Engineering Structures 45 (2012) 21–52

Contents lists available at SciVerse ScienceDirect

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

Review article

Pedestrian-induced lateral vibrations of footbridges: A literature review

E.T. Ingólfsson ⇑, C.T. Georgakis, J. Jönsson
Department of Civil Engineering, Technical University of Denmark, Building 118, Brovej, 2800 Kgs. Lyngby, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: The earliest scientific descriptions of excessive pedestrian-induced lateral vibrations are dated back to
Received 12 September 2011 the 1970s, but it was not until the beginning of the new millennium that bridge engineers fully compre-
Revised 23 March 2012 hended the potential negative effect of pedestrian crowds on long-span footbridges. Following the unex-
Accepted 23 May 2012
pected serviceability failures of Paris’ Solférino and London’s Millennium footbridges in 1999 and 2000, a
Available online 15 July 2012
new tract of research was initiated, focused on understanding the phenomenon which has become
known as Synchronous Lateral Excitation (SLE). In this paper, a comprehensive review of studies related
Keywords:
to pedestrian-induced lateral vibrations of footbridges is provided, primarily focusing on studies pub-
Footbridges
Lateral vibration
lished within the last decade. Research in this field can generally be split into three categories; (i) full-
Human-structure interaction scale testing of existing bridges subject to crowd loading, (ii) laboratory studies on human-structure
Full-scale testing interaction between single pedestrians and laterally moving platforms and (iii) mathematical modelling
Load models of the pedestrian-induced load. It is shown herein, that a significant amount of research has been carried
Experimental investigations out within each of the three categories, but there is only limited interconnection, particularly between
Ground reaction forces the mathematical models on one side and the empirical observations on the other. The main purpose
of this review is to provide this link, through a detailed and critical review of publications within each
of the three categories.
Ó 2012 Published by Elsevier Ltd.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2. Early cases of excessive lateral bridge vibrations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3. Lateral footstep forces on a rigid surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4. Paris pont de Solférino and London Millennium Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1. Full scale testing of pont de Solférino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2. Full scale testing of the London Millennium Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3. Laboratory tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.4. Early analytical approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.5. Lessons learned from Paris and London. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5. Lateral footstep forces on a moving surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1. Pedestrian walking on moving surfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2. Quantification of lateral forces using instrumented platforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.3. Walking on laterally moving instrumented treadmills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6. Full scale measurements: case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.1. Changi Mezzanine Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.2. Nasu Shiobara Bridge (M-Bridge) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.3. Lardal footbridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.4. Coimbra Footbridge. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.5. Passarelle Simone de Beauvoir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.6. Clifton Suspension Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.7. Weil-am-Rhein footbridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.8. Summary of full-scale measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7. Developments in the modelling of pedestrian induced lateral excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

⇑ Corresponding author.
E-mail address: eti@byg.dtu.dk (E.T. Ingólfsson).

0141-0296/$ - see front matter Ó 2012 Published by Elsevier Ltd.

http://dx.doi.org/10.1016/j.engstruct.2012.05.038
22 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

7.1. Dynamics of a moving crowd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

7.1.1. Spatially unrestricted pedestrian flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.1.2. Human–human dynamic interaction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.2. Linear response models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2.1. Equivalent number of ‘perfect’ pedestrians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2.2. Stability criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2.3. Stationary response due to random crowds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.2.4. Stationary response due to random crowds including spatial correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.2.5. Peak response due to random crowds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3. Nonlinear dynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3.1. Modified Arup models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3.2. Amplitude dependent DLF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7.3.3. Modal coupling (autoparametric resonance) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7.3.4. Parametric resonance excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.3.5. Pedestrian phase synchronisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.3.6. Pendulum walking models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.3.7. Continuous hydrodynamic crowd modelling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
8. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

1. Introduction Despite recent research efforts, several questions relating to

With the promise of extensive urban regeneration, isolated seg-
ments of the landscape are being increasingly connected by archi-
tecturally novel and challenging footbridges [1–3]. This has led to a
boom in footbridge design and construction, allowing recent exam-
ples to easily match road bridges in terms of cost and span [4]. The
challenges in footbridge design lie in the fulfillment of the architec-
tural demands for long, light and slender structures. And whilst the
architects and engineering consultants are constantly improving
material usage and cost, many new footbridges are experiencing
excessive vibrations for which extensive retrofit costs are being in-
curred. The temporary closures of both pont de Solférino in Paris in
1999 and then the London Millennium Bridge in 2000, following
excessive pedestrian-induced lateral vibrations during their inau-
guration [5,6], are probably the most famous and publicised cases
of this. Subsequent research revealed that the potential negative
effect of pedestrian-induced lateral forces on footbridges was not
limited to the innovative design of the two bridges, but had been
observed on several other bridges of different shapes, sizes and
functions in the past.
An international conference devoted to the design and dynamic
behaviour of footbridges (also known as ‘‘Footbridge 2002’’) was
established in Paris in 2002 and with 69 paper contributions it at-
tracted researchers, engineers and architects from all around the
world. Several working groups were created in the beginning of
this century to define design guidelines for footbridges [7]. The first
international guide was published in 2005 by fib (fédération inter-
nationale du béton), dealing with the general design of footbridges
[8]. More recent documents, specifically dealing with the dynamic
behaviour of footbridges, include the widely used guideline from
the French road authority Sétra [6], the European research project
SYNPEX [9–12], the work commissioned by the UK Highway
Agency [13–17], which was later adopted in the UK National Annex
to Eurocode 1991-2:2003 [18,19] and the book entitled ‘‘Foot-
bridge Vibration Design’’ [20], which is based on the scientific con-
tributions of a workshop arranged during the third international
Footbridge conference (Footbridge 2008).
Today, an increasing number of engineers put an effort in the
design stage to accommodate the potential threat of pedestrian
induced lateral vibrations [21–24]. Typical countermeasures in-
volve the design of vibration mitigation devices such as Tuned
Mass Dampers (TMDs), with subsequent experimental modal
analysis and pedestrian response tests for tuning and verification
purposes [25–28].
lateral pedestrian induced vibrations still remain unanswered.
Many examples of bridges have been reported as ’lively’ and an
increasing number of full scale pedestrian crowd tests have been
performed to verify the existence of ‘‘Synchronous Lateral Excita-
tion’’ (SLE) and to determine the critical number of pedestrians
needed to trigger SLE. However, there is a general dispute about
the basic mechanism of pedestrian-induced lateral forces and sev-
eral hypotheses and pedestrian load models exist.
In recent years several papers have been published within the
field of vibration serviceability of structures. It is especially worth
mentioning the review devoted to footbridge structures by Živano-
vić et al. [29], in which a comprehensive overview of the entire
field of vibration serviceability is given. The review by Racic et al.
[30] focused on modelling of human induced walking forces, lar-
gely seen from a bio-mechanical point of view and that of Venuti
and Bruno [31] which is primarily focused on crowd modelling
for footbridge applications are worth highlighting.
The purpose of this paper is to provide a comprehensive review
of the state-of-the-art, with emphasis on published cases of pedes-
trian-induced lateral bridge vibrations, the results from experimen-
tal studies and the recent development in modelling of pedestrian
loading and its effect on structures. The content of this review is pre-
sented (nearly) chronologically and can be divided into five parts.
Following an introduction into early cases of lateral vibrations (Sec-
tion 2) and lateral ground reaction forces from pedestrians on a rigid
floor (Section 3), a thorough presentation of the work made during
the retrofit period of the Solférino Bridge in Paris and the London
Millennium Bridge is given (Section 4). Since 2002, more research-
ers have been devoted to the determination of pedestrian loads on
flexible footbridges and several attempts have been made to quan-
tify the interaction between a pedestrian and a laterally moving
structure. The main results from experimental studies on lateral
loads on a moving structure are summarised in Section 5 and results
from reported full scale measurements of lively pedestrian bridges
are presented in some detail in Section 6. In the final section, a de-
tailed description of the development of pedestrian load models
and response evaluation techniques is provided.
2. Early cases of excessive lateral bridge vibrations
To the authors’ knowledge, the earliest reported incidents of
excessive lateral vibrations induced by crowds are dated back to
the late 1950s, one involving a road/railway bridge in China (the
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
Fig. 1. Steel footbridge across Main at Erlach (a) and Toda Park Bridge in Toda City, Japan (T-Bridge) (b) and (c). (Pictures after http://www.karl-gotsch.de/Bilder (a), [44] (b)
and [45] (c).)
Wuhan Yangtze Bridge) in 1957 [32] and another one involving a
pedestrian suspension bridge in Kiev following its opening in
1958 [33].
In fact, several large road bridges around the world have suf-
fered from this problem during exceptional crowd events, such
as opening day events [34], public demonstrations or festive events
[5,35,36]. Even the 120 year old Brooklyn Bridge in New York City
swayed remarkably when traversed by crowds of pedestrians dur-
ing a power black out, leading to several complaints from con-
cerned citizens [37,38].
The first assessment of lateral crowd induced excitation was of-
fered by Petersen [39] (as reviewed by Bachmann and Ammann
[40]), who observed strong lateral vibrations of a steel arch foot-
bridge at Erlach in Germany (Fig. 1a), during crossing of about
300–400 pedestrians. The vibrations occurred on the 110 m long
main span of the bridge at frequency around 1.1 Hz and were ex-
plained as a consequence of a lateral sway of the centre of gravity
of the human body occurring at half the pacing frequency, result-
ing in resonant vibrations and a synchronisation of the step with
the oscillation of the bridge [41]. In this particular case, the vibra-
tion problem was solved by installing a horizontally acting TMD.
One of the most cited incidents of excessive lateral vibrations
occurring in the last century is related to the Toda Park Bridge in
Toda City, Japan (T-Bridge) [42,43]. The bridge is a cable-stayed
bridge with the overall length of 179 m divided into a main span
(134 m) and a side span (45 m) (Fig. 1b). The frequency of the fun-
damental vibration mode was reported in the range 0.9–1.0 Hz
depending on the level of congestion. The bridge that connects a
stadium and a bus terminal was traversed by around 20,000 pedes-
trians following a boat race with up to 2000 people simultaneously
on the bridge (crowd density of approximately 2.1 ped/m2)
(Fig. 1c). During this event, the lateral acceleration of the bridge
girder was an order of magnitude larger than predicted when
assuming the resonance step frequency of all pedestrians and
mutually independent (random) phases. Video analysis show that
up to 20% synchronised their head movement to that of the bridge.
This observation was used to describe the phenomenon of human-
structure phase synchronisation (or ‘‘lock-in’’) as follows [42]:
‘‘First a small lateral motion is induced by the random lateral hu-
man walking forces, and walking of some pedestrians is synchron-
ised to the girder motion. Then resonant force acts on the girder,
consequently the girder motion is increased. Walking of more
23
pedestrians are synchronised, increasing the lateral girder motion.
In this sense, this vibration has a self-excited nature. Of course, be-
cause of adaptive nature of human being, the girder amplitude will
not go to infinity and will reach a steady-state’’. It is worth noting
that this description, although more detailed, is essentially similar
to that of Petersen [39], from 1972. It is implicitly assumed here
that all pedestrians synchronise in the same manner, i.e. with the
same relative phase between step frequency and bridge move-
ment, leading to an increase in the bridge vibration.
3. Lateral footstep forces on a rigid surface
During walking, the ground reaction force (GRF) occurs due to
acceleration (and deceleration) of the centre of mass (CoM) of
the body. The GRF is a three dimensional vector which varies in
time and space due to the forward movement of the person [30].
Early studies on the lateral component of the GRFs were carried
out by Harper et al. [46], which revealed that the horizontal lateral
component of the force was generally very small and that it is
caused by balancing of the body during walking. Andriacchi et al.
[47] showed that the peak force amplitudes (vertical and lateral)
increase numerically with an increase in the walking speed. Similar
observations were made more recently by Masani et al. [48]. Chao
et al. [49] measured single footstep forces from several persons and
found that the peak lateral forces (F1 to F3 in Fig. 2) are around 4–
5% of the body weight for men and little less for women, but with a
considerable difference between individuals.
Several parameters influence the shape of the GRFs which are
governed by large- intra and inter-subject variability. The intra-
subject variability is related to changes in the GRF from the same
person, measured at two different time instances, whereas the in-
ter-subject variability refers to the variability between different
people [29,30]. Variations in the gait parameters during continuous
walking is a form of intra-subject variability which causes random
fluctuations in the shape of the GRF from each footstep. However,
perfect periodicity in the walking is often assumed, as it implies
that the force time history of a series of consecutive steps can be
modelled as a Fourier series. In most cases, the vertical and hori-
zontal components of the GRF are treated separately. Herewith,
only the horizontal lateral component is discussed. For this compo-
nent, the fundamental harmonic is taken as the duration of two
consecutive steps (Fig. 2):
24 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

Sum

Lateral force [N]

40 Single footstep 40
20 20 Right foot
F2 F3
0 0
-20 -20
Left foot
-40 F1 -40 Tlateral
time [s] time [s]

Fig. 2. Typical shape of the lateral component of the walking force (figure after [29]).

X
n
F L ðtÞ ¼ Gj sinð2pjfw t  /j Þ ð1Þ
Table 1
j¼1
Parameters for PSD in Eq. (2) according to Pizzimenti and Ricciardelli [61].
The fundamental frequency, fw, of the Fourier series therefore j=1 j=2 j=3 j=4 j=5
equals the gait cycle frequency (defined as half the step frequency
Aj 0.96 0.73 0.879 0.55 0.74
fp = 2fw) and Gj and /j represent the load amplitude and phase an- B 0.0616 0.039 0.0288 0.037 0.025
j 
gle of load harmonic j, respectively. Often the load amplitude is de- e
F 2 =e
F2 0.81 0.050 0.277 0.047 0.072
Lj L
fined through the body weight normalised dynamic load factor k
e
F 2L (mean) 0.0012W2
(DLF), DLFj = Gj/W, where W is the body weight. According to Bach-
e
F 2L (characteristic) 0.0020W2
mann and Ammann [40] the values of the first five DLFs are
DLFj = {0.039, 0.01, 0.042, 0.012, 0.015}, j = 1    5. In a later publica-
tion by same authors, the values DLF1 = DLF3 = 0.1 are suggested
for design purposes [50]. Other researchers have similarly deter- As proposed by Ingólfsson and Georgakis [62], a synthetic
mined the DLFs from measured GRFs using instrumented force (pseudo-random) time series of lateral forces can be generated
plates [51–53]. from the spectral density:
Non-zero loads at even integer harmonics implies that the
N1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
walking is imperfect. The intra-subject variability is caused by
F L ðtÞ ¼ 2SF L ðfk ÞDf  cosð2pfk t þ wk Þ ð3Þ
small variations in the load pattern for each step during walking,
k¼0
e.g. due to difference between the GRF of a strong and weak leg NX
Harm
[54]. Brownjohn et al. [55] measured continuous vertical loads SF L ðfk Þ ¼ SF Lj ðfk Þ ð4Þ
using an instrumented treadmill and subsequently suggested that j¼1
the load could be treated as a narrow-band random process. 1 2
Later, Pizzimenti [56] used an instrumented treadmill to mea- Df ¼ ¼ ð5Þ
NDt T tot
sure the lateral component of the GRFs from 66 individuals and
Ricciardelli and Pizzimenti [57] defined DLFs for an average (per- where fk = kDf, k = 0    N  1 are the distinct frequencies for which
fectly periodic) footprint as the sum of the contributions in the the power spectrum ordinates are calculated, N is the total number
Fourier spectra of the measured force in a narrow band around of data points, NHarm is the total number of load harmonics, and Ttot
the frequency of the respective harmonic. The characteristic values is the total duration of the time series. The parameters wk are ran-
(with 95% probability of non-exceedance) of the first five DLFs domly generated phase angles, drawn from a uniform probability
were reported as DLFj,k = {0.04, 0.0077, 0.023, 0.0043, 0.011}, distribution. Therefore, this method can reproduce the spectral con-
j = 1    5. In their model, the band width was taken as Dfj = pfjj tent of the load, but not the temporal features (exact shape and
fw (jfw: frequency of jth load harmonic, fj: a structural damping, as- temporal correlation).
sumed 1%) following the approach of Ohlsson [58] and Eriksson Racic and Brownjohn [63] used an instrumented treadmill to
[59]. measure all three components (vertical, horizontal lateral and lon-
Ingólfsson et al. [60] confirmed these observations as the DLFs gitudinal) of the GRF during continuous walking of 85 volunteers. A
for the so-called ‘‘Narrow-band model’’ are very similar to those re- stochastic time domain model was proposed, which can be used to
ported by Ricciardelli and Pizzimenti [57]. They also showed that generate a realistic time-history of the lateral component of the
the DLFs depend strongly on the selected band-width, which is de- GRF, both in terms of spectral content and temporal features such
fined through the structural damping ratio. As the structural as variations in the duration of individual gait cycles.
damping ratio is generally unknown during design, a ‘‘broad-band According to Sahnaci and Kasperski [64], force plate measure-
model’’, where all contributions in the Fourier spectral between ments only capable of measuring single footsteps introduce a bias
individual harmonics are included, may therefore be more appro- in the GRF opposed to natural walking, as the test subjects have to
priate. The characteristics value of the DLFs (using the bandwidth adjust their walking speed and step length when approaching the
Dfj = fw) were reported as DLFj,k = {0.073, 0.01, 0.034, 0.007, 0.016}, force plate. For this reason they used a platform supported on load
j = 1    5. cells, allowing measurements of GRFs for several consecutive foot-
Due to the non-deterministic nature of the loading, a frequency- steps. The fundamental DLF from 251 (195 male and 56 female)
domain representation was offered by Pizzimenti and Ricciardelli test subjects is shown in Fig. 3.
[61] through a characteristic Power Spectral Density (PSD) for The investigations reviewed in this section are related to the
the first five load harmonics. The general (non-dimensional) form lateral GRFs measured on rigid surfaces, but for bridges with low
was given as: natural frequencies, the effect of the lateral motion on the GRF
(  2 ) and the change in the crowd dynamics due to human-structure
SF Lj ðf Þ  f 2Aj f =ðjfw Þ  1
¼ pffiffiffiffiffiffiffi exp 2 ð2Þ interaction must be accounted for. Until the beginning of this mil-
e
F2 2pBj Bj lennium, only few studies existed regarding the effect of a laterally
Lj
moving surface on the GRFs. However, with the serviceability fail-
where Aj and Bj are parameters determined by the data fit and e
F 2Lj is ures of pont de Solférino in Paris and the Millennium Bridge in Lon-
the area of the PSD around the jth harmonic, see Table 1. don a new tract of research was established, devoted to the
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
Fig. 3. Measured DLFs for the first load harmonic (figure after [54]).
understanding of the interaction between a pedestrian and a later-
ally moving surface and in particular the effect of moving platform
on the lateral GRF.
4. Paris pont de Solférino and London Millennium Bridge
On December 15, 1999, the pont de Solférino footbridge (now
called Passarelle Léopold-Sédar-Senghor), a 140 m long steel arch
footbridge across the Seine in Paris (Fig. 4a), was opened to the
public for crossing. On the opening day unexpected lateral oscilla-
tions were observed and the bridge was subsequently closed to the
public. A comprehensive test program was undertaken which in-
volved modal testing of the structure, pedestrian crowd tests and
installation of 14 TMDs followed by vibration testing and monitor-
ing of the bridge. In November 2000, the bridge was reopened after
almost a year of closure.
The London Millennium Bridge (Fig. 4b), which connects St.
Paul’s Cathedral with the Tate Modern Gallery is the first entirely
new bridge across the Thames in London since Tower Bridge was
completed in 1894 [65]. The bridge is a shallow suspension bridge
in three spans; a south span of 108 m, a central span of 144 m and
a north span of 81 m. The bridge deck consists of aluminium box
sections creating a very light superstructure (2 t/m) [5].
On June 10 2000, between 80,000 and 100,000 people gathered
to cross the bridge on its opening day, with up to 2000 people on
the deck at any one time [5,35]. Large amplitude vibrations in four
different vibration modes were reported; on the Southern span at a
frequency of around 0.8 Hz, at the central span in the first and sec-
ond lateral vibration modes at 0.48 Hz and 0.95 Hz, respectively,
and more rarely on the Northern span at a frequency of around
1 Hz. On June 12 2000, it was decided to close the bridge while a
retrofit solution could be developed and implemented. During
the next 18 months an extensive test program, similar to that in
Fig. 4. Pont de Solférino in Paris (a) (picture after http://www.mimram.com/) and the London Millennium Bridge (b).
25
Paris was undertaken. One of the main observations made on the
opening day was that the bridge exhibited an instability-type
behaviour. The bridge vibrated excessively when congested by a
large crowd of people, but if the number of pedestrians was re-
duced or if they stopped walking, the bridge vibration would re-
duce substantially [5,66].
On June 14, an interesting attempt to explain the cause of the
excessive vibrations was provided by Cambridge Professor Brian
Josephson, in a letter to The Guardian [67], stating that ‘‘the prob-
lem has little to do with crowds walking in step: it is connected
with what people do as they try to maintain balance if the surface
on which they are walking starts to move’’.
4.1. Full scale testing of pont de Solférino
The full scale testing of the bridge identified three critical vibra-
tion modes; a horizontal (lateral) mode with coupled torsional
movement at frequency 0.81 Hz and two torsional modes at
1.94 Hz and 2.22 Hz respectively. The modal mass of the first vibra-
tion mode was reported as approximately 400 t [68]. The charac-
teristics of the first six vibration modes are shown in Table 2.
The damping ratios (of the empty structure) were relatively small,
but increased considerably with the presence of stationary pedes-
trians (in total 116 people) [69]. This human-structure interaction
for passive (standing) people is well known [70], whereas recent
research suggests that also walking pedestrians add damping to
perceptibly vibrating structures in the vertical direction [71–73].
The first series of crowd tests was performed in February 2000
using up to 122 test subjects engaging in various types of rhythmic
activities. Dziuba et al. [69] report that a large number of pedestri-
ans could produce a considerable lateral response of the first mode
with acceleration amplitudes up to 0.6 m/s2 (displacement
24 mm). A total of 14 TMDs (weighing 2500 kg) were installed
on the bridge, 6 of which were designed to suppress horizontal
vibrations. As such, the modal frequency of the fundamental lateral
mode dropped to around 0.7 Hz and damping increased to 3.5%.
In 2002, another test campaign was arranged, involving up to
386 volunteering pedestrians. In the main crowd tests (with inac-
tive TMDs), the number of pedestrians on the bridge was gradually
increased to a maximum of between 207 and 229 people. Based on
the measured responses, an equivalent number of resonance
pedestrians was defined by Charles and Bui [74] as:
rffiffiffi R tþT
p 2 FðsÞqð _ sÞds
Neq ¼ t
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð6Þ
2 T R tþT
F 1p t
_ sÞ2 ds
½qð
where T is the modal period, F(t) is the lateral modal force and F1p is
the amplitude of the first load harmonic of an equivalent single res-
onant pedestrian walking at a frequency that matches the modal
frequency. The modal displacement is denoted q(t) and a dot repre-
26 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

Table 2
Modal properties of pont de Solférino without added damping.

