optimization
method. IEEE Access 9, 71244–71261 (2021)
12. Azizi, M., Baghalzadeh Shishehgarkhaneh, M., Basiri, M.: Design optimization of truss
structures by crystal structure algorithm. AUT J. Civil Eng. (2022)
13. Gandomi, A.H., et al.: Bat algorithm for constrained optimization tasks. Neural Comput. Appl.
22(6), 1239–1255 (2013)
14. Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61 (2014)
15. Mirjalili, S., Lewis, A.: The whale optimization algorithm. Adv. Eng. Softw. 95, 51–67 (2016)
16. Abdel-Basset, M., et al.: Nutcracker optimizer: a novel nature-inspired metaheuristic algorithm
for global optimization and engineering design problems. Knowl.-Based Syst. 262, 110248
(2023)
17. Gandomi, A.H., Yang, X.-S., Alavi, A.H.: Cuckoo search algorithm: a metaheuristic approach
to solve structural optimization problems. Eng. Comput. 29(1), 17–35 (2013)
18. Azizi, M., Talatahari, S., Gandomi, A.H.: Fire hawk optimizer: a novel metaheuristic algorithm.
Artif. Intell. Rev. 1–77 (2022)
19. Shishehgarkhaneh, M.B., et al.: BIM-based resource tradeoff in project scheduling using fire
hawk optimizer (FHO). Buildings 12(9), 1472 (2022)
20. Rahmani, A.M., AliAbdi, I.: Plant competition optimization: a novel metaheuristic algorithm.
Expert. Syst. 39(6), e12956 (2022)
21. Mirjalili, S., et al.: Salp swarm algorithm: a bio-inspired optimizer for engineering design
problems. Adv. Eng. Softw. 114, 163–191 (2017)
22. Azizi, M.: Atomic orbital search: a novel metaheuristic algorithm. Appl. Math. Model. 93,
657–683 (2021)
23. Azizi, M., Mohamed, A.W., Shishehgarkhaneh, M.B.: Optimum design of truss structures with
atomic orbital search considering discrete design variables. In: Handbook of Nature-Inspired
Optimization Algorithms: The State of the Art, pp. 189–214. Springer (2022)
Hybrid Social Network Search and Material Generation Algorithm … 69

24. Chou, J.-S., Truong, D.-N.: A novel metaheuristic optimizer inspired by behavior of jellyfish
in ocean. Appl. Math. Comput. 389, 125535 (2021)
25. Gandomi, A.H., Alavi, A.H.: Krill herd: a new bio-inspired optimization algorithm. Commun.
Nonlinear Sci. Numer. Simul. 17(12), 4831–4845 (2012)
26. Hayyolalam, V., Pourhaji Kazem, A.A.: Black widow optimization algorithm: a novel meta-
heuristic approach for solving engineering optimization problems. Eng. Applicat. Artif. Intell.
87, 103249 (2020)
27. Goodarzimehr, V., et al.: Special relativity search: a novel metaheuristic method based on
special relativity physics. Knowl. Based Syst. 257, 109484 (2022)
28. Nouhi, B., et al.: The fusion–fission optimization (FuFiO) algorithm. Sci. Rep. 12(1), 1–44
(2022)
29. Ting, T., et al.: Hybrid metaheuristic algorithms: past, present, and future. Recent Adv. Swarm
Intell. Evolut. Comput. 71–83 (2015)
30. Talatahari, S., Goodarzimehr, V.: A discrete hybrid teaching-learning-based optimization
algorithm for optimization of space trusses. J. Struct. Eng. Geo-Techniq. 10(1), 55–72 (2020)
31. Azizi, M., et al.: Optimum design of fuzzy controller using hybrid ant lion optimizer and Jaya
algorithm. Artif. Intell. Rev. 53(3), 1553–1584 (2020)
32. Kaveh, A., Zolghadr, A.: Truss optimization with natural frequency constraints using a
hybridized CSS-BBBC algorithm with trap recognition capability. Comput. Struct. 102, 14–27
(2012)
33. Mafarja, M.M., Mirjalili, S.: Hybrid whale optimization algorithm with simulated annealing
for feature selection. Neurocomputing 260, 302–312 (2017)
34. Mirjalili, S., Hashim, S.Z.M.: A new hybrid PSOGSA algorithm for function optimization. In:
2010 International Conference on Computer and Information Application, IEEE, 2010
35. Jabbari, M., et al.: A parameter tuned hybrid algorithm for solving flow shop scheduling
problems with parallel assembly stages. Sustain. Operat. Comput. 3, 22–32 (2022)
36. Shi, X.H., et al.: An improved GA and a novel PSO-GA-based hybrid algorithm. Inf. Process.
Lett. 93(5), 255–261 (2005)
37. Kaveh, A., Rahmani, P., Eslamlou, A.D.: An efficient hybrid approach based on Harris Hawks
optimization and imperialist competitive algorithm for structural optimization. Eng. Comput.
38(2), 1555–1583 (2022)
38. Mortazavi, A.: Size and layout optimization of truss structures with dynamic constraints using
the interactive fuzzy search algorithm. Eng. Optim. 53(3), 369–391 (2021)
39. Smith, J., et al.: Creating models of truss structures with optimization. ACM Trans. Graphics,
21 (2002)
40. Tsiptsis, I.N., et al.: Structural optimization employing isogeometric tools in particle swarm
optimizer. J. Build. Eng. 24, 100761 (2019)
41. Brütting, J., et al.: Design of truss structures through reuse. Structures 18, 128–137 (2019)
42. Kazemzadeh Azad, S., Aminbakhsh, S.: High-dimensional optimization of large-scale steel
truss structures using guided stochastic search. Structures, 33, 1439–1456 (2021)
43. Awad, R.: Sizing optimization of truss structures using the political optimizer (PO) algorithm.
Structures 33, 4871–4894 (2021)
44. Li, Y., Wang, S., Han, M.: Truss structure optimization based on improved chicken swarm
optimization algorithm. Adv. Civil Eng. 2019, 6902428 (2019)
45. Pierezan, J., et al.: Chaotic coyote algorithm applied to truss optimization problems. Comput.
Struct. 242, 106353 (2021)
46. Khodadadi, N., Mirjalili, S.: Truss optimization with natural frequency constraints using
generalized normal distribution optimization. Appl. Intell. 52(9), 10384–10397 (2022)
47. Lemonge, A.C.C., et al.: Multi-objective truss structural optimization considering natural
frequencies of vibration and global stability. Expert Syst. Appl. 165, 113777 (2021)
48. Jawad, F.K.J., et al.: Sizing and layout optimization of truss structures with artificial bee colony
algorithm. Structures 30, 546–559 (2021)
49. Liu, J., Xia, Y.: A hybrid intelligent genetic algorithm for truss optimization based on deep
neutral network. Swarm Evol. Comput. 73, 101120 (2022)
70 M. Saraee et al.

50. Yücel, M., Bekdaş, G., Nigdeli, S.M.: Optimization of truss structures with sizing of bars
by using hybrid algorithms. In: Intelligent Computing & Optimization, Springer International
Publishing, Cham, 2022
51. Omidinasab, F., Goodarzimehr, V.: A hybrid particle swarm optimization and genetic algorithm
for truss structures with discrete variables. J. Appl. Comput. Mech. 6(3), 593–604 (2020)
52. Kaveh, A., Talatahar, S.: A hybrid particle swarm and ant colony optimization for design of
truss structures. Asian J. Civil Eng. (BUILDING AND HOUSING) (2008)
53. Xu, Z., Cui, Y., Li, B.: Truss structure optimization design based on FE-PSO-SQP algorithm.
In: New Approaches for Multidimensional Signal Processing. Springer Nature Singapore,
Singapore (2023)
54. Cao, H., et al.: Sizing and shape optimization of truss employing a hybrid constraint-handling
technique and manta ray foraging optimization. Expert Syst. Appl. 213, 118999 (2023)
55. Nguyen, N.T.T., et al. An enhanced hybrid jaya algorithm for size optimization of truss structure
under frequency constraints. In: Advances in Engineering Research and Application. Springer
International Publishing, Cham (2023)
56. Kaveh, A., Rajabi, F.: Optimum structural design of spatial truss structures via migration-based
imperialist competitive algorithm. Scientia Iranica 29(6), 2995–3015 (2022)
57. Dastan, M., et al.: Hybrid teaching–learning-based optimization for solving engineering and
mathematical problems. J. Braz. Soc. Mech. Sci. Eng. 44(9), 431 (2022)
58. Goodarzimehr, V., Omidinasab, F., Taghizadieh, N.: Optimum design of space structures using
hybrid particle swarm optimization and genetic algorithm. World J. Eng. (2022). ahead-of-
print(ahead-of-print)
59. Do, D.T., et al.: A hybrid arithmetic optimization algorithm and differential evolution for
optimization of truss structures subjected to frequency constraints. J. Sci. Technol. Civil Eng.
(STCE)-HUCE 16(2), 22–37 (2022)
60. Dang, K.D., et al.: A single step optimization method for topology, size and shape of trusses
using hybrid differential evolution and symbiotic organisms search. Comput. Struct. 270,
106846 (2022)
61. Nguyen-Van, S., et al.: A novel hybrid differential evolution and symbiotic organisms search
algorithm for size and shape optimization of truss structures under multiple frequency
constraints. Expert Syst. Appl. 184, 115534 (2021)
62. Jafari, M., Salajegheh, E., Salajegheh, J.: Optimal design of truss structures using a hybrid
method based on particle swarm optimizer and cultural algorithm. Structures 32, 391–405
(2021)
63. Altun, M., Pekcan, O.: Optimum sizing of truss structures using a hybrid flower pollinations.
In: Dey, N. (ed.) Applications of Flower Pollination Algorithm and its Variants, pp. 113–137.
Springer Singapore, Singapore (2021)
64. Degertekin, S.O., et al.: Large-scale truss-sizing optimization with enhanced hybrid HS
algorithm. Appl. Sci. 11(7), 3270 (2021)
65. Kaveh, A., Talatahari, S., Khodadadi, N.: Hybrid invasive weed optimization-shuffled frog-
leaping algorithm for optimal design of truss structures. Iranian J. Sci. Technol. Transac. Civil
Eng. 44(2), 405–420 (2020)
66. Talatahari, S., Goodarzimehr, V., Taghizadieh, N.: Hybrid teaching-learning-based optimiza-
tion and harmony search for optimum design of space trusses. J. Optim. Indust. Eng. 13(1),
177–194 (2020)
67. Jafari, M., Salajegheh, E., Salajegheh, J.: An efficient hybrid of elephant herding optimization
and cultural algorithm for optimal design of trusses. Eng. Comput. 35(3), 781–801 (2019)
68. Talatahari, S., Bayzidi, H., Saraee, M.: Social network search for global optimization. IEEE
Access 9, 92815–92863 (2021)
69. Talatahari, S., Azizi, M., Gandomi, A.H.: material generation algorithm: a novel metaheuristic
algorithm for optimization of engineering problems. Processes 9(5), 859 (2021)
70. Wang, D., Zhang, W., Jiang, J.: Truss optimization on shape and sizing with frequency
constraints. AIAA J. 42(3), 622–630 (2004)
Hybrid Social Network Search and Material Generation Algorithm … 71

71. Lingyun, W., et al.: Truss optimization on shape and sizing with frequency constraints based
on genetic algorithm. Comput. Mech. 35, 361–368 (2005)
72. Gomes, H.M.: Truss optimization with dynamic constraints using a particle swarm algorithm.
Expert Syst. Appl. 38(1), 957–968 (2011)
73. Marinaki, M., Marinakis, Y., Stavroulakis, G.E.: Fuzzy control optimized by a multi-objective
differential evolution algorithm for vibration suppression of smart structures. Comput. Struct.
147, 126–137 (2015)
74. Farshchin, M., Camp, C., Maniat, M.: Multi-class teaching–learning-based optimization for
truss design with frequency constraints. Eng. Struct. 106, 355–369 (2016)
75. Goodarzimehr, V., et al.: Structural design with dynamic constraints using weighted chaos
game optimization. J. Comput. Design Eng. 9(6), 2271–2296 (2022)
76. Tejani, G.G., Savsani, V.J., Patel, V.K.: Adaptive symbiotic organisms search (SOS) algorithm
for structural design optimization. J. Comput. Design Eng. 3(3), 226–249 (2016)
77. Lin, J., Che, W., Yu, Y.: Structural optimization on geometrical configuration and element
sizing with statical and dynamical constraints. Comput. Struct. 15(5), 507–515 (1982)
78. Miguel, L.F.F., Miguel, L.F.F.: Shape and size optimization of truss structures considering
dynamic constraints through modern metaheuristic algorithms. Expert Syst. Appl. 39(10),
9458–9467 (2012)
Development of a Hybrid Algorithm
for Optimum Design of a Large-Scale
Truss Structure

Melda Yücel, Gebrail Bekdaş, and Sinan Melih Nigdeli

Abstract In the present study, the optimum design of a structural model as a 72-bar
truss was handled to detect the minimum weight. By this scope, the main objective
function value as the total weight is tried to be minimized through that all bar members
of the truss structure were optimized without grouping to observe section areas by
considering as design parameters. Within this process, two different metaheuristics
including nature-inspired, and swarm-based algorithms known as the flower pollina-
tion algorithm (FPA) and the Jaya algorithm (JA) were combined and hybridized to
enhance the optimization performance of JA. On the other hand, it is observed and
detected that the performances of algorithms show an increment in which iteration
and population numbers. With this respect, it became more suitable and effective that
the optimization process is implemented with the best-selected algorithm in terms
of optimal performance, and successful parameters for the algorithm can be deter-
mined to observe numerous design variables belonging to an extremely large-scale
structural model.

Keywords Metaheuristics · Flower pollination algorithm · Jaya algorithm ·

Hybridization · Truss structures

M. Yücel
Institute of Graduate Studies, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul, Turkey
G. Bekdaş · S. M. Nigdeli (B)
Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]
G. Bekdaş
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 73

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_5
74 M. Yücel et al.

1 Introduction

While structural design problems are tried to overcome, some conditions require to
be controlled, or some assumptions or acceptances must be applied for the designing
of structures. The main cause is that real engineering problems are not linear, and due
to that these always cannot suit a specific line/solution or equation. In other saying
structural models in real life have nonlinearity behavior. In this regard, to make true
the mentioned requirements or expressions can be considered like a task, which
must be realized by structural engineers. Therefore, some issues such as engineers’
experience and knowledge, etc. play an important role to obtain the best conditions
for designs. However, this is not possible to realize to find the most proper solution
with classical hand calculations. Also, computer design programs should be applied
iteratively by engineers. But these have time-consuming stages and cannot be turned
back when any error is made. For this reason, it is always not possible to say that
these are the best solutions or the rightest designs.
In recent times, advanced computer-codding applications or calculation methods
are performed to ensure the models are optimal by preventing these problems. Opti-
mization methodologies are one of these, and the usage of metaheuristic methods
based on nature, physical or chemical processes of materials, human memory, or
genetics came into prominence in this process. Because, these are very easy to use,
effective, understandable, and successful in finding the best solution, because of have
quick stages and error controllability advantages.
From this point of view, civil especially structural engineering is one of the
widespread application areas of metaheuristics. For example, the best cross-section
can be determined, or the optimal steel bar area can be found for a beam or column,
the weight or cost can be minimized for any structural member, the optimum amount
of structural materials can be analyzed, etc.
In this regard, in the mentioned structural engineering area, optimization applica-
tions are realized by utilizing many different metaheuristic algorithms. For instance,
genetic algorithm (GA) [1], artificial bee colony algorithm (ABC) [2], teaching–
learning based optimization (TLBO), harmony search (HS) and bat algorithm (BA)
[3], particle swarm optimization (PSO) combined with HS [4]; Jaya algorithm (JA)
[5]; crow search algorithm (CSA) [6], TLBO and flower pollination algorithm (FPA)
[7], GA and PSO [8], and hybrid approach based on HS [9] were benefited to mini-
mize of cost and/or structural mass via determining of optimum properties belonging
to structural members as beam, column, footing etc. including reinforced concrete,
steel composite materials; ant colony optimization (ACO) algorithm [10], charged
system search algorithm (CSS) [11], biogeography-based optimization algorithm
(BBO) [12], PSO algorithm [13]; more than five metaheuristics [14], black hole
algorithm (BH) [15] are handled about determination of the lowest carbon emis-
sion rate besides minimum cost by optimizing of geometry for retaining walls with
different structural designs; HS [16]; GA [17, 18], FPA [19, 20]; FPA, TLBO and JA
[21], differential evolution (DE) algorithm [22]; CSA, whale optimization algorithm
(WOA), and grey wolf optimization (GWO) [23] are also investigated for optimum
Development of a Hybrid Algorithm for Optimum Design … 75

designing of some devices like tuned mass or liquid dampers, base-isolation systems,
active control mechanisms etc., which are benefited to control and damp of dynamical
effects as wind forces.
On the other respect, in the structural engineering area, there also exist numerous
optimization applications realized for suitable and proper designs, which can provide
the required design targets like not exceeding the allowable compression and tension
stresses to prop the nodal displacements for truss structures at a specific level, etc. In
this regard, various applications have been realized for truss systems, which makes
possible either structural weight and also cost minimization or optimal placement
for node points (layout optimization), besides topology optimization. One of these
is a study conducted by Çerçevik et al. [24] where size and shape optimization was
realized in the way of minimization of weight under multiple natural frequency
constraints by using the firefly algorithm (FA) and HS for different truss structures.
Vu [25] and also Kaveh and Ghazaan [26] benefited from the methods, which are
known as DE and vibrating particles systems (VPS), respectively, for optimizing truss
structures (by providing minimum weight) to improve their dynamic performance
of them under frequency constraints, too. On the other hand, a method was proposed
to find the optimum truss structures with minimum weight by using three variations
of CSA [27]. Furthermore, various applications related to the topology optimization
of trusses were tackled with different metaheuristics. For example, Cui et al. [28]
also carried out an application for topology optimization by benefiting from GA
with the prevention of local buckling constraints besides overlapping of bars that it
is previously accomplished by controlling design constraints such as displacement
and stress. At the same time, Hosseinzadeh et al. [29] examined the combination of
a metaheuristic method called an electromagnetism-like mechanism algorithm with
an approach as migration strategy (EM-MS) to make real optimization for layout
and size belonging to five different truss models under natural dynamic frequency
constraints. PSO was utilized to generate the best truss structure with the aim of sizing
and layout arrangement by optimizing of cross-sectional areas of bars and coordinates
of each node, besides providing minimum weight through conforming to different
design conditions and laws [30]. In addition to these, two different algorithms were
arranged as a hybrid tool by integrating GWO and DE intended for optimization of
fully-stressed truss structure as shape (geometrical) and size besides topology [31].
Within the scope of this study, intended for the lightest 72-bar truss structure,
section areas belonging to each of the bar members are tried to find with an optimiza-
tion approach by developing a hybridized version of nature-inspired and population-
based metaheuristic methods, that was generated with the usage of two kinds as
Jaya algorithm and Flower pollination algorithm. The current problem is based on
an extremely large-scale structural design model and the consideration of this effec-
tive and precise method is required to analyze it. At the same time, it is aimed that
the best/suitable process time is determined with the evaluation of various iteration
numbers and population numbers by applying two different cases for observation of
optimum section area results and minimum weight values, besides collaboration of
optimization methods with each other.
76 M. Yücel et al.

2 Jaya Algorithm and Flower Pollination Algorithm

Jaya algorithm (JA), which is named after a Sanskrit word that meant victory, was
first proposed by Rao in 2016. Also, the working logic of JA is based on glorification,
which has an approach to reaching the best point by going far from the worst point
[32]. However, the Jaya algorithm can search the solutions between only the best and
worst solutions namely points. For this reason, the other solutions sometimes can be
ignored, while the optimistic solutions are trying to find. Although the taking time
for the operation of the optimization process is pretty short, and adjusting specific
parameters is not required similar to other metaheuristics, it cannot be so effective
and successful to find the best solution optimally.
In this meaning, there is one formulation to be used in the optimization process
by JA as below Eq. (5.1):
( | |) ( | |)
X i,new = X i, j + rand X i,gbest − | X i, j | − rand X i,wor st − | X i, j | (5.1)

where rand function provides to generate a random number ranging between 0

and 1. Also, X i,new and X i, j means to the value of the new namely updated and
jth (old) solution among the whole population corresponding to ith design variable,
respectively. Moreover, X i,gbest and X i,wor st are the best and worst solutions’ values
for this variable in terms of the objective function.
The flower pollination algorithm (FPA), which was proposed by Xin-She Yang
in the year of 2012, was created by taking as the basis of an approach depending on
the pollination capability of flowers. In this algorithm, there are several conditions
to be applied for mathematical optimization problems. These consist of the type of
pollination as global or local search, constancy of flower, selection of search by using
switch probability parameter (sp), etc. [33].
While the optimization process is realized, two alternative scenarios can be applied
according to the value of the FPA parameter expressed as sp (∊[0,1]), which deter-
mines the search type through be tuned this value. Here, the first one is known as
global search, which becomes possible between only different flowers of the same
plants by any pollinators including fish, insects, birds, etc., or wind through applying
a random distribution called L évy (Eq. (5.2)). Also, the local search can realize in
own structure of a specific flower. These processes can be expressed via Eq. (5.3) as
mathematical.
( )
1
+ (rand)−1.5 e(− 2rand )
1
Lé vy = √ (5.2)
2π
{ ( )
if sp > rand X i, j + Lé vy( X i,gbest − X)i, j Global search
X i,new = (5.3)
else X i, j + rand X i,m − X i,n Local search

Here, X i,m and X i,n present the two different random solutions (mth and nth) selected
from the population community.
Development of a Hybrid Algorithm for Optimum Design … 77

3 Hybridization of Metaheuristics

In the current study, two different population-based metaheuristic algorithms were

combined by benefiting from some properties and searching behaviors of them.
In this respect, the flower pollination algorithm (FPA) and JA were combined and
transformed into a hybrid method symbolized as JA&FPA.
These metaheuristic optimization algorithms also have some inspirations
including natural life processes and community-based properties. For example, FPA
used the pollination ability of flowers with the help of pollinators like flies, insects,
wind, etc. to simulate optimization processes for solving especially nonlinear real-
life problems. In this scope, to generate an advanced and powerful methodology
for both overcoming this kind of problem and developing the performance with the
success of the JA algorithm, a different approach was proposed by combining the
basic phase of JA and the local search phase of the FPA method. In this meaning, the
details for the working principle of the hybridized algorithm as JA&FPA is presented
as a flowchart in Fig. 1 [34].

Fig. 1 Flowchart for JA&FPA hybrid algorithm

78 M. Yücel et al.

4 Design of Truss Structure

In this book chapter, a structural model as a large-scale truss with 72-bar was inves-
tigated to generate the best design namely the optimal model by minimizing the total
weight. In Fig. 2, the details of the structural model in terms of geometry, and design
parameters can be seen as a plan, and 3-d display, respectively.
In this chapter, two different cases for optimization modeling of truss structure
were realized to determine the minimized structural weight by finding the best
parameters of structure as cross sections of bars. In this regard, the design prop-
erties as parameters and constants are presented in Table 1. On the other hand,
different constraint functions and limitations were handled for designing structures
in the way of providing suitability to design requirements. Two different constraints
were utilized for nodes and bars as controlling displacements, and compressive
together with tensile stresses, respectively (Table 2). For this reason, the multiple
loading conditions are applied on nodes of the truss that are given in Table 3. More-
over, the best parameter values as optimum results for all design properties can be
ensured without exceeding any constraints by applying the penalization of objective
functions.

Fig. 2 Design model and

variables of the 72-bar truss
structure [35]

Table 1 Design details for optimum design of truss structure

Property Symbol Limit/value Unit
Design Cross-section of bars Abar 0.1–3.0 inch2
parameters
Design Elastic modulus Es 107 psi
constants Weight per unit of volume for steel ρs 0.1 lb/inch3
Number of bars – 72 –
Number of nodes – 20 –
Development of a Hybrid Algorithm for Optimum Design … 79

Table 2 The design constraints applied on nodes of truss

Structural member Description Constraints Unit
Nodes Displacement
All Constraints for displacements δ < |∓0.25| inch
arisen on nodes
Bars Compression Tension
stress stress
A1–72 Constraints for stresses arising σc > −25000 σt < +25000 psi
on bars

Table 3 Values of the multiple loading conditions on nodes

Case Node Loading Unit
Px Py Pz
1st 17 5000 5000 −5000 lb/inch2
2nd 17 0 0 −5000
18 0 0 −5000
19 0 0 −5000
20 0 0 −5000

5 Numerical Investigations for 72-Bar Truss

In the current chapter, an optimization process was designed for the modeling of
a truss model with 72-bar. In this respect, the optimization process was applied in
two different cases by adjusting algorithm parameters. Also, all of these processes
were applied with the usage of both the independent algorithm as JA and the hybrid
algorithm as JA&FPA. In this regard, firstly, for the 1st and 2nd cases, the changing
of iteration and population numbers are evaluated in terms of detection of the best
weight, respectively. Here, in the 1st case (Fig. 3, Tables 4 and 5), the range of
iteration numbers is taken in 500–120,000 in the way of applying a total of 34
different alternatives (by using constant population as 20). As to the 2nd case (Fig. 4,
Tables 6 and 7), population numbers are evaluated as 3, 5, and 10–30 by increasing 10
together with 100,000 iteration numbers constantly. These two cases were presented
below sections, respectively.

6 Conclusions

In the current study, a large-scale structural model of a 72-bar truss was tried to
optimize by providing the best weight value as minimized. For this reason, all of the
design parameters as structural bar sections were obtained as optimized by utilizing
80 M. Yücel et al.

440.0000
420.0000
400.0000
Minimum weight (lb)

380.0000
360.0000
340.0000
320.0000
300.0000

Iteration number
JA&FPA JA

Fig. 3 The changing of the minimized weight values along iteration numbers is increasing

an advanced metaheuristic algorithm called JA&FPA. Moreover, two different opti-

mization cases were arranged by benefiting variable values of population and also
iteration numbers towards the classic version of JA and JA&FPA hybrid algorithm,
too.
According to the obtained results for the 1st case as changing of iteration numbers,
the most successful algorithm is JA&FPA in terms of the minimization of total struc-
tural weight (304.9611 lb). Although JA&FPA used fewer iteration steps according
to classical JA, it can be more effective for the minimization of weight.
On the other hand, in the 2nd case, the change in population numbers is evaluated
in terms of obtaining of best weight values. At this time, the most effective algorithm
is JA&FPA like the 1st case. Between the results of minimized weights provided by
JA and JA&FPA, there is an extremely significant difference at 3.15 lb. In this respect,
JA performance can be extremely improved in terms of minimizing the total weight.
However, it can be thought that the most successful case is changing population
numbers. Because the best value for total weight is smaller in comparison to the case
changing of iteration numbers.
Moreover, the convergence performance of the JA&FPA algorithm is more steady
in the minimization of weight for the truss model. Also, JA&FPA can reach the
mentioned best results more quickly than the classic version of JA.
Development of a Hybrid Algorithm for Optimum Design … 81

Table 4 The best optimization results are determined by JA besides evaluations of statistical
parameters
Bar Min. weight Bar Min. weight Bar Min. weight Bar Min. weight
A1 1.7610 A19 1.7005 A37 0.4587 A55 0.2786
A2 0.1001 A20 0.2652 A38 0.2979 A56 0.4252
A3 2.3456 A21 2.1035 A39 0.7212 A57 0.5699
A4 0.1388 A22 0.2594 A40 0.2695 A58 0.3839
A5 0.1000 A23 0.7451 A41 0.1000 A59 0.8746
A6 0.7444 A24 0.1010 A42 0.8071 A60 0.1000
A7 0.1001 A25 0.7679 A43 0.1000 A61 0.6557
A8 0.8090 A26 0.1090 A44 0.7615 A62 0.1000
A9 0.8284 A27 0.1477 A45 0.7403 A63 0.1000
A10 0.1006 A28 0.7604 A46 0.1000 A64 0.6670
A11 0.7184 A29 0.1000 A47 0.7712 A65 0.1002
A12 0.1005 A30 0.7143 A48 0.1000 A66 0.8415
A13 0.1000 A31 0.1000 A49 0.1007 A67 0.1000
A14 0.1000 A32 0.1003 A50 0.1001 A68 0.1000
A15 0.1000 A33 0.1000 A51 0.1000 A69 0.1000
A16 0.1001 A34 0.1000 A52 0.1000 A70 0.1000
A17 0.1000 A35 0.1000 A53 0.1000 A71 0.9089
A18 0.1000 A36 0.1002 A54 0.1025 A72 0.1000
Best weight 305.0709
Mean weight 305.0895
Std. dev. for 0.0088
best weight
Population 20
number
Iteration 100,000
number
82 M. Yücel et al.

Table 5 The best optimization results were determined by the hybrid algorithm as JA&FPA besides
evaluations of statistical parameters
Bar Min. weight Bar Min. weight Bar Min. weight Bar Min. weight
A1 1.7927 A19 1.6370 A37 0.4280 A55 0.1018
A2 0.1000 A20 0.2863 A38 0.3177 A56 0.2905
A3 2.3597 A21 2.1013 A39 0.7421 A57 0.3825
A4 0.1047 A22 0.2934 A40 0.3108 A58 0.5742
A5 0.1002 A23 0.7289 A41 0.1001 A59 0.3722
A6 0.7205 A24 0.1014 A42 0.7805 A60 0.9184
A7 0.1009 A25 0.7713 A43 0.1008 A61 0.1005
A8 0.8056 A26 0.1256 A44 0.7211 A62 0.6607
A9 0.8062 A27 0.1345 A45 0.7443 A63 0.1000
A10 0.1002 A28 0.7719 A46 0.1007 A64 0.1002
A11 0.7177 A29 0.1067 A47 0.7875 A65 0.6386
A12 0.1004 A30 0.7131 A48 0.1003 A66 0.1005
A13 0.1000 A31 0.1005 A49 0.1000 A67 0.8983
A14 0.1005 A32 0.1000 A50 0.1008 A68 0.1019
A15 0.1001 A33 0.1000 A51 0.1000 A69 0.1000
A16 0.1003 A34 0.1001 A52 0.1001 A70 0.1000
A17 0.1002 A35 0.1001 A53 0.1002 A71 0.1029
A18 0.1007 A36 0.1005 A54 0.4280 A72 0.8984
Best weight 304.9611
Mean weight 304.9869
Std. dev. for 0.0138
best weight
Population 20
number
Iteration 65,000
number
Development of a Hybrid Algorithm for Optimum Design … 83

1150.0000
1050.0000
950.0000
850.0000
Minimum weight (lb)

750.0000
650.0000
550.0000
450.0000
350.0000
250.0000
3 5 10 20 30
Population number

JA&FPA JA

Fig. 4 The decreasing of the minimized weight values along population numbers is changing

Table 6 The best optimization results are determined by JA besides evaluations of statistical
parameters
Bar Min. weight Bar Min. weight Bar Min. weight Bar Min. weight
A1 3.000 A19 1.7026 A37 0.4795 A55 0.2547
A2 0.100 A20 0.2542 A38 0.2724 A56 0.4011
A3 2.996 A21 2.1343 A39 0.7603 A57 0.6282
A4 0.1437 A22 0.1632 A40 0.3548 A58 0.4259
A5 0.1001 A23 0.7034 A41 0.1001 A59 0.9128
A6 0.7017 A24 0.1000 A42 0.7733 A60 0.1001
A7 0.1000 A25 0.7822 A43 0.1000 A61 0.6461
A8 0.8203 A26 0.1000 A44 0.7812 A62 0.1000
A9 0.7531 A27 0.1403 A45 0.7428 A63 0.1000
A10 0.1056 A28 0.7297 A46 0.1000 A64 0.5976
A11 0.6775 A29 0.1000 A47 0.7651 A65 0.1000
A12 0.1205 A30 0.7099 A48 0.1000 A66 0.8322
A13 0.1002 A31 0.1000 A49 0.1000 A67 0.1000
A14 0.1000 A32 0.1000 A50 0.1000 A68 0.1000
A15 0.1000 A33 0.1000 A51 0.1000 A69 0.1000
A16 0.1000 A34 0.1000 A52 0.1004 A70 0.1000
A17 0.1000 A35 0.1000 A53 0.1000 A71 0.8671
(continued)
84 M. Yücel et al.

Table 6 (continued)
Bar Min. weight Bar Min. weight Bar Min. weight Bar Min. weight
A18 0.1000 A36 0.1000 A54 0.1000 A72 0.1015
Best weight 308.4800
Mean weight 308.5093
Std. dev. for 0.0141
best weight
Population 10
number
Iteration 100,000
number

Table 7 The best optimization results were determined by the hybrid algorithm as JA&FPA besides
evaluations of statistical parameters
Bar Min. weight Bar Min. weight Bar Min. weight Bar Min. weight
A1 1.8660 A19 1.6899 A37 0.4019 A55 0.2990
A2 0.1025 A20 0.2955 A38 0.3272 A56 0.3654
A3 2.2738 A21 1.9927 A39 0.7335 A57 0.5858
A4 0.1044 A22 0.3047 A40 0.2857 A58 0.3911
A5 0.1047 A23 0.7584 A41 0.1030 A59 0.9106
A6 0.6863 A24 0.1018 A42 0.8213 A60 0.1021
A7 0.1009 A25 0.7334 A43 0.1034 A61 0.6489
A8 0.8113 A26 0.1283 A44 0.7317 A62 0.1001
A9 0.8057 A27 0.1168 A45 0.7177 A63 0.1001
A10 0.1035 A28 0.7707 A46 0.1075 A64 0.6523
A11 0.7305 A29 0.1000 A47 0.8072 A65 0.1025
A12 0.1039 A30 0.7399 A48 0.1036 A66 0.9197
A13 0.1002 A31 0.1009 A49 0.1006 A67 0.1055
A14 0.1004 A32 0.1000 A50 0.1000 A68 0.1008
A15 0.1005 A33 0.1000 A51 0.1000 A69 0.1001
A16 0.1021 A34 0.1002 A52 0.1030 A70 0.1003
A17 0.1000 A35 0.1009 A53 0.1003 A71 0.8822
A18 0.1005 A36 0.1016 A54 0.1000 A72 0.1037
Best weight 305.3269
Mean weight 305.4213
Std. dev. for 0.0406
best weight
Population 30
number
Iteration 100,000
number
Development of a Hybrid Algorithm for Optimum Design … 85

References

1. Senouci, A.B., Al-Ansari, M.S.: Cost optimization of composite beams using genetic
algorithms. Adv. Eng. Softw. 40(11), 1112–1118 (2009)
2. Ozturk, H.T., Durmus, A.: Optimum cost design of RC columns using artificial bee colony
algorithm. Struct. Eng. Mech. 45(5), 643–654 (2013)
3. Bekdaş, G., Nigdeli, S. M.: Optimum design of reinforced concrete beams using teaching-
learning-based optimization. In: 3rd International Conference on Optimization Techniques in
Engineering (OTENG’15), November 2015, 7–9 (2015)
4. Talaei, A.S., Nasrollahi, A., Ghayekhloo, M.: An automated approach for optimal design of
prestressed concrete slabs using PSOHS. KSCE J. Civ. Eng. 21(3), 782–791 (2017)
5. Kayabekir, A.E., Bekdas, G., Nigdeli, S.M., Temür, R.: Investigation of cross-sectional dimen-
sion on optimum carbon fiber reinforced polymer design for shear capacity increase of
reinforced concrete beams. Int. J. Theor. Appl. Mech. 3 (2018)
6. Ozbasaran, H.: Optimal design of I-section beam-columns with stress, non-linear deflection
and stability constraints. Eng. Struct. 171, 385–394 (2018)
7. Yucel, M., Nigdeli, S.M., Bekdaş, G.: Artificial neural networks (ANNs) and solution of
civil engineering problems: ANNs and prediction applications. In: Artificial Intelligence and
Machine Learning Applications in Civil, Mechanical, and Industrial Engineering, pp. 13–37.
IGI Global, USA (2020)
8. Chaudhuri, P., Maity, D.: Cost optimization of rectangular RC footing using GA and UPSO.
Soft. Comput. 24(2), 709–721 (2020)
9. Yücel, M., Nigdeli, S.M., Bekdaş, G.: Generation of sustainable models with multi-objective
optimum design of reinforced concrete (RC) structures. Structures 40, 223–236 (2022)
10. Ghazavi, M., Bonab, S.B.: Optimization of reinforced concrete retaining walls using ant colony
method. In: Norbert, V., Bernd, S., Daniel, S., Gerhardt B. (eds.) Geotechnical Safety and Risk.
ISGSR 2011. Karlsruhe: Bundesanstalt für Wasserbau, pp. 297–306 (2011)
11. Kaveh, A., Behnam, A.F.: Charged system search algorithm for the optimum cost design of
reinforced concrete cantilever retaining walls. Arab. J. Sci. Eng. 38(3), 563–570 (2013)
12. Aydogdu, I.: Cost optimization of reinforced concrete cantilever retaining walls under seismic
loading using a biogeography-based optimization algorithm with Levy flights. Eng. Optim.
49(3), 381–400 (2017)
13. Moayyeri, N., Gharehbaghi, S., Plevris, V.: Cost-based optimum design of reinforced concrete
retaining walls considering different methods of bearing capacity computation. Mathematics
7(12), 1232 (2019)
14. Yepes, V., Martí, J.V., García, J.: Black hole algorithm for sustainable design of counterfort
retaining walls. Sustainability 12(7), 2767 (2020)
15. Yücel, M., Kayabekir, A.E., Bekdaş, G., Nigdeli, S.M., Kim, S., Geem, Z.W.: Adaptive-hybrid
harmony search algorithm for multi-constrained optimum eco-design of reinforced concrete
retaining walls. Sustainability 13(4), 1639 (2021)
16. Park, W.: Optimal design of a hybrid structural control system using a self-adaptive harmony
search algorithm. J. Comput. Struct. Eng. Inst. Korea 31(6), 301–308 (2018)
17. Liu, Y., Wu, J., Donà, M.: Effectiveness of fluid-viscous dampers for improved seismic
performance of inter-storey isolated buildings. Eng. Struct. 169, 276–292 (2018)
18. Pal, S., Hassan, A., Singh, D.: Optimization of base isolation parameters using genetic
algorithm. J. Stat. Manag. Syst. 22(7), 1207–1222 (2019)
19. Yucel, M., Bekdaş, G., Nigdeli, S.M., Sevgen, S.: Estimation of optimum tuned mass damper
parameters via machine learning. J. Build. Eng. 26, 100847 (2019)
20. Yücel, M., Bekdaş, G., Nigdeli, S.M.: Metaheuristics-based optimization of TMD parameters
in time history domain. In: Optimization of Tuned Mass Dampers, pp. 55–66. Springer, Cham
(2022)
21. Ulusoy, S., Bekdas, G., Nigdeli, S.M.: Active structural control via metaheuristic algorithms
considering soil-structure interaction. Struct. Eng. Mech. 75(2), 175–191 (2020)
86 M. Yücel et al.

22. Deastra, P., Wagg, D., Sims, N.: Optimum design of a tuned-inerter-hysteretic-damper (TIHD)
for building structures subject to earthquake base excitations. In: EURODYN 2020: Proceed-
ings of the XI International Conference on Structural Dynamics. September 2020. European
Association for Structural Dynamics (EASD), pp. 1501–1509 (2020)
23. Çerçevik, A.E., Avşar, Ö., Hasançebi, O.: Optimum design of seismic isolation systems using
metaheuristic search methods. Soil Dyn. Earthq. Eng. 131, 106012 (2020)
24. Miguel, L.F.F., Miguel, L.F.F.: Shape and size optimization of truss structures considering
dynamic constraints through modern metaheuristic algorithms. Expert Syst. Appl. 39(10),
9458–9467 (2012)
25. Vu, V.T.: Weight minimization of trusses with natural frequency constraints. In: 11th World
Congress on Structural and Multidisciplinary Optimisation. 7–12 June 2015, Sydney, Australia
(2015)
26. Kaveh, A., Ghazaan, M.I.: Vibrating particles system algorithm for truss optimization with
multiple natural frequency constraints. Acta Mech. 228, 307–322 (2017)
27. Javidi, A., Salajegheh, E., Salajegheh, J.: Enhanced crow search algorithm for optimum design
of structures. Appl. Soft Comput. 77, 274–289 (2019)
28. Cui, H., An, H., Huang, H.: Truss topology optimization considering local buckling constraints
and restrictions on intersection and overlap of bar members. Struct. Multidiscip. Optim. 58(2),
575–594 (2018)
29. Hosseinzadeh, Y., Taghizadieh, N., Jalili, S.: Hybridizing electromagnetism-like mechanism
algorithm with migration strategy for layout and size optimization of truss structures with
frequency constraints. Neural Comput. Appl. 27(4), 953–971 (2016)
30. Mortazavi, A., Toğan, V., Nuhoğlu, A.: Weight minimization of truss structures with sizing
and layout variables using integrated particle swarm optimizer. J. Civ. Eng. Manag. 23(8),
985–1001 (2017)
31. Panagant, N., Bureerat, S.: Truss topology, shape and sizing optimization by fully stressed
design based on hybrid grey wolf optimization and adaptive differential evolution. Eng. Optim.
50(10), 1645–1661 (2018)
32. Rao, R.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7(1), 19–34 (2016)
33. Yang, X.S.: Flower pollination algorithm for global optimization. In: International Confer-
ence on Unconventional Computing and Natural Computation. Springer, Berlin, Heidelberg,
pp. 240–249 (2012)
34. Yücel, M., Bekdaş, G., Nigdeli, S.M.: Optimization of truss structures with sizing of
bars by using hybrid algorithms. In: International Conference on Intelligent Computing &
Optimization. Springer, Cham, pp. 592–601 (2021, December)
35. Bekdaş, G., Yucel, M., Nigdeli, S.M.: Evaluation of metaheuristic-based methods for optimiza-
tion of truss structures via various algorithms and Lèvy flight modification. Buildings 11(2),
49 (2021)
Structural Control Systems and Tuned
Mass Damper Optimization by Using
Jaya and Hybrid Algorithms

Muhammed Çoşut, Sinan Melih Nigdeli, and Gebrail Bekdaş

Abstract In this section, as well as explaining the building control systems and their
properties, the displacement and acceleration values were found in the case that TMD
is not used and TMD is placed, according to the earthquake record that affects the
building most negatively, by affecting past earthquake records on the single degree
of freedom system. By using the Matlab & Simulink programs, the displacement
optimization was performed according to the time history domain using the Hybrid
(TLBO-Jaya) and Jaya algorithms. In this optimization process, firstly, the TMD mass
was determined as 5% of the mass of the structure, after that. the constant values,
constraint values and the necessary information for the algorithm were entered to
solve the system. Optimal intervals were determined by assigning the TMD period
and damping ratio as variables. In the case of optimization of the single degree
of freedom system, it is seen that there are great differences in displacement and
acceleration in the system. Furthermore, while for the same system operated using
different algorithms, the values of the Hybrid algorithm and the Jaya algorithm for
variables under loadings were compared for system.

Keywords Structural control system · Passive tuned mass damper · Metaheuristic

algorithm · Optimization · Jaya algorithm · Hybrid algorithm

1 Introduction

Circumstances such as building design systems, different types of structures, material

development and techniques [1], which continue to develop from the past to the
present, reveal that there are significant increases in the performance of the system
to be designed. Such developments are constantly renewing themselves and will

M. Çoşut · S. M. Nigdeli (B) · G. Bekdaş

Istanbul University - Cerrahpaşa, 34320, Avcılar İstanbul, Turkey
e-mail: [email protected]
G. Bekdaş
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 87

G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_6
88 M. Çoşut et al.

contribute to the sustainability of the system [2] by ensuring that the system is
designed safely [3] and continues its life, even in cases such as the most undesirable
ground conditions and the seismic characteristics of the location. Although buildings
built or planned to be built (there are certain limitations in their designs according to
their intended use) are designed under the building design standards of the country in
which they are located, it is seen that the effectiveness of the system and the behaviour
of the system against external loads affecting the building will change if some building
control systems are used. Building control systems may differ according to the type
of building, its design, or the protection of the system from which loads. In other
words, while rubber-based isolation is used in some structures, active, passive or
mixed control systems are used in some structures.
After the most suitable system is selected by considering the costs of this usage
diversity, the compatibility between the structure and the control system used is
examined in detail by using analysis programs, and its suitability is checked. In
some cases, it can be used by choosing special structural materials due to the location
characteristics of the structure rather than the cost. The implementation of selections
in this way is carried out to ensure that the structures remain on the safe side under
dynamic loads such as earthquakes or winds. Accordingly, earthquake and wind load
risk reduction will be realized. Bekdaş and Niğdeli [4] found the optimum parameter
values of the TMD system under seismic loads using Harmony Search. Farshidianfar
and Soheili [5] analysed the TMD calculations of high-rise buildings according to
the Tabas and Kobe earthquakes using Ant Colony Optimization. Araz and Kahya
[6] performed the displacement optimization of serial-tuned mass dampers using the
Simulated Annealing method. Niğdeli and Bekdaş (2014) [7] made mass damper
optimization to prevent the collision of adjacent buildings under earthquake effects.
In this chapter, the types of building control systems that can be used in different
building types are explained, and in addition, in case of using TMD in a single degree
of freedom system, the displacement optimization is carried out by applying the
Hybrid [8] and Jaya [9] algorithms. In this optimization process, FEMA P-695 [10]
far-fault data, which has 44 different earthquake records affecting the system, were
used and the earthquake record with the most adverse conditions for the structure
was selected and its design was provided. While creating the algorithms, all the
necessary data for the analysis of the system were written into the program and the
displacement optimization was done according to the time domain.

2 Vibration

The oscillation of a certain speed and magnitude under the force of a system is called
vibration. While vibrations can occur in many different ways on the structure, it is
expected that the structure will dampen by reacting to these vibrations. Structural
vibrations are divided into some classes. There are different types of these classifi-
cations: undamped free vibration motion, damped free vibration motion, undamped
forced vibration motion and damped forced vibration motion. While free vibration
Structural Control Systems and Tuned Mass Damper Optimization … 89

is created by giving the initial motion to the structure, forced vibration is created by
applying varying forces.
In the case of free vibration movement of the undamped system, it is a type
of movement that does not have any external forcing and damping feature on the
structure, and the vibration movement continues from the moment it starts.
The free vibration motion of the damped system is the damped motion without any
external force on the structure. In any loading affecting the structure, the structure
will absorb the energy after vibrating for a certain period.
In the forced vibration motion of the undamped system, there is no damping when
there is an external force.
In the forced movement of the damped system, there is external stress as well as
damping.
Transient vibrations are called the highest dynamic loadings that occur due to
impact and explosion effects. However, stable vibrations are studied in general at
periodic loadings.

3 Structure Control Systems

The natural damping values of traditional structures are very low; therefore, in the
case of using mechanical materials, the damping rate value of the system is increased,
allowing it to perform better against seismic loads [11, 12]. Although the struc-
tures are designed considering sufficient strength, sufficient rigidity, and sufficient
ductility, sudden loads such as earthquakes and strong winds are possible structural
risks and extra precautions should be taken [13, 14]. While possible wind formations
can be predicted in some regions, such winds have started to be seen frequently even
in regions where no strong winds are expected with climate change.
The damping systems, which started to be used in the field of the mechanical
industry in the 1950s, started to be applied to high-rise buildings in 1969. Before
being used in high-rise buildings, it was created in low-rise buildings and adapted to
high-rise buildings. Here are some of the processes and work involved in making it
more harmonized:
• Madsen et al. [15] had a solution which is dampers in shear walls to transform
the control system from to low-rise building to a high-rise building
• Integration of dampers in the outrigger system is proposed [16–19].
• Kidokoro [20] suggests that the self-mass damper is inspired by the pendulum
movement and clock.
Figure 1 delineates the structure reaction diagram when the control system is
applied and not applied.
In the case of using building control systems in buildings that may be subjected to
any loading, it ensures that the relative floor displacement of the structure is reduced,
and thus, the structure is designed to remain on the safe side without damage under
the impact of seismic forces. In addition, in the case of using seismic isolation;
90 M. Çoşut et al.

Control
System
Seismic Loads Structure

Structure Reaction

Fig. 1 Structure reaction diagram

No Structural Control System With Structural Control System Seismic Izolation System

Fig. 2 Structure motions

the interaction between the structure and the ground decreases, the relative floor
displacements decrease, and at the same time, with the decrease of the structure-
ground interaction, a decrease in the acceleration value felt in the building is observed.
The damper type and properties should be designed according to the displacement
value of the structure as a result of the calculations and by finding certain distances
between adjacent buildings to prevent collisions.
Figure 2 shows three different buildings which have various properties such as no
control system, with the control system as well as with seismic isolation. It can be
observed when a structural control system is utilized, displacement is less than the
building which no structural control system.

3.1 Passive Control System

The passive control system is placed at the base, inside or on the structures. By
extending the construction period thanks to viscoelastic and friction tools [21], it
increases the energy absorption capacity of the structure against earthquakes or
different external loads, while ensuring the preservation of the equilibrium state
of the structure, preventing regional stresses in the sections and ensuring the safety
of the structure. Minimizing the damages that will occur in the structure depends on
Structural Control Systems and Tuned Mass Damper Optimization … 91

the mechanical property of the material to be used and will ensure that the vibration
levels decrease proportionally [22].
There are several different types of passive energy dampers and specifications;
bending metal dampers, friction-type dampers, tuned mass dampers and tuned liquid
dampers, viscous and viscoelastic dampers. Each of them has different properties and
performs energy distribution according to the property that it is good. To exemplify,
Taipei 101 and CN Tower TV Antenna building can be seen in Fig. 3.
Tuned Mass Damper (TMD)
TMD, the vibration control device [23] found by Frahm in 1909, ensures that the
vibrations remain at certain levels in the system in which it is used. As a result
of the studies, Den Hartog presented the formulas which are Eqs. (1) (frequency
ratio, f opt respect to mass ratio, μ) and (2) (damping ratio, ξ d ) for dampers in 1928.
Warburton defined the frequency and damping ratio in Eqs. (3) and (4) in 1982. Sadek
et al. [24] developed these equations and used them in 1997. With the development
of the system over time, it has started to be applied more efficiently in building
systems. This system is a structural control system created as a result of connecting
the spring and damping element, which are in harmony with each other, to a mass.
Since TMD generally performs well against wind and earthquake loads, it will be
quite appropriate to use it in structures exposed to such loads.
The working principle of this structural element is that the structure will make
certain displacements in a certain direction under seismic loads. Vibrating at the
same frequency as the structure, the TMD moves in the opposite direction of the
movement of the structure, compressing the spring element and the energy acting on

(a) (b)

Fig. 3 a Taipei 101, b CN Tower TV Antenna

92 M. Çoşut et al.

the system together with the damping element is damped to a certain extent. In this
way, it has been observed that there is a significant reduction in the displacement of
the structure.
1
f opt = (1)
1+μ
√
3μ
ξd = (2)
8(1 + μ)
( )
1 − μ2
f opt = (3)
1+μ
√
μ(1 − μ4 )
ξd = (4)
4(1 + μ)(1 − μ2 )

While Fig. 4 illustrates how the TMD system is applied, Table 1 shows the build-
ings which are applied TMD. There are several buildings that TMD is utilized to
design and built. To stay safe, designers prefer to use this mass damper system in
their structure.
Tuned Liquid Damper (TLD)
The TLD working principle is similar to the TMD working principle. TLD is created
by putting a certain amount of liquid (usually water) into the tank connected to the

Fig. 4 TMD application kd

md
cd

Storey: n

Storey: 2

Storey: 1

GROUND
Structural Control Systems and Tuned Mass Damper Optimization … 93

Table 1 TMD applications

Structure Height-Length Location Year
(m)
CN Tower TV 553 Toronto, Canada 1973
Antenna
John Hancock 240 Massachusetts, 1976
Tower USA
Chifley Tower 209 Sydney, Australia 1993
Al Taweelah 70 Abu Dabi, UAE 1993
Hotel Burj 321 Dubai, UAE 1997
Al-Arab
Petronas Twin 452 Kuala Lumpur, 1998
Towers Malaysia
Akashi Kaikyo 3911 Honshu-Shikoku, 1998
Bridge Japan
Park Tower 257 Chicago,ABD 2000
Taipei 101 508.2 Taipei, Taiwan 2003
Almas Tower 360 Dubai, UAE 2008
Tokyo Skytree 634 Tokyo, Japan 2012
23 Marine 392 Dubai, UAE 2012
Shanghai 632 Shanghai, China 2014
Tower
432 Park 426 New York, USA 2015
Avenue
Ping an 599 Shenzhen, China 2017
Finance Center

system. Although the movement of the TMD mass increases the damping ratio of
the structure, a TLD absorbs structural energy through the viscous actions of liquid
and wave break.
Friction Type Dampers
Friction, which has a significant mechanism to dissipate energy, has been used in
automotive brake systems to distribute kinetic energy for a long time. The develop-
ment of this system which is used in construction contributes to transforming from
kinetic energy to heat energy; therefore, the building will absorb energy under the
effects of earthquake or wind excitation. This damping system is used with a cross-
link arrangement of metals and steel alloys. Friction damper application can be used
as Fig. 5.
Yielding Metal Dampers
In energy dissipation, the use of inelastic behaviour is provided by utilizing the
hysterical behavior of metals. They are designed to endure the bending, shearing and
axial forces resulting from floor offsets. Metal dampers are generally designed as
94 M. Çoşut et al.

Fig. 5 Friction damper

application

rectangular, triangular, or X-shaped, and importance is given to the equal distribution

of stresses in the material [25]. Working in the system for a long time, and being
resistant to heat differences and environmental conditions makes its use widespread.
Some metal dampers are shown in Fig. 6.
Viscoelastic Dampers
When viscoelastic dampers are utilized in the structure, they contribute to not only
increasing the damping ratio but also rise the lateral rigidity. It transforms the energy
provided in shear stresses caused by plate movements into damping energy. The

Fig. 6 Metal damper

applications
Structural Control Systems and Tuned Mass Damper Optimization … 95

situation caused by the movements between the metal plates used in the system works
with the stretching principle of the elastomers. Its first application was used to prevent
vibrations in aircraft. After that, the system structure was developed and started to
be used in lots of engineering fields, and it is widely used in civil engineering. The
Twin Towers, which were first used in civil engineering, were completed in 1969
[26].
Viscous Dampers
Situations that might damage the structure, such as stresses that may occur under
the influence of earthquake and wind loads, and stresses at the junction points, are
reduced by viscous movement. In the viscous damper, the fluid contained in the
piston chamber is converted into heat energy with the movement and displacement
of the liquid and its absorption is made. As a fluid material; silicone, lead or oil are
used.

3.2 Active Control System

Active control systems contribute to reduction quantities such as structure deflec-

tion, forces, acceleration as well as velocities [27]. They are designed with control
computers separately from TMD. It makes calculations using algorithms according to
the force magnitudes affecting the structure and determines the movement, direction
and intensity of the control system. The change of these properties is done through
the properties of the system, namely the stiffness change and the mass damper. Some
of the most important reasons for using active control systems are:
• It ensures that the vibrations remain at a certain level according to the load that
will affect the structure.
• It takes up less space compared to passive mass dampers.
• While passive control systems are calculated and designed according to the pre-
experienced load, active control systems determine the behaviour of the system
according to the magnitude of the force when the force acts on the structure.
• Preservation of valuable materials inside the building
Some problems encountered when using these systems are:
• Late detection of seismic loads,
• The energy source to which the control system is connected must provide
continuous energy,
• The cost is too high.
• The structures to be applied are not linear
• Some uncertainties in the structure parameters
Table 2 gives information about the buildings which are used for ATMD.
96 M. Çoşut et al.

Table 2 The ATMD’s applications

Structure Height (m) Location Year
Ando Nishikicho 68 Tokyo, 1993
Japan
C Office Tower 130 Tokyo, 1993
Japan
Kansai International 86 Osaka, 1993
Airport Japan
Rokko Island Procter 117 Kobe, 1993
and Gamble Japan
Incheon International 100.4 Incheon, 2001
Airport Control Tower Kore
Shanghai World 492 Shanghai, 2008
Financial Centre China
Canton Tower 600 Guangzhou, 2010
(Guangzhou TV Tower) China

The active control system consists of active tuned mass damper, active beam
control, and active rigidity control. In addition, the operating principle of the active
control system is shown in Fig. 7.
Active Tuned Mass Damper (ATMD)
The ATMD system developed towards the end of the twentieth century began to be
applied in building systems. It is one of the control systems developed against the
seismic force effects of high-rise buildings. The fact that it can be adjusted through the
actuator and sensor used in the system enables this system to work more efficiently
than other systems. In addition, it performs effectively in most mode situations due
to its wide frequency range. Figure 8 shows the ATMD application.

Calculation of
Sensors Loads

Load Operators

Control Force

Seismic Loads Structure

Structure Reaction

Fig. 7 Active control system principle of work

Structural Control Systems and Tuned Mass Damper Optimization … 97

Fig. 8 ATMD application kd

cd md
actuator
Sensor

Storey: n

Storey: 2

Storey: 1

GROUND

Active Tendon Control

The system, which consists of prestressed cables and actuators, sends signals to the
control mechanism through sensors placed in different parts of the structure. As a
result of the forces acting on the structure, the sensors in this control mechanism will
apply the necessary force to reduce the loading after calculating using algorithms
according to the signal status of the loading. In this way, the self-adjustment of the
system according to the magnitude of the forces acting on the structure plays an
important role in preventing the structures from being damaged or collapsing under
any load. The active beam control system shows in Fig. 9.

Fig. 9 Active beam control X(t)

system

Cable

Actuator

X(t) u(t)
98 M. Çoşut et al.

