See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329269865

Optimum design of a cable-stayed steel footbridge with three dimensional

modelling and control devices

Article  in  Engineering Structures · February 2019

DOI: 10.1016/j.engstruct.2018.11.038

CITATION READS

1 205

2 authors:

Fernando Ferreira Luis M. C. Simões

University of Coimbra University of Coimbra
8 PUBLICATIONS   25 CITATIONS    144 PUBLICATIONS   548 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reliability-based Design View project

Optimum design of Stiffened Metal structures View project

All content following this page was uploaded by Fernando Ferreira on 29 November 2018.

The user has requested enhancement of the downloaded file.

 Engineering Structures 180 (2019) 510–523

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Optimum design of a cable-stayed steel footbridge with three dimensional T

modelling and control devices
Fernando Ferreira , Luís Simões
⁎

Department of Civil Engineering, University of Coimbra, Portugal

ARTICLE INFO ABSTRACT

Keywords: Cable-stayed footbridges exhibit significant flexibility and lightweight leading to vibration prone structures.
Cable-stayed Safety and comfort of the pedestrians crossing the bridge are guaranteed by limiting both the horizontal and
Structural control vertical dynamic response of the bridge. In particular, the control of the synchronous lateral excitation (also
Multi-objective optimization known as ‘lock-in’) is a governing criterion in the design of long span footbridges. One of the most common ways
Integrated design
to mitigate the dynamic response is to install control devices. In the traditional approach both the structure and
Viscous dampers
Footbridges
control devices are designed separately. This paper concerns the integrated optimum design of a cable stayed
Lock-in steel footbridge and its control devices (here viscous dampers) by using three dimensional analysis. The opti-
mization algorithm finds the minimum cost solution which simultaneously satisfies all the static and dynamic
design criteria. Both ultimate and service limit states are considered, such as stresses throughout the structure,
buckling of the structure and the members, displacements, accelerations and the dynamic stability of the
structure when subject to synchronous lateral excitation. The analysis model includes both cable and geometric
nonlinearities.

1. Introduction Coimbra, Portugal [5,6], the ‘Mayfly’ Footbridge in Szolnok, Hungary

Long span footbridges are flexible and highly sensitive structures to
pedestrian induced vibration. In particular the lateral instability of a
footbridge under synchronous lateral crowd loading (also known as
‘lock-in’) is a governing factor in longer span footbridges. The most
famous event was the very high lateral vibration that occurred during
the opening day of the London Millennium Bridge in 2000. In the
subsequent studies Dallard et al. [1] analyzed several long span foot-
bridges with different structural systems and concluded that, in parti-
cular, the 80–200 m span range are vulnerable to the ‘lock-in’ phe-
nomenon. In this span range, one or more lateral vibration modes are
likely to have a frequency within the critical ‘lock-in’ range of
0.5–1.25 Hz. One of the most economical ways to mitigate the bridge
dynamic response is to install damping devices, in particular viscous
dampers and tuned mass dampers which was the solution found to
retrofit the Millennium Bridge in London. In 2006 Setra [2] published a
comprehensive technical guide that addresses the footbridge dynamic
fundamentals and design.
There are several examples of footbridges, where the designers have
implemented control devices in order to mitigate the dynamic response,
for example the Millennium Bridge in London [1], the Changi Mezza-
nine Bridge in Singapore [3,4], the Pedro and Inês Footbridge in
[7], the Stockton Infinity Footbridge in United Kingdom [8] and the
‘Ponte del Mare’ in Pescara, Italy [9].
The optimum design of cable-stayed bridges can be defined as
minimizing the structural cost while simultaneously satisfying all the
standard requirements (in this work the European codes of practice
[2,10–12] are satisfied), which are divided in static and dynamic cri-
teria.
Cable-stayed bridges have long been considered as an efficient de-
sign alternative [13,14]. Previous research works started with the op-
timization of the cable tensioning for steel, concrete and composite
cable-stayed bridges [15–26]. The optimization of bridge geometry and
structural cross sections was also considered in previous works [27–36].
The optimal design considering the three dimensional behavior was
also investigated [37–43]. Erection stages and seismic hazard were
considered in the design using both a spectral and time-history-based
procedure [40]. Camara and Efthymiou [44] analyzed the transverse
seismic response of three dimensional cable stayed bridges for different
span ranges.
The optimization of cable-stayed bridges has been published for the
dynamic response and the use of passive and semi-active control de-
vices in a cable-stayed bridge subject to seismic events [45] and cable-
stayed footbridges [46,47] using two dimensional models. The results

⁎
Corresponding author.
E-mail address: fferreira@uc.pt (F. Ferreira).