Description Frequency (Hz) Damping (empty) (%) Damping (with 116 people) (%) Modal mass (t)
a a
Symmetric horizontal mode 0.81 (0.70) 0.38 (3.5) 1.6 400
First anti-symmetric vertical mode 1.22 – – –
First anti-symmetric torsional mode 1.59 0.2–0.5 – –
First symmetric vertical mode 1.69 0.49 – –
First symmetric torsional mode 1.94 0.50 1.36 –
Second symmetric torsional mode 2.22 0.28 1.60 –
a
After installation of 6 horizontal TMDs.

0.6
0.5
Degree of synchronisation (-)

0.4
0.3
Acceleration (m/s2)

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
Time (s)

Fig. 5. Lateral acceleration response and equivalent number of pedestrians during circulations of an increasing number of pedestrians walking randomly at slow speed (figure
after [74]).

sents a differentiation with respect to time. The physical interpreta-
tion of Neq is the number of uniformly distributed and synchronised
equivalent pedestrians, which input the same amount of energy
into the mode per oscillation period as that of the entire group
[74]. The modal pedestrian force was derived from the measured
acceleration response.
In Fig. 5, an acceleration time history is shown from one of the
crowd tests together with the so-called ‘‘equivalent degree of syn-
chronisation’’, defined as Neq/N; N being the total number of pedes-
trians. It was concluded that synchronisation of the pedestrians’
movement to that of the bridge occurred when the acceleration
reaches a threshold of about 0.1–0.15 m/s2. Until then, the pedes-
trian walking was reported as random, i.e. the pedestrian walking
characteristics were unaffected by the structural motion. It was
further found that around 140 pedestrians were needed to obtain
acceleration response exceeding this threshold.
4.2. Full scale testing of the London Millennium Bridge
The retrofit solution adopted on the Millennium Bridge consists
of 37 viscous dampers and 29 pairs of vertically acting TMDs. Prior
to their installation, a modal identification of the structure was
performed to validate the theoretical models and to test the effect
of three prototype dampers. Four lateral and four vertical vibration
modes were identified for the empty structure (Table 3) [75–77].
On December 19 2000, 275 Arup employees were used for
crowd tests on the structure; see Fig. 6. In total, 14 different pedes-
trian crowd tests were carried out, most during which the number
of pedestrians on the bridge was gradually increased. In Fig. 7, two
different lateral acceleration response time histories are shown
(measured in two different test series); (a) the response of the
northern span (NL1) and (b) the response of the central span (CL1).
The test showed that for a certain number of pedestrians, the
response was limited, but a small increase beyond a critical num-
ber resulted in diverging response amplitudes. From simple energy
considerations, it could be shown that the magnitude of the corre-
lated (modal) pedestrian force, Fcorr, being the component (of the
total modal force) in phase with the modal velocity (Fig. 8) could
be obtained as [5]:
€0
1 dq
F corr ¼ F D þ 2M ð7Þ
x0 dt
where FD is the damping force (cq_ 0 for linear viscous damping), M is
the modal mass, x0 is the angular modal frequency, c the viscous
damping constant, and q_ 0 and q
€0 are the amplitudes of the velocity
and acceleration respectively. An important finding was that the

Table 3
Modal properties of the Millennium Bridge without added damping (table reproduced with permission from Arup Partnership [75]).

Description Frequency (Hz) Damping (%) Modal mass (tonne)

First lateral mode of central span (CL1) 0.48–0.49 0.75–0.77 128–130
First lateral mode of Southern span (SL1) 0.80–0.81 0.6–0.7 172
Second lateral mode of central span (CL2) 0.95–0.99 1.3 145–148
First lateral mode of northern span (NL1) 1.04 0.32 113a
Third vertical mode of central span (CV3) 1.15–1.16 0.80 155
Fourth vertical mode of central span (CV4) 1.54–1.55 0.55 140
Fifth vertical mode of central span (CV5) 1.89–1.91 0.58–0.65 135
Sixth vertical mode of central span (CV6) 2.32–2.33 0.95 135
a
Estimated from FE analysis due to poor quality measurements.
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
Fig. 6. Crowd tests on the London Millennium Bridge (shown with permission from
Arup Partnership).
magnitude of Fcorr was almost linearly related to the amplitude of
the modal velocity (Fig. 9), which suggests that pedestrians act as
negative viscous dampers.
Dallard et al. [78] explained that the excessive oscillations were
caused by synchronous horizontal loading, as a consequence of fre-
quency ‘‘lock-in’’ between the gait-cycle frequency and the fre-
quency of the lateral movement. In a later publication, Dallard
et al. [5] defined the term ‘‘Synchronous Lateral Excitation’’ (SLE),
which is widely used in the literature when referring to excessive
pedestrian-induced vibrations. The velocity proportional feature of
the back-calculated pedestrian-induced force (Fcorr) was attributed
to an increased degree of synchronisation amongst the pedestrians
as the vibration amplitude grows, causing a positive force feed-
back. However, an important (and contradictory) observation
was that the vertical response of the structure at twice the modal
frequency did not show any disproportional increase which would
be expected as a consequence of a collective synchronisation of the
stepping frequencies within the crowd.
In January 2002, final pedestrian tests were carried out after
installation of the damping devices and since its reopening no
excessive vibrations have been reported for the bridge.
4.3. Laboratory tests
A series of laboratory tests were undertaken during the
18 month retrofit period of the London Millennium Bridge. The
experimental campaign involved walking on spot on a shaking ta-
ble at the University of Southampton [5] and walking across a
Fig. 7. Lateral accelerations measured on the Millennium Bridge during different crowd tests. (a) Vibrations of the northern span at frequency 1.0 Hz (figure from [35]). (b)
Vibrations of the centre span at frequency 0.48 Hz (reproduced with permission from Arup Partnership).
27
Fig. 8. Illustration of the phase angle between a force F and structural displacement
u.
40
30
F corr (N/person)
20
10
0
Velocity (m/s)
Fig. 9. Correlated pedestrian force per person as function of the local velocity of the
structure, derived from the response in Fig. 7b as observed on the Millennium
Bridge (figure after [5]).
7.3 m laterally driven test platform at Imperial College (Fig. 10a)
[79]. It was reported that the lateral DLF increases with the vibra-
tion amplitude but is vibration frequency independent. The tests
also revealed that the probability of ‘‘lock-in’’ increases with the
vibration amplitude, from between 30% and 40% at 5 mm ampli-
tudes up to around 80% at 30 mm [80]. A summary of published
test results is shown in Fig. 11. However, the number of people
used in the tests and the statistics of the results were not published
in the papers.
A research project related to the Solférino bridge was initiated
by the French Road Directorate and involved a suspended labora-
tory platform (7 m by 2 m) with a changeable lateral frequency
in the range 0.5–1.1 Hz (Fig. 10b) [74]. The dynamic (modal) pe-
destrian force was derived from the displacement response and
its derivatives, obtained through numerical differentiation. Both
28 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

Fig. 10. Platform tests commissioned Arup (a) and by the French Road Directorate (b) (pictures after [74] and http://www.arup.com/millenniumbridge/).

1
0.15 a b
0.8

0.1 0.6

0.4
0.05
0.2

0 0
0 10 20 30 0 10 20 30

Fig. 11. DLF (a) and probability of ‘‘lock-in’’ (b) for pedestrians walking on a vibrating surface (figure after [5]).

transient response tests and tests where a treadmill was placed on
the platform were carried out with the main conclusions that the
mean value of the first load harmonic is 35 N, the peak value of
the pedestrian force varies from 20 to 100 N. By analysing the nat-
ure of the energy input (through the cycle-by-cycle average of the
modal force multiplied by the sign of the velocity), it was con-
cluded that pedestrians consistently input energy into the bridge
when the acceleration reaches a threshold of 0.15 m/s2 [6]. The ef-
fect was explained as lock-in (or synchronisation) between the pe-
destrian gait cycle frequency and the movement of the platform,
triggered by the threshold acceleration. Interestingly, no correla-
tion between the instantaneous amplitude of the force from a sin-
gle pedestrian and the lateral velocity of the platform was
observed, even at large vibration amplitudes [74].
4.4. Early analytical approaches
Arup’s hypothesis, supported by the pedestrian crowd tests, is
that the lateral pedestrian force has a component in phase with
(correlated pedestrian force) and proportional to the velocity of
the bridge during ‘‘lock-in’’; see Fig. 9. In this case, the correlated
lateral force per person can be expressed as [5]:
_ Based on the results of the laboratory experiments at Imperial
F corr ðtÞ ¼ cp uðtÞ
where cp = 300 Ns/m is the velocity proportional lateral force coef-
_
ficient and uðtÞ is the lateral velocity of the bridge. This finding
led to Arup’s stability criterion which can be expressed in terms
of a critical number of pedestrians that will cancel the inherent
structural damping and eventually lead to excessive vibrations [5]:
4pff0 M
Ncr ¼ RL ð9Þ
cp 1L 0
½UðxÞ2 dx
The quantities, f, M, f0, U(x) represent the modal damping ratio,
mass, frequency and shape of the lateral mode in question and L
is the overall bridge length.
A different interpretation was given by Danbon and Grillaud
[68], who define the load per unit length of the correlated (or syn-
chronised) portion of the pedestrians as:
Fðx; tÞ ¼ Gbq/ðuÞ cos xt ð10Þ
where G = 23 N is defined as an average value of the amplitude of
the first load harmonic from a pedestrian walking on a stationary
surface, b is the width of the footbridge, q is the crowd density (per-
sons/m2) and /(u) is the portion of the pedestrians who are syn-
chronised with the lateral movement of the bridge. The
parameters in the model were determined from fitting to measured
responses on the Solférino bridge. The synchronisation function /
(u) was chosen as a stepwise linear function, such that the synchro-
nisation is constant at low amplitudes (i.e. / = 0.05 in the range
u < 1–2 mm) and increases linearly to a maximum constant value
in the range / = 0.80–1.0 for amplitudes larger than 5–6 mm.
ð8Þ
College, Newland [81,82] presented a model in which the pedes-
trian-induced force is proportional to the displacement amplitude
of the structure rather then the velocity. Mathematically, a feed-
back model was expressed in terms of the Fourier transforms
X(ix) and Y(ix) of the modal excitation force and the modal dis-
placement respectively:
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
YðixÞ ¼ HðixÞfXðixÞ þ aðixÞYðixÞg
The complex frequency response function and the modal force
exerted by the pedestrians per unit modal displacement are de-
noted H(ix) and a(ix) respectively. Furthermore, if people will
naturally fall into step with each other and adjust their phases such
as to increase the motion of the bridge (i.e. such that the load is in
phase with velocity), a stability criterion similar to the one in Eq.
(9) is obtained.
In a later paper, Newland [83] proposed that the load trans-
ferred to the pavement by the walker is modelled as two inertia
force terms: ‘‘The first arises from the natural displacement of a
person’s centre of mass while walking on a stationary pavement,
and the second from the additional displacement that occurs as a
consequence of movement of the pavement’’ [83]:
€ðtÞ þ m0 abu
FðtÞ ¼ m0 by € ðt  DÞ
where y(t) is the natural movement of the pedestrian’s centre of
mass, u(t) is the movement of the pavement, m0 is the mass of a sin-
gle pedestrian and D is a time lag. The coefficient a = 2/3 is based on
the results from the platform test at Imperial College (see Fig. 11)
and b is a synchronisation factor. The damping ratio required for a
stable bridge was derived as f > 12 abM p =M, where M and Mp are
the modal bridge and pedestrian masses respectively.
McRobie and Morgenthal [84] presented a non-dimensional
‘‘Pedestrian Scruton Number’’ as a measure for the susceptibility
of a pedestrian bridge to excessive vertical vibrations. For lateral
vibrations Newland [81] defined this quantity as:
2fM
Scp ¼
Mp
Thereby, Arup’s stability criterion in Eq. (9) can be rewritten in
terms of a minimum Scruton number:
cp
Scp >
2pf0 m0
Newland’s stability criterion can also be written this way, to
yield Scp > ab [83].
4.5. Lessons learned from Paris and London
The main conclusion from the investigations carried out by the
French and the UK research teams was that a transition point exists
for which a large increase in the lateral bridge response occurs as a
consequence of a small increase in the number of people on the
bridge. Dallard et al. [5] introduced the concept of negative pedes-
trian damping, such that the critical number of pedestrians needed
to cancel the inherent structural damping defines the transition
between stability and instability. A frequency-independent pedes-
trian damping coefficient cp was introduced, from which a stability
boundary could be derived (Eq. (14)). For design purposes, the UK
national annex to Eurocode 1991-2:2003 [18], suggests a modified
version of this stability boundary, assuming that instability cannot
occur for natural frequencies higher than 1.7 Hz. In Fig. 12 the sta-
bility criterion in Eq. (14) is shown with the modified UK NA ver-
sion. In addition, the constant boundaries proposed by Newland
[83] are depicted for two values of the synchronisation coefficient
b.
The French research teams [6,68,74] stated that people tend to
synchronise their gait cycle frequency to the frequency of
the structural movement as the vibrations become perceptible
(accelerations around 0.10–0.15 m/s2). Below this threshold, the
pedestrian behaviour was explained as random and motion inde-
pendent. Based on the research related to pont de Solférino, a
guideline for the dynamic design of footbridges, was issued by
the French road authorities, Sétra [6]. To avoid the possibility of
29
ð11Þ
ð12Þ
Fig. 12. Stability boundaries for avoiding excessive pedestrian-induced lateral
vibrations.
excessive horizontal vibrations, Sétra stipulates that the accelera-
tion response should be less than the lock-in threshold, defined
as alim = 0.1 m/s2. The proposed load model is based on the
assumption of a random crowd and presented in terms of an equiv-
alent sinusoidal area load acting at the natural frequency f0 of the
mode in question:
cF 0 pffiffiffiffi
FðtÞ ¼ Nw cosð2pf0 tÞ ð15Þ
A
ð13Þ where A is the surface area of the bridge deck, N the number of
pedestrians on the bridge, F0 the dynamic load amplitude of a single
pedestrian (35 N), w 2 [0, 1] a factor that reduces the load for fre-
quencies away from the average gait cycle frequency and c is a fac-
tor that depends on the type of crowd. For sparse and dense crowds
ð14Þ pffiffiffi
c ¼ 10:8 f, where f is the modal damping ratio and for very dense
crowds c = 1.85. The direction of the load is defined to correspond to
the sign of the mode shape. By imposing the condition that the
bridge’s steady-state acceleration should be less that the lock-in
threshold, a stability boundary can be written in terms of the pedes-
trian Scruton number. For a sinusoidal mode shape, the following
condition should be fulfilled:
rffiffiffiffi
4 cF 01
Scp > w ð16Þ
p m0 alim N
where m0 is the body mass of a single pedestrian. In Fig. 12, the sta-
bility boundary for a very dense crowd (c = 1.85) is shown, both for
N = 200 and N = 2000.
As shown here, the number of uncorrelated pedestrians needed
to trigger perceptible vibrations is similar to the concept of a crit-
ical number of pedestrians needed to cancel the inherent structural
damping (as proposed in [5]), although the interpretation of the
problem is different. However, the large differences in the stability
boundaries shown in Fig. 12, both in magnitude and frequency
dependency, clearly illustrate the need for further research within
the area.
5. Lateral footstep forces on a moving surface
When the walking surface moves laterally, an additional non-
negligible force component is potentially generated due to the
interaction between the movement of the pedestrian’s body centre
of mass and that of the structure. In this section, the most signifi-
cant contributions dealing with this human-structure lateral inter-
action and the quantification of pedestrian-induced lateral forces
on moving structures are reviewed.
30 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
5.1. Pedestrian walking on moving surfaces
In the field of biomechanics, few studies exist on the effect of
lateral surface movement on the walking characteristics. Kay and
Warren Jr. [85] used a visual stimulation to study the temporal cor-
relation between lateral sinusoidal oscillations of the visual field
(at frequencies between 0.075 and 1.025 Hz) and the lateral move-
ment of the test person’s neck. They concluded that the pedestrian
gait cycle frequency locked to the driving frequency in a large
range of frequencies (0.65–0.925 Hz). The visual stimulus was a
narrow hallway which vibrated at a very large (fictive) amplitude
of around 0.32 m. Recently, McAndrew et al. [86] presented studies
of both visual stimulation as well as physical perturbations of a
treadmill during walking. The treadmill was driven into lateral
oscillations at combinations of four distinct frequencies (0.16,
0.21, 0.24 and 0.49 Hz), each with different amplitudes (50, 40,
70 and 25 mm). Interestingly, it was reported that both the step
width and the frequency increased during the lateral vibrations,
whereas the step length decreased. Furthermore, the lateral move-
ment of the neck (C7 vertebral) was monitored and its PSD fea-
tured peaks at each of the vibration frequencies as well as at half
the pacing frequency.
For walking on a laterally oscillating surface, Brady et al. [87]
introduced the terms ‘‘Fixed to Base’’ and ‘‘Fixed in Space’’ to dis-
criminate between pedestrians that translate laterally with the
base and those who hold a fixed position in space and allow the
treadmill to move beneath them. Experiments carried out at low
frequency (0.2–0.3 Hz) and large amplitude (127 mm) lateral
vibrations showed that the step width generally increased as a
consequence of the vibrations. Pedestrians who were fixed in space
had a tendency to step wider with the left foot when the base was
moving to the right and narrower when it was moving to the left.
For pedestrians fixed to the base, the opposite tendency was true.
Video analysis was used by Fujino et al. [42] to track the lateral
head motion of randomly selected pedestrians during congested
periods on the T-Bridge in Japan. They found that the amplitude
of the lateral motion of the head increased with the lateral bridge
amplitude and they concluded that in order to maintain body bal-
ance, people widen their gait. On the T-Bridge, the steady-state
vibration amplitude reached approximately 10 mm causing 20%
of the pedestrians to synchronise their head movement (frequency
and phase) to that of the bridge [42]. According to Fujino et al. [42],
laboratory experiments using a laterally moving platform showed
that people synchronise their walking to a lateral motion for
amplitudes in the range 10–20 mm.
Yoshida et al. [88] estimated that during vibration amplitudes
of 9 mm the average correlated pedestrian force amplitude on
the T-Bridge was around 3.3 N. This estimate is based on video
analysis of the pedestrian movement during a crowd event. It
should be noted that this value is considerably lower than the
forces estimated from the Millennium Bridge (Fig. 9). Yoshida
et al. [89] reported that on average 60% of the pedestrians walked
at frequencies in the range 0.9 Hz ± 0.1 Hz (where 0.9 Hz is the fre-
quency of the bridge movement). However, this study did not re-
veal the number of people which were phase synchronised to the
movement of the bride.
Shortly after the Millennium Bridge incident, McRobie et al. [90]
constructed a suspended platform (weighing 1.2–2.0 t) equipped
with a treadmill having an adjustable lateral frequency in the
range 0.7–0.9 Hz. It was observed that people tend to spread their
feet further apart and walk at the same frequency (with constant
phase) as that of the platform. In addition, it was reported that
the load amplitude could reach 300 N with the component in
phase with the platform velocity (the correlated pedestrian force)
up to 100 N at 100 mm vibration amplitudes [90]. The authors ar-
gue that the increase in the overall correlated force observed on
the Millennium Bridge could be attributed to an increased proba-
bility in gait transition (more pedestrians changed gait) as the
vibration amplitude increased, rather than a velocity proportional
load.
5.2. Quantification of lateral forces using instrumented platforms
Rönnquist [91] measured lateral GRFs of pedestrians crossing a
4 m long suspended platform with adjustable lateral frequencies
(0.75 Hz, 0.84 Hz, 0.95 Hz and 1.14 Hz). For each platform fre-
quency, three different step frequencies were tested and the total
of 1087 footsteps were recorded, but using only four different test
persons. The study revealed that the lateral load increases with the
lateral acceleration of the platform and also as the gait cycle fre-
quency approaches the natural frequency of the platform [91].
Rönnquist and Strømmen [92] defined an equivalent DLF as a func-
tion of the pedestrian detuning away from the resonance fre-
quency (fw  fn) and the structural acceleration:
DLFeq ¼ 0:145  0:1
( "  2 #! )
1 fw  fn
 exp  0:45 þ 1:5 exp  € 1:35
u 0 ð17Þ
2 0:07
where u € 0 is the amplitude of the structural acceleration and fw  fn
is the pedestrian de-tuning from the natural frequency of the struc-
ture. Rönnquist [91] also defined the first four DLFs of the lateral
force obtained from a Fourier analysis of two consecutive footsteps.
The DLF of the first harmonic was almost identical to that in Eq. (17)
and is therefore not repeated here. The DLF for the higher harmon-
ics were DLF2 ¼ 0:010 þ 0:008u € 0 and DLF3 = 0.015 and DLF4 = 0.005.
No information was provided regarding the phase of the load har-
monics. Furthermore, the platform used in the study was light
and the measured footstep forces are applicable to accelerations
in excess (up to 2 m/s2) of what is normally considered acceptable
on a footbridge.
Butz [52] used a laterally driven platform equipped with four
integrated force plates (Fig. 13a) to measure GRFs from 98 different
persons and to quantify their degree of synchronisation with the
platform. The platform (length 12 m) was driven in a sinusoidal
lateral motion at frequencies ranging from 0.6 to 1.5 Hz with dis-
placement amplitudes in the range 3–40 mm. The lateral GRFs
were measured for slow, normal and fast walking speeds. Three
different degrees of synchronisation were reported; (1) walking
is completely synchronised and such to produce negative damping
(i.e. total force lies in the upper half-plane in Fig. 8), (2) walking is
not completely synchronised, but the pedestrian changes to syn-
chronised walking (with negative damping) during a passage,
and (3) not synchronised or negatively synchronised (i.e. adds
damping to structure). It was found that persons with a natural
walking frequency (i.e. as measured on a fixed floor) within
±0.1 Hz from the lateral vibration frequency, will potentially syn-
chronise to the structural motion. Butz [52], further reports that
the measured DLFs tend to increase (slightly) with the lateral plat-
form acceleration amplitude, but due to large scatter in the data,
the following values were suggested for design:
8
< 0:04;
> unsynchronised pedestrians or fixed floor
DLF1 ¼ 0:055; u€0 6 0:5 m=s2
>
:
€0 > 0:5 m=s2
0:075; u
ð18Þ
A different approach was taken by Nakamura et al. [93], who
used an instrumented shaking table to investigate walking on the
spot when subject to lateral vibrations with frequencies ranging
from 0.75, 0.87, 1.00 Hz and 1.25 Hz at displacement amplitudes
10–70 mm. Based on the results from five tests subjects, it was
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52 31

