Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
ContentslistsavailableatSciVerseScienceDirect
Wear
j o ur n a l ho me p a g e :w w w . e l s e v i e r . c o m / l o c a t e / w e a r
Development
of
a
wear
model
for
the
prediction
of
wheel
and
rail
profile
evolution
in
railway
systems
M.
Ignesti
a,
M.
Malvezzi
b,
L.
Marini
a,
E.
Meli
a,∗,
A.
Rindi
a aDepartmentofEnergyEngineering,UniversityofFlorence,ViaS.Martan.3,50139Firenze,Italy bDepartmentofInformationEngineering,UniversityofSiena,ViaRoman.56,53100Siena,Italya
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received25May2011 Receivedinrevisedform 22December2011 Accepted3January2012 Available online 24 January 2012 Keywords:
Multibodymodeling Wheel–railcontact Wheel–railwear
a
b
s
t
r
a
c
t
Thepredictionofthewearatthewheel–railinterfaceisafundamentalproblemintherailwayfield, mainlycorrelatedtotheplanningofmaintenanceinterventions,vehiclestabilityandthepossibilityof researchingspecificstrategiesforthewheelandrailprofileoptimization.InthisworktheAuthorspresent amodelspecificallydevelopedfortheevaluationofthewheelandrailprofileevolutionduetowear, whoselayoutismadeupoftwomutuallyinteractivebutseparateunits:avehiclemodelforthedynamic analysisandamodelforthewearestimation.Thefirstoneismadeupoftwopartsthatinteractonline duringthedynamicsimulations:a3DmultibodymodeloftherailwayvehicleimplementedinSimpack Rail(acommercialsoftwarefortheanalysisofmultibodysystems)andaninnovative3Dglobalcontact model(developedbytheAuthorsinpreviousworks)forthedetectionofthecontactpointsbetweenwheel andrailandforthecalculationoftheforcesinthecontactpatches(implementedinC/C++environment). Thewearmodel,implementedintheMatlabenvironment,ismainlybasedonexperimentalrelationships foundinliteraturebetweentheremovedmaterialandtheenergydissipatedbyfrictionatthecontact. Itstartsfromtheoutputsofthedynamicsimulations(positionofcontactpoints,contactforcesand globalcreepages)andcalculatesthepressuresinsidethecontactpatchesthroughalocalcontactmodel (FASTSIMalgorithm);thenthematerialremovedduetowearisevaluatedandthewornprofilesofwheel andrailareobtained.Thisapproachallowstheevaluationofboththequantityofremovedmaterialand itsdistributionalongthewheelandrailprofilesinordertoanalyzethedevelopmentoftheprofilesshape duringtheirlifetime.
Thewholemodelisbasedonadiscreteprocess:eachdiscretestepconsistsinonedynamicsimulation andoneprofileupdatebymeansofthewearmodelwhile,withinthediscretestep,theprofilesare supposedtobeconstant.Thechoiceofanappropriatestepisfundamentalintermsofprecisionand computationalload.Moreoverthedifferenttimescalescharacterizingthewheelandrailwearevolution requirethedevelopmentofasuitablestrategyfortheprofileupdate:thestrategyproposedbytheAuthors isbasedbothonthetotaldistancetraveledbytheconsideredvehicleandonthetotaltonnageburdenon thetrack.TheentiremodelhasbeendevelopedandvalidatedincollaborationwithTrenitaliaS.p.A.and ReteFerroviariaItaliana(RFI),whichhaveprovidedthetechnicaldocumentationandtheexperimental resultsrelatingtosometestsperformedwiththevehicleDMUAln501MinuettoontheAosta-PreSaint Didierline.
© 2012 Elsevier B.V. All rights reserved.
1. Introduction
Thewearatthewheel–railinterfaceisanimportantproblem
intherailwayfield.Theevolutionoftheprofileshapeduetowear
hasadeepeffectonthevehicledynamicsandonitsrunning
stabil-ity,leadingtoperformancevariationsbothinnegotiatingcurves
and instraight track. Thereforethe originalprofiles have tobe
∗ Correspondingauthor.
E-mailaddresses:ignesti@mapp1.de.unifi.it(M.Ignesti),malvezzi@dii.unisi.it
(M.Malvezzi),marini@mapp1.de.unifi.it(L.Marini),meli@mapp1.de.unifi.it
(E.Meli),rindi@mapp1.de.unifi.it(A.Rindi).
periodicallyre-establishedbymeansofturning:particularly,from
asafetyviewpoint,thearisingofacontactgeometrywhichmay
compromisethevehiclestabilityorincreasethederailmentriskhas
tobeavoided.Areliablewearmodelcanalsobeusedtooptimize
theoriginalprofilesofwheelandrailandtoobtainamoreuniform
wear.Inthiswaytheoverallamountofremovedmaterialcanbe
reducedinordertoincreasethemeantimebetweentwo
mainte-nanceintervalsand,atthesametime,thedynamicalperformance
ofthewheel–railpaircanbekeptapproximatelyconstantbetween
twoturnings.
