applied

sciences
Article
Heuristic Techniques for the Design of Steel-Concrete
Composite Pedestrian Bridges
Víctor Yepes 1 , Manuel Dasí-Gil 1 , David Martínez-Muñoz 1 , Vicente J. López-Desfilis 2 and
Jose V. Martí 1, *
1 Institute of Concrete Science and Technology (ICITECH), Universitat Politècnica de València,
46022 Valencia, Spain
2 Department. of Continuous Medium Mechanics and Theory of Structures, Universitat Politècnica de
València, 46022 Valencia, Spain
* Correspondence: jvmartia@cst.upv.es; Tel.: +34-96-387-7000; Fax: +34-96-387-7569




Received: 5 June 2019; Accepted: 4 August 2019; Published: 9 August 2019


Abstract: The objective of this work was to apply heuristic optimization techniques to a steel-concrete
composite pedestrian bridge, modeled like a beam on two supports. A program has been developed
in Fortran programming language, capable of generating pedestrian bridges, checking them, and
evaluating their cost. The following algorithms were implemented: descent local search (DLS),
a hybrid simulated annealing with a mutation operator (SAMO2), and a glow-worms swarm
optimization (GSO) in two variants. The first one only considers the GSO and the second combines
GSO and DLS, applying the DSL heuristic to the best solutions obtained by the GSO. The results
were compared according to the lowest cost. The GSO and DLS algorithms combined obtained
the best results in terms of cost. Furthermore, a comparison between the CO2 emissions associated
with the amount of materials obtained by every heuristic technique and the original design solution
were studied. Finally, a parametric study was carried out according to the span length of the
pedestrian bridge.

Keywords: pedestrian bridge; composite structures; optimization; metaheuristics; structural design

1. Introduction
Nowadays, society’s concern about the impact of activities is rising, not only their economic
influence, but also the environmental impact. The construction sector is one of the most carbon intensive
industries [1] due to the need for large amounts of materials and, henceforth, large amounts of natural
resources. Therefore, researchers are investigating how to achieve cost efficient and environmentally
sustainable processes for the construction industry. The term sustainable development was introduced
for the first time by the Brundtland Commission, defining it as, “development that meets the needs
of the present without compromising the ability of future generations to meet their own needs” [2].
Since then, countries have been raising awareness about the compromise to the future generations,
modifying their policies and demanding cheaper, ecofriendly constructions, without forgetting their
safety and durability. In essence, the demands of the governments are to reach solutions that reduce
their impact on the three main pillars: the economy, the environment, and society.
These demands translate into restrictions for constructors and designer. The former need to
carry out the constructions with new strategies to improve sustainability, while the designers have to
conceive their projects in a cheaper, ecofriendly way. This means profiting the materials and taking
maximum advantage of their characteristics, and maintaining durability and safety.
The traditional recommendation for the designers is to take a starting point for their designs.
Furthermore, there are lots of strict codes and regulations to ensure the safe and reliability of

Appl. Sci. 2019, 9, 3253; doi:10.3390/app9163253 www.mdpi.com/journal/applsci

Appl. Sci. 2019, 9, 3253 2 of 18

constructions, mostly on the structure field. There are many codes, which define bridge loads [3] and
concrete, steel, and composite bridge design [4–6]. In the same way, the European Committee for
Standardization (CEN) has advertised regulations related to structural bridge design [7–9] as has the
American Association of State Highway and Transportation Officials (AASHTO) [10]. In addition,
authors have been working to provide proper box-girder bridge designs [11]. This search for optimal
structures has led researchers to search for new forms of design, minimizing structural weight [12] and
economic cost [13].
Since the first relevant works carried out in the field of structural optimization [14,15], the interest
in the application of these techniques has grown a great deal, due to the different structural typologies
studied (steel structures [16], reinforced or pre-stressed concrete structures [17,18], or composite
structures [19–24]), as well as for the methods and algorithms used [25,26]. However, researchers
have focused, to a large extent, on the cost optimization of these types of structures. Some researchers
have stated that there is a relationship between the cost and environmental optimization, and cost
optimization is a good approach for the environmental one [27].
Optimization looks to find the values of the parameters that define the problem that allow us
to find the optimal solution. In the structural field, the problems tend to have too many variables to
analyze all the possible solutions. Therefore, the use of approximate methods that allow us to reach
optimal solutions have been studied. Research about heuristics and metaheuristics has been performed
recently [28], such as neural networks [29], Hybrid Harmony Search [30], genetic algorithms [31,32],
or simulated annealing [25,26]. Other authors have applied accelerated optimization methods, like
kriging [33], allowing to simplify the main structural problems.
Some researchers have already applied the multi-criteria optimization for bridge design [34],
considering other factors, besides the cost, like the security of the infrastructure and the CO2
emissions [27], the embodied energy [35], or the lifetime reliability [36]. However, these multi-criteria
methods have not been applied to composite bridges. Other researchers, such Penadés-Plà et al. [37],
have done a review of multi-criteria decision-making methods to evaluate sustainable bridge designs.
Nevertheless, if we focused on composite bridges there is a lack of knowledge.
In recent years, databases measuring the environmental impact of materials have been elaborated,
because of the importance of incorporating the design criteria to consider the impact on CO2
emissions [38–40]. Many researchers, such as Yepes et al. [41] and Molina-Moreno et al. [42], have
used these databases to study the difference between the cost and the CO2 emission optimization for
reinforced concrete (RC) structures.
The objective of this research was to study the differences between three heuristic optimization
techniques applied to a steel-concrete composite pedestrian bridge. Furthermore, the differences
between the CO2 emissions associated with the material amounts obtained from each heuristic were
analyzed and compared with the original structural design. Finally, a parametric study, according to
the span length, was performed.

2. Optimization Problem Definition

The problem proposed in this study is a single-objective optimization of a composite pedestrian
bridge. To reach this purpose, a program that determined the optimum values of the variables was
created. The objective was to minimize the objective function associated with the cost (1), satisfying
the constraints imposed on the problem, represented by the Equation (2):
rc
→  X → 
C x = pi ·mi x , (1)
i=1
→
G j x ≤ 0, (2)
Appl. Sci. 2019, 9, 3253 3 of 18

where vector x contains the design variables. To adjust the variables to real cases, these are discrete.
The total cost objective function is given in Equation (1), where pi is the prices of every construction
unit and mi the measurements obtained by the design variables. For example, for a random structure,
the vector x would contain the design variables of that random structure, mi would contain the amount
of materials associated with these variables, and these measurements, multiplied by their unit prices
(pi ), result in the total cost of the structure.
The values of cost for each variable contains the materials, labor and machinery. Because of the
characteristics of the construction materials, the maintenance was not taken into account in this study.
The construction place assumed in this paper was Valencia. The construction units included in this
study (rc ) were the volume of concrete, the amount of reinforcement steel, the amount of rolled steel,
and the amount of shear-connector’s steel.
The unit prices of the materials were taken from the price table of the College of Civil Engineers
of the Valencian Community for the year 2012. Furthermore, the CO2 emissions for each construction
unit were obtained from Molina-Moreno et al. [42]. The data of rolled steel and shear-connector steel
were taken to the BEDEC ITEC database of the Institute of Construction Technology of Catalonia [40].
Table 1 contains all the data on costs and CO2 emissions considered in this work.

Table 1. Prices and CO2 emissions.

