International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 706

ISSN 2229-5518

A review of structural size optimization

techniques applied in the engineering design.
ErsilioTushaj, Niko Lako
Abstract -The efficient solution that satisfies optimality condition is an important issue when analyzing the structural engineering design
problem. The new codes of structural design consist in methodologies that demand the total exploitation of the resources of the
construction materials. The goal of an optimal design is to achieve the best feasible solution according to a measure of effectiveness. This
demand, emphasizes the need for lower weights and total costs of the realization of the structure. The first analytical work in structural
optimization was by Maxwell in 1890, followed by Mitchell in 1904. The best applications during the years 30’-50’s were in the aircraft
industry, consisting of compressive loads and buckling constraints. Good developments followed with Schmit in 1960. He was the first to
offer a comprehensible analytical statement of the problem. Up to now have been published a lot of algorithms, which have similar
approaches but offer different results. It is difficult to identify which one is appropriate in solving a specific structural design problem. For
this purpose, more studies, surveys and analysis are necessary to give developments, advantages and disadvantages in the application of
these algorithms. The goal of this paper is to analyze and follow an up to date study of size optimization techniques used in the structural
optimization design, priory in steel constructions, basing on some highly peer reviewed studies, that were possible to be analyzed by the
author.

Index Terms - review, structural optimization, size optimization, engineering design.

————————————————————

1. INTRODUCTION

T he design process in structural optimization as announced iterations to guarantee the optimal solution. Efficiency and

IJSER
by professor Uri Kirsch [1] in 80’s, can be classified into robustness of the algorithms are the two bigger goals of re-
four stages: formulation of functional requirements, the searchers. [3]. Structural optimization is mostly not taught in
conceptual design stage, optimization and detailing. Iterative the bachelor or master programs for civil engineers, but it is
procedures or the application of algorithms are necessary considered as a complementary course in some PhD pro-
before the final solution is achieved. grams; such as the course followed by professor Bontiempi of
Typically, an optimal design problem consist of three issues: structural analysis at the University “La Sapienza”, in Rome.
(a) objective function, (b) design variables, and (c) constraints. [4]
Usually in the optimization problem the objective function
used is the total weight or the cost of the structure. The design 1.1. First steps in structural optimization.
variables are those parameters to be determined by the de- The first analytical works on structural optimization can be
signer in order to generate an optimal solution. In practical dated back to 1890 by Maxwell, followed by other studies of
applications achieving an optimum design should be carried Mitchell in 1904. The structural optimization techniques were
out with respect to a set of strength, stability and serviceability next developed in the aircraft and space industry, as a conse-
limitations. The classification of the structural optimization quence of restrictive requirements on the problem of the min-
problems related to the type of design variables involved, is imum weight design problem in the engineering design. Dur-
divided into three main categories: sizing, shape, and topolo- ing the years 30-50’, the availability of computers, made possi-
gy optimization. In sizing optimization the cross sectional ble the application of linear programming techniques to the
areas of the structural members are considered as design va- plastic design of frames.
riables. The process of sizing can be continuous or discrete. In
a continuous sizing optimization any positive value can be 1.2. Structural optimization after the 60’s.
assigned to the cross sectional areas of the [Link] practic- Lucien A. Schmit[5], was the first to offer a good statement on
al cases the structural members should be adopted from a set the use of mathematical programming techniques in solving
of available sections, so the design problem turns into a dis- the nonlinear inequality constrained problem of designing
crete sizing optimization. [2] elastic structures, when applying a multiplicity of loading
More efficient and robust algorithms are necessary to respond conditions. This work introduced a new philosophy of engi-
to the actual demand of design. Optimization problems may neering design which only in the 80’s began to be broadly ap-
fall into the trap of local minima, or may require too much plied. It indicated the feasibility of coupling finite element
structural analysis and nonlinear mathematical programming
————————————————
to create automated optimum design capabilities. In the 60’s,
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
the computational experience indicated that mathematical
University of Tirana in Albania. He currently teaches at POLIS University,
Tirana in [Link]@[Link]. programming (MP) techniques, applied to structural design,
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil were limited to only a few of dozen of design variables. At the
Engineering at the Polytechnic University of Tirana. He has a long expe- beginning the applications were limited to relatively small
rience in teaching and is author of various pubblications. nlako@[Link]
structures. In the late 1960’s an alternative approach, called
Optimality Criteria (OC), was presented in analytical form.
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 707
ISSN 2229-5518
The (OC) was largely intuitive and it was shown to be a very conceived as the main shortcomings of these techniques. A
effective design tool. It was independent of the problem size possible solution is by taking advantage of computing me-
and usually provided a near optimum design with a few thods, such as parallel or distributed methods of workload
structural analyses. This feature represented a remarkable im- amongst multiprocessors, which are connected to each other.
provement over the number of analysis required for (MP) me-
thods to reach the solution. The (OC) optimality criteria me-
2. GENERALS IN OPTIMIZATION TECHNIQUES.
thod was presented in two different approaches, as a physical
or a mathematical tool. The physical problem was based on 2.1. A General Formulation of the Optimization
stress constraints (fully stressed design – FSD) or displacement Problem.
one (Fully displacement design – FDD). An integrate version The general nonlinear constrained optimization problem can
also came regarding both stresses and displacements con- be stated as followed by [8]:
straints in “Fully Utility Design” (FUD. The mathematical ap- 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑥𝑥
proach was based on the Kuhn-Tucker conditions. 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑓𝑓(𝑥𝑥)
𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑡𝑡𝑡𝑡 𝑔𝑔�𝑥𝑥� = 0 𝑎𝑎𝑎𝑎𝑎𝑎 ℎ�𝑥𝑥� ≥ 0
1.3. Advances in structural optimization. where: 𝑥𝑥 = (𝑥𝑥1 , 𝑥𝑥2 , … , 𝑥𝑥𝑛𝑛 ) is a set of variables, which can be
Other optimization techniques have emerged in the last dec- binary, discrete or continuous.
ades. These methods do not require gradient information for 𝑓𝑓(𝑥𝑥)is the objective function.
the objective and the constraint functions, but use probabilistic The goal of the optimization process is to minimize the value
transition rules rather than deterministic ones. These tech- of 𝑓𝑓 related to 𝑥𝑥.
niques are called stochastic or meta-heuristic approaches,
since they search the optimal solution generating random 2.2. Different levels of structural optimization.
populations, based on some criteria or fitness functions. The Many authors agree on the following classification about the
idea behind them is to simulate a natural phenomenon, such different levels of optimization:

IJSER
as survival of the fittest, the immune system, swarm intelli- - size optimization deals with minimization (or
gence or the cooling process of molten metal through anneal- maximization) of one or more response variables
ing. A detailed review of these algorithms as well as a compar- (such as stresses, deformation, stiffness) acting on one
ison of their performance for discrete sizing problems is pro- or more design variables;
vided by Hasancebi[6]. These heuristic optimization methods - shape optimization aims to find the optimal shape of
have some advantages when compared to the deterministic domain, which is no longer fixed and become a
one. They separate the domain knowledge from search, mak- design variable itself;
ing them generally applicable to a wide variety of problem - topology optimization for continuum structures deals
formulations; with no limitation on the continuity of the with the number, position, shape of holes and
search space, since no gradient information is required. topology of the domain.
The state of art of Kazemzadeh[7], makes a distinction of the
methods in traditional and modern one. The traditional me- 2.3. Trends in structural optimization.
thods include the mathematical programming (MP) tech- In the process of solving these problems, it is important to be
niques and the optimality criteria (OC). The modern methods attentive to the objective function, because the right selection
include the non-deterministic approaches. Some of the most of it can bring to better results. The common objective func-
recognized modern algorithms are: the genetic algorithms tions used, as documented by the literature, are the following:
(GA), harmony search method (HS), simulated annealing (SA), minimization of weight, maximization of stiffness, minimiza-
particle swarm optimization (PSO), ant colony optimization tion of cost design and any combination of the previous. The
(ACO), ant bee colony (ABC) etc. minimization of weight is the most used, and often there is an
The meta-heuristics also present some disadvantages. They implicit assumption that the weight of a structure is the best
require significantly more computational resources than the measure for evaluating its cost. But this is not true, since doing
deterministic one. Research on convergence has shown that this, there is a risk of neglecting other important terms, beside
the number of evaluations required to reach a given solution, material cost, that contribute to define the final value. A recent
grows as a function of the square root of the size of the prob- modern point of view by some critics, is that the objective
lem. The slow rate of convergence towards the optimum and function should be evaluated in terms of the life-cycle cost of
the need of the high number of structural analyses are still the structure, taking in account more terms, such as the costs
of materials, fabrication, erection, maintenance, disassembling
at the end of structure life etc. So it’s clear that this approach,
————————————————
is more realistic, but encounters additional difficulties. Unfor-
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
University of Tirana in Albania. He currently teaches at POLIS University, tunately, constraint evaluation in the real world, involves
Tirana in [Link]@[Link]. many sources of imprecision and approximations that reflects
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil into the final results.
Engineering at the Polytechnic University of Tirana. He has a long expe-
rience in teaching and is author of various pubblications. nlako@[Link] Some authors agree that during the architectural-engineering
design process, the optimal version should not be the first goal

IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 708
ISSN 2229-5518
of the design team. It is necessary to explore various possible much more powerful than the ones that existed before. The
optimal solutions, which can also offer better quality from an same thing cannot be said also for numerical methods optimi-
aesthetic point of view.[9] zation methods, which despite their advance, are limited in
the academic research or are exclusive to specialized compa-
2.4. The algorithmic approach. nies. Practical applications are rare. These are encountered
much more in mechanics, aeronautics and electronics.
Their rarity is firstly due to the fact that the subject is really
Gradient or fitness
complex, requiring a deep knowledge and solid background
Selection of the algorithm.
function.
in many topics of numerical methods and also structural me-
chanics. Secondly there does not exist a unique formulation of
an optimization method that can be successfully applied to a
really large class of problems. The creation of a “multipur-
Formulation of the Initial pose” software that could manage sufficiently a wide set of
Candidate design
Problem. Design structural problems has not been possible yet. Baldock[10] has
reported a short review of some real world applications of
structural optimization and softs that have been used for this
Optimum achieved?
purpose.
NO Algorithms are programmed in softs. Some of the most used
YES are BASIC, MATLAB, FORTRAN, C++ etc. Other ad hoc softs
Structural analysis criteria.
have been developed to respond to practical applications such
Iteration finished. as: TOSCA2 (FE-Design), GENESIS3, OptiStruct4, BIGDOT,
Optimal solution. ABAQUS, OPTIMA, SODA, and others. A review of these soft
has been done by Vanderplaats[11]. Abishek[12] has docu-