Becauseofallthesereasons,thedevelopmentofamathematical
modelforthepredictionofthewearatthewheel–railinterface
representsapowerfultool.
0043-1648/$–seefrontmatter © 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.wear.2012.01.020
Fig.1.Generalarchitectureofthemodel.
InthisworktheAuthorswillpresentaproceduretoestimatethe
evolutionofthewheelandrailprofileduetowearbasedonamodel
thatcombinesmultibodyandwearmodeling.More specifically,
thegenerallayoutofthemodelconsistsoftwo mutually
inter-activeparts:thevehiclemodel(multibodymodeland3Dglobal
contactmodel)andthewearmodel(localcontactmodelandwear
evaluation and profiles update). Concerning thevehicle model,
themultibodymodel,implementedintheSimpackRail
environ-ment,accuratelyreproducesthedynamicsofthevehicle,taking
intoaccountallthesignificantdegreesoffreedom.The3Dglobal
contactmodel,developedbytheAuthorsinpreviousworks[1,2],
detectsthewheel–railcontactpointsbymeansofaninnovative
algorithmbasedonsuitablesemi-analyticprocedures andthen,
foreachcontactpoint,calculatesthecontactforcesthroughHertz’s
andKalker’stheory[3–5].Thankstothenumericalefficiencyofthe
newcontactmodel,thetwomodelsinteractdirectlyonlineduring
thesimulationofthevehicledynamics.
Asregardsthewearestimation,themodelisbasedonalocal
contactmodel(inthiscasetheKalker’sFASTSIMalgorithm)and
onanexperimentalrelationshipforthecalculationoftheremoved
material[6,7].Thewearmodel,startingfromtheoutputsofthe
vehiclemodel(contactpoints,contactforcesandglobalcreepages),
calculatesthetotalamountofremovedmaterialduetowearand
itsdistributionalongthewheelandrailprofiles.Theremovalofthe
materialiscarriedoutconsideringthethreedimensionalstructure
ofthecontactbodiesandthedifferenttimescalescharacterizing
thewearevolutiononthewheelandontherail.
Oneofthemostcriticalaspectsinthedevelopmentofawear
modelistheavailabilityofexperimentaldataforthevalidationof
themodel,becausethewearisalong-termphenomenonwhich
requiresseveralmonthsofmonitoringtocollectthedata.Ifonline
experimentalmeasurementcannotbecarriedout,theproblemcan
beovercomeusingtoolsprovidedbysoftware[8]orcarryingout
experimentalproofsonascaledtestrig[6].
Inthisworktheentiremodelhasbeenvalidatedbymeansof
theexperimentaldataprovided byTrenitaliaS.p.A.andRFI;the
dataconcerntheAosta-PreSaintDidierrailwaylineandthe
vehi-cleALSTOMDMUAln501Minuettowhich,inthisscenery,exhibits
seriousproblemsintermsofwear.
2. Generalarchitectureofthemodel
Thewholemodelconsistsoftwodifferentparts:thevehicle
modelandthewearmodel.Thegeneralarchitectureisshownin
theblockdiagraminFig.1.
Thevehiclemodelrepresentsthedynamicanalysisblockandis
composedbythemultibodymodelof thestudiedrailway
vehi-cle(inthisworktheALSTOMDMUAln501Minuetto)andthe3D
globalcontactmodelthat,duringthedynamicsimulation,interact
directlyonlinecreatingaloop.Ateachtimeintegrationstepthe
firstoneevaluatesthekinematicvariables (position,orientation
andtheirderivatives)relativetothewheelsetandconsequentlyto
eachwheel-railcontactpair.Startingfromthesequantities,the
secondonecalculatestheglobalcontactvariables(contactpoints
and contactforces, contactareasand globalcreepages).The 3D
globalcontactmodelisbasedbothonaninnovativealgorithmfor
thedetectionofthecontactpoints(developedbytheAuthorsin
previousworks[1,2])andonHertz’sandKalker’sglobaltheories
fortheevaluationofthecontactforces[3].Theglobalcontact
vari-ablesarethenpassedtothemultibodymodelinordertocarryon
thesimulationofthevehicledynamics.
The main inputs of the dynamic analysis block are the
rail-waytrackandthemultibodymodeloftheconsideredvehicle:in
thisresearchactivity,accordingtothespecificationsrequiredby
Trenitalia,thetrackisrepresentedbyastatisticalanalysisofthe
Aosta-PreSaintDidierlinebymeansofanensembleofNc
curvi-lineartracks,eachoflengthequaltoltrack.Thestatisticalanalysis
hasbeencarriedoutsplittingtheconsideredtrackinradiusclasses
andsuperelevationclasses.Thestatisticalapproachhasbeen
intro-ducedbecauseofthecomplexityandthelengthoftheAosta-Pre
SaintDidiertrack:forthesereasonsthesimulationoftheentireline
wouldhavecauseddifficultybothintermsofmultibodymodeling
andintermsofnumericalefficiency(computationalloadand
mem-oryconsumption).Theoutputsofthevehiclemodelaretheglobal
contactvariablesevaluatedduringalltheNcdynamicsimulations
andrepresenttheinputsofthewearmodel.