Fig. 10 Active rigidity control application

Active Rigidity Control

It creates a reaction against the incoming loads by changing the properties between
the beam and the conjunction. This response is determined and created as a result
of the calculation by the computer by taking the signal of the forces acting on the
structure. This system, which is used in high-rise buildings, consists of cylinders
and valves and reacts against the seismic loads. The opening status of the valve may
differ according to the floors. In other words, while the valve is open on one floor, the
valve can be closed on other floors. The fact that the valves are closed or open in this
way ensures the formation of rigidity differences on the floors. Figure 10 describes
the active rigidity control system.

3.3 Semi-Active Control System

The semi-active control system was recommended in about the 1920s. Active control
systems require a lot of energy to function properly and to prevent this [28], the semi-
active control system can be used. The working principle of these systems, which
can be supported by batteries and are not affected by power cuts in the building,
is to provide a time-varying control force by changing the characteristics of the
damping system. Examples of these systems are magnetorheological dampers. A
magnetorheological damper is a system filled with magnetorheological fluid that
is controlled by a magnetic field, usually using an electromagnet. This allows the
damping properties of the shock absorber to be constantly controlled by varying the
strength of the electromagnet. As the density of the electromagnet increases, the
viscosity of the fluid in the damper increases.
Structural Control Systems and Tuned Mass Damper Optimization … 99

3.4 Hybrid Control System

Hybrid control systems are created by using Passive control and Active control
systems together. Increasing the performance of the system under loads affecting
the structure [29] and the desire to reduce the costs of the system used is the most
important reasons for the creation of this system. It is created by adding a sensor,
actuator and controller to the passive control system. With the examinations, the
fact that the active control tools can be adjusted according to the load affecting the
structure and that the passive control systems are safer have been beneficial in the
effective use of mixed control systems. This has 2 different types of systems which
are a Hybrid mass damper (HMD) as well a Hybrid seismic isolation (HSI). Figure 11
illustrates the HMD in which the TMD connects the structure, after that the ATMD
links to the TMD.

4 Metaheuristic Algorithms

Algorithms developed by being inspired by many different events in nature [30] are
called metaheuristic algorithms. Teaching–learning based optimization (TLBO) is
inspired by the communication of teachers and students, Flower pollination opti-
mization (FPA) is inspired by local and global pollination of flowers, Ant colony
optimization (ACO) is inspired by the movements of ants and the fluid they secrete
in reaching the target, Harmony search (HS) is inspired by the harmonious combi-
nation of notes to create an effective and beautiful piece of music, while Shark smell
optimization (SSO) is inspired by sharks using their advanced sense of smell to
find their prey. These algorithms have been created as a result of formulating many
different situations by observing them and are used extensively in fields such as

Fig. 11 HMD system

Actuator
Spring
ATMD

TMD
Damping

Structure
100 M. Çoşut et al.

engineering, economics, software, and computers in general. When these algorithms

are used in optimization problems, they can reach the objective function in different
iteration numbers and in different times depending on the number of variables of the
problem and the size of the data to be examined. In addition, some algorithms reach
the objective function in a shorter time due to the formulas they contain, while some
algorithms reach the objective function in a shorter iteration and time than other
algorithms due to the 2-phase comparison system.

4.1 Jaya Algorithm

Jaya algorithm which is developed by Rao in 2016 is one of the most frequently
used algorithms for engineering problems today. One of the most important reasons
why it is preferred so much is that it does not need control parameters [30], while
the second most important feature is that it can reach the objective function in a
short time by determining the best and worst values of the objective function in the
iteration stage and using it to calculate the variables in the problem. The equation of
Jaya is illustrated in Eq. (5). There are necessary to find several values to calculate
the variable’s new value. X i, j is the value of the candidate solution, X i,gbest is the best
value of an objective function, X i,gwor st is the worst value of the objective function; in
addition, all values which are used to find new variable is in the initial matrix. Also,
rand () is assigned randomly.
( | |) ( | |)
X i,new = X i, j + rand() X i,gbest − | X i, j | − rand() X i,gwor st − | X i, j |
'
(5)

4.2 Hybrid Algorithm

Hybrid algorithms [31] are created when the developed metaheuristic algorithms
are combined with each other in 2 or more phases. It can be created using many
different algorithms, and in this way, much more efficient results can be obtained
in the optimization of problems compared to other algorithms. These results show
that better results can be obtained if the difference between the number of itera-
tions reached by the objective function and the mean standard deviation values are
compared, rather than the difference in the objective function value. Some hybrid
algorithms are TLBO-Jaya, TLBO-FPA (Teaching–Learning Based Optimization-
Flower Pollination Algorithm), ACO-GA (Ant Colony Optimization-Genetic Algo-
rithm), GAAP (Hybrid Ant Colony-Genetic Algorithm) as well as DE-BBO (Hybrid
Differential Evolution- Biography-Based Optimization).
There are several Hybrid Algorithm, and a lot of metaheuristic algorithms can be
combined to generate Hybrid Algorithms. One of them is TLBO-Jaya. TLBO has 2
Structural Control Systems and Tuned Mass Damper Optimization … 101

phases, whereas Jaya has 1 phase. That is why, the equation of Jaya is used in TLBO
process by replacing the learning phase.
Teaching phase equations are shown Eqs. (6) and (7). X i,or talama is the mean value
for the objective function.
( | |)
X i,new = X i, j + rand() X i,gbest − | X i, j | − (T F)X i,or talama (6)

T F = r ound(1 + rand()) (7)

Equation (8) is known as the learning phase in TLBO while Eq. (9) is used in
the Jaya algorithm process. In Eq. (8), a and b are assigned randomly. In order to
generate Hybrid Algorithm, the learning phase might be removed by Jaya’s equation,
which is shown in the equations as an arrow.
{ ( )
AFa < AFb , X i, j + r ()( X i,a − X i,b )
X i,new = (8)
AFa > AFb , X i, j + r () X i,b − X i,a
( | |) ( | |)
+ rand() X i,gbest − | X i, j | − r () X i,gwor st − | X i, j |
'
X i,new = X i, j (9)

5 Numerical Example

By using a Tuned Mass Damper in a single-degree-of-freedom system, the effect on

the structure and the changes in systemic movements were examined. Displacement
differences and changes in the acceleration of the force acting on the structure were
compared without using the control system and in the case of using the control
system. In addition, the necessary codes for this system have been written separately
for both the Jaya algorithm and Hybrid algorithms.
The hybrid algorithm was created as a result of the combination of TLBO and
Jaya algorithm and consists of 2 phases. It uses Jaya algorithm and learner phase of
TLBO.
Figures 12 and 13 show the single degree of freedom’s (SDOF) TMD application.
Table 3 displays structure features which are mass, damping coefficient and
rigidity.
Table 4 illustrates the constraint and intervals which can change by changing
structure features like structure period. According to the given structure properties,
the structure period is found as 0.288 s.
Earthquakes that are from FEMA P695 [10] are used for the given structure. Table
5 gives information about FEMA far-fault earthquake record, which provides us to
control the name of the earthquake as well as their motions. Looking at this table
in more detail, San Fernando earthquake, which was in 1971, is the earliest one.
102 M. Çoşut et al.

Fig. 12 TMD application

for SDOF: system-1
u+ud
kd
md
u cd
m

c
k k
2 2

P(t) u

Fig. 13 TMD application

for SDOF: system-2
P(t)

k kd

m md

c cd

u u+ ud

Table 3 Structure features

Symbol Value Unit
m 2924 Kg
c 1581 Ns/m
k 1,390,000 N/m

Table 4 TMD design

Explanation Values and intervals
properties and constraint
TMD mass (Structure mass) * 5%
TMD Max. Damping Ratio 0.5
TMD Min. Damping Ratio 0.01
Max. TMD period 1.5 x (Structure Period)
Min. TMD period 0.5 x (Structure Period)
max(|xd−x N |)T M Dwith
Stroke capacity constraint g(X ) = |x N |T M Dwithout ≤
st_ max
Structural Control Systems and Tuned Mass Damper Optimization … 103

All these earthquakes occurred between 1971 and 1999. In addition, they have two
components, acceleration value as well velocity value.
Furthermore, the maximum displacement for all earthquakes was recorded, and
they are shown in Table 6. The earthquake which is DUZCE/BOL000 in the table
was chosen to design the control system since the maximum displacement occurs.
Also, Table 6 shows the displacement of 100 iterations results for Hybrid and Jaya

Table 5 FEMA earthquake record

Earthquake Year Record information Motions
No Earthquake Component-1 Component-2 PGAmax (g) PGVmax (cm/
Location s)
1 1994 Northridge NORTHR/ NORTHR/ 0.52 63
MUL009 MUL279
2 1994 Northridge NORTHR/ NORTHR/ 0.48 45
LOS000 LOS270
3 1999 Duzce, DUZCE/ DUZCE/ 0.82 62
Turkey BOL000 BOL090
4 1999 Hector HECTOR/ HECTOR/ 0.34 42
Mine HEC000 HEC090
5 1979 Imperial IMPVALL/ IMPVALL/ 0.35 33
Valley H-DLT262 H-DLT352
6 1979 Imperial IMPVALL/ IMPVALL/ 0.38 42
Valley H-E11140 H-E11230
7 1995 Kobe, Japan KOBE/ KOBE/ 0.51 37
NIS000 NIS090
8 1995 Kobe, Japan KOBE/ KOBE/ 0.24 38
SHI000 SHI090
9 1999 Kocaeli, KOCAELI/ KOCAELI/ 0.36 59
Turkey DZC180 DZC270
10 1999 Kocaeli, KOCAELI/ KOCAELI/ 0.22 40
Turkey ARC000 ARC090
11 1992 Landers LANDERS/ LANDERS/ 0.24 52
YER270 YER360
12 1992 Landers LANDERS/ LANDERS/ 0.42 42
CLW-LN CLW-TR
13 1989 Loma Prieta LOMAP/ LOMAP/ 0.53 35
CAP000 CAP090
14 1989 Loma Prieta LOMAP/ LOMAP/ 0.56 45
G03000 G03090
15 1990 Manjil, Iran MANJIL/ MANJIL/ 0.51 54
ABBAR–L ABBAR–T
16 1987 Superstition SUPERST/ SUPERST/ 0.36 46
Hills B-ICC000 B-ICC090
(continued)
104 M. Çoşut et al.

Table 5 (continued)
Earthquake Year Record information Motions
No Earthquake Component-1 Component-2 PGAmax (g) PGVmax (cm/
Location s)
17 1987 Superstition SUPERST/ SUPERST/ 0.45 36
Hills B-POE270 B-POE360
18 1992 Cape CAPEMEND/ CAPEMEND/ 0.55 44
Mendocino RIO270 RIO360
19 1999 Chi-Chi, CHICHI/ CHICHI/ 0.44 115
Taiwan CHY101-E CHY101-N
20 1999 Chi-Chi, CHICHI/ CHICHI/ 0.51 39
Taiwan TCU045-E TCU045-N
21 1971 San SFERN/ SFERN/ 0.21 19
Fernando PEL090 PEL180
22 1976 Friuli, Italy FRIULI/ FRIULI/ 0.35 31
A-TMZ000 A-TMZ270

algorithms. Looking at the table in more detail, the displacements are decreased
considerably by TMD. Therefore, the structure is not influenced too much by any
dynamic loads, so the safety of the construction increases.
50 and 100 iterations were used for the optimization process to reach objective
function, which is related to the period of TMD and the damping ratio of TMD.
It can obviously be seen that displacement values without TMD are higher than
the structure which are utilized TMD. Hybrid and Jaya algorithms have the same
displacement for 100 iterations with 10 population numbers (Table 6) whereas they
have different displacement results for 50 iterations with 5 population numbers.
When 100 iterations were chosen for the optimization, Jaya and the Hybrid algo-
rithm reached the objective function. Thus, the result of displacements by using
TMD was found same for these algorithms. Nevertheless, when 50 iterations were
chosen for the optimization, Jaya and the Hybrid algorithm did not reach the objec-
tive function. Therefore, the result of displacements was found different for these
algorithms.
In order to compare Hybrid and Jaya algorithms, optimization was repeated 10
times and saved maximum displacement for 50 iterations with 5 population numbers.
After 10 runs, the maximum displacement and minimum displacement are chosen.
In addition, mean and standard deviation are found in these displacement results.
Table 7 delineates the values of the results.
Maximum displacement was found for the Hybrid and Jaya algorithms, at about
0.0602 m and 0.0707 m respectively.
Minimum displacement was found for the Hybrid and Jaya algorithms, at roughly
0.0419 m and 0.0466 m respectively.
The mean of 10 runs’ displacement was computed for the Hybrid and Jaya algo-
rithm, at approximately 0.05 m and 0.0547 m respectively. Therefore, the mean of
Structural Control Systems and Tuned Mass Damper Optimization … 105

Table 6 Earthquakes and

Earthquake no. Components Maximum displacemen
their maximum displacement
without TMD and by using No TMD With TMD
TMD All runs For 100 iterations
1 1 0.0271 0.0169
2 0.0377 0.0263
2 1 0.0229 0.0171
2 0.0236 0.0189
3 1 0.0455 0.0435
2 0.0213 0.0211
4 1 0.0160 0.0113
2 0.0304 0.0183
5 1 0.0206 0.0129
2 0.0201 0.0175
6 1 0.0617 0.0356
2 0.0288 0.0275
7 1 0.0350 0.0281
2 0.0463 0.0267
8 1 0.0079 0.0074
2 0.0115 0.0095
9 1 0.0146 0.0118
2 0.0361 0.0311
10 1 0.0087 0.0060
2 0.0097 0.0078
11 1 0.0148 0.0096
2 0.0198 0.0120
12 1 0.0287 0.0244
2 0.0212 0.0198
13 1 0.0732 0.0424
2 0.0212 0.0163
14 1 0.0400 0.0271
2 0.0289 0.0237
15 1 0.0277 0.0260
2 0.0396 0.0339
16 1 0.0204 0.0137
2 0.0151 0.0095
17 1 0.0250 0.0179
2 0.0169 0.0144
18 1 0.0302 0.0222
(continued)
106 M. Çoşut et al.

Table 6 (continued)
Earthquake no. Components Maximum displacemen
No TMD With TMD
All runs For 100 iterations
2 0.0469 0.0376
19 1 0.0275 0.0189
2 0.0191 0.0142
20 1 0.0275 0.0263
2 0.0323 0.0196
21 1 0.0195 0.0133
2 0.0110 0.0088
22 1 0.0300 00205
2 0.0205 0.0163

Table 7 Values after 10 runs for 50 iterations with 5 populations numbers

Maximum Minimum Mean Standard
Displacement Displacement Deviation
HYBRID 0.060226463 0.041869342 0.049886476 0.008663872
JAYA 0.0707 0.046601362 0.054665278 0.01239423

Hybrid algorithm was found to be closer than the Jaya algorithm in terms of objective
function result, which was 0.452 m.
Standard deviation results were found according to objective function displace-
ment. The standard deviation of Hybrid and Jaya algorithms were calculated, at
about 0.00866 and approximately 0.0124 respectively, which were too close to zero.
Nevertheless, the Hybrid algorithm is getting closer to zero compared to those of
Jaya.
Maximum and minimum displacements are shown in Table 7. The results of these
displacements in terms of variables which are period and damping ratio demonstrated
in Table 8. It can obviously be seen that different period and damping ratio values
were found due to displacements.
Table 9 shows the results according to 100 iterations. It can be seen that Hybrid
and Jaya which reached the objective function algorithm results have the same period
and damping ratio. The period was found as 0.379, and also the damping ratio was
found as 0.454 for 100 iterations with 10 population numbers in both Jaya and Hybrid
algorithms.
The earthquake acceleration figure can be seen in Fig. 14.
Algorithm results were used to find changes between displacement and acceler-
ations. When the optimization reaches the objective function, period and damping
ratio are found as 0.379 s and 0.454 respectively. Figure 15 and Fig. 16 are demon-
strated. For time histories and displacement and acceleration are observed at lower
values thanks to using TMD.
Structural Control Systems and Tuned Mass Damper Optimization … 107

Table 8 Period and damping

Displacements Period (s) Damping ratio
ratio results for displacements
(5 population and 50 iteration HYBRID Maximum 0.3556 0.2092
numbers) Displacement
Minimum 0.2852 0.1422
Displacement
JAYA Maximum 0.2145 0.2202
Displacement
Minimum 0.3423 0.4758
Displacement

Table 9 Hybrid and Jaya

Explanation 100 iterations
algorithm results
Period (s) Damping ratio
No control system 0.288 –
TMD HYBRID 0.379 0.454
JAYA 0.379 0.454

Fig. 14 Earthquake
acceleration (DUZCE/
BOL000)
108 M. Çoşut et al.

Fig. 15 Displacement of
structure (DUZCE/BOL000)

Fig. 16 Total acceleration

(DUZCE/BOL000)

6 Conclusion

Although the structure is built in accordance with the regulations, some loading
situations that are not taken into account in the design can cause regional damage by
creating excessive stresses on the system elements when the structure is encountered,
which triggers a collapse situation. Prevention of these and similar situations have
been carried out with building control systems. There are varieties of building control
systems that can be placed in many locations of the building, which makes it easy to
add these systems to pre-built buildings later. When these systems are added to the
structure; they can reduce the interaction between the structure and the ground, they
can reduce the relative stories displacements (allowing the displacements to remain
at a certain level), the structure will prevent the structure from being damaged by
Structural Control Systems and Tuned Mass Damper Optimization … 109

reducing the acceleration and speed of the acting load. By increasing the construction
period at a certain rate through the systems it contains, it will ensure that the structure
remains on the safe side.
Optimization of the TMD system was performed using Hybrid and Jaya algo-
rithms. As a result of this, the period and damping ratio value of TMD were found
depending on the amount of iteration. In addition, the displacement as well as accel-
eration of the structure were found and shown on graphs. It is clearly seen in the
graphs that the displacement and acceleration values generated in the structure when
no control system is used in the structure will take much larger values than in the
case of using TMD in the structure. Therefore, the structure will remain on the
safe side with a significant reduction in structure displacements and acceleration
values. Furthermore, the results of TMD can be found Jaya and Hybrid algorithms
which reached objective function in about 100 iterations. Looking at the Hybrid and
Jaya algorithm for different iteration numbers which are 50 iterations with 5 popu-
lation numbers, they have not reached the objective function. Moreover, this opti-
mization was repeated 10 times and saved for both algorithms. It was seen that the
Hybrid algorithm approached the objective function better than Jaya algorithm. As a
consequence, reaching the objective function is affected by iteration and population
numbers as well as algorithm types.

References

1. Isalgue, A., Lovey, F.C., Terriault, P., Ferran, Martorell, Torra, R.M., Torra, V.: SMA for
dampers in civil engineering. Mater. Trans. 47 (3), 682–690 (2006)
2. Vu, K.D.T., Nguyen, B.N.: Sustainable material awareness, belief and readiness in housing
development. Built Environment Journal 18(2), 88–96 (2021)
3. Utsev, T., Tiza, M.T., Sani, H.A., Sesugh, T.: Sustainability in the civil engineering and
construction industry: a review. J. Sustain. Constr. Mater. Technol. 7(1), 30–40 (2022)
4. Bekdaş, G., Niğdeli, S.M.: Estimating optimum parameters of tuned mass dampers using
harmony search. Eng. Struct. 33(9), 2716–2723 (2011)
5. Farshidianfar, A., Soheili, S.: Ant colony optimization of tuned mass dampers for earthquake
oscillations of high-rise structures including soil–structure interaction. Soil Dyn. Earthq. Eng.
51, 14–22 (2013)
6. Araz, O., Kahya, V.: Design of series tuned mass dampers for seismic control of structures
using simulated annealing algorithm. Arch. Appl. Mech., 4343–4359 (2021)
7. Nigdeli, S.M., Bekdas, G.: Optimum tuned mass damper approaches for adjacent structures.
Earthq. Struct. 7(6), 1071–1091 (2014)
8. Liao, T.W.: Two hybrid differential evolution algorithms for engineering design optimization.
Appl. Soft Comput. 10, 1188–1199 (2010)
9. Rao, R.V.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7, 19–34 (2016)
10. FEMA P-695. Quantification of Building Seismic Performance Factors. Washington
11. Montuori, R., Nastri, E., Piluso, V.: Theory of plastic mechanism control for eccentrically
braced frames with inverted Y-scheme. J. Constr. Steel Res. 92, 22–135 (2015)
12. Zeynali, K., Monir, H.S., Mirzai, N.M., Hu, J.W.: Experimental and numerical investigation of
lead-rubber dampers in chevron concentrically braced frames. Arch. Civil Mech. Eng. 18(1),
162–178 (2018)
110 M. Çoşut et al.

13. Davalos, C., Escudero, C., Cano, G., Hernandez, V., Cruz, S.: Damage modelling of hysteretic
energy dissipative devices in building seismic response control. Researchgate 73, 1883–1890
(2017)
14. Bekdaş, G., Niğdeli, S.M.: Mass ratio factor for optimum tuned mass damper strategies. Int. J.
Mech. Sci. 71, 68–84 (2013)
15. Madsen, L.P.B., Thambiratnam, D.P., Perera, N.J.: Seismic response of building structures with
dampers in shear walls. Comput. Struct. 81(4), 239–253 (2003)
16. Ahn, S. K., Min, K. W., Lee, S. H., Park, J. H., Lee, D. G., Oh, J. G.: Control of Wind-Induced
Acceleration Response of 46-Story R.C. Building Structure Using Viscoelastic Dampers
Replacing Outrigger System. In CTBUH Conference Seoul, 504–509 (2004).
17. Smith, R.J., Willford, M.R.: The damped outrigger concept for tall buildings. Struct. Design
Tall Spec. Build. 16(4), 501–517 (2007)
18. Joung, J., & Kim, D.: Experimental and Numerical Studies of a Newly Developed Semi-
active Outrigger Damper System. In CTBUH 9th World Congress Shanghai 2012 Proceedings,
790–796 (2011).
19. Asai, T., Chang, C.M., Phillips, B.M., Spencer, B.F., Jr.: Real-time hybrid simulation of a smart
outrigger damping system for high-rise buildings. Eng. Struct. 57, 177–188 (2013)
20. Kidokoro, R.: Self Mass Damper (SMD): Seismic Control System Inspired by the Pendulum
Movement of an Antique Clock. In The 14th World Conference on Earthquake Engineering,
(2008).
21. Öncü-Davas, S., Alhan, C.: Realibility of semi-active isolation under far-fault earthquakes.
Mech. Syst. Signal Process. 114, 146–164 (2019)
22. Aldemir, A., Ersin, A.: Depreme Dayanıklı Yapı Tasarımında Yeni Yaklaşımlar. Türkiye
Mühendislik Haberleri Sayısı 435, 81–89 (2005)
23. Bekdaş, G., Niğdeli, S.M.: Estimating optimum parameters of tuned mass dampers using
harmony search. Eng. Struct. 33, 2716–2723 (2011)
24. Sadek, F., Mohraz, B., Taylor, A.W., Chung, R.M.: A method estimating the parameters of
tuned mass dampers for seismic applications. Earthq. Eng. Struct. Dynam. 26(617), 617–635
(1997)
25. Şahin, M.: Depreme Dayanıklı Yapı Tasarımında Pasif ve Aktif Kontrol Sistemleri. Tasarım
Kuram, Sayı 1, 60–65 (1999)
26. Kayabekir, A.E., Bekdaş, G., Niğgeli, S.M., Geem, Z.W.: Optimum design of PID controlled
active tuned mass damper via modified harmony search. Applied Science 10, 2976 (2020)
27. Pourzeynali, S., Lavasani, H.H., Modarayi, A.H.: Active control of high-rise building structures
using fuzzy logic and egentic algorithms. Eng. Struct. 29, 346–357 (2007)
28. Gkatzogias, K.I., Kappos, A.J.: Semi-active Control Systems in Bridge Engineering: A Review
of the the Current State of Practise. Struct. Eng. Int. 26(4), 290–300 (2016)
29. Wang, L., Nagarajaiah, S., Shi, W., Zhou, Y.: Seismic performance improvement of base-
isolated structures using a semi-active tuned mass damper. Eng. Struct. 271, 114963 (2022)
30. Bekdaş G., Nigdeli M. N., Yücel M., Kayabekir A. E.: Yapay Zeka optimizasyon Algoritmaları
ve Mühendislik Uygulamaları. Seçkin Yayıncılık, Ankara (2021)
31. Ting, T.O., Yang, X.S., Cheng, S., Huang, K.: Hybrid metaheuristic algorithms: past, present
and future, recent advances in swarm intelligence and evolutionary computation, studies in
computational. Intelligence 585, 71–83 (2015)
Manta Ray Foraging and Jaya Hybrid
Optimization of Concrete Filled Steel
Tubular Stub Columns Based on CO2
Emission

Celal Cakiroglu, Kamrul Islam, and Gebrail Bekdaş

Abstract Concrete-filled steel tubular (CFST) columns exhibit favorable character-

istics and have been studied extensively particularly through experiments. However,
the CO2 emission in the production process of these structural members should be
reduced to minimize the environmental impact. At the same time, the performance
of these structures should be kept at a satisfactory level. This can be achieved using
metaheuristic optimization algorithms. The most commonly used indicator of struc-
tural performance for CFST columns is the ultimate axial load carrying capacity (Nu ).
This quantity can be predicted using various equations available in design codes and
the research literature. However, most of these equations are only applicable within
certain parameter ranges. A recently developed set of equations from the CFST liter-
ature was applied for the prediction of Nu due to its improved ranges of applicability.
Furthermore, novel metaheuristic algorithms called Manta Ray Foraging Optimiza-
tion and, Jaya algorithm are applied to the cross-section optimization of rectangular
CFST columns. The improvement of the structural dimensioning under Nu constraint
was demonstrated. The objective of optimization was to minimize the CO2 emission
associated with the fabrication of CFST stub columns. For different concrete classes
and load capacities, the optimum cross-sectional dimensions have been obtained.

Keywords Metaheuristic optimization · CFST columns · Axial compressive

strength · Jaya · Manta ray foraging

C. Cakiroglu
Department of Civil Engineering, Turkish-German University, Beykoz, 34820 Istanbul, Turkey
e-mail: [email protected]
K. Islam
Department of Civil, Geological and Mining Engineering, Polytechnique Montréal, Montréal,
Canada
e-mail: [email protected]
G. Bekdaş (B)
Department of Civil Engineering, Istanbul University - Cerrahpaşa, Avcılar, 34320 Istanbul,
Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 111
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_7
112 C. Cakiroglu et al.

1 Introduction

High ductility, rapid construction, and increased strength are a few of the qualities
making concrete-filled steel tubular (CFST) columns preferable in structural design.
However, it should be noted that the concrete constituents and steel constituents of
CFST columns can have considerable amounts of CO2 emission associated with their
production process. In line with the efforts of the construction industry to reduce its
carbon footprint design engineers should consider carbon emissions as an additional
factor besides the structural performance. The production of 1 kg of steel and 1 kg
of concrete are known to cause an emission of approximately 1.38 kg and 0.12 kg of
CO2 respectively [1]. Furthermore, Arama et al. [2] showed that the concrete volume
has a more decisive effect on the CO2 emission compared to the steel weight. Table
1 shows the amounts of CO2 emission related to different classes of concrete.
Based on extensive experimental studies various methods are developed and
included in the design codes to predict the load-carrying capacity (Nu ) of CFST
columns. However, for most of these methods, certain ranges of applicability are
defined so that they are not recommended to be used for variable values outside of
these ranges. For instance, the equations included in the AISC 360-16 code [3] are
recommended only for fy ≤ 525 MPa and for 21 MPa ≤ fc ' ≤ 70 MPa. The widely
accepted definition of Nu is the maximum axial compressive load if this load level is
reached at an axial strain of less than 1%. If the maximum load is observed at greater
strain levels, then Nu is defined as the load level at 1% axial strain [4–6].
The ultimate axial load carrying capacity prediction equations developed by Wang
et al. [6] obtain the Nu value by adding the structural capacities of the steel casing
and the concrete core as shown in Eq. 1 for circular CFST stub columns.

Nu = Ns + Nc , (1)

'
N s = ηs f y A s , N c = ηc f c A c , (2)
) (
0.14D
ηs = 0.95 − 12.6 f y−0.85 ln< 1, (3)
t
[ ( )0.04 ( ) ]( )
D ' 0.1 t f y 0.51
ηc = 0.99 + 5.04 − 2.37 fc >1 (4)
t D f c'

In Eq. 1, Ns and Nc stand for the load-carrying capacities of the steel casing and
the concrete inner part respectively. In Eq. 2, As and Ac are the areas of the steel

Table 1 CO2 emission of different concrete classes (Fantilli et al. [1])

Concrete class C25 C40 C60 C80
CO2 emission [kg/m3 ] 215 272 350 394
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 113

tube and the concrete core cross-sections respectively. Under the axial load, the steel
casing exerts a confining force on the concrete core which in turn causes the steel wall
thickness to decrease due to hoop stresses. Also, due to being a thin-walled structure,
the steel casing is prone to local buckling which adversely affects Nu . In Eq. 2, ηs
is a reduction factor that introduces these adverse effects into the overall structural
capacity. On the other hand, the confinement by the steel casing has a favorable
effect on the capacity of the concrete core. This favorable effect is quantified by the
amplification factor ηc in Eq. 2. The equations for ηs and ηc are given in Eqs. 3,4. In
Eq. 3, D and t are the diameter and wall thickness of the steel casing respectively as
depicted in Fig. 1. Table 2 shows the ranges of parameters in which Eq. 1 is valid.
Equations 1 and 2 are also applicable to columns with rectangular cross-sections.
In case of rectangular cross-sections, ηs and ηc are calculated as in Eqs. 5 and 6.
( ' )2
( ) D
ηs = 0.91 + 7.31 · 10−5 f y − 1.28 · 10−6 + 2.26 · 10−8 f y , (5)
t
( )1.3
t fy
ηc f c' Ac , ηc = 0.98 + 29.5( f y ) −0.48
K s0.2 (6)
(D ' f c' )

(a) (b)

Fig. 1 a Circular CFST cross-section; b rectangular CFST cross-section

Table 2 Parameter ranges

Diameter to thickness ratio 12 ≤ D/t ≤ 150
for which Eqs. (1–7) are
applicable [6] Width to thickness ratio 12 ≤ B/t ≤ 100
Height to width ratio 1 ≤ H/B ≤ 2
Yield stress of the steel tube 175 MPa ≤ fy ≤ 960 MPa
'
Compressive strength of the 20 MPa ≤ fc ≤ 120 MPa
concrete
Steel wall thickness [mm] 3 ≤ t ≤ 30
114 C. Cakiroglu et al.
( )
1 B − 2t 2
Ks = (7)
3 H − 2t
'
In Eqs. 5 and 6, D is the√equivalent diameter of the rectangular cross-section
'
which is calculated as D = B2 + H2 where B and H are the width and height of
the cross-section respectively as shown in Fig. 1b. Due to the rectangular shape of
the cross-section the concrete confinement is not as effective as in the circular form.
This lack of concrete confinement in rectangular cross-sections is quantified with the
equivalent confinement coefficient K s in Eq. 7.

2 Optimization Methodology

Metaheuristic optimization techniques are finding broad application in numerous

areas of engineering. To give a few examples of these applications the works of
Geem et al. [7–9] on water network design and vehicle routing, Kayabekir et al. [10,
11] on reinforced concrete retaining walls and plane stress systems, and Cakiroglu
et al. [12–15] on laminated composite plates, steel plate girders and retaining walls
can be mentioned. Furthermore, physics-inspired optimization algorithms such as the
atom stabilization algorithm [16–18], particle swarm optimization [19–22], genetic
algorithm [23] and evolutionary optimization algorithms [24, 25] have been widely
implemented in the optimization of various engineering systems. The following
sections introduce the algorithms implemented in this study for the optimal design
of rectangular CFST stub columns.

2.1 Jaya Optimization

In this study, a novel metaheuristic algorithm called the Jaya algorithm was imple-
mented to find the cross-sectional properties of a CFST column that lead to the
minimum CO2 emission. The Jaya algorithm was developed by Rao [26] as a general-
purpose optimization technique. The major difference of this algorithm compared to
the other state-of-the-art metaheuristic techniques is that the Jaya algorithm does not
depend on parameters that need to be tuned by the user. This is a great advantage since
it opens up the possibility of integration of this algorithm into the structural design
process by engineers that are not necessarily experts in the field of optimization. A
flowchart of the Jaya algorithm is given in Fig. 2.
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 115

Fig. 2 Flow chart of the

Jaya algorithm

The algorithm starts with the creation of an initial population where each randomly
generated population member represents a candidate for the optimum solution. In the
process of minimizing the CO2 emission, the side lengths of the rectangular cross-
section and the wall thickness are the design parameters while the yield stress of steel
and concrete compressive strength are kept constant throughout the optimization
process. In the case of circular cross-sections, the side lengths are replaced by the
outer diameter. In the Jaya optimization, each member of the population is an array
of design variables taking values within their predefined domain of allowable values.
In this process, Nu is kept above a certain predefined value which becomes one
of the optimization constraints. Additional constraints are introduced on the cross-
sectional dimensions as well. These variables can take any value continuously within
their predefined constraints. Once the initial population is generated, all members of
the population go through a Jaya iteration step which is described in Eq. 8.
( | |) ( | |)
Xt+1
i = xti + r1 g∗ − |xti | − r2 gw − |xti | (8)

In Eq. 8 xi t is the candidate solution vector in the t-th iteration, r1 , r2 ∈ (0,1)

are random numbers and g*, gw are the best- and worst-performing members of
the population respectively. After every iteration the newly generated vector xi t+1 is
compared to gw with respect to its CO2 emission value and in case the new vector
performs better, it replaces gw gw . Every newly generated vector is controlled with
respect to its Nu value and dimensions. In case a cross-sectional dimension is outside
its predefined upper and lower bounds this dimension is set equal to its lower or
116 C. Cakiroglu et al.

upper bound. After this operation if the Nu Nu value of the new vector is below the
lower bound for Nu then the newly generated vector is discarded.

2.2 Manta Ray Foraging Optimization (MRFO)

This algorithm is inspired by the intelligent foraging strategies of manta rays that
feed on planktons. The foraging strategies of Manta rays can be classified as chain
foraging, cyclone foraging, and somersault foraging [27]. The MRFO technique
combines the mathematical models of the chain foraging, cyclone foraging and
somersault foraging strategies of manta rays in one algorithm. Like many other
metaheuristic optimization algorithms, the starting point of MRFO is the generation
of a random population that fulfills certain optimization constraints. Afterward, the
algorithm enters into a loop of incremental improvements of this population. In every
iteration, the algorithm branches into one of cyclone or chain foraging based on the
value of a random variable in that particular iteration. Once the cyclone or chain
foraging operation is completed the population goes through somersault foraging
after which the values of the newly generated vectors are compared to the upper and
lower bounds. In case they exceed one of these boundaries they are assigned values
equal to the boundary values. Newly generated vectors replace old ones in case they
perform better.
The chain foraging strategy is based on the motion of manta rays in a head-to-tail
foraging chain toward areas with high plankton concentration. In this process every
population member except for the one in front of the chain (the first member) moves
towards the best performing member of the population and towards the individual
directly in front of it. The mathematical description of this motion is given in Eq. 9
where N denotes the size of the population and r ∈ (0, 1). A graphical depiction of
the chain foraging process is given in Fig. 3.
{ ( t ) ( t )
xit + r (xbest − xit ) + α(xbest − xit )ifi = 1
xit+1 = (9)
xit + r xi−1
t
− xit + α xbest
t
− xit ifi > 1
√
α = 2r |log(r)| (10)

In cyclone foraging manta rays move in a spiral pattern towards the plankton. In
addition to their motion towards the plankton, also each manta ray in the foraging
chain moves toward the individual in front of it. The cyclone foraging step further
branches into exploration and exploitation phases depending on the current number
of iterations. In the beginning stages of the algorithm the exploration operation is
performed in order to search for the global optimum. Equation 11 describes the
exploratory operation in cyclone foraging.
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 117

Fig. 3 Chain foraging

process

{ ( t ) ( t )
t
xrand + r (xrand − xit) + β( xrand − xit) ifi = 1
xit+1 = (11)
t
xrand + r xi−1
t
− xit + β xrand
t
− xit ifi > 1
T−t+1
β = 2er1 T · sin(2πr1 ) (12)

t
xrand = LB + r0 (UB − LB) (13)

In Eqs. 11 and 12, t and T are the current and maximum number of iterations
respectively and r1 ∈ (0, 1). In Eq. 13, r0 is another random number in (0,1) which
is used in generating a random vector between the lower bound vector LB and the
upper bound vector UB. On the late stages of the iterations when t/T > r0 ∈ (0, 1),
the exploitation operation is carried out which is described in Eq. 14.
{ ( t ) ( t )
t
xbest + r (xbest − xit ) + β(xbest − xit )ifi = 1
xit+1 = (14)
t
xbest + r xi−1
t
− xit + β xbest
t
− xit ifi > 1

Finally, after all population members go through the chain foraging and cyclone
foraging phases, the somersault phase begins. In this phase the population members
settle in a new position between their previous positions and another position symmet-
rical with respect to the position of the best population member. Mathematically
somersault iteration can be described as in Eq. 15 where S is the somersault factor
which determines the somersault range. In this study S is initialized as S = 2.

xit+1 = xit + S(r2 xbest

t
− r3 xit ) (15)
118 C. Cakiroglu et al.

In this study a new hybrid algorithm has been proposed by combining the Jaya
and Manta ray foraging algorithms. The Jaya iteration formula given in Eq. 8 has
been implemented in the exploration phase of the algorithm whereas the somersault
iteration given in Eq. 15 has been implemented in the exploitation phase.

3 Result Analysis

This section presents the steps of the CO2 emission minimization for rectangular
CFST stub columns. The optimization process is visualized for C25, C40, and C60
concrete classes separately. For each concrete class different lower bound values for
the ultimate axial load capacity (Nu,min ) are set and these values serve as optimization
constraints that guarantee a minimum level of acceptable structural performance.
In addition to Nu,min also dimensional constraints are in place in all stages of the
optimization. Using the Jaya algorithm, the cross-sectional width and height as well
as the steel casing wall thickness of the member are optimized while keeping Nu
above Nu,min at all times. The objective of the optimization is the CO2 emission
resulting from the production of a CFST stub column with unit height. In Fig. 4 the
results of the Jaya optimization are presented at four different levels of Nu,min for
columns with rectangular cross-section and a concrete class of C25. The curves in
Fig. 4 show CO2 emissions associated with the best performing member in all of the
population at each iteration step. From Fig. 4 it is clear that after the first 25 iterations
a convergence of the minimum CO2 emission is observed at all levels of Nu,min . For
the Nu,min threshold values of 5000 kN, 4000 kN and 2000 kN this convergence
is observed as early as after the first 10 to 20 iterations. Table 3 presents a list of
the minimum CO2 emission values together with the corresponding cross-sectional
dimensions for a constant yield stress of 800 MPa for the steel casing.
Figures 4, 5, and 6 visualize the development of the minimum CO2 emissions
obtained through the Jaya optimization process using C25, C40, and C60 concrete
classes respectively. For each concrete class the optimization process is repeated at
four different levels of the ultimate axial load capacity constraint Nu,min . The cross-
sectional configurations which do not satisfy the Nu,min constraints are not included in
the population of solution candidates during the Jaya iterations. In these optimizations
the cross-sectional height (H), width (B), and the thickness of the steel casing (t) are
the optimization variables, and modifying these dimensions leads to changes in the
concrete and steel volumes needed to manufacture the CFST stub columns which in
turn leads to different amounts of CO2 CO2 emissions. For each concrete class and
Nu,min constraint the corresponding cross-sectional dimensions are listed in Table 3.
Using the Jaya algorithm, average values of 50.4 kg, 48 kg, and 42.5 kg could be
obtained as minimum CO2 emissions corresponding to C25, C40, and C60 concrete
classes respectively.
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 119

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 4 Jaya optimization of a rectangular cross-section with respect to CO2 emission (C25)

Table 3 Minimum CO2 emissions for rectangular cross-section (Jaya)

Concrete class Nu,min [kN] Min. CO2 emission [kg] B [mm] H [mm] t [mm]
C25 6000 71.6 169 249 7.3
5000 59.1 118 183 9
4000 47.3 98 152 8.9
2000 23.6 70 108 6.2
C40 6000 67.7 233 275 4.8
5000 56.5 212 251 4.3
4000 45.2 190 225 3.9
2000 22.6 126 153 3
C60 6000 59.4 262 279 3
5000 49.6 229 247 3
4000 40 194 214 3
2000 20.8 115 135 3
120 C. Cakiroglu et al.

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 5 Jaya optimization of a rectangular cross-section with respect to CO2 emission (C40)

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 6 Jaya optimization of a rectangular cross-section with respect to CO2 emission (C60)
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 121

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 7 Manta ray foraging—Jaya hybrid optimization of a rectangular cross-section with respect
to CO2 emission (C25)

Figures 7, 8, and 9 visualize the development of the minimum CO2 emissions

obtained through the Manta ray foraging—Jaya hybrid algorithm using C25, C40, and
C60 concrete classes respectively. The optimal cross-sectional dimensions obtained
through this hybrid algorithm have been listed in Table 4. For concrete classes C25.
C40, and C60 optimal cross-sectional dimensions corresponding to four different
lower bounds of the axial load carrying capacity have been obtained. These lower
bound values have been varied between 6000 and 2000 kN as listed in Table 4.
The CO2 CO2 emission values corresponding to each optimal design have been
listed in Table 4. Using the Manta ray foraging—Jaya hybrid algorithm, average
values of 50.3 kg, 48 kg, and 42.5 kg could be obtained as minimum CO2 emissions
corresponding to C25, C40, and C60 concrete classes respectively.
122 C. Cakiroglu et al.

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 8 Manta ray foraging—Jaya hybrid optimization of a rectangular cross-section with respect
to CO2 emission (C40)

(a)Nu,min = 6000kN (b)Nu,min = 5000kN

(c)Nu,min = 4000kN (d)Nu,min = 2000kN

Fig. 9 Manta ray foraging—Jaya hybrid optimization of a rectangular cross-section with respect
to CO2 emission (C60)
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 123

Table 4 Minimum CO2 emissions for rectangular cross-section (Manta ray foraging—Jaya hybrid)
Concrete class Nu,min [kN] Min. CO2 emission [kg] B [mm] H [mm] t [mm]
C25 6000 70.9 225 302 5.4
5000 59.3 118 207 8.2
4000 47.3 98 152 8.8
2000 23.6 69 108 6.3
C40 6000 67.7 231 276 4.8
5000 56.5 211 250 4.4
4000 45.2 190 225 3.9
2000 22.6 126 153 3
C60 6000 59.4 262 279 3
5000 49.6 229 247 3
4000 40 194 214 3
2000 20.8 94 168 3

4 Conclusion

CFST members find widespread applications in the area of structural engineering

because of the enhanced mechanical properties that these members possess. However,
the production process of the concrete constituents and steel constituents of these
structures can be linked to considerable amounts of CO2 emissions. To reduce the
carbon footprint associated with the construction industry, there is an ongoing effort
to reduce the CO2 CO2 emission caused by the production of various construction
materials. At the same time, it is the responsibility of structural design engineers to
guarantee the performance of structures such as CFST columns in an optimum way so
that they satisfy certain load carrying capacity requirements. These requirements led
to an increasing interest in the application of meta-heuristic optimization techniques
to structural optimization problems in recent years since these techniques can solve
many engineering problems more effectively. The Jaya optimization algorithm is a
newly developed meta-heuristic algorithm that was shown to be suitable to tackle
structural design optimization problems. In this study the Jaya algorithm has been
combined with the Manta ray foraging algorithm which is another newly developed
metaheuristic algorithm and a hybrid algorithm has been proposed. These algorithms
have been applied to the problem of CO2 emission minimization by varying the
cross-sectional dimensions while keeping the structural capacity above a certain
level. The class of the core concrete was shown to have a significant influence on
the CO2 emission. The configurations with C60 concrete class had the least amount
of CO2 emission whereas, it was observed that the configurations with less concrete
compressive strength exhibited greater CO2 emission. Further research in this field
can include the study of CFST columns concerning the changes in the cross-sectional
slenderness properties and column heights. Also, concrete-filled double skin steel
124 C. Cakiroglu et al.

tubular column behavior can be investigated. Furthermore, the behavior of different

materials such as fiber-reinforced composites as outer casing can be analyzed.

References

1. Fantilli, A.P., Mancinelli O., Chiaia B.: The carbon footprint of normal and high-strength
concrete used in low-rise and high-rise buildings. Case Stud. Constr. Mater. 11, 1–7, e00296
(2019)
2. Arama, Z.A., Kayabekir, A.E., Bekdaş, G., Geem, Z.W.: CO2 and cost optimization of rein-
forced concrete cantilever soldier piles: a parametric study with harmony search algorithm.
Sustainability 12(15), 1–24, 5906 (2020)
3. American Institute of Steel Construction (AISC): Specification for Structural Steel Buildings;
AISC 360-16. American Institute of Steel Construction, Chicago, IL, USA (2016)
4. Tao, Z., Wang, Z.B., Yu, Q.: Finite element modelling of concrete-filled steel stub columns
under axial compression. J. Constr. Steel Res. 89, 121–131 (2013)
5. Uy, B., Tao, Z., Han, L.H.: Behaviour of short and slender concrete-filled stainless steel tubular
columns. J. Constr. Steel Res. 67(3), 360–378 (2011)
6. Wang, Z.B., Tao, Z., Han, L.H., Uy, B., Lam, D., Kang, W.H.: Strength, stiffness and ductility
of concrete-filled steel columns under axial compression. Eng. Struct. 135, 209–221 (2017)
7. Geem, Z.W.: Harmony search optimisation to the pump-included water distribution network
design. Civ. Eng. Environ. Syst. 26(3), 211–221 (2009)
8. Geem, Z.W.: Particle-swarm harmony search for water network design. Eng. Optim. 41, 297–
311 (2009)
9. Geem, Z.W., Lee, K.S., Park, Y.: Application of Harmony Search to Vehicle Routing. Am. J.
Appl. Sci. 2, 1552–1557 (2005)
10. Kayabekir, A.E., Arama, Z.A., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Eco-friendly design
of reinforced concrete retaining walls: multi-objective optimization with harmony search
applications. Sustainability 12(15), 1–30, 6087 (2020)
11. Kayabekir, A.E., Toklu, Y.C., Bekdaş, G., Nigdeli, S.M., Yücel, M., Geem, Z.W.: A novel
hybrid harmony search approach for the analysis of plane stress systems via total potential
optimization. Appl. Sci. 10, 1–15, 2301 (2020)
12. Cakiroglu, C., Bekdaş, G., Geem, Z.W.L.: Harmony search optimisation of dispersed laminated
composite plates. Materials 13(12), 1–13, 2862 (2020)
13. Cakiroglu, C., Bekdaş, G., Kim, S., Geem, Z.W.: Optimisation of shear and lateral–torsional
buckling of steel plate girders using meta-heuristic algorithms. Appl. Sci. 10(10), 1–14 3639
(2020)
14. Cakiroglu, C., Islam, K., Bekdaş, G., Kim, S., Geem, Z.W.: Metaheuristic optimization of
laminated composite plates with cut-outs. Coatings 11, 1235 (2021). https://doi.org/10.3390/
coatings11101235
15. Cakiroglu C., Islam K., Bekdaş G., Nehdi M.L.: Data-driven ensemble learning approach for
optimal design of cantilever soldier pile retaining walls. Structures 51, 1268–1280 (2023).
https://doi.org/10.1016/j.istruc.2023.03.109
16. Biswas, A., Mishra K.K., Tiwari, S. and Misra, A.K.: Physics-inspired optimization algorithms:
a survey. J. Optim. 2013, Article ID 438152, 16 pages, (2013). (ESCI)
17. Biswas, A.: Atom stabilization algorithm and its real life applications. Journal of Intelligence
and Fuzzy Systems (JIFS), IOS Press 30(4), 2189–2201 (2016)
18. Biswas, A., Biswas, B. and Mishra, K.K.: An atomic model based optimization algorithm.
In: 2nd International Conference on Computational Intelligence & Networks (CINE 2016),
pp. 63–68. Bhubaneswar (11 Jan. 2016)
19. Biswas, A., Kumar, A., and Mishra, K.K.: Particle swarm optimization with cognitive avoidance
component. In: 2nd International Conference on Advances in Computing, Communications and
Informatics (ICACCI-2013), pp 149–154, Mysore, India (22–25 Aug. 2013a)
Manta Ray Foraging and Jaya Hybrid Optimization of Concrete Filled … 125

20. Biswas, A., Lakra, A.V., Kumar, S., Singh, A.: An improved random inertia weighted particle
swarm optimization. In: 2013 International Symposium on Computational and Business
Intelligence (ISCBI 2013), pp. 96–99, New Delhi (24–26 Aug. 2013b)
21. Biswas, A., Biswas, B., Kumar, A., Mishra, K.K.: Particle swarm optimization with time varying
cognitive avoidance component. Int. J. Comput. Sci. Eng., Inder Sci. 16(1), 27–41 (2018)
22. Biswas, A., Biswas, B.: swarm intelligence techniques and their adaptive nature with appli-
cations. In: Complex System Modeling and Control Through Intelligent Soft Computations
vol 219, ch. 9, pp. 253–271, by: Studies in Fuzziness and Soft Computing, Springer-Verlag,
Germany (2014)
23. Sarkar, D.N., Biswas, A.: Genetic algorithm based deep learning models: a design perspective.
In: 7th International Conference on Mathematics and Computing (ICMC 2021), Shibpur, India
(02–05 Mar. 2021)
24. Biswas, A., Biswas, B.: Visual analysis of evolutionary optimization algorithms. In: II Interna-
tional Symposium on Computational and Business Intelligence, ISCBI 2014, pp. 81–84, New
Delhi (7–8 Dec. 2014)
25. Biswas, A., Biswas, B.: Regression line shifting mechanism for analyzing evolutionary opti-
mization algorithms. In: Soft Computing (SOCO), Springer, vol. 21, no. 21, pp. 6237–6252
(2017)
26. Rao, R.V.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7, 19–34 (2016)
27. Zhao, W., Zhang, Z., Wang, L.: Manta ray foraging optimization: an effective bio-inspired
optimizer for engineering applications. Eng. Appl. Artif. Intell. 87, 103300 (2020). https://doi.
org/10.1016/j.engappai.2019.103300
Optimum Design of Dam Structures
Using Multi-objective Chaos Game
Optimization Algorithm

A. Jafari, M. Saraee, B. Nouhi, M. Baghalzadeh Shishehgarkhaneh,

and S. Talatahari

Abstract This chapter aims to present a novel methodology for determining the
optimal shape of double-curvature arch dams. The proposed approach differs from
previous research by employing a multi-objective optimization approach, which
generates a set of Pareto solutions instead of a single solution. The authors use
the MOCGO algorithm, a multi-objective version of the standard Chaos Game Opti-
mization (CGO) algorithm, to achieve this goal. The methodology involves the use of
parallel-working APDL-MATLAB codes that model, analyze, and obtain the fitness
functions of the arch dam, and interface with the MOCGO algorithm at every step of
the optimization process. The results were compared with MoPSO, NSGA-II, and
MoCSS approaches, and findings show that the proposed algorithm has the potential
to provide effective and robust solutions for finding the optimal shape of double-curve
arch dams.

Keywords Multi-objective Chaos Game Optimization · Double curvature arch

dam · Optimum design

1 Introduction

Arch dams are concrete structures with a base less than half its height and rely on
its curve to transmit some of the water load laterally into the valley margins. Arch
dams may only have one-fifth of the concrete of gravity dams of the same size.
Arch dams can be modeled and designed with a single or double curve [1]. Due to

A. Jafari · M. Saraee · S. Talatahari (B)

Department of Civil Engineering, University of Tabriz, Tabriz, Iran
e-mail: [email protected]
B. Nouhi · S. Talatahari
Faculty of Engineering & Information Technology, University of Technology Sydney, Ultimo,
NSW 2007, Australia
M. B. Shishehgarkhaneh
Department of Civil Engineering, Monash University, Clayton, VIC 3800, Australia

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 127
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_8
128 A. Jafari et al.

several influential factors in double-curvature arch dam geometry, the complexity of

numerical modeling, and the need for high computing costs, their design is catego-
rized as a complicated task. Finding the optimal solution for such tough challenges
is more difficult when various designs must be executed. The nonlinear character of
the relevant objective function and constraints of the arch dam optimization issue
and the need to incorporate diverse boundary conditions necessitate adopting unique
optimization techniques in dealing with this problem, considering some primary
constraints [2]. The behavior constraint is one of the most significant constraints.
Natural frequencies are essential factors that influence how dynamically the arch
dams behave. Various restrictions should be placed on the natural frequency range to
decrease the vibrational domain and avoid the resonance phenomena in the dynamic
response of arch dams [3].
Conventionally, the shape of an arch dam is designed using the designer’s exper-
tise, model testing, trial-and-error, and deterministic solution methods. Because of
limitations such as needing an appropriate starting point, the possibility of attaining
an inadequate local minimum, necessitating objective function continuity and differ-
entiability, and having poor performance when dealing with large-scale practical
and intricate situations, metaheuristic (MH) algorithms have gained much popu-
larity among academics in dealing with shape optimization problems in arc dams.
The designer should choose numerous schemes with varied patterns and change them
to create various acceptable shapes to get a superior shape [4]. Zacchei and Molina
[5] used the Bayesian theorem hypothesis to optimize a Spanish double-arch dam’s
shape. Aalami, Talatahari [2] employed the multi-objective charge system search
algorithm for double curvature arch dams’ shape optimization considering natural
frequencies and concrete volume. Seyedpoor, Salajegheh [6] used the simultaneous
perturbation stochastic approximation (SPSA) algorithm for arc dams’ shape opti-
mization. Zhang, Li [7] employed the so-called trial-load approach Xiamen Arch
dam’s shape optimization. Fengjie and Lahmer [8] utilized the genetic algorithm
(GA) in dealing with model-based shape optimization of arch dams; they concluded
that it is possible to find a reliability-based design that guarantees the safety and
serviceability of newly constructed arch dams. Takalloozadeh, Takalloozadeh [9]
used the prominent particle swarm optimization (PSO) method to solve arch dams’
shape on unsymmetrical valleys. Wang, Zhao [10] proposed a novel methodology
consisting of a sequential Kriging surrogate model and GA for arch dams’ shape opti-
mization. Furthermore, Talatahari, Aalami [11] evaluated the arch dam’s optimum
shape’s failure costs using the Hybrid Charged System Search algorithm. Talatahari,
Aalami [12] proposed a multi-objective optimization approach for the design of arch
dams, which aims to simultaneously minimize two objectives: the stress state and
the concrete volume required for the construction of the dam body.
Multi-objective optimization (MOO) refers to the process of optimizing multiple
objective functions concurrently in the field of multi-criteria decision-making. This
approach is used to solve complex problems that involve multiple objectives, and the
goal is to find a set of solutions that are optimal for all objectives simultaneously [13].
Multi-objective optimization is widely used in various fields such as engineering [14–
17], economics [18], and environmental management [19, 20], where decisions need
Optimum Design of Dam Structures Using Multi-objective Chaos Game … 129

to be made concerning multiple and often conflicting criteria. It allows decision-

makers to consider different aspects of the problem simultaneously, and find the
best possible solution that satisfies all objectives. In other words, MOO is a useful
technique for optimizing solutions in various conditions where multiple objectives
need to be considered. One of the main challenges is dealing with contradictions and
non-conformities that arise due to the existence of multiple objectives [21, 22].
This chapter utilizes a Multi-objective Chaos Game Optimization (MOCGO),
for shape optimization of double-curved arch dams considering the frequency and
concrete volume. Chaos Game Optimization (CGO) was developed by Talatahari
and Azizi [23] and is inspired by the fractals self-similarity and chaos game theory.
The concrete volume of the dam body, which should be minimized, and the various
natural frequencies of the arch dam, which should be maximized, are the two other
objective functions in the optimization problem.

2 Optimization Problem Formulation

In the current research work, two different objectives are deemed as follows [2].

2.1 Concrete Volume

The arch dam’s concrete volume ought to be minimized. It might be identified by

integration on the dam’s surfaces as follows [2]:
¨
Minimi ze : f it1 (X ) = |yu (x, z) − yd (x, z)|d A (1)
A

where, yd (x, z) and yu (x, z) show the arch dam’s downstream and upstream surfaces;
A displays the region that results from projecting the body of the arch dam on an xz
plan.

2.2 Natural Frequency

The arch dam’s natural frequencies ought to be maximized by the following equation:

Maximi ze : f it2 (X ) = f rn n = 1, 2, . . . , n f r (2)

where, n f r elucidates the natural frequencies’ number. X , design variables, can be

defined as follows:
130 A. Jafari et al.

X = {s, β, tc1 . . . tcn+1 , r u 1 . . . r u n+1 , r d 1 . . . r d n+1 } (3)

X comprises 2 + 3(n + 1) shape characteristics of the arch dam, where n shows

the dam height divisions’ number.
Numerous constraints, including geometrical and stability constraints, are taken
into consideration in this optimization problem. Equation (4) should be met to guar-
antee that the dam’s downstream and upstream faces do not cross over one another.
Additionally, the slope of the overhang at the upstream and downstream sides of the
dam should meet Eq. (5) in order to build amenities and have smooth cantilevers
across the height of the dam, as follows:
rdi
rdi ≤ rui → − 1 ≤ 0, i = 1, 2, . . . , 6 (4)
rui
s
s ≤ salw → −1≤0 (5)
sahw

where, S = cot(α) is the slope of the overhang at the downstream and upstream
sides of the dam, and rdi and rui elucidate the radius of curvature at the ith level in
the z-direction. For the previously indicated parameter, salw is the allowed absolute
value. The following equation must be accurate for the dam’s sliding stability:

ϕl ≤ ϕ ≤ ϕu (6)

where, ϕ shows the arch dam’s central angle at the ith level in the z-direction, which
ranges from 90° to 130° across the dam’s height.
There are also a few geometric criteria for double-curvature arch dams. One
polynomial of second order is provided in the equation below, and it is used to
calculate the curve of the upstream boundary illustrated in Fig. 1a:

γ2
g(z) = −γz (7)
2βh

where, h shows the dam’s height and z is the elevation at which the upstream face’s
slope equals zero (βh). The thickness of the crown cantilever may be calculated
using a polynomial function by segmenting the dam height into n equal parts.