Unit Measurements Cost (€) Emissions (kg CO2 )

m3 of concrete C25/30 93.71 224.34
m3 of concrete C30/37 102.41 224.94
m3 of concrete C35/45 105.56 265.28
m3 of concrete C40/50 111.64 265.28
kg of steel (B-500-S) 1.20 3.02
kg of shear-connector steel 2.04 3.63
kg rolled steel (S-355-W) 1.70 2.8
Concrete classification according to EN 1992

2.1. Design Variables

The structural solution was defined by the parameters and variables, the fixed and variable data,
respectively. In this work, the objective was to obtain an optimum steel-concrete composite pedestrian
bridge with a box-girder cross section of 38 m of span length, modeled like a beam on two supports.
The parameters are defined in Table 2. These values were considered fixed for the optimization.
The construction was carried out on the ground and then large tonnage cranes lifted the structure.
This constructive process made it possible to leverage the materials, because the steel and concrete
deflection beginnings were the same when the structure was put into service.
Figure 1 shows the deck cross section geometrical variables and the reinforcement. Table 3 shows
the limits defined for all the variables considered in this problem. The section is formed by two main
elements, one the one hand, the steel beam conformed to steel sheets, welded and bolstered with
longitudinal and transversal stiffeners. On the other hand, a reinforced concrete slab placed in the top
of the steel beam and connected to this element by steel shear-connectors. Optimization variables were
discrete, to bring the problem into line with reality. It was noted that some dimensions used may not
have been practical, but allowed the algorithm to visit feasible intermediate solutions to find better
optimal solutions. The total dimension of the problem was 1.67 × 1027 possible solutions, because of
this, the complete evaluation of the problem was unapproachable. The problem optimization was
carried out by heuristic techniques.
Appl. Sci. 2019, 9, 3253 4 of 18

Table 2. Parameters considered in the analysis.

Geometrical
Appl. Sci. 2019, 9, x FOR PEERParameters
REVIEW 4 of 20
Pedestrian bridge width B = 2.5 m
Number of spansvariation between steel and concrete
Termical ± 18 °C 1
Span length
Exposure related parameters 38 m
Material Parameters
External ambient conditions IIb
MaximumCodeaggregate size
related parameters 20 mm
Reinforcing steel
Code regulations EHE-08/IAP-11/RPX-95 B-500-S
Loading Parameters
Service working life 100 years

115 Reinforced concrete specific weight 25 kN/m3

2
116 Auxiliary
Figure 1 shows assembly triangulations
the deck cross self-weight
section geometrical 0.14 kN/m
variables and the reinforcement. Table 3 shows
117 the limitsLive Load
defined for all the variables considered in this problem. The section is 5 2
kN/mby two main
formed
118 elements,Dead
one the
loadone hand, the steel beam conformed to steel sheets, welded and
1.15 2
bolstered
kN/m with
119 longitudinal and transversal
Temperature stiffeners.
variation On thesteel
between otherand
hand, a reinforced concrete slab
concrete ◦ C in the top
±18placed
120 of the steel beam and connected to this element by steel shear-connectors. Optimization variables are
121 Exposure Related Parameters
discrete to bring the problem into line with reality. It is noted that some dimensions used may not be
122 External ambient conditions IIb
practical, but allow the algorithm to visit feasible intermediate solutions to find better optimal solutions.
123 The totalCode
dimension of Parameters
Related the problem is 1.67·10 possible solutions, because of this, the complete
27

124 evaluation of the

Code problem is unapproachable. The problem optimization
regulations has been carried out by
EHE-08/IAP-11/RPX-95
125 heuristicService
techniques.
working life 100 years
126
127

128
129 Figure 1. Box-girder geometrical variables and reinforcement
Figure 1. Box-girder geometrical variables and reinforcement.
130 Table 3. Box-girder geometrical variables description and values range
Table 3. Box-girder geometrical variables description
Concrete slab variables Range values
and values range.
Step size

CL
Concrete SlabVariables
Slab depth 15 to 30 cmRange Values 1 cm Step Size
CL CB Slab Slabdepth
edge depth 10 to 30 cm 15 to 30 cm 1 cm 1 cm
CB VL Slab edge depth
Lateral slab cantilever 0.5 to 0.625 m 10 to 30 cm 5 mm 1 cm
VL Lateral slab cantilever 0.5 to 0.625 m 5 mm
DT Transversal reinforcement diameter 6, 8, 10, 12 and 16 mm -
DT Transversal reinforcement diameter 6, 8, 10, 12, and 16 mm -
DLS DLS TopToplongitudinal reinforcement
longitudinal reinforcement diameter
diameter 6, 8,1610,
6, 8, 10, 12 and mm 12, and 16 mm - -
DLI DLI Bottom
Bottom longitudinal reinforcement
longitudinal reinforcement diameter
diameter 6, 8, 10, 12,6,
16,8,2010,
and12,
32 16,
mm20, and 32 mm
- -
SBTC Transversal reinforcement separation in span center 10 to 30 cm 1 cm
SBTC Tranversal reinforcement separation in span center 10 to 30 cm 1 cm
SBTA Transversal reinforcement separation in supports 5 to 25 cm 1 cm
NLS SBTA Tranversal reinforcement separation in supports
Number of top longitudinal reinforcement bars 5 to 25 cm 10 to 40 1 cm 1
NLI NLS Number
Number of bottom
of top longitudinal
longitudinal reinforcement
reinforcement bars bars 10 to 40 10 to 40 1 1
Metal BeamNumber
NLI Variables
of botom longitudinal reinforcement bars 10 to 40 1
CA Metal beam depth 1.086 to 2.375 m 1 cm
AAI Bottom flange width 1 to 1.5 m 1 cm
SRL Longitudinal stiffener spacing 0, 0.16, 0.26, 0.36, and 0.46 -
SRT Transversal stiffener spacing 1, 2, 3.8, 7.6, 38 m -
Appl. Sci. 2019, 9, 3253 5 of 18

Table 3. Cont.

EAS Top flange thickness 8 to 40 mm 2 mm

EAL Web thickness 8 to 40 mm 2 mm
EAI Bottom flange thickness 8 to 40 mm 2 mm
ERL Longitudinal stiffener thickness 8 to 40 mm 2 mm
ERT Transversal stiffener thickness 8 to 40 mm 2 mm
DP Shear-connectors diameter 16, 19, and 22 mm -
LP Shear-connectors length 100, 150, 175, and 200 -
STP Transversal shear-connectors spacing 5 to 25 cm 5 cm
SLPC Longitudinal shear-connectors spacing in mid span 30 to 50 cm 5 cm
SLPA Transversal shear-connectors spacing in supports 10 to 30 cm 5 cm
Mechanical Variables
FCK Concrete characteristic strength 20 to 35 MPa 5 MPa

2.2. Structural Analysis and Constraints

The structure was analyzed like a linear element. The model of the structure considered the shear
deformation and the effective flange width [6]. To obtain the stresses of the structure to check regulations
and recommendations constraints, a Fortran language program was implemented. This program
calculated the stresses in two sections: mid span and supports. In this structure, the highest stresses
occurred in these areas. Prior to the verifying limit states, the program needed to calculate stress
envelopes due to the loads. This program evaluated the stress envelopes due to a uniform load of
5 kN/m2 and the deck self-weight, including the bridge railing and asphalt (see Table 2). Note that
the thermal gradient [5] and the differential settling in each support were also taken into account.
The model implemented obtained the beam stresses and the transversal section tensions to assess the
structural design validity.
Once the stresses were obtained, a structural integrity analysis was performed. The ultimate
limit states (ULS) assessed the capacity of the structure against the flexure, shear, torsion, and the
combination of the stresses. It further considered the minimum reinforcements to resist the stresses
and the examination of the geometrical conditions. To evaluate the structural capacity, the regulations
employed to obtain the equations that allow the verification of the pedestrian bridge have been the
Spanish code on structural concrete [4], the Spanish recommendations for composite road bridge
project [6], and the code on the actions for the design of road bridges [3]. The serviceability limit states
(SLS) assessed the capacity of the structure to continue its service. The ULS considered in this study
were: bending, shear, torsion, shear–torsion interaction, and the stiffeners verification. It must be taken
into account that once the variables that define a frame solution have been chosen, then geometry,
the materials, and the passive reinforcement are defined. It should be noted that no attempt has been
made to calculate the reinforcement in such a way as to comply with common design rules. Such
common design procedures follow a conventional order for obtaining reinforcement bars from flexural
ULS, and then checking the SLS and redefining if required. While this order is effective, it ignores other
possibilities that heuristic search algorithms do not oversee. The seismic verifications are not necessary
due to the small value of the calculus acceleration for the location of the constructions. On the other
side, the SLS assessed were: deflections, vibrations, cracking, and web deformation. The vibration
limit state was verified in accordance with the restrictions for footbridges [4]. The SLS of cracking
included compliance with limitations of the crack width for existing durability conditions. With respect
to deflection, the instantaneous and time-dependent deflection was limited to 1/500th of the main span
length for the characteristic combination [4], and the frequent value for the live loads was limited to
1/1200th of the main span length [5]. Concrete and steel fatigue were not considered, as this ultimate
limit state is rarely checked in pedestrian bridges. In addition, the recommendations indicated in the
specialized bibliography [43–45] were considered.
The modulus implemented compared the structure model values with the values obtained from
the regulations equations. This modulus verified the demands of the safety, as well as those relating to
the aptitude for service requirement. Therefore, the limit states and the geometrical and constructability
Appl. Sci. 2019, 9, 3253 6 of 18

requirements must be guaranteed. The ULS checked that the ultimate load effects were lower than the
resistance of the structure, as seen in Equation (3):