https://doi.org/10.1016/j.engstruct.2018.11.038
Received 4 May 2018; Received in revised form 12 November 2018; Accepted 13 November 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
F. Ferreira, L. Simões
showed that the structural performance was significantly improved
when compared to the solution where the structure and control devices
were designed separately. To the authors knowledge no other works
considered the optimum design of cable-stayed bridges with control
devices. Nonetheless other authors have proposed optimization algo-
rithms to solve the integrated structure and control design. For example
Tzan and Pantelides [48] presented an optimization method to design
active frames subject to seismic excitation, by minimizing the structural
volume with displacement and stresses constraints. Khot [49] optimally
designed integrated space systems using multi-objective optimization
by minimizing volume, control force and time to suppress oscillations.
Cimellaro et al. [50] illustrated a two-stage optimization procedure for
designing active steel frames with an inelastic response. The technique
proved efficient in determining the optimal control/structural system.
In this work the finite element model is used for the analysis of steel
footbridges and an optimization algorithm is proposed to find the op-
timum design. The 3D response of the structure considers the six in-
ternal forces interaction for the static design and includes the lock-in
criteria, vertical accelerations and respective sensitivities for the dy-
namic design of the structure. Contributions from the vertical and
torsional modes of vibration which characterize the spatial behavior are
considered, the optimum viscous damper coefficients for the transversal
and vertical directions are sought and the number of cables varies in the
optimization routine. Moreover an analytical formulation for finding
the pedestrian ‘lock-in’ sensitivities to the design variables is also pro-
posed.
The static response of the structure when subject to the standard
loading considers the geometrical nonlinear effects including the P-
delta effects in the tower and deck and the cables Ernst modulus [51].
In this case the design variables (DVs) are divided in three categories:
1. Structural properties: In this case the deck, tower and cable cross
sections and cable prestress;
2. Geometry:
a. Cable arrangement: Different cable position in the deck and
tower resulting in harp, semi-harp, and fan arrangement;
b. Tower shape: The change of the tower geometry resulting in in-
verted ‘Y’, ‘H’ and ‘A’ shaped geometries;
3. Control device properties: The tower-deck connection stiffness and
damping properties are considered as design variables.
The use of a relatively high number of design variables (DVs) poses
a challenge for the optimization algorithm while simultaneously sa-
tisfying a high number of static and dynamic criteria.
2. Structural model
2.1. Bridge geometry, structural cross sections and control design variables
A bridge with 180 m total length (the main span being approxi-
mately 90 m long) and 4 m wide is used to illustrate the technique and
the crossing is positioned 10 m above foundation. The geometry design
variables (DVs) allow the algorithm to change the bridge elevation, in
particular the cable position DVs can results in different cable
Fig. 1. Bridge elevation.
511
Engineering Structures 180 (2019) 510–523
arrangements (harp and fan systems in particular) (Fig. 1). The tower
shape geometry DVs allow for a complete change in the tower eleva-
tion, resulting in inverted ‘Y’, ‘H’ and ‘A’ shaped towers for example
(Fig. 2). The tower shape affects the lateral structural stiffness in par-
ticular, which is one of the governing aspects of the design, for the
limitation of the lateral displacement when subject to wind action and
the control of the ‘lock-in’ effect.
The structural system under analysis consists of a cable stayed
footbridge with a triangular bicelular box deck (Fig. 3), towers with
rectangular hollow section (Fig. 4). The added concrete slab is only
intended to add weight to the bridge which is needed to control the
dynamic response and no composite action is considered. The deck is
modeled as a simple section, but it actually represents a stiffened steel
plate deck (Fig. 3).
The number of cables (Ncables) is a discrete design variable and the
bridge is assumed to have the same number of cables in lateral and
central span for each tower. All the cable areas and prestress are design
variables.
The foundations are assumed to be sufficiently stiff to neglect the
structural-foundation interaction, thus all translations and rotations are
fixed at the tower base. The left abutment is considered to constraint all
translations and the torsion of the deck, the same happens with the
right abutment with the exception of the longitudinal horizontal
movement which is released. This is a common design option in order
to allow the bridge longitudinal expansion.
The tower deck connection properties (stiffness and damping) are
considered as DVs which are governed by the static and dynamic re-
sponse of the structure. The reason for this is that the connection
properties completely change the natural frequencies, the damping, the
deck displacement (in particular for lateral wind actions) and the tower
stresses. These design variables correspond to the design of the viscous
dampers in the tower-deck connection.
The structure is made of steel class S355 and the cables peak tensile
stress is 1860 MPa, which were chosen between the most common in
bridge practice. The use of high strength cables is also useful as it in-
creases the tension to mass ratio which results in an increase of the
cables fundamental frequency and reduces the likelihood of a cable-
structure interaction.
The assumed unit costs are 2.4 €/kg for the deck, 3.8 €/kg for the
tower, 8 €/kg for the cables and 500 €/m3 for the added concrete slab.
A total of 27 + 4 × Ncables structural, geometry and control design
variables (DVs) are summarized in Table 1.
A parametric bridge analysis model generator tool was built in
Matlab. This tool takes the user inputs (bridge length, width, steel
classes, loading, etc.) and design variables to generate a three dimen-
sional finite element model of the bridge (Fig. 5). This model uses Ti-
moshenko beams with linearized geometric stiffness to take into ac-
count both the deck and the towers P-Δ effects, rigid links to connect
the cable ends to the deck center of gravity and bars with equivalent
Ernst modulus for the cables. For the dynamic analysis, distributed mass
Fig. 2. Tower elevation.
F. Ferreira, L. Simões
is considered for the deck and lumped mass for the deck and cables. The
deck was discretized with one element for each segment and the towers
used four elements below the deck and above the deck until the first
cable connection (above the first cable, the towers are discretized for
each cable intersection). This was considered important to have a good
approximation for the buckling factors and eigenmodes for the towers.
The total number of elements in the finite element model is
2 + 4 × Ncables for the deck, 6 + 8 × Ncables rigid links, 28 + 4 × Ncables
for the towers, 8 × Ncables cables, 8 beams connecting the towers and 4
links for the tower-deck connection totaling 48 + 24 × Ncables elements
(ex. 116 elements for a bridge with Ncables = 3). The number of nodes is
3 + 6 × Ncables for the deck and 34 + 4 × Ncables for the towers with a
total of 37 + 10 × Ncables nodes (ex. 67 nodes for Ncables = 3). Each
node has six degrees-of-freedom (DOF) leading to a total of
222 + 60 × Ncables for the entire structure.
This finite element model was validated using a deck shell model in
the commercial software SAP2000® with very similar results for the
modes of vibration.
The displacement vector u in the static analysis is found by solving
the nonlinear equilibrium equation:
Ku = f = > (Km + K c + K g) u = f (1)
where f is the force vector and K the stiffness matrix that is given by the
sum of the material, cables and geometric stiffness (Km, Kc and Kg re-
spectively). The material stiffness matrix Km includes the deck, tower
and rigid links elements stiffness and is constant for the analysis of all
load combinations. The Kc and Kg matrixes depend on the element
stress as Kc is dependent on the cables tension and Kg on the deck and
towers axial forces. In this case Kg is positive for positive axial forces
(tension) inside the elements, nonetheless since almost all members are
in compression Kg is actually a negative stiffness matrix.
The nonlinear equilibrium equation is solved by using the Newton-
Raphson method.
Fig. 4. Tower cross section above and below the deck.
512
Engineering Structures 180 (2019) 510–523
Fig. 3. Deck equivalent and real cross sec-
tion.
2.2. Static loading and design criteria
Four loading types are considered: Self Weight “Gk”; Imposed load
Qk = 5 kN/m2; Wind Wk = 1.6 kN/m2 (corresponding to a mean wind
velocity of vm = 34 m/s at the deck height according to Eurocode 1
[11]); and cable prestress “P”. The corresponding reduction and com-
bination factors are used according to the Eurocode 0 [10].
The wind loading is only considered as a static action and no dy-
namic effects (apart from the wind turbulence that is already included
in the Eurocode 1 [11] in Wk) are taken into account. It is known that
the wind dynamic effects are particularly important in long span and
torsional flexible bridges [52–57]. In this example, the combination of
the bicelular deck with the cables being connected at the deck edges,
leads to a high torsional stiffness and low vulnerability to the wind
dynamic phenomena (such as flutter, vortex-shedding, buffeting, etc.).
The force coefficients for the deck and towers are considered re-
spectively as ± 1.4 and ± 1.1 for all directions, meaning the wind can
induce pressures in both directions. The wind velocity is assumed
constant along the deck and towers below the deck. For the tower
members above the deck the wind velocity is assumed to be 15% higher
and constant along height. This procedure corresponds to a simplified,
yet conservative, approach to take into account the wind static actions
using the Eurocode 1 [11] rules. Actually, the force coefficient in a
given cross section varies when the geometry of the cross section
changes and there a number of papers which design bridge decks with
varying geometry considering both static and dynamic analysis [52–56]
but they fall outside the scope of the present work.
For the design, different loading cases in the tower and deck are
considered (Table 2) for both ultimate and service limit states (ULS and
SLS respectively). In Table 2 the ‘+’ signs corresponds to the action of
both wind and live load in the gravity direction, the ‘−’ sign correspond
to no live load and to upward wind. The ‘+’ and ‘−’ signs also corre-
spond to a factor of 0.9 and 1.1 respectively for the prestress “P” cable
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

Table 1
Problem DVs.
Design variables Designation DVs number Units

Geometry Lcentral/180 Central span to total length ratio (Fig. 1) 1 –

x1, x2 and x3 Define the cable position (Fig. 1) 3 m
z1/Lcentral Tower height to span ratio (Figs. 1 and 2) 1 –
z2/z1 Defines anchorage distance (Figs. 1 and 2) 1 –
y1 and y2 Defines tower geometry (Fig. 2) 2 m
Ncables Number of cables (Fig. 1) 1 –

Cont. ky, kz Stiffness and damping properties of the tower deck connection 2 kN/m
cy , c z 2 kNs/m

Cable Cable.s area Cable area in the side span Ncables cm2
Cable.c area Cable area in the central span Ncables cm2
Cable.s prestress Cable prestress in the side span Ncables kN
Cable.c prestress Cable prestress in the central span Ncables kN

Section sizing H,B,t,e1,e2,deck Deck cross section (Fig. 3) 5 mm

H,B,t,e1,tower Tower cross section below and above the deck (Fig. 4) 4 mm
H,B,t,e2,tower 4 mm
hslab Height of the concrete slab (added mass) 1 cm

Total 27 + 4 × Ncables

anchored in that deck portion (for the ULS combinations only). The
deck load arrangement corresponds to: 1-left lateral span; 2-left side of
the central span; 3-right side of the central span and 4-right lateral
span.
For the ULS combinations, a torsional eccentricity of 20 cm is also
considered for the deck in the positive y direction for the actions with
‘+’ sign and in the opposite direction with the ‘−’ sign. For example the
“Qk” load corresponds not only to a downward uniform distributed
force of 20 kN/m (4 m × 5 kN/m2 for the elements with ‘+’ sign and 0
for the elements with ‘−’ sign) but also to a 4 kNm/m uniform torsion
(0.2 m × 20 kN/m) in the deck.
The structural imperfections are considered to be negligible when
compared to the magnitude of the live loads so they are not considered
in this work.
Fig. 6 gives an example of the ULS-4 load arrangement to better
illustrate the design combinations. The wind action is subdivided in y
and x directions (‘Wy’ and ‘Wx’ respectively) and the wind pressure in
each surface is given by the product of the force coefficients, char-
acteristic and combination values.
The wind may be associated with different force coefficients, for
instance it can lead to an upward or downward force in the deck for the
same combination. This means that the wind pressure signal is de-
termined according to the worse case stress envelope and this is a
conservative procedure for the wind effects on the structure because,
the wind is not a governing loading in this example. For wind vibra-
tional prone structures, in particular long span and torsional flexible
structures, a more complex approach should be followed [52–55].
The design criteria (DC) are all the ultimate and service limit states
(ULS and SLS) according to the Eurocode 3 [12] procedure, in parti-
cular resistance of cross section, buckling resistance of members, ver-
tical and horizontal deflections (Table 3). In the optimization algorithm
all the DC are normalized (divided by their limit value) to define a
design factor (DF), which must be smaller than 1 to guarantee all the
structural requirements.
The deck and tower buckling is found for load combinations ULS-1
to ULS-8 by solving the eigenvalue problem (Eq. (2)). It could be argued
that ULS-1 and ULS-2 have the highest total loading and are therefore
more likely to be governing to the buckling stability and that ULS-3-8
could be ignored for this analysis. Pedro and Reis [58] have shown that
in certain situations, other load combinations lead to lower buckling
loads and, for this reason, the authors of this work opted to include all
the load combinations that increase the vertical downward action (ULS-

Fig. 5. Finite element model example with the deck, towers and rigid links represented with solid lines and cables with dashed lines.