ThedynamicsimulationshavebeenperformedinSimpackRail.
Morespecifically,themultibodymodelhasbeenbuiltusingdirectly
the SimpackRail environmentwhile the globalcontact model,
implementedinC/C++,hasbeencustomizedbymeansofthe
Sim-packUserRoutinemodule(implementedinFORTRANenvironment)
thatallowstohandletheinteractionbetweenSimpackandroutines
definedbytheuser.Thewearmodelismadeupofthreedistinct
phases:thelocalcontactmodel,thewearevaluationandtheprofile
update.Initiallythelocalcontactmodel(basedbothonHertz’slocal
theoryandonsimplifiedKalker’salgorithmFASTSIM),startingfrom
theglobalcontactvariables,evaluatesthelocalcontactvariables
(contactpressuresand localcreepages) and dividesthecontact
patchintoadhesionareaandcreeparea.Then,thedistributionof
removedmaterial(hypothesizingthecontactindryconditionsas
requiredbyTrenitaliaandRFI)iscalculatedbothonthewheeland
ontherailsurfaceonlywithinthecreepareausingan
experimen-tallawbetweentheremovalmaterialandtheenergydissipatedby
frictionatthecontactinterface[6,7].Finallytheprofilesofwheel
andrailareupdated:thenewprofilesaretheoutputsofone
dis-cretestepofthewholemodelloopandhavetobepassedbackto
thevehiclemodelinordertocontinuethewearcycledescribedin
Fig.1andtosimulatethevehicledynamicswithupdatedprofiles.
Theevolutionofthewheelandrailprofilesisthereforeadiscrete
process.Inthisresearchthechoiceofthediscretestepsisoneof
themainissuesandhastoconsiderthedifferenceoftimescales
betweenthewheelandrailwearevolutionrate(aswillbeclarified
inthefollowing).Forthewheelwearthefollowingconsiderations
arevalid:
1. thetotalmileagekmtottraveledbytheconsideredvehiclehas
beensubdividedinconstantstepsoflengthequaltokmstep;
2.withineachdiscretestepofthewholemodel(correspondingto
kmstepkilometerstraveledbythevehicle)thewheelprofileis
Table1
Inertiapropertiesofthemultibodymodel.
MBSbody Mass(kg) Rollinertia
(kgm2) Pitchinertia (kgm2) Yawinertia (kgm2) CoachM 31,568 66,700 764,000 743,000 CoachT 14,496 30,600 245,000 236,000 Bogiem 3306 1578 2772 4200 Bogiet 3122 1674 3453 5011 Wheelsetm 2091 1073 120 1073 Wheelsett 1462 1027 120 1027
Thedepthoftherailweardoesnotdependonthedistance trav-eledbyvehiclebutonthenumberofvehiclesmovingonthetrack. Thereforeadifferentapproachforevaluatingthediscretestepfor therail,basedonthetotaltonnageburdenonthetrackMtot,is
needed:
1.dividingthetotaltonnageMtotbythevehiclemassMv,the
cor-respondingvehiclenumberNtothasbeencalculated;thenNtot
hasbeensubdividedinconstantstepsequaltoNstep;
2.withineachdiscretestep(correspondingtoNstepvehicles
mov-ingonthetrack)therailprofileissupposedtobeconstant. Finallythefollowingconsiderationsholdbothforthewheeland theraildiscretizationsteps:
• thenumberofdiscretizationstepsaffectsthemodelprecisionand thecomputationalload.Moreparticularly,increasingthestep number,themodelprecisionincreasesbut,atthesametime,the computationalloadincreasestoo:agoodcompromisemustbe researched;
• varioustypesofprofileupdatestrategiesareavailablein litera-ture[9,10]:theconstantstepandtheadaptivestepstrategiesare
themainones.Inthefirstoneaconstantupdatestepisdefined,
whilethesecondoneisbasedonthedefinitionofathreshold
valuethatimposesthemaximummaterialquantitytoremoveat
eachupdateoftheprofiles.Thetwomethodshavebeencompared
andthefirstonehasbeenchosenduetothefollowingreasons:
1.thephysicalphenomenonofthewearevolutionhasusuallyan
almostlinearcharacteristicandthusiswellsuitedtoaconstant
updatestep;
2.thetwomethodsleadtoverysimilarsystemevolutions(both
qualitativelyandquantitatively)butthefirstoneis
computa-tionallymoreefficient.
ThewearmodelhasbeenfullyimplementedintheMatlab
envi-ronment.