Fig. 13. (a) Laterally driven platform at RWTH Aachen (picture after [9]). (b) Instrumented treadmill on a shaking table at Tongji University (picture after [32]). (c) Laterally
driven treadmill at the University of Reggio Calabria (picture after [56]).

found that the weight-normalised force increased almost linearly

with the vibration amplitude at all frequencies from around 0.10 a
at 10 mm amplitude to around 0.16 at 70 mm. A synchronised pe- 4
10
3 x

S{Ff }, S{Ff } [N s]
destrian was defined as one walking with the same frequency as

2
that of the shaking table and the tests revealed that at the frequen-
2
cies 0.87 Hz and 1.00 Hz the pedestrian tended to synchronise, the

w
probability for this being 20% and 40–50% in the two cases respec- 1
tively. This observation is generally in agreement with those of

L
Butz [52]. 0
0.9
0.85
5.3. Walking on laterally moving instrumented treadmills 0.8 1.5
1.25
0.75 1
fL [Hz] 0.7 0.75
The utilisation of instrumented treadmills for the determination 0.65 0.5
0.25 f [Hz]
of pedestrian-induced forces offers some advantages opposed to 0
platforms of limited length as it offers the possibility to measure
Fig. 14. Examples of the PSDs of the lateral pedestrian force as a function of the
continuous time-histories of the pedestrian-induced forces.
lateral vibration frequency (fL) where the solid line represents F fw and the dot-
Sun and Yuan [32] fixed an instrumented treadmill on a force dashed line F fL for fixed vibration amplitudes of 15 mm (figure after [61]).
plate which in turn was fixed onto a shaking table, Fig. 13b. Seven
different pedestrians were asked to walk (freely) on a treadmill at
two different walking speeds 1.0 m/s (3.6 km/h) and 0.83 m/s one centred around the walking frequency and its higher harmon-
(2.99 km/h), subjected to different combinations of lateral vibration ics and the second one occurring at a frequency equal the vibration
frequencies (0.65–1.2 Hz) and amplitudes (4–50 mm). A simple frequency and denoted ‘‘the self-excited force’’ [61]. The pedes-
equation was proposed for the DLF of the first load harmonic as a trian-induced lateral force was therefore written as the sum of
function of the vibration amplitude, u0 (in (m)) of the structure: these two components:

DLFðu0 Þ ¼ 1:18u0 þ 0:05 for u0 < 0:05 m ð19Þ FðtÞ ¼ F fw ðtÞ þ F fL ðtÞ ð21Þ

It was observed during the tests when u0 > 0.05 m people could
not continue to walk steadily and had to hold the handrail to main-
tain their balance, hence the limit in Eq. (19). Sun and Yuan [32] re-
port that for small vibration amplitudes, the relative phase between
the pedestrian movement and that of the treadmill is variable (non-
constant), but as the amplitude increases the phase becomes (al-
most) constant and the gait cycle frequency changes to the vibra-
tion frequency. Further, they find that on average the pedestrian
load is 140.8° (S.D. 17.9°) ahead of the treadmill motion. Based on
their studies a qualitative equation for the probability of synchroni-
sation as function of the vibration amplitude (u0) and vibration fre-
quency (f0) was proposed [32]:
u0 2
qs ðu0 ; f0 Þ ¼ ec2 ðf0 1Þ
u0 þ c1
with c1 and c2 as unknown parameters.
Pizzimenti [56] constructed an instrumented treadmill which
could be driven in a sinusoidal lateral motion at predefined combi-
nations of frequency and amplitude (Fig. 13c). In a pilot study, the
loads from five test subjects were obtained at three different vibra-
tion amplitudes (15 mm, 30 mm and 45 mm) and at five different
lateral frequencies in the range 0.60–0.92 Hz. From the PSD of
the load, two different force components were observed, the first
where fw is the frequency of the first load harmonic (i.e. half the
pacing frequency) and fL is the lateral vibration frequency. In
Fig. 14 the PSD of the total lateral force for one test subject is shown
for 15 mm vibration amplitude at various vibration frequencies. The
self-excited force was further subdivided into an in-phase (with dis-
placement) and 90° out-of-phase (quadrature phase) lateral pedes-
trian load components as:
F fL ðtÞ ¼ DLFin W sinð2pfL tÞ þ DLFout W cosð2pfL tÞ ð22Þ
The mean value of the measured DLFs from the five test subjects
are shown in Fig. 15. According to the definition of the pedestrian-
induced load in Eqs. (21) and (22), DLFin > 0 implies that the pedes-
trian adds to the modal mass, whereas DLFin < 0 was explained as
ð20Þ added stiffness [61]. Similarly, DLFout > 0 corresponds to negative
damping. Two interesting phenomenon were observed. Firstly,
pedestrians seem to act as negative mass on the structure (or posi-
tive stiffness) over the entire frequency range. Secondly, for only
one combination of frequency and amplitude (0.92 Hz and
45 mm) did the pedestrians produce a large amount of negative
damping.
Motivated by the need of statistically reliable data for the pe-
destrian-induced lateral forces, Ingólfsson et al. [60] carried out
an extensive experimental campaign using the same instrumented
32 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

0.06 0.1
Positive mass u0 = 15 mm u0 = 15 mm
0.04 u0 = 30 mm u0 = 30 mm

0.02 u0 = 45 mm u0 = 45 mm
0.05
0

DLFout
Negative damping
DLFin

-0.02 0
-0.04

-0.06 -0.05
-0.08
Negative mass a Positive damping b
-0.1 -0.1
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
fL [Hz] fL [Hz]

Fig. 15. Average DLF for the in-phase (a) and out-of-phase (b) components of the self-excited pedestrian force (figure after [61]).

treadmill employed by Pizzimenti [61]. The lateral forces were
measured from 71 individuals at varying combinations of vibration
frequencies (0.33–1.07 Hz) and amplitudes (4.5–48 mm). By cover-
ing more than 55 km of walking distributed over almost 5000 indi-
vidual tests, the data comprises the largest database of pedestrian-
induced lateral forces on a moving surface that has been published
to date. The results from the pedestrian tests were presented in
terms of the velocity and acceleration proportional pedestrian load
coefficients cp and .p respectively. Positive values of cp and .p indi-
cate an overall decrease in the modal damping and mass of the
structure occupied by the pedestrian. The velocity proportional
coefficient was determined such that the work done by the mea-
sured pedestrian force on the lateral displacement of the treadmill
should equal the work done by an equivalent linear viscous dam-
per. The following expressions were derived for cp and .p [60]:
2
cp ¼ Re½SF u_ ðfL ÞDf
u_ 20
2 the experimental data impossible. Instead the load coefficients
.p mp ¼ €2 Re½SFu€ ðfL ÞDf
u0
where SF u_ and SF u€ are the cross spectral densities between the force
and the treadmill velocity (with amplitude u_ 0 ) and acceleration (with
amplitude u € 0 ) respectively, whereas Df is the frequency resolution of
the spectra. In Fig. 16, the results from the experimental campaign
are shown (a) and (b) as probability distributions and (c) and (d) as
frequency dependent mean values taken across all test subjects
and vibration amplitudes. The frequency axis is normalised with
the mean gait cycle frequency (f w ) of the pedestrian test subjects.
The following main conclusions were deduced from the study [60]:
1. Pedestrians extract energy from the structure (act as negative
dampers) at most of the frequencies that were tested.
2. At lower frequencies pedestrians decrease the overall modal
mass, but add to it at higher frequencies.
3. The load coefficients generally decrease (numerically) with
increasing vibration amplitude, suggesting a certain self-limit-
ð23Þ ing effect of the load.
4. Large scatter in the data makes a deterministic description of
ð24Þ
must be quantified through their probability distributions.

a b

c d

Fig. 16. Probability distributions of (a) damping proportional coefficient, cp, and (b) inertia proportional coefficient, .p. The coefficients mean value (cp and .
 p ) ± one standard
deviation (rcp and r.p ) are further shown as functions of the normalised frequency for (c) cp and (d) .p.
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
In a complimentary study, Ingólfsson et al. [94] analysed the
relative phase between the pedestrian movement and the move-
ment of the underlying surface. It was shown that human-struc-
ture synchronisation is not a pre-condition for the development
of positive values of the velocity proportional load coefficients,
cp, (i.e. negative damping) which may lead to excessive lateral
vibrations.
6. Full scale measurements: case studies
As modal identification techniques are increasingly improving
and both the hardware and software for vibration measurements
is becoming more compact and user friendly, full scale testing of
footbridges has become a common tool in the assessment of pedes-
trian-induced vibrations. This includes both the determination of
modal properties [95–97] and measurements of the response to
large pedestrian crowds.
In the latest version of the Eurocode (Annex A2 of EN 1990:2002
[98]) it is stated that if the pedestrian comfort criteria (defined as
0.2–0.4 m/s2 for lateral vibrations) is not satisfied with a significant
margin, provisions for installation of dampers may be necessary
and that ‘‘in such cases the designer should consider and identify
any requirements for commissioning tests’’.
In this section an extensive summary of various full scale mea-
surements of bridges susceptible to human induced lateral vibra-
tions have been collected and are presented in a chronological
order.
6.1. Changi Mezzanine Bridge
The Changi Mezzanine Bridge (Fig. 17) is a shallow arch steel
bridge of welded hollow circular and rectangular sections, with a
140 m main span (between the arch supports) and two 30 m side
spans. The bridge is situated at Changi International Airport in Sin-
gapore connecting two passenger terminals through the Rail Ter-
minal [99]. The bridge was designed by Skidmore, Owings and
Merrill LLP architects and Arup (New York) and construction began
in June 2000.
As a consequence of the problems with the opening of the Mil-
lennium Bridge in London, a study on the vibration serviceability
under human induced loading was commissioned, including modal
testing of the structure and response measurements to pedestrian
crowd loading. Two critical vibration modes were identified; a
symmetrical lateral vibration mode (LS1) at frequency 0.9 Hz with
damping ratio 0.4% of critical and modal mass around 453 t (esti-
mated from FE model) and a symmetric torsional mode (TS1) with
Fig. 17. The Changi Mezzanine Bridge at Singapore’s Changi Airport (Curtesy of
J.M.W. Brownjohn).
33
frequency 1.64 Hz, damping 0.4% of critical and an estimated mod-
al mass around 147 t [99]. It was estimated (using Arup’s formula
Eq. (9)) that around 140 people would cause divergent lateral
vibration amplitudes [100].
The pedestrian crowd tests, involving up to 150 volunteers,
were undertaken in February 2002. In the main test the number
of pedestrians was gradually increased until all 150 pedestrians
circled the bridge, walking at their own comfortable speed [101].
During the tests a disproportional increase in the amplitude of
LS1 occurred as the number of pedestrians increased. During a cir-
culatory tests (with 150 pedestrians on the bridge), the amplitude
of the lateral vibrations seemed to continually grow until the
pedestrians were asked to stop. At that point the lateral vibration
amplitude was 0.17 m/s2 (or 5.5 mm). A similar disproportionate
increase of the vertical response was not found and no distinct
peak in the vertical response at twice the frequency of the lateral
mode was observed as might be expected if the pedestrian footfall
was synchronised. Brownjohn et al. [101] noted that for a reduced
crowd size of 100 people, large build up of lateral vibrations oc-
curred before the pedestrians were asked to stop. After resuming
the walking, similar vibration levels were not reached again. This
variation in the critical number of pedestrians needed to trigger
instability in two nominally identical situations may be attributed
to randomness in the reaction of pedestrians to the lateral motion
of the bridge. Furthermore, variations in the spatial distribution of
pedestrians or the distribution of step frequencies (and phase an-
gles) within the crowd may also affect the critical number of
pedestrians. However, the tests revealed that between 100 and
150 people are sufficient to trigger large lateral vibrations, which
is in good agreement with the analytical prediction of 140 people,
using Arup’s formula (Eq. (9)).
Based on the dynamic analysis and the experimentally observed
instability, two lateral TMDs (500 kg each) were installed to in-
crease the damping of LS1 to 1.65% and thereby the (theoretical)
critical number of pedestrians to 560 people [101,102].
6.2. Nasu Shiobara Bridge (M-Bridge)
This very light (0.4 t/m2) suspension bridge across the Maple
Valley in Nasu Shiobara (M-Bridge) has a main span of 320 m
(Fig. 18) and is known to vibrate excessively when crowded by
pedestrians. The vibrations occur primarily in modes with frequen-
cies 0.88 Hz (third asymmetric mode) and 1.02 Hz (fourth symmet-
ric mode) dependent on the position of people on the bridge
[45,103].
In November 2002, controlled field tests were performed where
accelerometers were mounted, both on the waist of several test
persons and on the bridge. The pedestrians were asked to cross
the bridge during a period with normal traffic and therefore vary-
ing vibration amplitudes. The estimated crowd density during the
tests varied between 0.7 and 1.3 persons/m2, but generally with a
non-uniform spatial distribution [103].
According to Nakamura [103], the pedestrian gait cycle fre-
quency was generally synchronised to the girder vibration and
the pedestrian phase was around 120–160° ahead of the bridge
motion. Nakamura further explains that in some of the tests,
pedestrians lost balance and stopped walking, causing a ‘‘detun-
ing’’ of the pedestrian phase, but when the pedestrian started
walking again, he/she would synchronise to the girder movement
again. In Fig. 19 an example of a so-called synchronised pedestrian
and the aforementioned detuning is shown. It should be noted that
if the pedestrian walks so that the gait cycle frequency is close to
the modal frequency, the relative instantaneous phase of the pe-
destrian and that of the structure drift slowly apart. This causes
the pedestrian to alternately tune or detune to the structure.
34 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

Fig. 18. The Nasu Shiobara suspension footbridge in Japan (picture after [45]).

6
Position at L/4 Not tuned
Lateral displacement [cm]

Not tuned
4

-2

-4
Tuned
Girder Tuned
Pedestrian
-6

Time [s]

Fig. 19. Lateral displacement of the M-Bridge at quarter-span shown with the lateral displacement of the pedestrian (figure after [45]).

Nakamura [103] also investigated the subjective assessment of
the vibrations felt on the bridge during large lateral vibrations. At
(peak) vibration amplitudes about 0.30 m/s2 (or 10 mm at fre-
quency 0.88 Hz) the pedestrians could feel the vibrations and some
characterised them as uncomfortable, but without it affecting their
normal way of walking. At vibration amplitudes of 0.75 m/s2
(25 mm), some pedestrians had difficulties with walking and occa-
sionally touched the handrail. At 1.35 m/s2 (45 mm) people often
lost their balance and some even stopped walking.
6.3. Lardal footbridge
The Lardal footbridge is situated in a recreational area in Nor-
way, Fig. 20. When inaugurated in 2001 it gained local publicity
when considerable vibrations developed in the first lateral mode
(at 0.83 Hz) when traversed by pedestrian groups [104].
The bridge is a shallow arch glue-laminated timber bridge with
steel cable reinforcement in certain parts of the main span. The dis-
tance between the arch supports is 91 m. Additionally, there are two
approach spans, each 13 m long. It was found that the first vibration
mode has a damping ratio of 2.5%, a modal mass of 18 t and a mode
shape that could reasonably be approximated by a half sine with a
wavelength of 80 m. However, it was also observed, that the first lat-
eral mode had strong vertical and torsional components, with the
centre of rotation beneath the bridge deck, such that the motion
can be described similarly to that of an inverted pendulum [91].
During on-site measurements of the bridge vibrations it was
found to be extremely lively, with horizontal acceleration response
exceeding 1 m/s2 for as few as 40 pedestrians (Fig. 21). Based on
the measured acceleration response, a simple linear trend between
the number of pedestrians on the bridge and the peak acceleration
was observed [92]:

Fig. 20. The Lardal timber footbridge in Norway (picture after [92]).
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52 35

Fig. 21. Measured peak acceleration as function of the number of pedestrians on the Lardal footbridge (figure after [91]).