513
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

Table 2
Design combinations.
Deck load arrangement Tower 1 Wind Tower 2 Wind

Combination 1 2 3 4 x y x y Load combination

Ultimate Limit State – ULS ULS-1 + + + + + + + + 1.35 “Gk” + 1.5”Qk“ + 0.9 ”Wy“ + 0.3”Wx” + ”P”
ULS-2 + + + + + + + + 1.35 “Gk” + 1.5”QK“ + 0.3 ”Wy“ + 0.9”Wx” + ”P”
ULS-3 + − + − − + − + 1.35 “Gk” + 1.5”Qk“ + 0.9 ”Wy“ + 0.3”Wx” + ”P”
ULS-4 + − + − − + − + 1.35 “Gk” + 1.5”QK“ + 0.3 ”Wy“ + 0.9”Wx” + ”P”
ULS-5 − + + − + + − + 1.35 “Gk” + 1.5”Qk“ + 0.9 ”Wy“ + 0.3”Wx” + ”P”
ULS-6 − + + − + + − + 1.35 “Gk” + 1.5”QK“ + 0.3 ”Wy“ + 0.9”Wx” + ”P”
ULS-7 + − − + − + + + 1.35 “Gk” + 1.5”Qk“ + 0.9 ”Wy“ + 0.3”Wx” + ”P”
ULS-8 + − − + − + + + 1.35 “Gk” + 1.5”QK“ + 0.3 ”Wy“ + 0.9”Wx” + ”P”
ULS 9 to 14 Same as Comb 3 to 8 but with 1.0 “Gk” and towers “Wy” with opposing signs
ULS 15 to 28 Same span arrangement as Comb 1 to 14 but with 1.5 “Wk” and 1.05”Qk“

Service Limit State – SLS SLS-1 + − + − − + − + 1.0 “Gk” + 1.0”Qk“ + 0.6 ”Wy“ + 0.2”Wx” + “P”
SLS-2 + − + − − + − + 1.0 “Gk” + 1.0”Qk“ + 0.2 ”Wy“ + 0.6”Wx” + “P”
SLS-3 − + + − + + − + 1.0 “Gk” + 1.0”Qk“ + 0.6 ”Wy“ + 0.2”Wx” + “P”
SLS-4 − + + − + + − + 1.0 “Gk” + 1.0”Qk“ + 0.2 ”Wy“ + 0.6”Wx” + “P”
SLS-5 + − − + − + + + 1.0 “Gk” + 1.0”Qk“ + 0.6 ”Wy“ + 0.2”Wx” + “P”
SLS-6 + − − + − + + + 1.0 “Gk” + 1.0”Qk“ + 0.2 ”Wy“ + 0.6”Wx” + “P”
SLS-7 + + + + 1.0 “Gk” + ”P“

1 to 8). consequence the erection stages are not a governing aspect of the de-
sign. Nonetheless the final design was checked in order to guarantee
det(Km + K c + g K g) = 0 or (Km + K c) g = g Kg g (2) that it could resist during the construction process.

where g is the buckling eigenvector and λ g is the load factor (buckling
eigenvalue), meaning the ratio between the critical buckling and ap-
plied loads. As presented in Table 3 the load factor λ g must be higher
than 2 for all ULS-1 to 8 combinations.
The resistance of the tower and deck cross sections for each in-
dividual element follows Eurocode 3 [12] provisions, which requires
not only checking the Von Mises stress considering the internal forces
interaction (axial, torsion, bending and shear in both directions), but
also the buckling resistance of members. Some of the non-linear P-Δ
effects are already incorporated in the model, via the Kg matrix, but
each element is also checked for the internal forces and buckling in-
teraction considering the length of the elements as buckling length. It
would be also possible to use a more refined discretization and only
check the Von Mises stress, but that would require a higher computa-
tional effort for the analysis. Only the 12 worse combinations are
considered so, the number of DC is equal to the number of elements
times 12.
The structural deflections are also limited by considering standard
length to deflection ratios.
The total number of static design criteria in this example depends on
the number of cables (Ncables) and is in this case equal to 582 + 320
Ncables, for example, for a bridge with Ncables = 3 the number of static
design criteria is equal to 1542 (582 + 320 × 3).
The erection stages are not directly taken into account because, in
footbridges, the live to dead loads ratio is substantially higher and in
2.3. Dynamic loading and design criteria
The dynamic load models for pedestrians bridges is still an area of
intensive research nowadays. Human induced loads are different for
each individual and even vary from time to time for the same person.
This type of action has a random character for various reasons, for
example the pedestrian weight, gait and even the left and right steps are
different [59]. The behavior of a single pedestrian changes when he is
within a crowd and his reaction will be different when the ground is
moving. There are a high number of research works that study the
pedestrian loading, the pedestrian-pedestrian interaction and the pe-
destrian-structure interaction [60–63] and, so far, different load modes
have been proposed by different researchers.
The incorporation of this state-of-art load models in the interna-
tional standards is still a work in progress. Racic et al. [62] have
compared the different load models with the experimental findings and
concluded that Setra guidelines [2] have the best match with the ex-
perimental observations. These guidelines provide a load model for the
single pedestrian and determine the crowd response via multiplying the
single pedestrian response by an equivalent number of pedestrians.
In this example, the Setra guidelines [2] are followed, nonetheless
the parametric tool in this work allows the user to input different load
models. For the dynamic loading a crowd density of 1.5 pedestrians/m2
is considered for the evaluation of the ‘lock-in’ limit state and 1.0 pe-
destrians/m2 for the evaluation of the vertical acceleration comfort

Fig. 6. Bridge elevation with the load arrangement example for the ULS-4 combination (only the loads in the x-z plan are drawn).

514
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

Table 3
Static DC.
Static design criteria Number of DC

ULS criteria Deck buckling The critical buckling load of the deck and tower must be higher than 2 times the applied ULS loads 12
Tower buckling 12
Deck resistance The combined bending, shear, axial forces and torsion resistance must be checked according to Eurocode 3 [12] 24 + 48 Ncables
Tower resistance section 6 336 + 48 Ncables
Cable resistance Cable tension must be lower than 74% of the peak tensile strength 64 Ncables

SLS criteria Tower uy displacement Structural deflections must be lower than L/250 for load combinations, L/300 for isolated actions according to 72 + 12 Ncables
Tower ux displacement Eurocode 3 [12] 84 + 12 Ncables
Deck uy displacement 21 + 36 Ncables
Deck uz displacement 21 + 36 Ncables
Cable tension Cable tension must be inside the [5–50% ] range for all combinations 64 Ncables

Total 582 + 320 Ncables

limits. The single pedestrian time history loading in the vertical direc- Mu
¨ + Cu + Ku = f (4)
tion Fv (t ) is considered as a sum of three harmonic forces (Eq. (3)) and u
Considering the state space vector x defined as x = [ ], the 2nd
the crowd response is found by multiplying the individual response by u
the equivalent number of pedestrians 1.85 N =1.85 4 × 180 × 1 50 . order equation (Eq. (4)) is replaced by a 1st order:

Fv (t ) = G0 + (fm ) G0 [0.4sin(2 fm t ) + 0.1sin(4 fm t /2) + 0.1sin(6 fm t /2)] x = Ax + Bf (5)

(3) where:
where G0 is the average pedestrian weight equal to 700 N in this case,
0 I 0
fm the walking frequency, t the time and (fm ) is a minus factor with A=[ ]; B = [ ]
M 1K M 1C M 1 (6)
unit value in the [1.7–2.1] Hz range decreasing linearly to zero at 1 and
2.6 Hz frequencies. here 0 and I stands for zero and identity matrixes respectively.
The lateral time history loading was not considered in the dynamic Eq. (5) can be solved using an analytical formulation (Eq. (7)) for a
analysis because it has low influence on the vertical vibration response given time step Δt given by:
and also the control of the lateral modes of vibration is implicit in the
t+ t
‘lock-in’ design criteria. The time history loading is placed in all the x (t + t ) = e t A x (t ) + e[(t + t) ] A Bf ( )d
deck (corresponding to a centered position of the pedestrians in the t (7)
structure) and cable end nodes (corresponding to the worse pedestrians’
Considering a linear force variation the solution can be put in the
eccentric position) where the vertical and torsional modes interaction is
following form:
considered. The analysis is carried out in all frequencies within the
[1–2.6] Hz range with a 0.01 Hz interval and at the structural eigen- x (t + t ) = Kbx (t ) + Kf Bf (t ) + K f B [f (t + t ) f (t )] (8)
frequencies within this range. The 200 highest peak responses in the
structure are considered as design criteria in this example. where:
The Setra guidelines [2] load model and design criteria are used and t A; Kf
a 2.5 m/s2 peak vertical acceleration is set as a design limit to guarantee Kb = e K f = A 1 (K b I); K = A 1( I)
f
t (9)
the minimum comfort limits. Alternatively, international standards
prescribe different criteria, such as frequency-dependent where the In this example the time step was considered as dependent on the
maximum peak acceleration depends on the frequency [64,65]. The walking frequency t = 1/(36fm ) .
software developed in this work is not bound to specific criteria and
allows the user to introduce different values, for example, frequency- 3.2. Analytical evaluation of synchronous lateral instability
varying acceleration limits.
The dynamic analysis model is described with more detail in the The synchronous lateral instability of a footbridge or ‘lock-in’ is a
following sections. phenomenon where most of the pedestrians in a footbridge synchronize
the footstep frequency to the vibrating frequency of the bridge. To this
3. Dynamic analysis model date this phenomenon is not associated with structural collapse and the
reason for being described as ‘instability’ is it’s mathematical funda-
3.1. Time history loading mental.
The studies carried out after the closure of the Millennium Bridge in
The dynamic time history analysis employed the analytical in- London [1] concluded that, for higher lateral oscillations, the pedes-
tegration method due to its reduction in computational effort. Initially, trians would create a reaction in the structure in phase with the motion
the mass, damping and stiffness matrixes and the force vector (M, C, K in order to keep their own balance. These studies concluded that the
and f respectively), are determined using the finite element model. The reaction was equivalent to each pedestrian being modeled as a negative
stiffness matrix K considers the tangent stiffness (the Kc and Kg are damper. The negative damping coefficient cp varies for different vi-
assumed constant), the damping matrix C is the sum of a Rayleigh bration frequencies and was considered as cp = 0.3 kN s/m for vibration
damping matrix (with 0.4% damping fixed at the 0.4 and 4 Hz fre- frequencies between 0.5 and 1 Hz and decreases linearly to zero at 0.3
quencies) and the influence of the dampers in the tower-deck connec- and 1.5 Hz respectively.
tion and the force vector f considers the time history loading (Eq. (3)) in Taking the experimental observations under consideration Dallard
the different possible positions. et al. [1] proposed a formulation to determine the critical number of
The evaluation of the structural response (u) needs solving the dy- pedestrians Ncr that induce the lateral instability of a bridge with a
namic equilibrium equation: single mode of vibration with semi-sinusoidal shape:

515
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

8 fm Simões and Templeman [66] proposed an optimization method to solve

Ncr =
cp (10) this problem, given that finding the active constraints set in more
conventional direct search algorithms is not efficient, the problem is
where m stands for the modal mass, ξ the structural damping and f the transformed in an equivalent minimax optimization problem (Eq. (13))
fundamental frequency of the bridge. and from then to a Kreisselmeyer-Stainhauser scalar function (Eq. (14)).
For structures with multiple modes of vibration, or when control This formulation consists in a convex approximation of the objective
devices are placed, the evaluation of the lateral instability poses a more and design factor constraints and allows transforming Eq. (12) into an
complex problem and it can be stated that the dynamic analysis is unconstrained optimization problem that depends on the convex ap-
mathematically equivalent to the evaluation of wind flutter instability. proximation parameter ρ, which must be increased throughout the
McRobie and Winslow [48] have presented an analytical evaluation optimization process to make the convex approximation closer to the
method for structures with multiple degrees of freedom. original nonlinear functions.
To evaluate the lateral instability, the state space matrix A (Eq. (6))
is expanded (Al) to cover the pedestrians negative damping (Eq. (11)). ¯)
Costit + 1 (dv
¯)
Minmaxi {DF i (dv 1, 1}
The determination of the critical number of pedestrians Ncr is found by Costit (13)
increasing its value until the real part of at least one of the eigenvalues
NDF ¯)
of the state space matrix Al is equal to zero (equivalent to a zero 1 (DF i (dv
¯ ) 1) (
Costit + 1 (dv
1)
Min ln [ e +e Costit ]
damping state) meaning that for a higher number of pedestrians the i=1 (14)
system would become unstable.
DF i (dv
¯)
After the sensitivities for all design criteria and cost ( ¯j
and
dv
0 I ¯ )
Al = Costit + 1 (dv
respectively) have been computed a linear Taylor series ap-
M 1K M 1 (C cp Ncr C un, lock in) (11) dv¯j
proximation of Eq. (14) is used. Eq. (15) consists of an unconstrained
where C un, lock in is the unit lateral negative damping matrix found by optimization problem that can be solved using standard methods such
assembling a unit pedestrian negative damping in the transverse di- as the conjugate direction method. The solution of this problem is one
rection times the pedestrian relative density throughout the structure. iteration of the complete solution of Eq. (14) and an iterative sequence
Since the eigenvalue analysis depends on cp , which in turn depends on of explicit approximation problems is performed until the design vari-
the structural frequency, the imaginary part of the eigenvalue needs to ables converge to a stable solution. Depending on the initial design the
be checked and if a different frequency is found, cp needs to be changed algorithm can converge to a safe solution (all DFi ≤ 1 and at least one
accordingly in the following iterations. DFi = 1) with minimum cost (Pareto solution) or to an unsafe solution
The objective is to find the Ncr value that leads to a null real part of (at least one DFi > 1) with minimum design factor violation.
one eigenvalue and this analysis is repeated for all eigenvalues inside
NDV
the critical lock-in range. Since it is not likely to have more than ten NDF [DF i (dv
¯ ) 1+ (
DF i (dv)
× dvj )]
1 dv¯j
modes inside the critical ‘lock-in’ range, only the first ten eigenvalues Min ln [ e j =1

are computed. i=1

¯) NDV ¯) ¯j
Costit + 1 (dv Costit + 1 (dv dv
[ 1+ ( × )]
Costit dv¯j Costit
4. Optimization strategy and sensitivity analysis +e j =1 ] (15)

4.1. Optimization strategy where NDV is the number of design variables, equal to 27 + 4Ncables in
this particular example (Table 1) and dvj is the jth element of the de-
sign variable change vector dv .
The objective of this work is to find the lowest cost structure that
This optimization strategy has been used in cable stayed bridges
satisfies all the standard design criteria (DC) defined in Sections 2.2 and
subject to earthquakes [45] and pedestrian induced loading [46,47]
2.3 such as cross section resistance, buckling, deflections, accelerations
using active devices considering only the two dimensional behavior.
and ‘lock-in’ assessment. The design criteria are put in a normalized
The extension of this algorithm to a footbridge with VDs considering
form by dividing each one to its maximum value (DCimax) and these
the three dimensional structure posed some difficulties which required
quotient is defined as the structural design factor DFi = DCi/DCimax. It is
to update the algorithm, in particular including the ‘lock-in’ criteria,
important to clarify that the European legislation [10–12] takes the
vertical response considering the torsional influence, finding the op-
characteristic values of the actions and increases them by partial fac-
timum damping coefficients, analysing a higher number of load com-
tors, the same happens for the resistance in which the characteristic
binations and considering the number of cables as a discrete design
value is also decreased by partial factors. After this, the safety check of
variable.
the elements is performed to a reference design factor of 1. There are
For the dynamic DC the key aspects are the mass, modal frequencies
other international standards, even outside the Structural Engineering
and damping and all DV influence, even if in a small scale, these three
scope, where both loads and resistance are taken with their mean values
properties. One of the objectives of the structural design is to shift the
and, in this case the design factors need to be set to a substantially
lateral and vertical frequencies as far as possible from the [0.5–1] Hz
higher value.
and [1.7–2.1] Hz ranges respectively. The optimization algorithm does
In this problem the number of structural elements depends on the
not take this directly into account as the information which is given is
number of cables so the total number of static plus dynamic design
the sensitivities (Eq. (15)). It can be very easily observed that the
factors NDF is equal to 792 + 320 × Ncables.
sensitivities are more meaningful when the mode is outside the critical
The problem is then formulated (Eq. (12)) as finding the optimum
range (in the decreasing trend) both for the ‘lock-in’ and the vertical
design variables vector that guarantees all the design criteria and si-
accelerations criteria. Take, for example, a particular situation where a
multaneously tries to reduce the Cost in all iterations (it).
critical ‘lock-in’ mode has a frequency of 0.51 Hz, in this case the ne-
all DF i (dv ¯) 1 ¯)
all DF i (dv 1 0 gative damping coefficient cp has a null sensitivity with respect to the
or mode frequency. By looking at Eq. (10) it becomes apparent that the
¯)
Costit + 1 (dv Costit ¯ )/ Costit
Costit + 1 (dv 1 0
sensitivity of the critical number of pedestrians with respect to the
(12)
frequency is positive (considering that the damping and mass variation
Both the Cost and the design factors are functions of dv and the are smaller), meaning that the algorithm would try to increase the
objective is to find the value that satisfies the conditions in Eq. (12). frequency (and by consequence Ncr) where, in fact it could do the exact

516
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

opposite. Another possibility is that, if in a particular iteration, the fictitious forces [

f
¯j (
Km
¯j +
Kc
¯j +
Kg
¯ j ) u], which allows obtaining
dv dv dv dv
design factor for the dynamic criteria is much higher than for the static
the displacement vector sensitivity u
.
criteria, the algorithm could try to increase the mass (in particular the ¯j
dv
hslab DV), which is inefficient for the static design as it adds weight to (
Km
+
Kc
+
Kg
) u + (Km + K c + K g)
u
=
f
¯j ¯j ¯j ¯j ¯j
the structure, but is a solution to decrease the dynamic response. These dv dv dv dv dv
Kg
issues can happen more often in the first iterations and to solve them <=>
u
= (Km + K c + K g) 1 [
f
(
Km
+
Kc
+ ) u]
¯j
dv ¯j
dv ¯j
dv ¯j
dv ¯j
dv (16)
two modifications are implemented and used in all iterations until the
worse dynamic design factor is lower than 1.1 (meaning that the so- Each element internal forces vector pi (N, Vz, Vy, T, My and Mz) is
lution is closer to being valid, but not yet optimal): computed by multiplying a specific matrix Di (different for each ele-
ment) by the displacement vector u (pi = Di u ) and the sensitivity
• Using two different Rayleigh damping matrixes, (with 0.4% analysis is given by:
damping at different points: 0.5 and 1 Hz for the ‘lock-in’ analysis
pi Di u
and 1.7 and 2.1 Hz for the vertical accelerations). This modification = u + Di
¯j
dv ¯j
dv ¯j
dv (17)
introduces an additional sensitivity for the damping in respect to the
frequency that is of advantage to the algorithm to shift the fre- Kg
Since the geometric and cables stiffness sensitivities ( and Kc
)
quencies outside of the critical range; ¯j
dv ¯j
dv