IJSER
mented MATLAB codes to be used in most known optimiza-
Fig. 1. A general flowchart of the optimization problem. tion problems.
The variables, objective function and the constraints represent
physical properties of the structure being optimized. The algo- 3. THE OPTIMIZATION ALGORITHMS.
rithms deal exclusively with the mathematical form of the This paragraph explains through a short introduction how the
problem interfaced with computer models representing the optimization algorithms have evolved. A review for the dis-
physical structure. The model is used to perform structural crete variable optimization is given by Arora[13]. A review of
analyses requested by the optimization algorithm. The interac- non-deterministic approaches with algorithmic steps was
tion scheme between the algorithm and the analysis tool de- done by Hare [14].
pends on the optimization method. Fig. 1shows a general
scheme of a gradient based algorithm. The algorithm gene- 3.1. Deterministic methods.
rates a design by assigning values to the optimization va- The complexity of applying structural optimization algorithm
riables. After being updated with new values, the structural to practical problems has always motivated the researchers to
model is used to perform analysis. The results are taken into develop more efficient and robust optimization techniques. A
account by the algorithm to generate new designs, as long as review of applications of deterministic methods for structural
the termination criteria are not met. Optimization time is re- optimization has been studied by Chain [15].
duced by choosing an algorithm that converges quickly to-
wards the optimal design. 3.1.1. Mathematical Programming (MP).
Mathematical programming techniques are among the most
2.5. Difficulties in practical applications. known classes of optimization techniques. The (MP) optimiza-
The computational methods have been successful in many tion problem is divided into: the unconstrained and the con-
fields of engineering in the recent decades. The finite element strained (MP) problem. Since material, strength and displace-
method, for example has proven to be of common use not only ments in structures are limited, consequently the problem is
in academic contexts but also between professional engineers. generally in the constrained form. The general nonlinear for-
Finite element applications are now essential tools for modern mulation of the constrained problem was given in 2.1. The
design. The advances in technology bring to us computers linear formulation, (LP) linear programming, can be solved
using graphical or simplex methods. The solution obtained is
———————————————— possible, only when a space with few variables is analyzed,
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic since calculations get difficult in correspondence with the
University of Tirana in Albania. He currently teaches at POLIS University, complexity of the problem [16]. Other solution methods have
Tirana in [Link]@[Link]. been developed, based on gradient analysis of the objective
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil
Engineering at the Polytechnic University of Tirana. He has a long expe- function. The basic idea is to move in the negative direction of
rience in teaching and is author of various pubblications. nlako@[Link] the gradient of the objective function to find a more promising
candidate design. Some of the gradient based methods are: the
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 709
ISSN 2229-5518
“Conjugate gradient method” by Fleetcher[17], the method of cluded some advantages of the (FSD). The (FSD) is more effi-
the feasible direction [18], the Sequential linear programming cient, it needs less iterations, and in general if offers better op-
by Kelley[19] etc. timal solutions, but is less effective with displacement con-
It is generally believed that (MP) techniques are not robust straints since it doesn’t consider them. The advantages of the
methods for optimum design of structural systems that ana- (SLP) with move limits are enhanced, when considering the
lyze numerous design variables. They can fall into the trap of right values of the move limits. (OC) algorithms have been
local minima if some conditions are not imposed in the right also applied in the topology optimization. Results have shown
way. difficulties in the first steps of applying the algorithm to the
problem, but greater effectiveness in obtaining results.[27]
Sequential Linear Programming (SLP) (1961)
The Sequential Linear Programming (SLP) with move limits, Fully Constrained Design (FCD) (2014)
introduced first by Stewart [20], has been a successful tool in In the last years the efficiency of traditional methods in find-
size optimization of steel truss structures by Stasa[21]. A pro- ing the optimal solutions has inspired the scientific communi-
gram called (OPT-KM) has been built to find the optimal solu- ty, from the University of Stanford, ETH Zurich and Phoenix
tion of the minimal weight with stress and displacement con- System to develop a new algorithm which can find the optim-
straints. The (OPT-KM), built in BASIC language, has been al solution through a similar process of the (OC). This algo-
compared to the (KM) algorithmic program using the (FSD) – rithm has been called the (FCD) “Fully Constrained Design”
“Fully Stressed Design” of the Optimality Criteria (OC). Ad- [28]. It has been proved to be efficient and robust, in compari-
vantages of the (SLP) are reported when approaching with son to other deterministic and non-deterministic algorithms.
different move limits. (See also 3.1.2).
3.2. Non-deterministic techniques.
Gradient Based Algorithm (2005) These techniques have been built based on other system phe-
In 2005, Gisbain[22] reported a gradient based algorithm, in nomena and follow search strategies in finding the optimum

