Experimental Vibration Analysis for Civil Engineering Structures

237

Lateral vibration of footbridges under crowd-loading: on the

modelling of the crowd-synchronization effects
J. Bodgi, S. Erlicher & P. Argoul
Institut Navier (LAMI-ENPC), Marne-la-Valle, France

ABSTRACT: In this paper, a 1D crowd model is proposed, aiming to properly describe the
crowd-flow phenomena occurring when pedestrians walk on a flexible footbridge. The crowd is
assumed to behave like a compressible fluid and the pedestrian flow is modeled in a 1-D
framework using the mass (of pedestrians) conservation equation. This crowd model is coupled
with a simplified model for the dynamics of the footbridge and an optimized modeling of synchronization effects through Kuramoto equation. Numerical simulations are presented to show
some preliminary results.
1 INTRODUCTION
Recently designed footbridges have shown to be sensitive to the human induced vibration (Millenium Bridge, London; Solfrino Bridge, Paris). Several experimental measurements allowed
this phenomenon to be better understood (Dallard et al. 2001; SETRA 2006). The crowd walking on a footbridge imposes to the structure a dynamic lateral excitation at a frequency close to
1 Hz. When the first mode of lateral vibration of a footbridge falls in the same frequency interval, a resonance phenomenon is activated, the oscillation amplitude increases and, if it becomes
large enough, pedestrians are forced to change their way of walking, up to the so-called structure-pedestrian synchronization. This phenomenon has been often experimentally detected and
also analyzed in several studies (Dallard et al. 2001; Venuti et al. 2005; ivanovic et al. 2005).
A series of simplified design rules for footbridges accounting for these effects was recently proposed (SETRA 2006).
Moreover, the behaviour of each single pedestrian is affected by the presence of the crowd
around him. In detail, when the pedestrian density is very low, the walk is free and characterized by the speed, the walk frequency, the step length, etc. slightly varying from a walker to
the other. Nonetheless, when the crowd density becomes higher, a single walker is forced to
synchronize his speed with that of the others. This kind of pedestrian-pedestrian synchronization
(or traffic effect) occurs when the walk is on a rigid floor and some crowd models were already developed (Hoogendorn 2004) for this case. However, very few existing studies concern
the simultaneous modeling of the pedestrian-pedestrian and the structurepedestrian synchronizations (Venuti et al. 2005; Bodgi et al. 2007). This coupled analysis is discussed in the present
contribution.
2 SOME EXPERIMENTAL RESULTS
2.1 Single pedestrians walking
Every step of the human walking is characterized by a time interval where both feet are in contact with the floor and an interval where only one foot touches the floor. One may define the beginning of a step as the beginning of a simultaneous contact period. The end of a step coincides

238

EVACES07

with the beginning of the following simultaneous contact period. For a given pedestrian, walking at constant speed, the time-length Ts of a step is approximately constant. The walking frequency is defined by: f s = 1 Ts . For a standard walking, one gets f s 2 Hz ( ivanovic et al.
2005). The force induced by a single pedestrian on the floor has a lateral component related to
the small lateral oscillations of the centre of gravity of the pedestrian during walk. It acts in the
direction perpendicular to the walk speed, and with opposite signs for each foot. Hence, the frequency of the lateral force is f lat = f s 2 1Hz . The processing of data provided by Decathlon
(Bodgi et al. 2006) concerning lateral walking forces show a quasi-periodic behavior, with a
typical right-left cycle like the one indicated in Figure 1. The periodicity of the lateral force
f l ( t ) suggests a Fourier harmonic decomposition and for the analysis of the crowd effect on
the lateral motion of footbridges, only the first harmonic (frequency flat) is retained in most studies; this will be also the case in this paper:
+

fl (t ) = i sin(2 i f lat t i ) 1 sin(2 f lat t 1 )

(1)

i =1

where i is the i-th Fourier coefficient, i is the i-th phase difference. f l ( t ) is always
bounded and less than 50 N . For an increasing walking speed, the frequency flat also increases.