RU ≥ Su , (3)

where Ru is the ultimate response of the structure and Su the ultimate load effects. For instance, the
ULS of the shear and torsion interaction reduced the shear resistance due to the effect of the torsion.
The SLS covers the requirements of functionality, comfort, and aspect (Equation (4)):
Appl. Sci. 2019, 9, x FOR PEER REVIEW 6 of 18
Cs ≥ Es , (4)
where Ru is the ultimate response of the structure and Su the ultimate load effects. For instance,
where the ULS of the shear
Cs is the permitted and torsion
value of the interaction reduced
serviceability the and
limit shearEresistance due to the effect of the
s is the value obtained from the model
torsion. The SLS covers the requirements of functionality, comfort, and aspect (Equation (4)):
produced by the SLS actions.
𝐶 ≥𝐸, (4)
3. Applied Heuristic
where Cs isSearch Methods
the permitted value of the serviceability limit and Es is the value obtained from the
model produced by the SLS actions.
3.1. Descent Local Search
3. Applied Heuristic Search Methods
This algorithm (Figure 2) begins by obtaining a random initial solution. Then, a small movement
is produced3.1.in
Descent Local Search
randomly chosen variables, increasing or decreasing them by a unit step. The algorithm
obtains the cost This algorithm
and (Figure 2) begins
the evaluation modulusby obtaining
check aifrandom initial solution.
the alternative Then,
fulfils thea constraints.
small movement If the cost of
is produced in randomly chosen variables, increasing or decreasing them by a unit step. The
the working solution is lower than the first and the new solution fulfills the restrictions, then it replaces
algorithm obtains the cost and the evaluation modulus check if the alternative fulfils the constraints.
the previous
If theone. This
cost of the process is continued
working solution is loweruntil nofirst
than the better
and solutions are found,
the new solution after
fulfills the a certain number
restrictions,
of iterations.
then it replaces the previous one. This process is continued until no better solutions are found, after a
certain number of iterations.

FigureFigure 2. Flowchart
2. Flowchart ofofthe
the descent
descent local search
local (DLS)
search process.
(DLS) process.
The movements set out in the study modified 1 to 25 variables. The number of iterations used
with no improvement were 100, 500, 1000, 2000, 5000, 10,000, 100,000, 500,000, 1 million, and 10
 Appl. Sci. 2019, 9, 3253 7 of 18

The movements set out in the study modified 1 to 25 variables. The number of iterations
used with no improvement were 100, 500, 1000, 2000, 5000, 10,000, 100,000, 500,000, 1 million, and
10 million. Figure 3 shows the average cost values according to the number of iterations stop criteria.
With movements of one, two, three, four, and five variables, the algorithm converged very quickly in
such a way that improvements were no longer achieved from 500, 5,000, 100,000, and 500,000 iterations,
respectively. Movements of more than five variables always improved the results, as the number of
iterations increased, starting to converge from 10,000 iterations. The most balanced evolution was
given toSci.
Appl. a 2019,
movement of five
9, x FOR PEER or six variables; given that it converged from 10,000 iterations.
REVIEW 8 of 20

196
197 Figure 3.
Figure 3. Average
Averagecosts
costsand
andnumber of iterations
number stop stop
of iterations criterion
criteria.

198 3.2.3.2. Hybrid

Hybrid SimulatedAnnealing
Simulated Annealing with
with aa Mutation
MutationOperator
Operator
199 TheThe hybrid
hybrid Simulated
simulated Annealing
annealing (SA)(SA)
withwith a mutation
a mutation operator
operator (SAMO2)
(SAMO2) is anisalgorithm
an algorithm
developed
200 by Martí
developed by Martí et al. [26] (Figure 4). This technique is used to combine the advantages of good
et al. [26] (Figure 4). This technique is used to combine the advantages of good convergence of
201 convergence of SA and the promotion of the diversity of the genetic strategy. SA, developed by
SA and the promotion of the diversity of the genetic strategy. SA, developed by Kirkpatrick et al. [46],
202 Kirkpatrick et al. [46], is based on the analogy of the thermodynamic behavior of a set of atoms to
203 is based
form on the analogy
a crystal. of the isthermodynamic
“Annealing” behavior
the chemical process of a set
of heating andofcooling
atoms atomaterial
form a crystal. “Annealing”
in a controlled
204 is the
fashion. Genetic algorithms seek the best solution through operators such as selection, crossover and seek
chemical process of heating and cooling a material in a controlled fashion. Genetic algorithms
205 the mutation.
best solution
Sokethrough
and Bingul operators
[32] havesuch as selection,
combined crossover,
effectively and mutation.
both algorithms. SAMO2Soke and Bingul
introduces the [32]
206 effectively combined
probabilistic both algorithms.
acceptance of poorer qualitySAMO2 introduces
solutions duringthe
theprobabilistic acceptance
process, allowing of poorer
it to escape fromquality
207 solutions
local optimums
during the and to finally
process, find highest
allowing quality from
it to escape solutions.
localTo do this, itand
optimums accepts worsefind
to finally solutions
the highest
208 quality
withsolutions.
a probability P , given by the expression of Glauber (5), where T is now a parameter
To do this, it accepts worse solutions with a probability Pa , given by the expression
a that
209 of Glauber
decreases(5),with the time and, thus, reducing the probability of accepting worse solutions,
where T is now a parameter that decreases with the time, thus reducing the probability from an
210 of accepting
initial value, T0.
worse solutions, from an initial value, T : 0

11
P𝑃a =
= ∆∆E . (5) (5)
1+
+𝑒e T

211
This method was applied using the following fixed variable movements: 2, 3, 4, 5, 6, and 12.
212 This method has been applied using the following fixed variable movements: 2, 3, 4, 5, 6 and 12.
213 TheThe
initial temperature was set by the method proposed by Medina [47]. Markov chain lengths of 5,000,
initial temperature was set by the method proposed by Medina [47]. Markov chain lengths of:
214 10,000, 20,000,
5,000, 10,000,and 30,000
20,000 andwere
30,000tested. In this
have been work,Inathis
tested. geometrical the typeofTthe
cooling of cooling
work, a geometrical = k·Ti was
i+1 type
215 adopted, considering k <
𝑇 = 𝑘 · 𝑇 has been adopted, considering k < 1, which has the advantage to prolong the final phase when
1, which has the advantage of prolonging the final phase of the search
216 the of
temperature
the search whenis low.
theThe coefficients
temperature k used
is low. were 0.80, k0.85,
The coefficients used0.90,
haveand been 0.95.
0.80,For the
0.85, stop
0.90 andcriteria,
0.95. two
217 wereForset
theinstop
thiscriteria,
study:two thatare
the
settemperature
in this study: was lesstemperature
that the than 0.001·T or than
is 0less that 0.001·T
during0 ora Markov
that during chain no
218 better
a Markov chain no
solution was found. better solution has been found.

219
 Appl. Sci. 2019, 9, x FOR PEER REVIEW
Appl. Sci. 2019, 9, 3253 8 of 18
8 of 18

Flowchart
Figure 4.Figure of the hybrid
4. Flowchart of thesimulated annealing
hybrid simulated with a mutation
annealing operatoroperator
with a mutation (SAMO2)(SAMO2)
process. process.

