A coupled model for train-track-bridge stochastic analysis with consideration of spatial
variation and temporal evolution
Xu, Lei; Zhai, Wanming; Li, Zili
DOI
10.1016/j.apm.2018.07.001
Publication date
2018
Document Version
Final published version
Published in
Applied Mathematical Modelling
Citation (APA)
Xu, L., Zhai, W., & Li, Z. (2018). A coupled model for train-track-bridge stochastic analysis with
consideration of spatial variation and temporal evolution. Applied Mathematical Modelling, 63, 709-731.
https://doi.org/10.1016/j.apm.2018.07.001
Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy
Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
Green Open Access added to TU Delft Institutional Repository
‘You share, we take care!’ – Taverne project
https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher
is the copyright holder of this work and the author uses the
Dutch legislation to make this work public.
ContentslistsavailableatScienceDirect
Applied
Mathematical
Modelling
journalhomepage:www.elsevier.com/locate/apmA
coupled
model
for
train-track-bridge
stochastic
analysis
with
consideration
of
spatial
variation
and
temporal
evolution
Lei
Xu
a,b,∗,
Wanming
Zhai
a,
Zili
Li
ba Train and track research institute, State-Key Laboratory of Traction Power, Southwest Jiaotong University, Chengdu 610031, China b The Section of Road and Railway Engineering, Delft University of Technology, Delft 2628, The Netherlands
a
r
t
i
c
l
e
i
n
f
o
Article history:
Received 30 September 2017 Revised 2 June 2018 Accepted 4 July 2018 Available online 29 July 2018 Keywords: Train-track-bridge interactions Random vibrations Monte-Carlo method Karhunen–Loève expansion Track irregularities
a
b
s
t
r
a
c
t
Due to random characteristics of system parameters and excitations, the dynamic as-sessment and prediction for the train-track-bridge interaction systems become rather complexissuesneedingtobeaddressed,especiallyconsideringthelongitudinal inhomo-geneityanduncertaintyofdynamicpropertiesinphysicsandcorrespondinglytheir tem-poralevolutions. In this paper, atemporal-spatial coupledmodel is developed to fully dealwiththedeterministically/non-deterministicallycomputationalandanalyticalmatters inthe train-track-bridgeinteractions withanovelty, where atrain-track-bridge interac-tionmodel is newly developed by effectivelycoupling the three-dimensional nonlinear wheel-railcontactmodelandthefiniteelementtheory,moreover,theMonte-Carlomethod (MCM)andKarhunen–Loèveexpansion(KLE)areeffectivelyunitedtomodeltherandom fieldoftrack-bridgesystems,andaspectralevolutionmethodaccompaniedbyatrack ir-regularityprobabilisticmodelareintroducedtoselectthemostrepresentativetrack irregu-laritysetsandtocharacterizetheirrandomevolutionsintemporaldimension.Intermsof randomvibrationanalysis,thehigh-efficiencyandeffectivenessofthisdevelopedmodel isvalidatedbycomparingtoarobustmethod,i.e., MCM.Apartfromvalidations, multi-applicationsofthetemporal-spatialcoupledmodelfromaspectsofdeterministic compu-tation,randomvibration,resonant analysisand long-termdynamicprediction,etc.,have beenfullypresentedtoillustratetheuniversalityoftheproposedmodel.
© 2018ElsevierInc.Allrightsreserved.
1. Introduction
Railwaybridges, asa kindofinfrastructure,are becoming increasingly importantinsupporting andguidingthe train-tracksystems.Especiallyinhigh-speedlines,theproportionofbridgesismuchhigherthancommonrailwaylinesthat are mainlysupported by the subgrade layer. Tospecific lines,even 90% over isoccupied by bridges forconservationof land andenvironmentprotection.Hencethetheoreticalmethodsandappliedtechnologiesrelatedtotheassessmentofdynamic performance ofrailway systems,when a train passesthrough the track/bridgestructures, have attractedmore andmore attentionsinlasttwodecades.
Comparing to expensively experimental studies in situ or lab, the dynamic simulations actualized by the computer programhavebecomeadominantstrategyinmostsituations.Tillnow,thedynamicmodelsdevelopedtocharacterizethe train-track-bridgeinteractionsarenumerous,butmostlyconcentratedonverticalvibrations[1–8],obviously,itlimitsthe
re-∗ Corresponding author at: Train and track research institute, State-Key Laboratory of Traction Power, Southwest Jiaotong University, Chengdu 610031, China.
E-mail address: leix_2014@my.swjtu.edu.cn (L. Xu).
https://doi.org/10.1016/j.apm.2018.07.001
Nomenclature
m r therailmassperunitlength
A r thecross-sectionalareaoftherail
W r thepolarmomentofinertiaoftherailcross-section
L thelengthofthebeamelement
l r thedistancebetweentwoadjacentrailpads
E r theYoung’smodulusoftherail
(I ry,I rz) theflexuralmomentofinertiaaboutthe Y -axisand Z- axisofthecrosssectionoftherailrespectively
k rt thetorsionalrigidityoftherailcrosssectionaroundthe X -axis
m s themassoftheslabperunitvolume
(h s,b s,l s) theheight,widthandlengthoftheslabtrackelementrespectively
E s theYoung’smodulusoftheslabtrack
μ
s thePoissonratiooftheslabtrackI sz themomentofinertiaofthetrackslabaround Z -axis
m p themassofthepier
(k px,k px,k px) thesupportingstiffnesscoefficientsofthesubgradein X -, Y -and Z -axis,respectively
(c px,c px,c px) thesupportingdampingcoefficientsofthesubgradein X -, Y -and Z -axis,respectively
(k rp,z,k rp,y) theverticalandlateralstiffnessoftherailpadrespectively
(c rp,z,c rp,y) theverticalandlateraldampingoftherailpadrespectively
b thelateraldistancebetweenthecontactpointofrailpad-slabtrackandtheleft-sideborderoftheslab along X -axis
(k ca,z,k ca,y) theverticalandlateralstiffnesscoefficientsoftheCAM
(B r,H r) thecentrallateralandverticaldistancesbetweentheslabtrackandthegirder
(k p,z,k p,y) theverticalandlateralstiffnessreflectingthepropertiesofthebearing
a 0 thehalfofthehorizontaldistancebetweentwocontactpoints
(r li,r ri) therollingradiusoftheleftandrightwheelsetrespectively
thenominalrollingangularvelocityofthewheelset (I wY,I wZ) themomentofinertiaofthewheelsetaround Y -and Z -axis
M w themassofthewheelset
R i theradiusofthecurvatureoftherailcorrespondingtothe i thwheelset
φ
wi theangleofsuperelevationcorrespondingtothecenterofthe i thwheelset(
ϕ
˙wi,ϕ
¨wi) thefirst-orderandsecond-orderderivativesofφ
ir0 thenominalrollingradiusofthewheelset
˙
λ
wi thefirst-orderderivativeofthecurvature¯g theaccelerationofgravity V therunningspeedofthevehicle M c themassofthecarbody
(I cX,I cZ) themomentofinertiaofthecarbodyaround X -and Z -axisrespectively
R c theradiusofcurvaturewithregardtothecentroidofthecarbody
φ
c theangleofsuperelevationcorrespondingtothecentroidofthecarbody¨
ϕ
c thesecond-orderderivativeofφ
c˙
ς
c thefirst-orderderivativeofthetrackcurvatureH tw theverticaldistancebetweenthecentroidofthebogieframeandthecenterofthewheelset
H bt thevertical distance between thecentroid of the bogie frame andthe bottom plane of the secondary
suspension
H cb theverticaldistancebetweenthecentroidofthebogieframeandtheupperplane ofthesecondary sus-pension
(k sz,k sy,k sx) thesecondarysuspensionstiffnessinvertical,lateralandlongitudinaldirections
(k pz,k py,k px) theprimarysuspensionstiffnessinvertical,lateralandlongitudinaldirections
(d s,d p) thesemi-horizontaldistanceofthesecondaryandprimarysuspensionrespectively
l c thesemi-longitudinaldistancebetweenbogies
l t thesemi-longitudinaldistancebetweenwheelsetsinabogie
(l h,l v) thelateralandverticaldistancebetweenthewheel-railcontactpointandthecentroidoftherail
searchscopessincetheissuesonlateralstabilityandsafetyofsystemcomponentshavegraduallybecomenotableconcerns inrailwayengineering, especially inconditionsofhighspeed operations.Accounting forthis, moreandmoreresearchers starttofocusonbuildingthethree-dimensionaltrain-track/bridgeinteractionmodels,seeforexample,Zhaietal.[9]made a pioneeringworkin comprehensivelyconsidering thecoupleddynamics betweenavehicle andthetracks,in whichthe
three-dimensional(3-D)nonlinearcontact/creepinwheel-railinteractionsareintroducedinfocusedmanner,later,by adopt-ingthesamefundamentals,extensiveresearchesintrain-track-bridgecoupleddynamicsareputintopractice[10,11]; more-over,Refs.[12–17]conductedsignificantworkon3-Dtrain-bridgeinteractions,butneglectingtheeffectsoftrackstructural participationsonsyntheticvibrations,additionally,onlylinearwheel-railinteractionsareaccountedfor;Zengetal.[18] con-structed arather comprehensivemodelfor train-slabtrack-bridge interactions, butstill, thecoremechanismhighlighting thecomplexwheel-railinteractionsislinearizedtounifythesystemcomponentsbyenergy-variationalprinciple[19].