Figure 1.Lateral walking force on a rigid floor, for several walking speeds.

The analysis of these experimental data shows that the first harmonic amplitude 1 varies between 30 and 35 N, confirming the results of SETRA (2006) and it can be related to the walking
velocity v by a linear relationship: 1 = 0.6191 v + 35.5171 [N]. This flat frequency is often
very close to the lateral modal frequency of footbridges. Hence, footbridges may be subject to
large oscillations because their behavior is often slightly damped even if the pedestrian lateral
force is relatively small. In the case of non-rigid floor, the lateral force amplitude of the human
walk is assumed to be similar to that of the rigid-floor case. This strong assumption is supported
by the experimental results of SETRA (2006). Conversely, the walking velocity and frequency
are affected by the structure vibrations, as seen in the next Subsection.
2.2 Influence of the structural oscillation on pedestrians walking.
When the amplitude of the lateral vibrations acceleration for the footbridge is larger than a certain threshold: a min = 0.1 m / s 2 , the lateral vibrations become perceptible for pedestrians, who
tend to change their walking frequency to synchronize their walk with the structures oscillations (ivanovic et al. 2005). If a certain number of pedestrians is synchronized with the structure (and therefore, with each other), the total lateral force they produce further increases the
structures oscillations, inducing other pedestrians to synchronize their walk. If the lateral velocity of the footbridge floor reaches u& max = 0.25m / s , pedestrians stop walking, otherwise they
would loose equilibrium (Venuti et al. 2005).

239

Experimental Vibration Analysis for Civil Engineering Structures

2.3 Interactions between pedestrians.

For low crowd densities c = 0.3 p / m 2 , every single pedestrian walks freely, with an average speed v M = 1.5 m s , slightly varying from one walker to the other. When the crowd density is higher than c , the walking velocity decreases in order to avoid collisions between pedestrians. For very high crowd densities ( M = 1,6 1,8 pedestrians m 2 ), pedestrians
stop walking (Venuti et al. 2005).
3 SIMPLIFIED FOOTBRIDGE MODELING
The lateral motion of a footbridge can be approximately represented by a linear Euler-Bernoulli
beam equation with viscous damping (SETRA 2006):

ms ( x)

2u
u
4u
2u
(
,
)
(
)
(
,
)
(
)
(
,
)
(
,
)
(
,
)
( x, t )
x
t
+
c
x
x
t
+
k
x
x
t
=
F
x
t

m
x
t
l
p
t 2
t
x 4
t 2

(2)

where x is the coordinate along the beam axis; t the time; u(x,t) the lateral beam displacement;
ms(x) the mass per unit length of the beam [kg/m]; c(x) the viscous damping coefficient [N.s
/m2]; k(x) the stiffness per unit length [ Nm 2 ] ; Fl(x,t) the pedestrian lateral force per unit
length [ N / m ] and m p ( x ,t ) the linear mass of pedestrians [ kg / m ] .
Under the assumption of doubly hinged beam of length L, the first lateral mode shape can be
approximated by a sinus having the half-period equal to L: this assumption is not approximated
when ms, mp, c and k are constant along the beam axis. Since this mode plays a major role in the
lateral
footbridge
dynamics,
the
solution
is
assumed
of
the
form
u ( x, t ) = U (t ) 1 ( x) = U (t ) sin ( x / L ) , i.e. the contribution of the other modes is neglected,
and Eq. (2) becomes
*
m* ( t )U&&( t ) + c*U& ( t ) + k *U ( t ) = Fl ( t )

(3)

L
d 2 1
*
where k = k (x)
(x)dx , c = c(x) 12 (x)dx , Fl* ( t ) = Fl ( x ,t ) 1 ( x )dx and
2
dx
0
0
L

m* (t ) = ms* (t ) + m p* (t ) = [ms ( x) + m p ( x, t ) ] 12 dx
0

One can observe that the mass m* ( t ) is the sum of the structural mass contribution plus the total mass of pedestrians walking on the footbridge deck and having the instantaneous distribution
given by m p ( x ,t ) . Hence, the instantaneous modal frequency and damping ratio of the system
constituted by the footbridge and pedestrians read:

1
f ps (t ) =
2

2 L m p ( x, t ) 2 x
k
sin
= f s 1 +
dx
m (t )
L
L 0 ms

, (t ) =

c*
2 m* k *

(4)

The mass m p ( x ,t ) and the force Fl ( x ,t ) must be defined to solve Eq. (3). Their definition is
related to the approach used for modeling the crowd. Two different approaches are under study:
(i) an Eulerian approach consisting in a macroscopic modeling of the crowd which is considered as a whole and (ii) a Lagrangian approach consisting in modeling each single pedestrian
of the crowd as a moving particle interacting with the others. Under after, only the first approach is presented.

240

EVACES07

4 AN EULERIAN CROWD MODEL (ECM)

An Eulerian crowd model (ECM) postulates that the crowd behaves like a compressible fluid
(Venuti et al. 2005). This kind of analysis is intended to represent the pedestrian behavior for
high crowd densities. The crowd motion is characterized by it local density ( x ,t ) and its local
speed v( x ,t ) . In general three equations govern the motion of a fluid, i.e. the mass conservation

+ ( .v ) = 0
t x

(5)

the dynamic equilibrium and a constitutive law. However, for traffic flow modelling, it is usual
to substitute these last two equations with a simpler closure equation, relying on and v . For
the pedestrian flow, the following closure equation is adopted (Venuti et al. 2005):

v( ,u& ) = g( ) h( u& )

(6)

where

c
u&
vM

1 &

exp(
(
)
(
)
)

1
umax
&
g( ) =
, h( u ) =
c
M
c
vM ( 1
) > c

0
1 exp( )

u& < u&max

u& u&max

The crowd velocity v( , u& ) depends on the v() function, representing the interaction between
pedestrians described above. ( x, t ) is a parameter describing the traffic conditions. The mechanical analogy suggests for the role of a viscosity parameter of the equivalent crowd fluid.
A larger value indicates more difficult traffic conditions (highly viscous crowd fluid). Figure
2 presents the crowd speed on a rigid floor ( h ( u& ) = 1 ) for different values. Experimental data
lead to the values of being between 0 and 10 (Venuti et al. 2005).

Figure 2- Walking speed vs. crowd density for different -values

The second function h( u& ) 1 allows accounting for the influence of the lateral footbridge vibration on the crowd speed, in the sense that it imposes a speed reduction when pedestrians
walk on a footbridge undergoing large oscillations. In addition, it is assumed that after a stop
occurring when u& u&max , the pedestrian speed remains zero during five seconds. After this delay, they begin to walk if u& < u& max , otherwise, the stop lasts five more seconds. Observe that
Eq. (6) governs the crowd speed but nothing is said about the pedestrians-structure synchronization phenomenon which also involves the phases of pedestrians and of the structure.
Once the density (x,t) is known, the linear mass density follows
m p ( x, t ) = m1 p ( x, t ) l( x) , where l( x) is the deck width and m1p is the mass of a single pe-

destrian. The total force is defined as the product of the force of a single pedestrian and an
equivalent synchronized pedestrian number:

241

Experimental Vibration Analysis for Civil Engineering Structures

Fl ( x, t ) = f%l (t ) n p ,eq ( x, t )

(7)

f%l (t ) = 1 sin ( 2 f ps ( t ) t ) is deduced from Equation (1) for flat=fps and

n p ,eq ( x, t ) = ( x, t )l ( x ) S (u& , ) is the equivalent number of pedestrians defined to give a
fictitious total force Fl ( x, t ) acting at the modal frequency and phase of the structure whose ef-

where

fects (structural motion) are the same as those of the true total force, deriving from the true nonsynchronized pedestrians (see SETRA (2006)).
The coefficient S ( u& , ) [ 0,1] , introduced to represent the synchronization effects, is the
ratio between the number of synchronized pedestrians per unit length and the number of pedestrians per unit length. According to the assumptions of Venuti et al. (2005), the special case S=1
occurs when all the pedestrians walk with the frequency and phase of the structural velocity U& .
This case is the most severe for the structure, in the sense that the same number of pedestrians
always produces a less important structural motion when they are not in this situation of full
synchronization. According to Venuti et al. (2005), one has

