Vibrationserviceabilityofafootbridgeunderverticalpedestrianload

Stanaivanovi

ResearchAssociate
VibrationEngineeringSection
DepartmentofCivilandStructuralEngineering,TheUniversityofSheffield
SirFrederickMappinBuilding,MappinStreet,Sheffield,S13JD,UK

ABSTRACT

Vibrationserviceabilityoffootbridgesunderhumaninduceddynamicloadingattractedalotofattentionof
the research community in the last several years. Most often, crowd loading scenario and humanhuman
andhumanstructureinteractionareanalysed.However,betweenthecrowdloadingscenarioandthesingle
pedestrianthat mostoftenfeaturesinthe currentdesign guidelines,thereareatleasttwootherloading
scenariosthatcouldberelevantforfootbridgedesign:groupsofpeoplewalkingtogetherandthenormal
pedestrian traffic, which is usually related to the situation where the peoples movement is not spatially
restricted.Ontopofthisthevandalloadingshouldbeconsideredaswell.
Thispaperpresentsacombinednumericalandexperimentalstudyofaboxgirderfootbridgethatislivelyin
theverticaldirection.Anattemptismadetoanalysemostrelevantloadingscenariosforthisbridgeresiding
inthecitycentreofthecapitalofMontenegro.Theloadingscenariosarepresentedinformofariskmatrix.
Experimental data for some loading scenarios are collected and compared with numerical simulations
(basedeitheronthecurrentguidelinesoronsomeuncodifiedprocedures).Itcouldbeconcludedthatthe
existing design guidelines should be used with caution, and that the alternative methods of vibration
serviceabilityevaluationmightoftenbepreferable.
1

INTRODUCTION

Traditionally,thedesignguidelinesforvibrationserviceabilityoffootbridgesintheverticaldirectionrelyon
estimationofvibrationresponse to singlepersonwalking at apacingfrequencythat matchesthenatural
frequencyoftherelevantvibrationmode (BSI,1978; OHBDC,1983).Although this loadingscenarioisthe
worst case scenario for a single person since it induces the resonance, it is not necessarily plausible. For
example, the relevant mode could have the natural frequency outside the comfortable range for walking
frequencies(1.52.4Hz),andtheresonantscenariowouldproduceconservativeestimatesofthevibration
response.Ontheotherhandthereisanopinionthattheresonantsinglepersonloadingscenariopresentsa
simple means of predicting the response to multiperson traffic. This hypothesis needs verification and a
robustsinglepersonmodelthataccountsforfrequencydependenceoftheamplitudeofthewalkingforce
aswellasthereducedamplitudeforthehigherharmonics,whicharecurrentlynotaccountedforinBS5400
(BSI,1978)andOntariocode(OHBDC,1983).

Withsomehighprofileserviceabilityfailuresoffootbridgesinthelastdecade,suchasSolferinobridgein
Paris(Setra,2006)andMillenniumBridgeinLondon(Dallardetal.,2001),camerealisationthatattentionin
thedesignshouldbepaidtootherloadingscenarios,suchassparse,denseandverydensecrowdsfeaturing
inaFrenchguideline(Setra,2006).
Apart from having reliable load models for various loading scenarios, it is important that criteria for
evaluation of response of human (walking or standing) vibration receiver are clearly defined. This is still
subject of research, especially due to the fact that data about acceptability of vibrations perceived by
movingpeoplearesparse(ivanovietal,2005a).
Aconvenientwaytoapproachvibrationserviceabilityissueinfootbridgedesignistoconstructariskmatrix
thatcombinesinformationaboutthehazardassociatedwithaparticularloadingscenarioanditsprobability
ofoccurrence(McRobieandMorgenthal,2002).Thebridgeshouldbedesignedinsuchawaythatthemost
frequenteventscauseacceptablevibrationsformostpeople,whilethevibrationlevelcouldbeallowedto
behigherfortheeventsthatrarelyoccur.
Thispaperpresentscurrentlyavailablemeansofvibrationestimationforafootbridgeexposedtodifferent
loadingscenarios.Theevaluationofacceptabilityofthisresponsetofootbridgeusersisoutsidethescopeof
thisstudy.Thepaperstartswiththedescriptionofthefootbridgestructureanalysed,andpossibleloading
scenarios.Itthenproceedstoanalysingthesinglepersonloadingscenario,normalpedestriantraffic,groups
ofpeoplewalkingtogether,runningandvandalloading,andcrowdloading.Thisisfollowedbyconclusions.
2

STRUCTURALDESCRIPTIONANDEXPECTEDTRAFFICSCENARIOS

ThefootbridgeanalysedisasteelboxgirdershowninFigure1aresidinginthecitycentreofPodgorica,the
capital of Montenegro. Its length is 104m, with 78m between inclined columns. Employing FRFbased
modal testing (ivanovi et al., 2006) the first vertical (1V) vibration mode (Figure 1b), that is symmetric,
with natural frequency of 2.04Hz, modal mass of 58000kg and extremely low modal damping ratio of
0.22%wasidentifiedasthesourceofbridgeliveliness.Inafreedecaytestitwasfoundthatthedamping
increases to 0.26% for higher vibration magnitudes such as those generated by humanwalking. The
footbridgerespondstonormalpedestriantrafficdominantlyinmode1V(ivanovietal.,2005b).

Mode shape amplitude

1.2
1.0
0.8
0.6

(b)

Mode
shape

0.4
0.2
0.0
-0.2

(a)

fn=2.04Hz
=0.26%
m=58000kg

Half sine

78m
0

20

40
60
80
Bridge length [m]

100

120

Figure1:(a)Footbridgephotographand(b)modalpropertiesofthefundamentalmodeofvibration.
Thebridgeislocatedinthecitycentre,andthereforeisfrequentlyexposedtomultipersontrafficorafew
friends crossing together. Apart from this, during quiet daytime periods as well as during nighttimes, it
mightbeexposedtosinglepersonloading.Theseevents,therefore,couldbeconsideredasthosethatare
highlyprobableanditisexpectedthatfootbridgeperformwellintheseconditions.Additionally,itcouldbe
expectedthatthebridgeisoccasionallycrossedbyjoggersandcrowds,andveryrarelybydensecrowdsor
exposed to vandal loading. Based on this consideration, a risk matrix could be constructed, as shown in
2