IJSER
MATLAB environment that applied to the optimization of design. There is a rising popularity of these techniques as a
steel structures. The algorithm applies minimization of the consequence of the fact that they don’t require gradient infor-
stiffness matrix of the truss structure. The process combined mation, thus they are independent from the population under
size and topology optimization. analysis. They can deal with discrete and continuous design
variables and can handle better problems with a high number
3.1.2. Optimality Criteria (OC). of variables. They are not too difficult in coding, compared to
Another class of non-deterministic structural optimization deterministic methods. Some reviews on these methods have
techniques covers the optimality criteria (OC) methods. In been done in the last years by Kazemzadeh[7], Kicinger[29],
order to generate an optimum design, a recursive algorithm is Lamberti[30], Hare [14], and others.
employed to update the structural members for satisfying the Basically the aim of these methods is the location of the global
(OC). Early works on the algorithm are due to Prager[23] and optimum, by generation of candidate solutions in an iterative
Venkayya[24]. Then Bazaraa[25] realized the necessary opti- way. The fundamental idea is to seek the vicinity of more
mality criteria in Mathematical Programming in the presence promising candidate designs, found so far to drive the search
of differentiability. So it was possible to develop another ver- towards more reliable solutions. It starts with an initial popu-
sion of (OC), based on the Kuhn-Tucker conditions, and Vir- lation randomly selected design. Then the other populations
tual Work in searching the optimal solution [26]. are generated in base of a rule. In order to investigate their
Later, numerous variants of the optimality criteria methods quality, each candidate design is evaluated with respect to the
have been applied to optimum design of pin-jointed and objective function of the problem.
frame structures. It is worth mentioning that the fully stressed Once fitness of each candidate design is computed, new can-
design (FSD) can be also considered as a simple stress-ratio didates can be generated using the obtained information from
optimality criteria technique which can only deal with stress the formerly generated designs. The generation of a new pop-
constraints. An extension of the FSD that handles both stress ulation is guided by a mechanism or an operator and it is ite-
and displacement constraints is the fully utilized design ratively performed until a predefined termination criteria is
(FUD), which is capable of generating a feasible solution obtained. The last iteration is expected to be the optimal or an
through a small number of structural analyses. acceptable near optimal result.
Stasa[21], comparing the (OC) algorithm in the “fully stressed
design” version, with the (SLP) with move limits; has con- 3.2.1. Evolutionary algorithms (EA).
————————————————
These type of methods use a so called evolutionary computa-
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
tion (EC), which uses computational models of processes of
University of Tirana in Albania. He currently teaches at POLIS University, evolution and selection. These are mechanisms of Darwinian
Tirana in [Link]@[Link]. evolution and natural selection encoded in evolutionary algo-
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil
rithms. These algorithms are divided in: a. Evolution strategies
Engineering at the Polytechnic University of Tirana. He has a long expe-
rience in teaching and is author of various pubblications. nlako@[Link] (ES) (1965) by Rechenberg[31]; b. Evolution programming (EP)
(1966) by Fogel[32], c. Genetic Algorithm (GA) (1975) by Hol-
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 710
ISSN 2229-5518
land [33], and d. Genetic Programing (GP) (1992) by Koza[34]. The most known algorithm in this category is the Particle
swarm optimization (PSO).
Genetic Algorithms (GA) (1975)
The most known of the (EA)-s, are the genetic algorithms Particle Swarm Optimization (PSO) (1995)
(GA), which have found a wide spectrum of applications in The PSO algorithm, proposed by Kennedy [39], is an impor-
diverse engineering disciplines. The (GA) uses concepts from tant search technique with good applications in the field of
evolutionary biology. A genetic algorithm with standard com- structural design optimization. The (PSO) is based on the so-
ponents referred to as simple GA is outlined as follow [35]: cial behavior of animals, concerned with grouping by social
Step [Link] population: A design variable is encoded as a bi- forces that depend on both memory of each individual as well
nary string, such as 1001001. Initial population incorporates a as the knowledge gained by the swarm. The particles through
predefined number of individuals. the search space and their positions are updated using the
Step [Link]: For each individual a decoding is used to map current positions, the velocity vector and a time step. A swarm
all substrings to some integer values representing the se- consists of a predefined number of particles referred to as
quence numbers of standard sections. swarm size.
Step 3. Evaluation and fitness. Once an individual is decoded, All the particles are analyzed with the values of design va-
analysis is done in order to obtain the structural response un- riables that they represent and their objective function values
der external loads. Each individual is assigned with a fitness are calculated. A particle’s best position is referred to as par-
score, which indicates the merit of the individual with respect ticle’s best and is stored separately for each particle in a vector
to the overall population. B. On the other hand, the best feasible position located by any
Step 4. Selection and reproduction. A selection is done where particle since the beginning of the process is called the global
individuals of high fitness scores are selected and reproduced. best position, and it is stored in a vector G. The values are up-
Step 5. Crossover. The selected and reproduced individuals are [Link] velocity vector of each particle is updated consider-
paired, in base of a crossover probability. Step 6. Mutation and ing the particle’s current position, the particle’s best position