Apart from the developments on constructing train-bridge dynamic simulation models, researchers gradually ex-tend the general deterministic computations into random vibration analysis in light of the random characteristics of the train/track/bridge systems on physical properties and mechanical status. With regard to the random analysis of train-track/bridge interactions, there are mainly three types of methods, i.e., Monte-Carlo method (MCM) [6,20,21], pseudo-excitationmethod(PEM)[17,18,22,23]andprobability densityevolutionmethod(PDEM)[16,24],wherethesystem parametersorexcitations,e.g,trackirregularitiesandseismicmotions,etc.,areassumedtoberandomprocessesstatically characterized by probability densityfunction (PDF) and spectral densities. Undoubtedly,thesework has takensignificant stepstowardstheroadofrandomanalysis,however,farmorescientificresearchesneedtobeprobedinto,e.g.,
• Randomsimulationinvolvingwiththelongitudinaluncertaintyofdynamicpropertiesoftrack/bridgesystemsatspatial dimension.Thoughfragmentaryreportsinvestigatingtheeffectsofrandomtrackbedstiffnessonvehicle-track interac-tionshavebeenpresented[25],aunitedandhigh-efficientmethoddealingwithrandomsimulationandcombinationof multiplevariablesisrarelypresented;
• Ergodiccharacterizationofthesystemexcitations.Forexample,trackirregularities,perhapsthemostimportant excita-tion oftrain-track/bridge systems,holdrandom nature.InRefs.[16,18],onlytheexcitationofspecific trackirregularity spectrums, whicharejuststatisticalstatusoftheraildeformation atoneprobability,isconsidered,certainly,basedon whichthefullresponsesofsystemcomponentscannotberevealed;
• Thestrategiestocharacterizetheevolutionofsystemdynamiccharacteristicsaffectingthelong-termbehaviorsof train-track/bridgesystems.
Previously,XuandZhai[26]hadproposedatheoreticalprototypetoexpand thegeneralrandomvibrationanalysisinto temporal-spatial stochastic analysis withrespect to the vehicle-track systems, in which the randomness andcorrelation betweenlongitudinal system parameters are considered but being limited to normal distribution, moreover, the ergodic simulationoftrackrandomirregularitiesaresolvedbyaprobabilisticmethod.Thisresearchnoticesthefeaturesofrailway linesinlarge-scaleconstructionandlongitudinalunevenness/randomnessinsystemparametersandtrackirregularities,the temporal-spatial stochastic model is therefore established to wholly consider the possibility ofcombination over system parameters and excitations, and achieving far more comprehensive dynamics results that are necessities in reliability assessment,trackmaintenanceanddesign,etc.
Basedon thefundamentals ofthework inRef. [26], thestudies aimingat constructinga generalmodelforachieving the temporal-stochastic analysis of train-track-bridge systems will be further extended and perfected. The main content of this paper can be illustrated by: firstly, a more advanced 3-D nonlinear train-track-bridge dynamic model will be developed by fully modellingthe tracks andthe bridges asan integratedone, while the interactions between the train and the track-bridge systems are characterized by the complex nonlinear wheel-rail forces by correcting the deficiency ofRef.[18,27]; secondly,a unitedmethodwillbe providedto randomlysimulatethesystemvariables thatare correlated andfollowingarbitraryprobabilitydistribution, basedonMCMandKarhunen–Loève expansion(KLE);thirdly,a statistical strategyisdevelopedtodescribetheevolutionoftrackrandomirregularitiesaccompaniedbya simpleintroductionofthe trackirregularityprobabilistic model;fourthly, the theoretical framework forconstructing the temporal-spatial stochastic modelfortrain-track-bridgeinteractionsispresented;finally,numericalstudiesandconclusionswillbefurtherpresented.
2. Modellingfortrain-track-bridgeinteractions 2.1. Basic instructions
Tothethree-dimensional(3-D)train-track-bridgemodel(Fig.1),thefollowinginstructionsshouldbe madeinadvance, namely
(1) Thetrainconsistsofthefrontandrearmotorcarandseveraltrailercarsmodelledasrigidbodiesofacarbody,two bogieframe,fourwheelsetsandlinearsuspensionsystems,andthevehiclesmoveataconstantvelocity V alongthe tracks;
(2) Thebridgeismodelledassimplysupportedgirderbridges;thetracksrestingonthebridgesarechosenasslabtracks modelledbythin-plateelement,whilethetracksonrigid-subgradesaremodelledbycommonlyusedballastedtracks. (3) Onlythepresentationoverthedynamicequationsofmotionforthetrain-track-bridgeinteractionswillbeillustrated
Fig. 1. Three-dimensional train-track-bridge interaction model (end view). 2.2. Dynamic equations of motion for train-track-bridge interactions
Based onfundamentals offiniteelement theory andenergy-variational principle,one can derive the3-D equationsof motioninsub-matrixformforthetrain-trackbridgeinteractionsystemas
⎡
⎢
⎢
⎣
Mtt 0 0 0 0 0 Mrr 0 0 0 0 0 Mss 0 0 0 0 0 Mbb 0 0 0 0 0 Mpp⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎣
¨ Xt ¨ Xr ¨ Xs ¨ Xb ¨ Xp⎤
⎥
⎥
⎥
⎦
+⎡
⎢
⎢
⎣
Ctt Ctr 0 0 0 Crt Crr Crs 0 0 0 Crs Css Csb 0 0 0 Cbs Cbb Cbp 0 0 0 Cpb Cpp⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎢
⎣
˙ Xt ˙ Xr ˙ Xs ˙ Xb ˙ Xp⎤
⎥
⎥
⎥
⎦
+⎡
⎢
⎢
⎣
Ktt Ktr 0 0 0 Krt Krr Krs 0 0 0 Krs Kss Ksb 0 0 0 Kbs Kbb Kbp 0 0 0 Kpb Kpp⎤
⎥
⎥
⎦
⎡
⎢
⎢
⎣
Xt Xr Xs Xb Xp⎤
⎥
⎥
⎦
=⎡
⎢
⎢
⎣
Ft Fr Fs Fb Fp⎤
⎥
⎥
⎦
(1)wherethesubscripts“t”,“r”,“s”,“b” and“p” denotethetrain,rail,slabtrack,girderandpier,respectively. 2.2.1. The methodologies used to couple the train-track-bridge systems
Surveying this train-track-bridge coupled system, one can discover that the connection of all system components includingthewheel-railinterfacialinteractioncanbeequivalentlytreatedasaformofspring-dashpotcontact.