Table1.Theprobabilityofloadingscenariosisestimatedbasedontheexperiencefromtrafficmonitoring
onthebridge,whiletheseverityofvariousscenariosisautomaticallyassociatedwiththatatthediagonalof
the risk matrix. If the bridge were not already built, then any realisation after calculation that a traffic
scenario of prescribed probability (normally possible to estimate in the design stage based on pedestrian
traffic flow and density of the populated area) produces more severe vibrations than expected (upper
triangleoftheriskmatrix)wouldrequirethemodificationsinthedesigntomitigatetheextremevibrations.
If,however,thecalculatedresponsesarelessseverethanexpected(lowertriangleoftheriskmatrix)then
the design solution is either acceptable or could even be modified to allow for greater slenderness, and
thereforereductioninmaterialcost.
Table1:Riskmatrixforpossibleloadingscenarios.
Severity Low

(vibrationsacceptablefor
mostpedestrians)

Probability

Moderate

High

(vibrationsacceptable
forcertainpercentageof
people)

(vibrationsdonotneedtobeacceptable,
butpanicshouldbeavoidedandintegrity
ofthestructurepreserved)

High

Singlepedestrianloading
Normaltraffic,Groups

Moderate

Joggers,Crowd

Low

Vandals,Densecrowds

SINGLEPEDESTRIANLOADING

Estimationofthevibrationresponsegeneratedbyasinglepedestrianhasbeenapartoffootbridgedesign
guidelinessincetheirintroduction30yearsago.Thissectionstartswithpresentationofvibrationresponse
in analytical form, under an assumption that the relevant mode shape could be described by a halfsine
function.Thisisfollowedbypresentationofmeasuredvibrationresponseandthatcalculatedaccordingto
availabledesignprocedures.
3.1

Analyticalsolution

Ifasinglemodedominatesthefootbridgeresponseandonlyasingleforcingharmonicoffrequency isof
interestthentheresponseofthebridgecouldbefoundbysolvingthewellknowndifferentialequationof
thesecondorder:
(1)

&&
x(t ) + 2n x&(t ) + n2 x(t ) =

F (t )

wherem, and n aremodalmass,dampingratioandcircularnaturalfrequencyrelatedtorelevantmode

of vibration, respectively, &&
x(t ), x& (t ) and x(t) are modal acceleration, velocity and displacement of the
structure, respectively, while F(t) is the modal force obtained by multiplying the humaninduced force by
themodeshape,transformedfromspatialtothetimedomainusinginformationaboutthe(average)speed
offootbridgecrossingv:
(2)

F (t ) = (DLF W )sin(t ) (x) = A sin(t ) (vt ).

DLFistheratiobetweenamplitudeAofthedynamicforceandthepedestriansweightW, isthecircular
frequencyoftherelevantforcingharmonic(usuallyfirstorsecond),while isthemodeshape.Assuming
that the mode shape is a half sine function (which is often the case in analytical procedures defined by
differentauthors):
(3)

(x) = sin

x
L

= sin

vt
L

whereListhelengthofthebridge,thenequation(1)couldbetransformedinto:
(4)

&&
x(t ) + 2n x& (t ) + n2 x(t ) =

A
vt A
v
v

sin(t )sin
cos
=
t cos +
t .

m
L
2m
L
L

Assumingthatthestructureislinear,twotermsontherighthandsidecouldbesolvedseparatelyandthe
principle of superposition applied to get the total response (Inman, 2001). The closed form solution for
acceleration,assumingzeroinitialconditions,duetothefirsttermis:
(5)

n f0
f0
&&

x1 (t ) = 2n2 e nt 2 0 2 cos (d t )

t
sin
sin
(
)
(
)
d
d
+

d ( 2 + 2 )
d ( 2 + 2 )
f

f
f
2n e nt 02 d 2 sin (d t ) 2 n 0 2 cos (d t ) 2 0 2 cos (d t ) +
+
+
+

f 2

f
f
+ e nt 02 d 2 cos (d t ) + n2 d 02 sin (d t ) + 2 d 02 sin (d t )
+
+
+

f
2 0 2 2 cos ( t ) + 2 sin ( t )
+

where

f0 =

A
v
; d = n 1 2 ; = ; = n2 2 ; = 2n .
2m
L

Theclosedformsolution &&
x2 (t ) foraccelerationduetothesecondterminequation(4)isthesameasthat
A
v
presentedinequation(5),with f0 =
and = + . Therefore,thegeneralsolutiontoequation(4)
2m
L
is:
(6)

&&
x(t ) = &&
x1 (t ) + &&
x2 (t ).

Thesolutionisderivedforanarbitraryforcingfrequency.
While the solution might look complicated, it is actually not difficult to code, and therefore is worth
implementing instead of making unnecessary simplifications that might lead to erroneous vibration
estimates.
Theassumptionthatvibrationmodehashalfsineshapeallowedforcalculationoftheclosedformsolution.
For Podgorica bridge this assumption seems appropriate for the main 78m long span (Figure 1b). It is
interestingtoseehowthisapproximationaffectstheaccuracyofthesolution.Figure2ashowstheresponse
according to equation (6), with peak response of 0.31m/s2, which is a nearly 20% overestimation of the
peakresponseof0.26m/s2calculatedusinganumericalprocedureandtakingintoaccountthemeasured
modeshape(Figure2b).Bothresponsesarecalculatedunderanassumptionofforceamplitudeof180N,
stepfrequencyof2.04Hzandpedestrianspeedof1.84m/s,whichareparameterstakenfromBS5400(BSI,
1978).Althoughtheanalyticalsolutionoverestimatestheactualresponseitmightsometimesbeconvenient
touseitinanalyticalstudies.Theanalyticalsolutionisalsoshownheresinceitisabittedioustoderive,soit
couldbeusedasareferenceinsomefutureworks.
The further calculation in this paper will be done using measured mode shape, to avoid errors due to
assumptionofhalfsinemodeshape.