∑
n+1
tc (z) = L i (z)tci (8)
i=1

where, tci demonstrates the thickness of the centre vertical section at the level it and
z is a Lagrange interpolation formula. z could well be represented as:
Optimum Design of Dam Structures Using Multi-objective Chaos Game … 131

Fig. 1 Schematic of arch

dam: a Crown Cantilever
profile b parabolic shape of
an elevation

∏
n+1
z − zm
L i (z) = (9)
m=1,m/=i
zi − z

where, z i shows the ith level’s z coordinate. The horizontal portion of a parabolic
arch dam is shaped by the following two parabolas, as illustrated in Fig. 1b, for the
goal of a symmetrical canyon and arch thickening from crown to abutment.

x2
yu (x, z) = + g(z) (10)
2ru (z)

x2
yd (x, z) = + g(z) + tc (z) (11)
2rd (z)

where, the parabolas of the upstream and downstream faces are yu (x, z) and yd (x, z),
respectively. According to the following equations of the nth order, ru and rd are the
radius of curvature for the upstream and downstream curves in the z-direction and
may be interpolated by L i (z).

∑
n+1
ru (z) = L i (z)rui (12)
i=1
132 A. Jafari et al.

∑
n+1
rd (z) = L i (z)rdi (13)
i=1

where, rui and rdi are, respectively, the relevant ru and rd values at the controlling
levels.

3 Developed Method

3.1 Single Objective Chaos Game Optimization

Single objective Chaos Game Optimization (CGO) was inspired by Euclidean frac-
tals, which is a branch of Mathematics, [11]. The aim of using chaos game theory
is to find the patterns of points by rolling a die many times to predict later points.
Talatahari and Azizi [11] used the chaos game theory to introduce CGO. The CGO
algorithm is explained briefly using the Sierpinki triangle as a simple example as
follows: The beginning process of the CGO algorithm is started by considering three
points as the vertices of the main triangle. Each of the vertices and the roll are
contributed specific color. By rolling the die depending on what color comes up, the
random points which are named seeds moved half the distance toward which color
comes up. And by rolling the die for the second time, previous results are utilized
for the positions of the next seed. In several iterations, depending on problems the
Sierpinki triangle is organized as depicted in Fig. 2.
The mathematical explanation of CGO is as follows:
The primary positions of the eligible point are defined by a random value as
follows:
( ) { i = 1.2. . . . .n
j j j j
xi (0) = xi.min + rand. xi.max − xi.min . (7)
i = 1.2. . . . .d

j
The primary eligible positions are noted by xi (0) and also the minimum and
maximum permitted values by considering the search space boundary for the ith

Fig. 2 Different scales of Sierpinski triangles

Optimum Design of Dam Structures Using Multi-objective Chaos Game … 133

j j
solution and the jth decision variable are noted xi.min xi.max . And rand is considered
in a limitation between [0 1].
It is worth mentioning that each of the solution candidates, X i , in the search
process uses three special points as the vertices of the Sierpinki triangle: the position
of the best solution candidates (G B), the position of the mean values of some accept-
able primary points M G i and solution candidateX i . For regenerating new triangles to
find new points, the CGO algorithm utilized 3 seeds and a die. The seed was defined
in the X i , G B and the M G vertices respectively as depicted in Fig. 3.
Based on which seed was positioned on the vertices and which colors come up,
the seeds moved toward the G B or the M G i or X i . These processes were formulated
as follows:

Seed i1 = X i + αi × (βi × G B − γi × M G i ) i = 1.2. . . . .n (8)

Seed i2 = G B + αi × (βi × X i − γi × M G i ) i = 1.2. . . . .n (9)

Seed i3 = M G i + αi × (βi × X i − γi × G B) i = 1.2. . . . .n (10)

Fig. 3 Position update for seeds in the search space, [11]

134 A. Jafari et al.

Fig. 4 Update the position

of the fourth seed in the
search space, [11]

where, αi ,βi , γi are a random integer of 0 or 1 to define the seeds’ movement limita-
tions and rolling a die possibility. It is worth mentioning that,αi as exploitation and
exploration controller could be adjusted by following relations:
⎧
⎪
⎪ Rand
⎨
2 × Rand
αi = (11)
⎪
⎪ (δ × Rand) + 1
⎩
(∊ × Rand) + (ε)

where, Rand, δ.ε, were utilized as a random number in the range of [0 1]. In addition,
the mutation phase in CGO is considered as shown in Fig. 4.
The mutation phase was formulated as follows:
( )
Seed i4 = X i xik = xik + R . k = [1.2. . . . .d] (12)

where, k is considered as a random integer in the range of [1 d] and also R was

utilized as a uniform random number in the range of [0 1].
Optimum Design of Dam Structures Using Multi-objective Chaos Game … 135

3.2 Multi-Objective Chaos Game Optimization

The Multi-Objective Chaos Game Optimization, abbreviated as MOCGO, uses the

following concepts [24, 25]:
The algorithm utilizes an archive that contains all non-dominated Pareto optimal
responses. An archive controller is an essential component when a solution is archived
or when the archive is full. Additionally, only a limited number of solutions are
allowed to save in the archive. As part of the optimization process, we compare
non-dominated solutions obtained up to each step with the existing archive. With the
leader selection feature, the current best solutions from the archive are selected as
the search’s leaders. An archive member is omitted from the grid mechanism and a
new solution is added. Increasing the number of solutions in a hypercube increases
the probability of removing a solution. A solution is randomly deleted from the
most crowded segments in the archive, if it is full, in order to make space for a new
solution. When it is necessary, a solution can be added outside the hypercubes in
an exceptional circumstances. To accommodate the new solutions, all components
are extended. This allows several alternative solutions to have their components
changed. By increasing the number of segments, the algorithm can create more
distinct solutions and better identify which solutions are most crowded. This allows
the algorithm to more accurately determine which solutions should be removed in
order to make room for new ones. Furthermore, extending the components allows the
algorithm to adapt to situations where it is necessary to create alternative solutions
with different components.

4 Numerical Results

This study aims to assess the effectiveness of the MOCGO methodology by utilizing
the Morrow Point arch dam as a benchmark problem. To achieve this goal, a numerical
model of the dam-reservoir system is established. The accuracy and validity of the
proposed model are then confirmed by comparing it with experimental and analytical
findings from previous literature. Finally, the developed methodology is employed
to optimize the problem.
In order to apply the MOCGO algorithm to optimize the dam, objective functions
are established. Based on the previous research work [2], this paper utilizes the
inverse of frequencies multiplied together as the second objective function:

∏
n fr
1
f it2 (X ) = n = 1, 2, . . . , n f r (13)
n=1
f rn

Similar to [2], in this study, the first 10 natural frequencies of the dam-reservoir
system are utilized. By using a limited number of natural frequencies, the opti-
mization process becomes more tractable and computationally feasible, while still
136 A. Jafari et al.

capturing the essential aspects of the system’s performance and safety. The selection
of the specific frequencies to be included in the optimization process depends on
various factors, such as the dam’s design, location, and potential hazards.

4.1 Arch Dam Modeling [2]

The analysis of the arch dam in this study is conducted using a hybrid approach,
which combines parallel working MATLAB and Ansys Parametric Design Language
(MATLAB-APDL) codes as shown in Fig. 5. The lower and upper bounds of the
design variables are determined using established design approaches, as described
in reference [26]. This study utilizes a finite element model for accurately simu-
lating the behavior of an arch dam-reservoir system. The model incorporates 8-node
solid elements to model the dam body and an 8-node fluid element to represent the
reservoir. The model considers only the interaction between the dam and reservoir
and assumes a rigid foundation rock. The dam is modeled as a linear 3D structure,
while the fluid medium is assumed to be homogeneous, isotropic, irrotational, and
inviscid with linear compressibility. The fluid–structure interaction (FSI) effects are
considered between the reservoir and dam body, as well as the reservoir walls. The
number of nodes and elements may vary during the optimization process, depending
on changes in the dam’s dimensions and mesh generation requirements for each
analysis, [2].

Fig. 5 Arch dam finite element modeling

Optimum Design of Dam Structures Using Multi-objective Chaos Game … 137

4.2 Results and discussion

In addition to the proposed MOCGO, three well-studied multi-objective optimization

methods, namely the multi-objective particle swarm optimization (MoPSO) [27],
the non-dominated sorting genetic algorithm (NSGA-II) [28] and multi-objective
charged system search [2] were also selected to compare the performance of the new
algorithm. To ensure fairness in the comparison of these methods, twenty runs were
executed with different random initial populations, and the best-obtained solutions
were recorded as the final design. To ensure computational equivalence, the same
number of search iterations and population size were used for all algorithms.
The Pareto fronts of all algorithms are displayed in Fig. 6. These diagrams show the
Pareto fronts in which the concrete volume of the dam body ( fit 1 ) and a combination
of reversed frequencies (fit 2) are the main objectives of the problem. From Fig. 6,
it is clear that almost all points obtained by MOCGO are better than the previous
ones found in ref. [2] using MoPSO, NSGA-II, and MoCSS. Extreme values of the
Pareto solutions are provided in Table 1. In addition, Fig. 7 shows the corresponding
volume-frequency diagrams obtained by each method. According to this figure, the
new algorithm is capable of finding better results in many parts of the search space.
The improvement can be attributed to the fact that the new algorithm is capable of
exploring a larger area of the search space before settling on a solution. Furthermore,
this algorithm is more efficient in terms of time and resources because there is no
need to adjust any parameters.
To make a fair comparison of the results and make informed decisions based
on decision-maker criteria, the researchers considered five different scenarios, with
coefficients ranging from 0.1 to 0.9, [2]. The results of the decision-making process
are presented in Table 2. In MOCGO, the best volumes are produced for scenarios A,
B, and C, whereas in MoCSS, the best volumes are produced for scenarios D and E,
which is not so far off from the result of MOCGO. This suggests that both MOCGO
and MoCSS are able to effectively optimize for a range of decision-maker criteria.

5 Conclusion

An optimization algorithm based on the multi-objective chaos game has been used
for determining the optimal shape of double-curvature arch dams in this chapter. It
involves the development of a set of parallel-working APDL-MATLAB code modules
for modeling, analyzing, and obtaining the fitness function of the arch dam, as well
as interacting with MOCGO at each step of the optimization process. For the optimal
shape of double-curve arch dams, this approach is expected to provide effective and
robust solutions. The first objective function is the concrete volume, while the second
objective function is a combination of reversed natural frequencies of the dam. The
optimal design strategies of the dam are obtained by minimizing the two objective
functions simultaneously. The MOCGO technique is used to tackle the problem.
138 A. Jafari et al.

Fig. 6 Pareto front of all methods

Table 1 Comparison of the results obtained by selected algorithms

Optimization methods MoCSS, [2] NSGA-II, [2] MoPSO, [2] MOCGO
Obtained extreme values fit1 fit2 fit1 fit2 fit1 fit2 fit1 fit2
Vol (1e5) 1.099 3.340 1.099 3.288 1.099 3.280 1.110 2.928
Fr1 2.572 2.922 2.514 2.924 2.572 2.918 2.487 2.908
Fr2 3.104 3.305 3.101 3.304 3.104 3.302 3.080 3.275
Fr3 3.691 4.332 3.708 4.326 3.691 4.326 3.634 4.245
Fr4 4.433 5.196 4.421 5.167 4.433 5.196 4.384 5.056
Fr5 5.015 6.171 4.845 6.160 5.015 6.174 4.883 6.083
Fr6 5.689 6.664 5.572 6.675 5.689 6.638 5.506 6.556
Fr7 6.699 7.362 6.568 7.360 6.699 7.364 6.513 7.322
Fr8 7.008 7.716 6.943 7.712 7.008 7.713 6.936 7.677
Fr9 7.370 7.784 7.376 7.782 7.370 7.786 7.374 7.730
Fr10 7.460 7.885 7.404 7.878 7.460 7.886 7.340 7.869

The MOCGO approach can achieve a trade-off between the two objective functions.
In order to demonstrate the effectiveness of the proposed methodology, it has been
applied to the Morrow Point dam, and the results obtained by the MoCSS, MoPSO,
Optimum Design of Dam Structures Using Multi-objective Chaos Game … 139

a) b)

c) d)

e) f)

f)

Fig. 7 Pareto fronts of all methods; a fit1-Fr1, b fit1- Fr2, c fit1- Fr1, d fit1- Fr4, e fit1- Fr5, f fit1-
Fr6, g fit1- Fr7, h fit1- Fr8, i fit1- Fr9, j fit1- Fr10

and NSGA-II methodologies are compared with those obtained by the proposed
methodology. Findings suggest that MOCGO outperforms its competitors.
140 A. Jafari et al.

g) h)

i) j)

Fig. 7 (continued)
Table 2 Different possible scenarios corresponding solutions
Scenario Importance of criteria Possible priority Selected solution by MTDM
weights Vol Fr1 Fr2 Fr3 Fr4 Fr5 Fr6 Fr6 Fr7 Fr8 Fr10
MoCSS, [2]
A C1 >> C2 [0.9,0.1] 306,505.148 2.91 3.29 4.30 5.12 6.14 6.61 7.35 7.69 7.76 7.86
B C1 > C2 [0.7,0.3] 264,968.249 2.90 3.26 4.24 5.00 6.09 6.52 7.35 7.61 7.73 7.84
C C1 ≈ C2 [0.5,0.5] 198,943.337 2.84 3.21 4.10 4.80 5.97 6.12 7.31 7.34 7.72 7.72
D C1 < C2 [0.3,0.7] 150,206.901 2.72 3.15 3.93 4.62 5.65 5.80 6.98 7.31 7.55 7.67
E C1 << C2 [0.1,0.9] 109,942.578 2.571 3.10 3.69 4.43 5.01 5.69 6.70 7.00 7.36 7.46
NSGA-II, [2]
A C1 >> C2 [0.9,0.1] 306,734.745 2.918 3.292 4.302 5.122 6.145 6.603 7.360 7.690 7.759 7.870
B C1 > C2 [0.7,0.3] 269,714.224 2.904 3.273 4.248 5.005 6.101 6.481 7.353 7.620 7.740 7.830
C C1 ≈ C2 [0.5,0.5] 226,966.502 2.881 3.237 4.156 4.858 6.025 6.324 7.344 7.418 7.724 7.774
D C1 < C2 [0.3,0.7] 171,357.449 2.779 3.180 4.025 4.729 5.885 5.908 7.155 7.339 7.694 7.708
E C1 << C2 [0.1,0.9] 132,956.454 2.643 3.132 3.852 4.548 5.326 5.716 6.838 7.246 7.435 7.633
MoPSO, [2]
A C1 >> C2 [0.9,0.1] 314,937.556 2.920 3.296 4.311 5.156 6.165 6.616 7.362 7.701 7.778 7.877
B C1 > C2 [0.7,0.3] 271,749.572 2.902 3.275 4.259 5.040 6.109 6.503 7.356 7.636 7.748 7.848
(continued)
Optimum Design of Dam Structures Using Multi-objective Chaos Game …
141
Table 2 (continued)
142

Scenario Importance of criteria Possible priority Selected solution by MTDM

weights Vol Fr1 Fr2 Fr3 Fr4 Fr5 Fr6 Fr6 Fr7 Fr8 Fr10
C C1 ≈ C2 [0.5,0.5] 227,666.621 2.886 3.240 4.160 4.864 6.031 6.354 7.342 7.425 7.722 7.767
D C1 < C2 [0.3,0.7] 181,474.149 2.824 3.187 4.033 4.747 5.929 5.971 7.191 7.334 7.675 7.713
E C1 << C2 [0.1,0.9] 150,504.796 2.754 3.143 3.908 4.610 5.678 5.813 6.977 7.306 7.533 7.671
MOCGO
A C1 >> C2 [0.9,0.1] 271,186.374 2.905 3.282 4.278 5.064 6.116 6.536 7.364 7.674 7.750 7.856
B C1 > C2 [0.7,0.3] 229,625.894 2.886 3.256 4.205 4.947 6.086 6.358 7.377 7.479 7.753 7.788
C C1 ≈ C2 [0.5,0.5] 206,805.547 2.870 3.240 4.167 4.874 6.085 6.219 7.353 7.403 7.743 7.748
D C1 < C2 [0.3,0.7] 165,531.618 2.787 3.189 4.065 4.733 5.972 5.970 7.129 7.378 7.689 7.699
E C1 << C2 [0.1,0.9] 127,080.347 2.645 3.124 3.855 4.592 5.422 5.762 6.787 7.242 7.543 7.601
A. Jafari et al.
Optimum Design of Dam Structures Using Multi-objective Chaos Game … 143

References

1. Kaveh, A., Mahdavi, V.: Colliding bodies optimization for design of arch dams with frequency
limitations. Iran Univ. Sci. Technol. 4(4), 473–490 (2014)
2. Aalami, M.-T., Talatahari, S., Parsiavash, R.: Frequency-volume optimization of arch-dams
using charged system search method. arXiv preprint arXiv:2208.12920 (2022)
3. Gholizadeh, S., Seyedpoor, S.M.: Shape optimization of arch dams by metaheuristics and neural
networks for frequency constraints. Scientia Iranica 18(5), 1020–1027 (2011)
4. Kaveh, A., Ghaffarian, R.: Shape optimization of arch dams with frequency constraints by
enhanced charged system search algorithm and neural network. Int. J. Civ. Eng. 13(1), 102–111
(2015)
5. Zacchei, E., Molina, J.L.: Shape optimization of double-arch dams by using parameters
obtained through bayesian estimators. Iran. J. Sci. Technol., Trans. Civ. Eng. 43(4), 649–662
(2019)
6. Seyedpoor, S., Salajegheh, J., Salajegheh, E.: Shape optimal design of arch dams including
dam-water–foundation rock interaction using a grading strategy and approximation concepts.
Appl. Math. Model. 34(5), 1149–1163 (2010)
7. Zhang, X.-F., Li, S.-Y., Chen, Y.-L.: Optimization of geometric shape of Xiamen arch dam.
Adv. Eng. Softw. 40(2), 105–109 (2009)
8. Fengjie, T., Lahmer, T.: Shape optimization based design of arch-type dams under uncertainties.
Eng. Optim. 50(9), 1470–1482 (2018)
9. Takalloozadeh, M., Takalloozadeh, M., Ghaemian, M.: Shape optimization of concrete arch
dams considering abutment stability. Sci. Iran. 21(4), 1297–1308 (2014)
10. Wang, Y.-Q., et al.: Shape optimization of single-curvature arch dam based on sequential
kriging-genetic algorithm. Appl. Sci. 9(20), 4366 (2019)
11. Talatahari, S., Aalami, M.T., Parsiavash, R.: Risk-based arch dam optimization using hybrid
charged system search. ASCE-ASME J. Risk Uncertain. Eng. Syst., Part A: Civ. Eng. 4(2),
04018008 (2018)
12. Talatahari, S., M. Aalami, and R. Parsiavash, Multi-objective optimization of double curvature
arch dams subjected to seismic loading using charged system search. arXiv preprint arXiv:
2207.04366, 2022.
13. Khodadadi, N., et al.: Multi-objective crystal structure algorithm (MOCryStAl): introduction
and performance evaluation. IEEE Access 9, 117795–117812 (2021)
14. Jangir, P., et al.: MOMPA: Multi-objective marine predator algorithm for solving multi-
objective optimization problems. Evol. Intel. 16(1), 169–195 (2023)
15. Peng, J., et al.: Multi-objective integrated optimization study of prefabricated building projects
introducing sustainable levels. Sci. Rep. 13(1), 1–17 (2023)
16. Wang, Z., Zhang, N., Cheng, Q.: Multi-objective optimization-based reasonable finished state
in long-span cable-stayed bridge considering counterweights. In: Structures. Elsevier (2023)
17. Azizi, M., et al.: Multiobjective atomic orbital search (MOAOS) for global and engineering
design optimization. IEEE Access 10, 67727–67746 (2022)
18. Jadidi, O., Zolfaghari, S., Cavalieri, S.: A new normalized goal programming model for multi-
objective problems: a case of supplier selection and order allocation. Int. J. Prod. Econ. 148,
158–165 (2014)
19. Meng, Y., et al.: Multi-objective optimal dispatching of demand response-enabled microgrid
considering uncertainty of renewable energy generations based on two-level iterative strategy.
Energy Rep. 9, 1842–1858 (2023)
20. Wang, Q., et al.: Multi-objective optimization of IoT-based green building energy system using
binary metaheuristic algorithms. J. Build. Eng. 68, 106031 (2023)
21. Khodadadi, N., F. Soleimanian Gharehchopogh, and S. Mirjalili, MOAVOA: a new multi-
objective artificial vultures optimization algorithm. Neural Comput. Appl. 34(23), 20791–
20829 (2022)
22. Dhiman, G.: MOSHEPO: a hybrid multi-objective approach to solve economic load dispatch
and micro grid problems. Appl. Intell. 50(1), 119–137 (2020)
144 A. Jafari et al.

23. Talatahari, S., Azizi, M.: Chaos game optimization: a novel metaheuristic algorithm. Artif.
Intell. Rev. 54(2), 917–1004 (2021)
24. Nouhi, B., et al.: Multi-objective material generation algorithm (MOMGA) for optimization
purposes. IEEE Access 10, 107095–107115 (2022)
25. Khodadadi, N., Talatahari, S., Dadras Eslamlou, A.: MOTEO: a novel multi-objective thermal
exchange optimization algorithm for engineering problems. Soft Comput. 26(14), 6659–6684
(2022)
26. Varshney, R.S., CONCRETE DAMS. 1978, U.S.A.: Easton’s Books, Inc. (Mount Vernon, WA,
U.S.A.).
27. Coello, C.C., Lechuga, M.S.: MOPSO: A proposal for multiple objective particle swarm opti-
mization. In: Proceedings of the 2002 Congress on Evolutionary Computation. CEC’02 (Cat.
No. 02TH8600). IEEE (2002)
28. Deb, K., et al.: A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans. Evol.
Comput. 6(2), 182–197 (2002)
Machine Learning
The State of Art in Machine Learning
Applications in Civil Engineering

Yaren Aydin, Gebrail Bekdaş, Ümit Işıkdağ, and Sinan Melih Nigdeli

Abstract Machine learning (ML) is one of the methods used by the artificial intel-
ligence approach. Machine learning is used to teach machines how to handle data
more efficiently. The purpose of machine learning is to learn from the data. Thanks
to machine learning, a certain result can be reached without the need for an expert
on the subject. It has very common usage areas from the financial sector to e-mail
analysis. Machine learning is also widely used in civil engineering. In this study,
machine learning is explained in historical development, and general terms and the
studies that have been done are summarized.

Keywords Machine learning · Supervised learning · Unsupervised learning ·

Algorithms · Performance metrics

1 Introduction

In recent years, machine learning has become the basis of information technologies
and has become a part of life. Machine learning (ML) includes algorithms developed
to find the model and its parameters that best represent the available data. Learning
is the acquisition of knowledge, skills, and abilities from experience gained through
trial and error [1]. Artificial intelligence (AI) eliminates the programmer and the

Y. Aydin · G. Bekdaş (B) · S. M. Nigdeli

Department of Civil Engineering, Istanbul University - Cerrahpaşa, Avcılar, 34320 Istanbul,
Turkey
e-mail: [email protected]
Y. Aydin
e-mail: [email protected]
S. M. Nigdeli
e-mail: [email protected]
Ü. Işıkdağ
Department of Informatics, Mimar Sinan Fine Arts University, Şişli, 34427 Istanbul, Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 147
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_9
148 Y. Aydin et al.

Fig. 1 Artificial
intelligence, machine
learning, and deep learning
scope chart [2]

expert by imitating a certain part of the human brain. As a result of the progress
of artificial intelligence applications, machine learning, and its applications have
emerged. Although traditional programming can imitate human behavior to a certain
extent, when it comes to artificial intelligence, the concepts of machine learning and
deep learning (DL) often come to mind instead of traditional programming. The
relationship between these concepts is as follows (Fig. 1).
Artificial intelligence can be defined as the imitation of human cognitive features
such as learning and problem-solving by machines [3]. Artificial intelligence appli-
cations generally include machine learning and deep learning sub-fields. Machine
learning, known as predictive analytics or statistical learning, is an area of research
at the intersection of AI and computer science [4].
Machine learning is an approach that predicts that algorithms that automati-
cally extract information from existing data become better with experience and that
with less programmer intervention, improvement with experience increases. More
specifically, machine learning is concerned with automatically detecting meaningful
patterns in data and using the detected patterns for specific tasks [5].
The overall goal of machine learning is to recognize patterns in data that inform
how unseen problems are treated. For instance, in a complex self-driving car, a
computer learning about hazards must transform large amounts of data from sensors
to determine how it should control a car [6].
Machine learning is a multi-disciplinary field with a wide array of research areas
strengthening its presence. ML can easily overcome the complexity of real-world
problems. Machine learning can be applied to design and program open algorithms
with high-performance output, such as email spam filtering, traffic prediction, face
and shape detection, and medical diagnostics. As can be seen in Fig. 2, ML is used
in various fields.
Traditional programming is based on the computer execution of sequences of
commands entered by the programmer. Machine learning is based on learning. In
traditional programming, the inputs and the operations to be applied to these inputs
are coded one by one by the programmers as commands. But in machine learning,
the inputs and outputs are given by the programmers, and the operations to be done
to achieve the desired outputs of these applications are found by the computer and
The State of Art in Machine Learning Applications in Civil Engineering 149

Fig. 2 Application fields of machine learning [7]

Fig. 3 Generalized machine learning methodology [8]

recorded in a data structure called a model. The stages of developing a machine

learning model are given in Fig. 3.

2 Historical Development of Machine Learning

The terms artificial intelligence and machine learning have been studied and applied
by various expert groups for over 60 years. The mathematical foundations of machine
learning are based on algebra, statistics, and probability [9].
150 Y. Aydin et al.

The concept of neural networks was born in 1943 with the work of McCulloch,
a logician, and Pitts, a neuroscientist, at the earliest [10]. They created the first
mathematical model of a neural network. In 1949, Donald Hebb published The
Behavior Organization, which introduced theories about the interaction between
neurons so important to advancing machine learning [11]. In 1950 Alan Turning
created the “Turning Test” to test the intelligence of machines. To pass the turning
test, the machine must be able to convince a human that it is talking to a human and
not a machine. In 1952, Samuel created a machine that can play and self-learning
Chess was demonstrated to the public on television [12]. And ML first coined with
Arthur Samuel [13] in 1959 pioneered various machine learning techniques [14].
In 1967, Cover and Hart proposed the Nearest Neighbour Algorithm, which can be
considered the beginning of Pattern Recognition [15]. In 1979, Stanford students
invented the Stanford Cart, a remote-controlled car that could move autonomously
and sense and avoid objects [16].
Machine learning has advanced further after 1990 and has become the most
successful method used by the artificial intelligence approach and present technology
terms [17]. In 2006, Geoffery Hinton coined the term deep learning to describe a
new neural network architecture that uses multiple layers of neurons for learning. In
2012, Google’s Jeff Dean developed GoogleBrain, a deep neural network that recog-
nizes patterns in video and images. In 2017, Google proposed smartphones based on
ML and DL algorithms, such as Google Lens and Google Nexus. Apple presented
the Home Pod, an interactive machine-learning device [9]. Machine learning has
become an important method in decision-making and the most successful subfield
of AI. Thanks to ML, large data sets can be easily processed and a decision-making
mechanism can be provided between these sets. ML technology is developing and
new algorithms are written by researchers in many fields from medicine to industry,
from technology to art [18].

3 Machine Learning Types

Training is the process of automatic model building and training data is the data used
for training purposes. Thanks to the trained model, skills are gained regarding the
mapping of input variables to output and predictions can be made for new inputs. [19].
The model is trained and tested with the test data set and the predictions made are
evaluated [8]. If sufficient accuracy and evaluation metrics are obtained, the model is
considered complete. Otherwise, the model is retrained by evaluating the metrics and
using options such as adding/removing variables, increasing the data set, or reducing
the size. After this stage, the model is evaluated again and the steps are repeated until
the demanded results are obtained.
Machine learning is a highly multidisciplinary field that builds on ideas from
statistics, computer science, engineering, and crowded scientific also mathematical
disciplines [20]. Many applications such as fraud detection, text, and document classi-
fication, speech processing applications, credit card, computer vision, network attack
The State of Art in Machine Learning Applications in Civil Engineering 151

Fig. 4 Types of machine learning (adapted from [24])

detection, and self-driving car applications are performed with machine learning. The
problems are tackled using machine learning techniques [21].
Machine learning systematically applies algorithms to synthesize fundamental
relationships between data and information [22]. According to the learning types,
machine learning has four groups supervised learning, unsupervised learning,
reinforcement learning, and semi-supervised as unsupervised learning, supervised
learning, semi-supervised learning, and reinforcement learning [23]. Machine
learning types are given in detail in Fig. 4.

3.1 Supervised Learning

The supervised learning model works based on the input–output example. A model
is developed with samples consisting of input and output parameters. While training
the models, labeled data is used. The aim of this learning is that the machine can learn
the path to a known result by experiencing a particular result [25]. After establishing
the relationship between the input and output parameters, the model is tested and
its success in estimating the data it did not know before is checked. The supervised
learning workflow is shown in Fig. 5.
Supervised learning has applications such as predictive analysis based on regres-
sion and classification, natural language processing, automatic image classification,
and sentiment analysis [27].
152 Y. Aydin et al.

Fig. 5 Supervised learning workflow [26]

3.1.1 Performance Evaluation of Supervised Learning

Cross Validation

Cross-validation (CV) is a data resampling method used to assess the true prediction
error of the model and to adjust its parameters [28]. Cross-validation is used to deal
with generalization and overfitting errors. The current dataset is split into training
and testing, which will be used in the learning process and performance evaluation,
respectively [29].
CV has purposes such as estimating the performance of the learned model using
a single algorithm from the available data and comparing the performance of two or
more different algorithms [30]. The various CV charts differ in the way they split
the sample data. The most widely used method is k-fold cross-validation [31].

K-Fold Cross Validation

K-fold cross-validation decomposes the folds into layers so that the class distribution
of records in each layer is approximately the same as the initial data. It is usually
recommended that k should be 10 for relatively low bias and variance. [32]. In the
first iteration in Fig. 6, 1 box in the blue part is tested and 9 boxes are trained. Thus,
the first iteration ends and the success rate is achieved. The same process is applied
to the other iterations so that all of the data enters both training and testing.

Stratified K-Fold Cross Validation

In this cross-validation method, folds are created with a criterion. In general, the
criterion is to create folds in the form of an equal proportion of the result variable.
For example, in Fig. 7, 3 diabetics and 7 non-diabetics in every 10 rows were selected
as the criteria for diabetes. Thus, validation is created by creating a constancy in the
resulting output. The proportions of the class to be classified are equally distributed
on each floor.
The State of Art in Machine Learning Applications in Civil Engineering 153

Fig. 6 10-Fold cross-validation [33]

Fig. 7 Stratified 5-fold cross-validation [34]

154 Y. Aydin et al.

(a) (b)

Fig. 8 a Classification, b regression [38]

Leave One Out Cross Validation

Leave-Out-One-Out (LOO) CV is straightforward as the training dataset is created
by including all individuals except one, while the test set contains the individual
not included. Therefore, there are different training and test sets for each individual
in the whole dataset. This method wastes minimum data as only one individual is
removed from the training set [35].

Train Test Split

The most accurate machine learning algorithm is tried to be selected by making basic
modeling experiments on the training dataset. It consists of the observations that are
sampled the most over this data set [36]. The data set is divided into training and test
sets (usually with ratios of 0.7/0.3) and the success of the model is measured. Using
the training data, the best values of the control parameters of the various machine
learning models used are determined. While the data set is distributed as training
and test data set in certain ratios, irregularities that may occur negatively affect the
performance of the model. To solve this problem, Stone [37] developed the k-fold
cross-validation method in 1974. When more data is needed, the data is divided into
two training data and test data. To test how well it learns, the machine is presented
with test data, which is 30% of the data it has never seen before. Using accuracy
measurement metrics, the machine’s predicted value is compared with the actual
value, and the result is how many percent correct guesses are obtained.
The two most common types of supervised learning are classification and regres-
sion (Fig. 8). In classification, entries are categorized into a particular class. In regres-
sion, the label is continuous and the relationships between the variables are looked at
[25]. The main difference between classification and regression is that their outputs
are categorical and numerical respectively.
The State of Art in Machine Learning Applications in Civil Engineering 155

In traditional supervised learning, learning is performed from a large number of

training examples. Each training example has a label indicating the desired output for
the event described in the example. In the classification problem, the label indicates
the category in which the relevant sample is included, while in the regression; refers
to a real-valued output such as temperature, altitude, or price [39].

3.2 Unsupervised Learning

Unsupervised learning is referred to as descriptive or undirected classification. Unsu-

pervised learning is learning where data without labeled output is available. [40].
Some of the features that algorithms learn come from the data [41]. Unsupervised
learning discovers hidden patterns in the data [42]. Since no system controls the
learning according to the entered data, the machine determines the output itself [25].
Let X = (x 1 , …, x n ), be a set of n samples (or points), with x i ∈ X for all i ∈ [n]
T
:= {1,2, …, n}. It is often convenient to define (n × d) − matrix X = (x iT )i∈[n]
containing data points as rows. In this case, the purpose of unsupervised learning is
to find interesting structures in the X data [43].
The most common types of unsupervised learning are clustering and dimension-
ality reduction are shown in Fig. 9.

3.2.1 Clustering

Clustering is the grouping of observations according to similar characteristics.

Instead of classifying or predicting the value of a target variable, clustering algo-
rithms attempt to divide the entire dataset into relatively homogeneous subgroups.

(a) (b)

Fig. 9 a Clustering, b dimensionality reduction [38]

156 Y. Aydin et al.

Thus, similarity to observations within the cluster is increased while similarity to

observations outside the cluster is minimized [44].
Clustering is more difficult than supervised classification as there are no labels
attached to the patterns in clustering. In supervised classification, labels allow data
objects to be grouped as a whole. Since there are no labels in clustering, it is difficult
to determine to which group a model belongs [45].

3.2.2 Dimensionality Reduction

Dimension reduction is the task of reducing the number of inputs. Dimension reduc-
tion is the task of reducing the number of inputs. Dimension reduction techniques
facilitate the utilization of data for successful learning. This pre-processing is done
to make the data more ready for use before applying it to another algorithm [46].

3.3 Semi-Supervised Learning

A semi-supervised learning model is a combination of supervised and unsupervised

models. These areas of machine learning can also be fruitful with data mining,
where unmarked data that is already retrieved data tagged is a process. The standard
methods of supervised machine learning algorithms are for the labeled datasets, and
each record contains result information [47]. Only a subset of the training data is
labeled [48]. This learning improves learning performance by using a limited amount
of labeled data and a large amount of unlabeled data [49].

3.4 Reinforcement Learning

The training data is only given as feedback in the form of reward and punishment
to an AI agent with interaction. This feedback improves performance on the learned
task [50]. It is based on trial and error and has no output [51]. In this learning, the
goal of the model is to win the most rewards [25]. The learning system of the robot
is the reward-punishment system it receives according to the situation it encounters.
The schematic form of reinforcement learning is in Fig. 10.
There are different algorithms already developed for reinforcement learning prob-
lems. The most well-known use of reinforcement learning is models developed for
playing games and controlling robots [3].
The State of Art in Machine Learning Applications in Civil Engineering 157

Fig. 10 Reinforcement learning [42]

4 Machine Learning Algorithms

4.1 Random Forest (RF)

Random Forest was proposed by Leo Breiman in 1997 [52]. Random forest is based
on a supervised learning technique. This algorithm is an ensemble learning method.
It can be used both in classification and regression. The parameters of the random
forest are not complex. It is based on the logic of using a combination of many tree
estimators instead of using a single classifier tree to eliminate problem complexity and
improve model performance [53]. With more trees, it shows better performance [54].
As a result of the combination of many tree estimators, some trees may make correct
predictions while others may produce incorrect predictions. Although this may seem
like a disadvantage, all trees can predict the correct output when combined. RF has
decision trees that examine subsets of the given dataset. It averages the input dataset
to improve the prediction accuracy [55]. It is an extension of the bagging predictor
algorithm [56]. The schematic representation of the RF algorithm is given in Fig. 11.

4.2 Decision Tree (DT)

The best performance of the decision tree in speed and frequency has made it the most
widely used machine learning technique by researchers in classification. Decision
trees usually operate in two stages, the tree creation, and classification stages. It was
during classification the rules for classification are applied from the tree. The process
of deciding root to leaf followed the branches [42]. The diagram in Fig. 12 below
shows decisions and results in a tree format of a decision tree.
158 Y. Aydin et al.

Fig. 11 Random forest example [42]

Fig. 12 Structure of a decision tree [57]

4.3 Naive Bayes (NB)

This classification algorithm supports Bayes’ theorem and assumes that features
within classes are irrelevant [58]. It is based on the fact that if the class is known, the
properties can be predicted correctly and if the class is not known, the properties given
by the Bayes rule can be predicted for the given class [42]. While it has advantages
such as computing the training fast and showing success, it also has the disadvantage
that the success of the results of analysis depends on the records [42].
The State of Art in Machine Learning Applications in Civil Engineering 159

Fig. 13 Support vector

machine [61]

4.4 Support Vector Machine (SVM)

Support vector machine can be used for classification and regression problems [58].
SVM is used in many fields, including bioinformatics and the analysis of chemical
data [59]. The SVM has good sensitivity even with a small number of samples
and performs well [60]. SVM classifies data points by finding a hyperplane in n-
dimensional space. [54]. The margin is the largest distance between data points and
the goal is to find the plane with the maximized margin. A margin among the classes
was drawn (Fig. 13). Edges are also considered to maximize the distance between
the edge and the class, minimizing classification errors [25].

4.5 K-Nearest Neighbors (KNN)

The KNN algorithm was first mentioned by Fix and Hodges in 1951 and Cover
and Hart advanced the algorithm in 1967 [62]. KNN algorithm can be easily used in
classification and regression problems. The disadvantage is that the success decreases
when the data size increases [42]. The algorithm predicts the class of its unlabeled
neighbors using its labeled neighbors. The k parameter and the distance affect the
performance of the model. [25]. The KNN algorithm works by classifying each new
sample from its K-nearest neighbor according to the majority tag. This method works
well at a reasonable distance and with a small number of data points in the training
set [31].
A simple KNN example on a two-dimensional plane is shown in Fig. 14. In the
example, the nearest neighbors for K = 3 and K = 8 of a new data point in a two-class
dataset are specified. When the majority voting method is applied, it is seen that the
new sample is included in Class 2 for both cases [42].
160 Y. Aydin et al.

Fig. 14 K-nearest neighbor

algorithm example
demonstration [42]

4.6 Logistic Regression (LR)

Logistic regression is used to get the probability ratio in the occurrence of

various illustrative variables. It allows various illustrative variables to be analyzed
concurrently, meantime decreasing the effect of confusion factors [63].

5 Evaluation of Classifier Performances

Confusion matrix, performance metrics, and AUC-ROC curve approaches are used
to evaluate classification methods [64]. The confusion matrix allows the performance
of the classification to be evaluated. Performance metrics are determined according
to the correct and incorrect numbers of class predictions.

5.1 Confusion Matrix

The confusion matrix is used to determine which of the methods used to measure
the quality of algorithms used in machine learning classification studies gives better
results. By using confusion matrices, the relationships between the reference data,
that is, the ground truth data, which are accepted as correct, and the automatic classi-
fication results corresponding to these data can be categorically compared with each
other [65]. The confusion matrix is one of the common methods used to measure
machine learning model performance. The success of the model and the errors caused
by the classifier are shown in the form of a table in matrix format. This matrix is a
2-dimensional table. One of the dimensions shows the actual values, while the other
The State of Art in Machine Learning Applications in Civil Engineering 161

Fig. 15 Confusion matrix

shows the predicted values. Although it is mostly used in binary classification prob-
lems, a Confusion Matrix can also be created in multi-class classification problems.
It is a good measure by which models can explain the overlap in class properties and
which classes are most easily confused. The Confusion Matrix for binary classifi-
cation is shown in Fig. 15. The confusion matrix has actual values in the rows and
predicted values in the columns.
TP (True Positive) indicates the case of correctly predicting positive observations
with a positive true class value. FP (False Positive) indicates the case when obser-
vations with negative true class value are correctly predicted as positive. TN (True
Negative) indicates the case when observations with a negative true class value are
correctly predicted as negative. FN (False Negative) shows that observations with
positive true class values are incorrectly evaluated as negative as a result of the
prediction [66]. The sum of these four values gives the total number of samples to
be classified.

5.2 Performance Metrics

5.2.1 Accuracy

The most common and simple way to measure model performance is model accuracy
[67]. The confusion matrix is used when calculating accuracy.

TP +TN
Accuracy = (1)
T P + T N + FP + FN
162 Y. Aydin et al.

As can be seen from Eq. 1, it is a measure of how often the classifier makes
a correct prediction. In analyses using classification algorithms, it is not sufficient
to examine only the accuracy rate as a success criterion. It may not give correct
information, especially in unbalanced datasets.

5.2.2 Recall (True Positive Rate) (TPR)

The recall metric is an indicator of falsely predicted true positives. In a multiclass

and unbalanced classification problem, it is the ratio of true positives to the sum of
true positives and false negatives [68]. The calculation of the recall value is shown
in Eq. 2.

TP
Recall = (2)
T P + FN

5.2.3 Specificity (False Positive Rate) (FPR)

The specificity corresponds to the proportion of negative data points that are consid-
ered false positives relative to all negative data points [69]. The specificity is obtained
by dividing the false positive by the sum of the false positive and true negative.

FP
Speci f icit y = (3)
FP + T N

5.2.4 Precision

The Precision performance metric, also known as a positive predictive value, is a

measure of how accurately predicted from all classes. It is desired and expected to
be as high as possible. This metric determines how many of the positive estimates
are truly positive. Precision value is especially important when False Positive (FP)
prediction is costly. Precision is the ratio of true positives to the sum of true positives
and false positives [68]. The calculation of the precision value is shown in Eq. 3.
Precision indicates how many of all predicted positive examples were predicted
correctly.

TP
Pr ecision = (4)
T P + FP
The State of Art in Machine Learning Applications in Civil Engineering 163

5.2.5 F1 Score

More accurate results are obtained when recall and precision metrics are evaluated
together. For this purpose, the F1 score is defined. The F1 score is calculated by
taking the harmonic mean of recall and precision [67].

Pr ecision ∗ Recall
F1 Scor e = 2 ∗ (5)
Pr ecision + Recall

5.2.6 Receiver Operating Curve (ROC)

The ROC curve is a graph of positive false rate and recall of tests. The ROC curve is
often a form of performance evaluation for different kinds of classification problems.
The AUC values found in the ROC curve determine the overall accuracy values. It
shows how accurately the models used in the ROC curve can classify. When it is
desired to evaluate the methods used in the ROC curve, how close or 1 the AUC
values of the methods used in the model are, these values help to determine the most
suitable model [67].
A graph is created with TPR on the y-axis and FPR on the x-axis. The curve
passing through the graph in Fig. 16 and the area under the curve are important.
The area under this curve is called Area Under Curve or Are Under the ROC Curve.
Success can be calculated from the size of this area. Based on this, the matrix works.
The value of the ROC curve with a perfectly straight line is 1.0. If the field is cut in
the middle with a 45-degree line (50% success), it is a very unsuccessful prediction.

Fig.16 ROC and AUC [66]

164 Y. Aydin et al.

As advantages of the ROC curve, the ROC curve shows all possible cut-off points,
and the tests can be visualized in a graph according to the ROC curve. For these
reasons the evaluation is facilitated, in some cases sensitivity may come to the fore
or selectivity may be preferred to sensitivity. Even when these two situations are
encountered, ROC curves find suitable cut-off points [71].

6 Evaluation of Regressor Performances

As for the problem classification (two-class, multi-class) problem, it is mentioned

above that it is used in the measurement of success (precision, recall, F1, error matrix,
etc.). is working.

6.1 R2 (Coefficient of Determination)

R2 is a measure of fitness for linear regression models. The R2 (coefficient of deter-

mination) shows how close the data are to the fitted regression line [69]. R2 refers to
how successfully the model explains the total change and is the power of the analysis.
A high R2 indicates a good regression model fit.
∑n
(yi − xi )2 E x plained V ariation
R = 1 − ∑i=1
2
n =1− (6)
i=1 (yi − x i )
2 T otal V ariation

xi and yi are the predicted and actual values for the ith observation, respectively. x i
is the average of predicted values, and there is n number of observations [72].

6.2 Mean Absolute Error (MAE)

The absolute error is the difference between the actual and predicted value. The MAE
is the average of the absolute values of each difference between the actual value and
the predicted value for the sample. A lower MAE value means more success [69].
∑n
i=1 |yi − xi |
M AE = (7)
n
xi and yi are the predicted value and the actual value for the ith observation,
respectively.
The State of Art in Machine Learning Applications in Civil Engineering 165

6.3 Mean Squared Error (MSE)

Mean Square Error (MSE) is commonly used and is the mean square loss per sample
over the entire dataset. To calculate the MSE, all the squared losses for individual
samples are added up and then divided by the number of samples. A lower value
indicates better accuracy [69].

∑n
(yi − xi )2
MSE = (8)
i=1
n

6.4 Root Mean Squared Error (RMSE)

Root Mean Squared Error is the average squared root error between actual observation
and output [51]. The lower the measures, the more accurate the prediction results
[70].
⌜
| n
|∑ (yi − xi )2
RMSE = √ (9)
i=1
n

6.5 Mean Absolute Percentage Error (MAPE)

Mean absolute percentage error (MAPE) is a measure of the prediction accuracy of

the method used for prediction [74].
∑n
1 i=1 |yi − xi |
M AP E = × 100 (10)
n yi

7 Overfitting and Underfitting

Fitting the model to the data as undersaturated or oversaturated is important because

neither an under-fitting nor over-fitting model is desired for a successful machine
learning model.
In the visuals in Fig. 17, for example, those who have diabetes are shown with a
cross and those who do not are shown with a circle. In such a case, as Fig. 17a, when
166 Y. Aydin et al.

(a) Under-fitting, (b) Appropriate-fitting (c) Over-fitting

Fig. 17 Schematic illustration of three models for classification [75]

the curve and these are separated and the equation of this curve is a sufficient model
to make this distinction, it cannot make a very accurate classification when the shape
of the curve is correct. The reason for this may be an under-trained, insufficiently
trained model, the wrong algorithm chosen, etc. There is a model in this data that is
not well-fitted. The equation of this line cannot be predicted very well.
This problem also arises when the training data set is too small or when the
population data cannot be represented. A bad model cannot make good predictions
because it does not fit the training data well. This means that predictions using unseen
data are weak, as people unfamiliar with the training dataset are perceived as strangers
[35].
The Fig. 17b model is well-trained, although it throws a few errors. If the equation
of this curve is known, a correct distinction can be made.
Overfitting (Fig. 17c) occurs when the machine learning model learns the training
dataset but does not perform on a new dataset. It occurs when it learns the training
dataset very well but performs poorly on datasets it has not seen. Due to the overfitting
problem, success with new data is not achieved and the generalizability of the model
is hampered [35].

8 Machine Learning in Civil Engineering

There are many studies of machine learning in civil engineering. The outcomes
of these studies are beneficial for researchers in civil engineering for the timely
and effective evaluation of engineering problems. It is frequently used in various
sub-branches of civil engineering. In this section, recent work in the literature on
machine learning in civil engineering is examined. Although machine learning has
only recently been introduced in civil engineering, the results obtained are quite
good.
The State of Art in Machine Learning Applications in Civil Engineering 167

8.1 Structural Engineering

In structural engineering, computer software using methods such as the finite element
method is generally required for the calculation of design parameters. However,
analyses performed with this computer software require large amounts of memory
and time. By utilizing the advantages of machine learning, the results are reached in a
fast and reliable way. There are studies in structural engineering such as confinement
coefficients, compressive strength, carbonation, chloride diffusion, failure mode,
lateral drifts, long-term deflections, behavior under seismic effects, flexural strength,
axial capacity, structural damage, shear stress and plastic viscosity, optimum design,
moment capacity and ductility of different materials [76–92]. The use of machine
learning in structural engineering is listed in Table 1.

8.2 Geotechnical Engineering

Since the experiments in geotechnical engineering are mostly carried out in the field
and the laboratory, it is necessary to use approaches to determine the parameters in
a short time and close to reality. In addition, since geotechnical parameters depend
on many variables (environment, dynamic conditions, etc.), their calculation is more
difficult with traditional methods. The use of ML in geotechnical engineering has
become widespread day by day. There are studies on soil classification, correlation of
parameters, pile-bearing capacity, optimum design of retaining walls, leak-off pres-
sure, permeability coefficient, optimum drift capacity of retaining walls, compaction
quality, and soil liquefaction [93–103]. Table 2 lists some of the studies carried out
with machine learning.

8.3 Hydraulic Engineering

As a result of global warming and climate change, it is difficult to predict hydrological

data such as precipitation, floods, and rainfall in terms of time, cost, and labor. For this
reason, machine learning is necessary and has recently made significant progress in
hydraulic engineering. These issues such as dam flow, structural dam behavior, water
demand, reservoir water balance, water quality, and groundwater pollution have been
studied by machine learning [104–111]. Thanks to machine learning, structures that
require high costs (such as dams) can be designed more economically and safely.
The studies on the mentioned topics are listed in Table 3.
168 Y. Aydin et al.

Table 1 Machine learning studies in structural engineering

Study summary Authors Year
M5 model trees and MT were compared to predict the Naeej, M., Bali, M., Naeej, 2013
confinement coefficient in rectangular columns [76] M. R., & Amiri, J. V
Predicting the compressive strength of Chou, J. S., Tsai, C. F., 2014
high-performance concrete [77] Pham, A. D., & Lu, Y. H
Developing a machine learning-based carbonation Taffese, W. Z., Sistonen, E., 2015
prediction model [78] & Puttonen, J
Using a hybrid machine learning model for predicting Vu & Hoang 2016
the ultimate punching shear capacity of fiber reinforced
polymer reinforced slabs [79]
Building machine learning models for chloride diffusion Hoang, N. D., Chen, C. T., & 2017
prediction in cement mortar [80] Liao, K. W
Using machine learning techniques to identify the Mangalathu & Jeon 2018
classification of failure mode and the prediction of
beam-column joints [81]
Applying machine learning algorithms to classify the Siam, A., Ezzeldin, M., & 2019
walls and predict their lateral drifts [82] El-Dakhakhni, W
Proposing ML model to predict long-term deflections in Pham, A. D., Ngo, N. T., & 2020
reinforced concrete structures [83] Nguyen, T. K
Predicting the behavior of ductile reinforced concrete Hwang, S. H., Mangalathu, 2021
frame buildings under seismic effects [84] S., Shin, J., & Jeon, J. S
Developing a machine learning algorithm to predict the Kang, M. C., Yoo, D. Y., & 2021
compressive and flexural strengths of steel Gupta, R
fiber-reinforced concrete [85]
Predicting the axial capacity of fiber-reinforced polymer Cakiroglu, C., Islam, K., 2022
reinforced concrete columns [86] Bekdaş, G., Kim, S., &
Geem, Z. W
Combining optimization and machine learning tools to Yucel, M., Nigdeli, S. M., & 2022
minimize potential structural damage through optimal Bekdaş, G
modeling of tuned mass dampers (TMDs) [87]
Predicting shear stress and plastic viscosity of Cakiroglu, C., Bekdaş, G., 2022
self-compacting concrete utilizing ensemble machine Kim, S., & Geem, Z. W
learning techniques [88]
Designing optimum (minimizing total cost of wall Bekdaş, G., Cakiroglu, C., 2022
thickness) cylindrical walls utilizing ensemble learning Islam, K., Kim, S., & Geem,
methods [89] Z. W
Developing machine learning models for predicting the Cakiroglu, C., & Bekdaş, G 2023
shear strength of recycled aggregate concrete beams [90]
Using optimization and machine learning to predict the Aydın, Y., Bekdaş, G., 2023
design variables of an eco-friendly concrete column [91] Nigdeli, S. M., Isıkdağ, Ü.,
Kim, S., & Geem, Z. W
(continued)
The State of Art in Machine Learning Applications in Civil Engineering 169

Table 1 (continued)
Study summary Authors Year
Integrating machine learning method to evaluate the Guo, B., Lin, X., Wu, Y., & 2023
moment capacity and ductility of the compression Zhang, L
yielding beam with T section [92]

Table 2 Machine learning studies in geotechnical engineering

Study summary Authors Year
Developing a Unified Soil Classification System Gambill, D. R., Wall, W. A., 2016
prediction model via Random Forest algorithm [93] Fulton, A. J., & Howard, H. R
Correlating parameters such as density, index of Puri, N., Prasad, H. D., & 2018
compression, etc. with soil parameters determined in the Jain, A
laboratory and the field via machine learning models [94]
Predicting the pile-bearing capacity via gaussian process Momeni, E., Dowlatshahi, M. 2020
regression [95] B., Omidinasab, F., Maizir,
H., & Armaghani, D. J
Interpreting Cone Penetration Tests data with different Rauter, S., & Tschuchnigg, F 2021
machine learning models [96]
Predicting leak-off pressure with machine learning Choi, J. C., Liu, Z., Lacasse, 2021
algorithms [97] S., & Skurtveit, E
Developing predictive models for optimal dimensions of Bekdaş, G., Cakiroglu, C., 2022
reinforced concrete retaining walls [98] Kim, S., & Geem, Z. W
Predicting and investigating the permeability coefficient Tran, V. Q 2022
of soil with a machine learning algorithm [99]
Predicting drift capacity of RC walls via machine Aladsani, M. A., Burton, H., 2022
learning algorithm [100] Abdullah, S. A., & Wallace,
J. W
Compaction quality evaluation of subgrade using Wang, X., Dong, X., Zhang, 2022
machine learning model [101] Z., Zhang, J., Ma, G., &
Yang, X
Soil classification via machine learning algorithms [102] Aydın, Y., Işıkdağ, Ü., 2023
Bekdaş, G., Nigdeli, S. M., &
Geem, Z. W
Predicting soil liquefaction potential using different Vasegh, M., 2023
machine learning models [103] Dehghanbanadaki, A., &
Motamedi, S

8.4 Construction Management

While the construction of a building in the construction sector is still at the idea stage,
the correct establishment of the time and cost model will prevent many problems in
advance. This is possible with machine learning. Construction management issues
such as delay risk, construction cost, project characteristics, injury severity, risk anal-
ysis, construction requirements, and leading indicators were examined with machine
170 Y. Aydin et al.

Table 3 Machine learning studies in hydraulic engineering

Study summary Authors Year
Predicting the amount of dam inflow by combining Hong, J., Lee, S., Bae, J. H., 2020
machine learning models [104] Lee, J., Park, W. J., Lee, D., …
& Lim, K. J
Predicting and Interpreting structural dam behavior Mata, J., Salazar, F., Barateiro, 2021
with machine learning models [105] J., & Antunes, A
Predicting water demand utilizing machine learning Shuang, Q., & Zhao, R 2021
models [106]
Simulating reservoir water balance using a machine Latif, S. D., Ahmed, A. N., 2021
learning algorithm [107] Sherif, M., Sefelnasr, A., &
El-Shafie, A
Predicting the water quality classification via machine Malek, N. H. A., Wan Yaacob, 2022
learning models [108] W. F., Md Nasir, S. A., &
Shaadan, N
Detection of flood with machine learning [109] Tanim, A. H., McRae, C. B., 2022
Tavakol-Davani, H., &
Goharian, E
Integrating Bayesian and machine learning approaches An, Y., Zhang, Y., & Yan, X 2022
to recognition groundwater contamination source
parameters [110]
Predicting rainfall utilizing machine learning models Baljon, M., & Sharma, S. K 2023
[111]

learning models [112–119]. Table 4 lists the studies carried out with machine learning
in the construction business.

8.5 Transportation Engineering

Forecasts such as road traffic accidents, traffic flow, and road pavement assessment
involve complex operations. Machine learning is useful to create an efficient forecast
even in complex operations. Road pavement, energy consumption, classifying vehi-
cles, road traffic accidents, vehicular traffic flow, pavement condition, and tempera-
ture changes in asphalt mixtures, which are the fields of transportation engineering,
were examined by applying machine learning. [120–127]. Table 5 lists the topics
related to the use of machine learning in transport engineering.
The State of Art in Machine Learning Applications in Civil Engineering 171

Table 4 Machine learning studies in construction management

Study summary Authors Year
Predicting project delay risk using machine learning Gondia, A., Siam, A., 2020
algorithms [112] El-Dakhakhni, W., &
Nassar, A. H
Predicting the construction price index by applying Nguyen, P. T 2021
machine learning algorithms [113]
Predicting delay in construction utilizing decision tree and Erzaij, K. R., Burhan, A. 2021
naïve Bayesian classification algorithm [114] M., Hatem, W. A., & Ali,
R. H
Classifying project features using machine learning [115] Fan, C. L 2022
Predicting injury severity using machine learning [116] Gondia, A., Ezzeldin, M., 2022
& El-Dakhakhni, W
Analyzing risk with hybrid machine learning [117] Fitzsimmons, J. P., Lu, R., 2022
Hong, Y., & Brilakis, I
Predicting requirements in construction projects with Golabchi, H., & Hammad, 2023
machine learning [118] A
Producing predictions for leading indicators in different Gondia, A., Moussa, A., 2023
working areas and all project life cycles with machine Ezzeldin, M., &
learning [119] El-Dakhakhni, W

Table 5 Machine learning studies in transportation engineering

Study summary Authors Year
Predicting the properties of asphalt concrete Katanalp, B. Y., Yildirim, Z. B., 2019
with waste coal admixture [120] Karacasu, M., & Ibrikci, T. (2019)
Predicting highway energy consumption [121] Cansız, Ö. F., Ünsalan, K., & 2020
Erginer, İ. (2020)
Classifying vehicles with machine learning Li, C., & Xu, P 2021
[122]
Assessing prediction model designs for road Bokaba, T., Doorsamy, W., & Paul, 2022
traffic accidents [123] B. S
Predicting vehicular traffic flow [124] Olayode, I. O., Severino, A., 2022
Campisi, T., & Tartibu, L. K
Evaluating pavement condition [125] Sholevar, N., Golroo, A., & Esfahani, 2022
S. R
Modeling road traffic accidents [126] Megnidio-Tchoukouegno, M., & 2023
Adedeji, J. A
Evaluating temperature changes in asphalt Wang, X., Pan, P., & Li, J 2023
mixtures [127]
172 Y. Aydin et al.