From the application of the SAMO2 algorithm, graphs of trajectories of the cost, according to
From the application of the SAMO2 algorithm, graphs of trajectories of the cost, according to the
the number of iterations or the time, were obtained, as in Figure 5. This figure shows the trajectory
number of iterations or the time, were obtained, as in Figure 5. This figure shows the trajectory of one
of one of the rehearsed processes, where a correct operation of the algorithm is appreciated, initially
of the rehearsed processes, where a correct operation of the algorithm is appreciated, initially
accepting high worsening, which decreases as the process progresses, focusing the search on solutions
accepting high worsening, which decreases as the process progresses, focusing the search on
with similar or lower costs, which divides the process into an initial diversification phase and a final
solutions with similar or lower costs, which divides the process into an initial diversification phase
intensification phase.
and a final intensification phase.
Appl. Sci. 2019, 9, 3253 9 of 18
Appl. Sci. 2019, 9, x FOR PEER REVIEW 9 of 18

Figure Trajectory
5. 5.
Figure Trajectorycost
costand
and temperature-iterations.
temperature-iterations.

3.3. Glow-worm Swarm

3.3. Glow-worm Optimization
Swarm (GSO)
Optimization (GSO)
The glow-worm
The glow-worm swarm
swarm optimization
optimizationalgorithm
algorithm mimics
mimics the thebehavior
behavior of of a firefly
a firefly swarm
swarm and and
was was
proposed
proposed by Krishnanand
by Krishnanand and andGhose Ghose
[48].[48]. Glow-worms
Glow-worms produce
produce a natural
a natural light
light that
that is is
usedusedasasa signal
a
signal
to attract to attractEach
a partner. a partner. Each glow-worm
glow-worm carries ancarries
amount anof amount of luminescence,
luminescence, which we whichwillwe callwill call
“luciferin”.
“luciferin”. that
It is considered It is considered
the maximum that the maximum
distance distance
at which thisatluminescence
which this luminescence
is perceived is perceived
is limited is by a
limited by a maximum radial value, which we will call the sensitivity radius rs. So, the decision range
maximum radial value, which we will call the sensitivity radius rsi . So, the decision rangei for each
for each glow-worm is also delimited by a maximum radial value r d, that complies with 0 < r d ≤ rs,
glow-worm is also delimited by a maximum radial value ri d , that complies with 0 < ri d ≤ rs , which we
which we will call a decision radius. One glow-worm i considers another firefly j as its neighbor if j
will call a decision
is within radius.
its decision One rglow-worm
radius id and the leveli considers
of luciferinanother thanjthat
firefly
j is greater as its
of neighbor
i. if j is within its
decision radius r i and the level of luciferin j is greater than that of i.
The r d decision
i d radius allows the selective interaction of neighbors and helps the disjointed
The ri d decision
formation radius allows
of sub-branches. the selective
Each firefly interaction
selects, through of neighbors
a probabilistic and helps
mechanism, the disjointed
a neighbor, who
formation
has a of sub-branches.
higher Each firefly
value of luciferin selects,
and moves through
towards it. aThese
probabilistic
movements,mechanism,
which areabased neighbor,
solelywho
on has
localvalue
a higher information and and
of luciferin the selective interaction
moves towards of neighbors,
it. These movements,allow which
the swarm of fireflies
are based solelytoon belocal
subdivided into disjointed subgroups, that address, and are found in multiple
information and the selective interaction of neighbors, allow the swarm of fireflies to be subdivided optimums of the given
multimodal
into disjointed function. The
subgroups, thatprocess
address,can and
be summarized
are found in as multiple
follows: optimums of the given multimodal
function. The process
1. Initially canofben summarized
a swarm as follows:
feasible glow-worms is generated and distributed in the search space. Each
glow-worm has assigned the initial luciferin value l0 and the initial sensitivity radius rs;
1. Initially a swarm
2. Depending onof n previous
the feasible glow-worms
luciferin li andisthegenerated
objective and distributed
function in luciferin
value, the the search space. Each
is updated
glow-worm has assigned the initial luciferin
as is shown on Equation 6. The luciferin value decaysvalue l 0 and constant 𝜌 (0 < 𝜌 < 1) simulates
the initial sensitivity radius rs ; the
2. decreaseon
Depending in the
luciferin level luciferin
previous over time,liandandthe theluciferin
objective enhancement constant
function value, theγluciferin
(0 < γ < 1)isisupdated
the
as isproportion
shown onofEquation
the improvement
(6). Theinluciferin
the objective
value that glow-worm
decays constant ρ to
adds < luciferin.
(0 its ρ < 1) simulates
J(xi(t)) is the
the value
decrease of the objective
in luciferin function
level over time,ofand
the the
glow-worm
luciferini at iteration j: constant γ (0 < γ < 1) is the
enhancement
proportion of the improvement 𝑙 (𝑡) =in(1the
− 𝜌) · 𝑙 (𝑡 − 1)
objective · 𝐽 𝑥 (𝑡) ; adds to its luciferin. J(xi (t))
+ 𝛾glow-worm
that (6)is the
value of the objective function of the glow-worm i at iteration j:
3. Each glow-worm uses a probability sampling mechanism to move towards a neighbor with a
higher luciferin value. For each glow-worm i, the probability of moving to a neighbor j is given
by Equation 7, where Ni(t) lisi (tthe
) =set − ρneighbors
(1 of + γ·J
)·li (t − 1) of the(xi (t));
glow-worm i in the iteration t, dij (6)
represents the Euclidean distance between glow-worms i and j in iteration t. rid(t) is the decision
3. Eachratio
glow-worm usesi ainprobability
of glow-worm iteration j: sampling mechanism to move towards a neighbor with a
higher luciferin value. For each glow-worm i, the probability of moving to a neighbor j is given by
𝑙 (𝑡) − 𝑙 (𝑡)
𝑝 (𝑡) = i (t) is the set;of 𝑁 (𝑡) , 𝑁 (𝑡)
𝑗 ∈neighbors 𝑑 (𝑡) < 𝑟 (𝑡)i in
= 𝑗:glow-worm (𝑡) <
; 𝑙 the 𝑙 (𝑡) ; t, dij represents
(7)
∑ ∈ N
Equation (7), where
( ) 𝑙 (𝑡) − 𝑙 (𝑡)
of the iteration
i
the Euclidean distance between glow-worms i and j in iteration t. r d (t) is the decision ratio of
glow-worm i in iteration j:

l j (t) − li (t) n o
pij (t) = P ; j ∈ Ni (t) , Ni (t) = j : dij (t) < rid (t) ; li (t) < l j (t) ; (7)
k∈Ni (t) lk (t) − li (t)
Appl. Sci. 2019, 9, 3253 10 of 18

4. During the movement phase, the glow-worm i moves to glow-worm j. Equation (8) describes the
model
Appl. Sci. 2019,of
9, xthe
FORmovement glow-worm at any given moment, where xi (t) is the location10
of a
PEER REVIEW ofofthe
18
glow-worm i at iteration t and s is the step factor constant:
4. During the movement phase, the glow-worm i moves to glow-worm j. Equation 8 describes the
x j (moment,
t) − xi (t) where xi(t) is the location of the
!
model of the movement of a glow-worm at any given
x (t + 1) = x (t) + s· ; (8)
glow-worm i at iteration t and is is the stepi factor constant:
kx j (t) − xi (t)k
𝑥 (𝑡) − 𝑥 (𝑡)
5. (𝑡 + 1) = the
Once the movement is𝑥 finished, 𝑥 (𝑡) + 𝑠 · of the radial sensor
update ; range is carried out by the (8)
𝑥 (𝑡) − 𝑥 (𝑡)
expression of Equation (9), where β is a constant parameter and nt is another parameter that
5. Once thethe
controls movement
number ofisneighbors:
finished, the update of the radial sensor range is carried out by the
expression of Equation 9, where β isn a constant n parameter  and  nt is ooanother parameter that
i i
d
( +
controls the number ofrneighbors:
t 1 ) = min rs , max 0, rd
( t ) + β· nt −  N i ( t )  . (9)

𝑟 (𝑡 + 1) = 𝑚𝑖𝑛 𝑟 , 𝑚𝑎𝑥 0, 𝑟 (𝑡) + 𝛽 · (𝑛 − |𝑁 (𝑡)|) . (9)