IJSER
termination. Mutation is applied by randomly altering a gene and its global best position. The position vector of each par-
of 0 to 1 with a probability of less than a given value. The new ticle is then updated with the updated velocity vector. Steps
population replaces the old one and the steps are repeated are repeated until a predefined number of iterations N.
until a number of iterations is reached.
Applications of the evolutionary algorithms have been re- Ant Colony Optimization (1991)
ported widely in the scientific community. The above Simple Ant colony optimization (ACO) technique is inspired from the
(GA), has been improved several times leading to better re- way that ant colonies find the shortest route between the food
sults. (ES) applied to truss structures by Hasancabi[6], showed source and their nest. The technique has been developed by
a minor weight of the structure at a minor time. The efficiency Colorni and Dorigo[40]. Other publications with application of
and the robustness of the (ES) has been proven by the study. the improved algorithms, have reported good results too.
The four typologies of the evolutionary algorithms have the
same structure, they try to mimic the evolutionary process in Artificial bee colony algorithm (ABC) (2005)
nature, and the Theory of Darwin. They change each one with Another novel meta-heuristic algorithm is a nature-inspired
the others in the way the process is realized and the fitness method, the artificial bee colony (ABC). This algorithm was
function is applied. proposed in its original version by Karaboga[41]. Sonmez[42]
Some advantages of enhanced versions of the genetic algo- has used a discrete ABC algorithm for optimum design of
rithms (GA) combined with other deterministic algorithms, truss structures with 582 members and has reported a good
have been documented too [36]. Genetic algorithms gave good performance of the algorithm compared to the others known
results in the layout optimization of truss structures, where meta-heuristic techniques.
topology, shape and size optimization is executed at the same
time[37].
Sisko[38] developed a design center system based on decision
system support for conceptual design at early stages of the Enhanced Honey Bee Mating (EHBMO) (2017)
construction project, considering architectural and structural The Honey bee mating is a swarm based algorithm, where the
aspects at the same time. search algorithm is inspired by the process of mating in ho-
neybees. The (EHBMO) is a very recent optimization algo-
3.2.2. Swarm Intelligence Algorithms. rithm, which applied an enhanced version of the original one.
————————————————
The algorithm is reported to be very competitive with other
meta-heuristic methods analyzed. The (EHBMO) uses the con-
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
University of Tirana in Albania. He currently teaches at POLIS University, cept of giving weight to distant candidates which are slightly
Tirana in [Link]@[Link]. less feasible than the current local candidate, but may hold
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil information about the global optimal solution of the problem.
Engineering at the Polytechnic University of Tirana. He has a long expe-
rience in teaching and is author of various pubblications. nlako@[Link] The robustness of the algorithm has been proven by the study.
[43]
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 711
ISSN 2229-5518
and the current global best solution on new candidate solu-
3.2.3. Physical related Algorithms tions. Further, a multiphase search strategy has been em-
These algorithms are related to the imitation of physical phe- ployed to increase the quality of the final solution. The effi-
nomena, in processes that build an algorithm that finds the ciency of the BB–BC algorithm was compared to previously
optimal solution of a problem. reported (GA), (PSO) and (ACO) based approaches.

Simulated annealing (SA) (1985) Charged system search (CSS) algorithm (2010)
The first publication on Simulated Annealing processes was Charged system search (CSS) has been proposed by Kaveh
done in 1985 by Cerny[44]. Simulated annealing (SA) extends and Talatahari in 2010 [51]. The algorithm has been employed
its process to the annealing of physical systems applied in for the solution for optimum design of skeletal structures in-
thermodynamics. A physical system initially at a high-energy cluding three trusses and two frame structures. The study has
state is cooled down gradually until its minimum energy level reported advantages of (CSS) in comparison to the other meta-
is reached. This process can be simulated to solve optimization heuristic methods. Some enhanced versions of the (CSS) have
problems. Some applications of this algorithm in steel struc- been applied for configuration optimization of truss struc-
tural optimization have demonstrated good results[45]. An tures[52].
efficient simulated annealing algorithm for design optimiza-
tion of truss structures has been reported by Lamberti[46]. 3.2.4. Other stochastic algorithms.
Stochastic optimization methods are processes, where the
Harmony search algorithm (HS) (2001) search for the optimal solution is done by generating new
Harmony search is a non-deterministic method, inspired by populations that satisfies better the result, using some random
the improvisation process of musicians, firstly proposed by variables.
Geem[47]. Further developments and applications have been
reported of the (HS). A survey for this purpose was done by Guided Stochastic Search Algorithm (GSS) (2014)