€ 0;max ¼ 0:024Nped
u ð25Þ A total of eight TMDs were designed to control pedestrian induced
vibrations, one for the first lateral vibration mode (with mass
Ronnquist et al. [105] report that as the lateral movements in-
4920 kg) and seven for vertical modes in the frequency range
crease, more people tend to synchronise to the motion of the struc-
1.55–3.06 Hz [26].
ture but a direct quantification of the number of synchronised
In April 2006, both ambient and free vibration tests were per-
pedestrians or the distribution of phase angles were not provided.
formed prior to installation of the glass in the handrails and the
timber deck. These tests revealed that the natural frequency of
6.4. Coimbra Footbridge
the first lateral vibration mode is around 0.91 Hz with a damping
ratio as low as 0.5–0.6% and modal mass around 205 t (determined
The Coimbra Footbridge (also known as the Pedro and Inês foot-
from an updated FE model) [108]. The ambient vibration test
bridge), across the river Mondego in the city of Coimbra, Portugal,
showed a large variation in the estimated damping and the free
was built in the period 2005–2006 and inaugurated in November
vibration tests were used to narrow the values [109]. During con-
2006. The bridge is a shallow arch bridge with a total length of
trolled pedestrian crowd tests (Fig. 22b), the peak midspan acceler-
274.5 m divided into a main span (110 m), two side spans (64 m)
ation was around 0.2 m/s2 for 70 pedestrians but rose to about
and shorter approach spans at each end connecting the river bank
1.2 m/s2 for 145 pedestrians (Fig. 23a) [107,110]. This observation
to the bridge (Fig. 22a). The bridge deck is constructed as a steel-
matches the re-calculated critical number of pedestrians (Ncr = 73
concrete composite box girder (width 4 m) and the arches are
from Eq. (9)), when using the measured modal characteristics
made of steel box girders. The bridge is peculiar in that it is anti-
and the updated FE model as input parameters. A redesign of the
symmetric about a longitudinal axis, with each half of the bridge
dampers was performed and the final solution involved six individ-
consisting of two half-arches that are offset 4 m from each other
ual lateral TMDs (total mass 14.8 t) to provide a theoretical equiv-
and meet to form an 8 by 8 m square at midspan [106].
alent damping ratio of 7.8% [107]. This is considerably more
At the design stage, the bridge was found susceptible to both
damping than anticipated during the design.
lateral and vertical pedestrian induced vibrations and a compre-
Subsequent long-term monitoring [25], showed that during
hensive study on its serviceability was undertaken by the Labora-
the first 12 months the acceleration response of lateral vibration
tory of Vibration and Monitoring from the University of Porto [26].
was always less than the comfort threshold, defined as 0.1 m/s2
The analysis identified one lateral vibration mode prone to exces-
[97]. The measured daily maximum acceleration is shown in
sive lateral vibrations, at a frequency of between 0.70 and 0.78 Hz.
Fig. 23b.
The critical number of pedestrians calculated according to Eq. (9)
predicted that 145 pedestrians (a crowd density of 0.2 person/
m2) were required to trigger excessive pedestrian-induced lateral 6.5. Passarelle Simone de Beauvoir
vibrations when assuming a modal damping ratio of 1%. The for-
mula was also used inversely, to determine the required amount This 304 m long footbridge in Paris is the 39th bridge across the
of damping (5%) to avoid this for a crowd density of 1 person/m2. River Seine (Fig. 24). It was designed by Feichtinger Architects and

Fig. 22. Coimbra Footbridge (a) and pedestrian crowd tests (b) (pictures after [107]).
36 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

0.10
1.4
a 0.09 b
1.2
Maximum acceleration [m/s2]

Vibration amplitude [m/s2]

0.08
1
0.07
0.8 0.06

0.6 0.05

0.4 0.04

0.03
0.2
0.02
0 0.015
0 20 40 60 80 100 120 140 160 0 20 60 100 140 180 220 260 300 352
Number of pedestrians Time [days]

Fig. 23. Results from pedestrians crowd tests (a) (after Magalhaes et al. [110]), and maximum daily acceleration of mode 1 during the first year of service (b) (after Cunha
et al. [97]).

Fig. 24. Passarelle Simone de Beavoir (during design known as Passarelle Bercy-Tolbiac) across River Seine in Paris (picture after [111]).

RFR Ingénieurs and inaugurated on July 13, 2006. It features a ple occupied the structure simultaneously. Furthermore, a great
190 m main span which is a combined shallow arch and a stress amplitude dependency of the damping was observed in mode 1,
ribbon with a walkway on both levels that join approximately at increasing from less than 1% of critical at low vibration amplitudes
quarter spans [22]. During the design, nine modes were deemed (<5 mm) to values of around 2.5% at amplitudes of around 30 mm.
susceptible to human-induced loading, three of which were pre- An increase in the structural damping has been observed over time
dominantly lateral. The frequencies of these lateral modes were and therefore no further measures have been made. Instead a long-
calculated as 0.46 Hz (mode 1), 0.96 Hz (mode 3) and 1.12 Hz term vibration monitoring system has been installed, which will
(mode 5) respectively, with mode 3 being localised to one of the hopefully provide valuable data for future serviceability assess-
approach spans of the bridge. ment of the bridge.
An extensive series of tests and damping provisions were com-
missioned in order to secure the serviceability of the structure. The
experimentally determined frequencies and damping ratios were
0.56 Hz and 0.56% and 1.12 Hz and 0.53% for mode 1 and 5 respec-
tively. No information regarding the modal masses has been pro-
vided. Viscous dampers were installed near the bridge supports,
but subsequent modal identification revealed that the damping
in modes 1 and 5 only rose to 0.77% and 0.58% respectively [27].
Pedestrian crowd tests were carried out using 120 volunteering
participants and involved groups of varying sizes (from 20 to 120
people) circulating on the bridge (Fig. 25). Both modes 1 and 5
were excited during the crowd tests with maximum displacement
of 30 mm in mode 1 (0.37 m/s2) for a crowd of 80–100 people
walking randomly. Hoorpah et al. [27] report that twice this ampli-
tude was reached when a group of 60 people walked in step using a
metronome to control the step frequency. Although large vibration
amplitudes can be reached during controlled walking tests, these Fig. 25. Random walking of 120 pedestrians during crowd tests on 10 July 2006
levels were not observed on the opening day where about 400 peo- (picture after [27]).
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
6.6. Clifton Suspension Bridge
The Clifton Suspension Bridge in Bristol was designed by Isam-
bard Kingdom Brunel and completed in 1864 (Fig. 26). The bridge
has a main span of 214 m and a bridge deck with a width of 9.5 m.
The Clifton Suspension Bridge is a two-lane road bridge with pe-
destrian walkways on either side of it [112].
Ambient vibration measurements revealed 27 vibration modes
in the range 0.2–3.0 Hz, four of which were predominantly lateral.
In particular, two lateral vibration modes that react strongly to
pedestrian induced loading were identified; L2 with natural
frequency of 0.524 Hz and a damping ratio of 0.58% and L3 at
0.746 Hz with a damping ratio of 0.68%. The other two lateral
modes have natural frequencies and damping ratios of 0.240 Hz
and 3.68% (L1) and 0.965 Hz and 3.51% (L4) respectively.
During a 2 week monitoring of the bridge in the summer of
2003, two occasions were of particular interest, with both occur-
ring during the annual Bristol International Balloon Fiesta, where
the bridge is traditionally traversed by large crowds of pedestrians.
Based on videos from a security camera monitoring the entrance of
the bridge, it was estimated that during an approximately 2 h per-
iod, between 150 and 450 people were constantly on the bridge
causing large vibrations in modes L2 and L3 with L2 being the
dominant one. The maximum measured accelerations were
0.13 m/s2 (11.3 mm) in mode L2 and 0.11 m/s2 (4.5 mm) in mode
L3 respectively. It was noted that the lateral vibrations of the
bridge experienced an instability-type behaviour. Small vibration
amplitudes were recorded until the number of pedestrians reached
about 200 people. In a short period of time the number of people
on the bridge doubled, whereas the displacement amplitude in
the low frequency mode L2 increased by an order of magnitude.
Subsequently, the large response of mode L3 was initiated and
for a while both vibration modes reacted strongly to the pedestrian
induced loading, Fig. 27a.
It was noteworthy that vibration mode L4 with a natural fre-
quency of 0.965 Hz did not react (strongly) to the pedestrian in-
duced loading, which according to Macdonald [112] is attributed
to its high level of damping.
Macdonald used a similar methodology to that of Dallard et al.
[35] to determine the correlated pedestrian force (or in-phase with
velocity force) and, as shown in Fig. 27b and c, a near linear rela-
tionship between the force and the bridge velocity was observed.
Fig. 26. The Clifton Suspension Bridge in Bristol, UK (Courtesy of John Macdonald).
37
The value of the velocity proportional load coefficient, cp (from
Eq. (8)), was found to be lower than the value of 300 Ns/m, as re-
ported from the Millennium Bridge. Macdonald noted that this dif-
ference is partially caused by inaccuracies in the modal mass and
damping estimates.
The observation made on the Clifton Suspension Bridge was
used to argue that, although the response behaviour of the bridge
fits the assumption of negative damping, the mechanism does not
necessarily involve synchronisation of step frequencies to the nat-
ural frequencies of the bridge. The fact that the vibrations occurred
simultaneously in two vibration modes (L2 and L3) and initiated in
the lower frequency mode with a frequency further away from
comfortable step frequency, suggests that a different mechanism
than synchronisation is responsible for these vibrations. Further-
more, the measured vertical bridge response did not show any fre-
quency peaks at twice the frequency of the lateral modes, which
would be a consequence of collective synchronisation of the step
frequencies.
6.7. Weil-am-Rhein footbridge
The Weil-am-Rhein footbridge, also known as the ‘‘Dre-
iländerbrücke’’ (Three Country Bridge) was inaugurated in the
summer of 2007 and connects the German town Weil-am-Rhein
to the town Huningue in France at the border to Basel in Switzer-
land. It is a steel-arch bridge with a suspended walkway which fea-
tures a clear span of 230 m (Fig. 28). It consists of two steel arches,
a main arch which spans the river in a vertical plane and second
arch with a smaller cross section that inclines to almost meet the
main arch at midspan [113].
The bridge designers identified three vibration modes suscepti-
ble to pedestrian induced lateral loading, all featuring symmetrical
bending of the bridge deck (modes 3–5). A subsequent modal iden-
tification of the bridge revealed the frequencies of these modes as
0.90 Hz, 0.95 Hz and 1.00 Hz, about 10% higher than predicted by
the designers [114]. The damping is reported low (i.e. <1%) and
Franck [115] estimated the modal mass of mode 5 to be around
105 t.
Strobl et al. [113] used a value cp = 94 Ns/m to calculate the crit-
ical number of pedestrians, corresponding to a correlated lateral
force of 5 N per pedestrian at approximately 0.3 m/s2 (i.e. 20% of
the load amplitude from pedestrians walking on a rigid surface
38 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

550 22
a No. of pedestrians
500 Mode L2 amplitude 20

450 Mode L3 amplitude 18

400 16

Number of pedestrians

Amplitude [mm]
350 14

300 12
250 10

200 8

150 6

100 4

50 2

0 0
22:15 22:30 22:45 23:00 23:15 23:30 23:45 00:00 00:15
Time [hh:mm]

5 ver e 6
Force amplitude per pedestrian [N]

b it o s c
e st f ncrea
B i 5
4 i t ial
In t fit
4 Bes
3
3

2 2
1
1
Whole record, 0
20 cycle average
0 Initial increase in Whole record,
amplitude, every -1 every cycle
cycle
-1 -2

Velocity amplitude [m/s] Velocity amplitude [m/s]

Fig. 27. Results from full-scale measurements on the Clifton Suspension Bridge showing (a) the lateral displacement amplitudes of mode L2 and L3 and the correlated
pedestrian force as function of lateral bridge velocity for (b) mode L2 and (c) mode L3 (figures after [112]).

Fig. 28. Weil-am-Rhein footbridge, designed by Feichtinger Architects and Leonhardt, Andrä und Partner (LAP) engineering bureau (picture after [111]).

25 N). The reason for selecting 5 N is based on the observations on slightly higher pace and to avoid pausing during the crossing of the
the Toda Park Bridge in Japan [42] where 20% of the pedestrians bridge. In this case, excessive lateral vibrations were observed in
were reported in synchrony with the bridge vibrations. The esti- mode 5 at 0.96 Hz with acceleration time history as shown in
mated number of pedestrians needed to trigger large lateral vibra- Fig. 29b and amplitudes of up to 1.7 m/s2 (peak-to-peak displace-
tions was reported as 500 people (density of 0.24 pedestrians/m2) ment of approximately 80 mm). It was further pointed out that
[113]. Conversely, the damping needed to secure stability for a for safety reasons the walking tests were stopped prematurely.
crowd density of 2.0 pedestrians/m2 was calculated as 16.5% of This happened while the vibration amplitude was constantly
critical, which would require a 10 t TMD [113]. increasing, thus amplitudes may have further developed if people
Prior to the opening of the bridge, pedestrian crowd tests were were allowed to continue walking. According to Mistler et al.
carried out with more than 800 volunteers (Fig. 29a). In the first [114], normal walking became impossible for vibration amplitudes
test, the crowd was asked to cross the bridge (from one side) at exceeding 1.0 m/s2 and many people started to sway, increased
their own normal walking pace. In this scenario, lateral accelera- their gait width and holding on to each other or the hand rail dur-
tion amplitudes of between 0.19 and 0.45 m/s2 were recorded ing this crowd test.
without any evidence of diverging vibration amplitudes. Mistler Although the bridge clearly reacts strongly to pedestrian in-
et al. [114] attribute this to the general slow walking speed of peo- duced lateral loading, no dampers have been installed. One reason
ple. In the second test, the group was asked to cross the bridge at a is that under normal walking conditions excessive lateral vibra-
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
a b
Acceleration [m/s2]
Uncritical
-1
0 50
Fig. 29. Crowd tests on 13 January 2007 (picture from [113]) (a) and measured midspan acceleration during a crowd test (figure after [114]) (b).
tions did not occur, but only when people were instructed to in-
crease their walking speed. Further, the owners believe that the
crowd size needed to trigger large vibrations is a rare event and
that lateral vibrations will therefore not be a problem on a daily
basis [28,113,115].
6.8. Summary of full-scale measurements
In new long span footbridges, the possibility of excessive lateral
vibrations poses a challenge in the design and the role of full scale
measurements of the real dynamic behaviour of the bridge prior to
its inauguration is increasingly becoming an integrated part of the
design process. In particular, for bridges that are deemed suscepti-
ble to human induced vibrations, full scale testing is vital for sev-
eral reasons.
 The structural response (and critical number of pedestrians)
depends on the inherent structural damping, which cannot
accurately be predicted without testing.
 External damping devices, such as TMDs, depend on an accurate
tuning to the structural properties [116], which is usually only
possible through experimental modal identification.
 Due to uncertainties in the phenomenon governing excessive
lateral vibrations and the limited amount of data from existing
bridges, controlled crowd tests are beneficial for investigating
the possibility of excessive vibrations and finding its trigger
(e.g. critical crowd density) and for the purpose of verifying
the selected mitigation strategy.
A well engineered and successfully completed test program,
cannot only further the understanding of the problem with lateral
footbridge vibrations, but also contribute to a more safe and eco-
nomic design, which is based on the actual dynamic properties of
the bridge instead of those estimated during the design. The Coim-
bra footbridge (Section 6.4) is a prime example of the importance
of full-scale measurements, as the total damper mass needed on
the bridge to suppress pedestrian-induced lateral vibrations was
much larger than anticipated during the design process. This was
a consequence of wrongly assumed damping during the design
and partly due to a desire to provide extra damping to take into ac-
count uncertainties related to the dynamic behaviour of the struc-
ture [107].
On several occasions, the formula proposed by Arup (Eq. (9)) has
provided reasonable estimates for the number of pedestrians
needed to trigger large amplitude lateral vibrations [101,107,112].
However, there is conflicting evidence as to the necessity of syn-
chronisation between the gait cycle frequency of the crowd and
the lateral movement of the bridge for the development of
motion-induced pedestrian forces in the form of negative damping
[101,112].
39
Free decay of bridge
0,3 m/s²
STOP
100 150 200
Time [s]
Provided that pedestrians generally act as negative dampers on
low-frequency lateral vibration modes, all bridges with frequencies
in the range 0.4–1.3 Hz, with or without externally added damping,
have a potential to suffer from excessive pedestrian-induced vibra-
tions. The main concern of the designers will then be to determine
the probability of occurrence of the critical crowd density such that
an informed decision about the amount of external damping
needed can be taken or even avoided altogether as was done with
the Weil-am-Rhein bridge.
7. Developments in the modelling of pedestrian induced lateral
excitation
A critical review of the models proposed for crowd-induced lat-
eral loading of footbridges is presented, classified according to
their nature. In particular, two effects have been identified when
modelling human induced vibrations; human–human and hu-
man-structure interactions. Human-structure interaction is the po-
tential change in the walking pattern of an individual (e.g. walking
speed, frequency, phase or step length or width) due to the move-
ment of the structure and human–human interaction is the change
due to the surrounding people.
A review of interdisciplinary studies related to the dynamics of
moving crowds and human–human interaction is provided in Sec-
tion 7.1, followed by a review of pedestrian loading models includ-
ing modelling of human-structure interaction.
In the modelling of pedestrians-induced lateral vibrations, the
end goal is either to determine the magnitude of the structural re-
sponse under a given crowd or to determine the critical number of
pedestrians needed to cause diverging vibration amplitudes. In re-
cent years, several new load models and response evaluation tech-
niques have been proposed, originating from a variety of different
scientific disciplines. Initially, a number of linear models are pre-
sented (Section 7.2) which fall in two categories; (1) stability crite-
ria, which provide a critical number of pedestrians needed to
trigger excessive vibrations and (2) response evaluation methods
(deterministic or random) for prediction of the vibration magni-
tude. Next, nonlinear models are treated (Section 7.3). These mod-
els can be split into four categories (Fig. 30); (1) models that rely on
parametric excitation of modes with frequencies lower than the
dominant frequency of excitation, (2) representation of the hu-
man-body as an inverted pendulum, (3) load with nonlinear veloc-
ity dependency (modified Arup models), and (4) direct resonance
excitation in which the frequency and/or phase angle of the load
depends on the structural response.
7.1. Dynamics of a moving crowd
A crowd of pedestrians is an extremely complex system and a
complete description of the walking pattern of each individual is
40 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

7.1.1. Spatially unrestricted pedestrian flow

The walking speed determines the time spent by each pedes-
trian on the footbridge and the importance of this parameter is fur-
ther illuminated through the geometric relationship between the
walking speed (vp), the step length (lp) and the step frequency
(fp), vp = lp fp. In relation to design of footbridges, walking speed
of pedestrians has been reported in the range from 0.75 m/s for
slow walking to 1.75 m/s for fast walking, [125]. In recent years
other researchers have investigated the walking characteristics of
individuals in order to map their natural choice of walking speed
and step frequency [31]. A summary of different studies related
to the average unrestrained walking characteristics of pedestrians
is shown in Table 4. Typically researchers report a relationship be-
tween the step frequency of a person as a function of the freely se-
lected (unrestricted) walking speed. A linear relationship of the
type:

fp ¼ av p þ b ð27Þ

is often used, e.g. by Pachi and Ji [126], who report b = 0 and

Fig. 30. A flow diagram illustrating various methods in the modelling of human-
induced lateral vibrations of footbridges.
only possible in a probabilistic sense. Several studies on the walk-
ing characteristics of crowds have been undertaken in the past, of-
ten in relation to urban design and planning of pedestrian facilities
[117,118] and in studies related to evacuation of structures during
catastrophic events [119–121]. Modelling of pedestrian flow can be
made on a microscopic level where each pedestrian is modelled
uniquely or macroscopically where the crowd is modelled through
its average characteristics [118]. A widely used approach to the
modelling of macroscopic pedestrian flow is the continuum ap-
proach where the governing parameters are the crowd density, q
(expressed in persons/m2), the (mean) crowd speed, v, and the pe-
destrian flux, q  v, (persons/m/s) [122,123].
For dense crowds, the relationship between the density and the
flux may be expressed using the mass conservation equation in its
Eulerian form either in 2D [122] or 1D (which is more relevant for
pedestrian bridges) [124]:
@q @
þ ðq  v Þ ¼ 0
@t @x
a = 1.33 m1 for men and a = 1.49 m1 for women. Zivanovic et al.
[127] use a similar expression with b = 0.355 Hz and a = 1.075 m1.
Their observations are based on data obtained from 939 different
pedestrians during a 6.5 h monitoring of an indoor footbridge. Based
on the freely selected walking characteristics of 116 pedestrians
walking along a 72 m corridor, a linear fit of the aforementioned type
to the data yielded b ffi 0 and a = 1.326 m1 [128]. Ingólfsson et al.
[129] used the results from different studies [130–132] to formulate
a power law for the relationship between step frequency, walking
speed and step length:
fp ¼ 1:62v 0:35
p ð28Þ
Kasperski [54] points out that both physical, psychological and
environmental parameters determine the walking speed of a per-
son and the average value measured in 20 different cities in Ger-
many is in the range 1.41–1.55 m/s for men and 1.36–1.46 m/s
for women. The observations were made for people walking freely
in the respective city centres [133].
7.1.2. Human–human dynamic interaction
As the crowd density increases, the possibility of free move-
ment becomes restricted due to the reduction of available space
and people will need to adjust their walking speed to the speed
of the surrounding crowd [31,137]. In an early study on the dy-
namic behaviour of footbridges, Wheeler [125,138] postulated that
when the number of people on a bridge is increased, a ‘‘regimenta-
tion into a common forward speed and pacing frequency’’ occurs,
ð26Þ
without people falling into step.

Table 4
Comparison between unrestrained average walking characteristics according to different studies.