Obviously the dynamicproblems of elastic systems can be also transformed asstatic equilibrium problemswhere d’ Alembertprincipleandtheeffectsofdampingforcesisrespectivelyintroducedandconsideredbyageneralform[19]
fe= fρ+fc− Fsign
(
u)
+p(
t)
+G (2)with fρ=υ
(
−ρ
u¨)
dυ
; fc=υ(
−cu˙)
dυ
,wherefe,fρ,fc,F,p(t ) andGare thevectorsofsystemelastic force,inertiaforce,damping force,Coulomb’s frictionforce, external interfereforce andgravity, respectively; u¨, u˙ and u are respectivelythe acceleration,velocityanddisplacementquantity;
ρ
and c respectivelydenotetheparametersofmassdensityanddamping coefficient.Based on the principle of virtual work, the equation below can be derived by accounting for the system virtual displacement
δ
u,thatis[19]δ
Ue=− υδ
u Tρ
u¨dυ
− υδ
u Tcu˙dυ
−δ
uTFsign(
u)
+δ
uTp(
t)
+δ
uTG (3) withU ρ=υδ
uTρ
u¨dυ
, Uc=υ
δ
uTc u˙dυ
, V F=uTFsign(u), V p=−uTp(t ), V g=−uTG,inwhich U eistheelasticstrainenergyofforce, Coulomb’s friction force and external interfere force, respectively; V g is the gravitational potential energy of the
systems.
Followingtheprincipleoffiniteelementmethod(FEM),itisknownthatu= NX,u˙=NX˙ andu¨=NX¨,inwhichNisthe equivalentshapefunction, X,X˙ andX¨ are respectivelythe displacement,velocity andaccelerationvector ofthe systems, thusEq.(3)canbetransformedas
δ
Ue+δ
XTMX¨+δ
XTCX˙ =δ
XTFs (4)with M=υ
ρ
NTmNmd
υ
, C=υc NcTNcdυ
, Fs=−NTFFsign(
NFX)
− NTpp(
t)
− NTGG, where Nm and Nc denote the equivalentshapefunctionsforthemassmatrixanddampingmatrixrespectively;NF,NpandNGdenotetheequivalentshapefunctions
forloaddistribution;
δ
U eisthevariationoftheelasticstrainenergy,whichhasdifferentexpressionsaccordingtothemotionmodes,forexample,
δ
U e=δ
XTυEI NTNdυ
X¨ fortheelastic deformationenergyofa beam,inwhich E and I denotethemodulusofelasticityandmomentofinertiarespectively,N denotesthesecondderivativeoftheequivalentshapefunction. From Eq.(4)one can observe that thedynamicmatrices canbe obtained by removing thevirtual displacement term
δ
X, thus thekey work in buildingthe equationsof motionfora dynamical systemlies inclarifying theshape functions andloadingvectors. Thedetailmethodologyandderivation processfromthepotential energyofa dynamicsystemtothe correspondinglydynamicmatriceshavebeenpresentedintheworkofXuetal.[28],whichcanbeconsultedforreference. Inthefollowingparts,thedisplacementvectorsandsub-matriceswillbegivendirectlywithoutfurtherderivationsfor brevity.2.2.2. Displacement vector
Thetotaltraindisplacementvector,Xt,canbeassembledas
Xt =
XMo,1 XTr,1 XTr,2 ... XTr,Nv XMo,2
T(5)
wherethesuperscript“T” denotesthetranspositionofthematrix;thesubscripts“Mo”,“Tr” denotethemotorcarandtrailer carrespectively,andXMo,i, i =1,2andXTr,j, j =1,2,..., N vdenotethesub-vectorsofthe i thmotorcarandthe j thtrailercar,
allwith35degreesoffreedom(DOF’s),asshowninAppendixA-TableA1.
Thedisplacementvectoroftherail,Xr,withorderof(2× Nr× nr)× 1,where N risthetotalnumberofrailbeamelement,
n risthenumberofDOFsofarail beamandthenumber“2” originatesfromtheleft- andright-side oftherails,can be
writtenas
Xr=
XLr,1 XLr,2 ... XLr,nr XRr,1 XRr,2 ... XRr,nr T(6)
wherethesubscript“Lr” and“Rr” denotestheleft-andright-siderailrespectively.TheDOFforeachnodeoftherailbeam elementhasbeenlistedinAppendixA-TableA2.
Thedisplacementvectoroftheslabtrack,Xs,withorderof(N s× ns)× 1,where N sisthetotalnumberofslabelement,
n sisthenumberofDOFsofaslabthin-plateelement,canbewrittenas
Xs=
Xs,1 Xs,2 ... Xs,Ns
T(7)
wherethe Xs,l, l =1,2,..., N s,denotes the displacementvector of the l thslab trackelement.The DOFsfor thethin-plate
elementoftheslabtrackshavebeenlistedinAppendixA-TableA3.
The displacement vector of the girder, Xb, with orderof (N b× nb)× 1, where N b is thetotal number of bridge girder
elements, n b isthenumberofDOFsofagirderelement,canbewrittenas
Xb=
Xb,1 Xb,2 ... Xb,Nb
T(8)
wheretheXb,k, k =1,2,..., N bdenotesthedisplacementvectorofthe k thgirderelement.TheDOFsforthegirderelement
arethesameastherailbeamelement.