S ( x ,t ) = Spp( ( x ,t )) + Sps( u&( x ,t ))

(8)

The first term accounts for the fact that for high densities pedestrians are synchronized each
other: for c , the walk is free, every pedestrian has a different frequency and/or phase and
the contribution to the total force is therefore null. The second term derives from the observation
that for large oscillations, pedestrians have a stronger tendency to be synchronized with the
structure. The function S ps ( u& ) has two branches (Figure 3-b): the first one is a quadratic approximation of the ARUP data (Dallard et al. 2001). It defines an increase of the synchronization when the structural oscillations are larger. However, after reaching a maximum for particular amplitude of structural oscillations, the synchronization (and the equivalent pedestrian
number) is assumed then to decrease, as represented by the decreasing branch.

a)

/M

Figure3- a) Coefficient Spp( ) , b) Coefficient Sps( u& )

u& b)

The previous definition of the synchronization coefficient (Equation 8) have some drawbacks: (i) the sum of two contributions is rather artificial. It is difficult to distinguish the effects of the two synchronizations in the case of a vibrating floor: the one between pedestrians
and the other with the vibrations of the structure ; the synchronization between pedestrians being induced by the synchronization with the decks vibrations (ii) the Spp coefficient is arbitrary
and not based on experimental data. Hence, a new definition of the synchronization coefficient
based on experimental data is proposed and accounts for the proposal of SETRA (2006), stating
that the lateral force induced by a crowd of N pedestrians on a rigid floor can be approximated
by the force induced by Neq pedestrians walking in place and having the same phase and frequency. The new synchronization coefficient reads:

Sp( ( x ,t )) if u&&( x ,t ) a min = 0.1m / s 2

S ( x ,t ) =
otherwise
Sps( u&( x ,t ))

(9)

242

EVACES07

where, recalling that is the instantaneous modal damping ratio, the synchronization coefficient
in the case of low-amplitude of the floor vibration is defined as follows

8.6 ( ( x ,t )l( x ) L ) if ( x ,t ) c
Sp( ( x ,t )) = N eq ( x ,t ) / N =
otherwise
1.75 ( ( x ,t )l( x ) L )

(10)

NUMERICAL EXAMPLE : THE MILLENNIUM BRIDGE

In this Section, the model with the improved definition of the Synchronization coefficient (Eq.
(equation 9) is implemented. Equation (3) is solved by a Runge-Kutta scheme of order 4-5, and
Eq. (5) is discretized by a finite difference scheme (Lax-Friedrichs), fulfilling the Lax Friedrichs
conditions for the ratio between the spatial step dx and the time step dt . The model is then
validated using the Millennium Footbridge main decks parameters. The pedestrian mass is assumed to be 75 kg, hence m p ( x ,t ) = 75 ( x ,t ) l . Moreover the following parameters are
adopted: ( x ,t ) = 0 , U ( 0 ) = U& ( 0 ) = 0 and ( x ,0 ) = 1.2 . Figures 4-a and 4-b represent

U& ( t ) and Fl ( t ) respectively. The initial high density of pedestrians induces a synchronisation
between pedestrians which leads to a lateral force of high amplitudes. As a consequence, the
structures lateral vibrations are more important. When the vibrations velocity is high
( u&( x ,t ) > u& max ), the pedestrians stop, hence the lateral force amplitude decreases and so do the
structural vibrations.

a)

b)

Figure 4 -a) Footbridge velocity U& ( t ) ; b) total lateral force Fl ( t ) induced by the crowd

