applied

sciences
Article
Optimisation of Shear and Lateral–Torsional
Buckling of Steel Plate Girders Using
Meta-Heuristic Algorithms
Celal Cakiroglu 1, *, Gebrail Bekdaş 2 , Sanghun Kim 3 and Zong Woo Geem 4, *
1 Department of Civil Engineering, Turkish-German University, Sahinkaya Cad 86, Istanbul 34820, Turkey
2 Department of Civil Engineering, Istanbul University, Cerrahpaşa, Istanbul 34310, Turkey;
bekdas@istanbul.edu.tr
3 Department of Civil and Environmental Engineering, Temple University, Philadelphia, PA 19122, USA;
sanghun.kim@temple.edu
4 College of IT Convergence, Gachon University, Seongnam 13120, Korea
* Correspondence: cakiroglu@tau.edu.tr (C.C.); geem@gachon.ac.kr (Z.W.G.)




Received: 24 April 2020; Accepted: 20 May 2020; Published: 25 May 2020


Abstract: The shear buckling of web plates and lateral–torsional buckling are among the major failure
modes of plate girders. The importance of the lateral–torsional buckling capacity of plate girders was
further evidenced when several plate girders of a bridge in Edmonton, Alberta, Canada failed in 2015,
because insufficient bracing led to the lateral buckling of the plate girders. In this study, we focus on
the optimisation of the cross-sections of plate girders using a well-known and extremely efficient
meta-heuristic optimisation algorithm called the harmony search algorithm. The objective of this
optimisation is to design the cross-sections of the plate girders with the minimum area that satisfies
requirements, such as the lateral–torsional buckling load and ultimate shear stress. The base geometry,
material properties, applied load and boundary conditions were taken from an experimental study
and optimised. It was revealed that the same amount of load-carrying capacity demonstrated by this
model can be achieved with a cross-sectional area 16% smaller than that of the original specimen.
Furthermore, the slenderness of the web plate was found to have a decisive effect on the cost-efficiency
of the plate girder design.

Keywords: lateral–torsional buckling; harmony search; plate girder; shear buckling; optimisation

1. Introduction
Plate girders are prone to buckling under various load combinations because they are made of
thin plate elements. The web part of these members carries the shear loads; therefore, shear buckling
is a critical phenomenon affecting the web plates of I-shaped plate girders. The out-of-plane shear
buckling resistance of plate girders is often increased by applying transverse stiffeners or corrugated
webs. While the web of the plate girders is the primary element resisting the shear buckling, the flanges
are the primary elements carrying the bending and torsional loads. Therefore, in addition to the
shear buckling analysis, the design of plate girders also involves the lateral–torsional buckling of
structural members. Figure 1 illustrates an example of lateral plate girder buckling during the process
of installation on a bridge.
The design of plate girders with the highest possible load-carrying capacity within cost and material
restraints consists of finding the optimal combination of plate thicknesses, web plate slenderness,
and stiffener spacing [1–3]. Furthermore, to obtain an adequate dimension of steel I-girders, both lateral–
torsional and shear buckling should be considered in all construction stages.

Appl. Sci. 2020, 10, 3639; doi:10.3390/app10103639 www.mdpi.com/journal/applsci

Appl. Sci. 2020, 10, 3639 2 of 14
Appl.
Appl. Sci.
Sci. 2020,
2020, 10,
10, xx FOR
FOR PEER
PEER REVIEW
REVIEW 22 of
of 14
14

Figure
Figure 1.1.
Figure 1. Lateral
Lateral buckling
Lateralbuckling of
bucklingof bridge
ofbridge plate girders
bridge plate girdersin
inEdmonton,
in Edmonton,Alberta,
Edmonton, Alberta,Canada
Alberta, Canada
Canada [4].
[4].
[4].

InIn
In this
thisthis study,
study,
study, the
the flange
flange
the and
flangeand web
andweb dimensions
dimensions of
webdimensions of aa doubly
doubly symmetric
doubly symmetricI-shaped
symmetric I-shaped
I-shaped plate
plate
plate girder are
girder
girder areare
optimised.
optimised.
optimised. For
For the
the
For optimisation,
theoptimisation, a novel
optimisation,aanovel meta-heuristic
novelmeta-heuristic optimisation algorithm
optimisationalgorithm
meta-heuristic optimisation called
algorithmcalled the
calledthe harmony
the harmony
harmony
search
search
search algorithm
algorithm
algorithm isis applied.
applied.
is Optimised
applied.Optimised cross-sectional
cross-sectional dimensions
Optimisedcross-sectional dimensionswere
dimensions wereobtained
were obtained
obtained with aa satisfactory
with
with a satisfactory
satisfactory
amount
amount
amount ofofof computational overhead
computationaloverhead
computational leading
overheadleading to
leadingtotoaaasignificant
significant
significant reduction
reduction in
reduction inthe
in thecross-sectional
the cross-sectionalarea.
cross-sectional area.
area.

1.1.1.1.
1.1. Shear
Shear Capacity
Capacity
Shear Capacity
Steel
Steel
Steel plates
platesplates
under under
under
shearshear
loadsloads
shear loads are
are known
are known known to
to exhibit
to exhibit exhibit aa significant
a significant amountamount
significant amount of
of load-carrying
of load-carrying load-carrying
capacity in
capacity
thecapacity in the post-buckling
in the post-buckling
post-buckling regime [5]. This regime
regime [5]. This
[5]. This
structural structural
structural
behaviour behaviour
wasbehaviour
thoroughly was thoroughly
wasinvestigated
thoroughly by investigated
investigated by
by
the researchers
the
the researchers
researchers and
and attributed
attributed to
to the
the existence
existence of
of tensile
tensile stresses
stresses acting
acting
and attributed to the existence of tensile stresses acting in the diagonal direction of the plates after in
in the
the diagonal
diagonal direction
direction of
of
the
the plates
plates after
after the
the onset
onset of
of shear
shear buckling
buckling [2,6,7].
[2,6,7]. These
These areas
areas along
along the
the diagonal
diagonal axis
axis of
of the
the web
web plate
plate
the onset of shear buckling [2,6,7]. These areas along the diagonal axis of the web plate where tensile
where
whereactstensile
tensile stresses
stresses acts
actsincan
can also
also2. be seen
beTheir in
in Figure
seenresearchFigure 2. Their research
research led led to
to the
the development of
stresses can also be seen Figure led2.toTheir
the development of variousdevelopment
models forofthe
various
various of models
models for the prediction
for the prediction of the post-buckling
of the post-buckling shear capacity of web plates based on the
prediction the post-buckling shear capacity of web platesshear basedcapacity of web of
on the concept plates basedfield
a tension on along
the
concept
concept of
of aa tension
tension field
field along
along the
the plate
plate diagonal.
diagonal. Tension
Tension field
field theory
theory is
is often
often used
used as
as an
an umbrella
umbrella
the plate diagonal. Tension field theory is often used as an umbrella term for these models. This theory
term
term for
for these
these models.
models. This This theory
theory is is based
based on on the
the observation
observation that that the
the stiffeners
stiffeners of of aa plate
plate girder
girder take
take
is based on the observation that the stiffeners of a plate girder take up the compressive stresses resulting
up
up the
the compressive
compressive stresses stresses resulting
resulting from from the
the shear
shear forces,
forces, and
and the
the web
web plate
plate resists
resists buckling
buckling owing
owing
from
to the shear forces, and thetensile
web plate resists buckling the owing to shear forces produced by tensile
to shear
shear forces
forces produced
produced by by tensile stresses
stresses forming
forming along
along the diagonal
diagonal axisaxis ofof the
the plate
plate [8,9].
[8,9]. Among
Among
stresses
the forming along the diagonal axis of the plate [8,9]. Among the models of tension field theory,
the models
models of of tension
tension field
field theory,
theory, the the theory
theory developed
developed by by Basler
Basler [6][6] gained
gained the the most
most widespread
widespread
theacceptance
theory developed by Basler [6] gainedthis the most widespread acceptance in the the
research community;
acceptance in in the
the research
research community;
community; this model model was was also
also described
described in in [10].
[10]. On
On the otherother hand,
hand, the
the
this model
model was
model developed also
developed by described
by Porter
Porter [11]in [10].
[11] and
and theOn the
the rotatedother
rotated stress hand, the
stress field model
field method developed
method developed
developed by by Porter
by Höglund [11]
Höglund [12] and
[12] arethe
are
rotated
includedstress
included infield
in the method developed
the European
European design
design codes
codesby[13].
Höglund [12] are included in the European design codes [13].
[13].