Study N Walking speed (m/s) Pacing rate (Hz)

Men Women Total Men Women Total
Matsumoto et al. [134] 505 – – – – – 1.99
Zivanovic et al. [135] 1976 – – – – – 1.87
Morgenroth [133] (after [54]) 6000 1.49 1.41 1.45 – – –
Ricciardelli and Pizzimenti [57] 116 1.44 1.37 1.41 1.81 1.86 1.84
Sahnaci and Kasperski [64] 251 1.37 1.36 1.37 1.80 1.91 1.82
Pachi and Ji [126]a 400 1.35 1.25 1.30 1.80 1.86 1.83
Pachi and Ji [126] b 400 1.46 1.37 1.42 1.97 2.03 2.00
Zivanovic et al. [127] 939 1.51 1.45 1.47 1.89 1.98 1.94
Butz [52] 98 – – – 1.80 1.89 1.84
Venuti and Bruno [136] c – 1.34aGaT – – –
a
Measured on two different footbridges.
b
Measured on floors in two different shopping malls.
c
The coefficient aG depends on the geographic location (1.05 for Europe, 1.01 for USA and 0.92 for Asia) and the coefficient aT depends on the travel purpose (1.20 for Rush
hour/Business, 1.11 for Commuters/Events and 0.84 for Leisure/Shopping).
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
Bertram and Ruina [139] used 12 test subjects to investigated
the change in the step frequency for an imposed change in the
walking speed. The results from their study were used by Bruno
and Venuti [140] to formulate the following law for the lateral forc-
ing frequency as a function of the walking speed:
 
fw ¼ 0:35v 3p  1:59v 2p þ 2:93v p =2 ð29Þ
Human–human interaction denotes the interaction (and possi-
bly synchronisation) that occurs within a crowd of pedestrians
and the change in walking pattern as a consequence of close prox-
imity of other pedestrians on the bridge. Probably the most pro-
nounced type of human–human interaction occurs when the
pedestrian density increases not allowing each pedestrian to walk
freely according to his/her own preferences. Tentative values in the
range 0.2–0.5 ped/m2 have been suggested as a limit value for free
unrestrained walking [8,12,137]. However, Butz et al. [141] re-
ported that during a harbour fair in Duisburg a pedestrian stream
with density of 0.3 ped/m2 occupied a footbridge and the average
step frequency in the group was about 1.5 Hz, considerably lower
than what is usually reported for unconstrained walking. This
observation is confirmed by the tests performed on the London
Millennium Bridge where a near linear decrease in walking speed
for increasing crowd density was observed, even at very low crowd
densities [142].
Experimental investigations relating to human–human interac-
tion were performed by Zoltowski [143] who found that the pedes-
trian step frequency decreases with decreasing crowd densities.
Butz [52] used a group of 18 persons who were asked to walk in
one group with density varying from 1.2 to 3.0 ped/m2 to show a
linear relationship between the forward speed of the group and
the average step frequency. Further, the standard deviation of the
step frequencies within the group generally increased with the
speed and decreased with the density of the group [52]. Andersen
[144] used up to 80 volunteers walking in six different conditions
(free walk to 1 ped/m2) to reveal that the average step frequency,
walking speed and stride length decreased for increasing crowd
density. However, the standard deviation of step frequency re-
mained nearly constant. The same study also showed that the dis-
tribution of individual phases was random with a near-uniform
distribution at all tested crowd densities, suggesting limited step
synchronisation as a result of human–human interaction. This
agrees with observations of Barker [17], which are based on 1200
pedestrians walking normally over different footbridges. In a re-
cent study it was shown that the standard deviation of individual
walking frequencies within a group tends to decrease with an in-
crease in the crowd density [145]. It was also shown that as the
crowd density increases, intra-subject variation in the walking fre-
quency of each individual tends to increase.
An excellent review of the interdisciplinary studies relating to
human–human interaction, was provided by Venuti and Bruno
[31,136] and Bruno and Venuti [140], where a general form of
the Kladek formula from vehicular traffic to represent the average
walking speed of the crowd as a function of the crowd density was
proposed:
  
ity of the model to predict the equivalent load amplitude for a large
1 1
v ¼ vM 1  exp c  ð30Þ
q qM
Here vM is the mean maximum speed, qM is the maximum den-
sity causing the crowd to stop, which depends on the travel pur-
pose and the geographic location [136]. The parameter c = bqM is
a fitting parameter determined by the type of pedestrian activity
(leisure b = 0.245, commuter b = 0.214 and rush hour b = 0.273).
Other pedestrian speed–density relations are given in the review
by Ishaque and Nolan [146] and Venuti and Bruno [31].
41
7.2. Linear response models
Linear response models are here classified as those where the
lateral vibrations are caused by direct resonance, i.e. by pedestrians
walking at frequencies close to the natural frequency of one or
more lateral vibration modes.
7.2.1. Equivalent number of ‘perfect’ pedestrians
The term ‘perfect’ pedestrian, refers to the assumption that the
lateral (or vertical) load time history from a series of consecutive
footsteps can be represented as a truncated Fourier series with a
fundamental period equal to the duration of two steps. As dis-
cussed in Section 3, this assumption implies that the load from
each footstep is perfectly replicated and thereby intra-subject var-
iability in the loading is neglected. The response from a group of
pedestrians can be calculated as a multiplication of the response
from single perfect pedestrian walking with a frequency that
matches the natural frequency of the mode in question. This meth-
od is particularly preferred in codes of practice and design guide-
lines due to its simplicity. The effect of the group is taken into
account through an effective number of pedestrians ne, that pro-
duces the same response as that of the entire group:
FðtÞ ¼ ne G1 sinð2pf0 tÞ ð31Þ
where f0 is the frequency of the mode in question and G1 is the dy-
namic load amplitude of a single pedestrian. The effective number
of pedestrians depends primarily on crowd specific and structural
parameters. For lateral human-induced vibrations, the effective
number of pedestrians has been deduced from video recordings of
the motion of pedestrians or through energy considerations. The
idea of an effective number of pedestrians has been implemented
pffiffiffiffiffiffi
in the Sétra
pffiffiffiffi guideline [6] as 10:8 Nf for a sparse or dense crowd
and 1:85 N for a very dense crowd. These results are based on sto-
chastic response simulations for vertical footbridge response and do
not take into account human-structure interaction. The movement
of the bridge, u, was accounted for by Danbon and Grillaud [68]
who defined the effective number of pedestrians through a bilinear
synchronisation coefficient /(u) as described in Section 4.4. Rönn-
quist [91] defined the equivalent number of pedestrians, by fitting
a mathematical expression to the ratio between measured (peak)
response from a group of N people to the simulated response of a
single pedestrian crossing at resonance:
 1:6 !
N
ne ¼ 35  34 exp  ð32Þ
60
In Fig. 31 a comparison between different definitions of the
effective number of pedestrians is shown. It is worth noting that
the equivalent load amplitudes vary by an order of magnitude be-
tween different models. Models that assume that a certain portion
of the pedestrians walk at a common frequency and phase (ne = Np
or 0.2Np) result in a much larger load amplitude than those that as-
sume common (or near common) p step
pffiffiffiffiffiffi ffiffiffiffiffiffi frequencies, but mutually
independent phases ( N p or 1:85 N p ). For a moderate number
of pedestrians (less than about 200) the model of Rönnquist [91]
(Eq. (32)) deviates from the random crowd assumptions. The valid-
number of pedestrians is questionable, as it is based on experimen-
tal results using only 60 people. The lowest amplitude
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiis obtained
in the Sétra model [6] for sparse crowds (10:8 0:005N p , assuming
f = 0.005), which assumes random step frequencies and mutually
independent phases.
7.2.2. Stability criteria
Arup’s stability criterion from Eq. (9) is probably the most com-
monly used method to assess the susceptibility of a bridge to
42 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

Fig. 31. Comparison between different definitions of the effective number of pedestrians.

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
excessive pedestrian-induced lateral vibrations. Other stability cri- Gðfn Þ pNfn
teria exist, such as that of Newland [83] (Section 4.4) or that of rq€n ¼ p ðfn Þ ð35Þ
4M n fn fw
Roberts [147,148], who utilised the equilibrium between the lat-
eral acceleration of the body centre of mass and the lateral force in- where G(fn) is the load amplitude at the natural frequency fn, and
duced by the pedestrians. For Np pedestrians equally distributed pfw ðfn Þ is the probability density function for the gait cycle fre-
over the bridge length L then, according to Roberts [147] the lateral quency (with mean lfw and standard deviation rfw ), evaluated at
force per unit length induced by the pedestrians becomes: the modal frequency fn. The method was derived for predicting ver-
tical response of footbridges, but can equally well be applied for lat-
Np mp @ 2 up ðx; tÞ eral vibrations in the absence of human-structure interaction.
¼ Fðx; tÞ ð33Þ
L @t 2 Roberts [149] presented a similar probabilistic response evalu-
ation technique where the load is modelled as N stationary and
where mp is the average pedestrian mass and up(x, t) is the lateral
uniformly distributed harmonic oscillators, each with amplitude
displacement of the centre of mass. By assuming that the move- G1. Under the assumption that the instantaneous phase of the re-
ment of the centre of mass is sinusoidal, Eq. (33) is solved together
sponse is randomly distributed with a constant probability density,
with the equation of motion governing the bridge vibration. A sta- the standard deviation (rq€n ) of the modal acceleration response to
bility criterion is obtained for the condition where the amplitude
unsynchronised pedestrians is obtained (with symbols from Eq.
of the bridge vibration exceeds that of the body centre of mass: (34)) as:
rffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 þ a2 Mn 2
n x Z L
Ncr ¼ RL ð34Þ N G1 Hn;av
2a mp
½Un ðxÞ dx x
2 2H
w n;av
rq€n ¼ U2n ðxÞdx ð36Þ
aL 0 2L M n 0

In Eq. (34), Mn, xn and Un are the modal mass, frequency and shape The intra-subject variability in the load can be accounted for by
of mode n and Hn,av is an averaged value of the dynamic amplifica- representing the pedestrian load through its PSD. Ricciardelli et al.
tion, accounting for variations in the step frequencies amongst the [150] proposed that for laterally stiff footbridges with uniformly
pedestrians and L is the bridge length. The proposed value of Hn,av distributed pedestrians and a crowd density low enough for avoid-
to be used in design is tabulated in [148]. Furthermore, Eq. (34) ac- ing human–human interaction, the stationary RMS value of the
counts for the possibility of non-uniform pedestrian distribution, modal acceleration could be calculated using the well-established
through the factor a 2 [0; 1], such that aL is the portion of the bridge white-noise approximation [151]:
occupied by pedestrians. The average angular walking frequency is sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z
denoted xw = 2pfw. 1 pNfn 1 L
rq€n ¼ SFe ðfn Þ ½UðxÞ2 dx ð37Þ
2M n fn L 0
7.2.3. Stationary response due to random crowds Z 1
Due to inter- and intra-subject variability, the response evalua- fn SFe ðfn Þ ¼ fn SF ðfn =fw Þpðfw Þdfw ð38Þ
0
tion is often treated probabilistically. One of the earliest attempts
is that of Matsumoto et al. [134] who showed that for a flow of where SFe(fn) represents the PSD of the equivalent load of one pe-
identical pedestrians arriving randomly to a bridge with flow destrian and SF(fn/fw) is the PSD of a single pedestrian from Eq.
intensity k (pedestrians/second), the stationary RMS value of the (2). It can be shown that Eqs. (35) and (37) are identical when the
pffiffiffiffiffiffiffiffi
(vertical) acceleration response is kT 0 that of the single pedes- excitation and the mode shapes are sinusoidal functions. In this
trian, with T0 being the passage time of the pedestrian. The model case, the PSD of a single person, SF ðfn Þ ¼ 12 G2 ðfn Þ dðfn  fw Þ (d(fn  fw)
of Matsumoto et al. [134] assumes identical pedestrians and there- being the Dirac delta function) is substituted into Eq. (38) to yield
fore the distribution of step frequencies in a real crowd is not ac- the approximation in Eq. (35).
counted for, which for lightly damped structures may lead to
overly conservative estimates of the response [129]. 7.2.4. Stationary response due to random crowds including spatial
The distribution of step frequencies within the crowd were ta- correlation
ken into account by Dallard et al. [5], who proposed the following Brownjohn et al. [55] pointed out that synchronisation amongst
approximate formula for the standard deviation of the vertical humans happens in clusters where small groups of pedestrians
footbridge response for a sinusoidal vibration mode: walk in step and that the correlation may decrease with the
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
distance between the pedestrians. With analogy to turbulent buf-
feting wind load, a coherence function coh(f, x1, x2) 2 [0, 1] was de-
fined. However, no suggestion for the shape of the coherence
function is provided and only the two extremes, no correlation
(coh(f, x1, x2) = d(x1  x2)) and full correlation (coh(f, x1, x2) = 1) are
analysed. An expression for the PSD of the acceleration response,
SU€ ðf ; xÞ, at point x along the bridge deck is derived as:
Z L Z L
2 2
SU€ ðf ; xÞ ¼ UðxÞ jHðf Þj SP ðf Þ Uðx1 ÞUðx2 Þcohðf ; x1 ; x2 Þdx1 dx2
0 0
 2 2
N G ðf Þ
SP ðf Þ ¼ pfw ðf Þ
L 2
where H(f) is the frequency response function (FRF) for accelera-
tion, SP(f) is the PSD of walking loads per unit length, N is the
number of people on the bridge, L is the bridge length, G(f) is
the amplitude of the relevant load harmonic and U(x) is the mode
shape function [55]. The RMS acceleration response is obtained
from the PSD through integration over the frequency domain
R
r2u€n ðxÞ ¼ 01 SU€ ðf ; xÞdt.
7.2.5. Peak response due to random crowds
In an attempt to distinguish between peak and RMS accelera-
tions, a spectral load model has been proposed using Monte Carlo
response simulations [52,152]. The 95% fractile of the lateral peak
modal acceleration amplitude (q €0;max ) is given as:
€0;max ¼ kq€;95%
q rq€
where kq€;95% is the peak factor which depends on the crowd density
(varies in the range 3.63–3.77). The variance of the acceleration re-
sponse, r2q€ , of modes with sinusoidal shape is given as:
CkF N
r2q€ ¼ k1 ðfn Þfnk2 ðfn Þ
M 2n
where C and kF are constants and k1(fn) and k2(fn) are both polyno-
mial functions with parameters determined from Monte Carlo re-
sponse simulations [152]. When the average gait cycle frequency
fw,m differs from the natural frequency fn of the mode in question,
a reduction factor is defined as [153]:
 
1 fw;m  fn
kred ¼ exp 
2 Bred
where Bred is a constant determined from simulations.
7.3. Nonlinear dynamic models
For the vibration serviceability of footbridges, the presence of
human-structure interaction is often modelled through nonlinear-
ities in the loading. A special type of dynamic non-linearity is mod-
al coupling, which exists in structures with a certain geometric
coupling between two (or more) vibration modes when the ratios
between their natural frequencies are integer (or near integer)
numbers.
7.3.1. Modified Arup models
Based on the observed pedestrian behaviour during the large
amplitude lateral vibrations of the T-Bridge, Nakamura [154] pro-
posed modifications to Eq. (8). The modification assumes that
pedestrians will reduce their walking speed, or completely stop,
when the lateral velocity of the bridge deck becomes large. There-
fore, the response of the bridge will not reach infinity but is limited
to a certain level. The following equation for the modal pedestrian
load was proposed:
43
_
F P ¼ k1 k2 H½qðtÞGðf 0 ÞM P g ð44Þ
_
qðtÞ
_
H½qðtÞ ¼ ð45Þ
_
k3 þ jqðtÞj
Gðf0 Þ ¼ 1:0 ð46Þ
_
where qðtÞ is the modal velocity of the bridge, MPg is the modal self
weight of the pedestrians and f0 is the frequency of the lateral vibra-
tion mode under consideration. The coefficient k1 = 0.04 is the fun-
ð39Þ damental DLF of the pedestrian (according to [40]) and k2 is the
percentage of synchronised (frequency and phase) pedestrians.
_
The function H½qðtÞ describes the nature of the synchronisation
where k3 should be determined by trial and error such that it corre-
ð40Þ
sponds to measured data [154]. The synchronisation coefficient,
k2 = 0.20, is based on the observations made on the T-Bridge and
those established from laboratory platform tests [44]. The parame-
ter k3 = 0.01 has been estimated from back-analysis of the measured
response of two different bridges (T-Bridge and the M-Bridge) un-
der different crowd conditions, but its determination is sensitive
to the accuracy of k1 and k2. It is worth noting that with these coef-
ficients, the load amplitude predicted by Eq. (8) is 48 N for a bridge
velocity of x_ ¼ 0:16 m=s, whereas the modified model in Eq. (44)
only predicts modal loads of around 5 N for the same velocity.
Ingólfsson and Georgakis [62] used the experimentally obtained
forces from the experimental campaign described by Ingólfsson
et al. [60] to define the total pedestrian-induced lateral force,
F(t), as an equivalent static force plus additional equivalent damp-
ing and inertia forces respectively, so that:
ð41Þ
FðtÞ ¼ F st ðtÞ þ cp ðf0 =fw ; u0 Þ  u_ þ mp .p ðf0 =fw ; u0 Þ  u € ð47Þ
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
equivalent damping equivalent inertia
The equivalent static force, Fst(t), was defined through its aver-
aged Gaussian shaped PSD (Eqs. (2)–(5)). The functions cp(f0/fw, u0)
and .p(f0/fw, u0) define the self-excited forces and its dependency
ð42Þ on the vibration frequency, f0, amplitude u0 and gait cycle fre-
quency fw. The model is based on measured forces from 66 pedes-
trian test subjects walking on a laterally driven treadmill at a range
of lateral vibration frequencies and amplitudes (Section 5) [60].
Apart from this frequency and amplitude dependency, the results
from the campaign were governed by a large randomness, thus
the pedestrian load coefficients were presented in a probabilistic
framework [62]:
ð43Þ cp ðf0 =fw ; u0 Þ ¼ h0 ðf0 =fw Þ þ h1 ðf0 =fw Þu0 þ X  h2 ðf0 =fw Þeh3 ðf0 =fw Þu0 ð48Þ
/3 ðf0 =fw Þu0
.p ðf0 =fw ; u0 Þ ¼ /0 ðf0 =fw Þ þ /1 ðf0 =fw Þu0 þ X  /2 ðf0 =fw Þe ð49Þ
The parameters h0, /0, h1 and /1 describe the development of
the mean load coefficient as functions of the lateral frequency
and amplitude and were determined by fitting a linear equation
to the measured mean values of cp and .p. The parameters h2, /2,
h3 and /3 were determined by fitting an exponential function to
the standard deviations of the measured values of cp and .p. The
stochastic variable X was introduced as a discrete-time Gaussian
Markov process. The main strength of the model is that the self-ex-
cited force components are based on an extensive experimental
campaign. Ingólfsson and Georgakis [62] demonstrated the appli-
cability of the model to predict excessive lateral vibrations of both
low frequency modes, i.e. around 0.5 Hz and modes with frequen-
cies closer to the natural gait cycle frequencies.
Based on the work of Ingólfsson [155], Gimsing and Georgakis
[156] present a stability criteria, in which the frequency-depen-
dent average value of cp can be obtained as a function of the mean
step frequency of the crowd. Ricciardelli et al. [157] similarly used
the data from the treadmill experiments of Ingólfsson et al. [60] to
quantify the pedestrian mass and damping coefficients (cp and .p).
A third order polynomial function was used to model the frequency
44 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52

dependency of the pedestrian damping coefficient at each of the
tested amplitudes. In addition, the coefficients of the polynomials
were obtained as functions of the amplitude. Thereby, the expected
pedestrian damping coefficient at zero amplitude could be defined
as:
 3  2  
cp0 f0 f0 f0
fp0 ¼ ¼ 1:77 þ 7:77  9:55 þ 3:19 ð50Þ
2x0 mp fw fw fw
where x0 is the angular vibration frequency and mp is the pedes-
trian mass. An expression for the acceleration proportional coeffi-
cient at zeros amplitude was also obtained, but by fitting a
different type of function through the data points:
   