9 Summary and Conclusions

Machine learning is a trendy tool for solving all kinds of problems and it is not a
method that is only useful in a specific area. In this chapter, the state of the art in
machine learning applications of civil engineering is summarized. After the intro-
duction to machine learning, a short historical summary of ML is given. In the third
section of this chapter, the types of ML such as supervised learning, unsupervised
learning, semi-supervised learning, and reinforcement learning are summarized. In
the fourth section, several potentially used ML algorithms are mentioned. Then,
the evaluation methods for ML are shortly given. After the ML methods are briefly
mentioned, the state of art in the application of civil engineering is used for different
branches of civil engineering. 17 papers that were done between 2013 and 2023
are listed for structural engineering. It is possible to see 11 studies in geotech-
nical engineering and 8 recent papers about hydraulics engineering are reviewed.
8 recent construction management papers are given in the chapter. Finally, 8 papers
about transportation engineering are listed. According to the research, the number
of studies in subbranches of civil engineering has a great increase in numbers to
estimate optimum results or complex solutions easily.

References

1. https://sozluk.gov.tr/ [Visit date: 20 February 2023].

2. Kaluarachchi, T., Reis, A., Nanayakkara, S.: A review of recent deep learning approaches in
human-centered machine learning. Sensors 21(7), 2514 (2021)
3. Russell, S. J., Norvig, P., & Davis, E. (2010). Artificial intelligence: a modern approach. 3rd
ed. Upper Saddle River, NJ, Prentice Hall.
4. Alpaydin, E. (2016). Machine learning: the new AI. MIT press.
5. So, A. M. C. (2020). Technical elements of machine learning for intellectual property
law. Artificial Intelligence and Intellectual Property.
6. Carleo, G., Cirac, I., Cranmer, K., Daudet, L., Schuld, M., Tishby, N., ... & Zdeborová,
L. (2019). Machine learning and the physical sciences. Reviews of Modern Physics, 91(4),
045002.
7. https://www.javatpoint.com/applications-of-machine-learning. [Visit date: 21 February
2023].
8. Ahmad, Z., Shahid Khan, A., Wai Shiang, C., Abdullah, J., & Ahmad, F. (2021). Network intru-
sion detection system: A systematic study of machine learning and deep learning approaches.
Transactions on Emerging Tele-communications Technologies, 32(1), e4150.
9. Alzubi, J., Nayyar, A., & Kumar, A. (2018, November). Machine learning from theory to
algorithms: an overview. In Journal of physics: conference series (Vol. 1142, p. 012012). IOP
Publishing.
10. Schmidhuber, J.: Deep learning in neural networks: An overview. Neural Netw. 61, 85–117
(2015)
11. Alkaşı, U. Machine learning based design of gap waveguides. MSc Thesis, Istanbul Technical
Univ., Istanbul, Turkey.
12. https://www.ibm.com/ibm/history/ibm100/us/en/icons/ibm700series/impacts/#:~:text=On%
20February%2024%2C%201956%2C%20Arthur,The%20computer%20won. [Visit date: 21
February 2023].
The State of Art in Machine Learning Applications in Civil Engineering 173

13. Samuel, A.L.: Machine learning. The. Technol. Rev. 62(1), 42–45 (1959)
14. Bowling, M., Fürnkranz, J., Graepel, T., Musick, R.: Machine learning and games. Machine
learning 63(3), 211–215 (2006)
15. Cover, T., Hart, P.: Nearest neighbor pattern classification. IEEE Trans. Inf. Theory 13(1),
21–27 (1967)
16. https://www.forbes.com/sites/bernardmarr/2016/02/19/. [Visit date: 21 February 2023].
17. Thai, H. T. (2022, April). Machine learning for structural engineering: A state-of-the-art
review. In Structures (Vol. 38, pp. 448–491). Elsevier.
18. https://chatbotsmagazine.com/ whats-machine-learning-it-s-expensive-slow-and-/ exclusive-
for-now-4512c47352ee. [Visit date: 21 February 2023].
19. Baştanlar, Y., & Özuysal, M. (2014). Introduction to machine learning. miRNomics: MicroRNA
biology and computational analysis, 105–128.
20. Ghahramani, Z. (2004). Unsupervised learning. Advanced Lectures on Machine Learning:
ML Summer Schools 2003, Canberra, Australia, February 2–14, 2003, Tübingen, Germany,
August 4–16, 2003, Revised Lectures, 72–112.
21. Mohri, M., Rostamizadeh, A., & Talwalkar, A. (2012). Foundations of machine learning.[Sl].
22. Awad, M., & Khanna, R. (2015). Efficient learning machines: theories, concepts, and
applications for engineers and system designers (p. 268). Springer nature.
23. Xu, N. (2019, April). Understanding the reinforcement learning. In Journal of Physics:
Conference Series (Vol. 1207, No. 1, p. 012014). IOP Publishing.
24. Shobha, G., Rangaswamy, S.: Chapter 8-Machine Learning Handbook of Statistics. Elsevier
(2018)
25. Aksoy, S. Classsification of VOC Vapors Using Machine Learning Algorithm. MSc Thesis,
Yildiz Technical Univ., Istanbul, Turkey, 2022.
26. Jason, C. A., & Kumar, S. (2020). An appraisal on speech and emotion recognition
technologies based on machine learning. language, 67, 68.
27. Gambella, C., Ghaddar, B., Naoum-Sawaya, J.: Optimization problems for machine learning:
A survey. Eur. J. Oper. Res. 290(3), 807–828 (2021)
28. Peter, S. C., Dhanjal, J. K., Malik, V., Radhakrishnan, N., Jayakanthan, M., Sundar, D.,
... & Jayakanthan, M. (2019). Encyclopedia of bioinformatics and computational biology.
Ranganathan, S., Grib-skov, M., Nakai, K., Schönbach, C., Eds, 661–676.
29. Karystinos, G.N., Pados, D.A.: On overfitting, generalization, and randomly expanded training
sets. IEEE Trans. Neural Networks 11(5), 1050–1057 (2000)
30. Refaeilzadeh, P., Tang, L., Liu, H.: Cross-validation. Encyclopedia of database systems 5,
532–538 (2009)
31. Lemm, S., Blankertz, B., Dickhaus, T., Müller, K.R.: Introduction to machine learning for
brain imaging. Neuroimage 56(2), 387–399 (2011)
32. Altınok, N. Predicting case closeability of legal debt collecti̇on agency with machine learni̇ng
teqniques and assignment of cases with closibility based assignment model: A case study on
telecomunication sector. MSc Thesis, Istanbul Technical Univ., Istanbul, Turkey.
33. https://neptune.ai/blog/cross-validation-in-machine-learning-how-to-do-it-right. [Visit date:
20 February 2023].
34. https://towardsdatascience.com/cross-validation-explained-evaluating-estimator-perfor
mance-e51e5430ff85. [Visit date: 20 February 2023].
35. Montesinos López, O.A., Montesinos López, A., Crossa, J.: Overfitting, Model Tuning, and
Evaluation of Prediction Performance. In: Multivariate statistical machine learning methods
for genomic prediction, pp. 109–139. Springer International Publishing, Cham (2022)
36. https://yigitsener.medium.com/makine-%C3%B6%C4%9Frenmesinde-train-validation-ve-
test-kavramlar%C4%B1-ile-python-uygulamas%C4%B1-a60214fb5448. [Visit date: 20
February 2023].
37. Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal
of the royal statistical society. Series B (Methodological), 111–147.
38. Beck, A., Kurz, M.: A perspective on machine learning methods in turbulence modeling.
GAMM-Mitteilungen 44(1), e202100002 (2021)
174 Y. Aydin et al.

39. Zhou, Z.H., Li, M.: Semi-supervised learning by disagreement. Knowl. Inf. Syst. 24, 415–439
(2010)
40. Kwon, D., Kim, H., Kim, J., Suh, S.C., Kim, I., Kim, K.J.: A survey of deep learning-based
network anomaly detection. Clust. Comput. 22, 949–961 (2019)
41. Abdullahi,N. Sound Based Location Detection Using Machine Learning Methods. MSc
Thesis, Fırat Univ., Elazığ, Turkey, 2022.
42. Paper, D. (2020). Introduction to scikit-learn. Hands-on Scikit-Learn for Machine Learning
Applications: Data Science Fundamentals with Python, 1–35.
43. Chapelle, O., Scholkopf, B., & Zien, A. (2006). Semi-supervised learning. 2006. Cambridge,
Massachusettes: The MIT Press View Article, 2.
44. Larose, D.T.: Data mining and predictive analytics. John Wiley & Sons (2015)
45. Saxena, A., Prasad, M., Gupta, A., Bharill, N., Patel, O. P., Tiwari, A., ... & Lin, C. T. (2017).
A review of clustering techniques and developments. Neurocomputing, 267, 664-681.
46. Harrington, P.: Machine learning in action. Simon and Schuster (2012)
47. Veena, S., Nerisai, M. V., Remya, J. V., & Sai Tejah, S. (2020). Sound classification system
using machine learning techniques. International Journal of Engineering Applied Sciences
and Technology.
48. Alkaşı, U. Machine Learning Based Design of Gap Waveguides. MSc Thesis, Istanbul
Technical Univ., Istanbul, Turkey, 2022.
49. Xu, Y., Shang, L., Ye, J., Qian, Q., Li, Y. F., Sun, B., ... & Jin, R. (2021, July). Dash: Semi-
supervised learning with dynamic thresholding. In International Conference on Machine
Learning (pp. 11525–11536). PMLR
50. Zhang, X.D.: A matrix algebra approach to artificial intelligence (2020)
51. Yao, Q., Yang, H., Bao, B., Yu, A., Zhang, J., Cheriet, M.: Core and spectrum allocation based
on association rules mining in spectrally and spatially elastic optical networks. IEEE Trans.
Commun. 69(8), 5299–5311 (2021)
52. Cutler, A., Cutler, D. R., & Stevens, J. R. (2012). Random forests. Ensemble machine learning:
Methods and applications, 157–175.
53. Akar, Ö., Güngör, O.: Classification of multispectral images using Random Forest algorithm.
Journal of Geodesy and Geoinformation 1(2), 105–112 (2012)
54. Zan, Ç. Ö. Prediction of Soil Radon Gas Using Meteorological Parameters With Machine
Learning Algorithms. MSc Thesis, Dokuz Eylül University, İzmir, Turkey, 2021.
55. Nizam, H., & Akın, S. S. (2014). Sosyal medyada makine öğrenmesi ile duygu anal-
izinde dengeli ve dengesiz veri setlerinin performanslarının karşılaştırılması. XIX. Türkiye’de
İnternet Konferansı, 1(6).
56. https://www.analyticsvidhya.com/blog/2018/06/comprehensive-guide-for-ensemble-mod
els/. [Visit date: 21 February 2023].
57. Salimi, A., Faradonbeh, R.S., Monjezi, M., Moormann, C.: TBM performance estimation
using a classification and regression tree (CART) technique. Bull. Eng. Geol. Env. 77, 429–440
(2018)
58. Mahesh, B.: Machine learning algorithms-a review. Int. J. Sci. Res. (IJSR) 9, 381–386 (2020)
59. Brereton, R.G., Lloyd, G.R.: Support vector machines for classification and regression.
Analyst 135(2), 230–267 (2010)
60. Şahinbaş, K.: A decision support proposal for imbalanced clinical data: Acute appendicitis
cases. PhD Thesis, Istanbul Univ., Istanbul, Turkey (2019)
61. https://doi.org/10.21275/ART20203995. [Visit date: 22 February 2023]
62. Elasan, S. Investigation of different decision trees and k-nearest neighbor methods in data
mining: An application on gynecology and birth data. PhD Thesis, Van Yüzüncü Yıl Univ.,
Van, Turkey
63. Sperandei, S.: Understanding logistic regression analysis. Biochem. Med. 24(1), 12–18 (2014)
64. Mete, M.O.: Development of mass property valuation model based on geographic information
systems integrated machine learning methods. PhD Thesis, Istanbul Technical Univ., Istanbul,
Turkey
The State of Art in Machine Learning Applications in Civil Engineering 175

65. Lillesand, T., Kiefer, R.W., Chipman, J.: Remote sensing and image interpretation. John
Wiley & Sons (2015)
66. https://bernatas.medium.com/roc-e%C4%9Frisi-ve-e%C4%9Fri-alt%C4%B1nda-kalan-
alan-auc-97b058e8e0cf. [Visit date: 22 February 2023].
67. Coşkun, C., Baykal, A.: An application for comparison of data mining classification
algorithms. Akademik Bilişim 11, 51–58 (2011)
68. Machine learning method based marine diesel engine modelling and decision-support system
setting. PhD Thesis, Istanbul Technical Univ., Istanbul, Turkey.
69. https://medium.com/deep-learning-turkiye/model-performans%C4%B1n%C4%B1-de%
C4%9Ferlendirmek-metrikler-cb6568705b1. [Visit date: 22 February 2023].
70. Classification with support vector machines and logistic regression: the example of Turkey’s
import and export. MSc Thesis, İnönü Univ., Malatya, Turkey.
71. Kumar, R., Indrayan, A.: Receiver operating characteristic (ROC) curve for medical
researchers. Indian Pediatr. 48, 277–287 (2011)
72. Yulisa, A., Park, S. H., Choi, S., Chairattanawat, C., & Hwang, S. (2022). Enhancement of
voting regressor algorithm on predicting total ammonia nitrogen concentration in fish waste
anaerobiosis. Waste Biomass Valorization, 1–18
73. Solano, E.S., Dehghanian, P., Affonso, C.M.: Solar radiation forecasting using machine
learning and ensemble feature selection. Energies 15(19), 7049 (2022)
74. https://en.wikipedia.org/wiki/Mean_absolute_percentage_error. [Visit date: 23 February
2023].
75. https://www.geeksforgeeks.org/underfitting-and-overfitting-in-machine-learning/. [Visit
date: 23 February 2023].
76. Naeej, M., Bali, M., Naeej, M.R., Amiri, J.V.: Prediction of lateral confinement coefficient in
reinforced concrete columns using M5’machine learning method. KSCE J. Civ. Eng. 17(7),
1714 (2013)
77. Chou, J.S., Tsai, C.F., Pham, A.D., Lu, Y.H.: Machine learning in concrete strength
simulations: multi-nation data analytics. Constr. Build. Mater. 73, 771–780 (2014)
78. Taffese, W.Z., Sistonen, E., Puttonen, J.: CaPrM: carbonation prediction model for reinforced
concrete using machine learning methods. Constr. Build. Mater. 100, 70–82 (2015)
79. Vu, D.T., Hoang, N.D.: Punching shear capacity estimation of FRP-reinforced concrete slabs
using a hybrid machine learning approach. Struct. Infrastruct. Eng. 12(9), 1153–1161 (2016)
80. Hoang, N.D., Chen, C.T., Liao, K.W.: Prediction of chloride diffusion in cement mortar using
multi-gene genetic programming and multivariate adaptive regression splines. Measurement
112, 141–149 (2017)
81. Mangalathu, S., Jeon, J.S.: Classification of failure mode and prediction of shear strength for
reinforced concrete beam-column joints using machine learning techniques. Eng. Struct. 160,
85–94 (2018)
82. Siam, A., Ezzeldin, M., El-Dakhakhni, W.: Machine learning algorithms for structural perfor-
mance classifications and predictions: Application to reinforced masonry shear walls. In
Structures (Vol. 22, pp. 252–265). Elsevier (2019, December)
83. Pham, A.D., Ngo, N.T., Nguyen, T.K.: Machine learning for predicting long-term deflections
in reinforce concrete flexural structures. J. Comput. Des. Eng. 7(1), 95–106 (2020)
84. Hwang, S.H., Mangalathu, S., Shin, J., Jeon, J.S.: Machine learning-based approaches for
seismic demand and collapse of ductile reinforced concrete building frames. J. Build. Eng.
34, 101905 (2021)
85. Kang, M.C., Yoo, D.Y., Gupta, R.: Machine learning-based prediction for compressive and
flexural strengths of steel fiber-reinforced concrete. Constr. Build. Mater. 266, 121117 (2021)
86. Cakiroglu, C., Islam, K., Bekdaş, G., Kim, S., Geem, Z.W.: Interpretable machine learning
algorithms to predict the axial capacity of frp-reinforced concrete columns. Materials
2022(15), 2742 (2022)
87. Yucel, M., Kayabekir, A. E., Nigdeli, S. M., & Bekdaş, G.: Optimum design of carbon fiber-
reinforced polymer (CFRP) beams for shear capacity via machine learning methods: optimum
prediction methods on advance ensemble algorithms–bagging combinations. In Research
176 Y. Aydin et al.

anthology on machine learning techniques, methods, and applications (pp. 308–326). IGI
Global (2022)
88. Cakiroglu, C., Bekdaş, G., Kim, S., Geem, Z.W.: Explainable ensemble learning models for
the rheological properties of self-compacting concrete. Sustainability 14(21), 14640 (2022)
89. Bekdaş, G., Cakiroglu, C., Islam, K., Kim, S., Geem, Z.W.: Optimum design of cylindrical
walls using ensemble learning methods. Appl. Sci. 12(4), 2165 (2022)
90. Cakiroglu, C., Bekdaş, G.: Predictive Modeling of recycled aggregate concrete beam shear
strength using explainable ensemble learning methods. Sustainability 15(6), 4957 (2023)
91. Aydın, Y., Bekdaş, G., Nigdeli, S.M., Isıkdağ, Ü., Kim, S., Geem, Z.W.: Machine learning
models for ecofriendly optimum design of reinforced concrete columns. Appl. Sci. 13(7),
4117 (2023)
92. Guo, B., Lin, X., Wu, Y., Zhang, L.: Machine learning-driven evaluation and optimisation of
compression yielded FRP-reinforced concrete beam with T section. Eng. Struct. 275, 115240
(2023)
93. Gambill, D.R., Wall, W.A., Fulton, A.J., Howard, H.R.: Predicting USCS soil classification
from soil property variables using Random Forest. J. Terrramech. 65, 85–92 (2016)
94. Puri, N., Prasad, H.D., Jain, A.: Prediction of geotechnical parameters using machine learning
techniques. Procedia Computer Science 125, 509–517 (2018)
95. Momeni, E., Dowlatshahi, M.B., Omidinasab, F., Maizir, H., Armaghani, D.J.: Gaussian
process regression technique to estimate the pile bearing capacity. Arab. J. Sci. Eng. 45(10),
8255–8267 (2020)
96. Rauter, S., Tschuchnigg, F.: CPT data interpretation employing different machine learning
techniques. Geosciences 11(7), 265 (2021)
97. Choi, J.C., Liu, Z., Lacasse, S., Skurtveit, E.: Leak-off pressure using weakly correlated
geospatial information and machine learning algorithms. Geosciences 11(4), 181 (2021)
98. Bekdaş, G., Cakiroglu, C., Kim, S., Geem, Z.W.: Optimal dimensioning of retaining walls
using explainable ensemble learning algorithms. Materials 15(14), 4993 (2022)
99. Tran, V.Q.: Predicting and investigating the permeability coefficient of soil with aided single
machine learning algorithm. Complexity (2022)
100. Aladsani, M. A., Burton, H., Abdullah, S. A., & Wallace, J. W: Explainable Machine Learning
Model for Predicting Drift Capacity of Reinforced Concrete Walls. ACI Struct. J. 119(3)
(2022)
101. Wang, X., Dong, X., Zhang, Z., Zhang, J., Ma, G., Yang, X.: Compaction quality evaluation
of subgrade based on soil characteristics assessment using machine learning. Transportation
Geotechnics 32, 100703 (2022)
102. Aydın, Y., Işıkdağ, Ü., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Use of machine learning
techniques in soil classification. Sustainability 15(3), 2374 (2023)
103. Vasegh, M., Dehghanbanadaki, A., & Motamedi, S. (2023). Enhanced Soil Liquefaction
Potential Estimation using Machine Learning and Web-Based Platform.
104. Hong, J., Lee, S., Bae, J. H., Lee, J., Park, W. J., Lee, D., ... & Lim, K. J. (2020). Development
and evaluation of the combined machine learning models for the prediction of dam inflow.
Water 12(10), 2927
105. Mata, J., Salazar, F., Barateiro, J., Antunes, A.: Validation of machine learning models for
structural dam behaviour interpretation and prediction. Water 13(19), 2717 (2021)
106. Shuang, Q., Zhao, R.T.: Water demand prediction using machine learning methods: a case
study of the Beijing–Tianjin–Hebei region in China. Water 13(3), 310 (2021)
107. Latif, S. D., Ahmed, A. N., Sherif, M., Sefelnasr, A., El-Shafie, A.: Reservoir water balance
simulation model utilizing machine learning algorithm. Alex. Eng. J. 60(1), 1365–1378 (2021)
108. Malek, N.H.A., Wan Yaacob, W.F., Md Nasir, S.A., Shaadan, N.: Prediction of water quality
classification of the Kelantan River Basin, Malaysia using machine learning techniques. Water
14(7), 1067 (2022)
109. Tanim, A.H., McRae, C.B., Tavakol-Davani, H., Goharian, E.: Flood detection in urban areas
using satellite imagery and machine learning. Water 14(7), 1140 (2022)
The State of Art in Machine Learning Applications in Civil Engineering 177

110. An, Y., Zhang, Y., Yan, X.: An Integrated Bayesian and machine learning approach application
to identification of groundwater contamination source parameters. Water 14(15), 2447 (2022)
111. Baljon, M., Sharma, S.K.: Rainfall prediction rate in Saudi Arabia using improved machine
learning techniques. Water 15(4), 826 (2023)
112. Gondia, A., Siam, A., El-Dakhakhni, W., & Nassar, A. H. (2020). Machine learning algorithms
for construction projects delay risk prediction. J. Constr. Eng. Manag., 146(1), 04019085.
113. Nguyen, P.T.: Application machine learning in construction management. TEM J. 10(3),
1385–1389 (2021)
114. Erzaij, K.R., Burhan, A.M., Hatem, W.A., Ali, R.H.: Prediction of the delay in the portfolio
construction using Naïve Bayesian classification algorithms. Civ. Environ. Eng. 17(2), 673–
680 (2021)
115. Fan, C.L.: Evaluation of classification for project features with machine learning algorithms.
Symmetry 14(2), 372 (2022)
116. Gondia, A., Ezzeldin, M., & El-Dakhakhni, W.: Machine learning-based decision support
framework for construction injury severity prediction and risk mitigation. ASCE-ASME J.
Risk Uncertain. Eng. Syst., Part A: Civ. Eng. 8(3), 04022024 (2022)
117. Fitzsimmons, J. P., Lu, R., Hong, Y., Brilakis, I.: Construction schedule risk analysis–a hybrid
machine learning approach (2022)
118. Golabchi, H., Hammad, A.: Estimating labor resource requirements in construction projects
using machine learning. Constr. Innov., (ahead-of-print). (2023)
119. Gondia, A., Moussa, A., Ezzeldin, M., & El-Dakhakhni, W.: Machine learning-based
construction site dynamic risk models. Technol. Forecast. Soc. Chang. 189, 122347 (2023)
120. Katanalp, B.Y., Yildirim, Z.B., Karacasu, M., Ibrikci, T.: Comparison of performance char-
acteristics of waste coal modified asphalt concrete using artificial neural networks and central
composite design methods. J. Eng. Sci. Des. 7(3), 680–688 (2019)
121. Cansız, Ö.F., Ünsalan, K., Erginer, İ: Modeling of highway energy consumption by artifi-
cial intelligence and regression methods. Uludağ Üniversitesi Mühendislik Fakültesi Dergisi
25(3), 1297–1314 (2020)
122. Li, C., Xu, P.: Application on traffic flow prediction of machine learning in intelligent
transportation. Neural Comput. Appl. 33, 613–624 (2021)
123. Bokaba, T., Doorsamy, W., Paul, B.S.: Comparative study of machine learning classifiers for
modelling road traffic accidents. Appl. Sci. 12(2), 828 (2022)
124. Olayode, I.O., Severino, A., Campisi, T., Tartibu, L.K.: Prediction of vehicular traffic flow
using Levenberg-Marquardt artificial neural network model: Italy road transportation system.
Commun.-Sci. Lett. Univ. Zilina 24(2), E74–E86 (2022)
125. Sholevar, N., Golroo, A., Esfahani, S.R.: Machine learning techniques for pavement condition
evaluation. Autom. Constr. 136, 104190 (2022)
126. Megnidio-Tchoukouegno, M., Adedeji, J.A.: Machine learning for road traffic accident
improvement and environmental resource management in the transportation sector. Sustain-
ability 15(3), 2014 (2023)
127. Wang, X., Pan, P., Li, J.: Real-time measurement on dynamic temperature variation of asphalt
pavement using machine learning. Measurement 207, 112413 (2023)
Machine Learning Application
of Structural Engineering Problems

Ayla Ocak, Sinan Melih Nigdeli, Gebrail Bekdaş, and Ümit Işıkdağ

Abstract The cross-sectional properties of the basic structural system elements,

such as columns and beams, are the basic structural design elements that need to be
determined sensitively. For the optimum design of such structural system elements,
it is necessary to minimize the displacement and volume by optimization. In this
study, the design of a tubular column and I-section beam element has been opti-
mized, and a section prediction model has been produced by the machine learning
method, which has been successfully applied in the risk and damage detection of
various engineering problems. For this purpose, optimum cross-section properties
were determined for different load conditions with the Jaya Algorithm (JA), which is
a metaheuristic algorithm. To minimize production errors arising from workmanship
in the production of structural system elements, cross-section parameters are divided
into classes covering certain dimensions. Different design combinations obtained
by optimization were converted into a data set and training for machine learning
was applied. With the trained data, a cross-section prediction model was produced
that predicts the cross-sectional properties of column and beam samples on a class
basis. When the results are examined, it is understood that the prediction models to
be produced with the optimum design data are suitable for use in determining the
cross-sectional properties of the structural system elements.

Keywords Artificial İntelligence · Machine learning · Optimization · Jaya

Algorithm

A. Ocak · S. M. Nigdeli · G. Bekdaş (B)

Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]
S. M. Nigdeli
e-mail: [email protected]
Ü. Işıkdağ
Department of Informatics, Mimar Sinan Fine Arts University, İstanbul 34427, Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 179
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_10
180 A. Ocak et al.

1 Introduction

Artificial intelligence refers to various software made to bring several features unique
to the human brain to a machine. In today’s world, it is used in many areas such as
health, economy, finance, media, and telecommunication. The main idea of artificial
intelligence technologies emerged from the idea of intelligentizing machines. In this
context, phones, tablets, computers, and similar devices that people use as commu-
nication tools have become a source of commercial income as “smart devices” by
providing various human-specific features such as listening, perceiving what one
hears, making suggestions, and finding directions. In recent years, artificial intelli-
gence technologies have been successfully applied in the prediction of production
materials, and risk and damage detection as a solution to engineering problems.
When artificial intelligence studies in the field of engineering are examined, it is
seen that it is frequently used in the detection of material strength, performance,
tracking, and faults such as bearing failure detection in industrial plants, detection
of superficial defects of cold rolled steel strips, monitoring of thermal events in
reactors, quality control of automatic production lines, modeling of CNC machine
tools [1–5]. In the field of structural engineering, it is frequently preferred for deter-
mining the capacities of structural elements and materials such as structural damage
assessment, beam member bending estimation, steel plate buckling load estimation,
final bearing capacity estimation of concrete-filled steel pipes, concrete compressive
strength estimation [6–10].
Artificial intelligence is a technology that has a wide range of research and includes
different application techniques. Among them, machine learning stands out in terms
of the way it processes and uses data. Machine learning is defined as the process of
gaining human-specific behaviors to a machine, just as it is at the base of artificial
intelligence. The difference that makes it a different technique here is that the machine
is trained with some data to gain the ability to make inferences. By entering some
data into the machine, it is expected that each data will be perceived by the machine
and it will be successful in estimating the new data. For this purpose, using various
validation techniques, the data is divided into two training and test data, and the
prediction performance of the trained data and the tested data is turned into a table.
The resulting table shows how many of the correct and incorrect outputs the machine
predicts as correct or incorrect, and how many incorrect or correct predictions of the
incorrect result. in this way, the predictive success of the split data for training and
testing becomes comparable.
Machine learning is a method of generating a good predictive model that can
be applied to all parts of civil engineering, including construction, transporta-
tion, mechanics, geotechnical, materials, hydraulics, and construction management.
Performance in solving various problems such as the estimation of the bearing
capacity of foundations and soil geotechnical parameters, the extent of water spillage
effects, flood estimation, measurements and measurements of damage, estimation of
seismic isolation systems under earthquake load, estimation of cracks in asphalt pave-
ments and construction costs was supported by the operation performed [11–18]. The
Machine Learning Application of Structural Engineering Problems 181

possibilities of machine learning in terms of application areas give the impression

that it is suitable for the design of separation units. The design of the structural
system elements that make up the column, beam, floor, and a similar structure is
important in terms of building safety and cost. In machine test applications for such
elements, shear, stress, and axial load capacity estimations are generally studied [19–
22]. Machine learning applications in the dimensioning of structural carrier units are
a lighting method that should have an open structure specified in the literature.
Optimization is a method that structural engineers often resort to getting the
highest efficiency from the structural design. The aim here is to determine the most
appropriate parameters in line with the objective function determined by the problem.
In optimization applications, the optimum solution is sought by using various algo-
rithms. Metaheuristic algorithms are a method used in the optimization process and
are among the sub-branches of artificial intelligence. It is based on the idea of mathe-
matical modeling of intuitive methods that can be inspired by the instinctive behavior
of living things in nature as well as natural events such as pollination. Particle swarm
optimization (PSO) is inspired by the swarm behavior of living things, gray wolf opti-
mization (WAO) is inspired by the social hierarchy and hunting strategies of wolves,
bat algorithm (BA) is inspired by the echolocation of bats and flower pollination
algorithm (FPA) inspired by the pollination process of flowers. There are varieties
inspired by different events like the artificial bee colony algorithm (ABC) inspired by
the path it follows, the harmony search algorithm (HSA) inspired by the process of
composing the best harmony and the teaching–learning-based optimization (TLBO)
inspired by the learning and teaching process of a group of individuals [23–29]. In the
studies, lattice structures, steel plate shear wall frames, retaining walls, composite
plates, and similar elements were preferred as optimization methods in the optimum
design [30–34]. Another heuristic algorithm is the Jaya Algorithm (JA), which was
developed by Rao by reducing the two-stage TLBO algorithm consisting of a teaching
and learning process to a single stage [35]. Since this algorithm consists of a single
stage and does not contain a design factor, it is a simple and easily applicable,
advantageous algorithm that produces fast solutions. In structural engineering, the
optimization of control systems has been supported by studies that have remarkable
effects on the optimum design of truss structures [36–38].
Data is one of the most important parameters to achieve high prediction success in
machine learning applications. The optimum level of data to be used in the application
will be an advantage for the machine to offer the best solutions. The machine learning
method is divided into two main headings supervised and unsupervised. Studies with
supervised learning are divided into classification or regression problems according
to the type of problem and are used in algorithms accordingly. To make the work-
manship easy and applicable in the design of the structural elements, the design is
required to be in standard dimensions. Considering this aspect, it would be more
appropriate for the design problems to divide the dimensioning models into classes
and make class predictions in terms of ease of work and workmanship. Considering
this situation, it is recommended to use classification algorithms of machine learning
in estimating the cross-sectional dimensions of structural members.
182 A. Ocak et al.

In this study, a machine learning model has been developed for cross-section
predictions of column and beam examples, which are basic structural elements. For
this purpose, the optimum design properties were determined by optimizing the
structural elements under different load conditions with the Jaya algorithm. The
optimization process has been completed using the Jaya algorithm on the Matlab
program [39]. The obtained optimum design parameters and load cases were turned
into a data set and a machine-learning method was applied. For the machine trained
with optimum data, class prediction models were developed with machine learning
algorithms on the Python program for column and beam examples [40]. The success
and performance of machine learning algorithms were evaluated for both problems.

2 Methodology

2.1 Optimum Design of Structural System Elements

with Jaya Algorithm

Jaya algorithm is a metaheuristic algorithm developed by Rao in 2016 [35]. Its origin
is based on the teaching–learning-based optimization (TLBO) algorithm. It has been
developed by reducing the TLBO algorithm, which has two stages, to a single stage. In
Jaya optimization, similar to other heuristic algorithms, the objective function of the
problem is determined and the lower and upper limit values of the element properties
to be optimized are introduced to the program. For elements such as columns and
beams, these values are the section width, height, or thickness that is planned to be
optimized. After defining the limit values of the section properties of the element
to be optimized, the first solution matrix is randomly assigned between these limit
values. Then, for each new solution to be produced, the best and worst solution values
in the previous solution matrix are determined and used to produce new solutions.
The algorithm equation used to obtain new solutions is shown in Eq. 1.
( | |) | |
X i,new = X i, j + rand X i,best − | X i, j | − rand(X i,wor st − | X i, j |) (1)

X i,new in Eq. 1 is ith variable of new solution value, X i, j is ith variable of jth value
of the candidate solution in the initial matrix, X i,best and X i,wor st indicates the best
and the worst solution of ith value according to the objective function, respectively.
After this stage, the iteration process starts, each new solution is compared with
the old solutions, and the better solution is decided. The optimum result is reached
by repeating the solution updates with the amount of iteration.
Machine Learning Application of Structural Engineering Problems 183

2.2 Design Parameters for Tubular Column and I-Beam

Optimization

Tubular columns are elements with cross-sectional properties consisting of a center

and a diameter. In optimizing the cost of a tubular column under pressure, the objec-
tive function is to minimize these sections. A representative image of a tubular column
is shown in Fig. 1.
P in Fig. 1 denotes the pressure load, and l denotes the length of the column. In
the image of the A-A section taken on the column, di represents the inner diameter,
d0 , the outer diameter, d represents the center diameter and t represents the section
thickness of the column. The objective function f (d, t) used in the cost minimization
process for cost optimization of such columns is given in Eq. 2 [42].

Min f (d, t) = 9.8dt + 2d (2)

In the optimization of tubular columns, it is convenient to use the axial load

capacity and the buckling limit as constraints that can be defined for the design.
In the axial load capacity constraint, the case that the compressive stress does not
exceed the yield strength is taken into account. The equations to be used for the axial
load constraint (g1 ) and column buckling constraint (g2 ) is given in Eqs. 3 and 4,
respectively [42].

P
g1 = −1≤0 (3)
π d tσ y

Fig. 1 Tubular column and

cross-section detail [41]
184 A. Ocak et al.

8 P L2
g2 = −1≤0 (4)
π 2 Ed t(d 2 + t 2 )

E given in Eqs. 3 and 4 denotes the modulus of elasticity of the material, and σ y
the yield strength.
In steel I profile beam designs, the height of the beam, the width and thickness
of the flange, and the thickness of the web is the parameters that express the cross-
sectional properties of the I-shaped beams. The design of such beams is aimed to
prevent the deflection of the beam against horizontal and vertical loads. Optimization
of the section properties is necessary to avoid vertical displacements of the beam. In
addition to the height of the I-profile beams and the flange widths, the optimization of
the flange and web thicknesses is also important. Producing the material in sufficient
thickness is one of the design criteria that is effective in preventing deflection. A
representative illustration of an example of a steel beam with an I profile is given in
Fig. 2.
Figure 2, P indicates the vertical load, Q indicates the horizontal load, C L the
centerline, L is the length of the beam, b is the flange width of the beam, h is the beam
height, and t f and tw are the flange and web thicknesses of the beam, respectively.
In the optimization of the I-section beam, it is desired to minimize the deflection
of the beam. The displacement equation for a simple beam to be used in optimization
is given in Eq. 5. The objective function depending on the beam sections obtained
by calculating the inertia of the beam is shown in Eq. 6 [42].

P L3
f (x) = (5)
48E I

Fig. 2 I-section beam construction and design variables [43]

Machine Learning Application of Structural Engineering Problems 185

P L3
M i n f (h, b, t w , t f ) = ( ( )2 ) (6)
t w ( h−2t f )
3
bt 3f h−t f
48E 12
+ 6
+ 2bt f 2

E in Eqs. 5 and 6 represents the modulus of elasticity of the material and I

represents the moment of inertia of the beam.
To limit the beam section properties, two different design criteria can be set for
the I profile. One of them can be taken as the maximum area constraint (300 cm2 )
for the beam section and the other as the allowable moment stress (6kN/cm2 ). The
recommended beam section constraint g3 is given in Eq. 7 and the moment stress
constraint g4 is given in Eq. 8 [42].
( )
g 3 = 2bt f + t w h − 2t f ≤ 300 (7)

1.5 P Lh 1.5 Q Lb
g4 = ( )3 ( ( )) + 3 ( ) ≤6
t w h − 2t f + 2bt f 4t f + 3h h − 2t f
2 t w h − 2t f + 2t w b3
(8)

2.3 Class Prediction of Column and Beam Sections

with Machine Learning

Column and beam elements are carrier system parts that require precision and good
workmanship. In the production of these elements, although it is possible to estimate
the element dimensions at the numerical level depending on the characteristics and
correlation of the data, it is advantageous to apply the class-based estimation method
considering the convenience it will provide in production. Standard production of
carrier system elements, taking into account the speed and practicality of production,
reduces labor costs and minimizes the errors arising from the production to be made
in practice. For all these purposes, classification algorithms of machine learning are
suitable methods for use in education. While the classification of multiple sections
separately is possible with machine learning, they can also be classified at the same
time. Within the scope of this study, class-based cross-section prediction models
were produced by using Logistic Regression and Decision Tree algorithms, which
are machine learning algorithms, in training.
The Decision tree algorithm is an algorithm that queries and branches the data
and divides the data into small clusters with a tree-like appearance. The attributes of
the data are expressed as nodes and the results from the test data form the branches of
the tree [44]. As an example of the branching of this algorithm, when it is desired to
make a prediction model on column sections, these sections are branched according
to whether they are long, short, wide, or narrow, and then they are divided into short,
186 A. Ocak et al.

Fig. 3 Decision tree example representation

long, and medium branches, while these new nodes are sufficient for the section. The
answers are given to the question of whether there are any new sub-branches of the
tree. An example of a decision tree for a rectangular column is given in Fig. 3.
As seen in the figure, the features in the data are transformed into nodes, and the
responses corresponding to these features form the branches of the tree.
The logistic regression algorithm is a supervised machine learning algorithm that
establishes a relationship between two data features. When estimating a dependent
variable, it makes an estimation based on the least squares algorithm [45, 46]. It is a
very successful method in classifications with two answers such as yes or no. Using
one of the predicted factors establishes a link for the estimation of the other factor.
Looking at the beam example, its use is expected to yield remarkable results if it is
desired to make a dual classification as thick and thin for the flange and web widths
of the beam.

3 Numerical Examples

3.1 Tubular Column

In the tubular column study, a machine learning model has been developed to estimate
the cross-section of the column under axial compression load. The cost minimization
has been made by using the objective function in Eq. 2 of the column center diameter
and section thicknesses from the Tubular column sections. The algorithm and design
parameters used in the optimization and the design lower and upper limit values are
given in Table 1.
In the optimization process, optimum cross-section parameters were obtained for
different stress, load, and column lengths, and data to be used in machine learning
were obtained. Then, for classification estimation, optimum section thicknesses (A,
B, C, D, E, F, G) and section center diameters were converted into 6 (H, J, K, L, M,
N) different classes and made categorically. The 10 lines of the generated data set
are given in Table 2, and the class distributions in the data are given in Table 3.
Machine Learning Application of Structural Engineering Problems 187

Table 1 Optimization
Symbol Definition Value
parameters and design limits
mt Maximum iteration number 10,000
pn Population number 15
E Modulus of elasticity(kgf/cm2 ) 0.85 × 106
L Column length (cm) 100–500
P External load (kgf) 100–5000
σy Strain (kgf/cm2 ) 100–500
tmin Minimum section thickness (cm) 0.2
tmax Maximum section thickness (cm) 0.9
dmin Minimum section center diameter (cm) 2
dmax Maximum section center diameter (cm) 14

Table 2 10 rows of tubular column dataset

Strain Length Load Thickness (cm) Center Fx
(kgf/cm2 ) (cm) (kgf) Diameter
(cm)
100 100 100 A H 7.92
100 300 100 C H 13.35831
100 400 500 A J 22.22907
100 2200 400 G K 84.28906
200 2800 100 F J 54.39317
300 900 100 C H 13.35831
300 4500 400 E K 60.24811
400 1800 200 B H 21.81489
500 800 200 A H 14.43357
500 3500 300 A K 36.6415

The table with the correlation matrix of the data is given in Table 4 and its graphical
representation is given in Fig. 4.
The cross-sectional properties of the tubular column were estimated with the
created data set. Since the section center diameter and section thicknesses require
two classifications, the class estimation of the two sections at the same time was
investigated by applying multiple classifications. Multi-output classification is an
algorithm for the simultaneous prediction of two or more features. The classification
result presents two or more outputs. An example of this can give a data set about fruits.
In this dataset, when one of the attributes is the type of fruit and the other is the color
of the fruit, the type (apple, pear, orange) and color (yellow, orange, red) of the fruit
can be estimated simultaneously with multi-classification. In the classification made,
multi-output classification is applied by choosing one of the classification algorithms.
In multiple-output classification applications, it is calculated by taking the precision
188 A. Ocak et al.

Table 3 Class distribution of

Class of t Number of data Section intervals
the dataset
A 438 0.2–0.3
B 100 0.3–0.4
C 84 0.4–0.5
D 84 0.5–0.6
E 72 0.6–0.7
F 82 0.7–0.8
G 324 0.8–0.9
Class of d Number of data Section intervals
H 424 2–4
J 258 4–6
K 302 6–8
L 125 8–10
M 56 10–12
N 19 12–14

Table 4 Correlation matrix for the tubular column

Strain Length Load Fx
Strain 1 0.1102 0.0018 −0.5249
Length 0.1102 1 −0.0014 0.5879
Load 0.0018 −0.0014 1 0.1243
Fx −0.5249 −0.5879 0.1243 1

Fig. 4 Correlation matrix

graph for a tubular column
Machine Learning Application of Structural Engineering Problems 189

Table 5 Accuracy metrics of

Algorithm Accuracy (%)
multi-output classification for
a tubular column Decision tree classification 42.27
K-Nearest neighbors classification 33.23
Logistic regression 22.40
Linear discriminant analysis 34.39
Naïve Bayes 22.29
Support vector machine 29.48

values of each predicted group and finding their average. The precision table of the
multi-output classification prediction models made with different machine learning
algorithms is given in Table 5.
Prediction models were reproduced by using the decision tree algorithm. The
highest precision is given in multi-output classification to develop the prediction
model separately for the two dimensions. In the separate classification, 70% of the
data was allocated for training and 30% for testing was applied. Whereas in multi-
output classification, training was applied to the data with tenfold cross-validation.
The obtained confusion matrices are shown in Figs. 5 and 6, respectively, for the
section thickness and section center diameter estimation model, and the accuracy,
precision, recall, and f1 scores are shown in Tables 6 and 7, respectively.

Fig. 5 Confusion matrix:

thickness prediction model
190 A. Ocak et al.

Fig. 6 Confusion matrix:

center diameter
prediction model

Table 6 Decision tree

Class Precision Recall F1-score
classification, accuracy,
precision, recall, and f1-score A 0.96 0.92 0.94
values for thickness B 0.50 0.56 0.53
C 0.48 0.39 0.43
D 0.33 0.48 0.39
E 0.38 0.39 0.38
F 0.35 0.36 0.36
G 0.86 0.82 0.84
Accuracy 0.72

Table 7 Decision tree

Class Precision Recall F1-score
classification, accuracy,
precision, recall, and f1-score H 0.96 0.98 0.97
values for center diameter J 0.86 0.80 0.83
K 0.85 0.84 0.84
L 0.81 0.85 0.83
M 0.72 0.81 0.76
N 0.67 0.50 0.57
Accuracy 0.88
Machine Learning Application of Structural Engineering Problems 191

3.2 I-Section Beam

A machine learning model has been developed for the cross-section prediction of
an I-section steel beam sample. The beams resist deflection by using the moment of
inertia. Since beam section width and height are used in the basic parameters of the
moment of inertia, it is quite difficult to determine the optimum section properties for
different load situations by optimization. The optimization process usually aims to
maximize the inertia of the beam based on the maximum limit. In this case, obtaining
data on a wide variety of section properties, with optimization for use in machine
learning for the beam, is an area to be explored. A solution to this problem can be
considered by applying different objective functions. Another feature that protects
I-section beams from deflection is the beam flange and web thicknesses. The main
purpose of optimization here is to prevent vertical displacement of the beam against
horizontal and vertical loads. In a classification of beam dimensions, the classification
of I-section beam thicknesses is a viable solution to generate a predictive model for
minimizing vertical displacements. In this case study, the flange and web thicknesses
of an I-section beam are optimized for different load cases. The algorithm and design
parameters and limits for optimization with the Jaya algorithm are given in Table 8.
In the optimization process, optimum cross-section dimensions for different stress
and load situations were obtained and thus these optimal dimensions were used
in forming the new dataset. Then, for classification estimation, optimum section
thicknesses were converted into 2 (thin and thick) different classes. 10 rows of the
produced data set are given in Table 9 and the class distributions of the data are given
in Table 10.

Table 8 Optimization
Symbol Definition Value
parameters and design limits
mt Maximum iteration number 500,000
pn Population number 15
E Modulus of elasticity (kN/cm2 ) 20,000
L Beam length (cm) 200
P Vertical load (kN) 100 ∼ 800
Q Horizontal load (kN) 1 ∼ 80
σ Moment stress (kN/cm2 ) 6
h min Minimum beam section height (cm) 10
h max Maximum beam section height (cm) 80
bmin Minimum beam section width (cm) 10
bmax Maximum beam section width (cm) 50
twmin Minimum beam web thickness (cm) 0.9
twmax Maximum beam web thickness (cm) 5
t f min Minimum beam flange thickness (cm) 0.9
t f max Maximum beam flange thickness (cm) 5
192 A. Ocak et al.

Table 9 10 rows of I-beam dataset

Vertical load Horizontal load Height Width Web thickness Flange Fx
(kN) (kN) (cm) (cm) thickness
100 1 80 50 Thin Thick 0.002179
100 27 63 31.43 Thin Thick 0.003777
120 19 68.60 20.54 Thick Thick 0.005147
120 74 80 50 Thick Thin 0.004035
140 78 80 50 Thin Thin 0.003161
150 3 65.55 32.22 Thick Thin 0.006583
210 3 80 34.80 Thin Thick 0.004698
240 67 79.86 40.73 Thick Thin 0.007783
250 24 80 50 Thick Thin 0.008407
290 12 63.70 25.21 Thick Thick 0.014492

Table 10 Class distribution

Class of tw Number of data Section intervals
of the dataset
Thin 4917 0.9–2.2
Thick 596 2.2–5
Class of tf Number of data Section intervals
Thin 3931 0.9–2.2
Thick 1582 2.2–5

The graphical representation of the correlation matrix of the data in Table 11 is

given in Fig. 7.
With the created data set, a predictive model was produced with machine learning
classification algorithms. Multiple classifications were performed for simultaneous
estimation of both web and flange widths. The precision levels of the algorithms used
in multi-output classification are shown in Table 12.
The logistic regression algorithm, which gives the highest precision in multi-
output classification, is also used in the estimation of web and flange thicknesses
separately, and the difference between multi-output classification is examined. The

Table 11 Correlation matrix for I-section beam

Vertical Horizontal Load Height Width Fx
Load
Vertical load 1 −0.0209 −0.0115 0.0157 0.0111
Horizontal load −0.0209 1 −0.0126 0.0174 0.0216
Height −0.0115 −0.0126 1 0.7252 −0.4452
Width 0.0157 0.0174 0.7252 1 −0.1895
Fx 0.0111 0.0216 −0.4452 −0.1895 1
Machine Learning Application of Structural Engineering Problems 193

Fig. 7 Correlation matrix

graph for I-section beam

Table 12 Accuracy metrics

Algorithm Accuracy (%)
of multi-output classification
for I-section beam Decision tree classification 75.81
K-nearest neighbors classification 84.44
Logistic regression 88.71
Linear discriminant analysis 87.74
Naïve Bayes 60.84
Support vector machine 60.68

estimation precision level was more successful for the separately classified web and
flange thicknesses. In a separate classification, 70% of the data was allocated to
training and 30% to testing, and training was applied. In multi-output classification,
training was applied to the data with tenfold cross validation. Confusion matrices of
the trained models are shown in Figs. 8 and 9 for web thickness and flange thickness
predictions, respectively, and accuracy, precision, recall, and f1 scores are shown in
Tables 13 and 14, respectively.

4 Discussion and Conclusions

In this study, column and beam sections, as an example of structural system elements,
are optimized for different load and stress conditions with metaheuristic algorithms.
Jaya Algorithm, a metaheuristic algorithm, was used in the optimization process.
194 A. Ocak et al.

Fig. 8 Confusion matrix for

the thickness prediction
model

Fig. 9 Confusion matrix for

the center diameter
estimation model

Table 13 Logistic regression

Class Precision Recall F1-score
classification, accuracy,
precision, recall, and f1-score Thick 1.00 0.12 0.22
values for web thickness (tw) Thin 0.90 1.00 0.95
Accuracy 0.91
Machine Learning Application of Structural Engineering Problems 195

Table 14 Logistic regression

Class Precision Recall F1-score
classification, accuracy,
precision, recall, and f1-score Thick 0.98 0.98 0.98
values for flange thickness (tf) Thin 0.99 0.99 0.99
Accuracy 0.99

The optimum values obtained from the optimization, the attributes used in the opti-
mization process, and machine learning techniques were applied, which were turned
into a data set.
In the column example, the section center diameter and section thicknesses are
divided into 6 and 7 classes at regular intervals, and the prediction model of the
machine is aimed to estimate sections in standard dimensions. The highest precision
values of the column machine learning model, in which the multiple classification
methods are applied, were obtained in the decision tree algorithm. A decision tree is
an algorithm that can be applied to both classification and regression problems. In
this study, since it is desired to create a class-based prediction model, the model was
created by using the decision tree algorithm in classification. In comparison with
other algorithms, it can be shown that the most successful algorithm is the decision
tree because splitting into many classes results in the formation of many branches
against the nodal points of the partitions. Up to 42% prediction accuracy is a good
prediction success rate. Since it undertakes a successful estimation task, it may be
appropriate to use this algorithm in numerical studies for regression analysis apart
from classification. When the prediction model success levels of two-column section
features are examined separately, it is seen that the prediction success of some classes
is 96%, while the section feature of another class has a precision of around 33%.
One of the main reasons for this situation is the averaging of precision values in
multiple classifications. Therefore, it can be said that the machine learning method
did not show a high performance due to the data characteristics, since the tubular
column sample was averaged over different prediction success rates. To overcome
this situation, increasing the number of data in the class with low estimation precision
or extending the training period may produce a solution.
In the study made with the beam example, the section width height, flange, and
web thicknesses of the beam were optimized with the Jaya algorithm. In the opti-
mization, the deflection of the beam is tried to be minimized. In the optimization to
prevent deflection due to beam inertia, the section width and height values were close
to the upper limit. Considering this situation, it was decided that it would be more
appropriate to use the web and flange dimensioning. In the alternative solution here,
beam heights and widths are used in training and a prediction model is produced by
machine learning to determine the flange and web thicknesses that affect the deflec-
tion of the beam. Beam web and flange thicknesses are divided into two classes thin
and thick. The logistic regression algorithm produced the most successful estimation
model in the multi-output classification. The fact that the web and flange thicknesses
have 2 classes is the reason why logistic regression is the classifier with the highest
precision. Other classification algorithms have also provided a remarkable level of
196 A. Ocak et al.

precision in estimating beam thicknesses. Logistic regression and linear discrimi-

nant analysis algorithms showed very close precision. The reason for this can be
shown that there is a linear relationship between them due to the small number of
classes and the data obtained by optimization. A logistic regression algorithm was
used to classify beam thicknesses separately. In the prediction models produced, a
91% accuracy level in web thickness and a 99% accuracy level in flange thickness
were obtained. However, the recall score obtained from the separate classification
estimated 100% correctly when the web width was “thin”, but it was insufficient
with 12% in the classification as “thick”. When the data distributions in Table 13 are
examined, a noticeable disproportion in the data class distributions draws attention.
Although the accuracy level is above 90%, it can be said that the reason why only
12% of the data that should be classified as “thick” are correctly classified is the
insufficient number of data expressed as “thick” in web thicknesses. It is expected
that this remarkable difference in the recall value will disappear by balancing the
number of data.
In line with the optimization and machine learning studies, it is shed light that
the application of artificial intelligence methods can be successful in estimating the
carrier system designs, and it is concluded that an artificial intelligence prediction
model with a high level of precision can be produced if the data is well analyzed and
organized.

References

1. Omoregbee, H.O., Edward, B.A., Olanipekun, M.U.: Bearing failure diagnosis and prognostics
modeling in plants for industrial purposes. J. Eng. Appl. Sci. 70(1), 1–17 (2023)
2. Chen, H., Nie, Z., Xu, Q., Fei, J., Yang, K., Li, Y., Liu, X. J. et al.: Intelligent detection and
classification of surface defects on cold-rolled galvanized steel strips using a data-driven faulty
model with attention mechanism. J. Comput. Inf. Sci. Eng.23(4), 041001 (2023)
3. Grelier, E., Mitteau, R., Moncada, V.: Deep learning-based process for the automatic detection,
tracking, and classification of thermal events on the in-vessel components of fusion reactors.
Fusion Eng. Des. 192, 113636 (2023)
4. Bono, F.M., Radicioni, L., Cinquemani, S.: A novel approach for quality control of automated
production lines working under highly inconsistent conditions. Eng. Appl. Artif. Intell. 122,
106149 (2023)
5. Zhu, M., Yang, Y., Feng, X., Du, Z., Yang, J.: A robust modeling method for thermal error of
CNC machine tools based on random forest algorithm. J. Intell. Manuf. 1–14 (2022)
6. Nguyen, N.M., Wang, W.C., Cao, M.T.: Early estimation of the long-term deflection of
reinforced concrete beams using surrogate models. Constr. Build. Mater. 370, 130670 (2023)
7. Luleci, F., Catbas, F.N., Avci, O.: Generative adversarial networks for labeled acceleration
data augmentation for structural damage detection. J. Civ. Struct. Heal. Monit. 13(1), 181–198
(2023)
8. Jayabalan, J., Dominic, M., Ebid, A. M., Soleymani, A., Onyelowe, K. C., Jahangir, H.:
Estimating the buckling load of steel plates with center cut-outs by ANN, GEP and EPR
techniques. Designs6(5), 84 (2022)
9. Khan, K., Biswas, R., Gudainiyan, J., Amin, M.N., Qureshi, H.J., Arab, A.M.A., Iqbal, M.:
PCA-based hybrid intelligence models for estimating the ultimate bearing capacity of axially
loaded concrete-filled steel tubes. Materials 15(18), 6477 (2022)
Machine Learning Application of Structural Engineering Problems 197

10. Adamu, M., Haruna, S.I., Malami, S.I., Ibrahim, M.N., Abba, S.I., Ibrahim, Y.E.: Prediction
of compressive strength of concrete incorporated with jujube seed as partial replacement of
coarse aggregate: a feasibility of Hammerstein–Wiener model versus support vector machine.
Model Earth Syst. Environ. 1–11 (2021)
11. Bagińska, M., Srokosz, P.E.: The optimal ANN Model for predicting the bearing capacity of
shallow foundations is trained on scarce data. KSCE J. Civ. Eng. 23, 130–137 (2019)
12. Puri, N., Prasad, H.D., Jain, A.: Prediction of geotechnical parameters using machine learning
techniques. Procedia Comput. Sci. 125, 509–517 (2018)
13. Mashhadi, N., Shahrour, I., Attoue, N., El Khattabi, J., Aljer, A.: Use of machine learning for
leak detection and localization in water distribution systems. Smart Cities 4(4), 1293–1315
(2021)
14. Chang, F.J., Hsu, K., Chang, L.C. (eds.): Flood forecasting using machine learning methods.
MDPI (2019)
15. Vitola Oyaga, J., Tibaduiza Burgos, D.A., Anaya Vejar, M., Pozo Montero, F. (2016). Structural
Damage detection and classification based on machine learning algorithms. In: Proceedings of
the 8th European Workshop on Structural Health Monitoring
16. Nguyen, H.D., Dao, N.D., Shin, M.: Machine learning-based prediction for the maximum
displacement of seismic isolation systems. J. Build. Eng. 51, 104251 (2022)
17. Ferrer-Espinoza, D., Atencio, E., Muñoz-La Rivera, F., Herrera, R.F.: Evaluation of the use
of cascade detection algorithms based on machine learning for crack detection in asphalt
pavements. Solid State Technol 64(2), 5588–5605 (2021)
18. Rafiei, M.H., Adeli, H.: Novel machine-learning model for estimating construction costs
considering economic variables and indexes. J. Constr. Eng. Manag. 144(12), 04018106 (2018)
19. Hou, R., Hou, Q.: Prediction of the shear capacity of ultrahigh-performance concrete beams
using neural network and genetic algorithm. Sci. Rep. 13(1), 2145 (2023)
20. Nasab, M. N., Jahangir, H., Hasani, H., Majidi, M.H., Khorashadizadeh, S.: Estimating the
punching shear capacities of concrete slabs reinforced by steel and FRP rebars with an ANN-
Based GUI toolbox. In: Structures, vol. 50, pp. 1204–1221. Elsevier. (2023)
21. Yao, G., Li, B., Xiong, X.: Deformation capacity of flexural-controlled SRC columns under
lateral cyclic load. Soil Dyn. Earthq. Eng. 169, 107902 (2023)
22. Rafiq Joo, M., Ahmad Sofi, F.: Unified approach for estimating axial-load capacity of concrete-
filled double-skin steel tubular columns of multiple shapes using nonlinear fe models and
artificial neural networks. Pract. Period. Struct. Des. Constr. 28(2), 04022074 (2023)
23. Kennedy, J., Eberhart, R.C.: Particle swarm optimization. In: Proceedings of IEEE International
Conference on Neural Networks No. IV, 27 November–1 December Perth, IEEE Service Center,
Piscataway, NJ, 1942–1948 (1995).
24. Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61 (2014)
25. Yang, X.S.: A new metaheuristic bat-inspired algorithm. In Nature-inspired cooperative
strategies for optimization (NICSO 2010), pp. 65–74. Springer, Berlin, Heidelberg (2010)
26. Yang, X.S.: Flower pollination algorithm for global optimization. In: Durand-Lose, J., Jonoska,
N. (eds.) Lecture Notes in Computer Science, vol. 27, pp. 7445, 240–249. Springer, London
(2012)
27. Karaboğa, D.: An idea based on honey bee swarm for numerical optimization (vol. 200,
pp. 1–10). Technical report-tr06, Erciyes University, engineering faculty, computer engineering
department (2005)
28. Geem, Z.W., Kim, J.H., Loganathan, G.V.: A new heuristic optimization algorithm: harmony
search. Simulation 76(2), 60–68 (2001)
29. Rao, R.V., Savsani, V.J., Vakharia, D.P.: Teaching-learning-based optimization: a novel method
for constrained mechanical design optimization problems. Comput. Aided Des. 43, 303–315
(2011)
30. Kaveh, A., Rad, A.S. Metaheuristic-based optimal design of truss structures using algebraic
force method. In: Structures, vol. 50, pp. 1951–1964. Elsevier (2023)
31. Daryan, A. S., Salari, M., Palizi, S., Farhoudi, N.: Size and layout optimum design of frames
with steel plate shear walls by metaheuristic optimization algorithms. In: Structures, vol. 48,
pp. 657–668. Elsevier (2023)
198 A. Ocak et al.