Figure
Figure 2.2.
Figure 2. Post-buckling deformation
Post-bucklingdeformation
Post-buckling of
of stiffened
deformation of stiffened plate
stiffened plate girder
plate girderunder
girder undertransverse
under transverseloading
transverse loading [14].
loading[14].
[14].
Appl. Sci. 2020, 10, 3639 3 of 14
Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 14

Figure
Figure 33 shows
shows aa schematic
schematic of of the
the tension
tension field
field on
on aa girder
girder panel
panel surrounded
surrounded by by flanges
flanges and
and
transverse stiffeners, where 𝑠 denotes the width of the tension field of a web plate under the ultimate
transverse stiffeners, where s denotes the width of the tension field of a web plate under the ultimate
postbuckling
postbucklingshear stressτu𝜏. The
shearstress . Theplate in in
plate Figure
Figure3 represents
3 representsthe the
unstiffened partpart
unstiffened of a of
plate girder
a plate web
girder
and
webisand
assumed to be simply
is assumed supported
to be simply (SS) at all
supported (SS)edges.
at all Basler
edges.[6] developed
Basler an equation
[6] developed that predicts
an equation that
τpredicts
u . This 𝜏 . This equation was later modified by Fujii [15], Gaylord [16], and Selberg [17] as follows:
equation was later modified by Fujii [15], Gaylord [16], and Selberg [17] as follows:

𝜏 = 𝜏 + 𝜎 1 τ−cr
! !
sinθd . (1)
τu = τcr + σ y 1 − . (1)
τy 2 + cosθd

Figure 3. Schematic of Basler’s tension field theory model [5].

Figure 3. Schematic of Basler’s tension field theory model [5].
Once τu is known, the ultimate postbuckling shear force Vu can be obtained as Vu = τu Dtw ,
Once 𝜏 is known, the ultimate postbuckling shear force V can be obtained as V = τ Dt ,
where D is the depth of the plate (see Figure 3), and tw is the thickness of the web plate.
where 𝐷 is the depth of the plate (see Figure 3), and 𝑡 is the thickness of the web plate.
In Equation (1), θd is the angle of the web panel diagonal, σ y is the yield strength of the plate
In Equation (1), 𝜃 is the angle of the web panel diagonal, 𝜎 is the yield strength of the plate
material, from which the shear yield strength τ y can be obtained as τ y = 0.6σ y . The elastic shear
material, from which the shear yield strength 𝜏 can be obtained as 𝜏 = 0.6𝜎 . The elastic shear
buckling strength τcr in Equation (1) is calculated as follows [18]:
buckling strength 𝜏 in Equation (1) is calculated as follows [18]:
kπ 2
𝑘𝜋 E𝐸
𝜏τcr =
=  2 , , (2)
12 ( 1 − ν 2 ) D𝐷 (2)
12(1 − 𝜈 ) tw
𝑡
where EEisisthe
where themodulus
modulusofof elasticity, ν isν the
elasticity, Poisson’s
is the Poisson’sratio, D/tw
ratio, D⁄ist theisslenderness and kand
ratio,ratio,
the slenderness is the
k
shear buckling coefficient, which can be calculated as a function of a/D (span-to-depth
is the shear buckling coefficient, which can be calculated as a function of a⁄D (span-to-depth ratio) ratio) and the
assumed boundary
and the assumed conditions
boundary of the web
conditions of plate.
the webTheplate.
equation for k for simply
The equation supported
for 𝑘 for boundary
simply supported
conditions is as follows [19]:
boundary conditions is as follows [19]:
5.0
k = 5.0 + . 2, (3)
𝑘 = 5.0 +(a/D( ⁄ )
), (3)

where a𝑎isisthe
where theclear
cleardistance
distancebetween
betweentransverse stiffeners.
transverse stiffeners.
1.2. Lateral–Torsional Buckling
1.2. Lateral–Torsional Buckling
Another critical failure mode that must be considered in the design of large I-girders is the
Another critical failure mode that must be considered in the design of large I-girders is the
lateral–torsional buckling. Most methods developed for the prediction of the lateral–torsional buckling
lateral–torsional buckling. Most methods developed for the prediction of the lateral–torsional
capacity are only applicable in the case of a uniform bending moment distribution and require a
buckling capacity are only applicable in the case of a uniform bending moment distribution and
moment gradient factor for the adjustment of the predicted capacities to the case of non-uniform
require a moment gradient factor for the adjustment of the predicted capacities to the case of non-
moment distributions [20].
uniform moment distributions [20].
The moment gradient factor 𝐶 incorporates the variation of the bending moment along
unbraced sections of the I-girder into the buckling load capacity prediction. According to Wong and
Driver [21], 𝐶 can be calculated for any non-uniform moment distribution along an unbraced span
as follows:
Appl. Sci. 2020, 10, 3639 4 of 14

The moment gradient factor Cb incorporates the variation of the bending moment along unbraced
sections of the I-girder into the buckling load capacity prediction. According to Wong and Driver [21],
Cb can be calculated for any non-uniform moment distribution along an unbraced span as follows:

4Mmax
Cb = q ≤ 2.5. (4)
M2max + 4M2A + 7M2B + 4M2C

where Mmax is the absolute value of the maximum bending moment, and MA , MB , and MC are the
absolute values of the bending moments at lengths of a/4, a/2, and 3a/4 along the unbraced span
of the I-girder, respectively, where a denotes the total length of the unbraced span. In this study,
the moment distribution between two stiffeners is assumed to be linear, corresponding to the unbraced
span of the girder beam shown in Figure 2. Once Cb is known, the critical bending moment for the
lateral–torsional buckling can be calculated as follows:

Mcr = Cb M0cr , (5)

where M0cr is the critical bending moment of an unbraced span under uniform bending moment,
which can be expressed as [22]:
s
π π2 ECw
!
M0cr = EI y GJ + . (6)
a a2

where G is the shear modulus, E is the modulus of elasticity, J is the St. Venant torsion constant, I y is
the moment of inertia with respect to the minor axis of the I-section, and Cw is the warping constant.
The equations for the calculation of J and Cw are given in the Appendix A.