The displacement vector of the pier, Xp, with order of (N p× np)× 1, where N p is the total number of bridge girder
elements, n p isthenumberofDOFsofagirderelement,canbewrittenas
Xp=
Xp,1 Xp,2 ... Xp,Nb T(9)
wheretheXp,q, q =1,2,..., N pdenotesthedisplacementvectorofthe q thgirderelement.TheDOFsforthepierelementare
lineardisplacementsalong X -, Y -and Z -axissincebeingassumedasmasselement. 2.2.3. Sub-matrices for the train
Themassmatrixofthetrain,Mtt,canbeassembledas
Mtt = Nv Tr=1 MTr,c
(
Tr,c,
Tr,c
)
+MTr,Gq(
Tr,Gq,
Tr,Gq
)
+MTr,Gh(
Tr,Gh,
Tr,Gh
)
+ nwr i=1 MTr,w,i(
Tr,w,i,
Tr,w,i
)
+ 2 Mo=1 MMo,c(
Mo,c,
Mo,c
)
+MMo,Gq(
Mo,Gq,
Mo,Gq
)
+MMo,Gh
(
Mo,Gh,
Mo,Gh
)
+ nwr
i=1
MMo,w,i
(
Mo,w,i,
Mo,w,i
)
(10)with
u= yu zu
ψ
uβ
uφ
u , Mu(
u,
u
)
=diag(
[mu mu Iu,z Iu,y Iu,x ])
,w,g= yw,g zw,g
ψ
w,gβ
w,gφ
w,g , Mw,g(
w,g,
w,g
)
=diag(
[mw,g mw,g Iw,z,g Iw,y,g Iw,x,g ])
,where thesubscript “u” will be substitutedby “c”, “Gq” and“Gh” representing the carbody,front bogie frame andrear bogieframerespectively;thesubscript“w” denotesthewheelset,“g” denotesthe g thwheelset; m isthemass, I x, I yand I z
denotemomentofinertiaaround X -, Y -and Z -axis;
[•] isageneraloperatorindicatingtheDOFnumbercorresponding tothematrices.
Thestiffnessmatrixofthetraincanbederivedas
Ktt= Nv Tr=1 KTr+ 2 Mo=1 KMo (11)
ThedetailedexpressionofEq.(11)hasbeenelaboratedinAppendixB[29].Additionally,thedampingmatrixofthetrain,
Ctt,hasalmostthesameexpressionasKtt.Itisjustneededtosubstitutethestiffnesscoefficient k withdampingcoefficient c .
2.2.4. Sub-matrices for the rail
Themassmatrixoftherail,Mrr,canbewrittenas
Mrr= Nr h=1 2 g=1 mr
lr 0 NT XrNXrd
ξ
+ lr 0 NT YrNYrdξ
+ lr 0 NT ZrNZrdξ
+ lr 0 Wr Ar NT θXrNθXrdξ
(12) with NXr=NθXr= ... 1−ξ
/L ...ξ
/L ...1×n r, NYr= ... 1− 3ξL2+2ξL3 ξ L− 2 ξ L 2 +ξL3L 3ξL2− 2ξL3 ξL 3 −ξL2L ... 1×nr , NZr= ... 1− 3ξL 2 +2ξL3 −ξL − 2ξL2+ξL3L 3ξL2− 2ξL 3 −ξL 3 −ξL 2 L ... 1×nr ,where N Xr, N Yr, N Zr and N θXr arethe shapefunctionsfordisplacement along X -, Y -and Z -axisandanglearound X –axis;
g =1,2denotetheleftandrightsiderailalong X -axis; N risthetotalnumberofrailbeamelements;
ζ
isthelocaldistancefromtheleftnodeoftherailbeamtoarbitrarypointswithinthebeamelementalong X -axis. Thestiffnessmatrixoftherail,Krr,canbewrittenas
Krr= Nr h=1
⎡
⎣
2 g=1 ErAr lr 0 [N Xr] T [NXr ]dξ
+ErIry lr 0 [N Zr] T [NZr]dξ
+2 g=1 ErIrz lr 0 [N Yr] T [NYr]dξ
+krt lr 0 [N θXr] T [Nθ Xr]dξ
⎤
⎦
(13)where N and N arethefirstandsecondderivativeoftheshapefunctionrespectively. 2.2.5. Sub-matrices for the slab tracks
Themassmatrixoftheslabtrack,Mss,canbewrittenas
Mss= Ns h=1 mshs
(
bs/2 −bs/2 ls/2 −ls/2 NT ZsNZsdξ
dζ
+bs ls 0 NT YsNYsdξ
)
(14)where
ζ
is the local coordinatealong the Y -axis; N Ys has the sameexpression as N Yr, while N Zs is chosen asthe shapefunctionofarectangularthin-plateelement.
Thestiffnessmatrixoftheslabtrack,Kss,canbewrittenas
Kss= Ns h=1
ls 0 bs 0 BTDBdζ
dξ
+EsIsz ls 0 NZsTNZsdζ
] (15) with B=−∂
2N Zs∂
ξ
2 ;∂
2N Zs∂
ζ
2 ;∂
2N Zs∂
ξ∂
ζ
T ,D= Esh3s 12(
1−μ
s2)
1μ
s 0μ
s 1 0 0 0(
1−μ
s)
/2 ,2.2.6. Sub-matrices for the girder
Becausethegirderisassumedtobebeamelementastherails,thusitssub-matriceshavethesameexpressionsasthe rail.
Tothemassmatricesofthegirder,Mbb,onecanderiveitbysubstitutingthe“mr”,“Ar” and“Wr” inEq.(12)with“mb”,
“Ab” and“Wb”,namely
Mbb= Ns h=1 mb
lr 0 NT XrNXrd
ξ
+ lr 0 NT YrNYrdξ
+ lr 0 NT ZrNZrdξ
+ lr 0 Wb Ab NT θXrNθXrdξ
(16)Tothe stiffnessmatrix of thegirder, Kbb, it can be obtainedby substituting the “Er”, “Ar”, “Iry”, “Irz” and “krt” in Eq.
(13)with“Eb”,“Ab”,“Iby”,“Ibz” and“kbt”,namely
Kbb= Ns h=1
⎡
⎣
EbAb lr 0 [N Xr] T [NXr]dξ
+EbIby lr 0 [N Zr] T [NZr]dξ
+EbIbz lr 0 [N Yr] T [NYr]dξ
+kbt lr 0 [N θXr] T [Nθ Xr]dξ
⎤
⎦
(17)BasedontheassumptionofRayleighdamping,thedampingmatrixCbbcanbecomputedby
Cbb=
α
Mbb+β
Kbb (18)with
α
=2ζ
bω
1ω
2/(ω
1+ω
2)andβ
=2ζ
b/(ω
1+ω
2).Where
ζ
bisthedampingratio,ω
1 andω
2arethefirsttwocircularfrequenciesofvibrationofthegirder.2.2.7. Sub-matrices for the pier
The pier is simplified asmass element, the mass, stiffnessand damping matrix of which, Mpp, Kpp and Cpp, can be
writtenas Mpp= Np h=1 diag
(
[mp mp mp ])
(19) Kpp= Np h=1 diag(
[kpx kpy kpz ])
(20) Cpp= Np h=1 diag(
[cpx cpy cpz ])
(21)2.2.8. Sub-matrices for the rail-slab track interactions
Thematricesforrail-slabinteractionsaremarkedwiththesubscripts“rs” and“sr”,whicharederivedbytheinteractions betweentherailsandtheslabtracksconnectedbytherailpads.