32. Degertekin, S.O., Lamberti, L., Ugur, I.B.: Discrete sizing/layout/topology optimization of
truss structures with an advanced Jaya algorithm. Appl. Soft Comput. 79, 363–390 (2019)
33. Gandomi, A.H., Kashani, A.R., Roke, D.A., Mousavi, M.: Optimization of retaining wall design
using evolutionary algorithms. Struct. Multidiscip. Optim. 55, 809–825 (2017)
34. De Almeida, F.S.: Stacking sequence optimization for maximum buckling load of composite
plates using a harmony search algorithm. Compos. Struct. 143, 287–299 (2016)
35. Rao, R.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7(1), 19–34 (2016)
36. Ocak, A., Bekdaş, G., Nigdeli, S.M.: A metaheuristic-based optimum tuning approach for
tuned liquid dampers for structures. Struct. Design Tall Spec. Build. 31(3), e1907 (2022)
37. Değertekin, S.O., Bayar, G.Y., Lamberti, L.: Parameter-free Jaya Algorithm for truss sizing-
layout optimization under natural frequency constraints. Comput. Struct. 245, 106461 (2021)
38. Bekdaş, G., Kayabekir, A.E., Niğdeli, S.M., Toklu, Y.C.: Transfer function amplitude mini-
mization for structures with tuned mass dampers considering soil-structure interaction. Soil
Dyn. Earthq. Eng. 116, 552–562 (2019)
39. The MathWorks, Matlab R2022b. Natick, MA (2022)
40. Python Software Foundation. PYTHON 3.11.0, USA (2022)
41. Rao, S.S.: Engineering Optimization Theory and Practise, 4th edn. Wiley, New Jersey, USA
(2009). ISBN: 978-0-470-18352-6
42. Bekdaş, G., Nigdeli, S.M., Yücel, M., Kayabekir A.E.: Yapay Zeka Optimizasyon Algoritmaları
ve Mühendislik Uygulamaları. Seçkin, Ankara, Turkey (2021)
43. Yang, X.S., Bekdaş, G., Niğdeli, S.M. (eds.): Metaheuristic and Optimization in Civil
Engineering. Springer, Switzerland (2016). ISBN:9783319262451
44. Morales, E.F., Escalante, H.J.: A brief introduction to supervised, unsupervised, and reinforce-
ment learning. In: Biosignal Processing and Classification Using Computational Learning and
Intelligence, pp. 111–129. Academic Press (2022)
45. Harrell, F.E.: Regression modeling strategies: with applications to linear models, logistic
regression, and survival analysis, vol. 608. Springer, New York (2001)
46. Mellit, A., Kalogirou, S.: Artificial intelligence techniques: machine learning and deep learning
algorithms. Handbook of artificial intelligence techniques in photovoltaic systems, pp. 43–83.
(2022)
Modeling Civil Engineering Problems
via Hybrid Versions of Machine Learning
and Metaheuristic Optimization
Algorithms

Vahdettin Demir, Esra Uray, and Serdar Carbas

Abstract The aim of this study is to develop hybrid solution models by integrating
metaheuristic optimization algorithms and machine learning technique. These hybrid
models are utilized to estimate bearing capacity of pile groups and lake levels, which
are common challenges to calculate in the geotechnical and hydrology designs of
civil engineering. To achieve this, Lake Beyşehir lake-water level observations and
various pile group designs are used as a dataset, which is divided into two parts; 75%
for training and 25% for testing. By employing improved hybrid models that combine
metaheuristic algorithms such as harmony search, artificial bee colony, and particle
swarm optimization with a machine learning technique called least squares support
vector regression (LSSVR), optimal values of kernel parameters are obtained reliably
and robustly. The results suggest that these hybrid models can be successfully applied
to complex real-world problems, as evidenced by nine evaluation metrics, including
mean absolute error (MAE), root mean squared error (RMSE), and determination
coefficient (R2 ), which showed satisfactory and reasonable performance.

Keywords Hybrid model · Least squares support vector regression ·

Metaheuristic optimization algorithms · Pile group bearing capacity · Lake level

V. Demir (B) · E. Uray

Department of Civil Engineering, KTO Karatay University, Konya 42020, Türkiye
e-mail: [email protected]
E. Uray
e-mail: [email protected]
S. Carbas
Department of Civil Engineering, Karamanoglu Mehmetbey University, Karaman 70200, Türkiye
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 199
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_11
200 V. Demir et al.

1 Introduction

Nowadays, as the solution to various civil engineering phenomena is affected

by management and sustainability processes, the solution process developed by
including this effect emerges as a complex issue. Researchers contributing to the
literature have shown a great interest in applying modern artificial intelligence (AI)
solution models and their hybrid versions promising potential for tackling chal-
lenging issues, to solve complex real-world engineering design problems [1, 2]. For
this reason, hybrid models generated with machine learning and various optimization
techniques are employed commonly by researchers in solving many intricated engi-
neering problems. Given the robustness, reliability, and effectiveness of the hybrid
models that result from the integration of optimization algorithms and artificial intel-
ligence models, these models demonstrate the capacity to address complex nonlinear
problems encountered in the field of civil engineering. Furthermore, hybrid solution
models based on AI have been recently utilized in the optimization of hyperpa-
rameters in machine learning in an attempt to reduce modelling time and increase
the accuracy of complex engineering problems such as civil engineering designs.
Since the hyperparameters affect the behaviour and performance of the model on
a vast scale, determining the appropriate parameter values is crucial [3]. Thereby,
the fundamental objective of this study is to create a hybrid model by integrating an
artificial intelligence model with nature-inspired metaheuristic optimization algo-
rithms for hyperparameter optimization and subsequently investigate it to address
two distinct problems in the field of civil engineering [4]. Within this scope, optimum
values of γ and α kernel parameters of Least Squares Support Vector Regression
(LSSVR) machine learning algorithm have been investigated by utilizing hybrid
algorithms via combining the harmony search algorithm (HS), artificial bee colony
algorithm (ABC), and particle swarm optimization (PSO) metaheuristic algorithms
with LSSVR.
In obtaining optimal values of γ and α, the datasets of lake-water levels of Lake
Beyşehir and the pile group have been employed and the performance of the hybrid
models have been evaluated with nine different metrics such as Error Sum of Squares
(SSE), Mean Squared Error (MSE), Mean Root Square Error (MRSE), Root Mean
Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Relative Error
(MARE), Coefficient of Determination (R2 ), Variance Account Factor (VAF), and
Adjusted Coefficient of Efficiency (E).

2 Design Problem Definition

2.1 Lake Level

Water resources engineering is a significant field of civil engineering that studies the
earth’s atmosphere and the hydrological cycle, which is the distribution and circula-
tion of water. The main purpose of water resources engineering is to determine the
Modeling Civil Engineering Problems via Hybrid Versions 201

quantity and quality of water resources within the framework of a management plan,
to protect, control, and to contribute to engineering design studies for the most effi-
cient use. Applications of water engineering include the design of hydraulic structures
such as stormwater and sewer pipes, dams, flood traps, regulators, and breakwaters,
management of waterways such as erosion protection and flood protection, and envi-
ronmental management such as the estimation of surface waters (river or lake). In
addition, hydroelectric power development, water supply, and irrigation are the main
applications of water resources engineering [5, 6].
The development of accurate lake-water level prediction with hybrid solution
models is important for water resource planning and management. In this study, the
estimation of Lake Beyşehir lake-water level as a water resources engineering design
problem is taken into consideration.
Lake-water level forecast is a vital requirement in water management operations
such as irrigation and drinking water supply management and decision-making,
fishing, tourism, railroad transportation, and a variety of other recreational and
socioeconomic activities [7]. Recently, techniques such as artificial intelligence
models, machine learning models, and deep learning have demonstrated an exten-
sive ability to model lake water levels without the need for experimental apparatus
and complex hydro-physical models based on physical principles and mathematical
equations.
For a comprehensive literature review and to see the trend of current studies,
the Scopus database was searched with the keywords “lake water level, forecast,
and hybrid and model, and machine and learning”. It has been observed that there
are 299 academic studies on similar subjects in total. These studies were limited
to 5 common keywords and a relationship map was obtained for the keywords of
the studies scanned using VOSviewer software [8]. A good search of the available
literature shows that these investigations include applications of the neural network
model, heuristic regression techniques, machine learning techniques, deep learning
techniques and their hybrid models.
In Fig. 11.1, while machine learning has a central position in the estimation of
lake levels, it has been associated with hybrid models in recent years. In addition, it is
observed that researchers have developed methods such as fuzzy logic and artificial
neural network with preprocessing-hybrid models such as wavelet in the past. In
recent years, methods such as deep learning and heuristic optimization algorithms
like particle swarm optimization seem to attract more attention by researchers. In
addition, it can be said that studies on lake-water level are closely related to water
resources, climate change, and drought studies. The fact that there are studies on lake-
water level and the investigation of techniques to develop them shows that the subject
is up to date. In addition, advanced versions support vector regression of (SVR) such
as LSSVR and optimization algorithms such as harmony search algorithm, artificial
bee colony algorithm, particle swarm optimization were not used as a focus in studies
distinguishes the current study from the literature.
Lake-water level information can help with regional water planning, water
management, hydropower plants, commercial navigation, domestic, agricultural,
202 V. Demir et al.

Fig. 11.1 Relationship map of keywords

industrial activities, environmentally friendly construction planning, and the sustain-

able life of all inhabitants of the lake basin [9]. Lake-water level fluctuation is depen-
dent on time series that can be expressed as the results of numerous hydrological and
climatological parameters such as rainfall, runoff, evaporation, temperature, ground-
water and other geoscience variables [10, 11]. However, providing all these param-
eters in the estimation of lake-water levels and determining the impact areas of the
stations is a difficult and expensive process. Instead, making estimations using past
lake-water levels is more practical and economical, as well as more advantageous in
terms of applicability [12].
Lake Beyşehir, Turkey’s largest freshwater lake, and the drinking water reservoir
is located in Central Anatolia between 31°17' and 31°44' E and 37°34' and 37°59'
N [13]. It is one of the most critical water sources, particularly for domestic and
irrigation purposes [14]. Lake Beyşehir has been a first-degree Specially Protected
Area since 1991, and it has been entirely surrounded by two National Parks, Beyşehir
and Kızıldağ, since 1993 [15]. Figure 11.2 depicts a map of the lake’s location. The
lake is approximately 45 km long, 24 km wide and covering an area of 1125 m2 [16].
Lake Beysehir is a tectonic lake located between the Sultan and Anamas Moun-
tains, which form two fault groups running northwest-southeast. Despite its tectonic
origin, a karst phenomenon was responsible for its creation. The lake’s northern and
southern coastlines are shallow, with an average depth of 8.5 m and a maximum
observed depth of 10 m. The lake’s water level and area dramatically change from
season to season and year to year. In addition to direct precipitation, Lake Beysehir
is fed by small surface streams and creeks from the Sultan Mountain and the Anamas
Mountain, and by flow through the Mesozoic calcareous cracks located at the bottom
of the lake. Water losses from the lake occur by surface evaporation, the sinkholes in
the north–south direction and the Beyşehir canal, which diverts the water from the
Modeling Civil Engineering Problems via Hybrid Versions 203

(a) (b)
Fig. 11.2 Study area; a country-wide, b basin

lake for irrigation [17]. The long-term hydrology of the region shows that tempera-
ture and evaporation data have decreased, while lake-water levels, precipitation and
groundwater levels have increased [18].
The trend graph of lake-water levels in the time series is given in Fig. 11.3. When
Fig. 11.3 is examined, an increasing trend is observed in lake-water levels around
the 1950s and 1980s, while those decrease in the following years. With the measures
taken by local authorities since 2008, increasing trends are seen in the data [11, 18].
When the boxplot graph of the data is examined, there are fluctuations around 1123,
on average around 1125–1121 m, according to years.
This study investigates the applicability of at Least squares support vector regres-
sion with various optimization techniques (harmony search algorithm, artificial bee
colony algorithm, particle swarm optimization) for Lake Beyşehir lake-water level
estimation with input variables of historical lake-water level records.

(a) (b)

Fig. 11.3 The time series graph a and boxplot b of Lake Beyşehir lake-water levels
204 V. Demir et al.

2.2 Pile Group Design

Geotechnical engineering is a field of civil engineering that deals with the study of
the behavior of earth materials, such as soil, rock, and groundwater, under various
physical and environmental conditions. The primary objective of geotechnical engi-
neering is to ensure that structures and their foundations are built safely, efficiently,
and sustainably, taking into account the geological and environmental conditions of
the site. Geotechnical engineers perform a wide range of tasks, including site inves-
tigation, laboratory testing of soil and rock samples, analysis and interpretation of
data, design of foundations and retaining structures, slope stability analysis, and risk
assessment.
Comprehensive literature search according to cases in which artificial intelli-
gence, machine learning, neural networks, and deep learning methods application
in civil and geotechnical engineering have been conducted and results have been
visualized via VOSviewer software for conflicting of keywords of the studies [8].
Researchers have frequently employed artificial intelligence-based methods such as
artificial neural networks, machine learning algorithms, and metaheuristic optimiza-
tion algorithms as seen from in Fig. 11.4 usage of them in civil engineering and
geotechnical engineering.
The study which is conducted by Baghbani et al. [19] shows that artificial intelli-
gence methods have been extensively employed in geotechnical engineering topics
like frozen soils and soil thermal properties, rock mechanics, subgrade soil and pave-
ments, landslide and soil liquefaction, slope stability, shallow and piles foundations,
tunneling and tunnel boring machine, dams, and unsaturated soils. The utilization of
algorithm-based methods in civil and geotechnical engineering studies is observed
to be extensive owing to the existence of complex real-world engineering design

Fig. 11.4 Literature review

Modeling Civil Engineering Problems via Hybrid Versions 205

problems. In literature, there is the common usage of these kinds of algorithms to

predict reliable design [20–22] or to achieve the optimum design of geotechnical
structure [23–25].
In geotechnical engineering, pile foundations have been generally preferred as
structure foundations that transmit the load to the depth of soil safely because of
the need to increase living space with the growing population, especially in weak-
characterized soils. Using pile foundations, it is possible to construct multi-story
buildings even though the presence of a weak soil environment. Pile group by using
artificial intelligence methods [26, 27] and optimum design of pile group via meta-
heuristic algorithms [28, 29] were investigated and safely and prosperous results
were acquired.
The working principle of piles under loads involves the transfer of loads from
the superstructure to the soil through the pile. When a pile is subjected to a load,
it transfers the load to the soil or rock mass through skin friction and end bearing.
In the case of a pile group, the behavior is more complex as the loads are shared
among the piles. The interaction between the piles in a group can result in a decrease
in the overall bearing capacity of the group, compared to the sum of individual pile
capacities.
In this study, pile group (PG) bearing capacity are considered for application of
hybrid algorithms. Specifically, the design of the pile group considers an environment
characterized by cohesion-less soil, where the angle of internal friction (Ø) and unit
volume weight (γk ) are key factors. In the design, different soil properties and pile
group dimensions which are pile diameter (D), pile length (Lk ), and space between
pile center to center (S) have been considered and the pile group design is shown in
Fig. 11.5.
In pile group design, the single bearing capacity of a single pile is determined
initially, and then the bearing capacity of the pile group is calculated considering
the total number of piles in the group and group efficiency [30]. In this section,

(a) (b)
Fig. 11.5 Pile group design; a plan and b section
206 V. Demir et al.

Meyerhof’s method [31] is explained for single bearing capacity in homogenous

sandy soil. The ultimate bearing capacity (Qu ) is calculated by summing the load-
carrying capacity of the pile point (Qp ) and frictional resistance derived from the
soil-pile interface (Qs ) (Eq. 11.1).

Qu = Q p + Qs
/
Q p = γk L k N q π D 2 4
/
Q lmt = 0.5Pa Nq tan(φ ' )π D 2 4
Q s = 0.5(q0 + q Lkc )π DL kc + q Lkc (L k − L kc )π DL kc
q Lkc = L kc γk K tan(δ) (11.1)

For bearing capacity factor (Nq ), the values presented by Meyerhof [31] as depending
on the variation of Ø is used. δ is soil-pile friction and its value is taken 0.8Ø in given
formula. K which depends on pile type is the effective earth pressure coefficient and
its value is taken as 1.3 in this formula. The load-carrying capacity of the pile point
(Qp ) should not be great than Qlmt value. Here, Pa means atmospheric pressure (100
kN/m2 ).
The skin frictional resistance, denoted by Qs, exhibits an increasing trend with
depth during the Lkc length but becomes constant thereafter. A conservative estima-
tion of Lkc , assuming it to be 15D, is typically employed in calculations. As a result,
the skin frictional resistance qLkc attains a constant value beyond the Lkc depth. Once
Qu of single pile is determined by aggregating Qp and Qs , it is advisable to apply an
appropriate factor of safety (FS) ranging from 2.5 to 4 to arrive the total allowable
load (Qall ) for each pile (Eq. 11.2). In this study, FS is taken as 3.0.

Qu
Q all = (11.2)
FS
The pile groups are often used to distribute the loads from the structure over a
larger area, which helps to reduce the stress on the individual piles and prevent them
from failure. The bearing capacity of a pile group can be determined by reducing the
total pile bearing capacity with a reduction factor smaller than one. The reduction
factor value of η depends on various factors such as the pile group configuration, pile
spacing (S), and the number of total piles in directions horizontal (Nx ) and vertical
(Ny ). The bearing capacity of the pile group (Qg ), whose center-to-center distance
between piles is taken as 2.5D and 3.5D may be determined via Eq. 11.3 according
to Converse-Labarre formulation [30].

Q g = ηQ u N x N y
[ ]
ϕ (N x − 1)N y + (N y − 1)N x /
η =1− ϕ = tan−1 (D S) (11.3)
90 Nx N y
Modeling Civil Engineering Problems via Hybrid Versions 207

3 Solution Algorithms

3.1 Machine Learning Algorithm

3.1.1 Least Squares Support Vector Regression (LSSVR)

In 1995, Vladimir Vapnik and his colleagues at AT&T Bell Laboratories created
LSSVR, which are used to determine the nonlinear relationship between input vari-
ables and output variables with the least amount of error [32]. Then final version
LSSVR proposed by Suykens and Vandewalle in 1999 and it is applied on chaotic
time series forecasting [33]. Basically, the LSSVR function can be expressed as in
Eq. 11.4.

y(x) = w T ϕ(x) + b (11.4)

where y is the value obtained in x, w is the coefficient vector, ϕ is the mapping

function, and b is the bias term derived by minimizing the upper bound of the gener-
alization error. LSSVR kernels include two alteration parameters (γ, α). The γ is
the regularization constant and the α is the width of the radial basis function kernel.
LSSVR is derived from SVR, which is an excellent technique for solving real-world
problems through the use of regression, function estimation, and classification. The
SVR is based on structural risk minimization (SRM), which gives the least amount of
inaccuracy in forecasting challenges. It is mostly used for signal processing, pattern
recognition, and nonlinear regression estimation [34].
In this current study, various hybrid LSSVR models are used for y and modeling
of Lake Beyşehir lake-water levels. Hence, compared with traditional models, the
LSSVR model is the best at eliminating noise and reduces computational labor.
The hybridization of the model enables the kernel parameters (γ, α) to be obtained
quickly. So, because of these advantages, conventional models can be replaced with
hybrid LSSVR models.

3.1.2 Evaluation Metrics

For evaluation metrics of hybrid LSSVR models, commonly used criteria in the liter-
ature are utilized [35, 36]. These evaluation metrics usually focus on the difference
between observed and predicted values. While this difference is close to zero in
criteria such as RMSE and MAE, close to one in criteria such as E and R2 or close
to 100 in VAF is considered a success criterion.
208 V. Demir et al.

Below are the equations for the evaluation metrics.

(1) Error Sum of Squares (SSE)

∑
n
( )2
SS E = i p − io (11.5)
i=1

(2) Mean Squared Error (MSE)

⌜
| n
1|∑
MSE = √ (i o − i p )2 (11.6)
n i=1

(3) Mean Root Square Error (MRSE)

⌜
| n
1|∑
M RS E = √ (i o − i p )2 (11.7)
n i=1

(4) Root Mean Square Error (RMSE)

⌜
| n
|1 ∑
RMSE = √ (i o − i p )2 (11.8)
n i=1

(5) Mean absolute error (MAE)

1 ∑ || |
n
M AE = i p − io | (11.9)
n i=1

(6) Mean Absolute Relative Error (MARE)

n | |
1 ∑ || i p − i o ||
M A R E = 100 (11.10)
n i=1 | i p |

(7) Coefficient of Determination (R2 )

∑
n
(i o − i p )2
i=1
R =1−
2
(11.11)
(i o − i o )2
Modeling Civil Engineering Problems via Hybrid Versions 209

(8) Variance account factor (VAF)

( )
var(i p − i o )
V AF = 1 − x100 (11.12)
var(i p )

(9) Adjusted Coefficient of Efficiency (E)

n |
∑ |
|i o − i p |
i=1
E(%) = 1 − | | (11.13)
∑n
|i o − i o |
i=1

In these equations, io and ip show the observed and predicted values, respectively.
n represents the number of data, and the line over io or ip represents the average. In
addition, scatter plots, violin plots and Taylor diagrams are utilized in this study to
compare methods. These images show graphically how closely the models match
the observations [37–39]. The statistical information of the violin diagram used in
the study is given in Fig. 11.6 [40].

Fig. 11.6 Statistics of the

violin diagram
210 V. Demir et al.

3.2 Metaheuristic Optimization Algorithms

Metaheuristic optimization algorithms are designed to solve complex optimization

problems without requiring any initial values. One of the main advantages of these
algorithms are their ability to find the approximate global solution by exploring
many directions without getting stuck in a local optimum. This is achieved through
a randomization process that introduces diversity into the search process, allowing
the algorithm to escape from local optima. It has been successfully used in various
fields, including engineering, economics, and computer science [41].

3.2.1 Harmony Search Algorithm (HS)

The HS is a robust metaheuristic optimization algorithm based on the principles of

musical harmony [42]. It was first introduced in 2001 by Geem et al. as a metaheuristic
optimization algorithm inspired by the process of musicians improvising to find the
best harmony during a musical performance. Figure 11.7 displays a detailed flowchart
outlining the steps of the HS algorithm.
The HS algorithm works by maintaining a set of solutions for number of design
variables (NVar), called the harmony memory (HM), which is updated at each
iteration (iter) to improve the quality of the solutions.
The HS algorithm generates new candidate solutions by combining existing design
variables in HM whose row number is defined as harmony memory size (HMS).
Harmony memory consideration rate (HMCR) and pitch adjusting rate (PAR) param-
eters play important roles in the improvisation of a new solution vector (NCV). If
HMCR is smaller than the random number (ran) ranging between [0,1], the existing
solution may update its value with a new one from HM (with the possibility of
HMCR). Otherwise, the new solution is taken from the design space (DS) with the
possibility of 1-HMCR. In the case of selecting the new solution from HM, it is
checked whether the NCV value is replaced by its nearest lower and upper neigh-
bours according to the possibility of PAR or 1-PAR. It is similar to how musicians
might create new harmonies by combining existing musical notes since the musi-
cian performs a note from memory or a note near to in his/her mind. End of this
improvisation process, the NCV quality is assessed according to objective function
value. This value is added to the harmony memory matrix only if it is superior to the
lowest value of the harmonic vector in the matrix, and this lowest value is replaced
by the new value. Until the current iteration number reaches the maximum iteration
number (max-iter), the improvisation of NCV and updating HM continue.
Modeling Civil Engineering Problems via Hybrid Versions 211

Fig. 11.7 The flowchart of HS algorithm

212 V. Demir et al.

3.2.2 Artificial Bee Colony Algorithm (ABC)

Karaboğa [43] proposed the ABC algorithm as a metaheuristic optimization algo-

rithm based on swarm intelligence promoting communication between individuals in
the population to ensure survival and meeting basic needs such as nutrition, defence,
and migration. In Fig. 11.8, the flowchart of ABC algorithm is presented, including
steps and mathematical formulations [44].
The ABC algorithm begins with the definition of population size (NPop), number
of food sources (NFS), and limit parameters, as well as the determination of initial
food sources (xij ) for design variable (N). Employer bees are then dispatched to the
food source to assess the fitness value (fitnessi ) of the food source by generating
a new food source (vij ) in the vicinity of the current food source. After the fitness
value of the new food source is determined using the objective function value (fi ), the
better food source is selected between xij and vij , taking into account the calculated
fitness value using greedy selection. If the new food source does not have a better
fitness value, the solution failure counter is incremented by one. However, if the
food source with a better fitness value is selected and memorized, a new solution
is developed, and the solution failure counter is reset. The employed worker bees
then share information about the food sources they discover near the hive with the
onlooker bees, who choose their food source based on the nectar amount, location,
and quality of the food sources. The onlooker bees consider the probability value
(pi ) calculated based on the fitness value of the food source during the selection
process. If the randomly generated number (Øij ) falls in the range [0,1] and is greater
than pi , onlooker bees generate a new food source (vij ) like worker bees. The non-
solution counter is updated depending on whether the new food source is better or
worse than the old food source, and this process continues until all onlooker bees are
dispatched to food sources. The worker and onlooker bees continue their cycles until
it is determined, by checking the non-solution counter, whether the nectar in a food
source has run out. If the solution development counter exceeds the limit value, the
food source is abandoned, the employed bee becomes a scout bee, and the process
of searching for a random food source commences. This loop continues until the
current iteration number reaches the maximum number of iterations, after which the
algorithm terminates.

3.2.3 Particle Swarm Optimization (PSO)

The PSO algorithm was introduced by Kennedy and Eberhart in 1995 [45] as a swarm
intelligence-based algorithm inspired by social metaphors such as flocks of fish and
birds. The PSO algorithm is a meta-heuristic optimization technique that mimics the
communication and foraging behaviors observed in animals.

In the PSO algorithm, each individual in the swarm is represented as a particle

that adjusts its position by utilizing its previous experience to attain the best position
Modeling Civil Engineering Problems via Hybrid Versions 213

Fig. 11.8 The flowchart of ABC algorithm

214 V. Demir et al.

in the swarm, ultimately resulting in the optimal target value. Figure 11.9 illustrates
a summary flowchart of the PSO algorithm.
The PSO algorithm employs a set of parameters, namely c1 , c2 (accelerate
constants for local and global, respectively) ω (inertia weight), NPop (population
size), and the maximum number of iterations, to conduct an iterative search for an
optimal outcome. Initially, the algorithm generates random positions and resets the
initial velocity for each particle. Subsequently, the algorithm evaluates the objective
function value in the problem definition for each particle. Based on the updated objec-
tive function value, the algorithm determines the best solution for both the individual
and the swarm. The algorithm updates the velocities and positions of the individuals

Fig. 11.9 The flowchart of PSO algorithm

Modeling Civil Engineering Problems via Hybrid Versions 215

over time by taking into account the values of c1 , c2 , and ω. The objective function
value is calculated by considering the updated speed and location information while
also ensuring adherence to the constraints. The algorithm operates until the stopping
criterion, which is reaching the maximum number of iterations, is met. It is repeated
until the best solutions for the individuals and the optimal solution for the swarm are
obtained in each iteration.

4 Implementation of Hybrid Optimization Algorithms

In this section, the employed datasets of lake-water levels of Lake Beyşehir and pile
group bearing capacity for obtaining optimal values of γ and α to implementation of
hybrid optimization algorithms, and design optimization problem are explained in
detail.

4.1 Hybrid Optimization Algorithms

In this study, hybrid optimization algorithms versions which are obtained by

combining a metaheuristic optimization algorithm and a machine learning algorithm
have been improved. While the LSSVR method has been considered as a machine
learning algorithm, the HS, ABC, and PSO algorithms have been used as meta-
heuristic optimization algorithms given in Figs. 11.7 , 11.8, and 11.9, respectively.
Improved hybrid algorithms named LSSVR-HS, LSSVR-ABC, and LSSVR-PSO
have been implemented to the data sets of LL and PG problems given in this section to
optimize the kernels parameters (γ and α) of the LSSVR. Figure 11.10 illustrates the
flowcharts of hybrid algorithms by explaining the implementation of metaheuristic
optimization algorithms with the LSSVR method.

4.2 Dataset of Lake Level(LL)

In this study, forecasting of monthly lake-water levels are investigated by using

four different metaheuristic optimization algorithms and LSSVR for forecasting one
month ahead of water levels of Lake Beyşehir. The data sample consists of a 73-year
(1950–2022) record of average lake-water level (m). 75% of all data is used for the
training phase and 25% of the remaining and more recent data are used for the testing
phase. Basic statistics of training and test data are given in Table 11.1.
216 V. Demir et al.

Fig. 11.10 The flowchart of hybrid algorithms

Table 11.1 Statistical information for LL problem

Data set Period Mean Minimum Maximum Standard Kurtosis coefficient Skewness coefficient
deviation
All 1950–2022 1123.13 1121.03 1125.50 1.009 −0.862 0.265
Training 1950–2004 1123.30 1121.03 1125.50 1.034 −1.035 0.079
Testing 2004–2022 1122.67 1121.15 1124.63 0.762 −0.169 0.333

Lake-water levels from the past months (lag times) are used as input data in
design problem modelling. Correlation, autocorrelation, and partial correlation anal-
yses are performed in the literature for the number of inputs. In this study, the rela-
tionship between the data at T time and the data at T-N months is examined with
autocorrelation graphs and correlation (Figs. 11.11 and 11.12).
The autocorrelation graph shows that the data in the time series have strong rela-
tionships with the data in the past months. This relationship shows a decreasing trend
until the 8th month and then increases until the 12th month. After the 12th month, it
decreases again and the similar periodicity continues. In the study, the data up to the
Modeling Civil Engineering Problems via Hybrid Versions 217

Fig. 11.11 Autocorrelation graph

Fig. 11.12 Correlation graph of LL

12th month, which is the last period of the increase, are used as input data of this
study.
Similarly, Fig. 11.11, the correlation graph in Fig. 11.12. shows strong and
increasing correlations up to 12 months. In addition, the correlation between T and
T-N months can also be seen. As a result, when both graphs were examined, it is seen
218 V. Demir et al.

that there is a strong relationship between the data of the past month up to 12 months
and the data at time T. For this reason, a prediction model with 12 inputs from the
1st to the 12th month and 1 output that expresses the same T in the modelling is
established in the study.

4.3 Date Set of Pile Group (PG)

In the synthetic dataset of the pile group, input parameters of angle of internal friction
(Ø), unit volume weight (γk ), pile diameter (D), pile length (Lk ), and space between
pile center to center (S) are taken as input parameters to predict bearing capacity
of pile group (Qg). A dataset consisting of 1200 data is created by utilizing values
within the range of the lower and upper bounds of input parameters shown in Table
11.2.
Based on the prescribed input parameters, Eq. 11.3 has been employed to calculate
the bearing capacity of a pile group, yielding a total of 1200 distinct values. 75% and
25% of the 1200 data are used in the training phase and testing phase, respectively.
It is suggested that the application of normalization in case of existing big differ-
ences in the ranges of different features [46]. For instance, normalization is useful
because pile group bearing capacity and pile diameter features have different units and
magnitudes. Pile group bearing capacity is typically measured in kilopascal (kPa),
while pile diameter is measured in meters (m). Without normalization, the pile group
bearing capacity could have a much larger range of values than the pile diameter,
which could lead to bias in the analysis. To provide a similar scale between [0–1]
for all features and output parameter, the data set was normalized using including
min–max scaling (Eq. 11.14).

' X − X min
X = (11.14)
X max − X min

Table 11.3 presents the descriptive statistics of datasets for all, training, and testing.
A correlogram which is a graphical representation of a correlation matrix is shown
in Fig. 11.13 to identify patterns and relationships between features and output.

Table 11.2 Data set property

Features Lower–upper bounds
Ø 30–36°
γk 18–20 kN/m3
D 0.80–1.40 m
Lk 15-30 m
S 2.5D-3.5D
Modeling Civil Engineering Problems via Hybrid Versions 219

Table 11.3 Statistical information for PG problem

Data set Data Mean Minimum Maximum Standard Median Kurtosis coefficient Skewness coefficient
deviation
All 1200 0.3473 0 1 0.2271 0.3192 -0.6293 0.5144
Training 900 0.3472 0 1 0.2270 0.3132 -0.6713 0.5141
Testing 300 0.3476 0.0044 0.9805 0.2278 0.3276 -0.4938 0.5181

Fig. 11.13 Correlation

graph of PG

While dark colored or close to 1.0 show strong correlation, light colored or close
to 0.0 means that correlation is weak or uncorrelated. Also, values greater than 0.0
corresponds positive effect, and values smaller than 0.0 have a negative effect.

4.4 Design Optimization Problem

As beforementioned, in this study, the LSSVR method is used to occur the hybrid
versions with metaheuristic optimization algorithms which are the HS algorithm,
ABC algorithm, and PSO algorithm, to optimize the kernels parameters (γ and α)
that are used on the prediction model for both design problems of LL and PG.
Design space (DS) by using different γ and α values ranging between [1, 100]
both of them have been formed for the design optimization problem of improved
hybrid solution models. The quality of initial solutions or new solutions generated
randomly corresponds to objective function of the design optimization problem.
Since achieving the optimum values of the parameters of LSSVR depends on the test
performance of the model trained for the data sets used, the evaluation metrics given
in Eqs. 11.5 to 11.13 has been considered as the objective function and constraints.
Design variables, objective function, and constraints of the optimization problem are
demonstrated in Eq. 11.15.
220 V. Demir et al.

In accordance with the provided equation, the summation of all evaluation metrics
constitutes the objective function, whereas each individual metric is considered a
distinct constraint.
[ ]T
Design variables:−
→x = γ, α
Minimize:
f(−
→
x ) = R 2 + SS E + M S E + M RS E + R M S E
+ M AE + M A R E + V AF/100 + E
Subject to:
g (−
→
[1−6]x ) = [SSE MSE MRSE RMSE MAE MARE]
g[7−9] (−
→
x ) = [1 − VAF/100 1 − E 1 − R2 ] (11.15)

Deb’s rules which is a simple and intuitive way to evaluate the feasibility of
a solution with respect to its constraints were utilized as the constraint-handling
technique during the optimization procedure [47]. In this process, constraint values
that don’t satisfy limits and is not equal to zero, are summed and a penalty value is
obtained for each solution. To determine the optimal solution in the implementation
of Deb’s Rule, three principles are employed the ensuing; (i) a feasible solution is
preferred to an unfeasible one, (ii) if both designs have suitable constraints, then the
design with a better objective function value is preferred, (iii) if neither design is
feasible, the design with lesser constraint violation is preferred among the two.
In determination of the optimum algorithms parameters for HS, ABC, and PSO
algorithms like HMS, HMCR, PAR, nPop, NFS, MR, c1 , c2 , and w, pressure vessel
design (PVD), which is one of the well-known benchmark engineering design prob-
lems, has been employed [48]. It is a common complex engineering design opti-
mization problem that involves determining the minimum cost of a pressure vessel,
which takes into account factors such as material, forming, and welding, while also
satisfying constraints such as the thickness of the heads and shell, prescribed working
pressure values, volume, and the length of the shell section (excluding the head). In
this particular case, the design variables are the thicknesses of the shell (Ts ) and head
(Th ), the inner radius (R), and the length of the cylindrical section of the vessel.
Many trails have been conducted considering suggested values for HS [49], ABC
[44], and PSO [50] algorithm parameters in the literature. Considered algorithm
parameters for these analyses are given in Table 11.4. Obtained optimum results
according to values in Table 11.4 are tabulated in Table 11.5.
Modeling Civil Engineering Problems via Hybrid Versions 221

Table 11.4 Metaheuristic

Metaheuristic Algorithm parameters
algorithm parameters
Algorithm
HS HMS: 20, HMCR:0.95, PAR:0.40,
ABC NPop: 50, NFS:NPop/2, limit:
NFSxN, MR: 0.40
PSO NPop: 50, c1 : 1.9, c2 : 2, w:
(MaxIter-iter)/MaxIter

Table 11.5 Optimum values and comparison of the best solutions in literature for PVD problem
Optimum solutions Ts Th R L f(x) ($)
HSPS 0.8022 0.3961 41.2146 188.0063 5975.2258
ABCPS 0.8615 0.4245 44.4230 149.8970 6067.2776
PSOPS 0.778391 0.384759 40.3312 199.839 5886.1585
Kannan and Kramer [51] 1.2500 0.6250 50.000 120.0000 7198.2000
Deb [52] 0.9375 0.5000 48.3290 112.6790 6410.3810
Coello [53] 0.8125 0.4375 40.3239 200.0000 6288.7445
Çarbaş and Saka [54] 0.8125 0.4375 42.0984 176.6366 6059.7143
PS: Present Study

5 Hybrid Optimization Algorithms Results

5.1 Optimization Analyses

Improved hybrid algorithms (LSSVR-HS, LSSVR-ABC, LSSVR-PSO) have been

applied for 1000 maximum iterations (MaxIter) and 30 independent run numbers for
all optimization analyses. The Tables 11.6 and 11.7 present the statistical assessments
obtained from optimization analyses that have fulfilled the prescribed constraints for
the LSSVR-HS, LSSVR-ABC, and LSSVR-PSO algorithms for LL and PG prob-
lems, respectively. In Tables 11.6 and 11.7, there are differences between the results
statistics of 30 runs and the best run. This difference shows the performance improve-
ment of the models in reaching the optimum level. When the methods are compared,
both problems show slightly differences. In this case, it can be interpreted that all
three optimization techniques successfully support each other and give consistent
results. In Figs. 11.14 and 11.15, fmin objective function values of 30 runs and the
iteration history graph of the best run are demonstrated, respectively. Furthermore,
optimal solutions of the LL problem in Fig. 11.16 and the PG problem in Fig. 11.17.
are given for different hybrid algorithms.
In Fig. 11.14., as the iterations change in HS for both problems, the feature of
searching for the best shows more variation compared to HS and ABC. It is also
seen that PSO responds less to iteration change. In Fig. 11.15, the fmin change of
222 V. Demir et al.

Table 11.6 Statistical results for fmin optimum values of 30 runs

Problem Hybrid algorithm LSSVR-HS LSSVR-ABC LSSVR-PSO
LL Best 3.58263571 3.58005481 3.58005481
Mean 5.02602678 3.59714006 3.58005481
Worst 6.31815550 3.77392104 3.58005481
StD 0.85452026 0.04597511 0.00000000
Median 5.20494421 3.58005481 3.58005481
PG Best 3.04532573 3.04518859 3.04518859
Mean 3.08221120 3.04528597 3.04518859
Worst 3.14777058 3.04758083 3.04518859
StD 0.03385287 0.00043865 0.00000000
Median 3.06616057 3.04518859 3.04518859

Table 11.7 Statistical results for fmin optimum values of iterations for best run
Problem Hybrid algorithm LSSVR-HS LSSVR-ABC LSSVR-PSO
LL Best 3.58263571 3.58005481 3.58005481
Mean 3.59449892 3.58814613 3.58955693
Worst 3.78035587 3.95117693 5.88002689
StD 0.04719249 0.05233472 0.10416052
Median 3.58263571 3.58005481 3.58005481
PG Best 3.04532573 3.04518859 3.04518859
Mean 3.04534923 3.04554556 3.04535304
Worst 3.04767603 3.07543725 3.11206087
StD 0.00023503 0.00302603 0.00225448
Median 3.04532573 3.04518859 3.04518859

best run according to iteration is shown. For both problems, asymptotic change is
observed after 20 iterations. Therefore, there was no need for modeling over 100
iterations. When the iteration reaches 20, significant fmin decreases are seen in both
problems. Although the increase in the iteration of the problems did not show very
large decreases for the HS and ABC models, the increase in the iteration for the PSO
decreased the fmin value.
In Figs. 11.16. and 11.17., the two kernel parameters of the LSSVR model are
shown in three dimensions, with α approaching 1 and γ approaching 100 as the
solution that gives the least error. While the hybrid models approach these parameters,
the variable distribution is seen in LSSVR-HS, followed by LSSVR-ABC, and the
model that differs least or generally remains constant in its estimation is observed as
LSSVR-PSO.
Modeling Civil Engineering Problems via Hybrid Versions 223

5.2 Performance of Hybrid Algorithms

In the optimization analyses, the evaluation metrics given in Eqs. 11.5 to 11.13
have been considered as design constraints. The mean of 30 runs for 9 evaluation
metrics obtained using hybrid algorithms which are LSSVR-HS, LSSVR-ABC, and
LSSVR-PSO is demonstrated in Table 11.8 and Table 11.9 for LL and PG problems,
respectively.
In Table 11.8 and Table 11.9, the acceptable results are obtained for both design
problems. Although the success of LSSVR is known in the literature in estimating
similar problems, reaching these solutions quickly and reliably with optimization
algorithms is seen as the success of this study. For example, in the study performed
with LSSVR in Lake Michigan in the USA, the RMSE MAE R2 parameters are
0.0427, 0.0332, and 0.9890, respectively [9]. Obtaining results close to these criteria
in this current chapter study, although not directly showing the success of the model,
can show the accuracy of hybridization.

Fig. 11.14 fmin optimum values for 30 runs

224 V. Demir et al.

6 Discussions

In this section, the solution model performances are discussed with graphs obtained
using various statistics. These graphs are very good at seeing small differences
between models [55]. For example, the violin graph is a graph drawn according
to the statistics in Fig. 11.6., and the observed data in the first row is represented,
while the following graphs are drawn according to the results obtained in the best
prediction models of the optimized kernel parameters (γ, α). The similarity between
these drawings shows the model performances. Another graph is the Taylor diagram,
which is obtained according to the observed data, correlation, RMSE (obtained by
Eq. 11.8) relationship. The observed data in the graph are represented by the red dot
positioned on the x-axis, and being close to this point indicates the success of the
models. In Fig. 11.20, Violin plots of the best results are given for the LL and PG
problems, and in Fig. 11.21 Taylor diagram are given for the same problems.
The evaluation metrics of 30 runs using LSSVR-HS, LSSVR-ABC, and LSSVR-
PSO algorithms have been presented in Figs. 11.18 and 11.19 for the LL and PG
problems, respectively.
In the LL design problem, it is very difficult to see the difference between the
methods visually. This situation is also observed in the Taylor diagram by placing

Fig. 11.15 fmin optimum values versus iteration for best run I
Modeling Civil Engineering Problems via Hybrid Versions 225

Fig. 11.16 Optimum values of LL problem for 30 runs; a LSSVR-HS, b LSSVR-ABC, c LSSVR-
PSO
226 V. Demir et al.

Fig. 11.17 Optimum values of PG problem for 30 runs; a LSSVR-HS, b LSSVR-ABC, c LSSVR-
PSO
Modeling Civil Engineering Problems via Hybrid Versions 227

Table 11.8 Evaluation

Hybrid algorithm LSSVR-HS LSSVR-ABC LSSVR-PSO
metrics of optimum design
for LL problem R2 0.995619 0.995639 0.995639
SSE 0.55516 0.55262 0.55262
MSE 0.00257 0.002558 0.002558
MRSE 0.003449 0.003442 0.003442
RMSE 0.050697 0.050581 0.050581
MAE 0.032621 0.032534 0.032534
MARE 0.000029 0.000029 0.000029
VAF (%) 99.5585 99.5605 99.5605
E 0.946905 0.947047 0.947047

Table 11.9 Evaluation

Hybrid algorithm LSSVR-HS LSSVR-ABC LSSVR-PSO
metrics of optimum design
for PG problem R2 0.997648 0.997655 0.997655
SSE 0.036542 0.036436 0.036436
MSE 0.000122 0.000121 0.000121
MRSE 0.000637 0.000636 0.000636
RMSE 0.011037 0.011021 0.011021
MAE 0.007512 0.007494 0.007494
MARE 0.033469 0.033363 0.033363
VAF (%) 99.76450 99.7652 99.76520
E 0.960714 0.960811 0.960811

the models on top of each other. HS and PSO algorithms are superior to the observed
data with a small difference, especially in terms of mean, deviation and end data in
the violin graph. This difference is observed more in low data. For the PG design
problem, ABC and PSO algorithms perform very close to each other in the Violin
graph, whereas in HS algorithm the differences are observed when the observed
data is low value. However, it is also seen that HS algorithm predicts higher values
closer to the observed values than ABC and PSO algorithms. According to the Taylor
diagram, the most successful results are observed in ABC, PSO and HS algorithms,
respectively.

7 Conclusions

In this study, LSSVR, one of the machine learning methods, and HS, ABC and
PSO, which are metaheuristic optimization algorithms, are used for LL and PG, two
different design problems investigated and predicted in water resources field of civil
engineering. Hybrid solution models are created by using optimization techniques
228 V. Demir et al.

2
R SSE MSE

MRSE RMSE MAE

MARE VAF E
Fig. 11.18 Evaluation metrics for 30 runs-based LL problem

to reach the optimum solution of the two kernel parameters (γ, α) of LSSVR. In this
study, 75% of the observed or produced data for LL and PG prediction are used in
the training phase and the remaining 25% is used in the testing phase. As evaluation
metrics, 9 different evaluation metrics and various statistical graphs are used, and
the following principal conclusions are deducted;
• In the solution of engineering design problems, the hybrid models are obtained
by integrating the LSSVR method with the HS, ABC and PSO algorithms, and
it is seen that the evaluation criteria obtained with this novel solution models are
mighty compatible with the literature.
Modeling Civil Engineering Problems via Hybrid Versions 229

2
R SSE MSE

MRSE RMSE MAE

MARE VAF E
Fig. 11.19 Evaluation metrics for 30 runs-based PG problem

• The hybridization of LSSVR with metaheuristic optimization algorithms has

transformed the time-consuming nested loops in modeling into meaningful loops,
provided the solution of the model to be understood and added robustness to the
solution model.
• The top-performing hybrid solution models for the LL design problem are
LSSVR-ABC and LSSVR-PSO, which achieved metric values of RMSE =
0.050581, MAE = 0.032534, and R2 = 0.995639.
230 V. Demir et al.

(a) (b)
Fig. 11.20 Taylor diagram; a LL problem, b PG problem

(a) (b)
Fig. 11.21 Violin graph; a LL problem, b PG problem

• Comparable to the LL design problem, LSSVR-ABC, and LSSVR-PSO hybrid

solution models demonstrate outstanding performance in resolving the PG design
problem. This is supported by their metric values of RMSE = 0.011021, MAE =
0.0074940.032534, and R2 = 0.997648, respectively.
Moreover, the five main limitations of this study can be stated as follows; (i)
the use of 73 years of actual data in the estimation of lake-water levels, which is a
researched topic in water resources engineering, (ii) using 1200 synthetic data for pile
group design, one of the researched topics in geotechnical engineering, (iii) the use
of correlation and autocorrelation analysis for input selection, (iv) the usage of three
different optimization techniques such as HS, ABC and PSO in the hybridization of
LSSVR, (v) the use of visual comparison criteria in addition to 9 different evaluation
metrics.
Modeling Civil Engineering Problems via Hybrid Versions 231

References

1. Lu, P., Chen, S., Zheng, Y.: Artificial intelligence in civil engineering. Math. Probl. Eng. (2012).
https://doi.org/10.1155/2012/145974
2. Wikipedia: Artificial intelligence (2023). https://en.wikipedia.org/wiki/Artificial_intelligence
3. Feurer, M., Hutter, F.: Hyperparameter Optimization, 3–33 (2019). https://doi.org/10.1007/
978-3-030-05318-5_1
4. Yaseen, Z.M., Ali, Z.H., Salih, S.Q., Al-Ansari, N.: Prediction of risk delay in construction
projects using a hybrid artificial intelligence model. Sustain 12:1514–12:1514 (2020).https://
doi.org/10.3390/SU12041514
5. Hydraulic and Water Resources Engineering | Civil Engineering—McGill University. https://
www.mcgill.ca/civil/undergrad/areas/water. Accessed 14 Mar 2023
6. Water Resources Engineering, 2nd Edition | Wiley. https://www.wiley.com/en-us/Water+Res
ources+Engineering%2C+2nd+Edition-p-9780470574164. Accessed 14 Mar 2023
7. Altunkaynak, A., Sen, Z.: Fuzzy logic model of lake water level fluctuations in Lake Van,
Turkey. Theor. Appl. Climatol. 90, 227–233 (2007)
8. VOSviewer: Welcome to VOSviewer. In: 2022 Cent. Sci. Technol. Stud. Leiden Univ.
Netherlands (2022)
9. Demir, V.: Enhancing monthly lake levels forecasting using heuristic regression techniques
with periodicity data component: application of Lake Michigan. Theor. Appl. Climatol. 148,
915–929 (2022). https://doi.org/10.1007/s00704-022-03982-0
10. Li, X.Y., Xu, H.Y., Sun, Y.L., Zhang, D.S., Yang, Z.P.: Lake-level change and water balance
analysis at lake Qinghai, West China during recent decades. Water Resour Manag 21, 1505–
1516 (2007). https://doi.org/10.1007/s11269-006-9096-1
11. Demir, V., Keskin, A.Ü.: Water level change of lakes and sinkholes in Central Turkey under
anthropogenic effects. Theor. Appl. Climatol. 142, 929–943 (2020). https://doi.org/10.1007/
s00704-020-03347-5
12. Şen, Z., Kadioğlu, M., Batur, E.: Stochastic modeling of the van lake monthly level fluctuations
in Turkey. Theor. Appl. Climatol. 65, 99–110 (2000). https://doi.org/10.1007/s007040050007
13. Şener, Ş, Şener, E., Nas, B., Karagüzel, R.: Combining AHP with GIS for landfill site selection: a
case study in the Lake Beyşehir catchment area (Konya, Turkey). Waste Manag. 30, 2037–2046
(2010). https://doi.org/10.1016/j.wasman.2010.05.024
14. Nas, B., Berktay, A., Aygun, A., Karabork, H., Ekercin, S.: Seasonal and spatial variability of
metals concentrations in Lake Beysehir, Turkey. Environ. Technol. 30, 345–353 (2009). https:/
/doi.org/10.1080/09593330902752984
15. Karabork, H.: Selection of appropriate sampling stations in a lake through mapping. Environ.
Monit. Assess. 163, 27–40 (2010). https://doi.org/10.1007/s10661-009-0813-0
16. Altindaǧ, A., Yiǧit, S.: Assessment of heavy metal concentrations in the food web of lake
Beyşehir, Turkey. Chemosphere 60, 552–556 (2005). https://doi.org/10.1016/j.chemosphere.
2005.01.009
17. Yarar, A., Onucyıldız, M., Copty, N.K.: Modelling level change in lakes using neuro-fuzzy and
artificial neural networks. J. Hydrol. 365, 329–334 (2009). https://doi.org/10.1016/j.jhydrol.
2008.12.006
18. Demir, V.: Trend Analysis of Lakes and Sinkholes in the Konya Closed Basin, in Turkey.
Springer, Netherlands (2022)
19. Baghbani, A., Choudhury, T., Costa, S., Reiner, J.: Application of artificial intelligence in
geotechnical engineering: a state-of-the-art review. Earth-Science Rev 228, 103991 (2022).
https://doi.org/10.1016/J.EARSCIREV.2022.103991
20. Koopialipoor, M., Ramesh Murlidhar, B., Hedayat, A., Jahed Armaghani, D., Gordan, B.,
Tonnizam Mohamad, E.: The use of new intelligent techniques in designing retaining walls 36,
283–294 (2020). https://doi.org/10.1007/s00366-018-00700-1
21. Yücel, M., Bekdaş, G., Nigdeli, S.M., Kayabekir, A.E.: An artificial intelligence-based predic-
tion model for optimum design variables of reinforced concrete retaining walls. Int. J. Geomech.
21, 04021244 (2021). https://doi.org/10.1061/(ASCE)GM.1943-5622.0002234
232 V. Demir et al.