3. Thevehiclemodel
Inthissectionabriefdescriptionofthevehiclemodel
(com-posedbythemultibodymodelandtheglobalcontactmodel)is
given. In particular the global contact model willallow to the
detectionofthecontactpointsbetweenthewheelandrailand,
subsequently,ofthecontactforcesandtheglobalcreepagesinthe
contactpatch.
3.1. Themultibodymodel
TheDMUAln501Minuetto,apassengertransportunitwidely
usedontheItalianRailways,hasbeenchosenasbenchmark
vehi-cleforthisresearch;thephysicalandgeometricalcharacteristicsof
thevehiclecanbefoundinliterature[11,12].InTable1theinertia
propertiesofthevehicleareshown:motorsandgearboxeshave
notbeenmodeledandtheirinertiapropertieshavebeenincluded
inthemotorbogieandinthemotorwheelset(indicatedinTable1
Fig.2.Globalviewofthemultibodymodel.
withBogiemandWheelsetmrespectively)inordertotakeinto
accounttheirdifferentinfluenceontheunsprungandsprungmass.
ThemultibodymodelhasbeenrealizedintheSimpackRail
envi-ronment(seeFig.2)andconsistsofthirty-onerigidbodies:
• threecoaches;
• fourbogies:theintermediateones,interposedbetweentwo
suc-cessivecoaches,aretrailerbogieswhiletheotheronesaremotor
bogies;
• eightwheelsets:twoforeachbogie;
• sixteenaxleboxes:twoforeachwheelset.
Therigidbodiesareconnectedbymeansofappropriate
elas-tic and damping elements;particularly thevehicleis equipped
withtwosuspensionstages.Theprimarysuspensionsconnectthe
wheelsetstothebogies(seeFig.3)andcomprisetwospringsand
two verticaldampers,whilethesecondarysuspensionsconnect
thebogiestothecoaches(seeFig.4)andcomprisethefollowing
elements:
• twoairsprings;
• sixdampers(lateral,verticalandanti-yawdampers);
• onetractionrod;
• therollbar(notvisibleinthefigure);
• twolateralbumpstops.
Table2
Mainlinearstiffnesspropertiesofthesuspensions.
MBSelement Longitudinalstiffness
(N/m) Lateralstiffness (N/m) Verticalstiffness (N/m) Rollstiffness (Nm/rad) Pitchstiffness (Nm/rad) Yawstiffness (Nm/rad) Primary 1,259,600 1,259,600 901,100 10,800 10,800 1000 Suspension Spring Secondary 120,000 120,000 398,000 – – – Suspension Airspring Secondary – – 2,600,000 – – – Suspension Rollbar
Fig.4.Bogieandsecondarysuspensions.
Boththestagesofsuspensionshavebeenmodelledbymeansof three-dimensionalviscoelasticforceelementstakingintoaccount allthemechanicalnonlinearities(bumpstopclearance,dampers androdbehavior).Themainlinearcharacteristicsofthe suspen-sionsareshowninTable2whilethenonlinearcharacteristicsare
imposedasafunctionofdisplacementandvelocityforthesprings
andthedampersrespectively(seeFigs.5and6).
Fig.5.Exampleofnonlinearcharacteristic:verticaldampingoftheprimary sus-pension.
3.2. Theglobalcontactmodel
Dynamicsimulationsofrailwayvehiclesneedareliableand
effi-cientmethodtoevaluatethecontactpointsbetweenwheeland
rail,becausetheirpositionhasa considerableinfluencebothon
thedirectionandonthemagnitudeofthecontactforces.Inthis
workaspecificcontactmodelhasbeenconsideredinsteadofthat
implementedinSimpackRailinordertoachievebetterreliability
andaccuracy[1,2].Theproposedcontactmodelisdividedintwo
parts:inthefirstonethecontactpointsaredetectedbymeansofan
innovativealgorithmdevelopedbytheAuthorsinpreviousworks
[1,2],whileinthesecondonetheglobalcontactforcesactingatthe
wheel–railinterfaceareevaluatedbymeansofHertz’sandKalker’s
globaltheories[3–5].
Thealgorithmforthecontactpointsdetectionstartsfromthe
standardideathatthecontactpointsmakestationarythedistance
betweenthewheelandrailsurfaces(seeFig.7(a));inmoredetails
thedistancehasapointofrelativeminimumifthereisno
pen-etrationbetweentheconsideredsurfaces,whileithasarelative
maximumintheothercase(seeSection3.2.1fortheanalytic
for-mulation of the problem).The main features ofthe innovative
adoptedalgorithmarethefollowing:
• itisafully3Dmodelthattakesintoaccountallthesixrelative
degreesoffreedom(DOF)betweenwheelandrail;
• itisabletosupportgenericrailwaytracksandgenericwheeland
railprofiles;
• itassuresageneralandaccuratetreatmentofthemultiple
con-tactwithoutintroducingsimplifyingassumptionsontheproblem
geometryandlimitsonthenumberofcontactpointsdetected;
Fig.6. Exampleofnonlinearcharacteristic:lateralbumpstopstiffnessofthe sec-ondarysuspension.