2. Method
The optimisation of the I-girder involves the minimisation of the cross-sectional area,
while maintaining the critical lateral–torsional buckling load Mcr and the ultimate post-buckling
shear stress τu above certain predetermined values. While, in practice, plate girders are often designed
by trial and error, the application of optimisation techniques can lead to better structural performance
and more economical designs. The parameters that can be varied in the process of optimisation are the
flange thickness t f , flange width b f , web plate thickness tw , depth of the web plate D, and unbraced
length between transverse stiffeners a. To start this process, certain constraints need to be placed on
the design parameters and the load-carrying capacity of the structure. These constraints ensure that
the critical lateral–torsional buckling load and the ultimate post-buckling stress obtained from the
optimisation process are greater than certain threshold values. In this study, these threshold values
are chosen to be the values of the bending moment capacity and the ultimate shear stress of the
girder beam tested by Mamazizi et al. [14]. The main finding of the tests carried out by Mamazizi
et al. [14] was that the ultimate shear stress equations of Eurocode 3 are not conservative for certain
ranges of web plate slenderness. In the next step, the design parameters for which constraints are
necessary are identified. One of these design parameters is the slenderness of the web plate, which has
a major impact on the ultimate post-buckling shear strength of the girder plate. The effect of this
parameter on the ultimate post-buckling shear resistance of plate girders has been studied extensively
both experimentally and numerically [23–25]. According to their slenderness ratio, plate girders are
classified as compact, noncompact, or slender; note that the slenderness of both the web and flange
needs to be considered [26]. Compact sections are defined as sections that allow the full development
of the plastic moment prior to the local buckling of the flange or the web, whereas the local buckling of
a non-compact section can occur before the full development of the plastic moment. Slender sections
are sections that would fail due to local buckling before reaching the yield stress. Figure 4 shows the
Appl. Sci. 2020, 10, 3639 5 of 14

classification of the beam sections according to their slenderness, together with the variation of the
corresponding
Appl. Sci. 2020, 10, xnominal
FOR PEERmoment.
REVIEW 5 of 14

Figure 4.
Figure Classificationof
4. Classification ofbeam
beamsections
sectionsaccording
according to
to their
their slenderness
slenderness [1].
[1].

In this
In this study,
study, thethe girder
girder beam
beam section
section is
is optimised
optimised under
under different
different slenderness
slenderness constraints.
constraints. In
In the
the
first optimisation attempt, the slenderness of the section is chosen according to the
first optimisation attempt, the slenderness of the section is chosen according to the girder tested by girder tested by
Mamazizi et
Mamazizi et al.
al. [14].
[14]. To
To apply
applythis
thisconstraint,
constraint,thethelimits
limitsof
ofthe
theslenderness
slendernessas asper
perAISC
AISC[19]
[19]are
areused.
used.
The slenderness limits for the flange and web are,
The slenderness limits for the flange and web are, respectively, respectively,
s
𝜆 = bf ≤ E, (7)
λf = ≤ , (7)
2t f σy
𝜆 = ⩽ 5.7s . (8)
D E
λ = 6 5.7 . (8)
A section is classified as slender if wany tone w of theseσ y limits is violated. According to these
slenderness limits, the girder used as a reference in this study has a slender web, but its flange is not
A section is classified as slender if any one of these limits is violated. According to these slenderness
slender. In the second optimisation attempt, both the flange and the web are constrained to be
limits, the girder used as a reference in this study has a slender web, but its flange is not slender. In the
noncompact.
second optimisation
Furthermore, theattempt,
stiffenerboth the flange
spacing and the web
is constrained to beare constrained
not less than 750 to be noncompact.
mm, which is equal to
Furthermore,
the stiffener spacing the stiffener spacing
considered is constrainedstudy
in the experimental to be ofnotMamazizi
less than et
750 al.mm,
[14].which
Furtheris equal to the
constraints
stiffener spacing considered in the experimental study of Mamazizi et al. [14].
are imposed on the lateral–torsional buckling load and the ultimate shear stress capacity of the Further constraints are
imposed on the lateral–torsional buckling load and the ultimate shear stress capacity
section. Based on the experimental studies of the girder plates conducted by Mamazizi et al. [14], the of the section.
MBasedandonτthevalues
experimental studiesare
of the section of the
notgirder
allowed plates conducted
to be less thanby Mamazizi
11.83 × 10 et Nmmal. [14],and McrMPa,
the64 and
τ u values
respectively.of the section are not allowed to be less than 11.83 × 10 10 Nmm and 64 MPa, respectively.
Under these conditions, the best and most optimised combination of the flange and web
dimensions is obtained the
Under these conditions, best and
through the most
harmonyoptimised
searchcombination
algorithm. of the flange
Table 1 lists and web dimensions
the experimental andis
obtained through the harmony search algorithm. Table 1 lists
optimised values of the cross-sectional dimensions and the stiffener spacing. the experimental and optimised values
of the cross-sectional dimensions and the stiffener spacing.
Table 1. Geometric and material properties used in the experimental study by Mamazizi et al. [14]
Table 1. Geometric and material properties used in the experimental study by Mamazizi et al. [14].
𝒕𝒇 [𝒎𝒎] 𝒃𝒇 [𝒎𝒎] 𝒕𝒘 [𝒎𝒎] 𝑫[𝒎𝒎] 𝒂[𝒎𝒎] 𝒇𝒚𝒇 [𝑴𝑷𝒂] 𝒇𝒚𝒘 [𝑴𝑷𝒂]
tf [mm] b250
f [mm] tw 2[mm] D [mm] a [mm] fyf [MPa] fyw [MPa]
15 800 750 235 210
15 250 2 800 750 235 210
Harmony Search Method
The use of meta-heuristic techniques for optimisation problems in various fields of science and
engineering has increased in recent years, especially in structural engineering. For example, the
harmony search and flower pollination algorithms have been employed for the optimum design of
truss systems [27,28], steel frames [29], cylindrical reinforced concrete walls [30], plane stress systems
[31], PID controlled active tuned mass damper [32], retaining walls [33], and for the stacking sequence
Appl. Sci. 2020, 10, 3639 6 of 14

Harmony Search Method

The use of meta-heuristic techniques for optimisation problems in various fields of science
and engineering has increased in recent years, especially in structural engineering. For example,
the harmony search and flower pollination algorithms have been employed for the optimum design
of truss systems [27,28], steel frames [29], cylindrical reinforced concrete walls [30], plane stress
systems [31], PID controlled active tuned mass damper [32], retaining walls [33], and for the stacking
sequence optimisation of laminated composite plates [34]. In addition to harmony search and flower
pollination, techniques such as random forest [35,36], gene expression programming [37], supervised
machine learning [38], neuro-swarm optimisation [39] and the imperialist competitive algorithm [40]
have also been applied to various engineering problems.
The harmony search algorithm was developed by Geem et al. [41] and has been widely adopted
for the optimisation of a water network design [42], a slope stability analysis [43], heat and power
systems [44], job shop scheduling [45], team orienteering [46], and vehicle routing [47]. The method was
initially designed with discrete valued data for musical composition and was then further developed
for application to the optimisation of continuous valued solution vectors, e.g., those encountered in
the dimensioning of structural components. A parameter-setting-free version of the harmony search
algorithm was developed by Geem and Sim [48]; this algorithm is more accessible and efficient because
the proper selection of algorithm-specific parameters is a difficult task.
The harmony search optimisation algorithm requires a predetermined number of design variables
and an objective quantity to be maximised or minimised. The design variables of a doubly symmetric
plate girder are the width of the flange (b f ), thickness of the flange (t f ), depth of the web plate (D),
thickness of the web plate (tw ), and stiffener spacing (a), and the objective quantity to be minimised is
the cross-sectional area of the girder beam. The harmony search optimisation process starts with the
generation of a certain number of design variable combinations, each of which is called a candidate
solution vector. This initial population of candidate solutions is randomly generated within predefined
design constraints. From any given population of solutions, the solution vectors that deliver the best
and worst results are identified. In the next step, based on certain rules, a new candidate solution is
generated and compared to the members of the previously generated population. If the new candidate
solution performs better than the worst performing solution vector in the population, the newly
generated solution vector is incorporated into the population and the previous worst-performing
solution vector is removed from the population. This procedure is repeated for a predetermined
number of iterations, and the convergence of the result is observed.
The process of generating the new candidate solution vector relies on the harmony memory
consideration rate (HMCR) and the pitch adjustment rate (PAR). In this analysis, these parameters
were treated as variable values and calculated using HMCR = 0.5 ∗ (1 − iter/maxiter) and
PAR = 0.05 ∗ (1 − iter/maxiter), respectively, where iter denotes the index of the current harmony
search iteration, and maxiter denotes the stopping criterion for this index. Once HMCR and PAR are
known, the new solution vector is calculated as follows:

k = [int](rand ∗ HMS),
xi,min + rand ∗ (xi,max − xi,min ), i f HMCR > rand
(
(9)
xi,new =
xi,k + rand2 ∗ PAR ∗ (xi,max − xi,min ), i f HMCR ≤ rand

where rand and rand2 are random numbers in the intervals (0, 1) and (−1/2, 1/2), respectively; HMS is
the total number of candidate solution vectors in the population; xi,k is the i-th design variable in the
k − th member in the population of candidate solution vectors; k is the integer value nearest to the
product rand ∗ HMS; xi,min and xi,max are the minimum and maximum values that of the i − th variable
in the population, respectively.
In the initial step of randomly populating the design vectors as well as in the subsequent
iteration steps, certain constraints need to be imposed on the design variables. Besides the slenderness
Appl. Sci. 2020, 10, 3639 7 of 14

constraints given in Equations (7) and (8), the minimum sizes are defined for the flange and the web.
These geometric constraints are in the form of lower bounds for the flange, and web dimensions and are
chosen to be of the same order of magnitude as the specimen dimensions in the experiment conducted
by Mamazizi et al. [14]. These lower bounds are listed in Table 2.

Table 2. Lower bounds for the flange and web dimensions in the harmony search optimisation.

tf [mm] bf [mm] tw [mm] D [mm]

5
Appl. Sci. 2020, 10, x FOR PEER REVIEW 100 1 200 7 of 14

3.3. Results
Results
The candidate
The candidate solutions
solutions thatthat delivered
delivered the theoptimum
optimumcross-sectional
cross-sectional dimensions
dimensions resulting
resulting in in aa
minimumarea
minimum areaobtained
obtained withwith
the the harmony
harmony searchsearch
algorithmalgorithm converged.
converged. In the
In the first first to
attempt attempt
optimise to
optimise the cross-sectional dimensions, the web was constrained to be
the cross-sectional dimensions, the web was constrained to be slender, and the flange was constrained slender, and the flange was
constrained
to be noncompact to be noncompact in order the
in order to replicate to replicate
conditions the ofconditions of the experimental
the experimental study [14]. studyIn this [14]. In
case,
this case, approximately
approximately 500 iterations
500 iterations were sufficient
were sufficient to reach convergence
to reach convergence to an optimum
to an optimum solutionsolution
with a
with a minimum
minimum area of theareacross-section.
of the cross-section.
Figure 5Figure
shows5the shows the convergence
convergence of the minimum
of the minimum area among area
among
the the population
population of candidateof candidate
solutionssolutions
in the firstin800theharmony
first 800 search
harmony search iterations.
iterations. As can be seen As can frombe
seen
the from the
figure, the solution
figure, the solution
stabilises at stabilises at a cross-sectional
a cross-sectional area of 5741 area mm2of 5741
and 𝑚𝑚at that
stays and level
staysfor at that
the
level for the
remaining remaining
iterations. iterations.
Although thisAlthough
optimisedthis valueoptimised value of the cross-sectional
of the cross-sectional area with a slender area webwithisa
slender web is 63%
approximately approximately
of the area 63% of theofgirder
the area
beam of the
testedgirder beam tested
by Mamazizi by[14],
et al. Mamazizi
it yieldset aal.slightly
[14], it
yields a slightly greater bending moment capacity and an ultimate
greater bending moment capacity and an ultimate shear stress 11% greater than the experimental shear stress 11% greater than the
experimental
specimen. specimen. Aof
A comparison comparison of the experimental
the experimental dimensions of dimensions of the cross-section
the cross-section with the
with the optimised
optimised dimensions
dimensions can be found caninbeTable
found 3. inAn Table 3. An observation
observation of the values of the values
listed listed3inshows
in Table Table that
3 showsthe
that the unbraced span length stayed unchanged after the optimisation,
unbraced span length stayed unchanged after the optimisation, which was expected, owing to the which was expected, owing
to the adverse
adverse effect ofeffect of increasing
increasing this quantity
this quantity on the lateral–torsional
on the lateral–torsional bucklingbuckling
capacity capacity of the beam.
of the girder girder
beam. Furthermore,
Furthermore, the optimised
the optimised cross-sectioncross-section
has greater has web greater web slenderness
slenderness (D/tw ) and (𝐷/𝑡flange)slenderness
and flange
(bslenderness (𝑏 /2𝑡
f /2t f ) values. From) values. From this observation,
this observation, it can be inferredit can that
be inferred that the
the increased increased slenderness
slenderness of the web
of theand
plate web theplate andcan
flanges thecontribute
flanges can to contribute to a moredimensioning
a more economical economical dimensioning
of the cross-sectionof theunder
cross-
section under some circumstances. The optimised cross section is
some circumstances. The optimised cross section is also analyzed using the software package CUFSMalso analyzed using the software
package CUFSM
(Constrained (Constrained and
and Unconstrained Unconstrained
Finite Strip Method). Finite
The Strip Method).mode
first buckling The first
of thebuckling
optimised mode crossof
the optimised
section togethercrosswithsection
the normaltogether
stresswith the normal
distribution can stress distribution
be seen in Figurecan6. be seen in Figure 6.

Figure5.5.Variation
Figure Variationin
inthe
theminimum
minimumcross-sectional
cross-sectional area
areaof
ofthe
theslender
slenderweb
webin
inthe
thefirst
first800
800harmony
harmony
search
searchiterations.
iterations.

Table 3. Geometric and mechanical properties of the optimised cross-section with slender web.

𝒃𝒇 [𝒎𝒎] 𝒕𝒇 [𝒎𝒎] 𝑫[𝒎𝒎] 𝒕𝒘 [𝒎𝒎] 𝒂[𝒎𝒎] 𝑴𝒄𝒓 [𝑵𝒎𝒎] 𝝉𝒖 [𝑴𝑷𝒂] A [𝒎𝒎𝟐 ]

Experimental
250 15 800 2 750 1.1834 ⋅ 10 64.017 9100
value
Appl. Sci. 2020, 10, x FOR PEER REVIEW 8 of 14
Appl. Sci. 2020, 10, 3639 8 of 14

Table 3. Geometric and mechanical properties of the optimised cross-section with slender web.

bf [mm] tf [mm] D [mm] tw [mm] a [mm] Mcr [Nmm] τu [MPa] A [mm2 ]

Experimental value 250 15 800 2 750 1.1834 · 1011 64.017 9100
Optimised value 265.4 8.59 1182 1 750 1.1835 · 1011 70.98 5741
Difference [%] 6.16 −43 48 −50 0 0.008 11 −37
Appl. Sci. 2020, 10, x FOR PEER REVIEW 8 of 14

Figure 6. Results of the analysis with finite strip method using the software package CUFSM for the
slender web.