22. Uray, E., Demir, V., Keskin, A.Ü., Tan, Ö.: Use of artificial neural networks in stability control
of cantilever retaining walls. In: International Civil Engineering and Architecture Conference,
pp. 469–478 (2019)
23. Kalemci, E.N., İkizler, S.B., Dede, T., Angın, Z.: Design of reinforced concrete cantilever
retaining wall using Grey wolf optimization algorithm. Structures 23, 245–253 (2020). https:/
/doi.org/10.1016/j.istruc.2019.09.013
24. Uray, E., Carbas, S., Geem, Z.W., Kim, S.: Parameters optimization of taguchi method inte-
grated hybrid harmony search algorithm for engineering design problems. Math. 10, 327 10:327
(2022).https://doi.org/10.3390/MATH10030327
25. Uray, E., Carbas, S., Erkan, I.H., Tan, O.: Parametric investigation for discrete optimal design
of a cantilever retaining wall. Chall J. Struct. Mech. 5, 108 (2019). https://doi.org/10.20528/
cjsmec.2019.03.004
26. Kumar, M., Samui, P.: Reliability analysis of settlement of pile group in clay using LSSVM,
GMDH, GPR. Geotech. Geol. Eng. 38, 6717–6730 (2020). https://doi.org/10.1007/S10706-
020-01464-6/FIGURES/11
27. Fatehnia, M., Amirinia, G.: A review of genetic programming and artificial neural network
applications in pile foundations. Int. J. Geo-Eng. 9, 1–20 (2018). https://doi.org/10.1186/S40
703-017-0067-6/TABLES/8
28. Öser, C., Temür, R.: (1AD) Optimization of Pile Groups Under Vertical Loads Using
Metaheuristic Algorithms. https://services.igi-global.com/resolvedoi/resolve.aspx?doi=104
018/978-1-5225-4766-2.ch013 276–298. https://doi.org/10.4018/978-1-5225-4766-2.CH013
29. Bekdaş, G., Arama, Z.A., Kayabekir, A.E., Geem, Z.W.: Optimal design of cantilever soldier
pile retaining walls embedded in frictional soils with harmony search algorithm. Appl. Sci., 10
(2020).https://doi.org/10.3390/app10093232
30. Das, B.: Principles of Foundation Engineering 6th Edition (2007)
31. Meyerhof, G.G.: Bearing capacity and settlement of pile foundations. J. Geotech. Eng. Div.
102, 197–228 (1976). https://doi.org/10.1061/AJGEB6.0000243
32. Cortes, C., Vapnik, V., Saitta, L.: Support-vector networks. Mach. Learn. 203, 20:273–297
(1995).https://doi.org/10.1007/BF00994018
33. Suykens, J.A.K., Vandewalle, J.: Least squares support vector machine classifiers. Neural
Process Lett. 9, 293–300 (1999). https://doi.org/10.1023/A:1018628609742/METRICS
34. Kisi, O., Shiri, J., Demir, V.: Hydrological time series forecasting using three different heuristic
regression techniques. Handb. Neural Comput., 45–65 (2017).https://doi.org/10.1016/B978-0-
12-811318-9.00003-X
35. Legates, D.R., McCabe, G.J.: Evaluating the use of ‘goodness-of-fit’ measures in hydrologic
and hydroclimatic model validation. Water Resour. Res. 35, 233–241 (1999). https://doi.org/
10.1029/1998WR900018
36. Yaseen, Z.M.: An insight into machine learning models era in simulating soil, water bodies and
adsorption heavy metals: review, challenges and solutions. Chemosphere 277, 130126 (2021).
https://doi.org/10.1016/j.chemosphere.2021.130126
37. Taylor, K.E.: Summarizing multiple aspects of model performance in a single diagram. J.
Geophys. Res. Atmos. 106, 7183–7192 (2001). https://doi.org/10.1029/2000JD900719
38. Citakoglu, H., Demir, V.: Forecasting solar radiation using deep learning: the case of Turkey.
In: International World Energy Conference, pp. 167–175 (2021)
39. Demir, V., Yaseen, Z.M.: Neurocomputing intelligence models for lakes water level forecasting:
a comprehensive review. Springer, London (2022)
40. Legouhy, A.: al_goodplot—boxblot & violin plot. In: MATLAB Cent. Mathworks (2021)
41. Arora, J.S.: Introduction to Optimum Design. Introd. to Optim. Des. (2017).https://doi.org/10.
1016/C2013-0-15344-5
42. Geem, Z., Kim, J., Loganathan, G.: A new heuristic optimization algorithm: harmony search.
Simulation 76, 60–68 (2001). https://doi.org/10.1177/003754970107600201
43. Karaboga, D.: An idea based on honey bee swarm for numerical optimization (2005)
44. Akay, B., Karaboga, D.: A modified artificial bee colony algorithm for real-parameter
optimization. Inf. Sci. (Ny) 192, 120–142 (2012). https://doi.org/10.1016/j.ins.2010.07.015
Modeling Civil Engineering Problems via Hybrid Versions 233

45. Kennedy, J., Eberhart, R.: Particle swarm optimization. Proc ICNN’95—International Confer-
ence Neural Networks 4, pp. 1942–1948. https://doi.org/10.1109/ICNN.1995.488968
46. Ali, P.J.M., Faraj, R.H.: Data normalization and standardization—google Dokümanlar. Mach.
Learn. Tech. Rep., 1, 1–6 (2014). https://doi.org/10.13140/RG.2.2.28948.04489
47. Deb, K.: An efficient constraint handling method for genetic algorithms. Comput. Methods
Appl. Mech. Eng., 186 (2000). https://doi.org/10.1016/S0045-7825(99)00389-88
48. Sandgren, E.: Nonlinear integer and discrete programming in mechanical design optimization.
J. Mech. Des. 112, 223–229 (1990). https://doi.org/10.1115/1.2912596
49. Lee, K.S., Geem, Z.W.: A new meta-heuristic algorithm for continuous engineering optimiza-
tion: harmony search theory and practice. Comput. Methods Appl. Mech. Eng. 194, 3902–3933
(2005). https://doi.org/10.1016/J.CMA.2004.09.007
50. Shi, Y., Eberhart, R.C.: Parameter selection in particle swarm optimization. Lect Notes Comput
Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics) 1447, 591–600 (1998).
https://doi.org/10.1007/BFB0040810/COVER
51. Kannan, B.K., Kramer, S.N.: An augmented Lagrange multiplier based method for mixed
integer discrete continuous optimization and its applications to mechanical design. J. Mech.
Des. 116, 405–411 (1994). https://doi.org/10.1115/1.2919393
52. Deb, K.: GeneAS: A robust optimal design technique for mechanical component design. Evol.
Algorithms Eng. Appl., 497–514 (1997).https://doi.org/10.1007/978-3-662-03423-1_27
53. Coello Coello, C.A.: Use of a self-adaptive penalty approach for engineering optimization
problems. Comput. Ind. 41, 113–127 (2000). https://doi.org/10.1016/S0166-3615(99)00046-9
54. Carbas, S., Saka, M.P.: Efficiency of improved harmony search algorithm for solving
engineering optimization problems. Iran Univ. Sci. Technol. 3, 99–114 (2013)
55. Bayram, S., Çıtakoğlu, H.: Modeling monthly reference evapotranspiration process in Turkey:
application of machine learning methods. Environ. Monit., Assess, 195 (2023).https://doi.org/
10.1007/s10661-022-10662-z
Comparison of Multilayer Perceptron
and Other Methods for Prediction
of Sustainable Optimum Design
of Reinforced Concrete Columns

Yaren Aydın, Gebrail Bekdaş, Sinan Melih Nigdeli, Ümit Işıkdağ,

and Zong Woo Geem

Abstract Machine learning has become a popular science in recent years, as it

produces concrete and fast solutions to solve many problems. The rapidly increasing
world population and the developments in technology have caused large greenhouse
gas emissions. The harmful effects of carbon dioxide emissions from cement in
concrete on climate change and global warming are quite remarkable. In this study,
the most commonly used machine learning (ML) models in the literature were
used for CO2 minimization of reinforced concrete columns. Harmony search was
employed to find the optimum dataset for machine learning. The performances of
these algorithms were compared and the best algorithm was tried to be found. As a
result of all, it is observed that Multilayer Perceptron (MLP) has higher performance
than other algorithms. The R2 of the MLP is 0.999. According to this result, it was
observed that MLP is the most successful ML model in the design of eco-friendly
reinforced concrete columns.

Keywords Machine learning · Reinforced concrete · CO2 emission · Columns ·

Harmony search

Y. Aydın · G. Bekdaş (B) · S. M. Nigdeli

Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]
Y. Aydın
e-mail: [email protected]
S. M. Nigdeli
e-mail: [email protected]
Ü. Işıkdağ
Department of Informatics, Mimar Sinan Fine Arts University, 34427 Şişli, Istanbul, Turkey
e-mail: [email protected]
Z. W. Geem
College of IT Convergence, Gachon University, Seongnam 1342, South Korea
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 235
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_12
236 Y. Aydın et al.

1 Introduction

As a result of the unplanned and uncontrolled use of resources in the world, resources
are consumed rapidly and irregularly. This creates the problem of not being able to
leave a livable environment to future generations. Developing technology for people
to have more comfortable living conditions brings more energy consumption [1]. As
a result of the rapidly increasing production in the world after the industrial revo-
lution, rapid growth and development have occurred in countries that have adapted
to the industrialization process. Especially in the twentieth century, the increase in
both quantity and quality of needs has led to the emergence of new inventions and
technological developments. As a result of this process, the natural environment and
resources have started to be polluted; nature has started to have difficulty coping with
this pollution [2]. In addition to the increase in the world population, the increase in
the standard of living in developing countries has led to a significant increase in total
world energy production in the recent period.
Climate change is the statistical or systematic change in the average behavior of
precipitation, temperature, wind, and pressure over a certain period. Human impacts
affect global climate elements and this is interpreted as climate change [3]. Between
2015 and 2022, the temperature reached the highest levels due to the continuous
increase in the number of greenhouse gases and heat accumulation [4]. Extreme heat
waves and droughts adversely affected many people [5].
Since the industrial revolution until today, greenhouse gases released into the
atmosphere have been increasing with various human activities such as the burning
of fossil fuels and industrial development. Due to this increase, the temperature
increase in the earth and the lower layers of the atmosphere is called global warming
[6].
In the 5th assessment report of the IPCC (Intergovernmental Panel on Climate
Change) [7], it was reported that carbon dioxide (CO2 ), methane (CH4 ), and
N2 O concentrations in the atmosphere have increased unprecedentedly in the last
800,000 years. Carbon dioxide (CO2 ), methane (CH4 ), and nitrous oxide (N2 O) are
greenhouse gases (GHGs). Increases in the level of greenhouse gases lead to higher
amounts of solar radiation being absorbed which causes climate change [8].
In terms of the greenhouse effect, it has been reported that the effect of CO2
on a rapid increase in Earth’s average surface temperature and climatic change is
60% [7]. Methane is the second largest greenhouse gas causing climate change.
Extraction and processing of fossil fuels, collection, and disposal of municipal solid
wastes, wastewater plants produce methane gas [9]. Nitrous oxide (N2 O) is one of the
greenhouse gases that cause global climate change and also damages the ozone layer
[10]. Table 1 represents the greenhouse gases and their contributions. From the table,
it can be seen that carbon dioxide has the highest impact and is highly man-made
than others. So, efforts should be made to minimize carbon dioxide emissions.
The concentration of carbon dioxide in the atmosphere is monitored by the number
of particles per million units (ppm). The value of 300 ppm was never exceeded in
the world until 1950. In 2013, CO2 levels surpassed 400 ppm for the first time in
Comparison of Multilayer Perceptron and Other Methods for Prediction … 237

Table 1 Greenhouse gases

Greenhouse Greenhouse Natural (%) Man-made (%)
and their contributions [11]
gases effect (%)
Water vapor 95.000 94.999 0.001
Carbon dioxide 3.618 3.502 0.117
(CO2 )
Methane (CH4 ) 0.360 0.294 0.066
Nitrous oxide 0.950 0.903 0.047
(N2 O)
Misc. gases 0.072 0.025 0.047
Total 1 99.72 0.28

recorded history. The rapid increase in the record amount of carbon dioxide in the
atmosphere after the 1950s has made the impact of the industrial revolutions even
more evident [12].
Carbon dioxide is a tasteless, colorless, odorless, non-combustible, and weakly
acidic gas with the chemical formula CO2 . It is heavier than air. The amount of
carbon dioxide in the atmosphere is 0.03%. Carbon dioxide production is made from
natural underground outlets and chemical methods in factories. Carbon dioxide can
be liquefied and solidified at high pressure and temperatures [13, 14].
The increase in world energy consumption leads to an increase in CO2 emis-
sions and global warming [15]. The International Energy Agency [16] reports that
the current trend in energy supply and use is unsustainable from economic, environ-
mental, and social perspectives and that energy-related carbon dioxide CO2 emissions
will more than double by 2050 unless decisive and lasting measures are taken, and
that increasing oil demand will raise security concerns in oil-supplying countries
[17]. The increase in CO2 emission increases environmental pollution and causes
deterioration of climatic and environmental natural balance with global warming.
Because the cost of environmental degradation in the long term causes an unsus-
tainable situation in economic and social terms, causing the destruction and even
depletion of natural resources. The damage to natural resources limits the livable
environmental conditions of both present and future generations. Therefore, CO2
emissions of countries are a major cause of concern in terms of global warming
[18]. Intense carbon emission prevents the sun’s rays from reflecting back to the
earth and causes an increase in temperature. Considering this situation, climate and
environmental problems are increasing day by day [19].
One of the activities carried out to raise awareness about the increase in carbon
dioxide concentration in the atmosphere in the last century is the calculation and
reduction of carbon footprint. Carbon footprint is the gas emission represented
directly or indirectly in the production and consumption activities of individuals
or organizations. it is measured by calculating carbon dioxide emissions [20].
In addition to the reasons mentioned above, the construction sector is also a sector
that causes environmental problems. Mankind has been using concrete as a material
in buildings for a very long time. Concrete is the most consumed material after water
238 Y. Aydın et al.

in the world because it is easy to shape, though to environmental influences, low-cost,

and practical to use and produce [21, 22]. Concrete is obtained by mixing cement
homogeneously with sand, gravel, and water [23]. As in every field, changes have
occurred in concrete over the years. To overcome the disadvantages of concrete such
as low tensile strength and brittleness, concrete is often used in combination with
steel reinforcements. By eliminating the disadvantages of concrete, which is a cheap
material, and steel, which is an expensive material, a composite material, reinforced
concrete (RC), has emerged with adherence and a usable material has been obtained.
In reinforced concrete, steel reinforcement resists tensile effects and ensures the
ductile behavior of the material. Concrete with high compressive strength prevents
buckling around the reinforcement and protects the material [23].
Cement is the raw material for concrete, the most widely used man-made material
on earth. Since the cement is used in concrete, the use of cement has also become
widespread and has become an important issue. The use of cement increases the
emission of carbon dioxide (CO2 ) into the atmosphere. Cement in concrete is an
important source of carbon dioxide (CO2 ) emissions [24]. For this reason, studies to
be carried out to reduce carbon emissions in the cement sector are important for a
livable world. Since the amount of carbon dioxide emission is higher in the cement
sector, the environmental impact is greater [25].
Table 2 presents that China takes first place in cement production. India follows
China. Cement is the source of carbon dioxide emission approximately 8% of the
carbon dioxide emission [24].
In reinforced concrete (RC) structures, reinforced concrete columns are one of
the most important structural systems under earthquake loads. Columns in reinforced
concrete structures are vertical bearing elements. It enables the structure to survive

Table 2 Cement production by selected countries (in metric tonnes) [26, 27]
Countries 2018 2019 2020 2021 2022
China 2,200,000 2,200,000 2,400,000 2,500,000 2,100,000
Brazil 53,000 55,000 61,000 65,000 65,000
China 2,200,000 2,200,000 2,400,000 2,500,000 2,100,000
USA 87,000 89,000 89,000 92,000 95,000
Egypt 81,200 76,000 42,000 40,000 51,000
India 300,000 320,000 295,000 330,000 370,000
Indonesia 75,200 74,000 65,000 66,000 64,000
Iran 58,000 60,000 68,000 62,000 62,000
Japan 55,300 54,000 51,000 52,000 50,000
Korea, Republic of 57,500 55,000 48,000 48,000 50,000
Russia 53,700 57,000 56,000 56,000 62,000
Turkey 72,500 51,000 72,000 76,000 85,000
Vietnam 90,200 95,000 98,000 100,000 120,000
Other countries (rounded) 870,000 900,000 760,000 810,000 850,000
Comparison of Multilayer Perceptron and Other Methods for Prediction … 239

both under vertical loads such as building weight and under horizontal effects such
as earthquakes. It is one of the most critical elements in terms of building safety.
Reinforced concrete columns have high axial strength, stiffness, and a high damping
ratio. Columns are the elements that ensure the safe transfer of vertical loads to the
ground and the carrying of horizontal loads, especially in buildings without reinforced
concrete shear walls or cross members. Any problem that may occur in the design
and/or manufacturing of the columns may jeopardize the safety of the structure by
itself [28].
There are square, rectangular, and circular types of columns. The seismic resis-
tance of the circular column may be better than the rectangular column. A rectangular
or square section is easier to construct than a circle [29].
From Fig. 1, the geometry of an RC rectangular column where the section width
and height, and the lateral and longitudinal reinforcements can be seen. b is the
section width, h is the section height and L is the total length of the column.
The increasing threat of global warming and climate change in the world has led
researchers to CO2 minimization in civil engineering applications.
Paya-Zaforteza et al. [30] used a metaheuristic algorithm for minimizing the CO2
of a reinforced concrete building frame. The methodology they used proved that
an environmentally friendly way can be followed in RC structural design. Yepes
et al. [31] applied optimization designs for minimizing CO2 emissions of reinforced
concrete retaining walls. The analyzes have shown that the reduction in cost and
CO2 emissions are in parallel. Camp and Huq [32] used hybrid optimization for
designing reinforced concrete frames with minimum cost or CO2 emission. With the
algorithm used, they produced both economical and environmentally friendly designs

Fig. 1 Geometry of RC column

240 Y. Aydın et al.

by reducing construction costs and CO2 emissions. Purnell [33] stated that CO2
emission can be reduced by adjusting the normal mix design parameters of concrete
and that reinforced concrete beams contain significantly lower carbon than steel
or wood composite beams. De Medeiros and Kripka [34] adopted an optimization
algorithm to optimize the financial cost and environmental costs (CO2 emissions,
energy consumption, etc.) of reinforced concrete columns. The results show that with
the minimization of financial costs, there is a decrease in environmental costs. Destrée
and Pease [35] compared a type of steel fiber-reinforced concrete with conventional
concrete flooring systems. The results showed that the steel fiber-reinforced concrete
type reduced CO2 emissions by 40.5%. Yepes et al. [36] optimized the cost and CO2
emissions of precast–prestressed concrete road bridges. The analysis showed that
optimal cost can cause reduce CO2 . Yeo and Potra [37] carried out optimization
approaches with the view to allow decision makers to sustainable and economic
design of reinforced concrete structures. The optimized CO2 footprint design resulted
in 5 to 10% fewer emissions than the cost-optimized design. Kaveh and Ardalani
[38] utilized an optimization algorithm to minimize the amount of CO2 emissions of
RC frames. The results show that the CO2 emission of reinforced concrete frames
can be reduced by slightly increasing the cost. Wang et al. [39] explored the concrete
manufacturing process’s potential effect on global warming. According to the results,
while under different functional units, the environmental performance of composite
and cast-in-situ floors varies. Zhu et al. [40] suggested that variations in the span
of RC slabs affect environmental sustainability. The study showed that composite
slabs are commonly preferred in engineering for eco-friendly sustainability. Paik
and Na [41], dealt with CO2 emissions by using a different slab system instead
of a normal reinforced concrete slab. The result showed that the void slab system
has 34% less emission than the ordinary RC slab. Kayabekir et al. [42] applied a
multi-objective harmony search optimization method for analyzing the eco-friendly
and cost-effective design of reinforced concrete cantilever retaining walls. Their
approach performed well to find both economic and ecological results. Arama et al.
[43] introduced parametric modeling of military pile walls based on CO2 and cost
optimization by the optimization algorithm. As a result of the analyses, both cost and
CO2 emission were minimized. Cakiroglu et al. [44] optimized the CO2 emission
of concrete-filled steel tubular columns. The results showed that concretes of the
lower concrete class are more economical and environmentally friendly. Cakiroglu
et al. [45] compared different concrete classes according to minimum CO2 emissions
from a metaheuristic algorithm. In their study, it was observed that high compressive
strength concretes increase CO2 emissions. Bekdaş et al. [46] used a methodology to
minimize the CO2 emission of reinforced concrete beams. As a result of their work,
using recycled members provides a sustainable design. Yücel et al. [47] studied to
reduce the carbon emission and total cost of reinforced concrete beams. As a result
of their study, it was seen that the use of C25 and C30 concrete is more advantageous
both environmentally and economically. Bekdaş et al. [48] applied an optimization
process to generate an eco-friendly and cost-effective structural model for a post-
tensioned axial symmetric reinforced concrete cylindrical wall. It was found that
increasing the number of post-tensioning loads in the optimum design reduces CO2
Comparison of Multilayer Perceptron and Other Methods for Prediction … 241

emissions. Aydın et al. [49] proposed optimization and machine learning pipeline
for the prediction of an eco-friendly design. The results showed that the machine
learning algorithms are effective in CO2 minimization.
In the field of civil engineering, machine learning has been frequently used in
sub-fields such as structures, geotechnics, hydraulics, construction management, and
transportation. Structural engineering includes the framing of structures, their anal-
ysis, and the design of these structures to withstand the stresses, loads, and pressures
of their environment and remain safe throughout their use. In 1989, Adeli and Yeh
[50] used artificial neural networks (ANN) to design steel beams, which is one of the
first applications of machine learning in structural engineering. Other applications of
structural engineering are structure health monitoring, predicting failure mode and
shear capacity of ultra-high performance concrete beams, evaluating flexural strength
of concrete, predicting concrete compressive strength, estimation of optimum tuned
mass damper parameters, and simulation of structural dynamics [51–55]. Due to the
inherent complexity of soils and geotechnical materials, the use of machine learning
to solve geotechnical estimation problems has become widespread. Machine learning
applications in geotechnical engineering are CPT data interpretation, assessment of
pile drivability, prediction of undrained shear strength, and estimation of deflections
soil classification [56–60]. Hydraulics is generally concerned with the control and
management of water resources. Machine learning applications are flow velocity
prediction in alluvial channels, hydraulic fracturing pressure prediction, anomaly
detection in dam behavior, predict the discharge coefficient of curved labyrinth over-
flows, and prediction of flood risks [61–65]. To follow a correct process in construc-
tion, parameters such as time and finance must be accurately estimated. Applications
of ML in the field of construction management delay risk prediction, assessment of
defect risk, predicting preliminary factory construction cost, predicting the occur-
rence of construction disputes, and prediction of accident construction [66–70]. In the
transportation field; traffic prediction, detection of cracks in road pavement, arrival
time forecasting, forecasting transportation demand, and freight vehicle travel time
prediction [71–75] are the applications of ML.
The increasing threat of global warming and climate change in the world has led
researchers to CO2 minimization. This study is aimed to provide CO2 minimiza-
tion in optimum design of RC columns by using machine learning methods. In this
chapter, it is mentioned that machine learning algorithms, a sub-branch of artificial
intelligence, can be utilized for sustainable and environmentally friendly reinforced
concrete column design. Section 2 provides detailed information about the materials
used in this study and which methods will be used for these materials. In this study,
analyses were performed for CO2 minimization using machine learning methods.
242 Y. Aydın et al.

2 Material and Methods

The environmentally friendly structural design in this study consists of two stages.
The first stage is to generate the data set through optimization algorithms. The second
stage is a machine learning application in which bending moment (M) and axial force
(N) are inputs and column section width (b), column section height (h) and section
reinforcement area (As ) are outputs and predicted.

2.1 Generating a Dataset via Optimization Algorithm

The dataset used in this study is generated via an optimization algorithm. The
Harmony Search algorithm [76] is employed by MATLAB [77] for the optimal RC
rectangular column dimensions and the total area of longitudinal reinforcement (As )
for a given bending moment and axial force to minimize CO2 emissions to ensure a
sustainable structural design of reinforced concrete columns.
The Harmony Search (HS) algorithm developed by Geem [76] was inspired by
jazz musicians improvising to find better melodies. HS algorithm based on the prin-
ciple of finding the best harmonic has been applied to many civil engineering opti-
mization problems such as reinforced concrete structures [42, 43, 78–92], structural
control [93–112], steel structures [113–115], and others [116–118]. Figure 2 shows
the flowchart of HS.
The aim is to minimize CO2 emissions. The objective function of this optimization
problem is given in Eq. (1):

min f (C O 2 ) = CC,C O 2 ∗ VS + C S,C O 2 ∗ W S (1)

The descriptions of the constants used in Eq. 1 are given in Table 3.

For each design variable in the optimal design problem, a valid range of values is
defined. From these values, the search space for the design variables of the algorithm
is created. Then the number of solution vectors of the memory matrix (HMS), the
memory matrix consideration ratio (HMCR), the itch adjustment rate (PAR), and the
maximum number of iterations are determined.
Then, the matrix called harmony memory (HM) (Eq. 2) is filled with randomly
generated variable values, and the corresponding objective function values are calcu-
lated. Here m corresponds to the number of design variables and the number of rows
of the HMS. In the harmony memory matrix, the solution vectors are evaluated
according to the objective functions.
⎡ ⎤
b1 h 1 A1s M 1 N 1 f (x 1 )
⎢ b2 h 2 A2s M 2 N 2 f (x 2 ) ⎥
⎢ ⎥
HM = ⎢ .. .. .. .. .. .. ⎥ (2)
⎣ . . . . . . ⎦
bm h m Am s M m
N m
f (x m
)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 243

Fig. 2 Flowchart of HS
algorithm [119]

Table 3 Descriptions of the

Constants Descriptions
constants
CC,CO2 Carbon dioxide emission per unit volume
VC The total volume of concrete
CS,CO2 Steel carbon dioxide emission per unit weight
Ws Total weight of steel

In Eq. 2, M, N, b, h, and As are the bending moment, axial force, column width,
column height, and the total area of the longitudinal reinforcement in the i-th solution
candidate, respectively.
Then, a new harmonic memory (HM) matrix is developed. In the harmonic search
method, the generation of a new solution vector is controlled by the two main param-
eters of this method (HMCR and PAR) as shown in Eq. 3. HMCR is a probability
value that directs the algorithm to select a value for a design variable either from the
harmonic memory or from the set of all values. The aim here is to perform a more
detailed search by providing transitions around a current solution. This phenomenon
in the harmony search method is based on two is known as the adjustment between
values (PAR). k is the randomly chosen existing solution (Eq. 3). k value in Eq. 3 is
found in Eq. 4.
244 Y. Aydın et al.

Table 4 Details of the dataset

Variables Data Type Min Max Mean Standard deviation
Inputs
Bending moment float64a 100.003 399.806 237.550 81.593
Axial force float64 1000.298 399.929 2387.248 831.938
Outputs
Cross section width float64 250 266.841 250.551 2.372
Cross section height float64 310.392 1000 646.801 202.609
Total reinforcement area float64 2127.295 7016.328 3615.801 1010.983
a float64: 64-bit double precision values [120]

{ ( )
X i,min + rand() ) i f H MC R > r1
( X i,max − X i,min
X i,new = (3)
X i,k + P A R X i,max − X i,min i f H MC R ≤ r1

k = ceil(rand ∗ H M S) (4)

In the last step, the new solution candidate is compared with the worst solution
candidate in HM. The process of creating a new solution candidate is repeated until
a predetermined stopping condition is met.

2.2 Data Description

The data for his study is generated via the HS algorithm as mentioned in 2.1. By
using the HS algorithm, b, h, and As values minimizing the total CO2 emission were
obtained for each combination of M and N. Thus, a dataset of 4429 configurations
has been generated via HS. Descriptive statistics of features and outputs of the dataset
are shown in Table 4.
Figure 3 is the histogram of the dataset. and M, N, h, and As variables (normal-like)
are distributed. Moreover, b has a value of about 250 mm.
Using the Seaborn [121] library of Python [122] programming language, the
correlation matrix in Fig. 4 was generated. A value close to 0 indicates that there is
no connection between these two variables. It can be seen that the lowest correlation
between them were b and As .

2.3 Machine Learning

Artificial intelligence is the technology of developing machines that are created

entirely by artificial means without making use of any living organism and that can
show human-like behaviors and movements [123]. Machine learning (ML), which
Comparison of Multilayer Perceptron and Other Methods for Prediction … 245

Fig. 3 Histogram plot of the dataset

Fig. 4 Correlation matrix

is a sub-unit of artificial intelligence, measures learning algorithms and the perfor-

mance of these algorithms. It constitutes the scientific field of study that enables
computer algorithms to learn like them based on the way people think [124]. Machine
learning focuses on learning relationships from data using efficient computational
algorithms. The convergence of mathematics and computer science stems from the
unique computational challenges of building statistical models from massive datasets
246 Y. Aydın et al.

that can contain billions or trillions of data points [125]. ML can model complex
systems quickly [126].
There are various methods used in the literature to apply machine learning. Suit-
able methods should be selected according to the variety and amount of data. Machine
learning algorithms have 4 basic categories for their intended use [127]. In super-
vised learning, the dependent variable is predicted using one or more independent
variables. Unsupervised machine learning algorithms are used when the data used for
training is not labeled or classified. Semi-supervised learning algorithms use small
amounts of labeled and large amounts of unlabeled data for training. Reinforcement
learning aims to train a function that generates output depending on the feedback
received from the environment using data [128]. As machine learning algorithms gain
experience, their accuracy and efficiency improve. This enables algorithm developers
to get better results and make low-error predictions for the future. Types of ML are
shown in Fig. 5.
In this study, supervised learning was used since the dependent variables were
continuous. Regression was also used since the prediction was aimed. Regression
is widely used in prediction [130]. The dataset used in this study has three outputs.
Python [122] programming language, Anaconda3 [131] open-source distribution,
and Spyder 5.2.2 are used. Scikit-learn library [132] is used for the ML process.
The machine learning models used for an eco-friendly reinforced concrete design
are described below.

Fig. 5 ML algorithms [129]

Comparison of Multilayer Perceptron and Other Methods for Prediction … 247

2.3.1 Featured Machine Learning Models

Foundational: Simple linear regression is a model structure that examines the relation-
ship between independent and dependent variables [133]. In decision tree regression,
the numerical outcomes of the dependent variables are predicted by using a tree-based
structure [134, 135]. The elastic net method is used to perform model selection and
parameter estimation simultaneously in linear or logistic regression models [136].
KNN regression method is a regression method based on the average of k nearest
neighbors [137]. The support Vector Regression method is aiming to find the function
with the lowest generalization error [138].
Multilayer Perceptron (MLP): As a result of the failure of single-layer sensors
in solving nonlinear problems, the concept of multilayer sensors has emerged. In
multilayer perceptrons, the input layer consists of an input layer, a hidden layer,
and an output layer. All layers are interconnected. There are transitions between
the layers called backpropagation and forward propagation. In the backpropagation
phase, the output and error values of the artificial neural network are calculated. To
minimize the error value found in the backpropagation phase, the weight values of the
connection between the layers are updated [139]. Figure 6 shows MLP architecture.
Ensemble: Random forest (RF) is a supervised learning algorithm that can be
used in classification and regression problems. RF is an ensemble learning method
that combines the predictions of separately trained trees and makes new predictions
by averaging these predictions [141].
Gradient boosting uses a sequential approach instead of building parallel trees to
obtain predictions. Thus, in the gradient boosting method, all decision trees improve

Fig. 6 Multilayer Perceptron (MLP) [140]

248 Y. Aydın et al.

the error by predicting the error of the previous decision tree [142]. Histogram-based
gradient boosting is a technique for training faster than gradient boosting for large
datasets [143].
In the voting method, the class predictions made by a large number of different
classifiers are subjected to voting, and the class with the most votes is presented
as the class prediction of the community. The simplest voting scheme is plurality
voting. According to this voting scheme, each base-level classifier casts one vote for
its prediction. The sample is categorized into the class with the most votes [144].
Stacking consists of two levels of modeling. The first of these (often called sub-
models) are models based on a single algorithm and the second is a model responsible
for combining the results provided by the sub-models to obtain a final prediction
[145]. Stacking is based on accepting the predictions of different types of classifiers
as inputs for the meta-classifier and producing a higher-performing prediction from
these predictions [146]. Voting and Stacking can be scheme seen in Fig. 7.
Extreme Gradient Boosting (XGBoost) is a machine learning technique based on
gradient boosting and decision tree algorithms. XGBoost algorithm can be used very
effectively in a large amount of data analyses since it has computational simplicity
compared to other ML methods [149]. CategoryBoosting (CatBoost) is a gradient-
boosting application that uses binary decision trees as basic determinants [150].
Adaptive Boosting (AdaBoost), one of the most widely used boosting algorithms,
is preferred more than other models thanks to its features such as having a higher
prediction rate than other algorithms, using memory at a lower level than other
algorithms, and being applicable [151].

(a) (b)

Fig. 7 The flowchart of a Voting and b Stacking method [147, 148]

Comparison of Multilayer Perceptron and Other Methods for Prediction … 249

Fig. 8 Illustration of the

multioutput regressor [157]

2.3.2 Multitarget Regression (MOR)

In this study, multioutput regression, also known as multi-target, multivariate, or

multi-response regression in the literature, is used to estimate multiple real-valued
output/target variables simultaneously [152–154]. Many studies have proved that
multi-output regression methods generally provide better forecasting performance
compared to single-output methods and preserve the relationship between outputs
since the compound dependencies between outputs in single regression are ignored
in multi-output methods [155, 156]. Figure 8 is the illustration of the multioutput
regressor.

2.4 K Fold Cross Validation

When separating the data set for the training and testing phase of learning algorithms,
it can be separated in various ways. The important thing here is not to use the same
data set for training and testing because after a while the algorithm may memorize the
data set and accurate performance measurement may not be possible. In the k-fold
cross-validation technique, the entire data set is divided into k subsets consisting
of equal or nearly equal amounts of data. During the creation of the subsets, the
data are randomized. Each of the subsets is run in turn as a test set, with the k-1
number of subsets being used as the training data set during each run. The model to
be tested is run k times and the entire dataset is tested once. The advantage of the
k-fold cross-validation technique is that all data is used once for training and testing
purposes [158]. Since the most preferred k value in the literature is 10, 10 is taken
in the study.
250 Y. Aydın et al.

Table 5 Performance
Constants Descriptions
metrics for regression ∑n
Coefficient of determination (yi −xi )2
R2 = 1 − ∑i=1
n
i=1 (yi −x i )
2

/
Root Mean Squared Error ∑n (yi −xi )2
RMSE = i=1 n
∑n
i=1 |yi −x i |
Mean Absolute Error M AE = n
∑n (yi −xi )2
Mean Squared Error MSE = i=1 n

2.5 Performance Criterion

Performance metrics are used to measure and compare the predictions made.
Different methods are used to measure the suitability of the prediction methods
to the data and the efficiency of the algorithms. Performance metrics that are widely
used in the literature will be used in this thesis. In this study, 4 basic performance
metrics used; determination coefficient (R2 ), root mean squared error (RMSE), mean
absolute error (MAE) and mean squared error (MSE) performance criteria were used
to evaluate the performance of the models in our study.
The coefficient of determination (R2 ) expresses the part of the changes in the
dependent variable that can be explained by the independent variable. R2 can take
values between 0 and 1 [159]. The model is successful when the measured value of R2
is close to 1. Root Mean Squared Error (RMSE) takes into account the effect of error
squares and is widely used to compare different estimation methods [160]. Mean
Absolute Error (MAE) is the mean of the absolute value of each difference between
the actual value and the predicted value [161]. Mean Squared Error (MSE) is the
average squared difference between observed and predicted values [162]. The calcu-
lation of these metrics is shown in Table 5. n represents the number of predictions;
yi , xi and x i represents the actual value for the ith observation, the predicted value
for the ith observation, and the average of predicted values, respectively. Figure 9
gives a summary of the ML process in this study.

3 Results and Discussion

As a result, the highest result was achieved with Multilayer Perceptron (MLP). The
column data set created with the Harmony Search algorithm was converted into
“.csv” format and prepared for use in the development environment. The data set
was divided into 70% training and 30% test data, and the k-fold cross-validation
method was used to minimize deviations and errors arising from the distribution of
the data set as training and test data. For Multilayer Perceptron (MLP) model was
scaled before being fed with the data and reinstated after the prediction.
Comparison of Multilayer Perceptron and Other Methods for Prediction … 251

Fig. 9 A summary of the applied methodology

After training the algorithms, their performances were measured on the test data
set. The determination coefficient (R2 ), root mean squared error (RMSE), mean
absolute error (MAE), and mean squared error (MSE) were taken into account.
The obtained values are shown in Table 6 and discussed in this section. The most
successful model parameters are shown among the models where the R2 value
approaches 1 and the MSE value approaches 0. Different numbers of hidden layers
and the number of processing elements in the hidden layer were selected for MLP.
When examining the performance metrics of the Foundational methods, it is seen
that Decision Tree has a high R2 and less RMSE, MAE, and MSE values but it is not
good as MLP. ElasticNet has the least R2 , so it is the most unsuccessful foundational
method in predicting. Among the Ensemble methods, Random Forest and CatBoost
have high R2 (99.8%) values and fewer error values. AdaBoost has an R2 value
of 0.964 but it is not performing well as other Ensemble methods. The highest R2
(99.9%)) and least RMSE, MAE, and MSE values belong to MLP.
In various combinations of the Voting method, the increase in the R2 value with the
addition of MLP compared to the other foundational methods shows again that MLP
has the best performance among the foundational methods. In the Voting method,
the highest performances are observed in combinations where boosting algorithms
252 Y. Aydın et al.

and MLP combinations are used together. The highest success in the combinations
in Voting was achieved with MLP via a different number of combinations.
In the Stacking method, the use of random forest and histogram gradient boosting
as the final estimator gave very close results. In this method, combinations using them
together have higher performance. However, the highest success in the combinations
in Stacking was achieved with MLP.

Table 6 Results of the ML algorithms

Model R2 NRMSE NAME NMSE
Foundational
DecisionTreeRegressor 0.996 29.915 14.465 2505.221
LinearRegression 0.550 245.971 202.094 170,863.191
ElasticNet 0.546 243.435 204.567 166,992.324
KNeighborsRegressor 0.617 154.277 104.274 54,986.445
SVR 0.016 406.631 286.463 371,551.833
MLP*
mlp1 hidden layer sizes (10,10)* 0.991 0.086 0.058 0.008
mlp2 hidden layer sizes (10,20) 0.993 0.078 0.051 0.006
mlp3 hidden layer sizes (10,10,15) 0.993 0.078 0.053 0.006
mlp4 hidden layer sizes (30,40,50) 0.995 0.064 0.043 0.004
mlp5 hidden layer sizes (10,20,30,40,50,60) 0.994 0.059 0.043 0.004
mlp6 hidden layer sizes (30,40,50) 0.995 0.066 0.043 0.004
mlp7 hidden layer sizes (10,20,30,40) 0.993 0.074 0.046 0.005
mlp8 hidden layer sizes (10,20,30,40,50) 0.994 0.069 0.042 0.004
mlp9 hidden layer sizes (100,100) 0.995 0.064 0.041 0.004
mlp10 hidden layer sizes (10,10)** 0.998 0.026 0.012 0.001
mlp11 hidden layer sizes (10,20) 0.999 0.026 0.012 0.0008
mlp12 hidden layer sizes (10,10,15) 0.998 0.027 0.012 0.0009
mlp13 hidden layer sizes (30,40,50 0.998 0.025 0.011 0.0007
mlp14 hidden layer sizes (10,20,30,40,50,60) 0.999 0.024 0.011 0.0008
mlp15 hidden layer sizes (30,40,50) 0.999 0.026 0.012 0.0008
mlp16 hidden layer sizes (10,20,30,40) 0.999 0.026 0.011 0.0009
mlp17 hidden layer sizes (10,20,30,40,50) 0.999 0.025 0.011 0.001
mlp18 hidden layer sizes (100,100) 0.999 0.027 0.011 0.0008
Ensemble
RandomForestRegressor 0.998 17.777 10.162 840.449
HistGradientBoostingRegressor 0.997 20.212 12.456 1090.250
GradientBoostingRegressor 0.995 33.965 22.829 3406.932
(continued)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 253

Table 6 (continued)
AdaBoostRegressor 0.964 106.479 88.300 30,811.579
CatBoostRegressor 0.998 15.207 10.012 622.920
XGBoostRegressor 0.997 14.492 14.034 1572.969
Voting
Combinations
ab,cb,xgb 0.994 39.426 31.835 4288.886
ab,dtr,cb 0.994 39.932 32.055 4362.423
hgbr,rfr,gbr 0.998 19.921 12.868 1211.681
cb,xgb,knr 0.956 53.600 37.371 7064.303
mlp1,rfr,gbr 0.997 0.046 0.028 0.002
mlp2,rfr,gbr 0.997 0.045 0.026 0.002
mlp1,hgbr,gbr,mlp2 0.996 0.053 0.032 0.003
mlp1,hgbr,gbr,mlp18 0.997 0.0391 0.023 0.001
gbr,hgbr,rfr,mlp3 0.998 0.040 0.023 0.001
gbr, dtr,rfr,mlp4 0.998 0.038 0.021 0.001
lm,dtr,knr,svr 0.738 177.226 137.481 74,641.302
lm,dtr,ent,knr 0.790 148.851 120.047 59,889.432
lm,svr,rfr,hgbr 0.852 150.574 117.956 56,440.011
dtr,lm,hgbr,ent 0.886 123.958 102.064 43,174.734
dtr,rfr,hgbr,gbr 0.998 19.608 12.668 1155.584
ent, rfr,hgbr,gbr 0.968 62.278 53.828 12,587.107
knr, rfr,hgbr,gbr 0.975 45.830 31.108 5089.051
mlp1,dtr,rfr,mlp4 0.997 0.046 0.028 0.002
svr,lm,hgbr,rfr 0.852 151.568 118.307 56,129.966
lm,knr,svr,rfr 0.788 178.366 137.091 75,001.233
gbr,hgbr,rfr,ent 0.969 66.271 53.565 12,658.159
ab,cb,xgb,gbr 0.995 35.337 27.246 3353.604
gbr,cb,xgb,knr 0.975 44.725 30.351 5025.545
lm,cb,xgb,knr 0.910 91.406 71.826 21,844.489
gbr,hgbr,rfr,mlp3,mlp16 0.998 0.040 0.023 0.001
gbr, dtr,rfr,mlp4 0.998 0.038 0.021 0.001
gbr,dtr,rfr,mlp4,mlp15 0.998 0.032 0.016 0.001
ent,dtr,lm„knr, svr 0.722 186.811 148.981 86,171.159
ab,cb,xgb,gbr,knr 0.979 51.8133714 38.458 6770.1621
ab,cb,xgb,gbr,dtr 0.996 29.9054851 23.133 2629.4288
gbr,cb,xgb,knr,lm 0.941 75.4987471 59.041 15,238.650
mlp4,dtr,rfr,mlp18,hgbr 0.998 0.032 0.017 0.002
(continued)
254 Y. Aydın et al.

Table 6 (continued)
mlp4,dtr,rfr,mlp14,hgbr 0.998 0.031 0.017 0.001
Stacking
Final Estimator = Gradient Boosting Regressor
lm, dtr,ent,knr,svr 0.996 30.292 18.912 2547.112
hgbr,gbr,rfr 0.998 17.952 11.102 881.042
gbr,hgbr,rfr,ent,lm,dtr,knr,svr 0.998 18.445 11.616 895.697
ab,xgb,ent 0.997 23.872 15.855 1550.0595
ab,cb,xgb 0.998 15.282 10.602 691.734
cb,xgb,dtr 0.998 15.633 10.740 689.732
mlp1,hgbr,gbr,mlp2 0.996 0.053 0.032 0.002
gbr,rfr,mlp4,mlp17
mlp1, hgbr, gbr,rfr,mlp4,mlp15 0.998 0.034 0.015 0.001
Final Estimator = Hist Gradient Boosting Regressor
lm, dtr,ent,knr,svr 0.996 30.388 17.929 2840.752
hgbr,gbr,rfr 0.997 19.249 11.326 1028.443
gbr,hgbr,rfr,ent,lm,dtr,knr,svr 0.997 18.023 11.443 1021.252
ab,xgb,ent 0.997 24.457 15.418 1551.257
ab,cb,xgb 0.998 16.760 10.821 697.945
cb,xgb,dtr 0.998 16.540 10.533 746.261
mlp1, hgbr, gbr,dtr,rfr,mlp4,mlp17 0.998 0.030 0.014 0.001
mlp1, hgbr, gbr,rfr,mlp4,mlp17 0.998 0.036 0.015 0.001
Final Estimator = Random Forest Regressor
lm, dtr,ent,knr,svr 0.996 30.151 18.697 2897.098
hgbr,gbr,rfr 0.998 18.106 11.824 939.706
gbr,hgbr,rfr,ent,lm,dtr,knr,svr 0.998 17.582 10.921 912.215
ab,xgb,ent 0.997 24.259 15.829 1663.914
ab,cb,xgb 0.998 16.421 10.817 706.292
cb,xgb,dtrr 0.998 16.380 10.326 718.070
mlp1,hgbr, gbr,rfr,mlp4,mlp15 0.998 0.026 0.011 0.001
Additional Hyperparameters:
*MLPRegressor, activation = ‘tanh’,solver = ‘adam’, random_state = 0
**activation = relu, solver = ‘lbfgs’, random_state = 0

Finally, the success of machine learning algorithms in estimating the parameters

of the eco-friendly reinforced concrete column was compared and it was seen that
good performances can be obtained. Promising results have been obtained for the
use of the algorithms used in the study for sustainable design.
Comparison of Multilayer Perceptron and Other Methods for Prediction … 255

4 Conclusion

In recent years, where carbon emission has been observed to be in a continuously

increasing trend, the damages to national economies, climates, and human health
should now be evaluated on a global scale. In this study, it is pointed to reduce the
increasing CO2 emission due to various reasons such as technology and industrializa-
tion. This study was carried out since the CO2 emissions released to the environment
in the production of reinforced concrete columns are very high due to the cement in
the content of concrete.
In this study, it is shown that machine learning techniques can be used to achieve
CO2 minimization. A dataset of labeled data and different regression algorithms
(Linear Regression, Decision Tree, K Neighbors, Elastic Net, Support Vector, Multi-
layer Perceptron, Gradient Boosting, Random Forest, Histogram Gradient Boosting,
Adaptive Boosting, Category Boosting, Extreme Gradient Boosting, Stacking,
Voting) were used in the study and the performance of the algorithms were compared
using determination coefficient (R2 ), mean squared error (MSE), root mean squared
error (RMSE) and mean absolute error (MAE). The dataset was generated with the
HS algorithm and consists of 2 inputs, 3 outputs, and 4,429 rows. In the regres-
sion algorithms, bending moment (M) and axial force (N) are inputs; the outputs
are column section width (b), column section height (h), and the reinforcement area
of the section (As ). Multioutput Regressor was applied. The k-fold cross-validation
method was used to prevent the algorithms from memorizing the training dataset and
obtaining incorrect results. The results show that Multilayer Perceptron variations
have the best accuracy performance among the applied algorithms with the highest
R2 and the lowest RMSE, RMAE, and RMSE.
Finally, the performance of machine learning algorithms in predicting CO2 emis-
sions was compared and it was seen that good performances can be obtained.
Promising results have been obtained for the use of the algorithms used in the study
for sustainable design.
256 Y. Aydın et al.

References

1. Öcal, C., İnce, H. H.: Changing needs and sustainable structural design. In: International
Construction Congress (2012)
2. Sarisoy, S., Yildiz, F.: The Relationship between CO2 emissions and economic growth: a panel
data analysis for developed and developing countries. Sosyal Bilimler Metinleri 2013(1), 1–19
(2013)
3. Philander, S.G.: Encyclopedia of global warming and climate change: AE (Vol. 1). Sage
(2008)
4. https://public.wmo.int/en/media/press-release/past-eight-years-confirmed-be-eight-war
mest-record [Visit date: 30 March 2023]
5. https://public.wmo.int/en/our-mandate/climate/wmostatement-state-of-global-climate [Visit
date: 30 March 2023]
6. Smyth, R.C., Hovorka, S.D., Meckel, T., Breton, C., Paine, J.G., Hill, G.R., ... Center, G.C.C.:
Potential sinks for geologic storage of carbon dioxide generated in the Carolinas. US Bureau
of Econ Geol, summary report, 1–14 (2007)
7. Rastogi, M., Singh, S., Pathak, H.: Emission of carbon dioxide from soil. Curr. Sci. 82(5),
510–517 (2002)
8. FCCC, U.: impacts, vulnerabilities and adaptation in developing countries. In: UN FCCC-
United Nations Framework Convention on Climate Change. Bonn (2007)
9. Rao, S., Riahi, K.: The role of non-CO2 greenhouse gases in climate change mitigation:
long-term scenarios for the 21st century. Energy J. (Special Issue# 3) (2006)
10. Saggar, S., Giltrap, D.L., Li, C., Tate, K.R.: Modelling nitrous oxide emissions from grazed
grasslands in New Zealand. Agr. Ecosyst. Environ. 119(1–2), 205–216 (2007)
11. Gaikwad, R.W., Gudadhe, M.D., Bhagat, S.L.: Carbon dioxide capture, tolerance and seques-
tration using microalgae-a review. Int J Pharm Chem Biol Sci [Internet] 6(3), 345–349
(2016)
12. https://climate.nasa.gov/climate_resources/24/graphic-the-relentless-rise-of-carbon-dioxide/
[Visit date: 30 March 2023]
13. https://www.vedantu.com/question-answer/carbon-dioxide-constitutes-about-003-of-the-
class-12-biology-cbse-5fd301ba164c3d4fbe5c98cf [Visit date: 30 March 2023]
14. https://unacademy.com/content/nda/study-material/chemistry/physical-and-chemical-proper
ties-of-carbon-dioxide/ [Visit date: 30 March 2023]
15. Lee, C.C., Chang, C.P.: New evidence on the convergence of per capita carbon dioxide emis-
sions from panel seemingly unrelated regressions augmented Dickey-Fuller tests. Energy
33(9), 1468–1475 (2008)
16. https://www.iea.org/ [Visit date: 30 March 2023]
17. Apergis, N., Payne, J.E.: The emissions, energy consumption, and growth nexus: evidence
from the commonwealth of independent states. Energy Policy 38(1), 650–655 (2010)
18. Heil, M.T., Selden, T.M.: Panel stationarity with structural breaks: carbon emissions and GDP.
Appl. Econ. Lett. 6(4), 223–225 (1999)
19. Sel, A., Tekgün, B.: Prediction of carbon dioxide emissions for turkey using Anfis method.
Visionary E-Journal/Vizyoner Dergisi 13(34) (2022)
20. Wiedmann, T., Minx, J.: A definition of ‘carbon footprint.’ Ecol. Econ. Res. Trends 1(2008),
1–11 (2008)
21. Monteiro, P.J., Miller, S.A., Horvath, A.: Towards sustainable concrete. Nat. Mater. 16(7),
698–699 (2017)
22. Şimşek, O.: Beton ve Beton Teknolojisi (Deneyler İlaveli), 6th ed.; Seçkin Yayıncılık: Ankara,
Türkiye, 2020; pp. 21–22
23. Doğangün, A.: Betonarme yapıların Hesap ve Tasarımı. Birsen Yayınevi, 844 s, İstanbul
(2012)
24. https://www.bbc.com/news/science-environment-46455844 [Visit date: 30 March 2023]
25. McCaffrey, R.: Climate change and the cement industry. Global cement and lime magazine
(environmental special issue) 15, 19 (2002)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 257

26. U.S. Geological Survey: Mineral commodity summaries, 2022. U.S. Department of the
Interior (2020)
27. U.S. Geological Survey: Mineral commodity summaries, U.S. Department of the Interior
(2022)
28. Akpinar, E., Bingöl, E.: Comparison of fiber and analytical methods for construction of biaxial
moment—axial load interaction diagrams for reinforced concrete column design Avrupa Bilim
ve Teknoloji Dergisi (32), 917–927 (2021)
29. https://www.civilera.com/post/why-rectangular-column-shape-preferred-over-circular-
column [Visit date: 30 March 2023]
30. Paya-Zaforteza, I., Yepes, V., Hospitaler, A., Gonzalez-Vidosa, F.: CO2 -optimization of
reinforced concrete frames by simulated annealing. Eng. Struct. 31(7), 1501–1508 (2009)
31. Yepes, V., Gonzalez-Vidosa, F., Alcala, J., Villalba, P.: CO2 -optimization design of reinforced
concrete retaining walls based on a VNS-threshold acceptance strategy. J. Comput. Civ. Eng.
26(3), 378–386 (2012)
32. Camp, C.V., Huq, F.: CO2 and cost optimization of reinforced concrete frames using a big
bang-big crunch algorithm. Eng. Struct. 48, 363–372 (2013)
33. Purnell, P.: The carbon footprint of reinforced concrete. Adv. Cem. Res. 25(6), 362–368
(2013)
34. de Medeiros, G.F., Kripka, M.: Optimization of reinforced concrete columns according to
different environmental impact assessment parameters. Eng. Struct. 59, 185–194 (2014)
35. Destrée, X., Pease, B.: Reducing CO2 emissions of concrete slab constructions with the
primecomposite slab system. Amer. Concrete Institute (ACI) Special Publication 299, 1–12
(2015)
36. Yepes, V., Martí, J.V., García-Segura, T.: Cost and CO2 emission optimization of precast–
prestressed concrete U-beam road bridges by a hybrid glowworm swarm algorithm. Autom.
Constr. 49, 123–134 (2015)
37. Yeo, D., Potra, F.A.: Sustainable design of reinforced concrete structures through CO 2
emission optimization. J. Struct. Eng. 141(3), B4014002 (2015)
38. Kaveh, A., Ardalani, S.: Cost and CO2 emission optimization of reinforced concrete frames
using enhanced colliding bodies optimization algorithm. Applications of metaheuristic
optimization algorithms in civil engineering, 319–350 (2017)
39. Wang, J., Tingley, D.D., Mayfield, M., Wang, Y.: Life cycle impact comparison of different
concrete floor slabs considering uncertainty and sensitivity analysis. J. Clean. Prod. 189,
374–385 (2018)
40. Zhu, C., Chang, Y., Su, S., Li, X., Zhang, Z.: Development of qL-EIV interactive curves
for comparison of the environmental performance of composite slabs and RC slabs from the
perspective of mechanical features. Sci. Total Environ. 683, 508–523 (2019)
41. Paik, I., Na, S.: Comparison of carbon dioxide emissions of the ordinary reinforced concrete
slab and the voided slab system during the construction phase: a case study of a residential
building in South Korea. Sustainability 11(13), 3571 (2019)
42. Kayabekir, A.E., Arama, Z.A., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Eco-friendly design
of reinforced concrete retaining walls: Multi-objective optimization with harmony search
applications. Sustainability 12(15), 6087 (2020)
43. Arama, Z.A., Kayabekir, A.E., Bekdaş, G., Geem, Z.W.: CO2 and cost optimization of rein-
forced concrete cantilever soldier piles: a parametric study with harmony search algorithm.
Sustainability 12(15), 5906 (2020)
44. Cakiroglu, C., Islam, K., Bekdaş, G., Billah, M.: CO2 emission and cost optimization of
concrete-filled steel tubular (CFST) columns using metaheuristic algorithms. Sustainability
13(14), 8092 (2021)
45. Cakiroglu, C., Islam, K., Bekdaş, G., Kim, S., Geem, Z.W.: CO2 emission optimization
of concrete-filled steel tubular rectangular stub columns using metaheuristic algorithms.
Sustainability 13(19), 10981 (2021)
46. Bekdaş, G., Nigdeli, S.M., Kim, S., Geem, Z.W.: Modified harmony search algorithm-based
optimization for eco-friendly reinforced concrete frames. Sustainability 14(6), 3361 (2022)
258 Y. Aydın et al.

47. Yücel, M., Nigdeli, S.M., Bekdaş, G.: Generation of sustainable models with multi-objective
optimum design of reinforced concrete (RC) structures. In Structures (Vol. 40, pp. 223–236).
Elsevier (2022)
48. Bekdaş, G., Yucel, M., Nigdeli, S.M.: Generation of eco-friendly design for post-tensioned
axially symmetric reinforced concrete cylindrical walls by minimizing of CO2 emission.
Struct. Design Tall Spec. Build. 31(13), e1948 (2022)
49. Aydın, Y., Bekdaş, G., Nigdeli, S.M., Isıkdağ, Ü., Kim, S., Geem, Z.W.: Machine learning
models for ecofriendly optimum design of reinforced concrete columns. Appl. Sci. 13(7),
4117 (2023)
50. Adeli, H., Yeh, C.: Perceptron learning in engineering design. Comput. Aided Civil Infrast.
Eng. 4(4), 247–256 (1989)
51. Entezami, A., Shariatmadar, H., De Michele, C.: Non-parametric empirical machine learning
for short-term and long-term structural health monitoring. Struct. Health Monit. 21(6), 2700–
2718 (2022)
52. Solhmirzaei, R., Salehi, H., Kodur, V., Naser, M.Z.: Machine learning framework for
predicting failure mode and shear capacity of ultra high performance concrete beams. Eng.
Struct. 224, 111221 (2020)
53. Sharma, N., Thakur, M.S., Upadhya, A., Sihag, P.: Evaluating flexural strength of concrete
with steel fibre by using machine learning techniques. Compos. Mater. Eng. 3(3), 201–220
(2021)
54. Asteris, P.G., Skentou, A.D., Bardhan, A., Samui, P., Pilakoutas, K.: Predicting concrete
compressive strength using hybrid ensembling of surrogate machine learning models. Cem.
Concr. Res. 145, 106449 (2021)
55. Yu, Y., Yao, H., Liu, Y.: Structural dynamics simulation using a novel physics-guided machine
learning method. Eng. Appl. Artif. Intell. 96, 103947 (2020)
56. Rauter, S., Tschuchnigg, F.: CPT data interpretation employing different machine learning
techniques. Geosciences 11(7), 265 (2021)
57. Zhang, W., Wu, C., Li, Y., Wang, L., Samui, P.: Assessment of pile drivability using
random forest regression and multivariate adaptive regression splines. Georisk: assessment
and management of risk for engineered systems and geohazards 15(1), 27–40 (2021)
58. Zhang, W., Wu, C., Zhong, H., Li, Y., Wang, L.: Prediction of undrained shear strength using
extreme gradient boosting and random forest based on Bayesian optimization. Geosci. Front.
12(1), 469–477 (2021)
59. Zhang, R., Wu, C., Goh, A.T., Böhlke, T., Zhang, W.: Estimation of diaphragm wall deflections
for deep braced excavation in anisotropic clays using ensemble learning. Geosci. Front. 12(1),
365–373 (2021)
60. Aydın, Y., Işıkdağ, Ü., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Use of machine learning
techniques in soil classification. Sustainability 15(3), 2374 (2023)
61. Kitsikoudis, V., Sidiropoulos, E., Iliadis, L., Hrissanthou, V.: A machine learning approach for
the mean flow velocity prediction in alluvial channels. Water Resour. Manage 29, 4379–4395
(2015)
62. Ben, Y., Perrotte, M., Ezzatabadipour, M., Ali, I., Sankaran, S., Harlin, C., Cao, D.: Real-time
hydraulic fracturing pressure prediction with machine learning. In: SPE Hydraulic Fracturing
Technology Conference and Exhibition. OnePetro (2020)
63. Salazar, F., Conde, A., Irazábal, J., Vicente, D.J.: Anomaly detection in dam behaviour with
machine learning classification models. Water 13(17), 2387 (2021)
64. Hu, Z., Karami, H., Rezaei, A., DadrasAjirlou, Y., Piran, M.J., Band, S.S., ... Mosavi, A.:
Using soft computing and machine learning algorithms to predict the discharge coefficient of
curved labyrinth overflows. Eng. Appl. Comput. Fluid Mech. 15(1), 1002–1015 (2021)
65. Razali, N., Ismail, S., Mustapha, A.: Machine learning approach for flood risks prediction.
IAES Int. J. Artif. Intell. 9(1), 73 (2020)
66. Gondia, A., Siam, A., El-Dakhakhni, W., Nassar, A.H.: Machine learning algorithms for
construction projects delay risk prediction. J. Constr. Eng. Manag. 146(1), 04019085 (2020)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 259

67. Fan, C.L.: Defect risk assessment using a hybrid machine learning method. J. Constr. Eng.
Manag. 146(9), 04020102 (2020)
68. Dang-Trinh, N., Duc-Thang, P., Nguyen-Ngoc Cuong, T., Duc-Hoc, T.: Machine learning
models for estimating preliminary factory construction cost: case study in Southern Vietnam.
Int. J. Constr. Manag., 1–9 (2022)
69. Ayhan, M., Dikmen, I., Talat Birgonul, M.: Predicting the occurrence of construction disputes
using machine learning techniques. J. Constr. Eng. Manag. 147(4), 04021022 (2021)
70. Koc, K., Ekmekcioğlu, Ö., Gurgun, A.P.: Accident prediction in construction using hybrid
wavelet-machine learning. Autom. Constr. 133, 103987 (2022)
71. Meena, G., Sharma, D., Mahrishi, M.: Traffic prediction for intelligent transportation system
using machine learning. In: 2020 3rd International Conference on Emerging Technologies in
Computer Engineering: Machine Learning and Internet of Things (ICETCE), pp. 145–148.
IEEE (2020)
72. Praticò, F.G., Fedele, R., Naumov, V., Sauer, T.: Detection and monitoring of bottom-up cracks
in road pavement using a machine-learning approach. Algorithms 13(4), 81 (2020)
73. Valatsos, P., Vafeiadis, T., Nizamis, A., Ioannidis, D., Tzovaras, D.: Freight transportation
route time prediction with ensemble learning techniques. In: 25th Pan-Hellenic Conference
on Informatics, pp. 52–57 (2021)
74. Plakandaras, V., Papadimitriou, T., Gogas, P.: Forecasting transportation demand for the US
market. Transp. Res. Part A: Policy Practice 126, 195–214 (2019)
75. Li, X., Bai, R.: Freight vehicle travel time prediction using gradient boosting regression
tree. In: 2016 15th IEEE International Conference on Machine Learning and Applications
(ICMLA), pp. 1010–1015. IEEE (2016)
76. Geem, Z.W., Kim, J.H., Loganathan, G.V.: A new heuristic optimization algorithm: harmony
search. simulation 76(2), 60–68 (2001)
77. MathWorks, T.: Matlab R2022a; The MathWorks: Natick. MA, USA (2022)
78. Akin, A., Saka, M.P.: Optimum detailed design of reinforced concrete continuous beams
using the harmony search algorithm. In: The tenth international conference on computational
structures technology, Paper 131, Stirlingshire, UK (2010)
79. Bekdaş, G., Nigdeli, S.M.: Optimization of T-shaped RC flexural members for different
compressive strengths of concrete. Int. J. Mech. 7, 109–119 (2013)
80. Jaberipour, M., Khorram, E.: A new harmony search algorithm for solving mixed–discrete
engineering optimization problems. Eng. Optim. 43(5), 507–523 (2011)
81. Bekdaş, G., Nigdeli, S.M.: The optimization of slender reinforced concrete columns. Proc.
Appl. Math. Mech. 14(1), 183–184 (2014)
82. Nigdeli, S.M., Bekdaş, G., Kim, S., Geem, Z.W.: A novel harmony search based optimization
of reinforced concrete biaxially loaded columns. Struct. Eng. Mech. 54(6), 1097–1109 (2015)
83. Nigdeli, S.M., Bekdaş, G.: Optimization of reinforced concrete shear walls using harmony
search. In: 11th International Congress on Advances in Civil Engineering, pp. 21–25 (2014)
84. Bekdaş, G., Nigdeli, S.M.: Modified harmony search for optimization of reinforced concrete
frames. In: 3rd International Conference on the Harmony Search Algorithm (ICHSA 2017),
Bilbao, Spain (2017)
85. Bekdaş, G.: Optimum design of axially symmetric cylindrical reinforced concrete walls.
Struct. Eng. Mech. 51(3), 361–375 (2014)
86. Bekdaş, G.: Harmony search algorithm approach for optimum design of post-tensioned axially
symmetric cylindrical reinforced concrete walls. J. Optim. Theory Appl. 164(1), 342–358
(2015)
87. Bekdaş, G., Nigdeli, S.M.: Optimum reduction of flexural effect of axially symmetric
cylindrical walls with post-tensioning forces. KSCE J. Civil Eng., 1–8 (2018)
88. Bekdaş, G.: New improved metaheuristic approaches for optimum design of posttensioned
axially symmetric cylindrical reinforced concrete walls. Struct Des. Tall Spec. Build. 27(7)
(2018)
89. Bekdaş, G., Nigdeli, S.M.: Multi-objective optimization of reinforced concrete footings using
harmony search. In: 23rd International Conference on Multiple Criteria Decision Making
(MCDM 2015), Hamburg, Germany. Academic (2015)
260 Y. Aydın et al.