0.4
Modal acceleration [m/s2]

L=78m
half sine mode shape

Modal acceleration [m/s ]

0.4
0.2

0.0

-0.2

-0.4

L=104m
measured mode shape
0.2 BS5400 model

0.0

-0.2

-0.4

(a)

10

20
30
Time [s]

40

3.2

Measuredresponsestosinglepersonexcitation

(b)

10

20

30
40
Time [s]

50

60

Figure2:Modalaccelerationcalculated(a)whenhalfsinemodeshapeisassumed,and(b)withmeasured
modeshape.

Themeasuredpeakaccelerationresponseofthebridgetosinglepersonexcitationwasupto0.5m/s2for
freewalkingandupto0.7m/s2formetronomecontrolledwalkinginresonanceperformedbyawelltrained
testsubject.Itisinterestingtocomparethesemeasuredresponseswiththosecalculatedusingsomedesign
procedures.

3.3

Responsetosinglepersonaccordingtovariouscalculationprocedures

The response estimate required in British (BSI, 1978) and Canadian (OHBDC, 1983) footbridge codes is
identical for this bridge, and produces a peak estimate of 0.26m/s2 (Figure 2b), as explained previously.
However,the modelin Eurocode 5(EC, 1997)for design oftimber bridges, definesasingle personinthe
same way as BS5400, except that the force amplitude is 280N. This produces the response of 0.40m/s2,
which is clearly much closer (than BS5400 estimate) to the maximum measured response of 0.5m/s2 for
free walking. However, it is still an underestimation, and this is purely due to differences in excitation
potentialbetweendifferentpeople.Therefore,itisofcrucialimportancetoupdatetheexistingmodelswith
the one that takes into account properties of human population (such as range of force magnitude) in a
morepreciseway,i.e.inprobabilisticmanner.
The most comprehensive database of forcing amplitudes is collected by Kerr (1998). He measured about
1000forcerecordsgeneratedby40testsubjects.Hefoundgreatvariabilityintheamplitudeandexpressed
the results in form of DLFs. He concluded that DLF1 is frequency dependent, and could be fitted by a
polynomial(Figure3a):
(7)

DLF1 = 0.2649 fp3 + 1.3206 fp2 1.7597 fp + 0.7613,

whiletheupperandlowerboundscovering 2 confidencebands(where isthestandarddeviation)are:

(8)

upperbound:0.5073 fp 0.4843
lowerbound:0.2613 fp 0.2495

where fp ispacingfrequencyinHz.
For walking at 2.04Hz the most probable DLF1 would therefore be 0.42, while the upper 2 confidence
boundis0.55.Thetwofactors,appliedtoaweightof700N,produceforceamplitudesof294Nand385N,
respectively(asopposedto180NinBS5400and280NinEC5).Takingwalkingspeed v asfunctionofpacing
frequency(PachiandJi,2005):
(9)

v = 0.71 fp = 0.71 2.04 = 1.45m/s

produces the vibration response of 0.48 and 0.63m/s2, for an average and dynamically quite efficient
pedestrian,respectively.Theoretically,thesetwovaluesapproximatelycover50%and95%ofresponsesto
pedestrians walking at resonance, if DLF distribution is assumed to be Gaussian. Therefore, if the single
personmodelisrealistictheneventhemostextremeresponsestosinglepersonwalkingcouldbecovered.
Note that the absolute maximum response of 0.7m/s2 measured with help of metronome could be
consideredastheresponsegeneratedbyatestsubjectbelongingto5%ofpeoplenotcoveredby95%limit
DLFvalueof0.55(dotsabovetheupperboundinFigure3a.
ThevariabilityofDLFforthesecondharmonicisshowninFigure3b,andcouldbeusedforfootbridgeswith
naturalfrequencyusuallyintherange3.04.8Hz.
0.8

0.25

0.7
0.20

0.6
DLF1

0.5

0.15

0.4
0.10

0.3
0.2
0.1
0.0

(a)

0.05
1.0

1.4
1.8
2.2
footfall rate [Hz]

2.6

0.00

(b)

2.0 2.5

3.0 3.5 4.0 4.5

2 x footfall rate [Hz]

5.0 5.5

Figure3:Dynamicloadingfactorsasfunctionsofstepfrequencyfor(a)firstand(b)secondharmonic(after
Kerr,1998).

3.4

Probabilisticestimationofresponsetosinglepersonloading

Force models based on a single harmonic assume that the walking activity generates a periodic force.
However,walkingisactuallyanarrowbandrandomprocessinwhichthereisadissipationofenergyaround
main harmonics (Brownjohn et al., 2004). This is the consequence of imperfections inhuman walking. To
accountforthis,multiharmonicforcemodelisdevelopedbyivanovietal.(2007).Thefrequencydomain
forcingcontentuptothefifthharmonicwasfittedtotreadmilldata(Brownjohnetal.,2004)includingall
thespectrallinesbetweentheharmonics(Figure4a),aswellasthesubharmonicsduetounbalancedgait
(Sahnaci and Kasperski, 2005). The model was then transformed to the timedomain assuming random
phases between the spectral lines. Due to its representation of the forcing frequency content up to
approximately10Hz,thismodelcouldbeusedforpredictionofmultimoderesponse.
The model described is used in Monte Carlo (MC) simulations to predict the vibration response to
pedestrianshavingdifferentstepfrequency,steplengthandgeneratingdifferentforcemagnitudes.These
parameters were taken from appropriate Gaussian distributions (ivanovi et al., 2007). As a result the
probabilistic estimate of the response to whole pedestrian population using the bridge was made. The
resultinformofprobabilityofnonexceedenceofaparticularvibrationlevelisshowninFigure4b.Itisclear
thatforthegiven bridge95% ofpeoplewould induce vibrations below0.35m/s2, while50% ofthemwill
inducevibrationsupto0.033m/s2only.Itisalsoclearthattherangeofvibrationthatcouldbeinducedbya
singlepersonisquitelarge,i.e.itcouldbealmostanythingbelow0.7m/s2.
The MC simulations were conducted using VSATs software developed in Vibration Engineerign Section at
SheffieldUniversity(ivanovietal.,2008a;2008b).