• Pursuing an independent search for the optimum damping coeffi- are not directly available and depend on the element internal forces pi
(tension for the cables and compression for the deck and tower), Eqs.
cients (cy and cz) using a logarithmic search, where each damping
value is multiplied by [1 × 10i] (i = −2:0.2:2). The objective was to (16) and (17) must be solved iteratively until convergence.
converge faster to a better damping efficiency. For the buckling load factors sensitivities an eigenvalue sensitivity
analysis of Eq. (2) is carried out:
The discrete DV (Ncables) is associated with a topology change and an Km Kc g g Kg g
( + ) + (Km + K c) = Kg g Kg
enumerative procedure is proposed. This DV is kept constant until the ¯j
dv ¯j
dv g ¯j
dv ¯j
dv g g
¯j
dv g ¯j
dv
algorithm converges to a local optimum and, after that, the algorithm
(18)
restarts with a ± 1 cable and the optimization process continues.
In this problem the domain is possibly non-convex and non- By multiplying all the elements of Eq. (18) in the left side by gT and
connected and there is no guarantee that the global minimum is rearranging the equation one obtains the eigenvalue sensitivity (Eq.
reached therefore, to find the best possible solution, a multi-start pro- (19)). The equation was simplified by assuming that the eigenvector
sensitivity is null ( 0 ) which creates a small error in the sensitivity
g
cedure is employed and the lowest cost and safe solution is chosen. The ¯j
dv
multi-start procedure requires the user to generate several initial solu- analysis, but is useful for faster computation and does not compromise
tions. Different initial geometries and the sectional design variables the quality of the solutions.
were considered with pre-designing formulas (for examples the towers
Km Kc Kg
were pre-designed so that the buckling load of the member was higher T T g T
g ( ¯j
+
¯j
) g + g g
¯j g =
¯j g Kg g
dv dv dv dv
than the total weight of the structure divided by four). The following
Km Kc T Kg
variations were considered in the multi-start strategy:
T
g g ( ¯j
dv
+ ¯ j) g
dv
+ g g dv
¯j g
=> =
¯j T

• Tower height to span ratio between 1/4 and 1/8;

dv g Kg g

• Tower shapes with inverted ‘Y’, ‘H’ and ‘A’ shaped geometries; (19)

• Different control devices properties with three different initial The determination of the dynamic analysis sensitivity (Eq. (20))
results from differentiation of Eq. (4). As it can be seen after de-
properties (stiff, intermediate and soft);
• The total number of cables was considered between 3 and 4 in the termining the stiffness, damping and mass derivatives, the structural
response derivative is found by solving a similar dynamic equilibrium
initial design.
f
equation than Eq. (4) (with fictitious forces dv¯ ¯ ü )
K C M
¯ u
dv ¯ u
dv dv j j j j
After the optimization process converges, the multi-start routine and is computed using the same time history method presented in
saves the final optimal solution to an independent database. This has a Section 3.1.
double purpose as it is common that different initial designs lead to a
similar solution after a couple of iterations so, if a solution is similar to a u u u¨ f K C M
K +C +M = u u u¨
¯j
dv ¯j
dv ¯j
dv ¯j
dv ¯j
dv ¯j
dv ¯j
dv (20)
previous obtained one it is important to not waste time analysing it
again. Also this tool is able to analyse different examples, so if the user In this case the sensitivity of the stiffness matrix does not consider
now wants a solution for a different bridge (with 200 m length for ex- the non-linear behavior, only the material stiffness sensitivity
ample) the optimum solutions for this example are also employed as
( dv¯ = dv¯m ).
K K
initial designs. j j
For the sensitivities of the lock-in design criteria the derivatives of
the eigenvalues of Al (Al l = l l ) must be obtained (Eq. (21)). The
4.2. Sensitivity analysis
sensitivity of the eigenvector is rather small ( l
¯j
dv
0 ) and allows
The first part of the sensitivity analysis is to find the finite element simplifying the equation. The eigenvectors are normalized so that
l = 1.
T
stiffness, mass and damping derivatives. This was done using a com- l
bination of the finite difference, the direct differentiation method and Al
T T l T l T l
the semi-analytical method [67]. This option allows to take advantage l ¯j l + l Al ¯j
= l ¯j l + l l
¯j
dv dv dv dv
of the analytical sensitivity computation in the simpler and direct
T Al
equations and the finite difference method in the remaining (ex. Ti- l l dv ¯j l Al
T
moshenko beam). The static and dynamic design criteria sensitivities => = =
¯j
dv T l ¯j
dv l
(21)
l l
allow the optimization algorithm to find the most promising design
variable changes (Eq. (15)) in each iteration. where l and l are respectively the eigenvalue and eigenvector of Al
The sensitivity analysis for the static loads (Eq. (16)) is done by and the sensitivity of the state space matrix are given by:
differentiation of Eq. (1) and consists of a structural analysis with

517
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

Al 0 0
=[ 2 MK 1 K 2 M (C C cp Cun, lock in ]
¯j
dv M ¯j
dv
M ¯j
dv
M ¯j
dv
cp Ncr C un, lock in) M 1( ¯j
dv ¯ j Ncr C un, lock in
dv
cp Ncr ¯j
dv
) (22)

An important aspect is that the sensitivities are found for the

number of pedestrians that induce lock-in to the relevant eigenvector (around iteration 10), the optimization process started to reduce the
l . The eigenvalue sensitivity ¯ is then divided in the real and ima- cost and converged to an optimum cost design (reference cost equal to
l
dvj
ginary parts sensitivities, the first allows determining the sensitivity of 664.769 €). This happened after 22 iterations with a constant number of
the critical number of pedestrians and the second gives the frequency f cables (Ncables = 3) and after this, the algorithm restarted with the same
derivative, which in turn gives the sensitivity of cp (Eq. (23)). This geometry and section sizing and changed to 32 cables (Ncables = 4). This
means that Eqs. (21)–(23) must be solved together iteratively until design did not guarantee safety by a small percentage (≈10%), but
c
convergence since the value of ¯p is not directly available. after 8 iterations converged to a safe and lower cost design (reference
dvj
cost equal to 655.752 €) (solid blue geometry in Fig. 8).
5f 1.5; f [0.3 0.5]Hz The optimum design is governed by both static and dynamic re-
cp = (0.3kNs/m) × 1; f [0.5 1.0]Hz quirements, in particular stresses and displacements for both deck and
2f + 3; f [1.0 1.5]Hz tower, tower buckling, vertical accelerations and ‘lock-in’. The fact that
5 Imag( l)
the solution is not only governed by the dynamic requirements is a good
2 ¯j ;
dv
f [0.3 0.5]Hz indicator that the control strategy was efficient and that reducing the
cp
=> = (0.3kNs/m) × 0; f [0.5 1.0]Hz cross sections any further is not possible for safety reasons.
¯j
dv 2 Imag( l ) The initial and optimum DVs are displayed in Table 4 (geometry and
2 ¯j ;
dv
f [1.0 1.5]Hz
control), Table 5 (cable prestress and area), Fig. 9 (tower cross section)
(23) and Fig. 10 (deck cross section) and the finite element model for both
designs is displayed in Fig. 8.
The definition of the critical number of pedestrians is defined by the The optimization algorithm alters the solution from a semi-harp
real part of the eigenvalue being zero. After determining the eigenvalue with ‘H’ shaped tower to a fan type bridge with inverted ‘A’ shaped
sensitivities with respect to the design variable changes ¯ l , the deri-
dvj towers. This design type was common in the vast majority of the so-
vative of the critical number of pedestrians Ncr
¯j
is then obtained by: lutions found using the multi-start procedure, meaning that this system
dv
is most certainly the most efficient structural type for this design. The
Real( l ) Real( l ) Ncr Ncr
Real( l ) = 0 = > + =0<=> bridge height substantially decreased and the central span slightly in-
¯j
dv Ncr ¯j
dv ¯j
dv
creased leading to a much slender structure.
The use of fan and semi-fan systems (ex. ‘A’ or inverted ‘Y’ shaped
Real( l) T Al
¯j
dv
Real( ¯j
l dv l)
= Real( l)
=
T Al
towers respectively) can be subject to a discussion, since in many
Real( l) (24)
Ncr l Ncr practical application it may not be possible or cost effective to anchor
all the cables at the same position and, if this is the case, a constraint
where Al
is given by differentiation of Eq. (11):
Ncr must be placed at the z2 DV, or an added cost for lower values of z2 in
Al 0 0 order to guarantee a minimum spacing for the cables. In this case a
=[ ]
Ncr 0 M 1 (cp C un, lock in) (25) relatively small number of cables are needed so there was no restriction
for the minimum spacing for the cables.
Eq. (25) is simplified considering that that the sensitivity of cp with The Bdeck design variable, in real practice can be constrained as the
cp
respect to Ncr is null ( N = 0 ). In mathematical terms a change in Ncr bridge width, or if Bdeck is lower than the width, a secondary structure is
cr
will result in a slight variation of the imaginary part of the eigenvalues required to materialize the deck and to attach the cables to the primary
and, in consequence of cp, nonetheless this variation is very small and structure. This issue could be easily solved by constraining Bdeck to a
the simplification avoids unnecessary iterative procedures without any constant value, making the optimization iterative process easier by
substantial loss of accuracy. decreasing one (important) design variable. However this work illus-
trates and tests the optimization algorithm so the authors opted to allow
5. Bridge design results Bdeck to change to a minimum value of 0.5 of the width of the bridge.
In terms of the tower-deck vertical connection properties (Table 4)
5.1. Design factor vs. cost evolution the stiffness of the transverse direction (ky) and the damping in the
vertical direction substantially (cz) increased. This was a consequence
A simple way to demonstrate the optimization process is to visualize of the need to control the dynamic response of the structure.
the design factor vs. cost evolution in all iterations. A multi-start
strategy was employed and different initial solutions lead to different 7,50,000 € 3

optimization processes, nonetheless it was observed that most of the 7,30,000 €

solutions converged to the same design and, in other cases, the algo- 2.5
rithm got trapped in cost inefficient designs. Given the non-convex and 7,10,000 €
Reference Cost