Fig.7.Distancemethod.
• itassureshighlynumericalefficiencymakingpossibletheonline
implementation within the commercial multibody software
(Simpack-Rail,Adams-Rail);inthiswayalsothenumerical
per-formanceofthecommercialmultibodysoftwareareimproved.
Twospecificreferencesystemshavetobeintroducedinorderto
simplifythemodel’sequations:theauxiliaryreferencesystemand
thelocalreferencesystem.TheauxiliarysystemOrxryrzrisdefined
ontheplaneoftherailsandfollowsthemotionofthewheelset
duringthedynamicsimulations:thexraxisistangenttothecenter
lineofthetrackintheoriginOr,whosepositionisdefinedsothat
theyrzrplanecontainsthecenterofmassGwofthewheelset,and
thezraxisisperpendiculartoplaneoftherails.Thelocalsystem
Owxwywzwisrigidlyconnectedtothewheelsetexceptforthe
rota-tionarounditsaxisandthexwaxisisparalleltothexryrplane(see
Fig.7(b)).Inthefollowing,forthesakeofsimplicity,thevariables
referredtothelocalsystemwillbemarkedwiththeapexw,while
thosereferredtotheauxiliarysystemwiththeapexr;the
vari-ablesbelongingtothewheelandtotherailwillbeindicatedwith
thesubscriptswandrrespectively.
Thankstothesereferencesystems,thedefinitionofthewheel
andrailgeometricalsurfacesis easy.Inthelocalreference
sys-temthewheelsetcanbeconsideredasasimplerevolutionsurface
arounditsaxisyw(seeEq.(2));thegeneratingfunction,indicated
byw(yw),issupposedtobeknown(inthiswork,theprofileORES
1002forasinglewheelhasbeenchosen(seeFig.8)).Similarlythe
trackcanbelocallydescribedintheauxiliaryreferencesystemas
anestrusionsurfacealongthexraxis(seeEq.(3));thegenerating
function,indicatedbyr(yr),isknown(theprofileUIC60,with
lay-ingangle˛pequalto1/20rad,hasbeenchosenforasinglerail(see
Fig.8)).
WithreferencetoFig.1,theglobalcontactmodelcanbethought
ofasablackboxhavingthefollowinginputsandoutputs:
• Inputs: the kinematic variables evaluated by the multibody
model,i.e.thepositionOrw,theorientationmatrixRrw,theabsolute
velocity ˙Orwandtheabsoluteangularvelocityωrwofthewheel
sys-temandtheanalogousquantitiesoftherailsystemOrr=0,Rr
r=I,
˙Orandωrr;
• Outputs:theglobalcontactvariablesrelativetothewheel–rail
interface,likethepositionsPrwandPrrofthecontactpoints,the
contactforces(thenormalcomponentNrandthetangential
com-ponentsTr
x andTyr),theglobalcreepagesεx,εyandεspandthe
contactpatch’sdimensionsa,b.
3.2.1. Thedistancemethodalgorithm
In this subsection thealgorithmusedfor detectingthe
con-tactpointswillbedescribed.Thedistancemethodalgorithm(see
Fig.7(a))isbasedonaclassicalformulationofthecontactproblem
inmultibodyfield:
nrr(Prr)∧nrw(Prw)=nrr(Prr)∧Rrwnww(Pww)=0 (1a)
nrr(Prr)∧dr=0 (1b)
where:
• Pw
wandPrrarethepositionsofthegenericpointonthewheel
sur-faceandontherailsurfaceexpressedintheirreferencesystems:
Pww(xw,yw)=
xw yw − w(yw)2−x2w T (2) Prr(xr,yr)=(xr yr r(yr))T (3) • nwwandnrraretheoutgoingnormalunitvectorstothewheeland
railsurfacerespectivelyandaredefinedasfollows:
nw w(Pww)= −
∂
Pww∂
xw ∧∂
Pww∂
yw∂
Pww∂
xw ∧∂
Pww∂
yw , n r r(Prr)=∂
Prr∂
xr ∧∂
Prr∂
yr∂
Prr∂
xr ∧∂
Prr∂
yr (4) • Rrwistherotationmatrixthatlinksthelocalreferencesystemto
theauxiliaryone;
• dristhedistancevectorbetweentwogenericpointsonthewheel
surface andontherailsurface (bothreferred totheauxiliary
system)anditisequalto:
dr(xw,yw,xr,yr)=Pwr(xw,yw)−Prr(xr,yr) (5)
wherePrwisthepositionofthegenericpointofthewheelsurface
expressedintheauxiliarysystem:
Prw(xw,yw)=Orw+RrwPww(xw,yw). (6)
Thefirstcondition(Eq.(1a))ofthesystem(1)imposesthe
paral-lelismbetweenthenormalunitvectors,whilethesecondone(Eq.