In this work, the GSO algorithm was applied to reach an optimum solution for a steel-concrete
In thispedestrian
composite work, thebridge.
GSO algorithm
The values wasof applied to reachused
the parameters an optimum solution
to apply this method forwere
a steel–concrete
0.5, 0.1, 0.5,
composite
2, 4 for ρ , γ, bridge.
0.25, andpedestrian β, nt , s, The
and values of the parameters
l0 , respectively. used tonumber
The maximum apply this method were
of iterations 0.5, 0.1,
was fixed at
0.5, 2,The
4000. 0.25, and 4offor
values 𝜌 , γ,
n and r0 β, nt, s,taken
were and las0, respectively. Thethe
different values; maximum numberinofthe
values adopted iterations was
study are fixed
shown
at Table
in 4000. 4.
The values of n and r0 were taken as different values; the values adopted in the study are
shown in Table 4.
Table 4. n and r0 adopted values.
Table 4. n and r0 adopted values.
n 10 20 30 40 50 60 80 100
r0 50 n 10
100 2015030 40 50 60 80 100
r0 50 100 150
In this work, the GSO algorithm was applied in two experiments. The first one only used the GSO
In this work, the GSO algorithm was applied in two experiments. The first one only used the
to reach the optimum solution, but in the second, the DLS algorithm was applied to the best solutions
GSO to reach the optimum solution, but in the second, the DLS algorithm was applied to the best
of the GSO to improve those solutions. Figures 6 and 7 show the results of the GSO and GSO with
solutions of the GSO to improve those solutions. Figures 6 and 7 show the results of the GSO and
DLS, respectively.
GSO with DLS, respectively.

Figure 6. Average cost to average iterations for the glow-worms swarm optimization (GSO) experiment.
Figure 6. Average cost to average iterations for the glow-worms swarm optimization (GSO)
experiment.
Appl. Sci. 2019, 9, 3253 11 of 18
Appl. Sci. 2019, 9, x FOR PEER REVIEW 11 of 18

Figure 7. Average cost to average iterations for GSO and DLS combination experiment.
Figure 7. Average cost to average iterations for GSO and DLS combination experiment.
4. Discussion
4. Discussion
4.1. Comparison of the Heuristic Techniques
4.1. Comparison of the Heuristic Techniques
To compare the results of the heuristic techniques, we focused on the cost obtained by each one.
To compare
In Table the results
5, the results of theamounts
of material heuristicandtechniques,
cost for we
eachfocused
solutiononare
theshown.
cost obtained by each that
The heuristic one.
In Table 5, the results of material amounts and cost for each solution are shown.
obtained the lowest cost is GSO combined with DLS. This result is related to the low amount of rolled The heuristic that
obtained
steel the lowest
achieved, due tocost
theisimportance
GSO combined with
of this DLS. This
material result is related
in steel-concrete to the low
composite amount of
pedestrian rolled
bridges.
steel achieved, due to the importance of this material in steel–concrete composite
As is seen in Table 5, the heuristics with lower values of cost (SA and GSO with DLS) have a lower pedestrian bridges.
As is seen
amount in Table
of rolled 5, the heuristics with lower values of cost (SA and GSO with DLS) have a lower
steel.
amount
The of rolled strength
concrete steel. was the same for DSL, SA, and GSO, but when GSO and DSL were combined,
The concrete
the geometry strength
variables of thewas
slabthe same for
decreased due DSL, SA,
to the and GSO,
increase in thebut when strength,
concrete GSO andleveraging
DSL were
combined,
the material. theThe
geometry
optimum variables
solutionof for
the aslab decreased
steel due toof
beam consists the increase
locating inarea
the the concrete strength,
of steel in a way
leveraging the material. The optimum solution for a steel beam consists of locating
that allows the mobilization of the highest possible mechanical arm. In order to reach the lowest cost, the area of steel
in a way that allows the mobilization of the highest possible mechanical arm.
the solutions obtained by the optimization algorithms looked for greater depths with lower amounts In order to reach the
lowest
of cost,increasing
material, the solutions obtained
the inertia andby the optimization
reducing the structurealgorithms
weight. looked for greater depths with
lower amounts of material, increasing the inertia and reducing the structure weight.
Table 5. Cost and material amount for the best heuristic solutions.
Table 5. Cost and material amount for the best heuristic solutions.
DLS SA GSO GSO and DLS
DLS SA GSO GSO and DLS
Rolled steel kg/m2 153.99 151.26 157.49 150.66
Rolled steel
% Rolled/record kg/m
%
2 153.99
2.21% 151.26
0.40% 157.49
4.54% 0.00% 150.66
% Rolled/record
Shear-connector steel kg/m% 2 2.21%
0.90 0.40%
0.90 4.54%
0.79 0.90 0.00%
% Shear-connector/record
Shear-connector steel %
kg/m 2 14.28%
0.90 14.28%
0.90 0.00% 0.79 14.28% 0.90
Concrete
% Shear-connector/record m3 %/m2 0.15
14.28% 0.15
14.28% 0.140.00% 0.14 14.28%
% Concrete/record % 7.52% 7.52% 0.83% 0.00%
Concrete m3/m2 0.15 0.15 0.14 0.14
Reinforcement steel kg/m2 22.57 22.22 25.50 22.23
% Concrete/record
% Reinforcement/record %% 7.52%
1.58% 7.52%
0.00% 0.83%
14.76% 0.04% 0.00%
Reinforcement
Cost steel kg/m
€/m2 2 22.57
304.82 22.22 313.18
299.77 25.50 297.76 22.23
% Cost/record
% Reinforcement/record %% 2.37%
1.58% 0.67%
0.00% 5.18% 14.76% 0.00% 0.04%
Cost The rows with percentages express the €/m 2 304.82
increase in the quantity of299.77
material with313.18 297.76
respect to the minimum.
% Cost/record % 2.37% 0.67% 5.18% 0.00%
The rows with percentages express the increase in the quantity of material with respect to the minimum.
Appl. Sci. 2019, 9, 3253 12 of 18

4.2. Sustainability Study

An analysis of the CO2 emissions associated with the amount of the materials obtained from
every cost heuristic optimization was carried out. In addition, a comparison with the original project
of this steel-concrete composite pedestrian bridge was performed. In Table 6, the values of cost and
CO2 emissions of every solution are compared.

Table 6. CO2 emissions and cost data comparison from the reference solution and the heuristics.

Reference DLS SA GSO GSO and DLS

Cost €/m2 399.10 304.82 299.77 313.18 297.76
% Cost/reference % - −23.62% −24.89% −21.53% −25.39%
CO2 emissions kg CO2 /m2 700.22 536.39 527.70 552.54 523.68
% Emissions/reference % - −23.40% −24.64% −21.09% −25.21%

As is seen in this table, the GSO and DLS combination heuristic obtained a reduction of 8.12% of
the CO2 emissions compared with the reference. This means that an improvement of 1 €/m2 produced
a reduction of the 1.74 kg CO2 /m2 .

4.3. Parametric Study

A parametric study for varying span lengths is presented with the GSO and DLS combination
optimization model. Five span lengths were considered: 28, 32, 38, 42 and 48 m. The characteristics that
were studied are the economy, the geometry, and the amount of materials. Tables 7 and 8 compile the
values of the features of the optimization solutions: Table 7 gives the main values of the geometry of the
structure, and Table 8 gives the values of the measurements of the amount of materials of the structure.

Table 7. GSO with DLS combination for 28, 32, 38, 42, and 48 m spans.

Span CL CA CT EAS EAL EAI FCK Total

(m) (m) (m) (m) (mm) (mm) (mm) (MPa) Depth/L
28 0.17 1.53 1.70 18 8 12 35 0.035
32 0.16 1.37 1.53 18 8 10 30 0.036
38 0.15 1.21 1.36 18 8 10 30 0.036
42 0.15 0.96 1.11 18 8 10 25 0.035
48 0.15 0.80 0.95 18 8 10 25 0.034

Table 8. GSO with DLS combination measurements of the materials for 28, 32, 38, 42, and 48 m spans.