Ncables= 4
Design Factor

potentially nonconnected domain, which is a feature of optimization 6,90,000 €

with dynamic constraints, some initial designs converged to this solu- 2
6,70,000 €
tion while others converged to a safe solution but with inefficient costs Ncables= 3
(up to 30% more expensive). It can be observed that none of the cross 6,50,000 €
1.5
sections have a safety margin and neither the dynamic DC so the cross Cost Design Factor
6,30,000 €
sections cannot be reduced any further. The optimization ended with
the optimum design (Fig. 7), 30 iterations being needed to find this 6,10,000 € 1
Pareto optimum. 0 5 10 15 20 25 30
Iteration
The initial design is a harp footbridge with ‘H’ shaped towers and a
total of 24 cables (Ncables = 3) (dashed red geometry in Fig. 8). The Fig. 7. Reference cost and DF evolution throughout the optimization process
initial design was unfeasible so, at first, the design factor improved with that resulted in the optimum design.
the consequence of higher cost. When a feasible design was found

518
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

a) 3d View

b) Bridge elevation

Fig. 8. Initial (dashed red) and optimum design geometry (solid blue). (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)

Table 4
Initial and optimum Design Variables (DVs).
DVs Initial Optimum Units

Geometry Lcentral/180 0.50 0.52 –

x1 10.0 5.7 m
x2 8.6 6.8 m
x3 4.3 4.4 m
z1/Lcentral 0.20 0.13 – Fig. 10. Initial and optimum design deck cross section (dimensions in mm).
z2/z1 0.2 0.02 –
y2 3.0 0.3 m
y1 6.0 7.6 m
Ncables 3 4 –
The tower cross sections decreased and converged to a square
hollow section SHS 410-410-12 both below and above the deck (Fig. 9).
Control ky 100 317 kN/m
The deck cross section slightly increased in terms of material and re-
kz 10 10 kN/m
cy 477 513 kNs/m duced the transverse inertia while increasing the vertical stiffness. As a
cz 19 731 kNs/m consequence the bridge became in the transverse direction slender
hslab 5.0 4.0 cm (50% reduction of the width) and, to compensate this effect, the tower-
deck connection transverse stiffness (ky) was increased threefold in
Caption: Underline: User-defined minimum allowed value.
order to guarantee the bridge transverse stiffness. Also the tower base
distance (y1) was higher in the optimum design and allowed to increase
Table 5
the tower stiffness in the transverse direction. The vertical connection
Initial and optimum cable prestress and area.
stiffness (kz) started with its minimum pre-defined value of 10 kN/m
Cable prestress (kN) Cable area (cm2) (which is almost the same as being null in practical terms) and re-
mained constant throughout the optimization process. This happens
Initial Optimum Initial Optimum
because an increase in this parameter would result in a rapid increase of
Side span Cable s.1 227 496 24 9.2 the negative bending moment in the deck at the tower intersection.
Cable s.2 309 473 33 8.7 Also, it would decrease the effect of the vertical damper (cz) by blocking
Cable s.3 220 293 24 5.3 the connection and, in consequence, lower damping would be added to
Cable s.4 – 202 – 2.7
the vertical modes.
Central span Cable c.1 209 205 22 2.8 Once the cable prestress Tc , the cable length Lc and mass mc are
Cable c.2 253 278 27 5.1
known it is possible to find the cable fundamental frequency
Cable c.3 302 484 32 8.2 1
Cable c.4 – 482 – 8.4 fc = 2L Tc . Since the cable-structure interaction was not considered in
c mc
the finite elements model the authors checked for the cables funda-
mental frequencies and concluded that they were between 3 and 12 Hz,
meaning that they fall outside the critical dynamic design ranges and
the cable structure interaction can be neglected.

Table 6
Cost distribution in the initial and final solutions.
Initial Optimum

Deck 73.4% 82.9%

Towers 12.9% 8.7%
Cables 12.0% 6.2%
Fig. 9. Initial and optimum design tower cross section dimensions in mm (left – Concrete 1.6% 2.2%
below deck, right – above deck).

519
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

1.3 Table 7
1.2 Optimum
Initial and optimum design factor for different design criteria.
Deck Design Factor

1.1 Initial
1 Design criteria Initial Optimum
0.9
0.8 Deck Buckling 0.48 0.58
0.7 Tower Buckling 0.95 0.99
0.6 Tower uy displacement 0.24 0.50
0.5 Tower ux displacement 0.74 0.93
0.4 Deck uy displacement 0.12 0.64
0 30 60 90 120 150 180 Deck uz displacement 0.69 0.86
Coordinate (m) Self-weight deck uz 1.21 0.89

Table 8
Initial and optimum cable tension design factor.

Fig. 11. Initial and optimum design factor along the bridge deck. Cables Initial Optimum

Cable s.1 0.42 0.91

Cable s.2 0.23 0.75
None of the cross sections dimensions were constrained by pre- Cable s.3 0.23 0.85
defined minimum or maximum values with the exception of the Bdeck Cable s.4 – 0.98
which converged to the predefined minimum value of 0.5 of the width Cable c.1 0.23 0.98
Cable c.2 0.24 0.92
of the bridge (0.5 × 4 m = 2 m in this case).
Cable c.3 0.22 0.75
Table 6 presents the cost distribution of the different structural parts Cable c.4 – 0.58
in both designs and it can be seen that, in this example, the optimiza-
tion algorithm increases the bridge deck cross section and, in con-
sequence, the cost. One of the possible explanations for this is that it
was necessary, in order to mitigate the structural dynamic response. section (Fig. 11);
• The tower stresses at the top, caused by high bending moments in
5.2. Optimal design static behavior the top of the tower. This is a result of the different value of the
horizontal component of the cable forces which impose high lateral
In this section, the static response of both the initial and optimum forces in the top of the tower (Fig. 12). This is one of the reasons
designs are displayed and analyzed. The deck and tower design factors why the algorithm changes the tower shape to an inverted ‘A’ in
which take into account all the elements internal forces (N, Vz, Vy, T, My order to center the cable anchorages as close as possible;
and Mz) interaction are presented in Figs. 11 and 12 respectively, a
resume of the most relevant static criteria (Table 3) is presented in The optimum design on the other hand is governed by:
Table 7 and the cables design factors are presented in Table 7. In
Figs. 11 and 12 only the design factor envelope is shown, and the same • The deck stresses in the middle of the central span, caused by the
option was adopted in Tables 7 and 8 where the worse value between high bending moment at this point (Fig. 11). The deck design is
the entire load combinations (Table 2) are presented. governed by the combination ULS-5 and 6 (Table 2) where the
It can be seen that the initial design is governed by: central span is loaded and the side spans are unloaded, being a
consequence of the influence line of the bending moment in the
• The deck stresses, in the entire length of the structure as a con- deck.
sequence of unbalanced internal forces with higher values at the • The tower high compression and bending moment below the deck
intersection with the tower caused by a higher compression in this (Fig. 12) and the tower buckling of the cross section above the deck
(Table 7).
20
By analysing Table 8, it can be seen that the cable resistance utili-
Optimum
zation is more appropriate. In the initial design the cable cross section
15
Initial
was higher than necessary (Table 5) and in the optimum design the
cable areas are reduced while satisfying the maximum allowable ten-
sion leading to a much more efficient design.
10 Fig. 13 presents the structural response for three static load com-
binations and it can be seen that the cable prestress induce a 26 cm
z Coordinate (m)

precamber initial configuration for the unloaded structure (Fig. 13b).

5 Since the structure is very light and flexible and live loads are higher
when compared to the weight of the structure, this is an expected
outcome of the design as having an initial upward deflection will lower
0 the displacements and stresses when the structure is subject to the live
0.4 0.6 0.8 1 1.2 loads. This option also increases the tension in the cables and conse-
quently reduces the cable nonlinear behavior thus increasing the bridge
-5 stiffness in the unloaded case. It is interesting to note slight deck torsion
in Fig. 13c that is a consequence of the considered torsion eccentricities
as explained in Section 2.2. Nonetheless it can be viewed that the tor-
-10 sion of the deck is negligible when compared to the magnitude of the
Tower Design Factor vertical displacements.
Fig. 12. Initial and optimum design factor along the bridge tower height.