Although the optimised cross-section with a slender web delivers good results from an
economical point of view, the 1-mm-thick web of this optimal configuration is not practical. To obtain
a better solution, a second attempt at optimising the cross-sectional dimensions was made by
confining the slenderness value of both the web plate and the flange within the noncompact range.
The harmony search process for this second case can be seen in Figure 7. The outcome of this second
attempt was6. satisfactory.
Figure Results of of the The
the analysisoptimised
analysis with finitecross-sectional
finite strip method
method usingarea was
using approximately
the software
software 84% of for
package CUFSM
CUFSM thatthe
of the
Figure 6. Results with strip the package for the
original configuration
slender
slender web.
web.
with the same load-carrying capacity. Table 4 lists all the cross-sectional
dimensions of the original optimised specimens. It can be observed that the flange slenderness of the
optimised
Although
Although cross-section
thetheoptimised is approximately
optimised cross-section
cross-section 2.5 times
with with athat
a slender of the
web
slender original
delivers
web good specimen.
deliversresults
good Nevertheless,
from fromthis
an economical
results an
flange
point ofslenderness
economical view, the of
point remains
1-mm-thick
view, the in1-mm-thick
the noncompact
web of thisweb ofrange
optimal according
configuration
this optimal toisEquation is(7).
not practical.
configuration not OnTo the other
obtain
practical. To hand,
a obtain
better
the web slenderness
solution,
a better asolution, a of
second attempt the at
second optimised
optimising
attempt cross-section
at the is the
43%cross-sectional
cross-sectional
optimising ofdimensions
that of the original
was made
dimensions specimen.
bywas This
madeweb
confining the
by
slenderness
slenderness
confining the is slenderness
alsoof
value inboth
the noncompact
the
value webofplate
bothrange
the according
and plate to
the flange
web Equation
within
and thethe (8). within
The first
noncompact
flange thebuckling
range. mode
The
noncompact of the
harmony
range.
optimised
search
The harmony cross
process section
for
search this withfor
second
process noncompact
case
this can
secondbeweb casetogether
seen in Figure
can with
be seen the
7.inThe normal
outcome
Figure stress
7. The of distribution
this
outcome second can be
of thisattempt
second
seen
was in
attempt Figure
satisfactory. 8. The optimised
was satisfactory. cross-sectional
The optimised area wasarea
cross-sectional approximately
was approximately84% of that 84%ofofthe original
that of the
configuration with the same
original configuration with load-carrying capacity. Table
the same load-carrying 4 listsTable
capacity. all the4 cross-sectional dimensions
lists all the cross-sectional
of theTable
original
dimensions 4.of
Geometric
optimised
the original and mechanical It
specimens.
optimised properties ofItthe
can be observed
specimens. canoptimised
that
be the
observedcross-section
flange thewith
thatslenderness noncompact
flange web.of the
of the optimised
slenderness
cross-section is approximately
optimised cross-section 𝒇 [𝒎𝒎]2.5 times
𝒃𝒇 [𝒎𝒎] is 𝒕approximately 𝑫[𝒎𝒎]that of the original
2.5 𝒕times
𝒘 [𝒎𝒎] that𝒂[𝒎𝒎] of the specimen.
original Nevertheless,
specimen.
𝑴𝒄𝒓 [𝑵𝒎𝒎] this
Nevertheless,
𝝉𝒖 [𝑴𝑷𝒂] flange
A [𝒎𝒎 𝟐this
]
slenderness
flange
Experimental remains
slenderness 250in the
remains noncompact
in
15 the range
noncompact
800 according
range
2 to Equation
according
750 to (7).
Equation
1.1834 × 10 On the
(7).other
On
64.017 hand,
the otherthe web
hand,
9100
value
slenderness
the web of the optimised
slenderness cross-section
of the optimised is 43% of that
cross-section of the
is 43% oforiginal
that of the specimen.
originalThis web slenderness
specimen. This web
Optimised
is alsovalue
in the is
slenderness also374
noncompact 8.97 according
range
in the noncompact 399 2.31
rangetoaccording
Equation 750
(8).
to The first
Equation 1.1857 × 10 first
buckling
(8). The mode 64.029
of themode
buckling 7629
optimised
of the
Difference
cross section [%]with 66
noncompact −54 web −52
together with 14
the normal 0 stress
optimised cross section with noncompact web together with the normal stress distribution can be 0.19
distribution can 0.02
be seen in −16
Figure 8.
seen in Figure 8.

Table 4. Geometric and mechanical properties of the optimised cross-section with noncompact web.
𝒃𝒇 [𝒎𝒎] 𝒕𝒇 [𝒎𝒎] 𝑫[𝒎𝒎] 𝒕𝒘 [𝒎𝒎] 𝒂[𝒎𝒎] 𝑴𝒄𝒓 [𝑵𝒎𝒎] 𝝉𝒖 [𝑴𝑷𝒂] A [𝒎𝒎𝟐 ]
Experimental
250 15 800 2 750 1.1834 × 10 64.017 9100
value
Optimised
374 8.97 399 2.31 750 1.1857 × 10 64.029 7629
value
Difference [%] 66 −54 −52 14 0 0.19 0.02 −16

Figure 7.7.Variation
Figure Variationin in
thethe
minimum cross-sectional
minimum area with
cross-sectional area noncompact web in the
with noncompact webfirst
in 2000 harmony
the first 2000
search
harmonyiterations.
search iterations.
 Appl. Sci. 2020, 10, 3639 9 of 14

Table 4. Geometric and mechanical properties of the optimised cross-section with noncompact web.

bf [mm] tf [mm] D [mm] tw [mm] a [mm] Mcr [Nmm] τu [MPa] A [mm2 ]

Experimental value 250 15 800 2 750 1.1834 × 1011 64.017 9100
Optimised value 374 8.97 399 2.31 750 1.1857 × 1011 64.029 7629
Difference
Appl. Sci. 2020, 10, x FOR[%] 66
PEER REVIEW −54 −52 14 0 0.19 0.02 9 of 14−16

Appl. Sci. 2020, 10, x FOR PEER REVIEW 9 of 14

Figure Figure 8. Results

8. Results of the analysis
of the analysis withstrip
with finite finitemethod
strip method using
using the the software
software packagepackage
CUFSMCUFSM
for the for the
Figure 8.web.
noncompact Results of the analysis with finite strip method using the software package CUFSM for the
noncompact web.
noncompact web.
Figure 9 shows a finite element model with the Huber–von Mises–Hencky stress distribution of
Figure 9 shows a 9finite
Figure showselement
a finite model
elementwith
model the Huber–von
with Mises–Hencky stress distribution ofof
the optimised girder with the noncompact web. It the
canHuber–von
be observed Mises–Hencky
that, under the stress distribution
concentrated load at
the optimised girder with
the optimised the noncompact
girderbuckling
with the of web.
noncompact It can be observed
web. occurs
It can benear that,
observed under
that, the concentrated
underItthe load
the mid-span, the local the web plate the top flange. hasconcentrated load
been recommended
at the mid-span, the local the
at the mid-span, buckling of the of
local buckling web plate
web occurs near near
the thetop top
flange. It hashasbeen
that plate girders should be designed with athe plate occurs
span-to-depth ratio in the range offlange.
10–12 It been
[26]. This range
recommended that plate that
recommended girders
plateshould
girdersbe designed
should with awith
be designed span-to-depth ratioratio
a span-to-depth in the range
in the of of
range 10–12
10–12
for the a/D ratio corresponds to a recommended range for the angle of the panel diagonal (θd ) between
[26]. This
[26]. This ◦range range
for the a⁄for the a⁄corresponds
D ratio D ratio corresponds to a recommended
to a recommended rangerange for the
for the angleangle
of of
thethe panel
panel
4.76 and 5.71◦(𝜃. Considering thisand
requirement, the optimised plate girder is suitable for span lengths of
diagonal (𝜃diagonal
) between )4.76°
between
and4.76°
5.71°. 5.71°. Considering
Considering this requirement,
this requirement, the optimised
the optimised plate
plate girderisis
girder
up tosuitable
4.7 m. for span lengths of up to 4.7 m.
suitable for span lengths of up to 4.7 m.