Span (m) Beam Rolled Steel (kg/m2 ) Slab Concrete (m3 /m2 ) Slab Reinforcement (kg/m2 )
28 194.10 0.17 30.87
32 165.64 0.15 24.63
38 150.66 0.14 22.23
42 135.37 0.14 19.80
48 126.56 0.14 17.84

The results of the parametric study led to practical rules for the preliminary design of
cost-optimized steel-concrete composite pedestrian bridges with box-girder cross sections isostatic
spans. The discussion of the results was carried out together, with a regression analysis. The functions
obtained were valid approximations within the range of the studied parameters. The extrapolation of
these results to other span lengths should be carried out carefully.
Figure 8 shows the average results of the cost of the structure per square meter of the steel-concrete
composite pedestrian bridge for distinct span lengths. The cost evolution as a function of the horizontal
span leads to a very good quadratic correlation. The average footbridge cost adjusted to C = 0.1954L2
− 8.0873L + 325.96 with a regression coefficient of R2 = 0.9994. The cost rising is produced by the need
 Appl. Sci. 2019, 9, x FOR PEER REVIEW 14 of 20
Appl. Sci. 2019, 9, x FOR PEER REVIEW 13 of 18

323 Figure8 8shows

Figure showsthe theaverage
averageresults
resultsofofthethecostcostofofthe
thestructure
structureperpersquare
squaremeter
meterofofthe the steel-
steel–
324 concrete compositepedestrian
concrete pedestrian bridgefor fordistinct
distinctspanspan lengths.The Thecost
costevolution
evolutionasasa afunction
functionofofthe the
Appl. Sci. composite
2019, 9, 3253 bridge lengths. 13 of 18
325 horizontal span leads to a very good quadratic correlation. The average footbridge cost adjusted to C =
horizontal span leads to a very good quadratic correlation. (The average footbridge cost adjust to C
326 = 0.1954L
0.1954L2 –− 8.0873L
2
8.0873L ++325.96
325.96with
withaaregression
regressioncoefficient
coefficientofofRR=2 =0.9994.)
2
0.9994.The
Thecost
costrising
risingisisproduced
producedby
327 the
for
by needs
larger
the of
forlarger
needamounts largerofamounts
materials
amounts of
oftomaterials to
satisfy the
materials satisfy the
to deflection
satisfy therequirements.
deflection requirements.
deflection requirements.
Note that the Note
Note
2 that
R that theR2R2
regression
the
328 regressioncoefficient
coefficient
regression coefficient
in Figure 8inisinFigure
Figure81,8isthis
almost isalmost
almost 1,1,this
indicates this indicates
a indicates
very avery
verygood
goodcorrelation.
gooda correlation. correlation.
The variations The
The variations
between
variations the
329 between
minimum
between the
theand minimum
the mean
minimum and
andcost the mean
theofmean cost
the pedestrian of the
cost of thebridge pedestrian
produced
pedestrian bridge produced
bridgebyproduced
the GSO by by
andthethe
DLS GSO
GSO and
combination
and DLS DLS
330 combination
are 0.42%. are
combination are 0.42%. 0.42%.
331

332
333 Figure
Figure8. Average cost
Figure8.8.Average
Averagecost
for different
costfor
fordifferent
span lengths.
differentspan
spanlengths.
lengths.
334 Figure 9 shows the mean values of the depth of the steel beam (CA) for different span lengths.
Figure 99shows the
themean values ofofthe depth
depthofofaccording
the
thesteel
steelbeam (CA) for different span lengths.
335 The Figure
depth of shows
the beam mean
has values
a good linear the
variation beam
to the (CA)
span for different
length of thespan lengths.
bridge. (The
The
Thedepth
depth of the
of beam
the beam has a
has good
a linear
good variation,
linear according
variation, to
according the
to span
the length
span of the
length bridge.
of the The
bridge.
336 average depth of the beam adjusts to CA = 0.0369L – 0.2179 with R = 0.9927) Again, the good
2

337 average
The depthdepth
average
correlation
of the
of beam adjusts
the beam
factor represents
to CA
adjusts
a functional CA = 0.0369L
= 0.0369L
torelation. − 0.2179
− 0.2179 with
with R2R=2 =0.9927.
0.9927.Again,
Again,the
thegood
good
correlation
correlationfactor
factorrepresents
represents aafunctional
functional relationship.
relationship.
338

339
340 Figure 9. Mean steel beam depth for different span lengths.

341 As shown in Figure Figure

Figure
10, 9. Meansteel
a9.function
Mean steel beamdepth
isbeam
found depth fordifferent
for
relating different spanlengths.
span
the thickness lengths.
of the slab with different span
342 lengths. Up to a certain span length, the slope of the parabola is smaller because the inertia of the slab
Asshown
As shownin in Figure
Figure 10,
10, aa function
function was
was found
found relating thethethickness of the slabslab
withwith
different span
343 is determined by the transverse flexion. Once the relating
slab stressesthickness of the
are determined by the different
longitudinal
lengths. Up to a certain span length, the slope of the parabola is smaller because the inertia of the slab
344 span lengths.the
deflection, Upslope
to a certain
of the span
curvelength, the slope
increases. of the parabola
(The average is smaller
slab thickness because
adjusts the inertia
to CL= 0.0001Lof2 –
is determined by the transverse flexion. Once the Once
slab stresses are determined by the longitudinal
345 the slab is
0.0067L determined
+ 0.2568 with R2by the transverse
= 0.9849 when the span flexion. the slab
length is larger thanstresses are determined
38 m). Related by the
to the compressive
deflection, the slope of the curve increases. The average slab thickness adjusts to CL = 0.0001L 2
to CL =−
346 longitudinal deflection,
strength of the concrete2the slope
of the of in
slab, theFigure
curve11 increases.
the relationThetoaverage slab thickness
the concrete compressiveadjusts
characteristic
0.0067L2 +
0.0001L − 0.2568
0.0067Lwith R = 0.9849
+ 0.2568 with Rwhen the span
2 = 0.9849 whenlength is larger
the span than
length 38 m. Related
is larger than 38 to
m.the compressive
Related to the
strength of the concrete of the slab, Figure 11 shows the relationship of the concrete compressive
characteristic strength and the span length. This relationship adjusts well to a quadratic function.
 Appl. Sci. 2019, 9, x FOR PEER REVIEW 14 of 18
Appl. Sci. 2019, 9, 3253 14 of 18
compressive strength of the concrete of the slab, Figure 11 shows the relationship of the concrete
Appl. Sci. 2019, 9, x FOR PEER REVIEW 15 of 20
compressive characteristic strength and the span length. This relationship adjusts well to a quadratic
The concreteconcrete
function. compressive strength to FCKto
adjustsadjusts = FCK 2 − 1.6959L
0.0292L0.0292L + 49.714 with R 2 = 0.9987.
strengthThe and the spancompressive strength has a =good adjust− to
1.6959L + 49.714 with R(The
2 2 =
347 length is shown. This relation a quadratic function
Note
0.9987. that
Note the highest concrete
that the highest compressive
concrete strength
compressive considered
strength for
considered this study was 35 MPa.
348 concrete compressive strength adjusts to FCK = 0.0292L 2 – 1.6959L + for thiswith
49.714 studyRwas 35 MPa.Note
2 = 0.9987).

349 that the highest concrete compressive strength considered for this study is 35 MPa.

350
351 Figure
Figure
Figure 10. 10.
10.MeanMean
Mean slabslab
slab thickness
thickness
thickness for different
fordifferent
for different span
span
span lengths.
lengths.
lengths.