(1b))requirestheparallelismbetweenthenormalunitvectorto
therailsurfaceandthedistancevector.
Alternatively,otherclassicalformulationsofthecontact
prob-lemcanbeused,byexampleimposingtheorthogonalitybetween
thedrvectorandthetangentialplanestothewheelandrailsurfaces
(respectivelyw andr);howeverthis approachleadstomore
complexandlessmanageablecalculationsandforthisreasonhas
notbeadopted.
The system (1) consists of six nonlinear equations in the
unknowns(xw,yw,xr,yr)(onlyfourequationsareindependentand
Fig.8. Wheelandrailprofiles.
ofthefourvariables(inthiscase(xw,xr,yr))asafunctionofyw,
reducingtheoriginal4Dproblemtoasimple1Dscalarequation.
Thereductionoftheproblemdimensionusingappropriate
ana-lyticalproceduresisthemostinnovativeaspectoftheproposed
method.ThesecondcomponentofEq.(1a)leadstothefollowing
equation:
r13
w(yw)2−xw2 =r11xw−r12w(yw)w(yw) (7)
where rij are the elements of the Rrw matrix. Calling A=r13,
B=w(yw),C=r11andD=r12w(yw)w(yw),thepreviousequation
becomes:
A
B2−x2w=Cxw−D. (8)
Hence,removingtheradicalandsolvingforxw:
xw1,2(yw)=
CD±
C2D2−(A2+C2)(D2−A2B2)A2+C2 ; (9)
ascanbeseen,therearetwopossiblevaluesofxwforeachyw.
FromthefirstcomponentofEq.(1a)thefollowingrelationfor
r(yr)holds: r(yr)1,2 = r21xr1,2(yw)−r22w(yw)w(yw)−r23
w(yw)2−xw1,2(yw)2 r32w(yw)w(yw)+r33 w(yw)2−xw1,2(yw)2 . (10)Ifr(yr)1,2isadecreasingmonotonousfunction(considering
sep-aratelythesidesofthetrack),Eq.(10)isnumericallyinvertible
andasinglepairyr 1,2(yw)existsforeachywvalue;otherwisethe
numericalinversionisstillpossiblebutwillproduceafurther
mul-tiplicationofthesolutionnumber.
BythesecondcomponentofEq.(1b)theexpressionofxr1,2(yw)
canbeobtained:
xr1,2(yw)=r11xw1,2(yw)+r12yw−r13
w(yw)2−xw1,2(yw)2. (11)
Finally,replacingthevariablesxw1,2(yw),xr1,2(yw)andyr1,2(yw)in
thefirstcomponentofEq.(1b),thefollowing1Dscalarequation
canbewritten: F1,2(yw) = −r
Gmz+r32yw−r33 w2−x w1,22−b + −Gmy+r21xw1,2+r22yw−r23 w2−x w1,22 −yr1,2 =0 (12)whereGwx,Gwy,Gwzarethecoordinatesofthewheelsetcenterof
massintheauxiliarysystem.Theexpression(12)consistsoftwo
scalarequationsinthevariableywthatcanbeeasilysolvedwith
appropriatenumericalalgorithms.
Theadvantagesofthisapproachbasedonthereductionofthe
algebraicproblemdimensionaremanyandcanbesummarizedas
follows:
• thereductionofthealgebraicproblemdimensionfrom4Dto1D
allowstoobtainanhighnumericalefficiencythatmakes
pos-sibletheonlineimplementationofthenewmethodwithinthe
multibodyvehiclemodels;
• theanalyticalapproachassuresanhighdegreeofaccuracyand
generality;
• the1Dproblemassuresaneasiermanagementofthemultiple
solutionsfromanalgebraicandanumericalpointofview;
• in1Dproblemalsoparticularlyelementarynumericalalgorithms
likethegridmethodarequiteefficient.
Thus, onceobtainedthegenericsolution(indicatedwiththe
subscript i) ywi of Eq. (12),the complete solution (xwi,ywi, xri,
yri)ofthesystem(1)andconsequentlythecontactpointsPrwi=
Prw(xwi,ywi)andPrri=Prr(xri,yri)canbefoundbysubstitution.
However,sinceEqs.(1a)and(1b)containirrationalterms,the
genericsolution(xwi,ywi,xri,yri)mustsatisfythefollowing
analyt-icalconditions:
• thesolutionmustbereal;
• thesolutiondoesnothavetogeneratecomplexterms(thatcould
becausedbytheradicalsintheequations);
• thesolutionmustbeaneffectivesolutionofthesystem(1)(check
necessarybecauseoftheradicalremovalbysquaring).
Furthermore,fromaphysicalpointofview,alsothenextchecks
Fig.9. Convexityconditions.