520
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

a) SLS-3

29cm

b) SLS-7

26cm

c) ULS-4

43cm

Fig. 13. Bridge elevation showing the static response for three different combinations. (a) SLS-3, (b) SLS-7 and (c) ULS-4.

8 in the tower-deck connection.

Optimum
7 An interesting observation was that, even after the width of the deck
Vertical acceleration (m/s2)

Initial being reduced substantially, the first frequency remains approximately

6
5
Limit
4
3 optimum design as can be seen in Table 4) make a very important
2 contribution to the lateral stiffness of the structure. Another effect is the
1 change from ‘H’ to ‘A’ shaped tower and the decrease of the tower
0
1 1.2 1.4 1.6 1.8 2 2.2
Pedestrians Walking Frequency (Hz)
Fig. 14. Initial and optimum design dynamic response when subject to a pe-
destrian induced vibration.
5.3. Optimal design dynamic properties
In this example, the dynamic criteria are a governing aspect of the
bridge design. Fig. 14 shows the initial and optimum design vertical
accelerations of the structure when subject to a crowd of pedestrians
and Table 9 presents the first structural modes of vibration. It can be
seen that the initial design violates the 2.5 m/s2 limit by a very high
amount as a consequence of the resonance of the 5th, 11th and 13th
modes of vibration (Table 9). The optimum design accomplishes the
acceleration limit and is governed by the 5 and 6th modes resonance. It
is interesting to see that the 5th mode of the optimum design is rela-
tively far from the critical range [1.7–2.1 Hz]. Since this mode could not
be damped adequately the optimization algorithm decreased the
structural frequency to reduce the vertical excitation of the structure.
On the other hand the 6th mode is controlled using the viscous dampers
constant (and in fact increases slightly) which is rather counter-in-
tuitive. In fact, the increase of the deck height and the transverse
connection stiffness ky (which more than triples from the initial to the
height, which in turn increases the cable transversal inclination and, in
2.4 consequence, the lateral stiffness of the structure.
As expected, the lateral instability of the footbridge was one of the
governing design criteria, in the initial design the crowd density that
induces the synchronous lateral instability of the structure was only
0.7 pedestrians/m2. In both cases the structure is so flexible in the
transverse direction that the first mode of vibration falls outside the
critical ‘lock-in’ range of [0.5–1 Hz] by having a lower frequency and
thus being more flexible.
It can be seen that in the initial design the transverse damper in the
tower-deck connection was not well designed since the first mode of
vibration was ‘overdamped’ and was not critical to the design, while no
damping was added to the 3rd mode that was inside the critical ‘lock-in’
range. In the optimum design the algorithm is efficient in finding a
compromise between the damping in the 2nd and 4th modes of vibra-
tion and was able to meet the design crowd density of 1.5 pedestrians/
m2.
Fig. 15 shows the most important modes of vibration that govern
the design in particular in the final solution.

521
F. Ferreira, L. Simões Engineering Structures 180 (2019) 510–523

Table 9
Initial and optimum design modes of vibration.
Initial Optimal

*
Mode Frequency (Hz) Type ξ (%) Frequency (Hz) Type ξ (%)*

1 0.26 Transverse 40.0% 0.27 Transverse 19.1%

2 0.82 Vertical 0.3% 0.63 Transverse 22.4%
3 0.97 Transverse 0.3% 0.71 Vertical 0.3%
4 1.01 Vertical 20.4% 1.05 Transverse 6.5%
5 1.61 Vertical 0.3% 1.32 Vertical 0.3%
6 1.86 Vertical 3.3% 1.80 Vertical 1.1%
7 1.86 Vertical 3.3% 1.80 Transverse 0.3%
8 1.89 Torsion 17.1% 1.96 Vertical 3.2%
9 1.89 Tower – local 0.3% 2.48 Vertical 6.6%
10 2.09 Torsion 15.9% 2.69 Transverse 0.3%
11 2.24 Vertical 0.3% 3.17 Vertical 27.8%
12 2.42 Torsion 0.3% 3.58 Vertical 0.4%
13 2.43 Vertical 0.3% 3.74 Tower – local 0.4%

Caption: Bold: Modes of vibration susceptible to vertical accelerations.

Underline: Modes of vibration susceptible to ‘lock-in’.
* The damping value for the modes is obtained from the eigenvalues of the state-space matrix A (Eq. (6)).

a) 1st Mode b) 2nd (optimum) and 3rd

(initial) Mode

c) 5th Mode d) 6th Mode

Fig. 15. Initial (dashed red) and optimum design (solid blue) relevant modes of vibration. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

6. Conclusions the deck.

An optimization algorithm is proposed for the integrated design of
three dimensional cable–stayed steel footbridges with control devices.
The algorithm modifies the structural geometry, section sizing, damper
properties and cable prestress, to find the minimum cost design that
meets all the defined criteria.
A 180 m total length (approximately 90 m central span) cable-
stayed footbridge is designed using the proposed methodology, in-
cluding the geometric non-linearity, dynamic induced loads and syn-
chronous lateral excitation instability. The objective is to illustrate how
such a complex problem can be solved efficiently.
The procedure was efficient in finding a stable and lower cost so-
lution. The algorithm was able to change the structural dynamic
properties and to design the dampers to control the most relevant
modes of vibration.
The results demonstrate that, for this example, the fan type bridge
with inverted ‘A’ shaped towers is the most efficient design. This option
avoids the large bending moments at the top of the tower and stresses in
The algorithm was able to shift the structural frequencies to fall
outside the critical ranges for the vertical accelerations and for the
‘lock-in’ phenomena. To meet the vertical acceleration and the ‘lock-in’
criteria in the remaining modes the algorithm changed the tower and
cable locations and modified the tower-deck stiffness and damping
properties. This is important for the dynamic design because if a stiff
tower-deck connection is employed the structural frequencies will in-
crease but no damping will be added, on the other hand if the con-
nection is free, the frequencies will decrease but, also, no damping will
be added. The connection properties are also important for the static
design criteria, in particular the deck displacements.
The algorithm is able to find an efficient solution that guarantees all
the standard ultimate and service limit states with minimum cost.
References
[1] Dallard P, Fitzpatrick A, Flint A, et al. The London millennium footbridge. Struct
Eng 2001;79(22):17–33.