Figure
Figure 9. Finite
9. Finite element
element modelofofthe
model theoptimised
optimised plate
plategirder
girderwith
withnoncompact web.web.
noncompact

In order
InFigure
order to extend
9.toFinite
extend thethe
element analysis
analysis
model totomore
of the morecases,
optimisedcases, and
and
plate to
tobetter
girder with understand
better understand
noncompact the effect
the
web. of moment
effect of moment
distribution on the structural performance, the harmony search process
distribution on the structural performance, the harmony search process is repeated under the assumption is repeated under the
Inoforderassumption
to extend
constant of constant distributed
the analysis
distributed loading to for
more loading for a cross-section
cases, and to with
a cross-section with
betteraunderstand a slender
slender web web
theand and
effect noncompact
of momentflange.
noncompact
distribution flange. Thestructural
moment gradient factor 𝐶 the is adjusted according to the new load case using under
Equation (4).
The moment gradient factor Cb is adjusted according to the new load case using Equation (4).the
on the performance, harmony search process is repeated In order
In order to incorporate the change in the value of 𝐶 , the minimum bending moment capacity
assumption of constant
to incorporate the distributed
change in the loading
value for
of Cab ,cross-section
the minimum with a slender
bending momentweb capacity
and noncompact
constraint was
constraint was also lowered accordingly. The outcome of the harmony search optimisation can be
flange.also
Thelowered
momentaccordingly.
gradient factor
seen in Figure The𝐶outcome
10. A minimum
is adjusted
of the
cross-sectional
according
harmony
area of 7250
tosearch
the new load case using
𝑚𝑚 optimisation
could be reached
Equation
canafter
be seen (4).
in Figure
347 harmony 10.
In order to
A minimumincorporate the
cross-sectional
search iterations. change in the value
area ofproperties
The geometric
2
of
7250 mmof could 𝐶 , the minimum
be reached
the optimised bending
after 347
cross-section moment
areharmony capacity
search
listed in Table iterations.
5. It can
constraint was
The geometricalso lowered
be observed that theaccordingly.
properties optimised webThe
of the optimisedplateoutcome
has of thein
cross-section
a decrease harmony
are listed
depth, search
anin Tableoptimisation
increase 5.
in It can be and
thickness canis be
observed less that
seen inthe
Figure 10. Athan
optimised
slender minimum
web theplate cross-sectional
has
original a decreasewhich
geometry, area
in of contrary
depth,
is 7250 𝑚𝑚to what
an increase could be observed
in was reached
thickness and after
inisthe347
less harmony
slender
load than the
case with
searchoriginal
iterations. The
concentrated geometric
geometry,load. properties of the optimised cross-section are listed
which is contrary to what was observed in the load case with concentrated load. in Table 5. It can
be observed that the optimised web plate has a decrease in depth, an increase in thickness and is less
slender than the original geometry, which is contrary to what was observed in the load case with
concentrated load.
 Appl.
Appl. Sci.Sci. 2020,
2020, 10,10, x FOR PEER REVIEW
3639 10 of
10 14
of 14
Appl. Sci. 2020, 10, x FOR PEER REVIEW 10 of 14

Figure 10. Variation in the minimum cross-sectional area with slender web under distributed loading
Figure
Figure10.
in the Variation
10.
first inin
Variation
600 harmonythe
theminimum
minimum
search cross-sectional
cross-sectional area
iterations. area with slenderweb
with slender webunder
underdistributed
distributed loading
loading
in the first
in the 600600
first harmony
harmony search
searchiterations.
iterations.
Table 5. Geometric properties of the optimised cross-section with slender web under distributed
Table 5. Geometric properties of the optimised cross-section with slender web under distributed loading.
Table
loading.5. Geometric properties of the optimised cross-section with slender web under distributed
loading. bf [mm] 2
𝒃𝒇 [𝒎𝒎] tf [mm] D [mm] 𝒕𝒘 [𝒎𝒎]
𝒕𝒇 [𝒎𝒎] 𝑫[𝒎𝒎] tw [mm]𝒂[𝒎𝒎] a [mm]
A [𝒎𝒎𝟐A] [mm ]
Experimental Experimental
value 𝒃𝒇 [𝒎𝒎] 𝒕15
250 𝒇 [𝒎𝒎] 𝑫[𝒎𝒎]
800 𝒕𝒘 [𝒎𝒎] 𝒂[𝒎𝒎] 750A [𝒎𝒎𝟐 ] 9100
250 15 800 2 2 750 9100
Optimised value
Experimental
value 263.83250 11.26 492.21
15 800 2 2.66 750 750 9100 7250
Difference value value 5.53263.83
Optimised
[%] 11.26
−25 492.21
−38 2.6633 750 0 7250 −20
Differencevalue
Optimised [%] 5.53
263.83 −25
11.26 −38
492.21 33
2.66 0
750 −20
7250
Difference [%] 5.53 −25 −38 33 0 −20
AsAscancanbe be
seen from
seen from thethe
presented
presented results,
results,the
theeffect
effectofofslenderness
slendernesson onthe
thestructural
structural performance
performance is
notisclearly
notAs discernable.
can
clearlybediscernable.
seen from Figure
the 11 shows
presented
Figure the results
results,
11 shows thethe of a parametric
effect
results ofofaslendernessstudy
parametric on that waswas
the that
study carried
structural out out
in order
performance
carried in
to visualise
order to visualise the effect of changing the web plate slenderness on the structural behaviour. Usingthe
is not the
clearly effect of
discernable. changing
Figure the
11 web
shows plate
the slenderness
results of a on the
parametric structural
study behaviour.
that was carriedUsing
out in
order
equations toinvisualise
the equations Section the
1.1,effect
in Section the
1.1,ofthe
changing
ultimate theshear
shear
ultimate stress τu was
web stress
plate 𝜏 calculated
slenderness on for
the different
was calculated structural behaviour.
values
for different values ofUsing
of web plate
web
the
plateequations
slenderness λw . in
slendernessEach 𝜆 curve
Section 1.1,
. Each the
incurveultimate
Figure shear
11 represents
in Figure stress 𝜏 was
a different
11 represents calculated
webfor
web plate
a different different
thickness,
plate values
while
thickness, of flange
the
while web
the
plate
flangeslenderness
slenderness slenderness
and flange 𝜆and
.thicknesses
Each
flangecurve in Figure
thicknesses
are 11 represents
are kept It
kept constant. can abedifferent
constant. It can be
observed web plate
observed
that, thickness,
that,web
at small whileweb
at small the
thicknesses,
flange slenderness
thicknesses,
increasing and
increasing
the web flange
the web
slenderness thicknesses
hasslenderness are
a favourablehas kept constant.
a favourable
effect It can be observed
effect shear
on the ultimate on thestress, that,
ultimate at small web
shearatstress,
whereas, higher
webthicknesses,
whereas,
thickness increasing
at higher
values, web the web has
thickness
slenderness slenderness
values, has aeffect.
slenderness
the opposite favourable effect effect.
has the opposite on the ultimate shear stress,
whereas, at higher web thickness values, slenderness has the opposite effect.