IJSER
Manjarres[48]. The GSS offer a design procedure, built ad hoc for steel
The idea comes from the process of creation, where musicians trusses, where the generation of the optimal solution is guided
make several harmonies until they find the desired one. It is by the virtual work and other criteria of responses of the algo-
the improvisation process, in which a musician tries to find rithm. The information provided through the structural analy-
the perfect harmony, examining a wide range of combinations. sis and design check stages are utilized for handling strength
Although (HS) algorithm is a random search technique similar constraints. The Virtual Work Theorem is applied to achieve
to the genetic algorithm and the particle swarm optimization, the displacement constraints. The optimization with minimum
it is considerably different from those population based evolu- weight is performed based on both strength and displacement
tionary methods due to its single evolving search memory. criteria. The method has been reported to be efficient in truss
Geem has used the algorithm for sizing optimization of truss structures compared to other algorithms, using from 10 to 349
structures and proved to be more efficient than conventional sizing variables. (GSS) is a recent method proposed by Ka-
mathematical methods and genetic algorithms. The algorithm zemzadeh and Hasancabi[53]
can also be employed for optimum design of other types of Stochastic search techniques perform random moves using
structures such as frame, plate or shell structures. Shabani[49] strategies taken from nature to locate the optimal solution us-
introduced recently in 2017, a new version of the harmony ing a single or a population of candidate designs. Since the
search, the “Selective Refining Harmony Search” (SRHS). (GSS) takes the response computation from generated designs,
it is possible to utilize such valuable information to the design
Big bang crunch optimization. (BB-BC) (2006) process, obtaining better results, in a shorter time.
Erol[50] in 2006, introduced a new meta-heuristic optimization
method called Big Bang–Big Crunch (BB-BC). The algorithm Tabu search. (TS) (1989)
has been proven to be efficient in tackling practical optimiza- The Tabu Search (TS) has been proposed by Glover[54] in
tion problems, and has become a popular meta-heuristic. 1989, and it works with other algorithms to overcome the local
Some publications have reported the design optimization of optimality. Its applications are reported mostly on discrete
planar and spatial truss structures, performed using a mod- constrained combinatorial optimization problems. The tabu
ified version of the algorithm. In order to increase the efficien- search uses a local neighborhood search procedure moving
cy of the (BB–BC) algorithm, a weighting parameter has been from a potential solution to another in the neighborhood, until
introduced to control the influence of both the center of mass some stopping criteria are satisfied. To avoid falling into a trap
of local minima, the solutions admitted to the new neighbor-
————————————————
hood, are determined through the use of memory structures.
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
University of Tirana in Albania. He currently teaches at POLIS University,
Tirana in [Link]@[Link]. 3.3. Comparison of structural size optimization
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil
Engineering at the Polytechnic University of Tirana. He has a long expe-
algorithms.
rience in teaching and is author of various pubblications. nlako@[Link] In the previous paragraphs only some of the most peer re-
viewed algorithms have been introduced. The variety of these
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 712
ISSN 2229-5518
techniques, available in the literature of size structural optimi- the non-deterministic methods, generate candidate designs
zation, make it difficult to choose the right method that is ne- using a fitness equation that is not influenced by the values of
cessary to be applied in the practical applications. the candidate optimal design, so may handle bigger size prob-
Most of the studies analyzed, demonstrate that if the algo- lems, and guarantee a greater effectiveness. As disadvantage
rithm is improved from the original form, it can lead to better they need more iterations.
solutions in a shorter time of iterations. If the criteria are cho- New algorithms are constantly being developed in the field of
sen in the right way, most of the recently developed algo- size structural optimization. These algorithms make it possible
rithms reported good results. to have faster optimal results. These algorithms can handle
The comparisons between the design optimizations for size problems with a higher number of variables. Some of the most
truss structures, have shown the superiority of Simulated An- peer reviewed algorithms have been analyzed in this study.
nealing (SA), Evolution Strategies (ES), Enhanced Genetic Al- The three types of optimization problems: size, shape and to-
gorithms (GAs), Fully Constrained Design (FCD), Guided Sto- pology; are considered more in an integrate application. Even
chastic Search (GSS), Enhanced Artificial Bee Mate Colony this can cause some difficulties in the first phases, but when
(EHBMO) etc. understanding the way to apply the algorithm, it can lead to
better results.
3.4. Practical applications. Some algorithms are more efficient and robust than others.
Some studies conclude that the structural optimization algo- Recently introduced algorithms, and old improved ones have
rithms are efficient tools to solve real world problems. [55]. reported enhanced results. The optimization of skeletal struc-
These have been mostly recorded in the automotive industry tures, was demonstrated to be achieved better with: Simulated
using topology optimization. Annealing (SA), Evolution Strategies (ES), Fully Constrained
(OPTIMA) System was developed by HKUST [56]. It has been Design (FCD), Enhanced Genetic Algorithm (GA), Enhanced
applied to the size optimization of some project for high Honey Bee Mate Optimization (EHBMO), Guided Stochastic
buildings in Honk Kong, using combinations of (OC), (GA) Search (GSS) etc.