3.5

Discussion

Singlepersonmodelsconsistingofasingleharmoniconlyareusefulmeansofquickestimateoffootbridge
response under a single pedestrian loading scenario. This scenario is highly relevant for designing
footbridges residing in not very busy environments. However, it should be kept in mind that the single
6

person models defined in BS5400 and Ontario code are outdated, primarily in regard to DLF and walking
speed parameters. Using new data, such as those collected by Kerr (1998) and Pachi and Ji (2005), these
parameterscouldeasilybeupdated,asdemonstratedinsection3.3.
Ifmorethanoneharmoniccontributetofootbridgeresponseinoneormoremodesofvibration,orwhen
more precise estimates of the response that take into account the dissipation of energy around main
harmonics and subharmoncis are needed, then the multiharmonic force model (ivanovi et al., 2007)
couldbeemployed,asdemonstatedinsection3.4.
Non-exceedence probability

harmonic 5

harmonic 4

subharmonic 5

harmonic 3

0.05

subharmonic 4

0.10

subharmonic 2

0.15

subharmonic 3

0.20

harmonic 2

harmonic 1

1.0

subharmonic 1

Fourier amplitude / weight

0.25

0.00

(a)

0.8
0.6
0.4
0.2
0.0

2
3
4
5
1
Frequency / Walking frequency

(b)

0.1

0.2 0.3 0.4 0.5

2
Acceleration [m/s ]

0.6 0.7

Figure4:(a)Frequencycontentofawalkingforcemeasuredonatreadmill.(b)Cumulativeprobabilityof
peakaccelerationforsinglepersonwalkingscenario.

NORMALTRAFFIC

Thissectionpresentsthemeasurementsofvibrationresponseundermultipersontrafficandcomparethem
withpredictionsofvibrationresponseaccordingtovariouseithercodifiedoruncodifedprocedures.Normal
traffic here is defined as a traffic without spatial restrictions, allowing pedestrians to move freely with a
preferred speed. This condition was mostly satisfied during monitoring tests to be described in the next
section.

4.1

Measurements

The Podgorica footbridge was subject to vibration monitoring under normal pedestrian traffic on two
different days. During total monitoring time of 4.5 hours, about 3000 people crossed the bridge, with an
averagenumberof14peoplepresentonthebridgeatanytime.Occasionallygroupsoftwoormorepeople
werecrossingthebridgewhilechatting,whichischaracterisedbythesamecrossingspeed.Fromacquired
videorecordsitwasfoundthatthesynchronisationintheirpacingfrequencywasrareexceptionratherthan
arule.
A typical 45minute response in 1V mode measured at its nodal (midspan) point is shown in Figure 5a
togetherwiththetotalnumberofpeoplepresentonthebridgeatanytime.TheRMSvalueofthewhole
signalwas0.133m/s2.Thepresenceofmovingpeopleslightlyreducedthevibrationfrequencyfrom2.04to
about 2.00Hz. Modal response in mode 1V was almost the only contributor to the total response during
monitoringsincetheothermodeswerehardlyexcitedbypedestriantraffic.
During normal multiperson pedestrian traffic the response went mainly up to 0.4m/s2, and quite rarely
reachedorexceeded0.6m/s2(Figure5a).Maximumnumberofpeopleonthebridgeatatimewasaround
80duringpedestriantrafficmonitoringtests,whichcorrespondstothedensityof0.26pedestrians/m2.
Based on 4.5 hours of traffic and vibration monitoring, the cumulative distribution of peak accelerations
(pereachvibrationcycle)isshowninFigure5b(solidline).Themaximumaccelerationinthesamefigureis
0.79m/s2,while95%ofpeakvaluesarebelow0.35m/s2.Meanofpeakaccelerationpercycleis0.16m/s2,
whilethemedianis0.14m/s2.
7

RMS=0.133m/s

1.0

30
20
2

1.0

10

(a)

0.5
0.0

1000

2000

3000

Time [s]

-0.5
-1.0

Probability of non-exceedance

40

Modal acc. [m/s ]

Number of people on the bridge

0.8
0.6
0.4
0.2
0.0
0

(b)

0.2

0.4

0.6

0.8

1.0

Peak acceleration per vibration cycle [m/s2]

Figure5:(a)Modalaccelerationandthecorrespondingnumberofpeopleduringa45minlongtest.(b)
Cumulativedistributionofthepeakaccelerationpervibrationcycleasmeasuredandcalculated.

4.2

Numericalmodels

Theavailablecalculationprocedures,eitherdefinedinthetimeorfrequencydomain,willbepresentedin
thissectionandtheiroutputscomparedwiththemeasuredresponses.
4.2.1

EmpiricalmodelinformofGEVdistribution

Whenmeasurementsoffootbridgeresponseundernormalpedestriantrafficareavailablethenanestimate
of the peak response as a function of return period could be made based on generalised extreme value
(GEV)distributionfitoftheexperimentaldata(ivanovietal.,2008c).Thisallowsforpredictionofthepeak
response (or so called return level) depending on the return period of interest (Figure 6a). Note that the
confidence intervals are broader (i.e. the confidence in response estimate is smaller) for higher return
levels,aswouldbeexpectedduetounavailabilityofexperimentaldataforhigherreturnlevels.
The property of underlying Weibull distribution, used in GEV fit, is that it has a finite maximum (Coles,
2001).Theoretically,thismaximumlevelcorrespondstozeroprobabilityofoccurrence(i.e.infinitereturn
period)andonPodgoricabridgeitisequalto1.27m/s2whichisabelievableestimatefornormalpedestrian
traffic. The existence of an absolute maximum in the theoretical model agrees well with the logical
expectationthatabridgehasaphysicallimitofmaximumvibrationlevelthatdependsonitsstiffness,mass
anddampingpropertiesaswellasabilitytoaccommodatealimitednumberofpedestrians.
4.2.2

Matsumotoetal.s(1978)model

Early researchon crowdloadingconductedbyMatsumotoetal. (1978) suggestedestimation of vibration

responsetonormalpedestriantrafficbymultiplyingtheresponsetoasinglepersonexcitingtheresonance
ofthefootbridgebyafactor N ,whereNisthenumberofpeopleonthebridgeatthesametime.This
dependencewasderivedbysummingresponsesofindividualpedestriansarrivingonthebridgeaccording
tothePoissondistributionandinducingthewalkingforceswithequal(resonant)frequenciesandrandom
phases.Assumingthatasingleaveragepedestrianinducedthepeakaccelerationof0.48m/s2(seeSection
3.3), the predicted peak response for mean number of 14 pedestrians would be 1.80m/s2, while for
maximumof80peopleitwouldbe4.29m/s2,thereforesignificantlyoverestimatingthemeasuredvalues.
Clearly, a single person response of 0.48m/s2 is quite good estimate of the response to normal traffic. It
seems that there is no room for significant increase of this response under multiperson loading, and
therefore Wheelers (1982) suggestion that the single person response is good enough estimate of
multipersontrafficseemsvalidonthisbridge,aslongasthesinglepersonismodelledwitharealisticDLF
dependentonwalkingfrequency.
8