90. Bekdaş, G., Arama, Z.A., Kayabekir, A.E., Geem, Z.W.: Optimal design of cantilever soldier
pile retaining walls embedded in frictional soils with harmony search algorithm. Appl. Sci.
10(9), 3232 (2020)
91. Yücel, M., Kayabekir, A.E., Bekdaş, G., Nigdeli, S.M., Kim, S., Geem, Z.W.: Adaptive-Hybrid
harmony search algorithm for multi-constrained optimum eco-design of reinforced concrete
retaining walls. Sustainability 13, 1639 (2021)
92. Toklu, Y.C., Bekdaş, G., Geem, Z.W.: Harmony search optimization of nozzle movement for
additive manufacturing of concrete structures and concrete elements. Appl. Sci. 10(12), 4413
(2020)
93. Bekdaş, G., Nigdeli, S.M.: Estimating optimum parameters of tuned mass dampers using
harmony search. Eng. Struct. 33(9), 2716–2723 (2011)
94. Bekdaş, G., Kayabekir, A.E., Nigdeli, S.M., Toklu, Y.C.: Tranfer function amplitude mini-
mization for structures with tuned mass dampers considering soil-structure interaction. Soil
Dyn. Earthq. Eng. 116, 552–562 (2019)
95. Zhang, H.Y., Zhang, L.J.: Tuned mass damper system of high-rise intake towers optimized
by improved harmony search algorithm. Eng. Struct. 138, 270–282 (2017)
96. Amini, F., Ghaderi, P.: Hybridization of harmony search and ant colony optimization for
optimal locating of structural dampers. Appl. Soft. Comput. 13(5), 2272–2280 (2013)
97. Kayabekir, A.E., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Optimum design of PID controlled
active tuned mass damper via modified harmony search. Appl. Sci. 10(8), 2976 (2020)
98. Bekdaş, G., Nigdeli, S.M.: Mass ratio factor for optimum tuned mass damper strategies. Int.
J. Mech. Sci. 71, 68–84 (2013)
99. Nigdeli, S.M., Bekdaş, G., Yang, X.S.: Optimum tuning of mass dampers by using a hybrid
method using harmony search and flower pollination algorithm. In: International conference
on harmony search algorithm, February. Springer, Singapore, pp 222–231 (2017)
100. Nigdeli, S.M., Bekdaş, G., Alhan, C.: Optimization of seismic isolation systems via harmony
search. Eng. Optim. 46(11), 1553–1569 (2014)
101. Keshtegar, B., Etedali, S.: Nonlinear mathematical modeling and optimum design of tuned
mass dampers using adaptive dynamic harmony search algorithm. Struct. Control Health
Monit. 25(7), e2163 (2018)
102. Bekdaş, G., Nigdeli, S.M.: Preventing the pounding of adjacent buildings with harmony
search optimized tuned mass damper. In: 3rd European Conference of Civil Engineering, pp
2–4 (2012)
103. Nigdeli, S.M., Bekdaş, G.: Optimum tuned mass damper design for preventing brittle fracture
of RC buildings. Smart Struct. Syst. 12(2), 137–155 (2013)
104. Bekdaş, G., Nigdeli, S.M.: Metaheuristic based optimization of tuned mass dampers under
earthquake excitation by considering soil-structure interaction. Soil Dyn. Earthq. Eng. 92,
443–461 (2017)
105. Nigdeli, S.M., Bekdaş, G.: Optimization of TMDs for different objectives. In: An International
Conference on Engineering and Applied Sciences Optimization, Kos Island, Greece, pp. 4–6
(2014)
106. Nigdeli, S.M., Bekdaş, G.: The effect of impulsive motions on optimum tuned mass damper
parameters. In: 11th International Conference on Vibration Problems, pp. 2–10 (2013)
107. Nigdeli, S.M., Bekdaş, G.: Optimum tuned mass damper approaches for adjacent structures.
Earthq. Struct. 7(6), 1071–1091 (2014)
108. Nigdeli, S.M., Bekdaş, G.: Optimum tuned mass damper design in frequency domain for
structures. KSCE J. Civil Eng. 21(3), 912–922 (2017)
109. Ocak, A., Bekdaş, G., Nigdeli, S.M. Comparison of classical and adaptive parameter setting
for harmony search on a structural optimization problem. In: Proceedings of 7th International
Conference on Harmony Search, Soft Computing and Applications: ICHSA 2022, pp. 363–
368. Singapore: Springer Nature Singapore (2022)
110. Ocak, A., Nigdeli, S.M., Bekdaş, G., Kim, S., Geem, Z.W.: Optimization of seismic base
isolation system using adaptive harmony search algorithm. Sustainability 14(12), 7456 (2022)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 261

111. Ocak, A., Nigdeli, S.M., Bekdaş, G., Kim, S., Geem, Z.W.: Adaptive harmony search for
tuned liquid damper optimization under seismic excitation. Appl. Sci. 12(5), 2645 (2022)
112. Ocak, A., Bekdaş, G., Nigdeli, S.M.: A metaheuristic-based optimum tuning approach for
tuned liquid dampers for structures. Struct. Design Tall Spec. Build. 31(3), e1907 (2022)
113. Degertekin, S.O.: Optimum design of steel frames using harmony search algorithm. Struct.
Multidiscip. Optim. 36, 393–401 (2008)
114. Saka, M.P., Hasançebi, O.: Adaptive harmony search algorithm for design code optimization
of steel structures. Harmony search algorithms for structural design optimization, 79–120
(2009)
115. Cao, H., Qian, X., Zhou, Y.L., Yang, H.: Applicability of subspace harmony search hybrid
with improved deb rule in optimizing trusses. J. Comput. Civ. Eng. 32(4), 04018021 (2018)
116. Kaveh, A., Abadi, A.S.M.: Cost optimization of a composite floor system using an improved
harmony search algorithm. J. Constr. Steel Res. 66(5), 664–669 (2010)
117. Cakiroglu, C., Bekdaş, G., Kim, S., Geem, Z.W.: Optimisation of shear and lateral-torsional
buckling of steel plate girders using meta-heuristic algorithms. Appl. Sci. 10(10), 3639 (2020)
118. Cakiroglu, C., Bekdaş, G., Geem, Z.W.: Harmony search optimisation of dispersed laminated
composite plates. Materials 13(12), 2862 (2020)
119. Yoo, D.G., Kim, J.H., Geem, Z.W.: Overview of harmony search algorithm and its applications
in civil engineering. Evol. Intel. 7, 3–16 (2014)
120. https://www.simplilearn.com/tutorials/python-tutorial/float-in-python [Visit date: 01 April
2023]
121. https://seaborn.pydata.org/ [Visit date: 01 April 2023]
122. http://python.org [Visit date: 01 April 2023]
123. Aydın, İ.H., Değirmenci, C.H.: Yapay zekâ. Girdap Kitap (2018)
124. Alpaydin, E.: Introduction to Machine Learning. Mıt Press (2020)
125. Deo, R.C.: Machine learning in medicine. Circulation 132(20), 1920–1930 (2015)
126. Khambra, G., Shukla, P.: Novel machine learning applications on fly ash based concrete: an
overview. Materials Today: Proc. (2021)
127. Shalev-Shwartz, S., Ben-David, S.: Understanding machine learning: From theory to
algorithms. Cambridge University Press (2014)
128. https://blog.turhost.com/makine-ogrenmesi-machine-learning-nedir/ [Visit date: 02 April
2023]
129. Thai, H.T.: Machine learning for structural engineering: a state-of-the-art review. In: Structures
(Vol. 38, pp. 448–491). Elsevier (2022)
130. Gültepe, Y.: A comparative assessment on air pollution estimation by machine learning
algorithms. Avrupa Bilim ve Teknoloji Dergisi 16, 8–15 (2019)
131. Anaconda3 [Computer Software]. Available online: https://anaconda.org/. Accessed on 03
April 2023
132. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M.,
Prettenhofer, P., Weiss, R., Dubourg, V., et al.: Scikit-Learn: machine learning in Python. J.
Mach. Learn. Res. 12, 2825–2830 (2011)
133. Wu, C.S.M., Patil, P., Gunaseelan, S.: Comparison of different machine learning algorithms
for multiple regression on black friday sales data. In: 2018 IEEE 9th International Conference
on Software Engineering and Service Science (ICSESS), pp. 16–20. IEEE (2018)
134. Patel, B.N., Prajapati, S.G., Lakhtaria, K.I.: Efficient classification of data using decision tree.
Bonfring Int. J. Data Mining 2(1), 06–12 (2012)
135. Rathore, S.S., Kumar, S.: A decision tree regression based approach for the number of software
faults prediction. ACM SIGSOFT Soft. Eng. Notes 41(1), 1–6 (2016)
136. Zou, H., ve Hastie, T.:Regularization and variable selection via the elastic net. J.R. Statist.
Soc. B. 67(2), 301–320 (2005)
137. Altman, N.S.: An introduction to kernel and nearest-neighbor nonparametric regression. Am.
Stat. 46(3), 175–185 (1992)
138. Özlem, A.K.A. ., Tunçeli, M.: Use of the support vector regression in medical data analysis.
Experimental Appl. Med. Sci. 2(4), 242–256
262 Y. Aydın et al.

139. Ayşe, A., Berberler, M.E.: Interface design for the solution of prediction and classification
problems with artificial neural networks. Acta Infologica 1(2), 55–73 (2017)
140. https://www.analyticsvidhya.com/blog/2020/12/mlp-multilayer-perceptron-simple-ove
rview/ [Visit date: 04 April 2023]
141. Denil, M., Matheson, D., De Freitas, N.: Narrowing the gap: Random forests in theory and in
practice. In: International Conference on Machine Learning, pp. 665–673. PMLR (2014)
142. Ayyadevara, V.K.: Pro Machine Learning Algorithms. Apress, Berkeley, CA, USA (2018)
143. https://machinelearningmastery.com/histogram-based-gradient-boosting-ensembles/ [Visit
date: 04 April 2023]
144. Džeroski, S., Ženko, B.: Is combining classifiers with stacking better than selecting the best
one? Mach. Learn. 54, 255–273 (2004)
145. Graczyk, M., Lasota, T., Trawiński, B., Trawiński, K.: Comparison of bagging, boosting and
stacking ensembles applied to real estate appraisal. In: Intelligent Information and Database
Systems: Second International Conference, ACIIDS, Hue City, Vietnam, March 24–26, 2010.
Proceedings, Part II 2, pp. 340–350). Springer, Berlin, Heidelberg (2010)
146. Wolpert, D.H.: Stacked generalization. Neural Netw. 5(2), 241–259 (1992)
147. Patil, D.R., Patil, J.B.: Malicious URLs detection using decision tree classifiers and majority
voting technique. Cybern. Inf. Technol. 18(1), 11–29 (2018)
148. https://medium.com/greyatom/introducing-ensemble-more-is-better-than-one-436a44
8350cf [Visit date: 04 April 2023]
149. Hayri, A.B.A.R.: Prediction of gold prices by xgboost and mars methods. Ekev Akademi
Dergisi 83, 427–446 (2020)
150. Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A.V., Gulin, A.: CatBoost: unbiased
boosting with categorical features. Adv. Neural Inf. Process. Syst., 31 (2018)
151. Moldagulova, A., Sulaiman, R.B.: Using KNN algorithm for classification of textual docu-
ments. ICIT 2017–8th International Conference on Information Technology, Proceedings,
665–671 (2017)
152. Appice, A., Džeroski, S.: Stepwise induction of multi-target model trees. In: Machine
Learning: ECML 2007: 18th European Conference on Machine Learning, Warsaw, Poland,
September 17–21, 2007. Proceedings 18, pp. 502–509. Springer, Berlin, Heidelberg (2007)
153. Breiman, L., Friedman, J.H.: Predicting multivariate responses in multiple linear regression.
J. Royal Statist. Soc. Series B (Stat. Methodol.) 59(1), 3–54 (1997)
154. Similä, T., Tikka, J.: Input selection and shrinkage in multiresponse linear regression. Comput.
Stat. Data Anal. 52(1), 406–422 (2007)
155. Borchani, H., Varando, G., Bielza, C., Larranaga, P.: A survey on multi-output regression.
Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 5(5), 216–233
(2015)
156. Han, Z., Liu, Y., Zhao, J., Wang, W.: Real time prediction for converter gas tank levels based
on multi-output least square support vector regressor. Control. Eng. Pract. 20(12), 1400–1409
(2012)
157. Chou, J.S., Truong, D.N., Che, Y.: Optimized multi-output machine learning system for
engineering informatics in assessing natural hazards. Nat. Hazards 101(3), 727–754 (2020)
158. Fushiki, T.: Estimation of prediction error by using K-fold cross-validation. Statist. Comput.,
21, 137–146 (2011); Barrett, J.P.: The coefficient of determination—some limitations. Amer.
Stat. 28(1), 19–20 (1974)
159. Crouse, J.D.: Comparing the equating accuracy from three data collection designs using
bootstrap estimation methods. The University of Iowa (1991)
Comparison of Multilayer Perceptron and Other Methods for Prediction … 263

160. Chai, T., Draxler, R.R.: Root mean square error (RMSE) or mean absolute error (MAE)?–
Arguments against avoiding RMSE in the literature. Geoscientific Model Dev. 7(3), 1247–
1250 (2014)
161. https://medium.com/@20_80_/mean-absolute-error-mae-machine-learning-ml-
b9b4afc63077 [Visit date: 04 April 2023]
162. https://statisticsbyjim.com/regression/mean-squared-error-mse/ [Visit date: 04 April 2023]
Artificial Intelligence and Deep Learning
in Civil Engineering

Ayla Ocak, Sinan Melih Nigdeli, Gebrail Bekdaş, and Ümit Işıkdağ

Abstract Artificial intelligence is a variety of software developed that imitates the

human brain to perform the tasks that the human brain can do. Aiming to mini-
mize human intervention, this software has a wide range of content that deals with
many problems such as perception, problem-solving, information transfer, planning,
natural language processing, and so on. As a purpose-oriented method that can be
used in many disciplines, it is preferred with high success rates, especially in the solu-
tion of engineering problems. Its sub-branches include machine learning, in which
machines are trained to extract information from the available data. Machine learning
and deep learning methods, which express more specific learning, make it possible
to create a powerful predictive model. In this study, deep learning methods, which
are a sub-branch of artificial intelligence and artificial intelligence, and the studies
in which these methods are used in civil engineering are explained.

Keywords Artificial İntelligence · Machine Learning · Deep Learning

1 Introduction

The advancement of science and technology brought about the industrial revolution
and mechanization with new inventions in the eighteenth and nineteenth centuries
and caused a decrease in human intervention in production and technology. This
transition to mechanization has created a universal mechanical language among all

A. Ocak · S. M. Nigdeli · G. Bekdaş (B)

Department of Civil Engineering, Istanbul University - Cerrahpaşa, 34320 Avcılar, Istanbul,
Turkey
e-mail: [email protected]
S. M. Nigdeli
e-mail: [email protected]
Ü. Işıkdağ
Department of Informatics, Mimar Sinan Fine Arts University, 34427 Şişli, Istanbul, Turkey
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 265
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_13
266 A. Ocak et al.

people. In the middle of the eighteenth century, the source of communication called
“shared technical vocabulary” was the beginning of an unpredictable change [1].
The fact that the work that can be produced by human hands can be done with the
help of machines has also made it easier to reach the needed resources by increasing
production. However, the idea that machines are devices that act systematically and
that the human brain cannot do the things it can do even though it imitates human
behavior paved the way for a second technological revolution. This idea of mech-
anization pushed the limits of human production and led to the development of
a machine that imitated the capabilities of the human brain. Progress in science
and technology has made it possible to transfer the abilities of the human brain
to perception, problem-solving, learning, and similar abilities to various software
called artificial intelligence. In other words, “intelligent machines” were produced
as a continuation of the industrial revolution. Afterward, artificial intelligence soft-
ware was divided into sub-branches and various artificial intelligence techniques
were developed.

2 Artificial Intelligence (AI)

Artificial intelligence is a new technology that is characterized as a kind of intelligent

machine that has emerged with the features of human intelligence such as learning,
perception, decision-making, problem-solving, and similar features of human brain
imitation. In one of his studies, John McCarthy defined artificial intelligence as “the
science and engineering of making intelligent computer programs” alongside intelli-
gent machines [2, 3]. In other words, it consists of simulated human brain experiments
to make machines smart [3]. In the mid-twentieth century, English mathematician
and computer scientist Alan Mathison Turing said, “Can machines think? He laid the
foundation of artificial intelligence with the Turing test he developed to answer the
question [4]. This test investigates the communication between a man and a woman in
two different rooms with a typewriter called a teletype, as in today’s electronic chat
rooms. Accordingly, while the interrogator initially communicates with a human,
then a machine replaces the person in front of him. The decision of the interrogator,
whether there is a machine or a human being on the other party, gives the result
of whether the machine is intelligent [5]. This imitation game created by Turing is
described as the first step of artificial intelligence.
Considering the living things in nature, there is an instinctive behavior and
learned behavior common to both humans and animals. While instinctive behavior
is observed in the reflex of sucking the breast milk of a newborn creature, this
symbolizes a learned behavior for a child who experiences burner from a fire and
understands that he should not approach it. In the behavior mechanism based on
artificial intelligence technology, there is a learned behavior and the ability to make
inferences. As it can be understood from this situation, to develop a non-instinctive
learned behavior mechanism, some knowledge must be added to the repertoire of
artificial intelligence. The breadth of this knowledge reinforces the inference power
Artificial Intelligence and Deep Learning in Civil Engineering 267

of artificial intelligence, just as it increases a person’s ability to distinguish between

right and wrong in line with their experiences. For this purpose, the process of trans-
ferring various data to the machine regarding the abilities to be gained from artificial
intelligence has had an important place in the development of artificial intelligence
technology and has been a pioneer in the development of a new branch of science.
This branch of science is called “Data Science” and researchers interested in this
science are called “Data Scientists”. It refers to the studies carried out to reveal
general knowledge from the existing data based on data science [6]. Obtaining the
knowledge used in artificial intelligence with data science has made these technology
fields an inseparable duo. Investigating features such as the comprehensibility, distri-
bution, and frequency of the data has led to the creation of more accurate artificial
intelligence technology. For example, the presence of unusual daily weather in data
consisting of a long-term weather report may reduce the success level of weather
forecasts made by artificial intelligence. Data science comes into play at this point,
allowing artificial intelligence to make a more accurate prediction by detecting the
extreme value in the data and removing it from the flow. Although data science and
artificial intelligence technologies have similar points, data science focuses more on
data analysis, data processing, and statistics, while artificial intelligence focuses on
data modeling, inference, and prediction from data.
Artificial intelligence is actively used in many fields such as computer science,
medicine, engineering, mathematics, biochemistry, physics and astronomy, phar-
macology, social sciences, business administration, management and accounting,
economics, and psychology in today’s world. With artificial intelligence, it has been
possible to explore different types of fields such as facial recognition, mastering a
game, estimation of landslide danger, cancer diagnosis from scans, predicting the
future effects of epidemics, earthquake, and location detection, human behavior
prediction from digital records, social network analysis and political orientation
prediction [7–14]. When the studies in the field of engineering are examined, it is
seen that areas such as the estimation of the compressive strength of the rocks, the
evaluation of land pollution and soil quality, the estimation of wind-based energy
production, the estimation of long-term braking events to reduce emissions in the
automotive sector, the stability estimation of debris breakwater structures, the paint
structure quality estimation are investigated [15–20].
In the field of civil engineering, artificial intelligence, and its sub-branches have
been frequently used in subjects such as structural control and monitoring of building
behavior, as well as prediction model production studies for soil and material proper-
ties. In soil studies on construction, modeling of hydraulic conductivity of sandy soils,
classification of different types of soils, liquefaction prediction of fine-grained soils,
and estimation of shear strength of soil have been investigated and successful predic-
tion models have been produced [21–24]. In the field of building materials, mechan-
ical properties of sand concrete, thermo-mechanical properties of recycled aggregate
concrete, compressive strength of concrete, compressive strength of construction
demolition waste geopolymers, the effect of cement-based materials on compres-
sive strength, performance evaluation of slag-based concrete at high temperatures,
concrete-filled steel pipes has been done on the estimation of areas such as the
268 A. Ocak et al.

load bearing capacity of the columns [25–31]. In the field of hydraulics, artificial
intelligence methods have been applied to determine the fragility of concrete-lined
rockfill dams under seismic loads, flow variables of sharply curved channels, the
flow velocity of open channel junction, and discharge coefficient of side weirs [32–
35]. In the fields of structural engineering and building management, building design
and performance evaluation, nominal shear capacity of a reinforced concrete wall,
shear strength of reinforced concrete beams, the seismic failure mode of reinforced
concrete shear walls, the fatigue life of bridge decks, cost of engineering services
in the construction industry, structural health monitoring for damage and detection
investigation topics were examined with artificial intelligence methods [36–42]. In
the field of structural mechanics, studies are carried out in the design and evaluation
of structural control systems. With artificial intelligence, the production of perfor-
mance estimation models related to the structure was provided for the evaluation
of the effectiveness of the tuned mass dampers (TMDs) used in bridges to prevent
wind-induced vibrations, the estimation of the fatigue performance of the metallic
damper under cyclical loading, the estimation of the optimally tuned mass damper
parameters, the evaluation of the reaction of structures exposed to fire and similar
extreme conditions [43–46].
Artificial intelligence is a comprehensive method that includes many methods.
While the application examples of artificial intelligence technologies, which are also
expressed as smart machines, are seen in the programming of robots, the machine
learning method is used to give the machine the ability to make inferences by training
with a certain data set. In addition to these methods, there are also sub-fields such
as expert systems, data mining, machine perception, natural language processing,
planning, and optimization [47]. The sub-branches of artificial intelligence and the
fields associated with artificial intelligence are shown in Fig. 1.
Machine learning is a sub-branch of artificial intelligence developed in the 1980s,
consisting of learning algorithms [48]. Its basic logic is to train and test the machine
with a data set and demonstrate its ability to make inferences from new data with what
it has learned. The aim here is to expose the machine to information, just as a person
working at a convenience store constantly learns about product prices while receiving
payment so that the cashier can predict a product at an approximate rate, as well as
the machine. As an example of this; in a data set consisting of various information

Fig. 1 Artificial intelligence and sub-branches associated with artificial intelligence [47]
Artificial Intelligence and Deep Learning in Civil Engineering 269

such as age, height, weight, and similar information of a group of people and the
results of the analysis, the characteristics and test results given for each individual
represent the inputs of the data set. In this data, the column in which each candidate’s
cancer patient status is expressed or not indicates the unit of the data set, which is
called output. When the machine in your hand is taught the inputs in this data and
the outputs about whether it is sick or not, it becomes possible with the machine
learning method at a high level of success for the machine to decide whether the
next person is sick or not. There are supervised, unsupervised, semi-supervised,
and reinforcement learning options. A deep learning method has been developed by
training machine learning with more specific and intensive data. Deep learning, a
sub-branch of machine learning, is a machine learning method that aims to make
sense of high-precision data.

2.1 Deep Learning

Deep learning is an artificial intelligence method, which consists of neural networks

resembling neurons in the human brain, focused on working like the human brain
and using large amounts of data. It directs people to think like the human brain by
using different sources such as text, sound, pictures, and so on as data. In today’s
world, digital assistants on smartphones and computers are used in many areas such
as voice search on the internet, face recognition systems used in phones and secu-
rity units, creating subtitles for various video applications, and personalized internet
advertisements according to people’s product preferences. In the defense industry
and security sector, it plays an important role in preventing accidents and detecting
dangerous elements with its aspects such as identifying objects that threaten secu-
rity and providing driver support in the limited visibility of vehicles in traffic. In
2006, an unsupervised learning method, which can be used to create feature layers
without labeled data, was introduced [49]. Figure 2 gives a schematic representation
of artificial intelligence, machine learning, and deep learning over the years.
Deep learning is a method that resembles artificial neural networks consisting of
layers. In other words, they are multilayer neural networks. In this method, unlike
machine learning, the inputs used do not have tags. For example; in a 4-feature dataset
consisting of crown and sepal lengths and widths of three different iris flower species,
the 5th column includes the iris flower species. When this data set is modeled with
machine learning, the machine trained with 4 features learns the type of iris flower
against the features in each row and produces a prediction model for deciding the
type of iris flower according to these features. Thus, the machine can determine the
type of iris flower by reading the properties of new data. In deep learning, unlike
machine learning, the data given as input to the machine is defined as a node, not a
feature. In this method is added as many nodes (4 for this example) as the column
with the number of features in the layer with the inputs. Each node is multiplied with
a random weight coefficient close to 0, and outputs are obtained through the transfer
function and activation functions. Starting from the first row in the data set, if the
270 A. Ocak et al.

Fig. 2 Artificial Intelligence, Machine Learning, and Deep Learning chronological display

estimated iris flower type output and the actual value are not the same, the error is
calculated and feedback is sent to the network. As a result, the weight coefficients are
updated. All rows of the data set are used for this task, and the weighting coefficients
of the attributes in the data reach a correct value. Figure 3 shows the artificial neuron
model and Fig. 4 shows the layers for iris data.
Backpropagation is one of the methods used in deep learning. In the deep learning
system, the weight coefficients to multiply each feature in the data set are determined
randomly. The margin of error of the determined weights in the prediction model is
important. Backpropagation is used to minimize this margin of error.
There are varieties of deep learning such as Convolutional Neural Networks
(CNNs) and Recurrent Neural Networks (RNNs). Convolutional Neural Networks

Fig. 3 Artificial neuron model [50]

Artificial Intelligence and Deep Learning in Civil Engineering 271

Fig. 4 Iris dataset deep learning layers

are a type of deep learning used in image processing that enables classification and
discrimination by capturing the features of images. In this method, images are used
as input. The pixels in which each image is reduced to small squares are learned
with layers and visual classification of objects is provided. The machine, which has
learned the features of the visuals it has divided into classes, can guess what the new
visual is when it encounters a new visual.
Recurrent Neural Networks is an artificial intelligence deep learning technology
that provides successful results in predicting the next step used for time series prob-
lems. The Recurrent Neural Networks method is frequently used in price deter-
mination and natural language processing of systems that are constantly changing
depending on time, such as the stock market. In deep learning, while the features
given as input are independent, in Recurrent Neural Networks, a relationship is
established between the inputs, and this relationship is remembered in the training
by memorizing.
The deep learning method is a technology used in many fields such as computer
science, medicine, physics, mathematics, materials science, biochemistry, genetics,
social sciences, and so on.

2.2 Deep Learning in Civil Engineering

Deep learning applications have attracted attention in civil engineering as well as

in other fields of science. The development of deep learning has been on the rise in
the twenty-first century because it requires large amounts of data due to its structure
and faster processors facilitating implementation. In the field of civil engineering,
many data sources such as observation reports obtained from buildings, displacement
records caused by various vibrations in buildings, flow-energy data obtained from
water structures such as hydroelectric power plants and dams, information reports
where construction management costs are stored, damage assessment data of struc-
tures have prepared the ground for deep learning applications. A large amount of
272 A. Ocak et al.

data sources in the construction industry has increased the use of deep learning in
this industry. It is frequently used especially in building damage detection, material
quality, and monitoring of building health.

2.2.1 Deep Learning Studies in Geotechnics

Deep learning technology started to be used intensively after 2006, and its application
to civil engineering studies gained frequency after ten years. Applications of deep
learning techniques in the department of geotechnical have been seen in subjects
such as bearing capacity of foundations, settlement of soil, soil permeability, slope
safety, slope stability, and soil classification [51–59]. A summary table of the studies
in the field of geotechnical is given in Table 1.

2.2.2 Deep Learning Studies in Construction Materials

Deep learning is an artificial intelligence method that allows the processing of visual
data. Thanks to this feature, image data from picture videos can be easily used as
training data. In civil engineering, it is known that the damage and determination of
the materials used in the construction process of the buildings for various reasons
are important for the health of the building. Construction materials are an area that
needs to be kept under control due to reasons such as cracks and deterioration that
may occur during and after application, reducing the quality of the material and
reducing the strength level. The application of deep learning techniques in the field of
construction materials ensures that negative factors such as time and cost are avoided
in the detection of material defects. When the studies in this field are examined, deep
learning methods are applied for many subjects such as the detection of concrete
defects, concrete compressive strength, corrosion, concrete slump and workability,
detection of surface cracks, changes in the mechanical properties of concrete under
high temperature, and so on. In Table 2, some of the studies on the construction
material are summarized.

2.2.3 Deep Learning Studies in Construction Management

Energy consumption of buildings is a parameter that should be considered in reducing

environmental impacts such as carbon emissions. In today’s world, for the design
of sustainable structures, both the construction materials and the energy embedded
in the structure must be well controlled. Deep learning methods are an artificial
intelligence technology preferred for the control of energy consumption and the
detection of negative situations. In addition to the energy monitoring advantages of
models created with deep learning methods in construction management, it acts as a
very good detector for the detection of dangerous worker behaviors and vehicle usage
in terms of construction safety. In deep learning studies in the field of construction
Artificial Intelligence and Deep Learning in Civil Engineering 273

Table 1 Deep learning studies in geotechnics

Study summary Authors
Estimating the carrying capacity of shallow foundations with deep Bagińska, M., &
neural networks (DNNs) when data is scarce has been investigated, Srokosz, P. E
and it has been observed that different numbers of neurons and layers
are more effective in the prediction performance of the model than the
number of layers [51]
The deep learning method has been used to estimate the shaft Lu, S. L., Zhang, N.,
resistance of cast in situ piles in reclaimed soils, it has been verified to Shen, S., Zhou, A., &
be efficient in determining the shaft resistance in offshore areas [52] Li, H. Z
Traditional methods and deep learning methods were compared for Azmoon, B., Biniyaz,
slope stability analysis, and an accuracy and calculation efficiency A., & Liu, Z
exceeding 99% was obtained [53]
In the tunneling process, two models were developed with deep Lee, H. K., Song, M.
learning for the determination of the parameters affecting the K., & Lee, S. S
settlement of the surface and the prediction of collapse against these
parameters, it was determined that the models predicted collapse at a
very good level [54]
The use of machine learning and deep learning methods to predict Tang, L., & Na, S
settlement in the ground as a result of tunneling and the applicability
of the prediction model were examined [55]
A deep learning method was applied to estimate the permeability of Liu, Y., Chen, S. J.,
consolidated silty clay [56] Sagoe-Crentsil, K., &
Duan, W
To avoid slope shifts, the deep learning method was applied to the Pandey, V. H. R.,
simulated model by changing various features, and a high-precision Kainthola, A., Sharma,
model was produced that predicts the safety factors [57] V., Srivastav, A., Jayal,
T., & Singh, T. N
For the evaluation of the soil taken from the excavation sites, a model Zhan, L. T., Guo, Q.
has been developed that can classify soils with an accuracy of close to M., Chen, Y. M., Wang,
90%, using deep learning and cone penetration testing [58] S. Y., Feng, T., Bian,
Y., … & Yin, Z. Y
In land reclamation studies, deep learning algorithms were used to Chen, X. X., Yang, J.,
predict the settlement of the ground, and the created model He, G. F., & Huang, L.
successfully predicted the settlement of the airport fill [59] C

management, subjects such as monitoring building health, monitoring of construction

workers, detection of unsafe behavior, price estimation of basic construction costs,
and the efficiency of construction equipment are covered. Some examples of the
content of the studies carried out are shown in Table 3.

2.2.4 Deep Learning Studies in Mechanics

Structures can be damaged by various dynamic effects such as wind and earthquakes.
Various analysis methods are used to determine the damage of such vibrations in
274 A. Ocak et al.

Table 2 Deep learning studies in construction materials

Study summary Authors
A good image-based approach is provided by detecting the Protopapadakis, E., &
defects of the concrete in the tunnels utilizing robotic Doulamis, N
controllers trained with deep learning [60]
The data of concrete block walls consisting of recycled fine Deng, F., He, Y., Zhou, S., Yu,
and coarse aggregate, water-cement ratio, and fly ash ratio Y., Cheng, H., & Wu, X
were used in deep learning, and more successful results were
obtained in estimating the compressive strength of concrete
compared to the classical neural network [61]
In the detection of corrosion, a prediction model was created Atha, D. J., & Jahanshahi, M. R
using different convolutional neural networks based on texture
and color analysis, and the proposed methods were compared
[62]
By applying deep learning with image data, a model was Ding, Z., & An, X
developed that takes into account the slump flow value and
funnel flow time of self-compacting concrete and predicts its
workability [63]
In the estimation of the compressive strength of concrete, the Shin, H. K., Ahn, Y. H., Lee, S.
model was trained with a deep convolutional neural network H., & Kim, H. Y
(DCNN) using images obtained from various digital images
and video recordings, and it was determined that the method
was an advantageous option [64]
In the detection of concrete surface cracks, a deep Yamane, T., & Chun, P. J
learning-based high-sensitivity crack detection model has
been developed in which the factors that negatively affect
detection such as mold marks, shadows, dirt, and tie rod holes
can be removed [65]
Convolution-based deep learning method was used to predict Chen, H., Yang, J., & Chen, X
the change in the mechanical response of fiber-reinforced
concrete at high temperatures, such as fire, and a successful
prediction model was developed [66]
By applying different temperatures to the concrete samples Hacıefendioğlu, K. E. M. A. L.,
produced in the laboratory, deep learning techniques were Akbulut, Y. E., Nayır, S. A. F.
applied to the data set including the strengths, and the A., Başağa, H. B., & Altunışık,
methods were compared to the strength of the concrete A. C
damaged by fire [67]
For the detection of concrete defects, an image processing Gonthina, M., Chamata, R.,
method with a convolutional neural network was used, and the Duppalapudi, J., & Lute, V
quantities of cracks, damage, and similar situations were
determined with the trained model [68]
A self-healing mortar has been produced, and a model has Jin, X., Haider, M. Z., Cui, Y.,
been developed that evaluates the ability to heal cracks with a Jang, J. G., Kim, Y. J., Fang, G.,
deep learning method with an accuracy of up to 99% [69] & Hu, J. W
Artificial Intelligence and Deep Learning in Civil Engineering 275

Table 3 Deep learning studies in construction management

Study summary Authors
In the security and control of tunnels, a deep learning system has been Makantasis, K.,
established with the images obtained from the camera images, and a flaw Protopapadakis, E.,
detector that makes fast predictions has been obtained [70] Doulamis, A.,
Doulamis, N., &
Loupos, C
In building energy consumption, a prediction model was produced with Amasyali, K., &
deep learning and machine learning-based algorithms that take into El-Gohary, N
account outdoor weather conditions, the deep learning model was more
successful in the prediction performance comparison [71]
An extreme deep learning approach has been proposed for the detection Li, C., Ding, Z.,
of building energy consumption, and the performance of the approach Zhao, D., Yi, J., &
has been compared with various machine learning-based methods [72] Zhang, G
To ensure safe working control of construction workers, a deep learning Fang, Q., Li, H., Luo,
method was applied using images obtained from far-field surveillance X., Ding, L., Luo, H.,
records, and the helmet uses the status of the workers was determined Rose, T. M., & An,
with high sensitivity [73] W
For the action definition of excavation excavators, the model was trained Kim, J., & Chi, S
from around 72 thousand excavation field images with convolutional
neural network (CNN) and double-layer long-short-term memory
algorithms, and 80–94% accurate action definition was obtained from
case studies [74]
A model that infers construction practice constraints from construction Zhong, B., Xing, X.,
regulations was trained with a hybrid deep learning method, and the Luo, H., Zhou, Q.,
performance of the model in inferring construction procedures was Li, H., Rose, T., &
successful [75] Fang, W
For structural health monitoring, deep learning was applied with a visual Zhang, Y., Sun, X.,
dataset consisting of images of tight and loose bolts, and a high-precision Loh, K. J., Su, W.,
prediction model was developed that predicts bolt damage in an Xue, Z., & Zhao, X
image-based structure [76]
Using the generic data set, a good detector was obtained from the Neuhausen, M.,
construction site videos from the models trained with deep learning and Pawlowski, D., &
machine learning techniques, and it was found that both methods gave König, M
successful results [77]
Deep learning was applied with image processing data to detect unsafe Fang, W., Love, P. E.,
behaviors in the construction site and to take precautions, it was Ding, L., Xu, S.,
recommended that construction companies use the model created to Kong, T., & Li, H
prevent accidents [78]
A convolutional neural network model has been developed to detect Feng, C. and Chiang,
changes in iron prices in the construction industry, and the accuracy of Y
the model has been tested with events affecting iron prices [79]
Deep learning was provided with the images of the railway bridge Park, S. M., Lee, J.
obtained by web browsing for the detection of the building components H., & Kang, L. S
in the construction area, it was observed that the shooting angle of the
images greatly affected the estimation sensitivity in object detection [80]
(continued)
276 A. Ocak et al.

Table 3 (continued)
Study summary Authors
Using deep learning to measure equipment efficiency, the model was Mahamedi, E.,
trained with kinematic and noise data, and an accuracy exceeding 99% Rogage, K., Doukari,
was obtained in estimating excavator performance [81] O., & Kassem, M

the structure. However, factors such as time and cost extend the determination of
properties such as seismic fragility of the structure over a long period. Another
structural problem is that the constructed structures have a limited lifespan. Among
the main reasons for this is the fatigue share created by environmental and dynamic
effects in the structure. Problems such as the inability to exhibit the same rigidity on
all floors over time in a structure subject to seismic excitation and the formation of soft
stories further reduce the life of the structure. The efficiency of deep learning methods
has proven to be an artificial intelligence suitable for use in this field. Determining
structural damage after various disasters only with visual data and video images
shows that deep learning can be adapted to structural mechanics problems. In the
field of structural mechanics, deep learning techniques are applied in many areas
such as fragility level detection, vibration-induced damage detection, determination
of soft-story structures, and fatigue life of structures. In Table 4, some of the studies
carried out with deep learning techniques in mechanics are summarized.

2.2.5 Deep Learning Studies in Transportation

Transportation branch researches the issues of establishing the necessary infras-

tructure in civil engineering, land, air, and water transportation, designing and
constructing transportation facilities, and operating them in a safe, economical, and
environmentally friendly manner. Deep learning techniques can be easily applied
in the field of transportation, such as damage detection of transportation resources,
detection of negative effects of dynamic vibrations on highway and railway lines,
and monitoring of transportation structures. Similar to the use of image processing
technology in construction materials, deep learning methods give successful results
in determining the damage to pavement surfaces as a result of various factors on high-
ways. Some of the deep learning studies in the field of transportation are summarized
in Table 5.

2.2.6 Deep Learning Studies in Hydraulics

Hydraulics branch provides the most appropriate project design and safe construc-
tion facility by examining natural disasters such as water structures, drinking water
and waste water channels, port facilities, abutment structures, hydroelectric power
plants, floods, and tsunamis. It carries out the construction of dams and similar water
structures with observation stations that collect precipitation data to keep various
Artificial Intelligence and Deep Learning in Civil Engineering 277

Table 4 Deep learning studies in mechanics

Study summary Authors
A damage detection model was developed using a deep neural network with Lin, Y. Z.,
dynamic responses of beam-like structures, and it was found that deep neural & Ma, H.
network technology gave positive results in damage detection [82] W
Parameters calculated from the finite element model of a real bridge are used in Darsono,
deep learning training for the detection and calibration of a reinforced concrete D., &
bridge against aging [83] Torbol, M
Low-level sensor data has been used in deep learning for structural damage Lin, Y. Z.,
detection, and it has been observed from the data that there are features that Nie, Z. H.,
increase the success of the method and damage locations [84] & Ma, H.
W
For the prediction of the fatigue life of the structures, the deep learning model Gulgec, N.
trained with acceleration data was tested with loading experiments on a beam S., Takáč,
suitable for the laboratory environment [85] M., &
Pakzad, S.
N
To reduce the damage of wind-induced vibrations to the structure, an aerodynamic Li, S.,
mitigating shape optimization model was created with a deep reinforcement Snaiki, R.,
learning method [86] & Wu, T
A convolutional neural network method trained with image data was used to detect Chen, P.
soft-skinned structures to reduce seismic risk, and parameter sensitivity limits were Y., Wu, Z.
determined by comparing them with the latest technology convolutional neural Y., &
network techniques [87] Taciroglu,
E
A method that uses seismic fragility analysis to examine the response of various Wang, M.,
disasters and vibrations due to aging in bridges, as an aid in increasing the Zhang, H.,
prediction success of deep learning, is proposed and its accuracy is supported by Dai, H., &
examples [88] Shen, L
To determine the fragility level created by the wind load on the transmission tower, Li, H.,
a prediction model was developed using the structural health monitoring data of the Zhang, W.
tower, and it was observed that the model could detect the fragility effectively [89] & Fu, X
To detect the vibration-induced damage caused by dynamic traffic loads on a bridge Sajedi, S.,
with a high-resolution trident instrument, damage data were generated from the & Liang,
vehicle transition simulation with real vehicle data, and a high estimation success X
was obtained as a result of training with deep learning techniques [90]
Three deep learning methods were used for the time-history response prediction of Li, H.,
vibrations arising from the movements of vehicles in bridge traffic, the fastest Wang, T.,
prediction model was observed in a deep feed-forward neural network (DFNN), Yang, J. P.,
while the best prediction performance was found in convolutional neural network & Wu, G
(CNN) [91]

water sources under control. Hydraulic engineering plays an important role in the
calculations and design of hydroelectric power plants to obtain clean energy from
water in the most efficient way. In addition, by examining the characteristics of
water under coastal engineering, safe construction of port structures and issues such
as tsunamis, and flood are the subjects of his research. Looking at the research areas,
278 A. Ocak et al.

Table 5 Deep learning studies in transportation

Study summary Authors
For the estimation of pavement friction level affecting road safety, Yang, G., Li, Q. J.,
training with pavement texture data was applied using a convolutional Zhan, Y., Fei, Y., &
neural network (CNN), and it was determined that deep learning Zhang, A
techniques were suitable for road speed and pavement friction [92]
For the prediction of vehicle-body vibrations on high-speed railway Ma, S., Gao, L., Liu,
lines, the model produced using an integrated deep-learning method X., & Lin, J
has successfully predicted vehicle-body vibrations at frequencies
below a certain value [93]
In the detection of asphalt pavement cracks that threaten highway Fei, Y., Wang, K. C.,
safety, deep learning methods with 3D pixel-based image processing Zhang, A., Chen, C.,
data have been applied, and it has been observed that the applied Li, J. Q., Liu, Y., … &
methods show good results in detecting thin cracks [94] Li, B
The deep learning method was used for moisture content determination Zhang, J., Yang, X., Li,
in asphalt pavements, and it was determined by experimental results W., Zhang, S., & Jia, Y
that the produced prediction model significantly detected moisture
damage [95]
A model trained with drone images was created for various road Samma, H., Suandi, S.
damage detection using a pre-trained convolutional neural network A., Ismail, N. A.,
(CNN) with a two-layer optimizer [96] Sulaiman, S., & Ping,
L. L
In subway tunnels, a model was created for the detection of water leaks Qiu, D., Liang, H.,
by using the hybrid supervised deep learning method with the image Wang, Z., Tong, Y., &
data obtained from mobile vision systems, and the hybrid deep learning Wan, S
method was more successful than other deep learning methods [97]
To detect and classify highway cracks, pre-trained convolutional neural Elghaish, F., Talebi, S.,
network (CNN) deep learning methods were used to produce a Abdellatef, E.,
prediction model with visual data, and it was observed that the Matarneh, S. T.,
proposed method successfully tested highway cracks in a short time Hosseini, M. R., Wu,
[98] S., & Nguyen, T. Q
A deep learning technique has been proposed for the detection of Wen, T., Ding, S.,
pavement deterioration on asphalt surfaces, and it has been determined Lang, H., Lu, J. J.,
that the model trained with the pavement deterioration dataset can Yuan, Y., Peng, Y., …
make inferences with an accuracy exceeding 80% and in a good time & Wang, A
[99]
For the structural health monitoring of bridges, in the detection of Ye, X., Wu, P., Liu, A.,
abnormal data caused by heavy rain and similar extreme events, image Zhan, X., Wang, Z., &
processing and convolutional neural network (CNN)-based deep Zhao, Y
learning was performed, and the accuracy of the technique was
evaluated with long span suspension bridge data [100]
To realize the designs to reduce the vibration created by subway trains Liu, W., Liang, R.,
while they are in motion, a vibration evaluation model has been Zhang, H., Wu, Z., &
developed by using a deep learning method to identify train-induced Jiang, B
vibrations and detect vibrations created by subway transitions [101]
Artificial Intelligence and Deep Learning in Civil Engineering 279

the applicability of deep learning techniques is examined by various studies. Chan-

nels built for the transmission of clean water and removal of harmful wastewater
are damaged by precipitation and seismic dynamic effects over time. Detection of
leaks and damages in water pipes, which adversely affect human health, is a difficult
and costly process to control, especially in underground networks. In addition to the
water channels, the cracks in the dam structures also threaten the living conditions in
the region. Deep learning techniques have yielded successful results in the detection
of leaks and damage to water structures, taking into account the data consisting of
water pressure, flow rate, and similar properties. In the research, it has been under-
stood that deep learning methods have a high potential in crack detection with visual
monitoring data and flood and leak prediction with the flow characteristics of water
and the change in precipitation amount. Some of the results obtained from a few
studies are given in Table 6.

2.2.7 Deep Learning Studies in Structures

Structures is a branch of civil engineering that deals with the construction and
design of buildings, bridges, towers, silos, and similar structures with steel, rein-
forced concrete, wood, metal, and similar materials, and examines the reactions of
structures to various static and dynamic effects. It works on the economical design
of buildings with sufficient strength, and a safe carrier system, suitable for the seis-
micity of the region and the ground structure. The task of the structural engineers
starts with the necessary static analysis during the design phase and continues with
the correct application, the protection, and monitoring of the health of the building,
which is affected by various dynamic and environmental factors after the application.
Applying deep learning techniques to solve such structural problems facilitates the
task of structural engineers and leads the way in the establishment of an autonomous
system. It is applied in many areas such as deep learning techniques, structural stress
and damage detection, design optimization, and cable force adjustment in the building
department. Summary information on the content of some studies conducted in Table
7 is presented.

3 Summary and Conclusions

In this study, artificial intelligence technology was examined in general, and deep
learning techniques, which is a branch of artificial intelligence, and its current appli-
cations in civil engineering were mentioned. In the literature research, it has been
supported by the studies that deep learning techniques, especially the issues such as
structural damage detection, and road and construction safety related to the image
processing area, can be easily estimated through recorded images and videos. It has
been determined that the deep learning method has a successful performance in the
detection of ordinary situations such as floods and floods and leaks in sewer systems
280 A. Ocak et al.

Table 6 Deep learning studies in hydraulics

Study summary Authors
To detect the faults in the sewer pipes, a model created using deep learning Cheng, J. C., &
techniques with closed-circuit television images has been trained, and a Wang, M
system that provides high-sensitivity detection has been produced [102]
It has been observed that unmeasured watershed attributes improve Ojha, R., &
hydrological predictions in the deep-learning estimation of flood quantities Tripathi, S
[103]
To ensure the cyber security of water distribution systems, a detection Taormina, R., &
model has been created by the deep learning method, and the trained model Galelli, S
has successfully predicted all cyber attacks that threaten the security of
hydraulic distribution networks [104]
In the water distribution system, deep learning and artificial intelligence Guo, G., Liu, S.,
algorithms were used to predict water demand for short periods, such as Wu, Y., Li, J.,
15 min, and the model created by deep learning was found to perform Zhou, R., & Zhu,
better in 15-min and 24-h water demand forecasting [105] X
The convolutional neural network (CNN) model, which best predicts eight Park, J., Lee, H.,
algal species, has been determined for the monitoring and classification of Park, C. Y., Hasan,
algae that negatively affect drinking water lines, and it has been presented S., Heo, T. Y., &
with experimental results that it is an effective model [106] Lee, W. H
A deep learning technique was applied to detect leaks in water distribution Cody, R. A.,
pipes using acoustic monitoring data, and the model succeeded in detecting Tolson, B. A., &
0.25 L per second leaks with an accuracy of up to 97% [107] Orchard, J
A model was prepared by applying hybrid deep learning training with Dong, S., Yu, T.,
channel network sensor data, and it was verified by performing a real flood Farahmand, H., &
event prediction test that the trained model was suitable for flood Mostafavi, A
forecasting and taking precautions for flood warnings [108]
The convolutional neural network (CNN) method was used to predict the Wang, H. W., Lin,
long-term temporal flood range and depth of the structures, and it was G. F., Hsu, C. T.,
observed that the model was successful and could obtain predictions within Wu, S. J., &
seconds [109] Tfwala, S. S
To predict the deformation of mortar masonry dams against time-varying Su, Y., Zheng, Z.,
factors such as temperature and aging, a deep learning model has been Lin, C., Lin, Y.,
constructed and it has been verified that it provides high-precision dam He, Q., Zhang, T.,
deformation prediction [110] & Huang, S
For the detection of damage such as corrosion, scaling, and tearing in Wang, N., Fang,
buried sewer pipes, a prediction model has been produced with various H., Xue, B., Wu,
deep learning and artificial intelligence algorithms, and a real-time damage R., Fang, R., Hu,
estimation platform has been proposed [111] Q., & Lv, Y

for the hydraulics branch, which is one of the civil engineering fields. Based on the
studies in the field of construction management, it is seen that deep learning is an
appropriate method for determining the conditions such as monitoring the health
of the structure, controlling the work and worker safety during the construction
phase, and evaluating the performance of construction equipment. In addition, it has
been determined by studies that construction materials are very effective in deter-
mining strength, cracks, and damage. The use of artificial intelligence technologies
Artificial Intelligence and Deep Learning in Civil Engineering 281

Table 7 Deep learning studies in structures

Study summary Authors
A design optimization based on deep learning neural networks that Ferreiro-Cabello, J.,
improve flooring design to reduce the negative environmental impacts Fraile-Garcia, E., de
of floors such as carbon emissions and embedded energy has been Pison Ascacibar, E. M.,
developed and tested with a case study [112] & Martinez-de-Pison,
F. J
Using deep learning technologies, a model has been produced that Gao, Y., & Mosalam,
predicts fragmentation, component condition, and damage level and K. M
type in the detection of structural damage by image processing [113]
Deep learning techniques have been used to determine the load Chen, G., Li, T., Chen,
conditions and damage conditions that cause plastic deformation of the Q., Ren, S., Wang, C.,
shell structures [114] & Li, S
By creating a deep learning model with various building images, it has Yu, Q., Wang, C.,
been tried to determine the structures with soft stories that may cause McKenna, F., Yu, S. X.,
destruction [115] Taciroglu, E., Cetiner,
B., & Law, K. H
With a model trained with deep learning techniques, the total stress of Wang, W., Shi, P., Chu,
steel components was tried to be estimated, and a proposal was H., Deng, L., & Yan, B
presented for stress detection, which is difficult to obtain with
conventional sensors [116]
A multi-layer perceptron deep learning method was used to adjust the Shan, D., Zhang, X.,
cable strength of cable-suspended bridges with long spans, and a fast Gu, X. & Li, Q
and high-precision model was produced [117]
A deep learning method has been used for structural health and a Guo, J., Wang, Q., &
successful prediction model has been developed that detects facade Li, Y
defects [118]
By using multilayer perceptron, long-short-term memory network, 1D Dang, H. V., Raza, M.,
convolutional neural network, and convolutional neural network deep Nguyen, T. V.,
learning techniques, the suitability of these techniques in detecting Bui-Tien, T., &
multiple damage situations was investigated and a table was created Nguyen, H. X
[119]
Column connections of reinforced concrete slabs were modeled with Derogar, S., Ince, C.,
various artificial intelligence, machine learning, artificial neural Yatbaz, H. Y., & Ever,
networks, and deep learning methods using drilling-shear experiment E
data, and the puncture shear strength could be estimated with sufficient
accuracy [120]
In the structural stress distribution, a model that performs stress Bolandi, H., Li, X.,
analysis of loaded steel plates is used convolutional neural networks Salem, T., Boddeti, V.
(CNNs) are constructed, and finite element simulation is used to N., & Lajnef, N
compare the performance of the model [121]
In earthquakes and similar extreme events, a prediction model was Bai, Y., Zha, B., Sezen,
created for the unmanned detection of damage to the structure by using H., & Yilmaz, A
image processing and deep learning methods, and remarkable results
were obtained [122]
282 A. Ocak et al.

in the classification of the soil and determination of the bearing capacity, such as
soil surveys, which is among the first processes for the construction of a construc-
tion, brings speed and accuracy to the project. Situations for the areas of structure
and structure mechanics expressing the subsequent processes of the construction,
having the most suitable carrier system, and protection against earthquake, wind,
and similar dynamic effects after construction are important. Deep learning seems
to be a good predictor and damage estimator in civil engineering problems such as
structural damage, fatigue life under dynamic effects, and detection of load states that
trigger structural deformation. Looking at the studies in the field of transportation,
crack detection that threatens road safety, strength estimation of coating materials,
and similar areas have been the areas where deep learning is frequently used. In light
of all the studies, it is understood that deep learning techniques are in a structure that
can be used easily at every stage of construction and produce the necessary fore-
casting models for the optimum level of efficiency to be obtained from construction
materials, as well as a good damage assessment expert and the determination of the
elements that threaten the structure safety. It can be said that it is a suitable method
for use in a long process, both at the beginning of the construction and after the
construction, until deciding on the life of the structure.