1:11 f0
.p0 ¼ 9:17 sin exp 2:89 þ 0:56
f0 =fw fw
In Fig. 32, the extrapolated normalised pedestrian damping (a)
and mass (b) coefficients (at zero amplitude) are depicted. It is
noted that for mp = 75 kg, fw = 0.95 Hz, the maximum value of
cp = 361 Ns/m is somewhat higher than the value reported by Dal-
lard et al. [5]. A direct comparison would not be reasonable though,
as the value of cp = 300 Ns/m, obtained from the Millennium Bridge
data is an average value, valid at all amplitudes and modal fre-
quencies that occurred during the crowd tests.
By extrapolating the damping coefficient to zero amplitude, it
is assumed that instability starts from a static configuration of the
bridge. At larger amplitudes, the numerical value of the pedes-
trian damping coefficient decreases. As shown in Fig. 7a, the max-
imum acceleration during the pedestrian crowd tests on the
London Millennium Bridge (northern span) was around 80 mg
(or about 22 mm). At this amplitude, the damping coefficient is
only about one-quarter of the value at zero amplitude (Fig. 32a)
or cp(22 mm) ffi 90 Ns/m. Without any prior information about
the bridge amplitude at which the disproportionate increase ini-
tiates, the extrapolated value can be used to obtain a conservative
estimate for the critical number of pedestrians.
7.3.2. Amplitude dependent DLF
Based on their experimental work (Section 5), Sun and Yuan
[32] proposed an expression for the total load from a group of N
pedestrians:
N  qs ðu0 Þ  DLFðu0 Þ  W u_
FðtÞ ¼ ðu cosðuÞ þ sinðuÞÞ
u0 x0
where DLF(u0) is displacement dependent (Eq. (19)) and qs(u0) is
the probability of synchronisation according to Eq. (20), x0 is the
angular frequency of the mode in question, u is the phase angle be-
tween pedestrian loading and displacement u(t) of the structure
and W is the pedestrian weight. This equation is nonlinear, since
both DLF and the probability of synchronisation are functions of
u0. Sun and Yuan [32] derive an analytical stability criterion, which
must be solved numerically in order to determine the possibility of
excessive pedestrian-induced vibrations.
7.3.3. Modal coupling (autoparametric resonance)
Blekherman [33] investigated an elastic pendulum with a linear
spring subject to vertical forcing (Fig. 33a). The equations of mo-
tions governing the movement (x and #) of the system comprised
two coupled nonlinear second order differential equations. Blekh-
erman [158] derived the steady-state response for the pendulum
and the spring motion respectively and found that in the case
xs = 2X = 2xp there are two regions of interest, a linear and a non-
linear region (Fig. 33b). The frequencies xs, xp and X represent the
ð51Þ angular frequencies of the pendulum mode, swing mode and the
external excitation respectively. In the linear region, the amplitude
of the spring mode increases linearly with the applied load,
whereas the amplitude of the pendulum mode is zero. When the
force amplitude reaches the bifurcation point, the amplitude of
the vertical vibrations is limited (and independent of the forcing
applied) and the vibration of the pendulum mode experience a ra-
pid increase (jump).
According to Blekherman [33], the vibrations on the London
Millennium Bridge could be explained by autoparametric reso-
nance between the second lateral mode at 0.95 Hz and the third
vertical mode 1.89 Hz on the central span. The main argument is
that there is a 2:1 ratio between the modal frequencies and the
amplitude of the lateral load experienced a jump for a small in-
crease in the vertical load (due to a small increase in the number
of pedestrians on the bridge). He further points out that a relation-
ship of 1:2 between the natural frequency of the first lateral mode
(0.9–1.0 Hz) and the third vertical mode (2.0 Hz) of the T-Bridge in
Japan was observed.
Fujino et al. [159] used a scaled 3-degree-of-freedom (3DOF)
model of a cantilever beam supported by a taut string and found
that autoparametric resonance could be obtained in this system
for a load acting at the natural frequency of the vertical beam
mode. This caused lateral vibrations in both the cable stays and
the beam at half the excitation frequency, similarly to the pendu-
lum investigated by Blekherman. However, Fujino et al. [42] ar-
gued that parametric excitation was not the cause for large
amplitude vibrations on the T-Bridge. The vertical vibration levels
ð52Þ in the girder which are needed to trigger the autoparametric reso-
nance were not observed on the bridge. They also concluded that
the lateral cable vibrations were triggered by lateral vibrations of
the bridge deck, and not parametrically by the vertical girder mo-
tion as anticipated by the experiments reported by Fujino et al.
[159].
In a later paper, Blekherman [158] used a double pendulum
model to represent the first anti-symmetric torsional mode at fre-

a b

Fig. 32. (a) Average pedestrian damping and (b) mass coefficients based on the study of Ricciardelli et al. [157].
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
a b
Fig. 33. (a) Vertically loaded elastic pendulum with a linear spring and (b) response characteristics of the pendulum for xs = 2X = 2xp.
quency 1.59 Hz and the first lateral modes at frequency 0.81 Hz of
Pont de Solférino. Because of the near 2:1 relationship between the
modal frequencies, it was claimed that the bridge could have a
similar behaviour to that of the elastic pendulum (Fig. 33b). Fur-
thermore, Blekherman used the results from the full scale mea-
surements of the pont de Solférino to support this observation.
For a small group of 16 pedestrians marching in step at the fre-
quency of the torsional mode, the dominant frequency in the re-
sponse was the same as the step frequency. When the group size
was increased to 61 pedestrians, large vibrations developed in
the lateral mode at half the pacing frequency and the response
was governed by vibrations of both modes with the lateral one
dominating. This suggests that there may exist a critical load
parameter (e.g. number of resonance pedestrians) that causes
divergent lateral vibration amplitudes, similar to what is predicted
for the double pendulum.
7.3.4. Parametric resonance excitation
If the lateral force is nonlinear, e.g. through a displacement
dependency, lateral vibration modes with natural frequencies
equal half the gait cycle frequency, may be excited into parametric
resonance.
Piccardo and Tubino [160] defined an equivalent time-domain
model, in which the lateral pedestrian-induced force is propor-
tional to the bridge displacement:
Fðx; tÞ ¼ k½DLF1 þ DLFin uðx; tÞgmp ðxÞ cos xw t
The parameter k defines the percentage of synchronised (fre-
quency and phase) pedestrians, DLF1 equals the load amplitude
in the absence of lateral motion, DLFin is the proportion of the body
weight in phase with the bridge displacement, g is the acceleration
of gravity, mp(x) is the distribution of pedestrian mass, u(x, t) is the
displacement of the bridge and xw is the forcing frequency (angu-
lar gait cycle frequency).
For this type of loading, the equation of motion is a non-homo-
geneous damped Mathieu equation, which is characterised by
parametric excitation [161]. Piccardo and Tubino [160] analyse
the characteristics of this equation using a perturbation technique
and find that the susceptibility to instability depends on the damp-
ing fj, frequency ratio xj/xw between the modal frequency and the
forcing frequency and the ratio of the mass of the synchronised
pedestrians kMpj to the modal mass Mj. Their proposed stability cri-
terion is given as:
v"ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
#2
u
u x2
t 4 j  1 þ 16 xj f2
2 2
M pj x w
< ð54Þ
Mj 2gkDLFin xw2 xw 2 j
The minimum number of people needed to cause instability is
obtained in the case of xw = 2xj. The stability criterion assumes
45
that a fraction k of the pedestrians synchronise their step frequency
(and phase) to the frequency xw. Piccardo and Tubino [160] used
the London Millennium Bridge as a benchmark test to verify the
applicability of the procedure and find that when assuming
that 30% of the pedestrians synchronise at frequency xw/(2p) =
0.96 Hz (pacing rate 1.92 Hz), which is exactly twice the frequency
of the first central span lateral mode (0.48 Hz), the stability criterion
in Eq. (54) compares well with the observations made on the bridge.
The stability criterion is very sensitive to the relationship be-
tween the step frequency and the modal frequency and a very ra-
pid increase in the critical number of pedestrians is observed when
xw deviates from 2xj. It is worth noting that the expression in (53)
states that the pedestrian load has a sinusoidal component
(DLFingmp(x)cos xwt) which is multiplied by the displacement time
history u(x, t).
Later, Tubino and Piccardo [162] used the results of Ingólfsson
et al. [60] to obtain a qualitative validation of their force model.
However, the assumed displacement proportionality has still not
been validated experimentally.
7.3.5. Pedestrian phase synchronisation
Phase synchronisation originates from the theory of coupled
oscillators, primarily known from large biological and chemical
systems [163]. The first attempt to describe the lateral vibrations
of the London Millennium Bridge within the framework of coupled
ð53Þ oscillators was given by Strogatz et al. [164] who state that ‘‘wob-
bling and synchrony are inseparable’’. Their model is based on a set
of coupled nonlinear ordinary differential equations which (1) de-
scribe the modal response of the structure and (2) the develop-
ment of the phase of each pedestrian in relation to that of the
bridge [164]:
GX N
_ þ x20 qðtÞ ¼
€ðtÞ þ 2fx0 qðtÞ
q sin Hj ðtÞ ð55Þ
M j¼1
_ j ðtÞ ¼ xwj þ cq ðtÞ sinðWðtÞ  Hj ðtÞ þ aÞ
H ð56Þ
0
The modal mass, damping and angular natural frequency are de-
noted M, f and x0 respectively, q(t) is the modal response, G = 30 N
is the pedestrian force amplitude, xwj is the angular gait cycle fre-
quency in the absence of bridge motion and Hj(t) is the phase, both
for pedestrian j. The parameter a, denotes the initial phase between
the pedestrian-induced load and the bridge displacement. The
change in phase is governed by the parameter c which denotes
the sensitivity of the pedestrian to bridge motion, the bridge ampli-
tude q0(t) and total phase W(t) + a.
Strogatz et al. [164] present a closed form solution for the critical
number of pedestrians that cause instantaneous synchronisation in
the simple case where a = p/2 and the (initial) distribution pxw ðxÞ
46 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
of walking frequencies is symmetric about the modal frequency x0
[164]:
4f K € þ x2w ðs  yÞ ¼ u
Ncr ¼ ð57Þ
p G c pxw ðx0 Þ
where K is the modal stiffness. The only unknown in this equation is
the coefficient c which was calibrated against the data from the Mil-
lennium Bridge to the value of 16 m1s1. The assumption that
a = p/2 is similar to that of Newland [83], which is a worst-case sce-
nario for the input force, i.e. being in phase with the modal velocity.
If the pedestrian gait cycle frequencies are normally distributed,
(with standard deviation rxw mean frequency equal the modal fre-
quency) the critical number of pedestrians becomes:
rffiffiffiffi
2 4fK
Ncr ¼ r ð58Þ
p G c xw
The basic idea in this type of load model, is that the walking fre-
quencies of the pedestrians (which initially are randomly distrib-
uted), lock-in to the frequency of the bridge if the external
stimulus (the vibration amplitude) is strong enough and/or if the
initial gait cycle frequency is close to the vibration frequency.
According to Abrams [165], the model described above will be va-
lid for modes with natural frequencies in the range 0.75–1.25 Hz.
In later publications by the same authors [166,167], the expres-
sion for the phase development in Eq. (56) was modified such that
the pedestrians react to the acceleration of the bridge rather than
its displacement.
It was shown that the critical number of pedestrians could be
obtained under similar simplified assumptions as described above
and that the parameters of the model could be calibrated against
the results from e.g. the Millennium Bridge. Abdulrehem and Ott
[168] showed that the critical number of pedestrians increases
quadratically with the average gait cycle frequency and the modal
frequency.
7.3.6. Pendulum walking models
At the first international footbridge conference in Paris 2002,
Barker [169] proposed a very simple load model for the lateral load
induced by a walking pedestrian. The basic assumption in the load
model is that the maximum lateral force can be determined by
resolving the ‘static’ vertical body force into a lateral force through
the leg inclination. It is further assumed that the centre of mass of
the body moves in a straight line along the longitudinal axis of the
bridge. In the absence of lateral bridge movement, this assumption
means that the lateral force is constant during one step cycle and
consists of an ideal square wave with frequency fw.
According to the model, large lateral responses may be expected
from a group of mutually independent pedestrians, whose average
gait cycle frequency is (and remains) away from the natural fre-
quency of the lateral mode. The author argued that this is, in fact,
what was observed on the Millennium Bridge in London and that a
dramatic review of load models used at the time (which are based
on the assumption of synchronisation) may be necessary, some-
what in line with the statement of Josephson [67]. It should be
noted however, that the main assumption that the centre of mass
of the body moves in a straight line, violates the equilibrium con-
dition, which states that a lateral GRF is caused by the inertia asso-
ciated with a lateral acceleration of the body.
Inspired by the work of Barker, Macdonald [170] developed a
load model that accounts for the lateral movement of the pedes-
trian’s centre of mass and a control scheme for regulating the gait,
based on research related to human balance. The basic assumption
is that the lateral force exerted by the pedestrian is obtained as the
overall inertia force associated with the movement of the body
centre of mass. The governing differential equation for the move-
ment of the pedestrians becomes:
y € ð59Þ
where y is the movement of the centre of mass, s is the lateral posi-
tion of the centre of pressure, u is the lateral bridge displacement
and xw is the angular gait cycle frequency. The force F exerted by
a pedestrian on laterally moving surface is thereby obtained as:
€Þ ¼ mp x2w ðs  yÞ
€þy
F ¼ mp ðu ð60Þ
where mp is the pedestrian body mass. It was shown that the lateral
movement of the centre of mass, in the absence of bridge motion, is
approximately sinusoidal. The acceleration however, is not sinusoi-
dal and the resulting lateral forces compare well in shape and mag-
nitude to measured lateral force time histories.
Interestingly, Macdonald argues that the lateral foot placement
is the most efficient means of maintaining balance rather than the
timing of the footstep, as studied by Johnson [171]. In other words,
pedestrians do not need to synchronise with the movement of the
bridge to walk comfortably on a vibrating footbridge. The balanc-
ing strategy from Hof et al. [172] involves only control of the lateral
position of the centre of pressure, i.e. s, which ensures that the cen-
tre of mass does not pass the centre of pressure with a certain mar-
gin of safety.
Through numerical simulations, Macdonald showed that by
using this model, a lateral force component at the natural frequency
of the bridge occurs. Further, it was shown that the force compo-
nents in phase with the velocity and acceleration of the bridge
respectively are proportional to displacement amplitude, suggest-
ing that pedestrians can be treated as added mass and damping.
The amplitude (and sign) of these force components, depend on
the natural frequency of the bridge. In the frequency interval 0.7–
1.7 Hz, the pedestrians act as negative dampers (Fig. 34a) and in
the interval 0.3–1.2 Hz the equivalent added mass per pedestrian
is negative (Fig. 34b) with up to 61% of the body mass [170].
Carroll et al. [173] present a microscopic model of the pedes-
trian crowd where each individual pedestrian is modelled as an in-
verted pendulum, utilising the model of Macdonald [170]. The two
dimensional motion of each pedestrian in the crowd is modelled
using the concept of ‘‘social forces’’, which govern the movement
(acceleration) of each pedestrian. Essentially, the social forces
determine the pedestrians desire to reach their destination and
to maintain their direction of walk and include repulsion forces,
which ensure that a minimum distance is kept to physical bound-
aries and the other pedestrians. Thereby, variations in both the
walking speed and direction of each pedestrian is obtained
depending on the selected characteristics of the social forces.
As shown in Fig. 34, the model is sensitive to the selected bal-
ance strategy and the resulting self-excited forces vary consider-
ably between the two control laws that were tested. From a
comparison between Fig. 34 and the results from the experimental
campaign of Ingólfsson et. al [60] (Fig. 16) and the extrapolation of
Ricciardelli et al. [157] (Fig. 32), the control law that is based on the
relative velocity of the centre of mass (solid curve in Fig. 34) ap-
pears to provide a closer match to the experimental results.
Carroll et al. [173] note that the balancing strategy of Hof et al.
[172] is based on pedestrian walking on a stationary surface and
question its applicability for walking on a laterally moving bridge.
It is argued that the pedestrian must be able to react actively to a
continuously changing inertia force which is caused by the bridge
oscillation and acts as an input to the vestibular system of the
body.
Erlicher et al. [174] suggest that the lateral pedestrian-induced
force can be modelled as the restoring force of a modified hybrid
Van der Pol/Rayleigh oscillator. By tuning the parameters of the
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52 47

a b

Fig. 34. Simulated equivalent damping coefficient (a) and added mass (b) per pedestrian shown with estimated values from full scale measurements. LMB: London
Millennium Bridge [5], CMB: Changi Mezzanine Bridge [101], CSB: Clifton Suspension Bridge [112], solid curve: control law based on relative velocity of the centre of mass,
dashed curve: control law based on absolute velocity of the centre of mass (after [170]).

oscillator, lateral GRFs as measured on a rigid surface were repro- € 0 is the envelope of the accelera-
where a = 3.14, b = 2.68, fr = fw/f0, u
duced with a good degree of accuracy. The model has been applied tion time history, qsync is the crowd density that corresponds to to-
on a laterally moving floor where initial analysis shows that the tal synchronisation (proposed as 1.8 ped/m2) of the pedestrians, qc
load amplitude of the fundamental harmonic increases with the is the upper limit for unconstrained free walk, proposed as 0.3 ped/
vibration amplitude and that the frequency of the oscillator is en- m2, and u€ c is the corresponding acceleration [178].
trained by the frequency of the moving floor [175]. The modelling of the pedestrian crowd is governed by Eq. (26),
Similarly, Paulissen and Metrikine [176] developed a mathe- where the free walking speed, determined from Eq. (30), is multi-
matical model of the interaction between crowds and laterally plied with a correction function g(u € 0 ), which lowers the walking
moving bridge decks. In the model, the pedestrian-structure dy- speed as the platform vibration increases [178]. To the authors’
namic system is described through coupled differential equations, knowledge, no studies have been performed to determine the ef-
one describing the movement of the structure and the other a set fect of structural vibrations on the walking speed of pedestrians
of non-linear oscillators representing the movement of the pedes- and only limited experimental evidence exists for the synchronisa-
trians. Similar to Trovato et al. [175], the pedestrian movement is tion coefficients. However, the authors of the load model per-
formulated with analogy to the Van der Pol theory and the cou- formed a thorough comparison between the simulated response
pling between the pedestrian movement and that of the structure and the measured response on the T-Bridge in Japan which showed
occurs through a set of coupling coefficients. Currently, the param- a good agreement, both in terms of vibration amplitude and degree
eters of the model are only based on qualitative assumptions, but of synchronisation [178,180].
the authors stress the need for a calibration of the model to actual Bodgi et al. [181,182] used the same framework, i.e. a macro-
measurements of pedestrian-induced lateral forces. scopic modelling of the pedestrian flow assuming that the crowd
behaves like a compressible fluid. The ’closure’ equation, which de-
7.3.7. Continuous hydrodynamic crowd modelling approach fines the relationship between the crowd density and speed is
With analogy to fluid dynamics, Venuti et al. [124,177] pro- adopted from Venuti et al. [177], whereas the pedestrian load is as-
posed a model of crowd-structure interaction on footbridges. The sumed slightly differently [182]:
footbridge is modelled as a 1D beam and the modal pedestrian
Fðx; tÞ ¼ G1 sinð2pf0 tÞqbSðu;
_ qÞ ð63Þ
load, Pðt; q; q;
_ q€Þ, is a function of the crowd mass density, q, and
of the modal bridge vibration q, making the equation of motion
Sp ðqÞ €ðx; tÞ 6 u
u €min
non-linear. Furthermore, The modal mass is the sum of a contribu- _ qÞ ¼
Sðu; ð64Þ
_
Sps ðuÞ otherwise
tion from the structural mass Ms and that of the pedestrian crowd
Mp. ( pffiffiffi pffiffiffiffiffiffiffiffiffi
The pedestrian force is modelled as the sum of three contribu- 8:6 f= qbL q 6 qc
Sp ðqÞ ¼ pffiffiffiffiffiffiffiffiffi ð65Þ
tions, Fs due to uncorrelated pedestrians (i.e. mutually indepen- 1:75= qbL otherwise
dent), Fps due to pedestrian-structure synchronisation and Fpp
due to synchronisation amongst pedestrians (pedestrian-pedes- where b is the width of the bridge deck, L is its length, f0 is the lat-
€ min ¼ 0:1 m=s2 is defined as a
eral frequency of the relevant mode, u
trian synchronisation) [178].
vibration perception threshold and Sps ðuÞ_ is a velocity dependent
The average gait cycle frequency of the crowd is determined as
a function of the forward crowd speed v, according to Eq. (29) synchronisation function of a similar form to that in Eq. (62). The
[179]. Following the studies reported by Pizzimenti [61], the force load amplitude is taken as a function of the walking speed,
induced by the synchronised portion of the pedestrians is divided G1 = 0.6191 Ns/m  v + 35.5171 N. According to Bodgi et al. [182],
_ qÞ defines the equivalent num-
the synchronisation coefficient Sðu;
into a component in phase with the velocity of the structure and
one in phase with the acceleration, both of which are assumed to ber of pedestrians walking with the same frequency as that of the
increase with an increase in the bridge oscillation [179]. The lateral mode in question. This produces the same load as the entire
remaining pedestrians are assumed to walk at the average step fre- crowd and is based on the tests reported in Sétra [6].
quency determined from Eq. (29). Of the remaining pedestrians, a In a later paper by Bodgi et al. [183] a different load model,
number are assumed to walk in phase due to pedestrian-pedes- which is based on the theory of coupled oscillators (discussed in
trian interaction and the rest walk at a random phase [178]. The Section 7.3.5), coupled with the Eulerian description of the crowd,
number of synchronised pedestrians is determined through the is proposed:
synchronisation coefficients Sps and Spp, defined as: Fðx; tÞ ¼ qðx; tÞbG1 sin HðtÞ ð66Þ
€ 0 ; fr Þ ¼ ½1  e
Sps ðu € 0 u
bðu €c Þ
½e 50ðfr 1Þ2 e20u_ 0 =p
 ð61Þ @H e _ s ðtÞ2 sinðWðtÞ  Hðx; tÞÞ
   ðx; tÞ ¼ xw þ Au ðtÞjUðxÞj½W ð67Þ
qsync þ qc @t 2
1
Spp ðqÞ ¼ 1 þ erfðaqM q  Þ ð62Þ W_ ðtÞ ¼ x0 ðtÞ ð68Þ
2 2
48 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
Here xw is the (initial) angular pacing frequency, Au is the maxi-
mum displacement of the footbridge during the last two oscilla-
tions, U(x) is the mode shape of the relevant mode with angular
frequency xs(t) and e is a constant that determines the sensitivity
of the phase coupling between the pedestrian and the structure.
In this framework, a different closure equation has been proposed:
1 @H
v ðx; tÞ ¼ lp C s ðx; tÞC q ðx; tÞ ðx; tÞ
p @t
where lp is the free step and the functions Cs(x, t), 2 [0, 1] and
Cq(x, t), 2 [0, 1] reduce the pedestrian step length with an increase
in structural vibrations and local crowd density respectively. The
parameters of the model were calibrated to find the same critical
number of pedestrians, as on the Millennium Bridge [183]. Further
analysis of the performance of this model is presented by Bodgi [184].
8. Summary and conclusions
The existence of a form of bridge instability, in which large lat-
eral vibrations can develop under crowd loading, has been ob-
served and verified on numerous occasions. The research carried
out during the temporary closures of the Millennium Bridge in
London and pont de Solférino in Paris revealed that for a certain
critical number of pedestrians (or critical acceleration threshold),
the vibration amplitude increases disproportionately for a small in-
crease in the number of pedestrians on the bridge. The research
also highlighted the importance of taking into account human-
structure interaction when modelling the effect of pedestrians on
flexible structures. Until recently, it was commonly accepted that
this instability-type behaviour, which had become known as Syn-
chronous Lateral Excitation (SLE), occurs due to synchronisation
between the movement of the pedestrian and that of the bridge.
More precisely, the gait cycle frequency of the synchronised pedes-
trians is assumed to coincide with the natural frequency and phase
of the bridge in such a way as to increase the bridge movement.
Early laboratory tests focused on quantifying the transition
from random to synchronised walking and determine the frequency
amplitude ranges for which this occurs. A number of investigators
have shown that human-structure synchronisation can occur when
the freely selected initial gait cycle frequency is close to the fre-
quency of the moving platform. However, a number of important
observations from full-scale measurements of existing bridges
question the necessity of this form of synchronisation for the
development of excessive lateral vibrations. Vibration modes with
natural frequencies well below the average pedestrian gait cycle
frequency have exhibited large vibrations on a number of occa-
sions. Simultaneous vibrations in two modes were reported on
the Clifton Suspension Bridge [5,101,112]. In addition, a dispropor-
tionate increase in the vertical response at twice the modal fre-
quency has never been reported, which would be a consequence
of a sudden collective synchronisation of the step frequencies
and phases within a crowd.
In fact, the term synchronisation has been subject to various
interpretations, e.g. the synchronisation of the head motion of
pedestrians to the girder vibrations of the T-Bridge [42] has been
used to support the assumption of synchronised stepping. How-
ever, as shown by McAndrew et al. [86], the motion of a persons
neck during walking on a laterally moving surface features peaks
in the PSD at the frequency of the motion although the pedestrian
gait cycle frequency is different. As such, synchronisation of the
head motion to the structural motion can occur in the absence of
synchronised stepping. Indeed, the results of Pizzimenti [56] and
later Ingólfsson et al. [60] show that pedestrian-induced lateral
forces occur at the frequency of the lateral bridge motion even
though the pedestrian gait cycle frequency is different. The
question whether a person’s step frequency is affected by the
lateral vibration and how it might depend on the frequency and
amplitude of motion is still debated and not well understood.
However, models that rely solely on phase-synchronisation be-
tween the pedestrians gait cycle frequency and the bridge natural
frequency are insufficient.
Based on the available data from full-scale measurements and
laboratory tests, it appears that human-structure interaction is
ð69Þ governed by a number of coexisting phenomena rather than one
single mechanism. Pedestrians generate self-excited forces over a
large range of frequencies and their phase and magnitude depend
strongly on the vibration characteristics of the structure as well as
their individual walking pattern. Pedestrian-structure synchroni-
sation can also occur in situations where the pedestrian gait cycle
frequency is close to the modal frequency, but the exact nature of
this synchronisation and its effect on the overall crowd load is still
unknown. The possibility of phenomena such as nonlinear modal
coupling, parametric excitation or excitation of lateral modes
through vertical loads being responsible for excessive bridge vibra-
tions is still open to debate. For a number of the bridges presented
in this review, some of the necessary conditions have indeed been
observed, e.g. integer relationship between natural vibration
modes [158], although definite conclusions regarding these phe-
nomena cannot be drawn.
As far as laboratory work is concerned, most published tests
have only investigated the effect of single pedestrians, walking at
a freely selected speed in unrestricted (or semi-restricted) circum-
stances, neglecting the possible effect of the surrounding crowd. In
addition, most tests are carried out under idealised vibratory con-
ditions, such as single frequency and constant amplitude sinusoi-
dal vibrations or on platforms where only a limited number of
consecutive steps were recorded. As such, the effect of continuous
variations in the vibration amplitudes (as due to mode shape ef-
fects) and/or simultaneous response in several modes (as observed
on Clifton Suspension Bridge) is still not well understood. A com-
prehensive experimental mapping of the walking forces under
multi-modal vibration may prove practically difficult as the num-
ber of tests to be carried out to cover a reasonable number of cases
may be excessively large. Another strategy is to use measured data
(on a fixed or moving floor) to calibrate a mechanical model of the
mechanism that generates the ground reaction forces, such as
an inverted pendulum model of the body centre of mass [169,
170,175]. Such models can subsequently be used in more general
settings, such as multi-modal response or combined vertical and
lateral motion.
To date, most researchers have focused on the quantification of
the pedestrian-induced loads, without explicitly investigating the
mechanisms that generate these loads. Such empirical models
can only be verified through extensive testing to cover a range of
different circumstances. In addition, inter- and intra-subject vari-
ability, combined with the possible effect of the crowd density
on the involved parameters further complicates the process of cal-
ibrating human-structure interaction models. As pointed out by
Venuti and Bruno [31], a number of simplified force and response
evaluation models have emerged from the need for simple design
rules. However, such models are usually calibrated against a lim-
ited number of empirical observations and cannot accurately de-
scribe the full spectrum of involved phenomena. On the other
hand, comprehensive models require more experimental valida-
tion due to their large number of variables.
With experimental data available, the ability of researchers to
test, calibrate and verify their loads models increases. As illustrated
herewith, a number of existing bridges have shown susceptibility
to excessive lateral vibrations and a number of full-scale tests have
been carried out in the past. The information obtained from
these tests has mostly been limited to the determination of the
critical number of pedestrians needed to trigger excessive lateral
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52 49