Thestiffnessmatrix,K˜r−s,withorderof(2× Nr× nr+N s× ns)× (2× Nr× nr+N s× ns),canbewrittenas
˜ Kr−s= 2(Nr+1) h=1
⎡
⎣
krp,z[(
[NZr] T ξ=0− [NZs]Tξ=0,ζ=b)(
[NZr]ξ=0− [NZs]ξ=0,ζ=b)
+(
[NZr]Tξ=0− [NZs]Tξ=0,ζ=bs−b)(
[NZr]ξ=0− [NZs]ξ=0,ζ=bs−b)
] +krp,y[(
[NYr]Tξ=0− [NYs]Tξ=0)(
[NYr]ξ=0− [NYs]ξ=0)
]⎤
⎦
(22)ThedampingmatrixC˜r−s canbe obtainedby substitutingthe stiffnesscoefficients k rp,z and k rp,y inK˜r−swithdamping
coefficients c rp,z and c rp,y.
It should be noted that the matrices of stiffness and damping are actually assembled ones, which will be further partitionedas
¯ Krr Krs Ksr K¯ss,1 =K˜r−s, ¯ Crr Crs Csr C¯ss1 =C˜r−s, (23)ThusthestiffnessanddampingmatrixoftherailwillbereplacedbyKrr+K¯rr andCrr+C¯rr.
2.2.9. Sub-matrices for the slab track-girder interactions
Thematricesforslab track-girderinteractions aremarked withthesubscripts“sb” and“bs”,which arederived by the interactionsbetweentheslabtracksandthegirderconnectedbythecementasphaltmortar(CAM).
Thestiffnessmatrix,K˜s−b,withorderof(N s× ns+N b× nb)× (N s× ns+N b× nb),canbewrittenas
˜ Ks−b= Ns h=1
kca,z ls 0 bs 0 NT Z,sbNZ,sbdζ
dξ
+kca,y ls 0 NT Y,sbNY,sbdξ
(24)Fig. 2. Wheel-rail coupled model [24] . with NZ,sb=NZs− NZb, NZb=
NZrBr− bs 2 +
ζ
NθXr , NY,sb=NYs− NYb, NYb= NYr Hr NθXr ,The dampingmatrix,C˜s−b,has almostthe sameexpression asK˜s−b justby substitutingthe stiffnesscoefficients“kca,z”
and“kca,y” withdampingcoefficients“cca,z” and“cca,y” .
InthesamemannerasEq.(23),thematricesofstiffnessanddampingshouldbepartitionedas
¯ Kss,2 Ksb Kbs K¯bb,1 =K˜s−b; ¯ Css,2 Csb Cbs C¯bb,1 =C˜s−b, (25)ThusthestiffnessanddampingmatrixoftheslabwillbereplacedbyKss+K¯ss,1+K¯ss,2 andCss+C¯ss,1+C¯ss,2.
2.2.10. Sub-matrices for the Girder–Pier interactions
The matrices for girder-pier interactions are marked with the subscripts “bp” and “pb”, which are derived by the interactionsbetweenthegirderandthepierconnectedbythecementasphaltmortar(CAM).
Thestiffnessmatrix,K˜b−p,withorderof(N b× nb+N p× np)× (N b× nb+N p× np),canbewrittenas
˜ Kb−p= Np h=1
kp,zNZ,bpT NZ,bp+kp,yNY,bpT NY,bp (26) with N Z,bp=N Zs− NZp, N Zb=1; N Y,bp=N Yb− NYp, N Yp=1;Thedampingmatrix,C˜b−p,can beobtainedbyreplacing thestiffnesscoefficients“kp,z” and“kp,y” in thecorresponding
stiffnessmatrixwithdampingcoefficients“cp,z” and“cp,y” .
Thematricesofstiffnessanddampingshouldbepartitionedas
¯ Kbb,2 Kbp Kpb K¯pp,1 =K˜b−p; ¯ Cbb,2 Cbp Cpb C¯pp,1 =C˜b−p, (27)Accordingly, thestiffness anddampingmatrixof thegirder willbe assembled asKbb+K¯bb,1+K¯bb,2 and Cbb+C¯bb,1+C¯bb,2, andthestiffnessanddampingmatrixofthepierwillbeobtainedbyKpp+K¯pp,1 and Cpp+C¯pp,1.
2.2.11. Load vectors
The wheel-rail contact geometries are almost the most important and complex part in the whole train-track-bridge dynamicmodel.Inthewheel-railcoupledmodel[30],theyaw,pitch,roll,transverseandbouncemotionsofthewheelset andthetransverse,bounceandtorsionalmotionsoftherailareconsideredwithacomprehensiveway,asshowninFig.2, thewheel-railcontactswillthereforebetreatedasathree-dimensional(3-D)nonlinearandasymmetryprobleminevitably.
Thedynamicloadactingonthewheelsetisderived bythework oftheforce/momentontransverse,bounce,roll,yaw andpitchmotionsofthewheelsetrespectively,thatis[24],
Fw= 2+Nv j=1 4 i=1 Fw,i (28) with Fw,i=
⎡
⎢
⎢
⎣
Fw,Y i Fw,Zi Fw,φi Fw,ψi Fw,βi⎤
⎥
⎥
⎦
=⎡
⎢
⎢
⎢
⎣
FYli+FYri+Mw¯gϕ
i− MwV2/Ri−Mwr0ϕ
¨wi FZli+FZri+Mw¯g+Mwa0ϕ
¨wi+MwV2/Riϕ
wi a0(
FZri− FZli)
− rliFYli− rriFYri+IwY(
β
˙wi−)(
ψ
˙wi+V/Ri)
−IwXϕ
¨wi a0(
FXli− FXri)
− a0ψ
wi(
FYli− FYri)
+MZli+MZri+IwY(
β
˙wi−)(
φ
˙wi+ϕ
˙wi)
−IwZVλ
˙wi rliFXli+rriFXri+ψ
wi(
rliFYli+rriFYri)
+MYli+MYri⎤
⎥
⎥
⎥
⎦
,wherethesubscript“i”, i =1,2,3,4,denotesthe i thwheelset; F Xliand F Xridenotethelongitudinalforcesactingonthe
left-andright-sideofwheel/railcontactinterface,respectively; F Yliand F Yridenotethetransverseforcesactingontheleft-and right-sideofwheel/railcontactinterface,respectively; F Zliand F Zridenotetheverticalforcesactingontheleft-and
right-sideofthewheel/railcontactinterface,respectively; M Zliand M Zriarethemomentofforcearound Z -axis,respectively; M Yli
and M Yriarethemomentofforce around Y -axis,respectively;
β
˙wi,ψ
˙wiandφ
˙wi denotethepitch,yawandrollingvelocity ofthe i thwheelset;ψ
wiistheyawangle.Theforcesactingonthecentroidofmassofthecarbodyaremainlyderivedfromtheworkofforce/momentatmotions oftransverse,bounce,rollingandyawwhenthevehiclepassesthroughthecurvedtracks,namely
Fc= 2+Nv i=1 Fc,i (29) with Fc,i=
⎡
⎢
⎣
Fc,Y i Fc,Zi Fc,φi Fc,ψi⎤
⎥
⎦
=⎡
⎢
⎣
Mcgϕ
c−McV2/Rc−Mc(
r0+Htw+Hbt+Hcb)
ϕ
¨c Mcg+McV2/Rcϕ
c+Mca0ϕ
¨c −IcXϕ
¨c −IcZVς
˙c⎤
⎥
⎦
,Theforces actingon the centroidof massofthe bogie framesare mainly derived fromthe work offorce/momentat motionsoftransverse,bounce,rollingandyawwhenthevehiclepassesthroughthecurvedtracks,namely
Fb= 2+Nv i=1 2 k=1 Fk (30) with
⎡
⎢
⎣
Fk,Y Fk,Z Fk,φ Fk,ψ⎤
⎥
⎦
=⎡
⎢
⎣
Mkgϕ
k−MkV2/Rk−Mk(
r0+Htw)
ϕ
¨k Mkg+MkV2/Rkϕ
k+Mka0ϕ
¨k −IkXϕ
¨k −IkZVς
˙k⎤
⎥
⎦
,where k =1and2representthefrontbogieframe,“Gq”,andrearbogieframe,“Gh”,respectively.