Peak acceleration [m/s ]

1.2
1.1

Mean pacing frequency [Hz]

2.0

Return level [m/s2]

1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3

(a)

0.2
0.1 0
10

0.9
0.8
0.7

1.9

0.6
1.8

0.5
0.4
0.3

1.7

0.2
0.1

1.6
10

10

10

Return period [s]

10

10

(b)

20 30 40 50 60 70
Number of people on bridge

80

Figure6:(a)Returnlevelestimateasafunctionofreturnperiod.The95%confidenceintervalsforthe
returnlevelsareshownasdashedlines.(b)Peakaccelerationasafunctionofmeanpacingrateofthe
pedestrianpopulationandnumberofpeopleonthebridge.
4.2.3

Setra(2006)model

With maximum pedestrian density of about 0.26 ped/m2 (for 80 people), the bridge is closest to Class III
defined in French Setra guideline (Setra, 2006). For this class the multiplication factor for single person
responseisdefinedas
(10)

10.8 N

( (x) dx ) / L = 10.8 1
L

0.0026 N 39.0 /104.0 = 0.21 N

where isthefactorrelatedtoriskofresonance, andNarethemodaldampingandtotalnumberof

people on the bridge, while (x) and L are the mode shape and footbridge length, respectively. Setra
guideline requires multiplying the stationary resonant response produced by a single person by this
multiplication factor to get response to normal traffic. Single person force having amplitude of 280N
generates the maximum response of 0.93m/s2. Therefore, for 80 people on the bridge the estimated
responsewouldbe 0.21 80 0.93 = 1.75 m/s2.However,ifmovingsinglepersonmodelproducing0.48m/s2
peak acceleration is used, the predicted response would be 0.90m/s2, which is in quite reasonable
agreementwithmeasuredlevelof0.79m/s2.
4.2.4

Designspectrumapproach

BasedonMonteCarlosimulationsofresponseofastandardbridgetoastandardpopulationofpeople
designspectrumrepresentingthepeakresponseofthebridgeinasinglevibrationmodeasafunctionof
naturalfrequencycouldbeconstructed(Ingolfssonetal.2008;GeorgakisandIngolfsson2008).Thedesign
spectraofthiskindaresimilartothoseusedbytheearthquakeengineeringcommunitywherepeakmodal
responsevaluesarepresentedasafunctionofmodefrequencyforaspecificinput.Thesquarerootsumof
squares (SRSS) approach is used to sum the modes. To account for differences between a bridge to be
designed and the standard one used while constructing the design spectra some correction factors for
differentparameters areintroduced. Thedifferencesbetweenthe parameters characterisingthe realand
standardpopulationsaredealtwithinasimilarmanner.
Response of Podgorica bridge to standard population estimated using the appropriate spectrum is
0.25m/s2. Taking the actual modal properties of the Podgorica bridge, and the fact that pedestrian
population crossing the bridge walked at a mean pacing rate of 1.87Hz, the correction factors related to
bridgelength,dampingratio,modeshape,modalmass,pedestrianarrivaltime,meanstepfrequencyand
return period for which calculation is being done (4.5 hours in this case)the peak response predicted for
9

average traffic of 14 people onthe bridge would be0.36m/s2, while with maximum of 80 people on the
bridgeitwouldbe0.86m/s2.Theseestimatesareingoodagreementwiththeexperimentalresults.
The method described could be used for estimation of the response to various arrival times as well as
differentmeanpacingratesinthepopulationofpedestriansusingthebridge.ThesearepresentedinFigure
6bisformofcontourcurves.
4.2.5

Scrutonnumber

Newland(2004)suggestedthatthefootbridgeserviceabilitycouldbeevaluatedusinganequivalentofthe
Scrutonnumber(originallyusedinwindengineering),definedas:
(11)

Sp =

2 M

where Sp ispedestrianScrutonnumber, ismodaldampingratio,Misbridgemassperunitlengthandm

is pedestrian mass per unit length. Higher value of structural damping and mass are beneficial, and they
producehigherScrutonnumber.Increasingthenumberofpeople,however,reducesScrutonnumber.This
is supposed to account for potential instability of the bridge under crowd, although the instability in the
vertical direction does not seem very probable (ivanovi et al., 2005b; Setra, 2006). Newland (2004)
suggestedthatScrutonnumbershouldexceedaprovisionalvalueof0.27forfootbridgetobestableunder
pedestrianload.
Scruton number was calculated for Podgorica bridge, which total mass is about 260 tonnes, assuming
presenceofmaximumof80pedestrians(75kgeach)uniformlydistributedacrossthebridgeandvalueof
0.23wasobtained.Thisisbelowtheprovisionallimitof0.27suggestingthatthebridgeispronetovibration
in the vertical direction (although this could not be related to instability phenomenon), which is in
agreementwithwhatwasobservedinpractice.
4.2.6

MonteCarlosimulations

Using VSATs software (ivanovietal., 2008a;2008b)Monte Carlosimulationfor 4.5 hoursofpedestrian

traffic were conductedassumingthearrival time of7 peopleperminutefromeither endofthebridge,a
little above the experimentally observed total of 11 crossings per minute. Each pedestrian was modelled
accordingtomodeldefinedinbyivanovietal.(2007).Distributionofamplitudeofmainharmonicsofthe
walkingforceaswellaspacingfrequencywasassumedtobenormal.Takingintoaccountmodalproperties
of the relevant vibration mode, an overestimation of the response would be obtained (ivanovi et al.
2005b). Since it seems that HSI occurs on this bridge when pedestrians feel perceptible vibrations in the
formofdampeningthemoutthenewsimulationsofthebridgeresponsewithdampinglevelincreasedfrom
0.26to0.6%wereconducted.Theresultinformofdistributionsforpeakaccelerationpercycleisshownin
Figure5b(dashedline).Itcanbeseenthatthedistributionisveryclosetothemeasuredone(solidlinein
Figure5b).
4.2.7