352
353 Figure 11. Average compressive strength for different span lengths.
Average compressive strength
Figure
Figure 11.11.
Average compressive strength forfor different
different span
span lengths.
lengths.
354 Regarding ratio of the amount of rolled steel (Rs) and the surface of the slab (Ss); Figure 12
355 Regarding
illustrates
Regarding thethe
the increasing ratio
ratio ofofofthe
the the amount
amount
amount ofofof rolled
rolled
rolled steel
steel
steel (R s ) and
needed
(R s) and to
the
theresist surface of of
thethe
the flexural
surface slab
(Ss(S s ) Figure
requirements.
slab ) Figure The
12 12
356 illustrates
slope
illustrates of thethe increase
curve
increase tensinintothe
the amountasof
increase
amount ofthe rolled
span
rolled steel
length
steel needed totoresist
increases
needed thethe
(The
resist flexural
mean requirements.
amount
flexural of rolledThe
requirements. slope
steel
The in
357 of the
relation
slope curve
of theto curvetends
the surface to increase
tendsoftoslab as the
adjusts
increase span
asto the Rs/Slength increases.
s = 0.01913L
span The mean amount of rolled
2 – 3.5553L + 154.96 with R2 = 0.9996). Although,
length increases. The mean amount of rolled steel in steel in relation
s /Sthe =surface ++ 154.96 with RR22==0.9996.
Rto 2
358 to the
the
relationratiosurface
to of volume
the of slab
surface adjusts
concrete
of slab (Vto
c) and
adjusts Rs s/S 0.01913L
s = 0.01913L slab−2fits
−3.5553L
a second
3.5553L order with
154.96 equation 0.9996.
that increasesHowever,
However, with
359 the the
the ratio
span
ratio of the
of length
the volumevolume
in theofsameof concrete
concreteway (V (V c
asc)rolled)
and theand the
steel surface
amount
surface slab
slabasfits fits
it seen a
a secondsecond
in Figure order
order13 equation
(The mean
equation that
that ratio increases
of the
increases
360 withwiththethe
volume ofspan
span length
concrete
length in in
inthe the
relation
same same way
to the
way as rolled
assurface
rolled steel
of slab
steel amount,
adjusts
amount, asto as
seenVcseen
/S =in0.0001L
ins Figure Figure
13.2The 13. The mean
– 0.0067L
mean ratio
+ 0.245).
ratio of
c /Ssof=slab same+
2
361 the ofvolume
the volume
Moreover, ofthe of concrete
ratio
concrete in relation
ofinreinforcing
relation steel
to the to thes)surface
(RF
surface of slabofadjusts
measured slab
peradjusts
to Vc/Sto
square =V
smeter
0.0001L 20.0001Lshows− +the
− 0.0067L 0.0067L
0.245.
362 0.245. Moreover,
tendency
Moreover, the ratiothe
as rolled of ratio
steel of reinforcing
amount
reinforcing and
steel steel
concrete
(RF (RF s ) measured
volume
s) measured (The per of
per ratio
square square meter
reinforced
meter of
steel
of slab slabinshows
shows thethe
relation tosame
same the
363 tendency as rolled steel amount and concrete volume. The ratio of reinforced steel in relation to thethe
tendency
surface of as
slab rolled steel
adjusts to amount
RF s / S s =and
0.0246L concrete volume. The ratio of reinforced
2 – 1.2009L + 32.516 with R2 = 0.9955), it shows in figure steel in relation to 14
s /Ss = 0.0246L − 1.2009L + 32.516 with R = 0.9955, as shown in Figure 14.
2 2
surface
surface of of slab
slab adjusts
adjusts totoRFRF s/Ss = 0.0246L 2 − 1.2009L + 32.516 with R2 = 0.9955, as shown in Figure 14.
Appl. Sci. 2019, 9, 3253 15 of 18
Appl. Sci. 2019, 9, x FOR PEER REVIEW 15 of 18
Appl. Sci. 2019, 9, x FOR PEER REVIEW 15 of 18
Appl. Sci. 2019, 9, x FOR PEER REVIEW 15 of 18

Figure 12. Variation

Figure ininthe
12. Variation theratio ofthe
ratio of therolled
rolled steel
steel amount
amount in relation
in relation to the
to the surface of surface of the
the slab with the slab
span.with
Figure 12. Variation in the ratio of the rolled steel amount in relation to the surface of the slab with the span.
the span.
Figure 12. Variation in the ratio of the rolled steel amount in relation to the surface of the slab with the span.

Figure 13. Variation in the ratio of the volume of concrete in relation to the surface of the slab with the span.
13. Variation
FigureFigure ininthe
13. Variation theratio ofthe
ratio of thevolume
volume of concrete
of concrete in relation
in relation to theofsurface
to the surface of the
the slab with theslab
span.with
Figure 13. Variation in the ratio of the volume of concrete in relation to the surface of the slab with the span.
the span.

Figure 14. Variation in the ratio of the reinforcement steel in relation to the surface of the slab with the span.
Figure 14. Variation in the ratio of the reinforcement steel in relation to the surface of the slab with the span.
Figure 14. Variation
Figure ininthe
14. Variation theratio ofthe
ratio of the reinforcement
reinforcement steel steel in relation
in relation to the
to the surface of surface of the
the slab with the slab
span.with
the span.
Appl. Sci. 2019, 9, 3253 16 of 18

5. Conclusions
Three heuristic algorithms were applied to a steel-concrete composite pedestrian bridge to find an
efficient design. DLS, SA, and GSO are used to automatically find optimum solutions. All procedures
are useful in the automated design of steel-concrete composite pedestrian bridges. Furthermore,
a parametric study was carried out. The conclusions are as follows:

• The GSO optimization algorithm obtained worse results than DLS and SA, but if we apply the
DLS to the best GSO solutions, then this combination of heuristic techniques reaches the lowest
cost solution;
• The results show the potential of the application of heuristic techniques to reach automatic designs
of composite pedestrian bridges. The reduction in costs exceeds 20%. It is important to note that
the current approach eliminates the need for experience-based design rules;
• The CO2 emission comparison showed that the reduction between the original structure and the
GSO with DLS combination optimized structure reduced the CO2 emission by 21.21%. Thus,
an improvement of 1 €/m2 produced a reduction of 1.74 kg CO2 /m2 . Therefore, the solutions that
are acceptable in terms of CO2 emissions are also viable in terms of cost, and vice versa;
• The results indicate that cost optimization is a good approach to environmentally friendly design,
as long as cost and CO2 emission criteria reduce material consumption;
• The parametric study showed that there is a good correlation between span length and cost,
amount of material, and geometry. This relationship could be useful for designers, to have a guide
to the day-to-day design of steel-concrete composite pedestrian bridges. However, the tendencies
of the thickness of the flanges and webs of the steel beam are not clear;
• The heuristic techniques look for lower amounts of materials, which allows the reduction of the
self-weight of the structure. In addition, the optimization algorithms look for an increase of the
depth of the section to improve their mechanical characteristics. For an optimized pedestrian
bridge, the relationship between the steel beam depth and span length takes a value of 1/27.

Author Contributions: This paper represents a result of teamwork. The authors jointly designed the research.
D.M.-M. drafted the manuscript. V.Y., M.D.-G., J.V.M., and V.J.L.-D. edited and improved the manuscript until all
authors are satisfied with the final version.
Acknowledgments: The authors acknowledge the financial support of the Spanish Ministry of Economy and
Business, along with FEDER funding (DIMALIFE Project: BIA2017-85098-R).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Liu, S.; Tao, R.; Tam, C.M. Optimizing cost and CO2 emission for construction projects using particle swarm
optimization. Habitat Int. 2013, 37, 155–162. [CrossRef]
2. World Commission on Environment and Development. Our Common Future; Oxford University Press:
Oxford, UK, 1987.
3. Ministerio de Fomento. Instrucción Sobre las Acciones a Considerar en el Proyecto de Puentes de Carretera (IAP-11);
Ministerio de Fomento: Madrid, Spain, 2011.
4. Ministerio de Fomento. Instrucción de Hormigón Estructural (EHE-08); Ministerio de Fomento: Madrid,
Spain, 2008.
5. Ministerio de Fomento. Instrucción de Acero Estructural (EAE-11); Ministerio de Fomento: Madrid, Spain, 2011.
6. Ministerio de Fomento. Recomendaciones Para el Proyecto de Puentes Mixtos de Carretera (RPX-95); Ministerio
de Fomento: Madrid, Spain, 1995.
7. European Committee for Standardization. Eurocode 1: Actions on Structures-Part 1-4: General Actions-Wind
Actions; European Committee for Standardization: Brussels, Belgium, 2005.
8. European Committee for Standardlzation. BS EN 1992-2: 2005—Eurocode 2: Design of Concrete
Structures—Part 2: Concrete Bridges—Design and Detailing Rules; European Committee for Standardization:
Brussels, Belgium, 2005.
Appl. Sci. 2019, 9, 3253 17 of 18