IJSER
and structural analysis. The “Kowloon Mega Tower”, in It was observed that ad hoc algorithm built for specific struc-
China, was built optimizing the material costs and distances of tures, offer the best results. So for example, the (GSS), point
the constructions. [36] 3.2.4, built for steel truss structures, showed a greater efficien-
ODA (Structural Optimization Design Analysism acronym cy and robustness compared to other algorithms.
software Inc.), is another commercial software, realization of There are only a few real world applications of size structural
Waterloo University, which applied (OC) criteria to optimization, since these algorithms are known priory in aca-
optimization. A collaboration of Ove Arup, Partners Honk demic contexts and only in very few specialized companies.
Kong ltd and Khust brought to life the North-Eastern Tower of There is a lack between the scientific community studies and
the Honk Kong Station, with objective function the minimal the real world applications in the construction industry. It is
weight. necessary to own a deep knowledge of structural mechanics
The Train Station roof of Florence, realized in 2002, with di- and numerical methods, in order to apply optimization tech-
mensions of 150m x 26m x 15m height, has been projected ap- niques to real world problems.
plying an Extended ESO algorithm [57]. Vast applications in the construction industry may bring to a
new way of designing, with more efficient structures, versus a
sustainable use of resources and less costs.
4. CONCLUSIONS. Further research is necessary to have a more comprehensive
The surveys and the conclusions given in this study are based
detailed state of the art of the structural size optimization al-
on the references analyzed. There is a widespread information,
gorithms.
with too many studies about the argument, and it is not possi-
ble to analyze all them in one paper.
The structural optimization algorithms were divided in two 5. REFERENCES
maxi-groups: the deterministic and non-deterministic me-
thods. The two bigger goals are: efficiency in the number of [1] U. Kirsch, Structural Optimization, Fundamentals and applications,
Springer-Verlag, (1993).
iterations and robustness in finding the optimal solution. The
deterministic methods have the advantage of requiring less [2] P. W. Christensen, An introduction to structural optimization,
iterations to the optimal design; but have the disadvantage of Ontario, Canada: Springer, (2009).
managing only problems with less than 100 variables. Instead [3] J. Hedengren, Course lectures in Optimization techniques in Engineering,
University of UTAH, (2016).
———————————————— [4] F. Bontempi, PhD course in structural optimization, Dottorato in
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic [Link].e Geotecnica, Universita' degli Studi di Roma "La
University of Tirana in Albania. He currently teaches at POLIS University, Sapienza", (2016).
Tirana in [Link]@[Link]. [5] L. A. Schmit, "Structural design by systematic syntesis," in 2nd
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil Conference on Electronic Computation, Pittsburgh, PA, (1960).
Engineering at the Polytechnic University of Tirana. He has a long expe-
rience in teaching and is author of various pubblications. nlako@[Link] [6] O. Hasançebi, S. Çarbas, E. Dogan, P. Erdal and M. P. Saka,
"Performance evaluation of metaheuristic search techniques in the
optimum design of real size pin jointed structures," Computer and
IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 713
ISSN 2229-5518
Structures, vol. 87, p. 284–302, (2009). [28] G. S. A. A. K. S. J. H. M. F. Forest Flager, "Fully Constrained Design:
[7] S. Kazemzadeh and A. Hasançebi, "Optimum Design of Skeletal A general and scalable method for discrete," Computers and Structures
Structures using Metaheuristics: A Survey of the state of the Art.," , p. (140) 55–65, 2014.
International [Link] Engineering & Applied Sciences (IJEAS), vol. 6, pp. 1-11, [29] R. Kicinger, T. Arciszewski and D. K. Jong, "Evolutionary
(2014). computation and structural design: A survey of the state of the art,"
[8] A. Fiacco and G. McCormick, Nonlinear Programming: Sequential vol. 83, (2005).
Unconstrained Minimization Techniques, New York: John Wiley and [30] L. Lamberti and C. Pappalettere, "Metaheuristic design optimization
Sons, (1968). of skeletal structures: a review,," (2012).
[9] A. Rolvnik, C. Mueller and J. Coenders, "State on the Art of [31] I. Rechenberg, "Cybernetic solution path of an experimental
Computational Tools for Conceptual Structural Design," in problem," Library translation No. 1122,, Farnborough, Hants., UK;,
Proceedings of the IASS-SLTE 2014 Symposium “Shells, Membranes and 1965.
Spatial Structures: Footprints”, (2014). [32] L. J. Fogel, A. J. Owens and M. J. Walsh, "Artificial Intelligence
[10] R. Baldock, Structural Optimization in Building Design Practice: Case through Simulated Evolution,," John Wiley, (1966).
studies in topology optimization of bracing systems, Cambridge: [33] J. Holland, Adaptation in Natural and Artificial Systems, (1975).
Dissertation, Cambridge University Engineering Department, (2007).
[34] J. R. Koza, Genetic Programming: On the Programming of Computers
[11] G. Vanderplaats, "Structural optimization for statics, dynamics and by Means of Natural Selection, USA: MIT Press. ISBN 0-262-11170-5,
beyond," J. Braz. Soc. Mech. Sci. & Eng., vol. 8, p. Rio de Janeiro, (2006). (1992).
[12] G. Abhishek, Numerical methods using Matlab, Apress, Springer, [35] D. E. Goldberg, "Genetic algorithms in search optimization and
(2015). machine learning.," (1989).
[13] J. S. Arora, "Methods for discrete variable structural optimization.," [36] C. M. Chan and P. Liu, "Design optimization of practical tall concrete
Recent Advances in optimal structural design, pp. 1-40, (2002). buildings using Hybrid Optimality Criteria and Genetic Algorithms.,"
[14] W. Hare, J. Nutini and J. Tesfamariam, "A survey of non-gradient in 8th International Conference on Computing in Civil & Building
optimization methods in structural engineering," (2013). Structures, CA, USA., (2000).
[15] C. Y. Phd thesis, Application of Deterministic Operations Research for [37] R. Cazacu and L. Grama, "Steel truss optimization using genetic