Brownjohnetal.(2004)model

Thismodeldefinesthepedestrianloadinginformofautospectraldensity(ASD)thatdependsonprobability
distributionofpedestrianpacingrates.TheASDoftheresponsecouldbeobtainedbymultiplicationofASD
forforce,squareofthefrequencyresponsefunctionforthestructureandthecoherencefunctionbetween
pedestrians at different locations taking also into account the ordinate of the mode shape at position of
eachpedestrian.ThesquarerootoftheresponseASDgivesestimateoftheRMSresponse.Sincecurrently
thereisnodataaboutcoherenceavailable,itcouldbeassumedthatitisequaltozero.Assumingdamping
of0.6%andthat38peopleispresentonthebridge,whichwasmaximuminthetestshowninFigure5a,the
10

RMSresponsepredictedisaround0.10m/s2.TheunderestimationofthemeasuredRMSvalue(Figure5a)is
expected,andcouldbecorrectedoncetheappropriateformofcoherencefunctionisavailable.Therefore
thecoherencefunctionrequiresadditionalresearchandexperimentalquantification.

4.3

Discussion

Thekeyparameterscharacterisingnormalpedestriantrafficarethearrivaltime(thatinfluencesthenumber
of peopleonthebridge atanytime,andthereforethepedestriandensity)andprobabilitydistribution of
theirpacingfrequencies.Theonlyguidelinethatcancopewiththistrafficscenarioandcouldgiveaquite
reasonable estimate of the measured response is Setra (2006) guideline. However, it should be said that
someimprovisationinitsimplementationisrequired.Additionally,responseobtainedbyacoupleofother
models, such as those defined by Brownjohn et al. (2004) and Georgakis and Ingolfsson (2008) are quite
promisingaswell.Thisgiveshopethatsomeoftheseprocedureswillbecomepartofdesigncodesandwill
beusedinfootbridgedesignondailybasis.However,forthistohappenverificationofalltheseprocedures
onotherfootbridgesisalsorequired.
Finally,aquestionarisesifthenotionthattheresponsetoa(dynamicallymostefficient)singlepersoncould
be used as prediction of the response to multi person traffic is a myth of reality. On Podgorica bridge it
seemsthat this notion is quite realistic, since themaximum estimate to single person loading of 0.7m/s2
(Figure4b)isquiteclosetothepeakresponsemeasuredduringmonitoringtime.However,toanswerthe
questionwithconfidence,moredatafromotherbridgesshouldbeusedforverificationpurpose.

GROUPLOADING

As previously described, some groups of people crossing the bridge together were noticed during traffic
monitoring. To account for this, the worst case scenario of groups of up to five people, all walking at an
averagepacingfrequencyof2.04Hzandwithspeedof1.45m/s,isconsidered.Eachpedestrianinagroupis
modelled using model by ivanovi et al. (2007). This means that random phases were induced between
peoplewalkingtogether.Foreachgroupsizethesimulationswererepeated200times.
Theresultsareshowninformofcumulativedistributionofpeakresponsesinducedbygroupsofdifferent
sizes (Figure7). Clearly, the response increases with group sizein a nonlinear manner. It again should be
emphasisedthatitismuchmoreprobablethatpeoplewillwalkwithdifferentstepfrequencieswhileina
group, than at the same pacing rate (equal to the natural frequency of the bridge). However, resonant
walkingfrequencyforallmembersofthegroupisassumedhereastheworstcasescenario.
1.0

Non-exceedance probability

0.8

0.6

0.4

0.2

0.0
0

0.5

1.0

1.5

2.0

2.5

Peak acceleration [m/s ]

Figure7:Cumulativeprobabilityofpeakresponsetouptofivepeoplewalkingtogetheratfrequencyof
2.04Hz.
11

RUNNINGANDVANDALLOADING

Running and jumping (or vandal loading) normally induce the dynamic forces that are much larger than
thoseinducedbywalking.Typicalfrequenciesfortheseactivitiesare1.93.3Hzforrunningand1.32.4Hz
for jumping (Bachmann, 2002). Therefore, there is a danger that some runners/jumpers could induce
resonanceinPodgoricabridgebymatchingitsnaturalfrequencyof2.04Hz.
Differentlyfromwalkingactivitythatischaracterisedbycontinuouscontactwiththeground,runningand
jumpingactivitiesarebothcomposedofacontactandaflyingphaseineachstep/jump.Thisisthereason
thattheyareoftenmodelledinasimilarway.BachamnnandAmmann(1987)suggestedahalfsinemodel
fordescribingthecontactphasewhiletheforceiszeroduringtheflyingphase:
(12)

k W sin ( t / tc ) fort t c
F (t ) = p

0fortc < t < T

where W is test subjects weight, tc is contact time, T is the period of running/jumping activity and kp is
dynamicimpactfactor,whichisafunctionoftheratiobetweencontacttimeandperiod(socalledcontact
ratio):
(13)

k p (tc / T ) =

2tc / T

This factor is calculated usingenergypreservation equation forflying phaseand the impulse theorem for
contactphase(EllisandJi,1997;Occhiuzzietal.,2008).
Thismodelisoneofraremodelsavailableforrunning/jumpingactivity,andwillbeusedhereforestimation
oftheresponseofPodgoricabridge.However,itisworthsayingthatBachmannandAmman(1987)warned
thatthismodel couldboth overandunderestimate theforcemeasuredexperimentally.Thisisprobably
partlyduetointerandintrasubjectvariability,andpartlyduetooversimplificationoftheforcedescription,
whichactuallycoulddiffersignificantlyfromahalfsineshape.