9. European Committee for Standardization. CEN/TC250 EN1993: Eurocode 3 Design of Steel Structures—Part 2:
Steel Bridges; European Committee for Standardization: Brussels, Belgium, 2006.
10. Aashto. AASHTO LRFD Bridge Design Specifications; American Association of State Highway and
Transportation Officials: Washington, DC, USA, 2012.
11. Ministerio de Fomento. New Overpasses: General Concepts; Ministerio de Fomento: Madrid, Spain, 2000.
12. Li, L.J.; Huang, Z.B.; Liu, F. A heuristic particle swarm optimization method for truss structures with discrete
variables. Comput. Struct. 2009, 87, 435–443. [CrossRef]
13. Martí, J.V.; Yepes, V.; González-Vidosa, F. Memetic algorithm approach to designing precast-prestressed
concrete road bridges with steel fiber reinforcement. J. Struct. Eng. 2015, 141, 04014114. [CrossRef]
14. Goldberg, D.E.; Samtani, M.P. Engineering Optimization via Genetic Algorithm; ASCE: Leston, VA, USA, 1991.
15. Cohn, M.Z.; Dinovitzer, A.S. Application of Structural Optimization. J. Struct. Eng. 1994, 120, 617–650.
[CrossRef]
16. Sarma, K.C.; Adeli, H. Cost optimization of steel structures. Eng. Optim. 2000, 32, 777–802. [CrossRef]
17. Sarma, K.C.; Adeli, H. Cost optimization of concrete structures. J. Struct. Eng. 1998, 124, 570–578. [CrossRef]
18. Torres-Machi, C.; Yepes, V.; Alcalá, J.; Pellicer, E. Optimization of high-performance concrete structures by
variable neighborhood search. Int. J. Civ. Eng. 2013, 11, 90–99.
19. Adeli, H.; Kim, H. Cost optimization of composite floors using neural dynamics model. Commun. Numer.
Methods Eng. 2001, 17, 771–787. [CrossRef]
20. Kravanja, S.; Šilih, S. Optimization based comparison between composite I beams and composite trusses.
J. Constr. Steel Res. 2003, 59, 609–625. [CrossRef]
21. Senouci, A.; Alansari, M. Cost Optimization of Composite Beams Using Genetic Algorithms. Adv. Eng. Softw.
2009, 40, 1112–1118. [CrossRef]
22. Kaveh, A.; Abadi, A.S.M. Cost optimization of a composite floor system using an improved harmony search
algorithm. J. Constr. Steel Res. 2010, 66, 664–669. [CrossRef]
23. Ramires, F.B.; De Andrade, S.A.L.; Da Silva Vellasco, P.C.G.; De Lima, L.R.O. Genetic algorithm optimization
of composite and steel endplate semi-rigid joints. Eng. Struct. 2012, 45, 177–191. [CrossRef]
24. Kaveh, A.; Behnam, A.F. Cost optimization of a composite floor system, one-way waffle slab, and concrete
slab formwork using a charged system search algorithm. Sci. Iran. 2012, 19, 410–416. [CrossRef]
25. Yepes, V.; Alcala, J.; Perea, C.; González-Vidosa, F. A parametric study of optimum earth-retaining walls by
simulated annealing. Eng. Struct. 2008, 30, 821–830. [CrossRef]
26. Martí, J.V.; Gonzalez-Vidosa, F.; Yepes, V.; Alcalá, J. Design of prestressed concrete precast road bridges with
hybrid simulated annealing. Eng. Struct. 2013, 48, 342–352. [CrossRef]
27. García-Segura, T.; Yepes, V. Multiobjective optimization of post-tensioned concrete box-girder road bridges
considering cost, CO2 emissions, and safety. Eng. Struct. 2016, 125, 325–336. [CrossRef]
28. Yepes, V.; Martí, J.V.; García-Segura, T.; González-Vidosa, F. Heuristics in optimal detailed design of precast
road bridges. Arch. Civ. Mech. Eng. 2017, 17, 738–749. [CrossRef]
29. García-Segura, T.; Yepes, V.; Frangopol, D.M. Multi-objective design of post-tensioned concrete road bridges
using artificial neural networks. Struct. Multidiscip. Optim. 2017, 56, 139–150. [CrossRef]
30. García-Segura, T.; Yepes, V.; Alcalá, J.; Pérez-López, E. Hybrid harmony search for sustainable design of
post-tensioned concrete box-girder pedestrian bridges. Eng. Struct. 2015, 92, 112–122. [CrossRef]
31. Li, X.G.; Wei, X. An improved genetic algorithm-simulated annealing hybrid algorithm for the optimization
of multiple reservoirs. Water Resour. Manag. 2008, 22, 1031–1049. [CrossRef]
32. Soke, A.; Bingul, Z. Hybrid genetic algorithm and simulated annealing for two-dimensional non-guillotine
rectangular packing problems. Eng. Appl. Artif. Intell. 2006, 19, 557–567. [CrossRef]
33. Penadés-Plà, V.; García-Segura, T.; Yepes, V. Accelerated optimization method for low-embodied energy
concrete box- girder bridge design. Eng. Struct. 2019, 179, 556–565. [CrossRef]
34. Yepes, V.; García-Segura, T.; Moreno-Jiménez, J.M. A cognitive approach for the multi-objective optimization
of RC structural problems. Arch. Civ. Mech. Eng. 2015, 15, 1024–1036. [CrossRef]
35. Martí, J.V.; García-Segura, T.; Yepes, V. Structural design of precast-prestressed concrete U-beam road bridges
based on embodied energy. J. Clean. Prod. 2016, 120, 231–240. [CrossRef]
36. García-Segura, T.; Yepes, V.; Frangopol, D.M.; Yang, D.Y. Lifetime reliability-based optimization of
post-tensioned box-girder bridges. Eng. Struct. 2017, 145, 381–391. [CrossRef]
Appl. Sci. 2019, 9, 3253 18 of 18

37. Penadés-Plà, V.; García-Segura, T.; Martí, J.V.; Yepes, V. A review of multi-criteria decision-making methods
applied to the sustainable bridge design. Sustainability 2016, 8, 1295. [CrossRef]
38. Goedkoop, M.; Spriensma, R. The Eco-Indicator 99: A Damage Oriented Method for Life Cycle Impact Assessment;
Prazska Energetika, A.S.: Prague, Czech Republic, 2001.
39. Alcorn, A. Embodied Energy and CO2 Coefficients for NZ Building Materials; Centre for Building Performance
Research: Dallas, TX, USA, 2003.
40. Catalonia Institute of Construction Technology. BEDEC ITEC Materials Database. Available online:
https://metabase.itec.cat/vide/es/bedec (accessed on 16 May 2019).
41. Yepes, V.; Martí, J.V.; García-Segura, T. Cost and CO2 emission optimization of precast-prestressed concrete
U-beam road bridges by a hybrid glowworm swarm algorithm. Autom. Constr. 2015, 49, 123–134. [CrossRef]
42. Molina-Moreno, F.; Martí, J.V.; Yepes, V. Carbon embodied optimization for buttressed earth-retaining walls:
Implications for low-carbon conceptual designs. J. Clean. Prod. 2017, 164, 872–884. [CrossRef]
43. Martínez-Calzón, J.; Ortiz-Herrera, J. Construcción Mixta Hormigón-Acero; Editorial Rueda: Alcon, Spain, 1978.
44. Viñuela-Rueda, L.; Martínez-Salcedo, J. Proyecto y Construcción de Puentes Metálicos y Mixtos; APTA:
Washington, DC, USA, 2009.
45. ACHE. Grupo de Trabajo 5/3 Puentes Mixtos Monografía M-10; ACHE: Chicago, IL, USA, 2016.
46. Kirkpatrick, S.; Gelatt, C.D.; Vecchi, M.P. Optimization by simulated annealing. Science 1983, 220, 671–680.
[CrossRef]
47. Medina, J.R. Estimation of Incident and Reflected Waves Using Simulated Annealing. J. Waterw. Port Coast.
Ocean Eng. 2001, 127, 213–221. [CrossRef]
48. Krishnanand, K.N.; Ghose, D. Glowworm swarm optimisation: A new method for optimising multi-modal
functions. Int. J. Comput. Intell. Stud. 2009, 1, 93–119. [CrossRef]

© 2019 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/).