Withregard tothetracksystems,rail isthecomponent thatdirectly bearsthedynamicloadsfromthe train,andthe dynamicforceactingonthetransverse,verticalandtorsionaldisplacementcanbeexpressedas
Fr= 2+Nv i=1 4 j=1 2 l=1 Fk,l, j (31) with Fk,l, j=
F Y,l j FZ,l j MT,l j = F YlNYr, j FZlNZr, j(
FZllhi+FYllvi)
NθXT, j ,wherethesubscript “l” denotestheleft orrightsideofthe rail;thesubscript “j” denotesthe j thwheel-railcontactpair; F Y,lj, F Z,ljand M T,ljdenotetheequivalentlateral,verticalandtorsionalforcesactingonthecentroidoftherail.
Accordingtoabovepresentations,onecanobtainedthatFt=Fc+Fb+Fw,Fs=0,Fb=0,Fp=0.
Tillnow,thedynamicsub-matricesandloadingvectorsinEq.(1)havebeenfullyrevealedtoeffectivelycouplethetrain, trackandbridgesystems asan entiresystem. Byadoptingnumericalintegration methods,thesystem responsesof train, trackandbridgecanbeobtainedsimultaneously.
Fig. 3. Comparisons with the model of Zhai et al. [1] on car body accelerations and the vibrations at the mid-span of the bridge (a. lateral acceleration of the car body; b. vertical acceleration of the car body; c. lateral displacement of the girder at the mid-span; d. vertical displacement of the girder at the mid-span).
2.2.12. Comparison with general solutions
For validating the practicality of this dynamical model compiled in a computer program, a comparison to a three-dimensionalmodelbyZhaietal.[10]isconducted. Itisassumedthatanentiretraingroupedbyfivevehiclesrunsonthe bridge(five-spansimplysupportedbeam-bridge)withaconstantvelocity(300km/h).
Fig.3presentsthe comparisons on the dynamic responses of the bridge at thesecond mid-span and car body accelera-tions,fromwhichonecanobservethatthedynamicresultsrespectivelyderivedfromthesetwomodelscoinciderelatively well witheach other.Theaccuratenessandpracticalityofthismodelisthereforeillustratedclearly.Howeverit shouldbe notedthat thereinevitablyexist slightdeviations betweentheresults becauseofthecompletelydifferentmethodsinthe modelconstructions,seeforexample:thecoupleranddraftsystembetweenthecarbodiesofvehicles ignoredin[10]are consideredeffectivelyinthismodel;inthismodel,thetrackandthebridgehavebeenunitedanentiresystem,whilethey aresolvedseparatelyin[10].
3. Characterizationofthetemporal-spatialstochasticityforthecoupledsystem
The model constructed in Section 2 is developed to characterize the train-track-bridge interactions with respect to deterministic systemparameters andexcitations, e.g.,track randomirregularities. Ithasbeen pointedout previously that thetrain-track-bridgesystemsareessentiallyrandominnature,especiallythetrack-bridgesystem,whichholdssignificantly inhomogeneous characteristics in the spatial domain, moreover, the random field of dynamic parameters is constantly evolvedcorrespondingtothecyclictrain-track-bridgeinteractions.
3.1. Joint simulation based on Monte-Carlo Method (MCM) and Karhunen-Loève Expansion (KLE) 3.1.1. Random field representation
Byneglecting thestochasticity ofthesystemcomponents in Y direction, therandom variables oftrack-bridgesystems canbeexpressedas
(x, z, t ),where x representsthelongitudinalabscissaofthetracks, z representstheverticalcoordinate and t representsthetimeprocess.Thustheunifiedfieldofsystemstructurescanbeexpressedby
T=T
((
x,z,t))
,∈R3 (32)
where R 3 representsthethree-dimensionalEuclideanspace,
representsthe definitiondomainofthe randomfield, and
the variable z is mainly applied to express the vertically layered characteristics of track and bridge structures, namely differentphysicalandmechanicalparametersofstructures.
Withrespecttothedeterminedstructuralparameters z υ,
υ
=1, 2, ..., M and thetime point t k, k =1,2,..., N ,thespatialdistributionof
(x, z, t ),whichisregardedtofollowacertaintypeofprobabilitydistribution,canbecharacterizedby
whereP(•)representsthepossibleprobability distributiontype,andV ˜ representsthecorresponding characteristic param-eters,e.g.,thestatisticalmeanandstandarddeviationofnormaldistribution,thedegreeofchisquaredistribution,etc.
Generally,thelongitudinalcoordinateofthe raillinerepresentedby parameter x will bewithahuge magnitudeonce considering the whole bridge sections existing in a railway line. Therefore, a longitudinal division towards the random spaceisconductedbyconsideringthelimitlengthofnumericalcalculations,namely
x=∩xl,xl∈
(
0,Stot/N] (34)where S totrepresentsthetotalrailwaylength, N representsthenumberofdivision,and l =1,2,..., N .
Insummary,therandomfieldoftrack-bridgesystemscanbefurthertransformedfromEq.(32)as
T=T
((
xl,zυ,tκ))
(35)3.1.2. Monte-Carlo simulation
Innumericalstudies, therandomfieldrepresentedby Eq.(35)isconventionallymodelledby MCMthat isperhapsthe mostrobust approachinrandom analysis.Foracertaintime point t κ,thespacevectorinEq.(35)canbe expoundedasa two-dimensionalmatrixform
T=
⎛
⎜
⎜
⎜
⎜
⎜
⎝
T(
u(
x1,z1,tκ))
T(
u(
x2,z1,tκ))
· · · T(
u(
xN,z1,tκ))
. . . ... ... ... T(
u(
x1,z1×H,tκ))
T(
u(
x2,z1×H,tκ))
· · · T(
u(
xN,z1×H,tκ))
. . . ... ... ... T(
u(
x1,zM×H,tκ))
T(
u(
x2,zM×H,tκ))
· · · T(
u(
xN,zM×H,tκ))
⎞
⎟
⎟
⎟
⎟
⎟
⎠
(M×H)×N (36)wherethesubscript“H” denotesthenumberforaspecificvariableinonesample.