6.1

RunningonPodgoricabridge

It is assumed that a single runner, weighing 750N, is crossing the bridge at frequency 2.04Hz and with
speed2m/s.Thecontactratioisassumedtobe1/3.Withtheseparametersthepeakresponseofthebridge
would be 1.82m/s2. Clearly, two synchronised runners would induce twice this value; three will induce
threetimesthisvalue,andsoon.
However,itisdifficulttoexpectthat,saytwojoggerswillsynchroniseperfectlywell,aswellasthattiming
ofeitherjoggerwillbeimpeccablesothatforceisrepeatedexactlywithperiodTandwiththesamecontact
time throughout the crossing. If contact ratio for each leg contact with the bridge for either jogger is
generatedrandomlyfromaGaussiandistributionhavingmeanof1/3andcoefficientofvariation(COV)of
5%, and the period of the force is generated from Gaussian distribution with mean of 1/2.04s with COV
equalto5%,thentheresponsecanvaryhugely.Theresponsetotwojoggerswascalculatedin1000runsto
account for randomness of the parameters used. The envelopes of some responses (four of them
emphasised just to represent the shape of the response envelopes) are shown in Figure 8a. Any peak
responsebetween0.3m/s2and3.3m/s2isfeasible,butwithdifferentprobabilityofoccurrence,asshown
inFigure8b.

12

70

3.0

60

Frequency of occurrence

Modal acceleration [m/s ]

3.5

2.5
2.0
1.5
1.0
0.5
0.0
0

(a)

10

20

30
Time [s]

40

50

60

(b)

50
40
30
20
10
0
0

0.5

1.0 1.5 2.0 2.5 3.0

Modal acceleration [m/s2]

3.5

Figure8:(a)Envelopesforresponsesfor50differentsimulationsoftwojoggers.(b)Histogramofpeak
accelerationresponsegeneratedbytworunners.

6.2

VandalloadingonPodgoricabridge

Usually,severalpeoplejumpinginunisonareconsideredasvandalloading.Bachmann(2002)suggeststhat
up to three people could synchronise their movement quite well, and therefore should be modelled as
perfectly in phase with each other. Apart from jumping, bouncing should also be considered as possible
vandalloading,especiallyinstructureswithnaturalfrequenciesabovetherangefornormaljumping,sayup
to4Hz.ForPodgoricabridge,thejumpingloadisrelevantsinceajumpingatfrequencyof2.04Hzisquite
comfortable.
Forasinglejumperweighing850N,andjumpingatfrequencyof2.04Hzfor50s,withcontactratioequalto
1/3(EllisandJi,1997),thepeakresponsecalculatedis4.1m/s2.Twoperfectlysynchronisedjumperswould
induce 8.2 while four of them would induce 16.4m/s2. These levels are quite high, dangerous and
unrealistic.Tosupportthisclaim,1000runsofsimulationinvolvingfourpeoplejumping,witheachjumpof
eachpersonhavingnormallydistributedcontactratioandperiod,thistimewithCOV=10%,wouldproduce
timehistoriesforsomeofwhichenvelopesarepresentedinFigure9a.Again,broadrangeofpeakvalues
wasobtained,withtheirprobabilityofoccurrenceshowninFigure9b.Anexperimentwith4peopletrying
to jump in a synchronised manner produced the acceleration of about 2m/s2, but they were ordered to
stopjumpingduetostrongvibrationofthefootbridge.
Frequency of occurrence

70

Modal acceleration [m/s ]

(a)

10

20
30
Time [s]

40

50

60
50
40
30
20
10
0

(b)

3
4
5
6 2 7
Modal acceleration [m/s ]

Figure9:(a)Envelopesforresponsesfor50differentsimulationsoffourjumpers.(b)Histogramofpeak
accelerationresponsegeneratedbyfourjumpers.

6.3

Discussion

Theguidanceonfootbridgedesignwithregardtorunningandvandalloadingissparseandrequirefurther
research. This especially applies to possibility of synchronisation during running and jumping in small
groups,tothemorerealisticmathematicaldescriptionoftheloading,andtoexpecteddurationofvandal
13

excitation. Also, the correlation between changes in period, contact ratio and force amplitude with each
step/jumpcouldbeinvestigated.
The numerical simulations shown in this paper are only an illustration how small (assumed) variation in
contactratioandperiodcouldinfluencethepeakresponse.Thisistypicalforbridgeswithsmalldamping.
Thisvariationshouldexperimentallybequantifiedbeforesomedesignrecommendationscanbemade.

CROWDS

Apartfromnormalpedestriantrafficwithunrestrictedpedestrianmovement,possiblescenariosaresome
mediumcrowdofdensityaround0.5ped/s2andquitedensecrowdswithdensity1ped/s2.Theonlydesign
guidelineabletocopewiththeseloadingscenariosisSetra(2006)guideline.
For the medium crowds of 156 people on the bridge at a time Setra (2006) predicts peak response of
1.26m/s2(usingactualpeakresponsetosinglepersoninsteadofrecommendedresonantresponse),while
for density of 1ped/m2, the predicted response would be 5.88m/s2. On the other hand, the design
spectrum model (Georgakis and Ingolfsson, 2008) would produce around 1.2 and 1.7m/s2, respectively.
Whiletheestimatesformediumcrowdaresimilar,theonesfordensecrowdarequitedifferent.Itseems
thatSetraestimateisheavilyconservativeinthiscase.
Finally, Monte Carlo simulations of medium and dense crowds were conducted using VSATs software
(ivanovi et al., 2008a; 2008b), and assuming an increased structural damping of 0.63%. The 45 minute
long time domain response with 1s RMS trend, its spectrum and the distribution of peak response per
vibrationcycleareshowninFigure10formediumandinFigure11fordensecrowds.Itcouldbeseenthat
the peak response estimates are around 1.64 and 2.48m/s2, respectively. However, these are probably
conservative estimates since it would be expected that the modal damping ratio is higher than assumed
0.63%duetoincreasednumberofpeopleonthebridgeandthehumanstructuredynamicinteraction.

CONCLUSIONS

Footbridge structures are likely to be exposed to various loading scenarios during their lifetime. It is
recommended to conduct a risk analysis for occurrence of different loading scenarios depending on
footbridge position, density of population in the area, alternative routes available, etc. Current design
guidancehowtomodelvariousloadingsislimited,withtheSetra(2006)guidelinecurrentlybeingthemost
advance one. However, certain degree of improvisation even when applying this guideline is required.
Therefore,thecurrentdesigncodesshouldbeusedwithcaution,andshouldalwaysbeaccompaniedwith
calculationsaccordingtomoreadvanceddesignprocedures.
This paper analysed several possible loading scenarios on a box girder footbridge, and compared results
from different numerical models available. While some advances are noticeable in modelling the crowd
loading,itseemsthatthereisstillresearchworktobedoneforestimatingoftheresponsetovandaland
runningloading.