522
F. Ferreira, L. Simões
[2] Setra/AFGC. Footbridges – assessment of vibrational behavior of footbridges under
pedestrian loading. Technical guide; October 2006.
[3] Brownjohn J, Fok P, Roche M, Moyo P. Long span steel pedestrian bridge at
Singapore Changi Airport – part 1: prediction of vibration serviceability problems.
Struct Eng 2004;82(16):21–7.
[4] Brownjohn J, Fok P, Roche M, Omenzetter P. Long span steel pedestrian bridge at
Singapore Changi Airport – part 2: crowd loading tests and vibration mitigation
measures. Struct Eng 2004;82(16):28–34.
[5] Caetano E, Cunha Á, Magalhães F, Moutinho C. Studies for controlling human-in-
duced vibration of the Pedro e Inês footbridge, Portugal. Part 1: assessment of dy-
namic behavior. Eng Struct 2010;32(4):1069–81.
[6] Caetano E, Cunha Á, Moutinho C, Magalhães F. Studies for controlling human-in-
duced vibration at the Pedro e Inês Footbridge, Portugal. Part 2: implementation of
tuned mass dampers. Eng Struct 2010;32(4):1082–91.
[7] Pálossy M, Szabó G, Szecsányi L. “Mayfly” footbridge, Szolnok – design, construc-
tion and dynamic behavior of the longest footbridge in Hungary. Steel Constr
2011;4(3):193–202.
[8] McRobie A, Winslow P. The lateral dynamic stability of Stockton infinity footbridge
using complex modes. Struct Eng Int 2012;22:545–51.
[9] Bursi S, Kumar A, Abbiati G. Identification, model updating, and validation of a
steel twin deck curved cable-stayed footbridge. Comput-Aid Civ Infrastruct Eng
2014;2014(29):703–22.
[10] CEN (European Committee for Standardization). Basis of structural design.
Eurocode 0, Brussels; 2002.
[11] CEN (European Committee for Standardization). Actions on structures. Eurocode 1,
Brussels; 2002.
[12] CEN (European Committee for Standardization). Design of steel structures.
Eurocode 3, Brussels; 2005.
[13] Hewson N. Prestressed concrete bridges: design and construction. London: Thomas
Telford Publishing; 2003.
[14] Leonhardt F. Cable stayed bridges with prestressed concrete. J Prestress Concr Inst
1987;32(5):1–30.
[15] Baldomir A, Hernandez S, Nieto F, Jurado JA. Cable optimization of a long span
cable stayed bridge in La Coruña. Adv Eng Softw 2010;41:931–8.
[16] Chen Q. Optimization of cable-stretching planning in the construction of cable-
stayed bridges. Eng Optim 1992;19:1–20.
[17] Furukawa K, Sugimoto H, Egusa T, et al. Studies on the optimization of cable
prestressing for cable-stayed bridges. Paper presented at the international con-
ference on cable-stayed bridges, Bangkok, Thailand. 1987.
[18] Furukawa K, Sakai I, Kumagai, et al. Optimization of cable forces in cable-stayed
prestressed concrete bridges based on minimum strain energy criterion. Paper
presented at the international conference on cable-stayed bridges, Bangkok,
Thailand. 1987.
[19] Granata M, Longo G, Recupero A, Marcello A. Construction sequence analysis of
long-span cable-stayed bridges. Eng Struct 2018;2018(174):267–81.
[20] Hassan M, Nassef S, El Damatty A. Determination of optimum post-tensioning cable
forces of cable-stayed bridges. Eng Struct 2012;44:248–59.
[21] Janjic D, Pircher M, Pircher H. Optimization of cable tensioning in cable stayed
bridges. J Bridge Eng 2003;8(3):121–7.
[22] Kasuga A, Arai H, Breen J, Furukawa K. Optimum cable-force adjustments in con-
crete cable-stayed bridges. J Struct Eng ASCE 1995;121(4):685–94.
[23] Martins A, Simões L, Negrão J. Cable stretching force optimization of concrete
cable-stayed bridges including construction stages and time-dependent effects.
Struct Multidiscip Optim 2015;51(3):757–72.
[24] Martins A, Simões L, Negrão J. Optimization of cable forces on concrete cable-
stayed bridges including geometrical nonlinearities. Comput Struct
2015;155:18–27.
[25] Song C, Xiao R, Sun B. Optimization of cable pre-tension forces in long-span cable-
stayed bridges considering the counterweight. Eng Struct 2018;2018(172):919–28.
[26] Sung Y, Chang D, Teo E. Optimum post-tensioning cable forces of Mau-Lo Hsi cable-
stayed bridge. Eng Struct 2006;28:1407–17.
[27] Hassan M. Optimization of stay cables in cable-stayed bridges using finite element,
genetic algorithm, and B-spline combined technique. Eng Struct 2013;49:643–54.
[28] Hassan M, Nassef S, El Damatty A. Optimal design of semi-fan cable-stayed bridges.
Can J Civ Eng 2013;40:285–97.
[29] Long W, Troitsky M, Zielinski Z. Optimum design of cable stayed bridges. J Struct
Eng Mech 1999;7(3):241–57.
[30] Lute V, Upadhyay A, Singh K. Computationally efficient analysis of cable-stayed
bridge for GA-based optimization. Eng Appl Artif Intell 2009;22:750–8.
[31] Martins A, Simões L, Negrão J. Optimum design of concrete cable-stayed bridges.
Eng Optim 2015. https://doi.org/10.1080/15732479.2014.931976.
[32] Martins A, Simões L, Negrão J. Optimum design of concrete cable-stayed bridges
with prestressed decks 5–6 Int J Comput Method Eng Sci Mech 2017;17:339–49.
[33] Ohkubo S, Taniwaki K. Shape and sizing optimization of steel cable-stayed bridges.
In: Hernandez S, Brebbia CA, editors. Proceedings of OPTI 91 – optimization of
structural systems and industrial applications. Cambridge, MA, USA: Elsevier
Applied Sciences; 1991.
[34] Simões L, Martins A, Negrão J. Optimization of concrete cable-stayed bridges with
discrete design variables. In: Schumacher A, Vietor T, Fiebig S, Bletzinger KU,
523
View publication stats
Engineering Structures 180 (2019) 510–523
Maute K, editors. Advances in structural and multidisciplinary optimization.
WCSMO 2017. Cham: Springer; 2018.
[35] Simões L, Negrão J. Sizing and geometry optimization of cable-stayed bridges.
Comput Struct 1994;52(2):309–21.
[36] Torii K, Ikeda K. A study of the optimum design method for cable-stayed bridges.
Paper presented at the international conference on cable-stayed bridges, Bangkok,
Thailand. 1987.
[37] Freire A, Negrão J, Lopes A. Geometrical nonlinearities on the static analysis of
highly flexible steel cable-stayed bridges. Comput Struct 2006;84:2128–40.
[38] Nazmy A, Abdel-Ghaffar A. Three-dimensional nonlinear static analysis of cable-
stayed bridges. Comput Struct 1990;34(2):257–71.
[39] Negrão J, Simões L. Optimization of cable-stayed bridges with three-dimensional
modeling. Comput Struct 1997;64(1–4):741–58.
[40] Simões L, Negrão J. Optimization of cable-stayed bridges subjected to earthquakes
with non-linear behaviour. Eng Optim 1999;31:457–78.
[41] Simões L, Negrão J. Optimization of cable-stayed bridges with box-girder decks.
Adv Eng Softw 2000;31:417–23.
[42] Simões L, Negrão J. Reliability optimum design of Glulam cable-stayed foot-bridges.
J Bridge Eng ASCE 2005.
[43] Somja H, Ville de Goyet V. A new strategy for analysis of erection stages including
an efficient method for creep analysis. Eng Struct 2008;30:2871–83.
[44] Camara A, Efthymiou E. Deck–tower interaction in the transverse seismic response
of cable-stayed bridges and optimum configurations. Eng Struct
2016;2016(124):494–506.
[45] Ferreira F, Simões L. Optimum design of a controlled cable stayed bridge subject to
earthquakes. Struct Multidiscip Optim 2011;2011(44):517–28.
[46] Ferreira F, Simões L. Optimum cost design of controlled cable stayed footbridges.
Comput Struct 2012;2012(106–107):135–43.
[47] Ferreira F, Simões L. Optimum design of a controlled cable stayed footbridge sub-
ject to a running event using semi-active and passive mass dampers. J Perform
Constr Facil 2018. [in press].
[48] Tzan S, Pantelides C. Convex model for seismic design of structures—II: design of
conventional and active structures. Earthquake Eng Struct Dyn
1996;1996(25):945–63.
[49] Khot N. Multicriteria optimization for design of structures with active control. J
Aerospace Eng (ASCE) 1998;11(2):45–51.
[50] Cimellaro G, Soong T, Reinhorn A. Integrated design of inelastic controlled struc-
tural systems. Struct Control Health Monit 2009;2009(16):689–702.
[51] Ernst JH. Der E-Modul von Seilen unter berucksichtigung des Durchhanges. Der
Bauingenieur 1965;40(2):52–5. [in German].
[52] Baldomir A, Kusano I, Hernández S, Jurado J. A reliability study for the Messina
Bridge with respect to flutter phenomena considering uncertainties in experimental
and numerical data. Comput Struct 2013;2013(128):91–100.
[53] Kusano I, Baldomir A, Jurado J, Hernández S. Reliability based design optimization
of long-span bridges considering flutter. J Wind Eng Ind Aerodyn
2014;2014(135):149–62.
[54] Kusano I, Baldomir A, Jurado J, Hernández S. The importance of correlation among
flutter derivatives for the reliability based optimum design of suspension bridges.
Eng Struct 2018;2018(173):416–28.
[55] Larsen A, Larose G. Dynamic wind effects on suspension and cable stayed bridges. J
Sound Vibrat 2015;334(2015):2–28.
[56] Montoya M, Hernández S, Nieto F. Shape optimization of streamlined decks of
cable-stayed bridges considering aeroelastic and structural constraints. J Wind Eng
Ind Aerodyn 2018;177:429–55.
[57] Xiuli X, Zhijun L, Weiqing L, Dongming F, Xuehong L. Investigation of the wind-
resistant performance of seismic viscous dampers on a cable-stayed bridge. Eng
Struct 2017;2017(145):283–92.
[58] Pedro J, Reis A. Simplified assessment of cable-stayed bridges buckling stability.
Eng Struct 2016;2016(114):93–103.
[59] Zivanovic S, Pavic A, Reynolds P. Probability-based prediction of multi-mode vi-
bration response to walking excitation. Eng Struct 2007;2007(29):942–54.
[60] Ingólfsson E, Georgakis C, Jönsson J. Pedestrian induced lateral vibrations of
footbridges: a literature review. Eng Struct 2012;2012(45):21–52.
[61] Lievens K, Lombaert G, Nimmen K, Roeck G, Broeck P. Robust vibration service-
ability assessment of footbridges subjected to pedestrian excitation: strategy and
applications. Eng Struct 2018;2018(171):236–46.
[62] Racic V, Pavic A, Brownjohn J. Experimental identification and analytical modeling
of human walking forces: literature review. J Sound Vib 2009;326(1–2):1–49.
[63] Ricciardelli F, Demartino C. Design of footbridges against pedestrian-induced vi-
brations. J Bridge Eng 2016;21(8).
[64] AISC. Floor vibrations due to human activity (Design guide 11). Chicago, USA:
American Institute of Steel Construction; 1997.
[65] Wesolowsky M, Irwin P, Galsworthy J, Bell A. Human-induced vibrations in
building. Int J High Rise Build 2012;1(1):15–9.
[66] Simões L, Templeman A. Entropy-based synthesis of pretensioned cable net struc-
tures. Eng Optim 1989;15:121–40.
[67] Cheng G, Liu Y. A new computation scheme for sensitivity. Eng Optim
1987;12(3):219–34.