Tothe actualrandom variables z υ,which areassumed tofollow an alreadyknownprobability densityfunction(PDF) f Z
(
z υ)
,thecumulativedistributionfunction(CDF),amonotonicfunction,canbederivedbyFZ
(
zυ)
=zυ
−∞fZ
(
zυ)
dzυ,υ
=1,2,...,M, (37)Thecorrelationmatrixbetween z υ canbeexpressedby
RM×M=
⎛
⎜
⎜
⎝
1 r12 ... r1M r21 1 ... r2M . . . ... ... ... rM1 rM2 ... 1⎞
⎟
⎟
⎠
(38)Forobtainingrandom seriesof z υ that satisfiesEqs.(37) and(38),based onalinear-nonlinear transformationmethod [31],thestepsforachievingthisgoalcanbefollowedby
(1)Let g υ bethevariablefollowingGaussiandistribution,thatis,
fG
(
gυ)
= 1 2πσ
gυ exp−
(
z−μ
gυ)
2 2σ
2 gυ (39)where
μ
gυ andσ
gυ respectivelyrepresentthestatisticalaverageandstandarddeviationwithrespecttothevariable g υ.Forsimplifyingthetransformation,itispracticaltoset
μ
gυ =0andσ
gυ =1.InthesamemannerasEq.(37),theCDFof g υcanbeobtainedby
FG
(
gυ)
=gi
−∞fG
(
gυ)
dgυ (40)BasedonEqs.(37)and(40),thenonlineartransformationbetween x υ and g υ canbeestimatedas
xυ=¯hυ
(
gυ)
=FZ−1(
FG(
gυ))
(41)where
(•)denotesthenonlineartransformationoperatorforargument(•); F -1(•)denotestheinversefunction.(2)Derivingthecorrelationcoefficient r gigj through r xixj, i, j =1,2,..., M, i = j ,thatis[30,31],
rxixj = E
(
xi−μ
xi)(
xj−μ
xj)
σ
xiσ
xj = ∞ −∞ ∞ −∞(
i(
gi)
−μ
xi)
σ
xi(
j(
gj)
−μ
xj)
σ
xj fGG(
gi,gj)
dgidgj (42) withFig. 4. PDF comparison between the results of MCM and MCM-KLE. fGG
(
gi,gj)
= 1 2π
1− r2 gigj exp −(
g 2 i− 2rgigjgigj+g 2 j)
2(
1− r2 gigj)
;μ
x=E(
x)
;σ
x=E(
x2)
− [E(
x)
]2,inEq.(42), E denotesthemathematicalexpectation, andtherelationshipbetween r gigj and r xixj hasbeenconnected
defi-nitely.ForsolvingtheintegrationofEq.(42),Newton–Raphsoniterationmethodisapplied[32],andthen,thecorrelation matrixR ˜M×Mwithregardto g icanbeconfirmedaccordingly.
(3)UsingMonte-Carlomethod,therandomvectorof g icanbeassembledas
(
G=(
G1,G2,...,GM)
L×M)
(43)Andthen,theCholeskydecompositionisconductedonR ˜M×M,andtheGwillbeupdatedby
G=GL×MCr,M×M (44)
withR ˜M×M=C gTC g,theupdatedrandomvectorGwillsatisfythecorrelationmatrixR ˜M×M.
(4)Finally,throughthenonlinear transformation,thestandardized randomvector Gcan betransformed intoarbitrary correlatedpseudorandomvariables,namelythespacevectoroftherandomfiledby
T=¯h
(
G)
(45)3.1.3. Dimension reduction based on Karhunen–Loève Expansion (KLE)
MCMisgenerallyaccompaniedbyratherlow computationalefficiencyandconvergenceintherandomanalysis.Inthe currentwork,Karhunen–LoèveExpansion(KLE)isfurtherappliedtoachieve adimensionalreductionfortherandom filed oftrack-bridgesystems.
BasedonthedefinitionofKLE[33],therandomfieldoftrackstructurescanbefurtherexpressedby
T≈ E[T]+ Ns i=1
λ
iuiζ
i (46)with E[
ζ
iζ
j]=δ
i j, whereδ
ij is the Kronecker delta, E[· ] denotes the expectation operation; E[ T] is the mean of thestochasticprocess,
ζ
iareorthonormalrandomvariables.Thesetofdeterministicfunctions
λ
ianduiareeigenvaluesandeigenvectors,respectively,withsatisfyingthefollowing equation,namelyCXX
μ
=λμ
(47)where CXX represents an bounded, symmetric and positive definite auto-covariance function of T. The projection basis
{
μ
1,μ
2,...,μ
Ns}
ischosenorthonormalinthesensethat,forallγ
and:(
μ
γ)
Tμ
=δ
γ(48)
TheKLEiscapableofcapturingthespatialcorrelationofthefields,guaranteeingtherandomicityofsignalsandgreatly reducingthecalculationsamplesby N s N.
FormoredetailsaboutKLE,Refs.[34,35]canbeconsultedforreferences.
Forvalidatingthe effectivenessofthe MCM-KLEmethod,Fig. 4plots aPDF comparisonbetweenthe randomsamples respectivelysimulatedby MCMandMCM-KLEwithregard totherandomvariable,i.e.,lateralstiffnessoffastener.Asseen fromFig.4,thestatisticalpropertiesforthesimulatingresultsofthesetwomethodcoincidewellwitheachother,however, thereare 2000samplesintheMCM, butbyapplyingMCM-KLE,the samplingnumberis reducedto174,thusthe united MCM-KLEmethodgreatlyincreasesthecomputationalefficiency.
3.2. Track irregularity probabilistic model
Xu and Zhai [36] proposed a track irregularity probabilistic model (TIPM) to select representative and realistic track irregularitysetsforahigh-efficientcharacterizationofthestatisticalpropertiesoftrackgeometries.
Theabbreviatedproceduresareillustratedas
1) DividethemeasuredtrackirregularitiesI(s )into N segments,namely
I
(
s)
=∪(
I(
s˜k))
(49)where s denotesthedistancealongthetrack, k =1,2,..., N ,s ˜k∈
(
S tot(
k − 1)
/N,S totk/N],2) Set (•)asthepowerspectraldensity(PSD)operator,andthePSDofI
(
s ˜k)
canbeobtainedby P k(
ω
)
=(
I(
s ˜k))
;3) Withaseriesofderivations,theprobabilitydensityfunction(PDF)of P k(
ω
)denotedby(P k(
ω
))canbederived.4) Basedontime-frequencytransformationapproach[35],theequivalenttimedomaintrackirregularitiesI˜
(
s ˜k)
usedasthe excitinginputsoftheinteractionmodelcanbeperformedon P (k ),andassuming(
P k(
ω
))
=(
I˜(
s ˜k))
.5) Withcognitionoftheprobabilitydistributionof P k(
ω
),itisratherconvenienttoselectrepresentativePSDfunction,i.e.,P r(
ω
),inwhich r istheselectedsampleofPSD, r =1,2,...,N ˜,usingrandomsimulationmethods.Generally,itisobviousthat N ˜ N,whilethestatisticalpropertiesofI(s ) inamplitudesandfrequenciesare definedwithoutinformationloss, andbeingcontainedin P r(
ω
).3.3. Temporal evolution of system dynamic characteristics
Astotheevolutionofsystemparameters,theexperimentalstudiesinsituorlabhaverarelybeenreportedyet,thusthe formulaicexpressions usedto program thevariation ofparameters againsttime cannot be revealedin thepresentstudy, and accordingly, the coefficients of variation (COV) are simply adopted to characterize the evolution of random system parameters,asshowninRef.[26].