Figure
Figure 11.11.
TheTheultimate
ultimateshear
shearstress
stressvariation
variation for
for different
different values
valuesofofweb
webslenderness.
slenderness.
Figure 11. The ultimate shear stress variation for different values of web slenderness.
Appl. Sci. 2020, 10, 3639 11 of 14
Appl. Sci. 2020, 10, x FOR PEER REVIEW 11 of 14

Inaddition
In additionto toslenderness
slenderness andandthe
themoment
moment gradient
gradient factor,
factor,the
theeffect
effectof
ofchanging
changingthe theangle
angleofof
thepanel
the paneldiagonal
diagonalon onthetheultimate
ultimate shear
shear stress
stress isis also analyzed. The θθd values
also analyzed. valuesininaawide
widerange
rangethat
that
alsocontains
also containsthe
theangles
angles corresponding
corresponding to to the
the recommended
recommended range range for
for the a⁄D ratio
the a/D ratioare
areused
usedin inthe
the
calculationsfor
calculations forττu . .The
Theresults
resultsof
ofthis
thisanalysis
analysisarearevisualised
visualisedininFigure
Figure12.12.This
Thisvisualisation
visualisationclearly
clearly
showsthe
shows thenonlinear
nonlinear relationship
relationship between τu and
between τ θand
d . θ .
Figure Figure
12 shows 12 shows
that that
lowering θ
lowering
d has a θ has a
favourable
on 𝜏
effect on τu at small angles up to about 20 . On the other hand, increasing θd beyond this threshold
favourable effect at small angles up
◦ to about 20°. On the other hand, increasing θ beyond
this threshold
value value
only slightly τu . increases τ .
only slightly
increases

Figure 12. The variation of the ultimate shear stress with respect to the angle of the panel diagonal.
Figure 12. The variation of the ultimate shear stress with respect to the angle of the panel diagonal.
4. Conclusions
4. Conclusions
The lateral–torsional buckling and shear capacity of plate girders are significant factors in the
design Theof lateral–torsional
plate girders. The buckling
buckling andofshear
platecapacity
girders of used
plateduring
girders the construction
are stage ofinthe
significant factors the
replacement
design of plate project of the 102
girders. TheAvenuebuckling overofGroat
plateRoad Bridge
girders used in Edmonton,
during the Alberta, Canada
construction (Figure
stage 1)
of the
isreplacement
a recent example of the impact of these structural failure modes on infrastructures.
project of the 102 Avenue over Groat Road Bridge in Edmonton, Alberta, Canada (Figure In addition
to1)satisfy the serviceability
is a recent example of the requirements,
impact of these structures
structuralshould also
failure be designed
modes in an economical
on infrastructures. way.
In addition
Intothis study,
satisfy theitserviceability
was demonstrated that by adequately
requirements, structuresdimensioning
should also bethe cross-section
designed in an of a girder beam,
economical way.
its cost can be significantly reduced without compromising the structural
In this study, it was demonstrated that by adequately dimensioning the cross-section of a girder performance. The harmony
search
beam, algorithm
its cost can was
be utilised
significantlyto optimise
reduced thewithout
cross-sectional
compromisingdimensions under twoperformance.
the structural different setsThe of
slenderness
harmony search constraints.
algorithm In the was first set, thetoflange
utilised slenderness
optimise was kept in the
the cross-sectional noncompact
dimensions under range,
two
while
different sets of slenderness constraints. In the first set, the flange slenderness was kept in the
the web was allowed to be slender. In the second set, the web slenderness was constrained in the
noncompact
noncompactrange. range,In addition
while the web to satisfying
was allowed certain slenderness
to be slender. In requirements,
the second set, thethecross-sections
web slendernesswere
constrained
was constrained to have buckling
in the moments
noncompact and In
range. ultimate
addition shear stresses atcertain
to satisfying least asslenderness
high as therequirements,
moment and
ultimate shear stress capacity of the specimen tested by Mamazizi et al.
the cross-sections were constrained to have buckling moments and ultimate shear stresses at least as [14].
highThe
as theoptimisation
moment and under the first
ultimate shearsetstress
of slenderness
capacity of constraints
the specimen resulted
testedinby a cross-sectional
Mamazizi et al.area [14].
corresponding to 63% of the cross-sectional area of the experimental specimen.
The optimisation under the first set of slenderness constraints resulted in a cross-sectional area Despite this favourable
outcome,
correspondingthe 1-mm-thick
to 63% ofweb theofcross-sectional
this configuration areawas of not
the practical.
experimentalThe following
specimen. optimisation
Despite this
attempt
favourablewithoutcome,
the secondthe set1-mm-thick
of slenderness webconstraints resulted in a cross-sectional
of this configuration was not practical. area approximately
The following
84% of the cross-sectional
optimisation attempt with areatheofsecond
the experimental specimen.
set of slenderness These two
constraints outcomes
resulted in ashow that a suitably
cross-sectional area
optimised
approximately cross-section
84% of thecan cross-sectional
lead to great savings area of inthe
material and costs.
experimental However,These
specimen. withouttwoa outcomes
rigorous
optimisation procedure,
show that a suitably it is difficult
optimised to design can
cross-section a cross-section
lead to great with suitable
savings dimensions;
in material it was However,
and costs. observed
that
without a rigorous optimisation procedure, it is difficult to design a cross-section with Further
either increasing or decreasing the slenderness of the web can lead to a better design. suitable
research
dimensions;should be conducted
it was observed to better
that eitherunderstand
increasingthe or effects of thethe
decreasing flange and webof
slenderness slenderness
the web can on lead
the
structural performance. Furthermore, additional constraints should be placed
to a better design. Further research should be conducted to better understand the effects of the flange on the design variables
to satisfy
and webthe requirements
slenderness on the forstructural
the span-to-depth
performance.ratio. Furthermore, additional constraints should be
placed on the design variables to satisfy the requirements for the span-to-depth ratio.
Appl. Sci. 2020, 10, 3639 12 of 14

Author Contributions: Methodology, G.B.; formal analysis (coding), C.C.; writing—original draft preparation,
C.C. and G.B.; writing—review & editing, S.K. and Z.W.G.; visualisation, C.C.; supervision, G.B., S.K. and Z.W.G.;
funding acquisition, Z.W.G. All authors have read and approved the final manuscript.
Funding: This work was supported by the National Research Foundation of Korea (NRF) through a grant funded
by the Korea government (MSIT) (2020R1A2C1A01011131) and by the Energy Cloud R&D Program through the
National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT (2019M3F2A1073164).
Conflicts of Interest: The authors declare no conflict of interest.

Appendix A
Torsional section properties of a doubly symmetric I-section [49]. St. Venant’s torsion constant:
 
2bf t3f + D + tf t3w
J=
3
Warping constant:
 2
D + tf b3f tf
Cw =
24

References
1. Ziemian, R.D. Guide to Stability Design Criteria for Metal Structures; John Wiley & Sons: Hoboken, NJ,
USA, 2010.
2. Lee, S.C.; Yoo, C.H. Strength of plate girder web panels under pure shear. J. Struct. Eng. 1998, 124, 184–194.
[CrossRef]
3. Gupta, V.K.; Okui, Y.; Nagai, M. Development of web slenderness limits for composite I-girders accounting
for initial bending moment. Doboku Gakkai Ronbunshuu A 2006, 62, 854–864. [CrossRef]
4. Available online: https://edmonton.talks.news/groat-road-bridge/ (accessed on 24 February 2020).
5. Glassman, J.D.; Moreyra Garlock, M.E. A compression model for ultimate postbuckling shear strength.
Thin-Walled Struct. 2016, 102, 258–272. [CrossRef]
6. Basler, K. Strength of plate girders in shear. J. Struct. Div. 1961, 87, 151–180.
7. White, D.W.; Barker, M.G. Shear resistance of transversely stiffened steel I-girders. J. Struct. Eng. 2008, 134,
1425–1436. [CrossRef]
8. Wagner, H. Flat Sheet Metal Girders with Very Thin Metal Web. Part II: Sheet Metal Girders with Spars Resistant to
Bending-Oblique Uprights-Stiffness; NASA Technical Report; NASA: Washington, DC, USA, 1931.
9. Wilson, J.M. On specifications for strength of iron bridges. Trans. Am. Soc. Civ. Eng. 1886, 15, 389–414.
10. American Association of State Highway and Transportation Officials. AASHTO LRFD Bridge Design
Specifications, Customary US Units: 2006 Interim Revisions; American Association of State Highway and
Transportation Officials: Washington, DC, USA, 2006.
11. Porter, D.M.; Rockey, K.C.; Evans, H.R. The collapse behavior of plate girders loaded in shear. Struct. Eng.
1975, 53, 313–325.
12. Höglund, T. Shear buckling resistance of steel and aluminum plate girders. Thin-Walled Struct. 1997, 29,
13–30. [CrossRef]
13. Dubina, D.; Landolfo, R.; Ungureanu, V. Design of Cold-Formed Steel Structures: Eurocode 3: Design of Steel
Structures. Part 1–3, Design of Cold-Formed Steel Structures; European Convention for Constructional Steelwork
(ECCS): Bruxelles, Belgium, 2012.
14. Mamazizi, S.; Crocetti, R.; Mehri, H. Numerical and experimental investigation on the post-buckling behavior
of steel plate girders subjected to shear. In Proceedings of the Annual Stability Conference Structural Stability
Research Council, St. Louis, MI, USA, 16–20 April 2013.
15. Fujii, T. On an improved theory for Dr. Basler’s theory. In Final Report of 8th Congress of International Association
for Bridge and Structural Engineering(IABSE), New York; IABSE: Zürich, Switzerland, 1968; pp. 479–487.
16. Gaylord, E.H. Discussion of K. Basler Strength of Plate Girders in Shear. Trans. ASCE 1963, 128, 712.
17. Selberg, A. On the shear capacity of girder webs, International Association for Bridge and Structural
Engineering. Publications 1974, 34, 145–155.
Appl. Sci. 2020, 10, 3639 13 of 14