ACKNOWLEDGEMENTS
TheauthorwouldliketothanktothefinancialsupportwhichcamefromtheUKEngineeringandPhysical
Sciences Research Council (EPSRC) for grant reference GR/T03000/01 (Stochastic Approach to Human
StructureDynamicInteraction).

14


Figure10:Theresponseestimateundermediumcrowdofpeople,withapproximatedensityof0.5ped/m2.

15


Figure11:Theresponseestimateunderdensecrowdofpeople,withapproximatedensityof1ped/m2.

REFERENCES
Bachmann,H. andAmmann, W. (1987).Vibrations in StructuresInduced by Man andMachines IABSE AIPC IVBH,
Zrich,Switzerland.
Bachmann,H.(2002)Livelyfootbridgesarealchallenge.IntheProceedingsoftheInternationalConferenceonthe
DesignandDynamicBehaviourofFootbridges,Paris,France,2022November.
Brownjohn,J.M.W.,Pavic,A.andOmenzetter,P.(2004)Aspectraldensityapproachformodellingcontinuousvertical
forcesonpedestrianstructuresduetowalking,CanadianJournalofCivilEngineering,31(1),6577.
BSI(1978)Steel,concreteandcompositebridges.Specificationforloads,BS5400:Part2,BritishStandardInstitution
1978;London.
Coles,S.(2001)AnIntroductiontoStatisticalModelingofExtremeValues,Springer,London.
Dallard P, Fitzpatrick AJ, Flint A, Le Bourva S, Low A, RidsdillSmith RM, Willford M. (2001) The London Millennium
Footbridge.StructuralEngineer,79(22),1733.
EC (1997) Eurocode 5, Design of Timber StructuresPart 2: Bridges, ENV 19952: 1997, European Committee for
Standardization,Brussels,Belgium.
Ellis,B.R.andJi,T.(1997)Theresponseofstructurestodynamiccrowdloads.BREDigestReport426.
Georgakis, C. T. and Ingolfsson, E. T. (2008) Vertical footbridge vibrations: the response spectrum methodology, In
Proceedingsofthe3rdInternationalConferenceFootbridge2008,Porto,24July.

16

Ingolfsson, E. T., Georgakis, C. T. and Svendsen, M. N. (2008) Vertical footbridge vibrations: details regarding
experimental validation of the response spectrum methodology, In Proceedings of the 3rd International Conference
Footbridge2008,Porto,24July.
Inman,D.J.(2001)EngineeringVibration,PrenticeHall2001,UpperSaddleRiver,NewJersey.
Kerr,S.C.(1998)Humaninducedloadingonstaircases,PhDThesis,UniversityCollegeLondon,MechanicalEngineering
Department,UK.
MatsumotoY,NishiokaT,ShiojiriH,MatsuzakiK.(1978)Dynamicdesignoffootbridges.IABSEProceedings1978,No.
P17/78,115.
McRobie, A. & Morgenthal, G. (2002). Risk Management for PedestrianInduced Dynamic of Footbridges, in the 1st
International Conference Footbridge 2002on the Design and Dynamic Behaviour of Footbridges, Paris, 2022
November.
Newland,D.E.(2004)Pedestrianexcitationofbridges.JournalofMechanicalEngineeringScience;218(5),477492.
Occhiuzzi,A.,Spizzuoco,M.andRicciardelli,F.(2008)Loadingmodelsandresponsecontroloffootbridgesexcitedby
runningpedestrians.StructuralControlandHealthMonitoring,15(3),349368.
OHBDC (1983) Ontario Highway Bridge Design Code. Highway Engineering Division; Ministry of Transportation and
Communication,Ontario,Canada.
Pachi,A.andJi,T.(2005)Frequencyandvelocityofpeoplewalking.TheStructuralEngineer2005;83(3),4640.
Sahnaci, C. and Kasperski, M. (2005) Random loads induced by walking, in Sixth European Conference on Structural
Dynamics(EURODYN),Millpress,Rotterdam,Paris,pp.441446.
Setra(2006)Guidemthodologiquepasserellespitonnes.Setra.Servicedetudestechniquesdesroutesetautoroutes,
Paris.
Wheeler, J. E. (1982) Prediction andcontrol of pedestrian inducedvibrationin footbridges. Journal of the Structural
DivisionASCE,108(ST9),20452065.
ivanovi, S., Pavic, A. and Reynolds, P. (2005a). Vibration serviceability of footbridges under humaninduced
excitation:aliteraturereview.JournalofSoundandVibration,279(12),174.
ivanovi, S., Pavic, A., Reynolds, P. and Vujovi, P. (2005b) Dynamic analysis of lively footbridge under everyday
pedestrian traffic, Sixth European Conference on Structural Dynamics, Vol. 1, pp. 453459, Paris, France, 47
September.
ivanovi,S.,Pavic,A.andReynolds,P.(2006)ModaltestingandfiniteelementmodeltuningofPodgoricafootbridge.
EngineeringStructures,28,857868.
ivanovi, S., Pavic, A. and Reynolds, P. (2007) Probabilitybased prediction of multimode vibration response to
walkingexcitation,EngineeringStructures,29(6),942954.
ivanovi,S.,Pavic,A.andBrownjohn,J.M.W.(2008a)Vibrationserviceabilityassessmentofslenderstructuresusing
VSATssoftware.26thInternationalModalAnalysisConference(IMACXXVI),Orlando,Florida,USA,47February.
ivanovi,S.,Pavic,A.andBrownjohn,J.M.W.(2008b)VibrationserviceabilityassessmentoffootbridgesusingVSATs
software,inProceedingsofthe7thEuropeanConferenceonStructuralDynamicsEURODYN2008,Southampton,79
July.
ivanovi,S.,Brownjohn,J.M.W.andPavic,A.(2008c)Vibrationperformanceoffootbridgesunderpedestriantraffic,
inProceedingsofthe3rdInternationalConferenceFootbridge2008,Porto,24July.

17