Tothetrackirregularities,trackmaintenancedepartmentwillperiodicallydetecttherailprofiledeformationsusingtrack inspection car. The trackirregularity probabilistic model mentioned inlast partis developed to extract the realistic and representativesamplesfromthemassivetrackirregularitydata,basedonwhichthestrategyonstatisticallycharacterizing thetemporalevolution oftrackrandomirregularities canbe furtherachievedby equivalentlytransforming theamplitude evolutionoftrackirregularitiesintoitspowerspectraldensityevolution,thedetailedprocedurescanbefollowedasbelow 1) Thepowerspectrumsoftrackirregularitiesattime t iand t i+1 denotedbyPς,ti
(
ω
)
andPς,ti+1(
ω
)
respectively,wherethesubscript “
ζ
” denotesthetrackirregularitytype,aresummarily obtainedby thespectral estimationmethod,indicating that theamplitude-frequencypropertiesofIς,ti(
s)
willbeevolvedfromPς,ti(
ω
)
toPς,ti+1(
ω
)
duringthetimeperiodof(t i,t i+1],thusthespectralevolutioncanbefiguredoutby
˜
Pς,(ti,ti+1 ]
(
ω
)
=Pς,ti+1(
ω
)
− Pς,ti(
ω
)
(50)2) Fortheanalyticalconvenience,aspectraldensitymatrixwithorderof E × Wcanbeassembledby
ς,i
(
q,ω
)
=∪Pς,ti,q(
ω
d)
E×W,q=1,2,...,E, (51)
wherethesubscript"i "and"q "denotesthetimeintervalofmeasurement(t i,t i+1]andthe q thtrackirregularitypower spectrumwithrespecttothe q thtrackportion, E isthetotalnumberofpowerspectrumsandthesubscript“W” denotes thetotalnumberoffrequencypoints.Inthesamemanner,aspectraldensityevolutionmatrix
˜ς,i
(
q,ω
)
canbeobtained by[ς,i+1(q ,
ω
)−ς,i(q ,
ω
)].3) Obviously, with large-scale and long-term measurements, namely i =1, 2, ..., T, T is a relatively large value, a com-pletely mapping relationbetween
ς,i(q ,
ω
) and˜ς,i
(
q,ω
)
can be numericallymatched, whereς
(q ,ω
)=∪ς,i(q ,
ω
)and
ς
˜(
q,ω
)
=∪ς
˜,i
(
q,ω
)
.4) LetPς,l(
ω
d)=min[ς
(q ,ω
d)]andPς,u(ω
d)=max[ς
(q ,ω
d)],wheremin[•]andmax[•]denotetheminimumandmax-imumoperators respectively.Because thespectraldensities arediscretely distributed,the amplitudedomain ofPς(
ω
d)canbedividedinto
Pς,g
(
ω
d)
∈[Pς,l(
ω
d)
+gP˜(
ω
d)
,Pς,l(
ω
d)
+(
g+1)
P˜(
ω
d)
] (52)withP˜
(
ω
d)
=int[Pς,u(ωd)−Pς,l(ωd)Z ],0≤ g≤ Z-1,
whereP˜isthediscreteintervalofspectraldensities; Z isthetotalpartitionnumber;int[·]isanoperatorusedtoobtain thelargestintegerbeingsmallerthanthenumberinthebracket;thesubscript "
ς
,g"denotesthefield ofspectraldensitiessatisfyingEq.(52).
5) It is certain that the spectral densities Pς,g
(
ω
d)
included inthe field ofς
,g will correspond to different evolutionvalues, P˜ς,g
(
ω
d)
∈ς
˜(
ω
d)
. Based on probabilistic statistics, the probability density function (PDF) of the evolutionsdenotedby f ς,ωd
(
P˜ς,g)
canbedeterminedbyP˜ς,g(
ω
d)
.6) Itisregardedthattheevolutionofthespectraldensitieswithin
ς
,gwillfollowthesamePDFcharacteristics.Therefore,acumulativeprobabilityindex p ς ∈(0,1)isintroducedtouniquelyconfirmtheevolutionvalueofspectraldensityover
Pς,ti
(
ω
)
,namely ˜ Pti(
ω
; pς)
=F −1 ς,ωd(
pς)
(53) with F ς,ωd(
P˜ς,g)
= P˜ς,g −∞ f ς,ωd(
P˜ς,g)
dP˜ς,g,Fig. 5. Modelling framework for train-track-bridge stochastic model. where F ς,ωd
(
·)
istheCDFof f ς,ωd(
·)
.7) OnthebasisofEq.(53),theevolutionspectrumPt
i+1
(
ω
; pς)
canbeobtainedbyPti+1
(
ω
; pς)
=Pς,ti(
ω
)
+P˜ti(
ω
; pς)
(54)8) Itshouldbenotedthat Pt
i+1
(
ω
; pς)
,whichisdeterminedby theprobabilityindex p ς andevolution domainP˜ς,g(
ω
d)
, isanuncertainquantity,thusitcanbeselectedbyrandomsimulationmethods.By implementingthe time-frequency transformation process [36], the evolution spectrum Pt
i+1
(
ω
; pς)
in Eq.(54) can be equivalentlytransformed intotime-domain track irregularitiesIti+1
(
s)
. Byloading Iti+1(
s)
intothe dynamicmodel,the dynamicperformanceoftrain-track-bridgesystemscanbeassessedinlightofdifferentevolutionarystatusofexcitations. 4. Constructionofthetemporal-spatialstochasticmodelTill now, the work related to the establishment of the dynamic model, random simulation and characterization of the systemparameters andexcitationsin temporal-spatial domain have all been accomplishedproperly. The essencefor constructingthetrain-track-bridgetemporal-spatialstochasticmodellaysonunitingtherandomphysical-mechanicalstatus ofthesystemsandthedynamicalinteractionmechanisms.
Summarily,themodellingframeworkcanbeillustratedasFig.5 5. Casestudies
Inthenumericalstudies,it isassumedthatthereare twomotorcarsinthe frontandrearpartofthetrain,andwith three trailer cars containedin the middle,the relatedparameters havebeen presented inRef. [18]; besides, a fivespan simplesupportbeambridge isconstructedby 32-meterbeamelements,therelateddynamicparametersofthebridgeare listedinRef.[18].
5.1. Case 1: characteristics of system responses for train-track-bridge interactions
Ratherdifferent fromcommon infrastructural systems,the railwaybridges normally assumedasbeam elementsoffer moreflexiblesupportstothetrain-tracksystemsingeneral,thusthedynamicresponseoftrain-track-bridgesystemsmight possessuniquecharacteristics.
Fig. 6 plots the time-varyingrail displacements by tracking the wheel-railcontact points. It can be seen fromFig. 7 that the time processesof rail displacements show significantly non-stationaryandasymmetry characteristics; when the wheelsetsarriveatthemid-spanofthebridge,theverticaldynamicdeflections D zrofthetrack-bridgestructures reachthe
maximumresponses; astothe raillateral displacement D yr,though, withoutthat significantly asymmetry feature as D zr,
wecanstillobservethatthe D yrhasobviouslydeviatedfromthenormalcentralline.
Moreover, Fig.7 illustrates the representative responses ofthe displacement andthe accelerationwithspecific to the point atthemid-spanofthebridge girder,fromwhich onecan observethat whenthefirst motorcarapproachestothe specific point,theverticaldeflectionwillbe graduallyincreasedtoreachthefirst peakwithloadingofahalfweightofa car,andthenthedeflectionwillreachthemaximumafterafullweightofthecaractingonthegirderofthebridge,when thetrain gradually passesthrough themid-span, thedeflectionofthebridge structureswill accordinglydecayto original