vibrations. However, important information such as the distribu- modego river in Coimbre. In: Proceedings of Footbridge 2005, second
international conference, Venice, 2005.
tion of step frequencies within the crowd and its correlation with
[26] Caetano E, Cunha Á, Da Fonseca A, Bastos R, Da Fonseca Jr A. Assessment and
either the structural movement or the crowd density have gener- control of human induced vibrations in the new Coimbra footbridge. In:
ally not been provided, although this information is vital for the Proceedings of Footbridge 2005, second international conference, Venice,
verification process. As most bridges that are susceptible to large 2005.
[27] Hoorpah W, Flamand O, Cespedes X. The Simon de Beauvoir footbridge in
amplitude human-induced vibrations are equipped with dampers Paris. Experimental verification of the dynamic behaviour under pedestrian
prior to their opening, there are often limited opportunities to car- loads and discussion of corrective modifications. In: Proceedings of
ry out the relevant pedestrian crowd tests. The development of Footbridge 2008, third international conference, Porto, 2008.
[28] Low A. Design for dynamic effects in long span footbridges. In: Proceedings of
efficient standardised test protocols and data post-processing tech- Footbridge 2008, third international conference, Porto, 2008.
niques to ensure that the most important information is extracted [29] Živanović S, Pavic A, Reynolds P. Vibration serviceability of footbridges under
from the bridge before the retrofit is installed is therefore an human-induced excitation: a literature review. J Sound Vib 2005;279(1–
2):1–74. http://dx.doi.org/10.1016/j.jsv.2004.01.019.
important challenge for the future. [30] Racic V, Pavic A, Brownjohn JMW. Experimental identification and analytical
modelling of human walking forces: literature review. J Sound Vib
References 2009;326:1–49. http://dx.doi.org/10.1016/j.jsv.2009.04.020.
[31] Venuti F, Bruno L. Crowd-structure interaction in lively footbridges under
synchronous lateral excitation: a literature review. Phys Life Rev
[1] Eyre J. Aesthetics of footbridge design. In: Proceedings of Footbridge 2002,
2009;6:176–206. http://dx.doi.org/10.1016/j.plrev.2009.07.001.
first international conference, Paris, 2002.
[32] Sun L, Yuan X. Study on pedestrian-induced vibration of footbridge. In:
[2] Strasky J. New structural concept for footbridges. In: Proceedings of
Proceedings of Footbridge 2008, third international conference, Porto, 2008.
Footbridge 2002, first international conference, Paris, 2002.
[33] Blekherman AN. Swaying of pedestrian bridges. J Bridge Eng 2005;10(2):
[3] Bardsley H, Menard R, Consigny F, Vaudeville B. The Bercy-Tolbiac footbridge
142–50. http://dx.doi.org/10.1061/(ASCE)1084-0702(2005)10:2(142).
in Paris. In: Proceedings of Footbridge 2002, first international conference,
[34] Wolmuth B, Surtees J. Crowd-related failure of bridges. Proc ICE: Civil Eng
Paris, 2002.
2003;156(3):116–23.
[4] Duguid B. Benchmark cost and value of landmark footbridges. In: Proceedings
[35] Dallard P, Fitzpatrick T, Flint A, Low A, Smith RR, Willford M, et al. London
of Footbridge 2011, fourth international conference, Wroclaw, 2011.
Millennium Bridge: pedestrian-induced lateral vibration. J Bridge Eng
[5] Dallard P, Fitzpatrick AJ, Flint A, Le Bourva S, Low A, Smith RMR, et al. The
2001;6(6):412–7.
London Millennium Footbridge. Struct Eng 2001;79(22):17–33.
[36] Swaying causes running wariness over Bosphorus Bridge, Hürriyet Daily
[6] Sétra. Footbridges. Assessment of vibrational behaviour of footbridges under
News & Economic Review, 18 October 2010. <http://www.hurriyetdailynews.
pedestrian loading. The Technical Department for Transport, Roads and
com/n.php?n=swinging-bosporus-bridge-alerts-experts-2010-10-18>
Bridges Engineering and Road Safety, November 2006.
[accessed 08.08.11].
[7] Hoorpah W. Footbridges conference: History, reason and the way forward. In:
[37] Point of collapse. The Village Voice, New York, USA, 27 August 2003. <http://
Proceedings of Footbridge 2008, third international conference, Porto, 2008.
www.villagevoice.com/2003-08-26/news/point-of-collapse/> [accessed 08.08.11].
[8] FIB. Guidelines for the design of footbridges, bulletin 32, Fédération
[38] Ye Q, Fanjiang GN, Yanev B. Investigation of the dynamic properties of the
international du béton (fib), November 2005.
Brooklyn Bridge. In: Sensing issues in civil structural health monitoring,
[9] Butz C, Sedlacek G. Bemessungskonzept für fußgängerinduzierte
Springer Netherlands, 2005. p. 65–72.
Brückenschwingungen (design concept for pedestrian-induced bridge
[39] Peterson C. Theorie der Zufallsschwingungen und Anwendungen (theory of
vibrations). Stahlbau 2007;76(6):391–400. http://dx.doi.org/10.1002/sTable
random vibrations and applications). Work Report 2/72, Structural
200710042 [in German].
Engineering Laboratory, Technical University of Munich, 1972 [in German].
[10] Butz C, Heinemeyer C, Keil A, Schlaich M, Goldack A, Trometer S, et al. Design
[40] Bachmann H, Ammann W. Vibrations in structures. Induced by man and
of footbridges – guidelines and background document, HIVOSS, rFS2-CT-
machine, 3rd ed. Structural Engineering Documents, International Association
2007-00033, 2007.
for Bridge and Structural Engineering (IABSE), Zürich, Switzerland, 1987.
[11] Heinemeyer C, Feldmann M. European design guide for footbridge vibration.
[41] Bachmann H. Case studies of structures with man-induced vibrations. J Struct
In: Proceedings of Footbridge 2008, third international conference, Porto,
Eng 1992;118(3):631–47.
2008.
[42] Fujino Y, Pacheco BM, Nakamura SI, Warnitchai P. Synchronization of human
[12] Butz C, Feldmann M, Heinemeyer C, Sedlacek G, Chabrolin B, Lemaire A, et al.
walking observed during lateral vibration of a congested pedestrian bridge.
Advanced load models for synchronous pedestrian excitation and optimised
Earthq Eng Struct Dyn 1993;22(9):741–58.
design guidelines for steel footbridges. Project report RFSR-CT-2003-00019,
[43] Nakamura SI, Fujino Y. Lateral vibration on a pedestrian cable-stayed bridge.
Research fund for Coal and Steel, 2008.
Struct Eng Int: J Int Assoc Bridge Struct Eng (IABSE) 2002;12(4):295–300.
[13] Barker C, DeNeumann S, Mackenzie D, Ko R. Footbridge pedestrian vibration
[44] Nakamura S, Kawasaki T. A method for predicting the lateral girder response
limits. Part 1: Pedestrian input. In: Proceedings of Footbridge 2005, second
of footbridges induced by pedestrians. J Construct Steel Res
international conference, Venice, 2005.
2009;65:1705–11. http://dx.doi.org/10.1016/j.jcsr.2009.03.003.
[14] Barker C. Footbridge pedestrian vibration limits. Part 3: Background to
[45] Nakamura SI, Kawasaki T. Lateral vibration of footbridges by synchronous
response calculations. In: Proceedings of Footbridge 2005, second
walking. J Construct Steel Res 2006;62(11):1148–60. http://dx.doi.org/
international conference, Venice, 2005.
10.1016/j.jcsr.2006.06.023.
[15] Mackenzie D, Barker C, McFadyen N, Allison B. Footbridge pedestrian
[46] Harper FC, Warlow WJ, Clarke BL. The forces applied to the floor by the foot in
vibration limits. Part 2: Human sensitivity. In: Proceedings of Footbridge
walking. 1. walking on a level surface, Tech. rep., National building studies –
2005, second international conference, Venice, 2005.
research paper 32. Department of Scientific and Industrial Research, 1961.
[16] Barker C, Daly AF. Dynamic response properties of footbridges: design
[47] Andriacchi TP, Ogle JA, Galante JO. Walking speed as a basis for normal and
proposals for pedestrian excitation. Unpublished project report UPR/ISS/48/
abnormal gait measurements. J Biomech 1977;10(4):261–8.
05, July 2005.
[48] Masani K, Kouzaki M, Fukunaga T. Variability of ground reaction forces during
[17] Barker C. Footbridge pedestrian vibration limits background to response
treadmill walking. J Appl Physiol 2002;92(5):1885–90. http://dx.doi.org/
calculation. Int J Space Struct 2007;22(1):35–43.
10.1152/japplphysiol.00969.2000.
[18] NA to BS EN 1991-2:2003 UK. National annex to Eurocode 1: Actions on
[49] Chao EY, Laughman RK, Schneider E, Stauffer RN. Normative data of knee joint
structure – Part 2: Traffic loads on bridges. British Standards Institution,
motion and ground reaction forces in adult level walking. J Biomech
London, May 2008.
1983;16(2):219–33.
[19] Barker C, Mackenzie D. Calibration of the UK National Annex. In: Proceedings
[50] Bachmann H, Pretlov AJ, Rainer H. Vibration problems in structures: practical
of Footbridge 2008, third international conference, Porto, 2008.
guidelines. Birkhäuser; 1996 [Chapter Appendix G: dynamic forces from
[20] Caetano E, Cunha A, Raoul J, Hoorpah W, editors. Footbridge vibration
rhythmical human body motions].
design. Taylor and Francis; 2009.
[51] Crowe A, Samson MM, Hoitsma MJ, van Ginkel AA. The influence of walking
[21] Moutinho C, Caetano E, Cunha A, Adao Da Fonseca A. Dynamic behaviour of a
speed on parameters of gait symmetry determined from ground reaction
long span stainless steel arch footbridge. In: Proceedings of Footbridge 2002,
forces. Hum Movement Sci 1996;15:347–67.
first international conference, Paris, 2002.
[52] Butz C, Beitrag zur Berechnung fußgängerinduzierter Brückenschwingungen
[22] Cespedes X, Manon S. Dynamic analysis of the Bercy-Tolbiac footbride. In:
(on the calculation of pedestrian-induced vibration of bridges). PhD thesis,
Proceedings of Footbridge 2005, second international conference, Venice,
RWTH Aachen, 2006 [in German].
2005.
[53] Sahnaci C, Kasperski M, Excitation of buildings and pedestrian structures
[23] Fournol A, Gerard F, De Ville V, Duchene Y, Maillard M. Dynamic behaviour of
from walking and running. In: Proceedings of experimental vibration analysis
Ceramique footbridge (Maastricht, NL). In: Proceedings of Footbridge 2005,
for civil engineering structures, Porto, 2007. p. 209–18.
second international conference, Venice, 2005.
[54] Kasperski M, Sahnaci C. Serviceability of pedestrian structures. In:
[24] Powell D, De Donno A, Low A. Design of damping systems for footbridges.
Proceedings of the 25th IMAC Conference, Orlando, Florida, 2007.
Experience from Gatwick and Ijburg. In: Proceedings of Footbridge 2005,
[55] Brownjohn JMW, Pavic A, Omenzetter P. A spectral density approach for
second international conference, Venice, 2005.
modelling continuous vertical forces on pedestrian structures due to walking.
[25] Alves R, Figueiras J, Pimentel M, Félix C, Dimanda A, Cunha A et al.
Can J Civil Eng 2004;31(1):65–77. http://dx.doi.org/10.1139/L03-072.
Instrumentation, monitoring and execution control of the footbridge over
50 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
[56] Pizzimenti AD. Analisi sperimentale dei meccanismi di eccitazione laterale
delle passerelle ad opera dei pedoni (experimental analysis of the lateral
pedestrian-induced mechanism of excitation of footbridges). Ph.D. thesis,
Department of Civil and Environmental Engineering, University of Catania,
2004 [in Italian].
[57] Ricciardelli F, Pizzimenti AD. Lateral walking-induced forces on footbridges. J
Bridge Eng 2007;12(6):677–88. http://dx.doi.org/10.1061/(ASCE)1084-
0702(2007)12:6(677).
[58] Ohlsson SV. Floor vibration and human discomfort. PhD thesis, Chalmers
University of Technology, Göteborg, 1982.
[59] Eriksson PE. Vibration of low-frequency floors – dynamic forces and response
prediction. PhD thesis, Chalmers University of Technology, Department of
Structural Engineering, Göteborg, March 1994.
[60] Ingólfsson ET, Georgakis CT, Ricciardelli F, Jönsson J. Experimental
identification of pedestrian-induced lateral forces on footbridges. J Sound
Vib 2011;330:1265–84. http://dx.doi.org/10.1016/j.jsv.2010.09.034.
[61] Pizzimenti AD, Ricciardelli F. Experimental evaluation of the dynamic lateral
loading of footbridges by walking pedestrians. In: Proceedings of the 6th
international conference on structural dynamics, Paris, 2005.
[62] Ingólfsson ET, Georgakis CT. A stochastic load model for pedestrian-induced
lateral forces on footbridges. Eng Struct 2011;33(12):3454–70. http://
dx.doi.org/10.1016/j.engstruct.2011.07.009.
[63] Racic V, Brownjohn JMW. Mathematical modelling of random narrow band
lateral excitation of footbridges due to pedestrians walking. Comput Struct
2012;90–91:116–30. http://dx.doi.org/10.1016/j.compstruc.2011.10.002.
[64] Sahnaci C, Kasperski M. Random loads induced by walking. In: Proceedings of
the 6th European conference on structural dynamics, Southampton, 2005.
[65] Cookson B. Crossing the river. Edinburgh: Mainstream Publishing; 2006.
[66] Fitzpatrick T, Dallard P, Le Bourva S, Low A, Smith RR, Willford M. Linking
London: The Millennium Bridge, Tech. rep., Royal Academy of Engineering,
2001.
[67] Josephson B. Out of step on the bridge, The Guardian, Wednesday 14 June
2000. <http://guardian.co.uk/theguardian/2000/jun/14/guardianletters3>
[accessed 08.08.11].
[68] Danbon F, Grillaud G. Dynamic behaviour of a steel footbridge.
Characterisation and modelling of the dynamic loading induced by a
moving crowd on the Solferino footbridge in Paris. In: Proceedings of
Footbridge 2005, second international conference, Venice, 2005.
[69] Dziuba P, Grillaud G, Flamand O, Sanquier S, Tétard Y. La passerelle Solférino
comportement dynamique (dynamic behaviour of the Solférino bridge). Bull
Ouvrages Métalliques 2001;1:34–57 [in French].
[70] Sachse R. The influence of human occupants on the dynamic properties of
slender structures. PhD thesis, University of Sheffield, UK, April 2002.
[71] Živanović S, Pavić A, Ingólfsson ET. Modeling spatially unrestricted pedestrian
traffic on footbridges. ASCE J Struct Eng 2010;136(10):1296–308. http://
dx.doi.org/10.1061/(ASCE)ST.1943-541X.0000226.
[72] Živanović S, Díaz IM, Pavić A. Influence of walking and standing crowds on
structural dynamic properties. In: Proceedings of the 27th IMAC Conference,
Orlando, USA, 2009.
[73] Jørgensen N. Human structure interaction: influence of walking pedestrians
on the dynamic properties of footbridge structures they occupy. MSc thesis,
Department of Civil Engineering, Technical University of Denmark, August
2009.
[74] Charles P, Bui V. Transversal dynamic actions of pedestrians. Synchronization.
In: Proceedings of Footbridge 2005, second international conference, Venice,
2005.
[75] Ove Arup and Partners International Ltd. London Borough of Southwalk
Millennium Bridge. Prototype test report 2: Damping test results,
(Unpublished internal report), February 2002.
[76] Pavic A, Armitage T, Reynolds P, Wright J. Methodology for modal testing of
the Millennium Bridge, London. Proc ICE: Struct Build 2002;152(2):111–22.
[77] Pavic A, Willford M, Reynolds P, Wright J. Key results of modal testing of the
Millennium Bridge, London. In: Proceedings of Footbridge 2002, first
international conference, Paris, 2002.
[78] Dallard P, Fitzpatrick T, Flint A, Low A, Smith RR, Willford M. Pedestrian-
induced vibration of footbridges. Struct Eng 2000;78(23/24):13–5.
[79] Hobbs RE. Test on lateral forces induced by pedestrians crossing a platform
driven laterally. Tech. rep., Civil Engineering Department, Imperial College
London, London, UK, a report for Ove Arup and Partners, August 2000.
[80] Willford M, Dynamic actions and reactions of pedestrians. In: Proceedings of
Footbridge 2002, first international conference, Paris, 2002.
[81] Newland DE. Pedestrian excitation of bridges – recent results. In: Proceedings
of the tenth international congress on sound and vibration; 2003. p. 533–47.
[82] Newland DE. Vibration of the London Millennium Bridge: cause and cure. Int J
Acoust Vib 2003;8(1):9–14.
[83] Newland DE. Pedestrian excitation of bridges. Proc Inst Mech Eng, Part C: J
Mech Eng Sci 2004;218:477–92.
[84] McRobie A, Morgenthal G. Risk management for pedestrian-induced dynamic
of footbridges. In: Proceedings of Footbridge 2002, first international
conference, Paris, 2002.
[85] Kay BA, Warren Jr WH. Coupling of posture and gait: mode locking and
parametric excitation. Biol Cybernet 2001;85:89–106.
[86] McAndrew PM, Dingwell JB, Wilken JM. Walking variability during
continuous pseudo-random oscillations of the support surface and visual
field. J Biomech 2010;43:1470–5. http://dx.doi.org/10.1016/j.jbiomech.
2010.02.003.
[87] Brady RA, Peters BT, Bloomberg JJ. Strategies of healthy adults walking on a
laterally oscillating treadmill. Gait Posture 2009;29:645–9. http://dx.doi.org/
10.1016/j.gaitpost.2009.01.010.
[88] Yoshida J, Abe M, Fujino Y, Higashiuwatoko K. Image analysis of human
induced lateral vibration of a pedestrian bridge. In: Proceedings of Footbridge
2002, first international conference, 2002.
[89] Yoshida J, Fujino Y, Sugiyama T. Image processing for capturing motions of
crowd and its application to pedestrian-induced lateral vibration of a
footbridge. Shock Vib 2007;14:251–60.
[90] McRobie A, Morgenthal G, Lasenby J, Ringer M. Section model tests on
human-structure lock-in. Proc ICE: Bridge Eng 2003;156(BE2):71–9.