IJSER
Structural Optimization, MASSACHUSETTS INSTITUTE OF algorithms and FEA," in The 7th International Conference
TECHNOLOGY, 2015. Interdisciplinarity in Engineering (INTER-ENG 2013), (2014).
[16] K. Murtty, Linear programming, New York: John Wiley & Sons, ISBN [38] G. M. Sisko, J. C. Miles and C. J. Moore, "Designer centered
0-471-09725, (1983). development of GA based DSS for Conceptual design of buildings," J.
[17] R. Fleetcher and C. M. Reeves, "Function minimization by conjugate of Computing in Civil Engineering, vol. 17, pp. 159-166, (2003).
gradients," (1964). [39] J. Kennedy and R. Eberhart, Particle swarm optimization. In: IEEE
[18] Zoutendijk, "Methods of feasible directions," Amsterdam, (1960). international conference on neural networks, IEEE Press, 1995.
[19] J. E. Kelley, "The Cutting Plane Method for Solving Convex [40] A. Colorni, M. Dorigo and V. Maniezzo, "Distributed optimization by
Programs," J. SIAM (8), pp. 702-712, (1960). ant colony," USA, (1991).
[20] G. Stewart, "A nonlinear programming technique for the optimization [41] D. Karaboga, An idea based on honey bee swarm for numerical
of continous processing systems," (1961). optimization., Erciyes University: Engineering Faculty, Computer
Engineering Department, 2005.
[21] K. Stasa, Projektimi Optimal i Kapriatave Metalike, Tirana, Albania: PhD
Dissertation, Universiteti Politeknik i Tiranes (UPT), (1994). [42] M. Sonmez, "Discrete optimum design of truss structures using
artificial bee colony algorithm," Struct Multidisc Optim , vol. 43, pp.
[22] P. Ghisbain, "Application of a Gradient-Based Algorithm to Structural
85-97, 2011.
Optimization," Master Thesis, MIT, Massachusets, (2009).
[43] M. R. Maheri, H. Shokrian and M. M. Narimani, "An enhanced honey
[23] W. Prager, "Optimality criteria in structural design," (1968).
bee mating optimization algorithm for design of side sway steel
[24] V. B. Venkayya, N. S. Khot and Berke, "Application of optimality frames," Advances in Engineering Software, vol. 109, pp. 62-72, (2017).
criteria approaches to automated design of large practical structures,"
[44] V. Cerny, "Thermo-dynamical approach to the traveling salesman
(1973).
problem: anefficient simulation algorithm.," J. Optim. Theor. Appl., vol.
[25] M. S. Bazaraa and J. J. Goode, "Neccessary Optimality Criteria in 45, pp. 41-51, (1985).
Mathematical Programming in the Presence of Differentiability," J. of
[45] S. Kirckpatrick, C. D. Gerlatt and M. P. Vecchi, "Optimization by
mathematical analysis and applications, vol. 40, pp. 609-621, (1972).
simulated annealing," Science, p. 671–80., (1983).
[26] G. I. Rozvany, Structural design via optimality criteria: the Prager
[46] L. Lamberti, "An efficient simulated annealing algorithm for design
approach to structural optimization., Dordecht: Kluwer Academic,
optimization of truss structures," Bari, (2008).
(1989).
[47] Z. E. Geem, J. H. Kim and G. Loganathan, "A new heuristic
[27] A. Shukla and A. Misra, "Review of optimality criterion approach
optimization algo- rithm: Harmony search.," vol. 76, no. 2, (2001).
scope, limititation and development in topology optimization.,"
(2013). [48] D. Manjarres, S. G. Lopez, I. L. Torres, J. Delser and M. N. Bilbao, "A
survey on applications of the harmony search algorithm," vol. 26,
(2013).
————————————————
[49] M. Shabani, S. A. Mirroshandel and H. Asheri, "Selective Refining
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic Harmony Search: A new optimization algorithm," vol. 81, (2017).
University of Tirana in Albania. He currently teaches at POLIS University,
Tirana in [Link]@[Link]. [50] O. K. Erol and I. Eskin, "A New optimization method: Big Bang–Big
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil Crunch," Adv Eng Software, vol. 37, pp. 106-111, (2006).
Engineering at the Polytechnic University of Tirana. He has a long expe- [51] A. Kaveh and S. Talatahari, "A novel heuristic optimization method:
rience in teaching and is author of various pubblications. nlako@[Link]
charged system search," Acta Mech., vol. 213, pp. 267-289, (2010).

IJSER © 2017
[Link]
International Journal of Scientific & Engineering Research Volume 8, Issue 8, August-2017 714
ISSN 2229-5518
[52] A. Kaveh and S. Talatahari, "A charged system search with a fly to
boundary method for discrete optimum design of truss structures,"
Asian J Civ Eng., vol. 11, pp. 277-293, (2010).
[53] S. A. Kazemzadeh, O. Hasançebi and M. P. Saka, "Guided stochastic
search technique for discrete sizing optimization," Computers and
Structures, vol. 134, p. 62–74, (2014).
[54] F. Glover, "Tabu search – part 1. ORSA," J. Comput. and Struct., vol. 1
(2), pp. 190-206, (1989).
[55] C. Luebkeman and K. Shea, "CDO: Computational design +
optimization in building practice," The Arup J. (3), (2005).
[56] C. M. Chan, "Advances in structural optimization of tall buildings in
Hong," in Japan - Korea Joint Symposium on Optimization of Structural
and Mechanical Systems, Kanazawa, Japan, (2004).
[57] C. Cui, H. Ohmori and M. Sasaki, "Computational Morphogenesis of
3D structures by extended ESO method," J. of the Inernational
Association for Shell and Spatial Structures, vol. 44 (1), pp. 51-61, (2003).

IJSER

————————————————
• ErsilioTushaj is a PhD Candidate in Civil Engineering at the Polytechnic
University of Tirana in Albania. He currently teaches at POLIS University,
Tirana in [Link]@[Link].
• Niko Lako is Professor of Steel Structures for the PhD Program of Civil
Engineering at the Polytechnic University of Tirana. He has a long expe-
rience in teaching and is author of various pubblications. nlako@[Link]

IJSER © 2017
[Link]