• themultiplesolutionsobtainedfromtheanalyticalresolutionof
Eq.(12)mustbeindividuatedanderasedbecausetheyhaveno
physicalmeaning;
• thefollowingconvexityconditionsmustbesatisfiedsothatthe
contactisphysicallypossible:
k1,wi+k1,ri>0
k2,wi+k2,ri>0
(13)
wherek1,wi,k2,wiarethenormalcurvatureofthewheelsurfacein
longitudinalandlateraldirection(referredtotheauxiliarysystem
andevaluatedintheithcontactpoint(xwi,ywi,xri,yri))whilek1,ri,
k2,riaretheanalogousquantitiesfortherailsurface.Becauseof
theproblemgeometry,thefirstoneofEq.(13)isalwayssatisfied
andthusonlythesecondonemustbeverified(seeFig.9).
• thegenericsolutionofthesystem(1)canbeaneffectivecontact
pointonlyifthenormalpenetration ˜pnbetweenthesurfacesof
wheelandrailisnegative(accordingtotheadoptedconvention),
i.e.theremustbeeffectivepenetrationbetweenthebodies:
˜pni=dri·nrr(Prri)=−d
r
i·nrw(Prwi)<0. (14)
3.2.2. Thecontactforces
Then,foreachcontactpoint,thenormalandtangentialcontact
forcesandtheglobalcreepagesonthecontactpatcharedetermined
(seeFig.10).
ThenormalforcesNr(expressedintheauxiliarysystem)are
calculatedbymeansofHertz’stheory[5,13]:
Nr=
−kh˜pn
+kv|
v
n| sign (v
n)−1 2 sign(˜p n)−1 2 (15) where:• ˜pnisthenormalpenetrationdefinedbyEq.(14);
• istheHertz’sexponentequalto3/2;
• kvisthecontactdampingconstant(kv= 105Ns/m);
•
v
n=V·nrristhenormalpenetrationvelocity(Visthevelocityofthecontactpointrigidlyconnectedtothewheelset);
Fig.10. Globalforcesactingatwheelandrailinterface.
• khisthehertzianconstant,functionbothofthematerial
prop-ertiesandofthegeometryofthecontactbodies(curvaturesand
semiaxesofthecontactpatch)[14,2].
Theglobalcreepagesε(longitudinalεx,lateralεyandspin
creep-ageεsp)arecalculatedasfollows:
εx= V· ir
˙Orw, εy= V· tr r Prr ˙Orw
, εsp= ωr w· nrr Prr ˙Orw
(16)
whereVisthevelocityofcontactpointrigidlyconnectedtothe
wheelset, ˙Orwisthewheelsetcenterofmassvelocity(takenasthe
referencevelocityforthecalculationoftheglobalcreepages),ωr
w
istheangularvelocityofwheelsetexpressedinauxiliarysystem,ir
istheunitvectorinlongitudinaldirectionoftheauxiliarysystem
andtrristhetangentialunitvectortotherailprofile.
The tangentialcontact forces ˜Tr
x, ˜Tyr and thespin torqueMspr
(expressedintheauxiliarysystem)arecalculatedbymeansofthe
Kalker’sglobaltheory:
˜
Tr
x=−f11εx, T˜yr=−f22εy−f23εsp
(17)
Mspr =f23εy−f33εsp (18)
wherethecoefficientsfijarefunctionbothofthematerialsandof
thesemiaxesofthecontactpatch:
f11=abGC11, f22=abGC22
f23=(ab)3/2GC23, f33=(ab)2GC33 (19)
inwhichGisthewheelandrailcombinedshearmodulusandCijare
theKalker’scoefficientsthatcanbefoundtabulatedinliterature
[3].Atthispoint,itisnecessarytointroduceasaturationonthe
tangentialcontactforces ˜Tr=[ ˜Tr
x T˜yr]T,inordertoconsiderthe
adhesionlimit(nottakeninaccountbythelinearKalker’stheory):
Tr≤cNr (20)wherecisthekineticfrictioncoefficient.Consequentlythe
satu-ratedtangentialforcesTrwillhavethefollowingexpression:
Tr=
˜Tr (21)inwhichthesaturationcoefficient
canbeevaluatedasfollows[15,16]: =
⎧
⎨
⎩
cNr ˜ Tr ˜ Tr cNr −1 3 ˜ Tr cNr 2 + 1 27 ˜ Tr cNr 3 if ˜Tr≤3cNr cNr ˜ Tr if ˜T r>3 cNr (22) where ˜Tr=˜T r .4. Thewearmodel
Inthissectionthethreephases,inwhichthewearmodelhas
beendivided,willbedescribedindetails:thelocalcontactmodel,
theevaluationoftheamountofremovedmaterial(assumingdry
contactconditions)andthewheelandrailprofileupdate.
4.1. Thelocalcontactmodel
Thepurposeofthelocalcontactmodelisthecalculationofthe
localcontactvariables(normalandtangentialcontactstressespn,
ptandlocalcreeps,allevaluatedwithinthecontactpatch)starting
fromthecorrespondingglobalvariables(contactpointsPrw,Prr,
con-tactforcesNr,Tr
x,Tyr,globalcreepageεandsemiaxesofthecontact
Fig.11.Contactpatchdiscretization.