[91] Rönnquist A. Pedestrian induced vibrations of slender footbridges. PhD thesis,
Norwegian University of Science and Technology, 2005.
[92] Rönnquist A, Strømmen E. Pedestrian induced lateral vibrations of slender
footbridges. In: Proceedings of the 25th IMAC Conference, Orlando, USA, 2007.
[93] Nakamura SI, Kawasaki T, Katsuura H, Yokoyama K. Experimental studies on
lateral forces induced by pedestrians. J Construct Steel Res 2008;64:247–52.
http://dx.doi.org/10.1016/j.jcsr.2007.05.011.
[94] Ingólfsson ET, Georgakis CT, Ricciardelli F, Procino L. Lateral human-structure
interaction on footbridges. In: Tenth international conference on recent
advances in structural dynamics, Southampton, 2010.
[95] Pavic A, Reynolds P. Modal testing of a 34 m catenary footbridge. Proc SPIE –
Int Soc Opt Eng 2002;4753(II):1113–8.
[96] Živanović S, Pavic A, Reynolds P. Modal testing and FE model tuning of a lively
footbridge structure. Eng Struct 2006;28(6):857–68. http://dx.doi.org/
10.1016/j.engstruct.2005.10.012.
[97] Cunha A, Caetano E, Moutinho C, Magalhães F. The role of dynamic testing in
design, construction and long-term monitoring of lively footbridges. In:
Proceedings of Footbridge 2008, third international conference, Porto, 2008.
[98] EN 1990:2002/A1:2005. Eurocode – Basis of structural design. European
Committee for Standardization, Brussels, amendment A1, December 2005.
[99] Brownjohn JMW, 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.
[100] Brownjohn JMW, Pavic A, Omenzetter P. Modeling and measuring dynamic
crowd loading on a long span footbridge. In: Proceedings of the 2004
international conference on noise and vibration engineering, ISMA, 2004. p.
751–65.
[101] Brownjohn JMW, 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.
[102] Brownjohn J, Zivanovic S, Pavic A. Crowd dynamic loading on footbridges. In:
Proceedings of Footbridge 2008, third international conference, Porto, 2008.
[103] Nakamura SI. Field measurements of lateral vibration on a pedestrian bridge.
Struct Eng 2003;81(22):22–6.
[104] Rönnquist A, Strømmen E, Wollebæk L. Dynamic properties from full scale
recordings and FE-modelling of a slender footbridge with flexible
connections. Struct Eng Int 2008;4:421–6.
[105] Rönnquist A, Wollebæk L, Bell K. Dynamic behaviour and analysis of a slender
timber footbridge. In: 9th World Conference on Timber Engineering, 2006.
[106] Da Fonseca A, Balmond C, Conceptual design of the new Coimbra fotbridge.
In: Proceedings of Footbridge 2005, second international conference, Venice,
2005.
[107] Caetano E, Cunha A, Moutinho C, Magalhães F. Lessons from the practical
implementation of a passive control system at the new Coimbra footbridge.
In: Proceedings of Footbridge 2008, third international conference, Porto,
2008.
[108] Caetano E, Cunha A, Moutinho C. Implementation of passive devices for
vibration control at Coimbra footbridge. In: Experimental vibration analysis
for civil engineering structures; 2007. p. 43–54.
[109] Magalhães F, Cunha A, Caetano E. Comparison of damping estimates from
ambient and free vibration tests in large structures. In: Experimental
vibration analysis for civil engineering structures; 2007. p. 307–16.
[110] Magalhães F, Cunha A, Caetano E. Dynamic testing of the new Coimbra
footbridge before implementation of control devices. In: Proceedings of the
25th IMAC Conference, Orlando, Florida USA, 2007.
[111] Feichtinger D. Bridge design. In: Proceedings of Footbridge 2008, third
international conference, Porto, 2008.
[112] Macdonald JHG. Pedestrian-induced vibrations of the Clifton Suspension
Bridge, UK. Proc ICE: Bridge Eng 2008;161(BE2):69–77. http://dx.doi.org/
10.1680/bren.2008.161.2.69.
[113] Strobl W, Kovaces I, Andrä HP, Häberle, Eine Fußgängerbrücke mit einer
Spannweite von 230 m (a footbridge with a span of 230 m), Stahlbau 76 (12)
(2007) 869–879, (in German).
[114] Mistler M, Heiland D. Lock-in-Effekt beings Brücken infolge
FuSSgängeranregung - Schwingungstest der weltlängsten FuSSgänger- und
Velobrücke (lock-in effect due to pedestrian excitation of bridges – vibration
test of the world’s longest pedestrian and bicycle bridge). In: D-A-CH Tagung,
Vienna, 2007 [in German].
[115] Franck L. Synchronous lateral excitation of footbridges. Semester project
report. Applied Computing and Mechanics Laboratory, Swiss Federal Insitute
of Technology, 2009. <http://imacwww.epfl.ch/Team/Lestuzzi/Semester_
Project/SA08_Footbridges_Franck.pdf>.
[116] Butz C, Dist J, Huber P. Effectiveness of horizontal tuned mass dampers
exemplified at the footbridge in Coimbra. In: Proceedings of Footbridge 2008,
third international conference, Porto, 2008.
 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
[117] Finnis KK, Walton D. Field observations of factors influencing walking speeds.
In: 2nd International conference on sustainability engineering and science,
2007.
[118] Papadimitriou E, Yannis G, Golias J. A critical assessment of pedestrian
behaviour models. Transp Res Part F 2009;12(3):242–55. http://dx.doi.org/
10.1016/j.trf2008.12.004.
[119] Hughes RL. The flow of human crowds. Annu Rev Fluid Mech 2003;35:169–82.
http://dx.doi.org/10.1146/annurev.fluid.35.101101.161136.
[120] Lee RSC, Hughes RL. Minimisation of the risk of trampling in a crowd. Math
Comput Simulat 2007;74:29–37. http://dx.doi.org/10.1016/j.matcom.
2006.06.029.
[121] Kuang H, Li X, Song T, Dai S. Analysis of pedestrian dynamics in counter flow
via an extended lattice gas model. Phys Rev E 78 (6). http://dx.doi.org/
10.1103/PhysRevE.78.066117.
[122] Hughes RL. A continuum theory for the flow of pedestrians. Transp Res, Part B
– Methodological 2002;36B(6):507–35.
[123] Wong L, Huang SC, Zhang M, Shu CW, Lam WHK. Revisiting Hughes’ dynamic
continuum model for pedestrian flow and the development of an efficient
solution algorithm. Transp Res Part B 2009;43:127–41. http://dx.doi.org/
10.1016/j.trb.2008.06.003.
[124] Venuti F, Bruno L, Bellomo N. Crowd dynamics on a moving platform:
mathematical modelling and application to lively footbridges. Math Comput
Model 2007;45:252–69. http://dx.doi.org/10.1016/j.mcm.2006.04.007.
[125] Wheeler JE. Prediction and control of pedestrian-induced vibration in
footbridges. ASCE J Struct Div 1982;108:2045–65.
[126] Pachi A, Ji T. Frequency and velocity of people walking. Struct Eng
2005;83(3):36–40.
[127] Živanović S, Racić V, El-Bahnsay, Pavić A. Statistical characterisation of
parameters defining human walking as observed on an indoor passerelle. In:
Experimental vibration analysis for civil engineering structures; 2007. p.
219–25.
[128] Ricciardelli F, Briatico C, Ingolfsson ET, Georgakis CT. Experimental validation
and calibration of pedestrian loading models for footbridges. In: Proceedings of
experimental vibration analysis for civil engineering structures, Porto, 2007.
[129] Ingólfsson ET, Georgakis CT, Jönsson J, Ricciardelli F. Vertical footbridge
vibrations: towards an improved and codifiable response evaluation. In:
Third international conference on structural engineering, mechanics and
computation, Cape Town, South Africa, 2007.
[130] Pansera A. Analisi sperimentale delle caratteristiche del cammino ed azione
dei pedoni sulle passerelle pedonali (experimental analysis of gait
characteristics and walking-induced actions on footbridges). BSc thesis,
University of Reggio Calabria, 2006 [in Italian].
[131] Terrier P, Turner V, Schutz Y. GPS analysis of human locomotion: Further
evidence for long-range correlations in stride-to-stride fluctuations of gait
parameters. Hum Movement Sci 2005;24(1):97–115. http://dx.doi.org/10.
1016/j.humov.2005.03.002.
[132] Ingólfsson ET. Pedestrian-induced vibrations of line-like structures. MSc
thesis, Department of Civil Engineering, Technical University of Denmark,
June 2006.
[133] Morgenroth O. Walking speed and pace of life. Report ’städtestudie’, TU
Chemnitz, Institute for Psychology; 2003 [in German].
[134] Matsumoto Y, Nishioka T, Shiojiri H, Matsuzaki K. Dynamic design of
footbridges. In: IABSE proceedings, vol. P-17/78; 1978. p. 1–15.
[135] Živanović S, Pavic A, Reynolds P, Vujović P. Dynamic analysis of lively
footbridge under everyday pedestrian traffic. In: Proceedings of the sixth
European conference on structural dynamics, 2005.
[136] Venuti F, Bruno L. An interpretative model of the pedestrian fundamental
relation. CR Mecanique 2007;335:194–200. http://dx.doi.org/10.1016/
j.crme.2007.03.008.
[137] Oeding D. Verrkhersbelastung und Dimensionierung von Gehwegen und
anderen Anlagen des Fußgangerverkhers (traffic load and design of sidewalks
and other pedestrian facilities). Tech. rep. 22, Strassenbau und
Strassenverkherstechnik, 1963 [in German].
[138] Wheeler JE. Crowd loading on footbridges. Tech. rep. no. 23, Main Roads
Department, Western Australia, 1981.
[139] Bertram JEA, Ruina A. Multiple walking speed–frequency relations are
predicted by constrained optimization. J Theor Biol 2001;209(4):445–53.
http://dx.doi.org/10.1006/jtbi.2001.2279.
[140] Bruno L, Venuti F. The pedestrian speed–density relation: modelling and
application. In: Proceedings of Footbridge 2008, third international
conference, Porto, 2008.
[141] Butz C, Magalhães F, Cunha A, Caetano E, Goldack A. Experimental
characterization of the dynamic behaviour of lively footbridges. In:
Proceedings of Footbridge 2005, second international conference, Venice, 2005.
[142] De Donno A, Powell D, Low A, Design of damping systems for footbridges.
Conceptual framework. In: Proceedings of Footbridge 2005, second
international conference, Venice, 2005.
[143] Zoltowski K. Pedestrian on footbridges, vertical loads and response. In:
Proceedings Footbridge 2008, third international conference, Porto, 2008.
[144] Andersen RK. Pedestrian-induced vibrations: human–human interaction.
MSc thesis, Department of Civil Engineering, Technical University of
Denmark, 2009.
[145] Ricciardelli F, Pansera A. An experimental investigation into the interaction
among walkers in groups and crowds. In: Tenth international conference on
recent advances in structural dynamics, Southampton, 2010.
51
[146] Ishaque MM, Noland RB. Behavioural issues in pedestrian speed choice and
street crossing behaviour: a review. Transport Rev 2008;28:61–85. http://
dx.doi.org/10.1080/01441640701365239.
[147] Roberts TM. Lateral pedestrian excitation of footbridges. J Bridge Eng
2005;10(1):107–12.
[148] Roberts TM, Synchronised pedestrian lateral excitation of footbridges. In:
Proceedings of the 6th international conference on structural dynamics, Paris,
2005.
[149] Roberts TM. Probabilistic pedestrian lateral excitation of bridges. In:
Proceedings of the institute of civil engineers: bridge engineering, vol. 158;
2005. p. 53–61.
[150] Ricciardelli F. Lateral loading of footbridges by walkers. In: Proceedings of
Footbridge 2005, second international conference, Venice, 2005.
[151] Sólnes J. Stochastic processes and random vibrations. Theory and
practice. Chichester, England: John Wiley & Sons; 1997.
[152] Butz C. Codes of practice for lively footbridges: State-of-the-art and required
measures. In: Proceedings of Footbridge 2008, third international conference,
Porto, 2008.
[153] Butz C. A probabilistic engineering load model for pedestrian streams. In:
Proceedings of Footbridge 2008, third international conference, Porto, 2008.
[154] Nakamura SI. Model for lateral excitation of footbridges by synchronous
walking. J Struct Eng 2004;130(1):32–7.
[155] Ingólfsson ET. Pedestrian-induced lateral vibrations of footbridges.
Experimental studies and probabilistic modelling. PhD Thesis, Department
of Civil Engineering, Technical University of Denmark, 2011.
[156] Gimsing N, Georgakis C. Cable supported bridges: concepts and design. 3rd
ed. London: John Wiley & Sons; 2011.
[157] Ricciardelli F, Mafrici M, Ingólfsson ET. Motion-dependent lateral walking
forces on footbridges. In: Proceedings of Footbridge 2011, fourth
international conference, Wroclaw, 2011.
[158] Blekherman A. Autoparametric resonance in a pedestrian steel arch bridge:
Solferino Bridge, Paris. J Bridge Eng 2007;12:669–76. http://dx.doi.org/
10.1061/(ASCE)1084-0702(2007)12:6(669).
[159] Fujino Y, Warnitchai P, Pacheco BM. An experimental and analytical study of
autoparametric resonance in a 3DOF model of cable-stayed-beam. Nonlinear
Dyn 1993;4:111–38.
[160] Piccardo G, Tubino F. Parametric resonance of flexible footbridges under
crowd-induced lateral excitation. J Sound Vib 2008;311:353–71. http://
dx.doi.org/10.1016/j.jsv.2007.09.008.
[161] Thomsen J. Vibration and stability. In: Advanced theory, analysis and
tools. Berlin, Heidelberg, Germany: Springer; 2003.
[162] Tubino F, Piccardo G. A loading model for the interpretation of lateral
vibrations of flexible footbridges. In: Proceedings of Footbridge 2011, fourth
international conference, Wroclaw, 2011.
[163] Strogatz SH, Stewart I. Coupled oscillators and biological synchronization. Sci
Am 1993;269:68–73.
[164] Strogatz SH, Abrams DM, McRobie A, Eckhardt B, Ott E. Crowd synchrony on
the Millennium Bridge. Nature 2005;438(7064):43–4.
[165] Abrams DM. Two coupled oscillator models: the Millennium Bridge and the
Chimera State. Phd thesis, Cornell University (August 2006).
[166] Eckhardt B, Ott E. Crowd synchrony on the London Millennium Bridge. Chaos
16. http://dx.doi.org/10.1063/1.2390554.
[167] Eckhardt B, Ott E, Strogatz SH, Abrams DM, McRobie A. Modeling walker
synchronization on the Millennium Bridge. Phys Rev E 2007;75:21110-
1–21110-10. http://dx.doi.org/10.1103/PhysRevE. 75.021110.
[168] Abdulrehem MM, Ott E. Low dimensional description of pedestrian-induced
oscillation of the Millennium Bridge. Chaos 19. http://dx.doi.org/10.1063/
1.3087434.
[169] Barker C. Some observations on the nature of the mechanism that drives the
self-excited lateral response of footbridges. In: Proceedings of Footbridge
2002, first international conference, Paris, 2002.
[170] Macdonald JHG. Lateral excitation of bridges by balancing pedestrians. Proc
Roy Soc A 2009;465:1055–73. http://dx.doi.org/10.1098/rspa.2008.0367.
[171] Johnson R. A mathematical modelling of the pedestrian induced lateral
vibrations of the Millennium Bridge. MSc thesis, University of Bristol, 2005.
[172] Hof AL, van Bockel RM, Schoppen T, Postema K. Control of lateral balance in
walking: experimental findings in normal subjects and above-knee amputees.
Gait Posture 2007;25(2):250–8. http://dx.doi.org/10.1016/j.gaitpost.2006.
04.013.
[173] Carroll SP, Owen JS, Hussein MFM. Crowd-bridge interaction by combining
biomechanical and discrete element models. In: Proceedings of the 8th
international conference on structural dynamics, Leuven, 2011.
[174] Erlicher S, Trovato A, Argoul P. Modeling the lateral pedestrian force on a
rigid floor by a self-sustained oscillator. Mech Syst Sig Process 2010;24(5):
1579–604. http://dx.doi.org/10.1016/j.ymssp.2009.11.006.
[175] Trovato A, Erlicher S, Argoul P. Modeling the lateral pedestrian force on rigid
and moving floors by a self-sustained oscillator. In: ECCOMAS thematic
conference on computational methods in structural dynamics and
earthquake engineering, Rhodes, Greece, 2009.
[176] Paulissen JH, Metrikine AV, Non-linear modeling of adaptive pedestrian
behavior on lively footbridges. In: Proceedings of the 8th international
conference on structural dynamics, Leuven, 2011.
[177] Venuti F, Bruno L, Bellomo N. Crowd-structure interaction: dynamics
modelling and computational simulations. In: Proceedings of Footbridge
2005, second international conference, Venice, 2005.
52 E.T. Ingólfsson et al. / Engineering Structures 45 (2012) 21–52
[178] Bruno L, Venuti F. Crowd-structure interaction in footbridges: Modelling,
application to a real case-study and sensitivity analyses. J Sound Vib
2009;323(1–2):475–93. http://dx.doi.org/10.1016/j-jsv.2008.12.015.
[179] Venuti F, Bruno L. A new load model of the pedestrians lateral action. In:
Proceedings of Footbridge 2008, third international conference, Porto,
2008.
[180] Venuti F, Bruno L. Synchronous lateral excitation on lively footbridges:
modelling and application to the T-Bridge in Japan. In: Proceedings of
Footbridge 2008, third international conference, Porto, 2008.
[181] Bodgi J, Erlicher S, Argoul P. Lateral vibration of footbridges under crowd-
loading: continuous crowd modelling approach. Key Eng Mater
2007;347:685–90.
[182] Bodgi J, Erlicher S, Argoul P. Lateral vibration of footbridges under crowd-
loading: on the modelling of the crowd-synchronization effects. In:
Experimental vibration analysis for civil engineering structures, Porto,
2007. p. 237–45.
[183] Bodgi J, Erlicher S, Argoul P, Flamand O, Danbon F. Crowd structure
synchronization: coupling between Eulerian flow modeling and Kuramoto
phase equation. In: Proceedings of Footbridge 2008, third international
conference, Porto, 2008.
[184] Bodgi J. Synchronization piétons-structure: application aux vibration des
passerelles souples (pedestrian-structure synchronization: Application to
vibration of flexible bridges), PhD thesis, Ecole Nationale des Ponts -
ParisTech (September 2008) [in French].