18. Timoshenko, S.P.; Gere, J.M. Theory of Elastic Stability; Courier Corporation: North Chelmsford, MA,
USA, 2009.
19. American Institute of Steel Construction (AISC). Steel Construction Manual, 14th ed.; American Institute of
Steel Construction (AISC): Chicago, IL, USA, 2011.
20. Helwig, T.A.; Frank, K.H.; Yura, J.A. Lateral-torsional buckling of singly symmetric I-beams. J. Struct. Eng.
1997, 123, 1172–1179. [CrossRef]
21. Wong, E.; Driver, R.G. Critical evaluation of equivalent moment factor procedures for laterally unsupported
beams. Eng. J. 2010, 47, 1.
22. Galambos, T.V.; Surovek, A.E. Structural Stability of Steel: Concepts and Applications for Structural Engineers;
Wiley Online Library: Hoboken, NJ, USA, 2008.
23. Subramanian, L.; White, D.W. Improved Noncompact Web-Slenderness Limit for Steel I-Girders. J. Struct. Eng.
2017, 143, 4016216. [CrossRef]
24. Estrada, I.; Real, E.; Mirambell, E. General behaviour and effect of rigid and non-rigid end post in stainless
steel plate girders loaded in shear. Part I: Experimental study. J. Construct. Steel Res. 2007, 63, 970–984.
[CrossRef]
25. Maquoi, R.; Skaloud, M. Stability of plates and plated structures: General report. J. Construct. Steel Res. 2000,
55, 45–68. [CrossRef]
26. Williams, A. Steel Structures Design ASD/LRFD; McGraw-Hill: New York, NY, USA, 2011.
27. Bekdaş, G.; Nigdeli, S.M.; Yang, X.-S. Sizing optimization of truss structures using flower pollination
algorithm. Appl. Soft Comput. 2015, 37, 322–331. [CrossRef]
28. Cheng, M.Y.; Prayogo, D.; Wu, Y.W.; Lukito, M.M. A Hybrid Harmony Search algorithm for discreet sizing
optimization of truss structure. Automat. Construct. 2016, 69, 21–33. [CrossRef]
29. Degertekin, S.O. Optimum design of steel frames using harmony search algorithm. Struct. Multidiscip. Optim.
2008, 36, 393–401. [CrossRef]
30. Bekdaş, G. Harmony search algorithm approach for optimum design of post-tensioned axially symmetric
cylindrical reinforced concrete walls. J. Optim. Theory Appl. 2015, 164, 342–358. [CrossRef]
31. 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. 2020,
10, 2301. [CrossRef]
32. 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. 2020, 10, 2976. [CrossRef]
33. 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. 2020, 10, 3232. [CrossRef]
34. De Almeida, F.S. Stacking sequence optimization for maximum buckling load of composite plates using
harmony search algorithm. Compos. Struct. 2016, 143, 287–299. [CrossRef]
35. Yu, Z.; Shi, X.; Zhou, J.; Chen, X.; Qiu, X. Effective Assessment of Blast-Induced Ground Vibration Using
an Optimized Random Forest Model Based on a Harris Hawks Optimization Algorithm. Appl. Sci. 2020,
10, 1403. [CrossRef]
36. Zhang, H.; Zhou, J.; Jahed Armaghani, D.; Tahir, M.M.; Pham, B.T.; Huynh, V.V. A Combination of Feature
Selection and Random Forest Techniques to Solve a Problem Related to Blast-Induced Ground Vibration.
Appl. Sci. 2020, 10, 869. [CrossRef]
37. Hajihassani, M.; Abdullah, S.S.; Asteris, P.G.; Armaghani, D.J. A Gene Expression Programming Model for
Predicting Tunnel Convergence. Appl. Sci. 2019, 9, 4650. [CrossRef]
38. Geem, Z.W.; Kim, J.H.; Loganathan, G.V. A new heuristic optimization algorithm: Harmony search. Simulation
2001, 76, 60–68. [CrossRef]
39. Xu, H.; Zhou, J.; Asteris, P.; Jahed Armaghani, D.; Tahir, M.M. Supervised Machine Learning Techniques to
the Prediction of Tunnel Boring Machine Penetration Rate. Appl. Sci. 2019, 9, 3715. [CrossRef]
40. Armaghani, D.J.; Asteris, P.G.; Fatemi, S.A.; Hasanipanah, M.; Tarinejad, R.; Rashid, A.S.A.; Huynh, V.V.
On the Use of Neuro-Swarm System to Forecast the Pile Settlement. Appl. Sci. 2020, 10, 1904. [CrossRef]
41. Chen, H.; Asteris, P.G.; Jahed Armaghani, D.; Gordan, B.; Pham, B.T. Assessing Dynamic Conditions of the
Retaining Wall: Developing Two Hybrid Intelligent Models. Appl. Sci. 2019, 9, 1042. [CrossRef]
42. Geem, Z.W. Particle-swarm harmony search for water network design. Eng. Optim. 2009, 41, 297–311.
[CrossRef]
Appl. Sci. 2020, 10, 3639 14 of 14

43. Cheng, Y.M.; Li, L.; Lansivaara, T.; Chi, S.C.; Sun, Y.J. An improved harmony search minimization algorithm
using different slip surface generation methods for slope stability analysis. Eng. Optim. 2008, 40, 95–115.
[CrossRef]
44. Vasebi, A.; Fesanghary, M.; Bathaee, S.M.T. Combined heat and power economic dispatch by harmony serach
algorithm. Int. J. Electr. Power Energy Syst. 2007, 29, 713–719. [CrossRef]
45. Gao, K.Z.; Suganthan, P.N.; Pan, Q.K.; Chua, T.J.; Cai, T.X.; Chong, C.S. Discreet harmony search algorithm for
flexible job shop scheduling problem with multiple objectives. J. Intell. Manuf. 2016, 27, 363–374. [CrossRef]
46. Tsakirakis, E.; Marinaki, M.; Marinakis, Y.; Matsatsinis, N. A similarity hybrid harmony search algorithm for
the Team Orienteering problem. Appl. Soft Comput. 2019, 80, 776–796. [CrossRef]
47. Geem, Z.W.; Lee, K.S.; Park, Y. Application of harmony search to vehicle routing. Am. J. Appl. Sci. 2005, 2,
1552–1557.
48. Geem, Z.W.; Sim, K.-B. Parameter-setting-free harmony search algorithm. Appl. Math. Comput. 2010, 217,
3881–3889. [CrossRef]
49. Kulak, G.L.; Grondin, G.Y. Limit States Design in Structural Steel; Canadian Institute of Steel Construction:
Alliston, ON, Canada, 2002; p. 63.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).