5 AN EULERIAN CROWD MODEL COUPLED WITH THE KURAMOTO

SYNCHRONIZATION MODEL (ECMK)
In this Section, the synchronization effect is analyzed in a more general way: instead of the synchronization coefficient, the phase difference between the structure and pedestrians is introduced as variable governing synchronization and its evolution is modelled by the so-called
Kuramoto differential equation (Pikowski et al. 2001). This procedure is illustrated by the simple example of a self-sustained oscillator entrained by an external force having a known and
constant frequency f e . The oscillator tends to synchronize its total phase (t ) with that of the
first one e (t ) = 2f e t + . The Kuramoto equation reads:

& (t ) = 0 + sin(e (t ) (t ))

(11)

& ( t ) ( 2 ) is its instanwhere 0 is the initial pulsation of the self-sustained oscillator, while
taneous frequency. The solution of this differential equation (11) verifies the fact that
& ( t ) ( 2 ) tends to the instantaneous frequency f . The parameter affects the synchroniza
e
tion, i.e. the convergence of

& ( t ) ( 2 ) toward f . For example, lets assume

243

Experimental Vibration Analysis for Civil Engineering Structures

& ( t ) ( 2 ) can be seen in Figthat f e = 1.03 Hz , 0 = 2 , and = 0.01 . The variations of

ure 5a. In the case of a greater coefficient = 8 , the synchronization occurs (Fig. 5b). The
switch from the non-synchronized to the synchronized regime depends on the value and on the
difference between the frequencies 0/2 and fs (Pikowski et al. 2001).

& ( t ) ( 2 ) versus time

Figure 5 -
In the case of the pedestrian-structure synchronization, one can assume that each pedestrian
behaves like an oscillator moving on the footbridge (Lagrangian approach), while the external
force acting on it is produced by the footbridges lateral vibrations. As seen before, the synchronization phenomenon affects the pedestrians velocity v(t ) and the lateral force fl (t ) induced
by one pedestrian. These two quantities depend on the total phase of the pedestrian (t ) according to the following relationships:

fl (t ) = 1 sin( (t ) ),

v(t ) =

lstep (t )

& (t )

(12)

& ( t ) ( 2 ) is the instantaneous lateral force frequency;

& ( t ) is the walking frewhere
quency; l step ( t ) is the length of the pedestrians step: the average value of lstep ( t ) for a normal
walk on a rigid floor is l0 = 0.8 m .
In the Eulerian version of the model, the variables must be considered locally, i.e.
(t ) ( x, t ) , etc. and the equations introduced previously must be modified in order to fit
the model. The new Eulerian model proposed is composed of previous equations (3), (5) plus
two new equations replacing equations (6) and (7). The first one links the local speed v( x ,t ) of

the crowd to the total phase (t ) through the local step length la ( x, t )

v ( x ,t ) =

l a ( x , t ) d
( x ,t )

dt

(13)

where l a ( x ,t ) = l0 C s ( x ,t )C ( x ,t ) depends on C s ( x ,t ) and C ( x ,t ) , which reflect the influence of the footbridges vibrations and of the local pedestrians density, respectively:

C s ( x, t ) =
0

u ( x, t ) 1
1
t
vmax

2u
( x, t ) < amin
t 2
u
if
( x, t ) > vmax
x

if

otherwise

if ( x, t ) < c
1

C ( x, t ) =
if ( x, t ) > max
0

1
otherwise
( x, t )
max

244

EVACES07

The second equation defines the force per unit length applied on the footbridge :

Fl ( x, t ) = [1 sin( ( x, t )) ] ( x, t ) l ( x )

(14)

& ( t ) = 2f (t ) . Finally, the

and the footbridge instantaneous structural vibration pulsation is
s
ps
Eulerian Kuramoto equation writes

( x, t ) = ( x, t ) + v ( x, t ) ( x, t ) = ( x, t ) + u ( x, t ) sin s (t ) ( x, t ) + 2 (15)
dt
t
x
where ( x ,t ) = ( x ,0 ) =0 in view of preliminary numerical analyses; ( x, t ) is the local
free walking frequency.
6 SIMPLIFIED VERSION AND NUMERICAL EXAMPLES
A simplified version of the model previously proposed has been implemented, where f si is replaced by

f ps in Equation (15).