References

1. Mokyr, J.: The Industrial Revolution and the economic history of technology: Lessons from the
British experience, 1760–1850. Quarterly Review of Economics & Finance 41(3), 295–295
(2001)
2. McCarthy, J. (2020, June 14). Retrieved from https://www-formal.stanford.edu/jmc//.
3. Sajja, P. S., & Sajja, P. S. (2021). Introduction to artificial intelligence. Illustrated Computa-
tional Intelligence: Examples and Applications, 1–25.
4. https://www.balikesir.edu.tr/~ieee/index.php/2019/10/29/yapay-zeka-artificial-intelligence/
# , [Visit date:20 February 2023].
5. French, R.M.: The Turing Test: the first 50 years. Trends Cogn. Sci. 4(3), 115–122 (2000)
6. Dhar, V.: Data science and prediction. Commun. ACM 56(12), 64–73 (2013)
7. Wright, J., Yang, A.Y., Ganesh, A., Sastry, S.S., Ma, Y.: Robust face recognition via sparse
representation. IEEE Trans. Pattern Anal. Mach. Intell. 31(2), 210–227 (2008)
8. Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., ... & Hassabis,
D. (2017). Mastering the game of go without human knowledge. nature, 550(7676), 354–359.
9. McKinney, S.M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., Shetty,
S.: International evaluation of an AI system for breast cancer screening. Nature 577(7788),
89–94 (2020)
10. Tien Bui, D., Tuan, T.A., Klempe, H., Pradhan, B., Revhaug, I.: Spatial prediction models
for shallow landslide hazards: a comparative assessment of the efficacy of support vector
machines, artificial neural networks, kernel logistic regression, and logistic model tree.
Landslides 13, 361–378 (2016)
11. Vaishya, R., Javaid, M., Khan, I.H., Haleem, A.: Artificial Intelligence (AI) applications for
the COVID-19 pandemic. Diabetes Metab. Syndr. 14(4), 337–339 (2020)
12. Perol, T., Gharbi, M., Denolle, M.: Convolutional neural network for earthquake detection
and location. Sci. Adv. 4(2), e1700578 (2018)
13. Kosinski, M., Stillwell, D., Graepel, T.: Private traits and attributes are predictable from digital
records of human behavior. Proc. Natl. Acad. Sci. 110(15), 5802–5805 (2013)
Artificial Intelligence and Deep Learning in Civil Engineering 283

14. Colleoni, E., Rozza, A., Arvidsson, A.: Echo chamber or public sphere? Predicting political
orientation and measuring political homophily in Twitter using big data. J. Commun. 64(2),
317–332 (2014)
15. Ferentinou, M., & Fakir, M. (2017, June). An ANN approach for the prediction of uniaxial
compressive strength, of some sedimentary and igneous rocks in eastern KwaZulu-Natal.
In ISRM European Rock Mechanics Symposium-EUROCK 2017. OnePetro.
16. Moradi, M. H., Sohani, A., Zabihigivi, M., Wagner, U., Koch, T., & Sayyaadi, H. (2022).
Machine learning and artificial intelligence application in land pollution research. Current
Trends and Advances in Computer-Aided Intelligent Environmental Data Engineering, 273–
296.
17. Dupré, A., Drobinski, P., Alonzo, B., Badosa, J., Briard, C., Plougonven, R.: Sub-hourly
forecasting of wind speed and wind energy. Renewable Energy 145, 2373–2379 (2020)
18. Emil, S., Tomas, M.: Introducing Compressed Mixture Models for Predicting Long-Lasting
Brake Events. IFAC-PapersOnLine 51(31), 840–845 (2018)
19. Saha, S., Changdar, S., & De, S. (2022). Prediction of the stability number of conventional
rubble-mound breakwaters using machine learning algorithms. Journal of Ocean Engineering
and Science.
20. Kebisek, M., Tanuska, P., Spendla, L., Kotianova, J., & Strelec, P. (2020). Artificial intelligence
platform proposal for paint structure quality prediction within the industry 4.0 concept. IFAC-
PapersOnLine, 53(2), 11168–11174.
21. Ur Rehman, Z., Khalid, U., Ijaz, N., Mujtaba, H., Haider, A., Farooq, K., Ijaz, Z.: Machine
learning-based intelligent modeling of hydraulic conductivity of sandy soils considering a
wide range of grain sizes. Eng. Geol. 311, 106899 (2022)
22. Aydın, Y., Işıkdağ, Ü., Bekdaş, G., Nigdeli, S.M., Geem, Z.W.: Use of Machine Learning
Techniques in Soil Classification. Sustainability 15(3), 2374 (2023)
23. Ozsagir, M., Erden, C., Bol, E., Sert, S., Özocak, A.: Machine learning approaches for
prediction of fine-grained soils liquefaction. Comput. Geotech. 152, 105014 (2022)
24. Nguyen, Q. H., Ly, H. B., Ho, L. S., Al-Ansari, N., Le, H. V., Tran, V. Q., ... & Pham,
B. T. (2021). Influence of data splitting on the performance of machine learning models in
prediction of shear strength of the soil. Mathematical Problems in Engineering, 2021, 1-15.
25. Zhang, J., Li, D., Wang, Y.: Toward intelligent construction: Prediction of mechanical prop-
erties of manufactured-sand concrete using tree-based models. J. Clean. Prod. 258, 120665
(2020)
26. Feng, W., Wang, Y., Sun, J., Tang, Y., Wu, D., Jiang, Z., ... & Wang, X. (2022). Prediction of
thermo-mechanical properties of rubber-modified recycled aggregate concrete. Construction
and Building Materials, 318, 125970
27. Feng, D.C., Liu, Z.T., Wang, X.D., Chen, Y., Chang, J.Q., Wei, D.F., Jiang, Z.M.: Machine
learning-based compressive strength prediction for concrete: An adaptive boosting approach.
Constr. Build. Mater. 230, 117000 (2020)
28. Shen, J., Li, Y., Lin, H., Li, H., Lv, J., Feng, S., Ci, J.: Prediction of compressive strength of
alkali-activated construction demolition waste geopolymers using ensemble machine learning.
Constr. Build. Mater. 360, 129600 (2022)
29. Li, Y., Li, H., Shen, J.: The study of the effect of carbon nanotubes on the compressive
strength of cement-based materials based on machine learning. Constr. Build. Mater. 358,
129435 (2022)
30. Toufigh, V., Palizi, S.: Performance evaluation of slag-based concrete at elevated temperatures
by a novel machine learning approach. Constr. Build. Mater. 358, 129357 (2022)
31. Bardhan, A., Biswas, R., Kardani, N., Iqbal, M., Samui, P., Singh, M.P., Asteris, P.G.: A novel
integrated approach of augmented grey wolf optimizer and ANN for estimating axial load
carrying capacity of concrete-filled steel tube columns. Constr. Build. Mater. 337, 127454
(2022)
32. Zhou, Y., Zhang, Y., Pang, R., Xu, B.: Seismic fragility analysis of high concrete-faced rockfill
dams based on plastic failure with support vector machine. Soil Dyn. Earthq. Eng. 144, 106587
(2021)
284 A. Ocak et al.

33. Gholami, A., Bonakdari, H., Zaji, A.H., Akhtari, A.A.: A comparison of artificial intelligence-
based classification techniques in predicting flow variables in sharply curved channels.
Engineering with Computers 36, 295–324 (2020)
34. Zaji, A.H., Bonakdari, H.: Velocity field simulation of open-channel junction using artificial
intelligence approaches. Iranian Journal of Science and Technology, Transactions of Civil
Engineering 43, 549–560 (2019)
35. Ebtehaj, I., Bonakdari, H., Zaji, A.H., Azimi, H., Khoshbin, F.: GMDH-type neural network
approach for modeling the discharge coefficient of rectangular sharp-crested side weirs.
Engineering Science and Technology, an International Journal 18(4), 746–757 (2015)
36. Sun, H., Burton, H.V., Huang, H.: Machine learning applications for building structural design
and performance assessment: A state-of-the-art review. Journal of Building Engineering 33,
101816 (2021)
37. Chou, J.S., Liu, C.Y., Prayogo, H., Khasani, R.R., Gho, D., Lalitan, G.G.: Predicting the
nominal shear capacity of a reinforced concrete wall in the building by metaheuristics-
optimized machine learning. Journal of Building Engineering 61, 105046 (2022)
38. Zhang, J., Sun, Y., Li, G., Wang, Y., Sun, J., & Li, J. (2022). Machine-learning-assisted shear
strength prediction of reinforced concrete beams with and without stirrups. Engineering with
Computers, 1–15.
39. Mangalathu, S., Jang, H., Hwang, S.H., Jeon, J.S.: Data-driven machine-learning-based
seismic failure mode identification of reinforced concrete shear walls. Eng. Struct. 208, 110331
(2020)
40. Fathalla, E., Tanaka, Y., Maekawa, K.: Remaining fatigue life assessment of in-service road
bridge decks based upon artificial neural networks. Eng. Struct. 171, 602–616 (2018)
41. Matel, E., Vahdatikhaki, F., Hosseinyalamdary, S., Evers, T., Voordijk, H.: An artificial neural
network approach for cost estimation of engineering services. Int. J. Constr. Manag. 22(7),
1274–1287 (2022)
42. Flah, M., Nunez, I., Ben Chaabene, W., Nehdi, M.L.: Machine learning algorithms in civil
structural health monitoring: a systematic review. Archives of computational methods in
engineering 28, 2621–2643 (2021)
43. Sun, Z., Feng, D.C., Mangalathu, S., Wang, W.J., Su, D.: Effectiveness assessment of TMDs
in bridges under strong winds incorporating machine-learning techniques. J. Perform. Constr.
Facil. 36(5), 04022036 (2022)
44. Bae, J., Lee, C.H., Park, M., Alemayehu, R.W., Ryu, J., Ju, Y.K.: Modified low-cycle fatigue
estimation using machine learning for radius-cut coke-shaped metallic damper subjected to
cyclic loading. International Journal of Steel Structures 20, 1849–1858 (2020)
45. Yucel, M., Bekdaş, G., Nigdeli, S.M., Sevgen, S.: Estimation of optimum tuned mass damper
parameters via machine learning. Journal of Building Engineering 26, 100847 (2019)
46. Naser, M.Z.: An AI-based cognitive framework for evaluating the response of concrete
structures in extreme conditions. Eng. Appl. Artif. Intell. 81, 437–449 (2019)
47. Blake, R.W., Mathew, R., George, A., Papakostas, N.: Impact of artificial intelligence on
engineering: Past, present and future. Procedia CIRP 104, 1728–1733 (2021)
48. Gruson, D., Helleputte, T., Rousseau, P., Gruson, D.: Data science, artificial intelligence, and
machine learning: opportunities for laboratory medicine and the value of positive regulation.
Clin. Biochem. 69, 1–7 (2019)
49. LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. nature 521(7553), 436–444 (2015)
50. https://www.cc.ntu.edu.tw/chinese/epaper/0038/20160920_3805.html. [Visit date: 20
February 2023].
51. Bagińska, M., Srokosz, P.E.: The optimal ANN Model for predicting the bearing capacity of
shallow foundations is trained on scarce data. KSCE J. Civ. Eng. 23, 130–137 (2019)
52. Lu, S. L., Zhang, N., Shen, S., Zhou, A., & Li, H. Z. (2020). A deep-learning method for evalu-
ating shaft resistance of the cast-in-site pile on the reclaimed ground using field data. Zhejiang
University. Journal. Science A: Applied Physics & Engineering, 21(6), 496–508.
53. Azmoon, B., Biniyaz, A., Liu, Z.: Evaluation of deep learning against conventional limit
equilibrium methods for slope stability analysis. Appl. Sci. 11(13), 6060 (2021)
Artificial Intelligence and Deep Learning in Civil Engineering 285

54. Lee, H.K., Song, M.K., Lee, S.S.: Prediction of Subsidence during TBM Operation in Mixed-
Face Ground Conditions from Realtime Monitoring Data. Appl. Sci. 11(24), 12130 (2021)
55. Tang, L., Na, S.: Comparison of machine learning methods for ground settlement prediction
with different tunneling datasets. Journal of Rock Mechanics and Geotechnical Engineering
13(6), 1274–1289 (2021)
56. Liu, Y., Chen, S.J., Sagoe-Crentsil, K., Duan, W.: Predicting the permeability of consolidated
silty clay via digital soil reconstruction. Comput. Geotech. 140, 104468 (2021)
57. Pandey, V.H.R., Kainthola, A., Sharma, V., Srivastav, A., Jayal, T., Singh, T.N.: Deep
learning models for large-scale slope instability examination in Western Uttarakhand. India.
Environmental Earth Sciences 81(20), 487 (2022)
58. Zhan, L. T., Guo, Q. M., Chen, Y. M., Wang, S. Y., Feng, T., Bian, Y., ... & Yin, Z. Y. (2023).
An efficient classification system for excavated soils using soil image deep learning and TDR
cone penetration test. Computers and Geotechnics, 155, 105207
59. Chen, X. X., Yang, J., He, G. F., & Huang, L. C. (2023). Development of an LSTM-based
model for predicting the long-term settlement of land reclamation and a GUI-based tool. Acta
Geotechnica, 1–14.
60. Protopapadakis, E., & Doulamis, N. (2015). Image-based approaches for tunnels’ defects
recognition via robotic inspectors. In Advances in Visual Computing: 11th International
Symposium, ISVC 2015, Las Vegas, NV, USA, December 14–16, 2015, Proceedings, Part
I 11 (pp. 706–716). Springer International Publishing.
61. Deng, F., He, Y., Zhou, S., Yu, Y., Cheng, H., Wu, X.: Compressive strength prediction of
recycled concrete based on deep learning. Constr. Build. Mater. 175, 562–569 (2018)
62. Atha, D.J., Jahanshahi, M.R.: Evaluation of deep learning approaches based on convolutional
neural networks for corrosion detection. Struct. Health Monit. 17(5), 1110–1128 (2018)
63. Ding, Z., An, X.: Deep learning approach for estimating workability of self-compacting
concrete from mixing image sequences. Adv. Mater. Sci. Eng. 2018, 1–16 (2018)
64. Shin, H.K., Ahn, Y.H., Lee, S.H., Kim, H.Y.: Digital vision-based concrete compressive
strength evaluating model using deep convolutional neural network. CMC-COMPUTERS
MATERIALS & CONTINUA 61(2), 911–928 (2019)
65. Yamane, T., Chun, P.J.: Crack detection from a concrete surface image based on semantic
segmentation using deep learning. J. Adv. Concr. Technol. 18(9), 493–504 (2020)
66. Chen, H., Yang, J., Chen, X.: A convolution-based deep learning approach for estimating
the compressive strength of fiber-reinforced concrete at elevated temperatures. Constr. Build.
Mater. 313, 125437 (2021)
67. Hacıefendioğlu, K. E. M. A. L., Akbulut, Y. E., Nayır, S. A. F. A., Başağa, H. B., & Altunışık,
A. C. (2022). Automated Estimation of Exposed Temperature and Strength Changing Ratio
for Fire-Damaged Concrete Using Deep Learning Method. Experimental Techniques, 1–18.
68. Gonthina, M., Chamata, R., Duppalapudi, J., & Lute, V. (2022). Deep CNN-based concrete
crack identification and quantification using image processing techniques. Asian Journal of
Civil Engineering, 1–14.
69. Jin, X., Haider, M.Z., Cui, Y., Jang, J.G., Kim, Y.J., Fang, G., Hu, J.W.: Development of
nanomodified self-healing mortar and a U-Net model based on semantic segmentation for
crack detection and evaluation. Constr. Build. Mater. 365, 129985 (2023)
70. Makantasis, K., Protopapadakis, E., Doulamis, A., Doulamis, N., & Loupos, C. (2015,
September). Deep convolutional neural networks for efficient vision-based tunnel inspec-
tion. In 2015 IEEE international conference on intelligent computer communication and
processing (ICCP) (pp. 335–342). IEEE.
71. Amasyali, K., & El-Gohary, N. (2017). Deep learning for building energy consumption
prediction. In Proceedings of the 6th CSCE-CRC International Construction Specialty
Conference.
72. Li, C., Ding, Z., Zhao, D., Yi, J., Zhang, G.: Building energy consumption prediction: An
extreme deep learning approach. Energies 10(10), 1525 (2017)
73. Fang, Q., Li, H., Luo, X., Ding, L., Luo, H., Rose, T.M., An, W.: Detecting non-hardhat-use
by a deep learning method from far-field surveillance videos. Autom. Constr. 85, 1–9 (2018)
286 A. Ocak et al.

74. Kim, J., Chi, S.: Action recognition of earthmoving excavators based on sequential pattern
analysis of visual features and operation cycles. Autom. Constr. 104, 255–264 (2019)
75. Zhong, B., Xing, X., Luo, H., Zhou, Q., Li, H., Rose, T., Fang, W.: Deep learning-based
extraction of construction procedural constraints from construction regulations. Adv. Eng.
Inform. 43, 101003 (2020)
76. Zhang, Y., Sun, X., Loh, K.J., Su, W., Xue, Z., Zhao, X.: Autonomous bolt loosening detection
using deep learning. Struct. Health Monit. 19(1), 105–122 (2020)
77. Neuhausen, M., Pawlowski, D., König, M.: Comparing classical and modern machine learning
techniques for monitoring pedestrian workers in top-view construction site video sequences.
Appl. Sci. 10(23), 8466 (2020)
78. Fang, W., Love, P. E., Ding, L., Xu, S., Kong, T., & Li, H. (2021). Computer vision and deep
learning to manage safety in construction: Matching images of unsafe behavior and semantic
rules. IEEE Transactions on Engineering Management.
79. Feng, C., Chiang, Y.: Hybridizing Deep Learning with Google Trends to Predict Rebar Price
Fluctuation in Taiwan. Chinese Journal of Civil and Hydraulic Engineering 33(8), 595–604
(2021)
80. Park, S. M., Lee, J. H., & Kang, L. S. (2022). A Framework for Improving Object
Recognition of Structural Components in Construction Site Photos Using Deep Learning
Approaches. KSCE Journal of Civil Engineering, 1–12.
81. Mahamedi, E., Rogage, K., Doukari, O., Kassem, M.: Automating excavator productivity
measurement using deep learning. Proceedings of the Institution of Civil Engineers-Smart
Infrastructure and Construction 174(4), 121–133 (2022)
82. Lin, Y. Z., & Ma, H. W. (2016, December). Dynamic response-based damage detection for
beam-like structures using deep learning. In Proceedings of the 24th Australasian conference
on the mechanics of structures and materials (ACMSM24), Perth, WA, Australia (pp. 6–9).
83. Darsono, D., & Torbol, M. (2017, June). Calibration of a reinforced concrete bridge using
deep learning. WCCM.
84. Lin, Y.Z., Nie, Z.H., Ma, H.W.: Structural damage detection with automatic feature extraction
through deep learning. Computer-Aided Civil and Infrastructure Engineering 32(12), 1025–
1046 (2017)
85. Gulgec, N.S., Takáč, M., Pakzad, S.N.: Structural sensing with deep learning: Strain estima-
tion from acceleration data for fatigue assessment. Computer-Aided Civil and Infrastructure
Engineering 35(12), 1349–1364 (2020)
86. Li, S., Snaiki, R., Wu, T.: A knowledge-enhanced deep reinforcement learning-based shape
optimizer for aerodynamic mitigation of wind-sensitive structures. Computer-Aided Civil and
Infrastructure Engineering 36(6), 733–746 (2021)
87. Chen, P.Y., Wu, Z.Y., Taciroglu, E.: Classification of soft-story buildings using deep learning
with density features extracted from 3D point clouds. J. Comput. Civ. Eng. 35(3), 04021005
(2021)
88. Wang, M., Zhang, H., Dai, H., & Shen, L. (2022, June). A deep learning-aided seismic fragility
analysis method for bridges. In Structures (Vol. 40, pp. 1056–1064). Elsevier.
89. Li, H., Zhang, W., Fu, X.: Fragility assessment of a transmission tower subjected to wind
load based on big data and deep learning. Chinese Journal of Civil Engineering 55(9), 54–64
(2022)
90. Sajedi, S., Liang, X.: Trident: A Deep Learning Framework for High-Resolution Bridge
Vibration Monitoring. Appl. Sci. 12(21), 10999 (2022)
91. Li, H., Wang, T., Yang, J.P., Wu, G.: Deep learning models for time-history prediction of
vehicle-induced bridge responses: A comparative study. Int. J. Struct. Stab. Dyn. 23(01),
2350004 (2023)
92. Yang, G., Li, Q.J., Zhan, Y., Fei, Y., Zhang, A.: Convolutional neural network–based friction
model using pavement texture data. J. Comput. Civ. Eng. 32(6), 04018052 (2018)
93. Ma, S., Gao, L., Liu, X., Lin, J.: Deep learning for track quality evaluation of high-speed
railway based on vehicle-body vibration prediction. IEEE Access 7, 185099–185107 (2019)
Artificial Intelligence and Deep Learning in Civil Engineering 287

94. Fei, Y., Wang, K. C., Zhang, A., Chen, C., Li, J. Q., Liu, Y., ... & Li, B. (2019). Pixel-level
cracking detection on 3D asphalt pavement images through deep-learning-based CrackNet-
V. IEEE Transactions on Intelligent Transportation Systems, 21(1), 273-284
95. Zhang, J., Yang, X., Li, W., Zhang, S., Jia, Y.: Automatic detection of moisture damages in
asphalt pavements from GPR data with deep CNN and IRS method. Autom. Constr. 113,
103119 (2020)
96. Samma, H., Suandi, S.A., Ismail, N.A., Sulaiman, S., Ping, L.L.: Evolving Pre-Trained CNN
Using Two-Layers Optimizer for Road Damage Detection From Drone Images. IEEE Access
9, 158215–158226 (2021)
97. Qiu, D., Liang, H., Wang, Z., Tong, Y., Wan, S.: Hybrid-Supervised-Learning-Based Auto-
matic Image Segmentation for Water Leakage in Subway Tunnels. Appl. Sci. 12(22), 11799
(2022)
98. Elghaish, F., Talebi, S., Abdellatef, E., Matarneh, S.T., Hosseini, M.R., Wu, S., Nguyen, T.Q.:
Developing a new deep-learning CNN model to detect and classify highway cracks. Journal
of Engineering, Design, and Technology 20(4), 993–1014 (2022)
99. Wen, T., Ding, S., Lang, H., Lu, J. J., Yuan, Y., Peng, Y., ... & Wang, A. (2022). Automated pave-
ment distress segmentation on asphalt surfaces using a deep learning network. International
Journal of Pavement Engineering, 1–14.
100. Ye, X., Wu, P., Liu, A., Zhan, X., Wang, Z., & Zhao, Y. (2022). A Deep Learning-based
Method for Automatic Abnormal Data Detection: Case Study for Bridge Structural Health
Monitoring. International Journal of Structural Stability and Dynamics.
101. Liu, W., Liang, R., Zhang, H., Wu, Z., Jiang, B.: Deep learning-based identification and uncer-
tainty analysis of metro train induced ground-borne vibration. Mech. Syst. Signal Process.
189, 110062 (2023)
102. Cheng, J.C., Wang, M.: Automated detection of sewer pipe defects in closed-circuit television
images using deep learning techniques. Autom. Constr. 95, 155–171 (2018)
103. Ojha, R., Tripathi, S.: Using attributes of ungauged basins to improve regional regression equa-
tions for flood estimation: A deep learning approach. ISH Journal of Hydraulic Engineering
24(2), 239–248 (2018)
104. Taormina, R., Galelli, S.: Deep-learning approach to the detection and localization of cyber-
physical attacks on water distribution systems. J. Water Resour. Plan. Manag. 144(10),
04018065 (2018)
105. Guo, G., Liu, S., Wu, Y., Li, J., Zhou, R., Zhu, X.: Short-term water demand forecast based
on deep learning method. J. Water Resour. Plan. Manag. 144(12), 04018076 (2018)
106. Park, J., Lee, H., Park, C.Y., Hasan, S., Heo, T.Y., Lee, W.H.: Algal morphological iden-
tification in watersheds for drinking water supply using neural architecture search for the
convolutional neural network. Water 11(7), 1338 (2019)
107. Cody, R.A., Tolson, B.A., Orchard, J.: Detecting leaks in water distribution pipes using a deep
autoencoder and hydroacoustic spectrograms. J. Comput. Civ. Eng. 34(2), 04020001 (2020)
108. Dong, S., Yu, T., Farahmand, H., Mostafavi, A.: A hybrid deep learning model for predictive
flood warning and situation awareness using channel network sensors data. Computer-Aided
Civil and Infrastructure Engineering 36(4), 402–420 (2021)
109. Wang, H.W., Lin, G.F., Hsu, C.T., Wu, S.J., Tfwala, S.S.: Long-Term Temporal Flood
Predictions Made Using Convolutional Neural Networks. Water 14(24), 4134 (2022)
110. Su, Y., Zheng, Z., Lin, C., Lin, Y., He, Q., Zhang, T., Huang, S.: A spatiotemporal hybrid model
for deformation of mortar masonry dams with time-varying factor. Journal of Hydropower
Engineering 41(11), 124–138 (2022)
111. Wang, N., Fang, H., Xue, B., Wu, R., Fang, R., Hu, Q., Lv, Y.: Automatic Damage Segmen-
tation Framework for Buried Sewer Pipes Based on Machine Vision: Case Study of Sewer
Pipes in Zhengzhou. China. Journal of Infrastructure Systems 29(1), 04022046 (2023)
112. Ferreiro-Cabello, J., Fraile-Garcia, E., de Pison Ascacibar, E.M., Martinez-de-Pison, F.J.:
Metamodel-based design optimization of structural one-way slabs based on deep learning
neural networks to reduce environmental impact. Eng. Struct. 155, 91–101 (2018)
288 A. Ocak et al.

113. Gao, Y., Mosalam, K.M.: Deep transfer learning for image-based structural damage
recognition. Computer-Aided Civil and Infrastructure Engineering 33(9), 748–768 (2018)
114. Chen, G., Li, T., Chen, Q., Ren, S., Wang, C., Li, S.: Application of deep learning neural
network to identify collision load conditions based on permanent plastic deformation of shell
structures. Comput. Mech. 64, 435–449 (2019)
115. Yu, Q., Wang, C., McKenna, F., Yu, S.X., Taciroglu, E., Cetiner, B., Law, K.H.: Rapid visual
screening of soft-story buildings from street view images using deep learning classification.
Earthq. Eng. Eng. Vib. 19, 827–838 (2020)
116. Wang, W., Shi, P., Chu, H., Deng, L., Yan, B.: Deep learning framework for total stress
detection of steel components. J. Bridg. Eng. 26(1), 04020113 (2021)
117. Shan, D., Zhang, X., Gu, X., Li, Q.: Cable Force Adjustment for Long-Span Cable-Stayed
Bridge Based on Multilayer Perceptron Deep Learning. Bridge Construction 51(1), 14–20
(2021)
118. Guo, J., Wang, Q., Li, Y.: Evaluation-oriented façade defects detection using the rule-based
deep learning method. Autom. Constr. 131, 103910 (2021)
119. Dang, H.V., Raza, M., Nguyen, T.V., Bui-Tien, T., Nguyen, H.X.: Deep learning-based detec-
tion of structural damage using time-series data. Struct. Infrastruct. Eng. 17(11), 1474–1493
(2021)
120. Derogar, S., Ince, C., Yatbaz, H. Y., & Ever, E. (2022). Prediction of punching shear strength of
slab-column connections: A comprehensive evaluation of machine learning and deep learning
based approaches. Mechanics of Advanced Materials and Structures, 1–19.
121. Bolandi, H., Li, X., Salem, T., Boddeti, V. N., & Lajnef, N. (2022). Bridging finite element
and deep learning: High-resolution stress distribution prediction in structural components.
Frontiers of Structural and Civil Engineering, 1–13.
122. Bai, Y., Zha, B., Sezen, H., Yilmaz, A.: Engineering deep learning methods on automatic
detection of damage in infrastructure due to extreme events. Struct. Health Monit. 22(1),
338–352 (2023)
Deep Learning-Based Framework
for Reconstruction and Optimisation
of Building Information
Models Containing Parametric Rules

Vincent J. L. Gan

Abstract The building information model (BIM) is maturing as a new paradigm

for storing information and exchanging knowledge about a building. BIM models
precisely show existing structural elements which can be conveniently extracted to
support condition monitoring, quality assessment and design optimisation of engi-
neering structures. However, current practice does not fully address the integration
of 3D modelling and computational intelligence to support the automatic reconstruc-
tion and optimisation of BIM models. To resolve this challenge, this paper presents
a framework for deep learning-based reconstruction and optimisation of BIM, with
a case study to demonstrate its application for the parametric design optimisation of
high-rise buildings. The proposed framework involves the development and appli-
cation of deep neural network for automated reconstruction of content-rich building
models subject to certain design rules. After obtaining the geometric details about
a building, the performance of the generated 3D models is assessed towards identi-
fying the optimum solution. In this study, attempts have been made to optimise the
wind flow of a high-rise residential building, as the wind plays an important role in
designing a high-rise. The findings provide a deeper understanding and interesting
insights into 3D reconstruction and optimisation of BIM subject to parametric design
rules.

Keywords Building information model · Machine learning · Geometric

modelling · Computational optimisation · High-rise building

V. J. L. Gan (B)
Department of the Built Environment, National University of Singapore, Singapore 117566,
Singapore
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 289
G. Bekdaş and S. M. Nigdeli (eds.), Hybrid Metaheuristics in Structural Engineering,
Studies in Systems, Decision and Control 480,
https://doi.org/10.1007/978-3-031-34728-3_14
290 V. J. L. Gan

1 Introduction

Due to rapid urbanisation, the design and construction of new buildings have reached
an unprecedented level in densely populated metropolises. Buildings nowadays have
more complex built forms [1], and understanding the impact of time-varying ambient
environment helps optimise building geometries and provide positive attributes to
maintain the buildings. For example, adaptive designs that can utilise naturally-
occurring wind force to create a better-built environment increasingly interests many
researchers [2]. In light of this, evolutionary algorithms or generative modelling
techniques were leveraged to generate different spatial-geometric configurations,
whereas a simulation-based approach served to evaluate the fitness of each candi-
date solution [3, 4]. The research suggests a persistent need to develop new strate-
gies to accurately reconstruct and optimise the building geometries to improve the
built environment. Eleftheriadis, Duffour [5] and Choi, Lee [6] proposed multi-
objective computational optimisation models to forecast and optimise the building’s
structural layout and element details to lower cost and carbon footprint. Instead
of generating a single optimum, the ultimate target of multi-objective optimisa-
tion was to search for a Pareto Front, which contains a set of non-dominated solu-
tions where multiple objective functions are balanced and equally good [7]. The
new optimum searching method, including Multi-Objective Particle Swarm Opti-
misation (MOPSO) [8], pattern search PSO with Hooke-Jeeves algorithms [9], etc.,
were proposed to support the exploration of the Pareto Front. However, traditional
heuristic-based optimisation methods face challenges which are gaining attention as
the industry is adopting artificial intelligence in computational optimisation.
Emerging digital technologies (such as computational intelligence) are pervasive
across the virtual design and construction of new facilities. Building information
modelling (BIM) is a cutting-edge technology that allows the digital representation
of physical assets in virtual simulation models [10]. It enables the virtual recon-
struction of the detailed building geometry, followed by assessing and optimising
buildings in the physical world to meet clients’ requirements [11]. Based upon this
digital framework, BIM has been increasingly used to digitally represent and improve
the performance of modern buildings [12]. Figure 1 displays the process taking laser
scanning to collect point clouds of real-world structures, followed by using compu-
tational intelligence to process the point cloud for automated reconstruction of BIM.
The 3D digital model accurately represents the as-built conditions, allowing archi-
tects, engineers and construction professionals to depict building conditions and plan
more effectively for future projects. Researchers [13, 14] critically reviewed state-
of-the-art approaches with BIM for building optimisation. Among these, Liu, Meng
[15] presented a design optimisation method with a BIM environment to extract the
project-based initial information to facilitate the evaluation of design candidates. The
proven design can be put into simulating different design strategies and realising the
optimal solution. Ramaji and Memari [16] presented an information interoperability
standard to underpin BIM applications in the design and construction of residential
buildings.
Deep Learning-Based Framework for Reconstruction and Optimisation … 291

Fig. 1 Automated 3D reconstruction of as-built BIM

With the adoption of BIM, robust simulation-based optimisation methods were

proposed to forecast and optimise the building performance. Based upon parame-
terisation, it is paramount to establish ways to generate BIM models for integrated
simulation and performance assessment for different candidate designs. For instance,
a study [17] on coupling BIM and genetic algorithms (GA) to optimise modular high-
rise housing was performed. The design variables, including building shape, struc-
tural form, and member size, were parameterised to generate BIM models [18, 19]
automatically. The structural components were defined as a set of numerical parame-
ters which manipulate the building geometry [18]. Integrating predefined parametric
relationships and design rules among structural components can be used for recon-
structing and optimising the BIM model [19]. To save computing resources, numer-
ical simulation was undertaken to evaluate the performance of alternative solutions
[20]. For example, Gan et al. [18] and Solnosky, Memari [21] proposed a structural
BIM optimisation, with the aid of finite element analysis and hybrid metaheuristic
algorithms (e.g. optimality criteria GA), to explore, simulate and identify the optimal
design solution.
BIM was leveraged in the construction planning of buildings, including cost
estimation [22], site logistics planning and control [23], and offsite manufac-
turing [24, 25]. Design BIM models provide geometrical and semantics such as
resources and manufacturers, which empower the precise prefabrication of building
components. However, these new approaches still show disadvantages in compu-
tational time and optimality of final solutions because of potential iterations of
simulation-based assessment in the optimisation process. Yet, possibilities to incor-
porate BIM and computational intelligence, especially for reconstructing and opti-
mising building design subject to wind effects, are largely unexplored. In addition,
buildings constructed nowadays have complex shapes and geometry. Understanding
the impact of building geometries and the surrounding environment (such as wind) is
essential yet lacking; therefore, a comprehensive study on deep learning-based BIM
reconstruction and optimisation is still much needed.
This paper presents a deep learning-assisted framework to underpin the recon-
struction and optimisation of BIM subject to specific parametric design rules, with
a case study to demonstrate its application for high-rise building optimisation. The
proposed new framework involves geometric modelling and 3D reconstruction of
structural elements in the design space. After obtaining the geometric details about a
building, a deep learning algorithm is developed and deployed to assess the 3D model
292 V. J. L. Gan

and its resulting performance. In this research, attempts have been made to opti-
mise the geometry and layour high-rise modular buildings. The suggested approach
combines computational simulation with neural network computing to quantitatively
analyse the wind flow of high-rise buildings. To examine the flow characteristics of
high-rise buildings in different forms, numerical simulation is used to create the
training dataset for a strong neural network model in accurate forecasting and opti-
misation. Details about the methdology and results are provided in the following
sub-sections.

2 Methodology

This section articulates the deep learning-assisted framework for reconstruction and
computational optimisation of BIM. BIM is considered a single information hub
containing all the necessary spatial, geometric, and material information extracted
to optimise building performance. Provided BIM 3D objects, the architect assem-
bles customised floor plans using a parametric object library containing different
standard units. It is advised to undergo an optimisation process for the distribu-
tion of the BIM objects to optimise the building’s overall performance subject to
certain geometrical constraints. Building design optimisation is implemented, which
involves establishing the optimal architectural layout plan for optimising the wind
environment, as it plays an essential role in constructing and optimising building
models. To save the computing resource, deep learning was leveraged in this study
for automatic performance assessment of airflow patterns associated with building
spatial-geometric features. A deep neural network (DNN) model was developed to
learn from different outdoor flow conditions and building spatial-geometric features
to interpolate the complex relationship between building physical characteristics
and airflow. We will accomplish this via supervised training, which iteratively tunes
the deep learning algorithms between data-driven prediction and high-resolution
physics-based computation.
In general, this study contributes to a new method by synthesising BIM with deep
learning to optimise the building performance subject to wind effects. A diverse set
of realistic designs was created first with BIM, followed by DNN prediction with
its supervised learning capacity on the problem structure to identify the optimum.
The proposed new approach will leverage the strength of deep learning and BIM
to enhance the automatic generation and optimisation of 3D building models. The
optimisation methods are described in the following sections.

2.1 Optimal Design Formulation

As shown in Fig. 2, the developed approach couples automated computer-generated

design and machine learning data-driven prediction to iteratively analyse, predict
Deep Learning-Based Framework for Reconstruction and Optimisation … 293

Fig. 2 Computer-generated parametric BIM model

and optimise the design performance in the physical world. The proposed method
starts by creating design BIM models using metaheuristic algorithms, which generate
building layout plans containing the distribution of BIM objects, building geometry,
and site constraints. This is followed by evaluating the performance of the developed
models using a deep neural network (DNN) to identify the optimal design option.
As there is a growing academic interest in improving building design, the DNN in
the study considers the wind effects on buildings.
To illustrate the proposed approach for automatic generation and optimisation
of BIM, this study considers building structures with standardised units, such as
residential units, which usually could meet the structural performance requirements
while giving a higher degree of design freedom for architectural layout planning.
As such, this study considers typical buildings composed of modular/standardised
units and space dimensions from a parametric object library. A structural system
with one or several concrete cores is used, in which the core provides overall
stability, whereas the modular/standardised units are clustered around the core(s).
The modular/standardised units are subject to compression force in these structural
systems.
Since proper design formulation is crucial to the quality of the final design, the
proposed method first identifies the parametric design variables and constraints.
The computer-generated design of buildings includes defining one or more spatial
variables to characterise the layout plan for the entire building. The focus of the design
problem is on seeking the optimal layout configuration to maximise the building
performance subject to regulatory and code-stipulated performance requirements, as
follows:
⎡ ⎤ϕ
∑I ∑ J ∑K ∑ J
τ =⎣ Ck ⎦
j j
Zi + wher e Z i → (m i , n i , ai , bi , θi ) (1)
i=1 j=1 k=1 j=1
294 V. J. L. Gan

j
Z i ∈ A(SC) i = 1, 2, . . . I ; j = 1, 2, . . . J. (2)

( L )
∑
A Fwl ≤ Ca (3)
l=1
mi
L
Car < ≤ Car
H
(4)
ni
∑ j j
If Z i ∩ Ck , D(Z i , Ck ) ≤ Cin_d j = 1, 2, . . . J. (5)

in which τ refers to a randomly generated layout plan consisting of multiple stan-

dardised modular units; i refers to a part of the building (such as a wing); j stands for
a particular modular unit; Z i j stands for a specific type of modular unit j in building
part i. C k refers to a certain type of connecting area k (such as an elevator lobby or a
staircase), and ϕ represents the orientation of the entire layout (with North being zero
degrees). As such, Eq. (1) describes the layout as the whole plan with a total number
of parts J and common area K at orientation ϕ. Each part i has a set of geometric
properties including the width mi , wing length ni , origin ai and bi (at the end of one
side of the corridor), and azimuth θ i (along the West–East direction). The size and
position of a wing w can vary by changing the width (mi ), length (ni ), origin (ai and
bi ) and rotation angle (θ i ). On the other hand, the geometric properties of wing w
are used to determine the exact types and positions of different apartments in a later
stage. Equation (2) limits all modular apartments in the layout plan fitting within
the site boundary A(SC), whereas Eqs. (3) and (4) define precisely the geometrical
constraints. C a refers to the maximum allowable Area for each part of the building,
whereas C ar L and C ar H stand for acceptable aspect ratios. Equation (5) represents
the code-stipulated accessibility constraint, in which C in_d is the maximum permis-
sible distance from a randomly selected modular unit to its closest connecting part
C k . The aforementioned mathematical formulation is used to guide the computer-
generated design process in generating the various layout configurations subjected
to specific design considerations. Upon establishing the explicit expression for the
design constraints, a DNN is used to evaluate the design performance progressively.

2.2 Building Behaviour Forecasting with Deep Neural Net

This section will present the proposed deep neural network (DNN) for forecasting
building performance concerning wind effects. First, we generate BIM models of
several typical high-rise buildings. Next, we use computational fluid dynamics (CFD)
to simulate wind flow in each apartment within the buildings. The data obtained
from both BIM and CFD simulations is then used as a training source for the DNN
model. We describe the apartment of a building floor plan using several criteria
assigned as input features for the DNN. The output for the DNN consists of the wind
Deep Learning-Based Framework for Reconstruction and Optimisation … 295

flow and air change per hour (ACH) for each apartment. By training the DNN with
abundant labelled data on ACH and varying apartment arrangements, we can predict
the wind flow and assess the building’s performance regarding wind effects. Different
model architectures and parameters are tested to determine the most promising DNN
configurations for prediction and assessment.
Due of its accessibility and current popularity in data research, a literate program-
ming tool is employed to construct the DNN model for regression analysis [26].
The DNN model is constructed using the Keras Application Programming Interface
(API) since it enables sequential definition of the neural network’s layers, activation
functions, and hyperparameter settings [27]. The training is carried out on a 2.20 GHz
Intel Core i5-5200U CPU, which takes 10 min to complete 500 epochs.
Dataset: Three sets of data are available for public housing, each consisting of
standard block shapes. In total, 457 samples are collected. Table 1 summarises the
amount of samples: the Cruciform shape has 153 samples, the T-shape has 144, and
the Y-shape has 160. Each plane shape has two wind directions such as 0° and 45°
for the Cruciform, 0° and 180° for the T-shape, and 0° and 180° for the Y-shape.
Four-floor levels were chosen (5th, 15th, 25th, and 35th floors), each comprising four
apartments: Studio Apartment A, Studio Apartment B, One Bedroom Apartment, and
Two Bedrooms Apartment. The dataset has consistent wind environment settings,
which serve as ground truth for the training. The dataset is split, with 70% used
for training, and the remaining points are allocated for cross-validation (15%) and
testing (15%).
Input features: In this case, the input features are derived from data gathered by the
CFD simulation regarding the ACH of each apartment and the building floor plan.
Feature selection is performed to make the data useful for the machine learning. The
selected features are used to depict the spatial distribution of apartments according to
their distances and orientation about the core of the building. Table 2 summarises the
input features used in this analysis, which are obtained from parametric modelling

Table 1 Training dataset used in this analysis

Shape Cruciform-shape T-shape Y-shape
Number of data 153 144 160
points
Prevailing wind 0 and 45 0 and 180 0 and 180
Floor Level 5th, 15th, 25th, 35th 5th, 15th, 25th, 35th 5th, 15th, 25th, 35th
Floor plan
296 V. J. L. Gan

Table 2 Input features used

Input features Type Options
in this analysis
Distance from the Continuous mm
building core (X1)
Building orientation Continuous ◦

(X2)
Apartment size (X3) Continuous Studio apartment A
(14.5 m2 )
Studio apartment B (22
m2 )
One bedroom
apartment (30.9 m2 )
Two bedroom
apartment (35.9 m2 )
Floor-to-floor height Continuous 5th, 15th, 25th, 35th
(X4)
Apartment Discrete 0 or 1
position—L(X5)
Apartment Discrete 0 or 1
position—R (X6)

to describe the position of the apartment. To make predictions more robust, one-
hot encoding is implemented for discrete variables. The features are then stored
in a tabular form (CSA) and normalised within the range of 0 to 1 by using the
Min–Max Normalisation to avoid the over-dominance of any unique feature. Such
pre-processing is critical to facilitating the convergence of gradient descent [28].

X i − min(X )
X i' = (6)
max(X ) − min(X )
'
where X i is the original value and X i is the normalised value.
DNN model architecture: The architecture of the DNN model has been optimised
using exhaustive searching to minimise the mean squared error (MSE) for enhanced
training performance and maximise the R2 score. The model shown in Fig. 3 consists
of four hidden layers, each comprised of 50, 80, 80, and 30 neurons. The number of
neurons as well as hyperparameters such as batch size and learning rate, have been
optimised using exhaustive searching. To prevent overfitting, batch normalisation is
performed prior to applying non-linear activation to both hidden and output layers
[29]. The rectified linear unit (ReLU) is used for the hidden layers. To capture the
non-linearity, the sigmoid function is selected for the output layer. A batch size of 15
has been shown to be the most effective for resolving the ACH prediction through
trial and error. MSE in Eq. (2) and R2 in Eqs. (3)-(5) are used to gauge how well the
DNN models perform.
Deep Learning-Based Framework for Reconstruction and Optimisation … 297

Fig. 3 Architecture of the DNN model

1∑
n
MSE = (yi − ŷi )2 (7)
n i=1

SSRegression
R2 = 1 − (8)
SSTotal
∑
n
( )2
SSRegression = yi − yRegression (9)
i=1

∑
n
SSTotal = (yi − y)2 (10)
i=1

in which n stands for the data points, yi and ŷi refer to the ground truth and prediction,
SSRegression and SSTotal are squared errors, yRegression represents the value of regression
prediction, and y refers to the mean value.
Six different optimisers are tested to determine the best DNN models’ perfor-
mance. The six optimisers, SGD, RMSprop, Adam, Nadam, Adagrad, and Adadelta,
all have unique characteristics and default learning rates. The model is evaluated
using three separate shapes (cruciform, T, and Y) and then combined into an overall
model. The final result of the experiment is the best hyperparameter settings for the
DNN model, as determined by the optimiser tests.
298 V. J. L. Gan

3 Results and Discussion

3.1 Performance Assessment Using Deep Learning

This section presents the Cruciform-, T-, and Y-shapes’ respective optimiser R2
scores and MSE comparisons. Below are described the specific R2 score, MSE, and
combined scenario for all forms.
Cruciform-shape: The findings in Fig. 4 show that RMSprop optimiser got the best
R2 of 0.769 and 0.681 for testing and entire datasets. Although after 500 epochs
Nadam, Adam, and RMSprop converge to comparable MSE values, it is noteworthy
that RMSprop surpasses the others in terms of the R2 score for this specific dataset.
RMSprop has the edge over the other optimisers when it comes to the R2 score on
this specific dataset.
T-shape: Adadelta optimiser is the best option, according to Fig. 5, since it achieves
the greatest R2 scores of 0.895 for the testing set and 0.917 for the entire dataset. R2
indicates that this model can make reliable predictions on unobserved data and is not
restricted to the training dataset. After 500 epochs, RMSprop, Adam, and Adadelta
all show a similar convergence. Their R2 ratings for the entire dataset are therefore
comparable.
Y-shape: With an R2 value of 0.774 and 0.896 for testing and overall datasets, the
Adadelta optimiser produces the best results, as shown in Fig. 6. Although the
Adadelta optimiser performs similarly to RMSprop, the learning curve often follows
the same pattern as the T-shape model.

Fig. 4 MSE comparison for Cruciform-shape

Deep Learning-Based Framework for Reconstruction and Optimisation … 299

Fig. 5 MSE comparison for T-shape

Fig. 6 MSE comparison for Y-shape

Combined model for three shapes: The Adam optimiser generates the highest R2
score, 0.817 for testing and 0.666 for the entire dataset. This model performs substan-
tially poorer in predicting unknown data on the testing dataset than the previous
three models, which individually assessed the data for each plane shape. This is
300 V. J. L. Gan

because each building layout has a different aerodynamic property. To avoid over-
fitting, capturing all the aerodynamic correlations in the three datasets necessitates
both a more complex architectural design and additional datasets.
In general, Adam outperforms all three geometric layout datasets. It may be
inferred that the reason the aforementioned optimisers outperform others, such as
SGD, despite a problem-based black-box model. These optimisers mostly fall into
the categories of the constant learning and adaptive learning rate. The remaining
ones are of the second kind, which, after a certain batch size, modifies the learning
rate. The disadvantage of constant learning rate is that it might go below the global
minimum. Adagrad belongs to the second group, but because it is usually used to
situations with sparse gradients, it performs the worst when compared to other opti-
misers. The multivariate non-linear regression does not provide a satisfactory fit as
a consequence.

3.2 Evaluation of Alternative Designs

DNN models can forecast the ACH of each apartment for a specific geometric layout
after the neural network development. To find the best floor plan with the highest
ACH, many preliminary floor plan concepts can be made and assessed. The T-shape
model was the neural network that was selected for the demonstration since it had
the greatest R2 score and the best prediction accuracy. Three design plans were made
with one wind direction in mind in addition to the initial setup for model training.
The four different apartment arrangements of a T-shaped building design are shown
in Fig. 7 and are grouped according to their apartment sizes. The same number of
apartments listed in Table 3 were used in each scenario.

3.3 Generation and Optimisation of BIM Models

Table 4 and Fig. 8 offer a thorough breakdown of ACH for each apartment. Optimising
high-rise building wind flow is the ultimate goal. As a result, the assessment is based
on the following standards: (1) the highest ACH; (2) the least amount of variation
across the four various apartment types based on the standard deviation. The example
with the greatest overall average ACH value, as shown in Table 4, is case C, in
which all two-bedroom apartments are allocated on the windward side, and all studio
apartments are spread on the leeward side. Despite the fact that the four distinct types
of apartments have significantly diverse airflow distributions overall, studio B and
the two bedroom apartment stand out.
Cases A and D have a less deviation for ACH, indicating that they are more
uniformly distributed than Cases B and C. Due of a superior wind flow, case A is
more effective than case D. The one bedroom apartment in case A, is also more
ventilated than the one in scenario C. This design benefits more inhabitants as a
Deep Learning-Based Framework for Reconstruction and Optimisation … 301

Fig. 7 Apartment layout for a T-shaped floor plan

Table 3 Configuration of the

Apartment Type Area (m2 ) Amount
layout plan for evaluation
Studio apartment A 14.5 3
Studio apartment B 22 4
One bedroom apartment 20.9 8
Two bedrooms apartment 35.9 3

consequence. Case A is the one that should be used in terms of the assessment
criteria.
Two trends related to the ACH may be seen in addition to comparisons between
various apartment configurations. Better wind flow is achieved by placing apartments
on the windward side and farther from the centre. As a result, it is advised to allocate
apartments to these spots. Based on these findings, it is inferred that it is advantageous
for the air change rate, and therefore the natural wind flow, to situate larger-sized
302 V. J. L. Gan

Table 4 ACH value for different apartments

Case Apartment type Number ACH Average ACH Standard Deviation
Case A 2 bedrooms apartment 3 10.77 16.97 4.97
1 bedroom apartment 8 18.11
Studio apartment A 3 11.95
Studio apartment B 4 23.11
Case B 2 bedrooms apartment 3 34.43 12.56 11.55
1 bedroom apartment 8 7.99
Studio apartment A 3 5.68
Studio apartment B 4 10.44
Case C 2 bedrooms apartment 3 9.21 17.58 11.92
1 bedroom apartment 8 11.06
Studio apartment A 3 15.01
Studio apartment B 4 38.85
Case D 2 bedrooms apartment 3 14.51 15.30 4.22
1 bedroom apartment 8 11.84
Studio apartment A 3 14.93
Studio apartment B 4 23.12

apartments on the windward side. From this vantage point alone, case C demonstrated
the greatest overall ACH of 17.58, with 16.97 using case A. A broad range of values
was found when the average ACH for each apartment type was calculated. Case
A did not exhibit the imbalanced distribution of air change rate, although ranking
second in average ACH. Thus, when it came to ACH distribution, case A won out
over case C even if case C produced the most ACH.

4 Conclusions

This study aims to develop an innovative deep learning-based framework for the
automatic reconstruction and optimisation of BIM. This research focuses on the
realisation of the whole process of performance assessment and optimisation of
buildings combined with BIM and deep learning algorithms. To demonstrate the
suggested strategy, the ACH optimisation models for complex-shaped residential
buildings are used. The T-shaped model is used to examine many alternative apart-
ment designs and choose the best arrangement. It is advised to analyse additional
geometric arrangements like cruciform and Y-shape for these datasets using a more
effective DNN model. This paper presents a case study about how to use BIM and
deep learning to underpin the computer-aided design of high-rise buildings. A new
computer-generated design method is discussed regarding its capacity to generate
the optimal layout plan using modular units. Given the layout plan, this study further
Deep Learning-Based Framework for Reconstruction and Optimisation … 303

Fig. 8 Evaluation of different T-shape models

investigates the possibility of deep learning algorithms to facilitate automated design

optimisation.
This study contributes to a new body of knowledge and is the foundation to drive
future research into the automatic generation of BIM using computational design.
Firstly, this study delivers a framework which combines the strength of BIM with
data-driven deep learning algorithms to improve the reconstruction and optimisa-
tion of BIM models. Secondly, this study explores a DNN model that can analyse
the latent spatial-geometric features of modular buildings and perform reasoning to
understand the relationship between design variations and indoor airflow patterns.
Because the mechanism between building morphology and indoor airflow has been
a long-standing problem, the proposed new method will establish a way to analyse
and optimise building performance to drive future research in related fields. Future
work of this study includes the structural integrity of buildings, which requires
more attention to the practical installation using appropriate connections between
modules and the proper design strategy of structural systems to enhance the overall
304 V. J. L. Gan

stability and robustness. Besides, as building systems are getting complex, there is
a growing interest in the application of novel AI or new IT technologies (such as
robotics, computer vision) to empower the 3D BIM reconstruction and optimisa-
tion by systematically integrating the design, construction and long-term facilities
operation.

References

1. Pan, W., Qin, H., Zhao, Y.: Challenges for energy and carbon modeling of high-rise buildings:
The case of public housing in Hong Kong. Resour., Conserv. Recycl. 123(Supplement C),
208–218 (2017)
2. Gan, V.J.L., et al.: Physics-based, data-driven approach for predicting natural ventilation of
residential high-rise buildings. Build. Simul. (2021)
3. Tafraout, S., et al.: Automatic structural design of RC wall-slab buildings using a genetic
algorithm with application in BIM environment. Autom. Constr. 106, 102901 (2019)
4. Mangal, M., et al.: Automated clash-free optimisation of steel reinforcement in RC frame struc-
tures using building information modeling and two-stage genetic algorithm. Autom. Constr.
126, 103676 (2021)
5. Eleftheriadis, S., et al.: Investigating relationships between cost and CO2 emissions in rein-
forced concrete structures using a BIM-based design optimisation approach. Energy Build.
166, 330–346 (2018)
6. Choi, J., et al.: Multi-objective green design model to mitigate environmental impact of
construction of mega columns for super-tall buildings. Sci. Total Environ. 674, 580–591 (2019)
7. Sandberg, M., et al.: Multidisciplinary optimisation of life-cycle energy and cost using a BIM-
based master model. Sustainability (Switzerland) 11(2) (2019.)
8. Delgarm, N., et al.: Multi-objective optimisation of the building energy performance: a
simulation-based approach by means of particle swarm optimisation (PSO). Appl. Energy
170, 293–303 (2016)
9. Kim, J., et al.: An optimisation model for selecting the optimal green systems by considering
the thermal comfort and energy consumption. Appl. Energy 169, 682–695 (2016)
10. Hamidavi, T., Abrishami, S., Hosseini, M.R.: Towards intelligent structural design of buildings:
a BIM-based solution. J. Build. Eng. 32, 101685 (2020)
11. Kaewunruen, S., Rungskunroch, P., Welsh, J.: A digital-twin evaluation of net zero energy
building for existing buildings. Sustainability 11(1), 159 (2019)
12. Zhang, J., et al.: BIM-enabled modular and industrialised construction in China. Procedia Eng.
145, 1456–1461 (2016)
13. Gan, V.J., et al.: Simulation optimisation towards energy efficient green buildings: current status
and future trends. J. Clean. Prod., 120012 (2020a)
14. Afzal, M., et al.: Reinforced concrete structural design optimisation: a critical review. J. Clean.
Prod., 120623 (2020)
15. Liu, S., Meng, X., Tam, C.: Building information modeling based building design optimisation
for sustainability. Energy Build. 105, 139–153 (2015)
16. Ramaji, I.J., Memari, A.M.: Information exchange standardisation for BIM application to
multi-story modular residential buildings. In: AEI 2015, pp. 13–24 (2015)
17. Gan, V.J., et al., Simulation-based evolutionary optimisation for energy-efficient layout plan
design of high-rise residential buildings. Journal of Cleaner Production, 2019c.
18. Gan, V.J.L., et al.: Parametric modelling and evolutionary optimisation for cost-optimal and
low-carbon design of high-rise reinforced concrete buildings. Adv. Eng. Inform. 42, 100962
(2019)
Deep Learning-Based Framework for Reconstruction and Optimisation … 305

19. Singh, M.M., Sawhney, A., Borrmann, A.: Modular coordination and BIM: development of
rule based smart building components. Procedia Eng. 123, 519–527 (2015)
20. Fahmy, M., Mahdy, M.M., Nikolopoulou, M.: Prediction of future energy consumption reduc-
tion using GRC envelope optimisation for residential buildings in Egypt. Energy Build. 70,
186–193 (2014)
21. Solnosky, R.L., Memari, A.M., Ramaji, I.J.: Structural BIM processes for modular multi-story
buildings in design and construction. In: 2nd Residential Building Design & Construction
Conference, pp. 201–215 (2014)
22. Barkokebas, B., et al.: Coordination of cost estimation for industrialised residential projects
through the use of BIM. In: Proceedings of the International Symposium on Automation and
Robotics in Construction (ISARC), p. 1. IAARC Publications (2015)
23. Bortolini, R., Formoso, C.T., Viana, D.D.: Site logistics planning and control for engineer-to-
order prefabricated building systems using BIM 4D modeling. Autom. Constr. 98, 248–264
(2019)
24. Abanda, F.H., Tah, J.H.M., Cheung, F.K.T.: BIM in off-site manufacturing for buildings. J.
Build. Eng. 14, 89–102 (2017)
25. Alwisy, A., Al-Hussein, M., Al-Jibouri,S.: BIM approach for automated drafting and design
for modular construction manufacturing. Comput. Civ. Eng. 2012, 221–228 (2012)
26. Kery, M.B., et al.: The story in the notebook: Exploratory data science using a literate program-
ming tool. In: Proceedings of the 2018 CHI Conference on Human Factors in Computing
Systems (2018)
27. Ketkar, N.: Introduction to keras. In: Deep Learning with Python, pp. 97–111. Springer (2017)
28. Aksoy, S., Haralick, R.M.: Feature normalisation and likelihood-based similarity measures for
image retrieval. Pattern Recogn. Lett. 22(5), 563–582 (2001)
29. Ioffe, S., Szegedy, C.: Batch normalisation: Accelerating deep network training by reducing
internal covariate shift. In: International conference on machine learning. PMLR (2015)