The pedestrians have the same initial frequency

( f lat ( x ,0 ) = 1 Hz ) and the same initial phase ( ( x ,0 ) = 0 ). The boundary conditions are
given by ( 0 ,t ) = ( x ,0 ) = 0 where 0 is a constant. In the case: 0 = 0.5 pedestrian / m 2 ,
several simulations for different values of have been made in order to study its influence on
the results.

Figure 6: =0.01 (a) Bridge displacement at mid-span; (b) Pedestrian density at different positions along
the bridge length; (c) Relevant frequencies vs. time

Figure 7: =5 (a) Bridge displacement at mid-span; (b) Pedestrian density at different positions along the
bridge length; (c) Relevant frequencies vs. time

Experimental Vibration Analysis for Civil Engineering Structures

245

The first case studied is for = 0.1 (Fig. 6). In this case, the synchronization is almost inexistent, and the solution is identical to the one obtained in the case of a beam under a sinusoidal external force of constant amplitude. The second case illustrated in Figure 7, for = 5 : the
synchronization does not occur, either. The third case for = 10 is shown in Figure 8, where
the synchronization of frequencies is evident.

Figure 8: =10 (a) Bridge displacement at mid-span; (b) Pedestrian density at different positions along the
bridge length; (c) Relevant frequencies vs. time

7 CONCLUSION
In this paper, the crowd flow has been described by a flow equation, where pedestrians are supposed to walk along straight trajectories parallel to the longitudinal dimension of the footbridge.
The non-stationary behaviour has been analyzed, as well as the synchronization phenomenon by
an Eulerian version of the Kuramoto equation. Preliminary results in the application of the
proposed modeling to real cases are presented. Work is in progress to optimize the evaluation of
the footbridge instantaneous frequency and to identify the synchronization parameter from the
experimental data of response of footbridges under pedestrian crowd (Solferino and BercyTolbiac footbridges in Paris).
REFERENCES
Bodgi, J., Erlicher, S. & Argoul, P. 2007. Lateral vibration of footbridges under crowd-loading : continuous crowd modeling approach. Key Engineering Materials 347: 685-690.
Bodgi, J., Erlicher, S. & Argoul, P. 2006. Vibrations et amortissements des passerelles pitonnes. JSI
2006, CDRom (in French).
Dallard, P., Fitzpatrick, A.J., Flint, A., Le Bourva, S., Low, A., Ridsdill Smith, R.M. & Willford, M.
2001. The London Millenium footbridge. The Structural Engineer 79(22): 17-32.
Hoogendoorn, S. 2004. Pedestrian flow modelling by adaptive control. Proc. TRB 2004 Annual Meeting,
CDRom, 12 pages.
Pikowski, A., Rosenblum, M., Kurths, J. 2001. Synchronization A universal concept in nonlinear sciences. Cambridge Univ. Press, Cambridge, UK.
SETRA 2006. Passerelles pitonnes - Evaluation du comportement vibratoire sous l'action des pitons.
Document technique Bagneux, 92 pages (in French).
Strogatz, S.H., Abrams, D.M., McRobie, A., Eckhardt, B., Ott, E. 2005. Theoretical mechanics: crowd
synchrony on the Millennium Bridge. Nature 438(7064):43-4.
Venuti, F., Bruno, L., Bellomo, N. 2005. Crowd-structure interaction: dynamic modelling and computational simulations. Proc. Footbridge 2005: 2nd Int. Conf., CDRom.
ivanovic, S., Pavic, A., Flint, A., Reynolds, P. 2005. Vibration serviceability of footbridges under human induced excitation: a literature review. Journal of Sound and Vibration 279: 1-74.

246

EVACES07