ThismodelisbasedontheKalker’slocaltheoryinthe
simpli-fiedversionimplementedinthealgorithmFASTSIM;thisalgorithm
contains an extremely efficient version (although necessarily
approximate)oftheKalkertheoryandthereforeiswidelyusedin
railwayfield[17].
Forthelocalanalysisanewreferencesystemisdefinedatthe
wheel–railinterfaceonthecontactplane(i.e.thecommontangent
planebetweenthewheelandrailsurfaces):thexandyaxesare
thelongitudinalandthetransversaldirectionofthecontactplane
respectively(seeFigs.11and13).Thealgorithmisbasedonthe
pro-portionalityhypothesisbetweenthetangentialcontactpressurept
andtheelasticdisplacementsu,bothevaluatedwithinthecontact
patch:
u(x,y)=Lpt(x,y), L=L(ε,a,b,G,) (23)
wheretheflexibilityL(functionoftheglobalcreepagesε,the
semi-axesofthecontactpatcha,b,thewheelandrailcombinedshear
modulusGandthewheelandrailcombinedPoisson’sratio)can
becalculatedasfollows:
L=|εx| L1+
εy
L2+c
εsp
L3
(ε2
x+ε2y+c2ε2sp)
1/2 (24)
withL1=8a/(3GC11),L2=8a/(3GC22),L3=a2/(4GcC23)andc=
√ ab
(theconstantsCij,functionsbothofthePoisson’sratioandofthe
ratioa/b,aretheKalker’sparametersandcanbefoundinliterature
[17]).
Thelocalcreepagesscanbecalculatedbyderivationconsidering
boththeelasticcreepagesandtherigidones:
s(x,y)=u(x,• y)+V
εx εy (25)whereV=
˙Orwisthelongitudinalvehiclespeed.Atthispoint
itisnecessarytodiscretizetheellipticalcontactpatchinagridof
pointsinwhichthequantitiespn,ptandswillbeevaluated.Initially
thetransversalaxis(withrespecttothemotiondirection)ofthe
contactellipsehasbeendividedinny−1equalpartsofmagnitude
y=2b/(ny−1)bymeansofnyequidistantnodes.Thenthe
longitu-dinalsectionsofthepatch(long2a(y)= 2a
1−y/b2)havebeendividedinnx−1equalpartsofmagnitude x(y)=2a(y)/(nx−1)by
meansofnxequidistantnodes(seeFig.11).Duetothisstrategythe
longitudinalgridresolutionisnotconstantbutincreasesnearthe
lateraledgesoftheellipse,wherethelengthsa(y)aresmaller.This
procedureprovidesmoreaccurateresultsrightnexttotheedgesof
theellipse,whereaconstantresolutiongridwouldgenerate
exces-sivenumericalnoise.Thevaluesofthenxandnyparametershave
toassuretherightbalancebetweenprecisionandcomputational
load;goodvaluesofcompromiseareintherange[2550].
Oncethecontactpatchisdiscretized,theFASTSIMalgorithm
allowstheiterativeevaluationofboththecontactpressuresvalue
pn,ptandthelocalcreepagesinordertodividethecontactpatch
inadhesionandslipzone.Indicatingthegenericpointofthegrid
with(xi,yj),1≤i≤nx1≤j≤ny,thenormalcontactpressurecanbe
expressedas: pn(xi,yj)= 3 2 Nr ab
1−x 2 i a2− y2 j b2 (26)whereNristhenormalcontactforce,whilethelimitadhesion
pres-surepAis: pA(xi,yj)=pt(xi−1,yj)−
εx εy x(yj) L ; (27)thus,knowingthevariablevaluesinthepoint(xi−1,yj),itispossible
topasstothepoint(xi,yj)asfollows:
if
pA(xi,yj)≤pn(xi,yj) arrow pt(xi,yj)=pA(xi,yj) s(xi,yj)=0 (28a) if pA(xi,yj)>pn(xi,yj) arrow pt(xi,yj)=pn(xi,yj)pA(xi,yj)/pA(xi,yj) s(xi,yj)= LV x(yj) (pt(xi,yj)−pA(xi,yj)) (28b)whereisthestaticfrictioncoefficient;Eqs.(28a)and(28b)hold
respectivelyintheadhesionandslipzone.Iteratingtheprocedure
for2≤i≤nxandsuccessivelyfor1≤j≤nyandassumingas
bound-aryconditionspt(x1,yj)=0,s(x1,yj)=0for1≤j≤ny(i.e.stressesand
creepageszerooutofthecontactpatch),thedesireddistributionof
pn(xi,yj),pt(xi,yj)ands(xi,yj)canbedetermined.
4.2. Thewearevaluation
Toevaluatethedistributionofremovedmaterialonwheeland
railduetowear(assumingdrycontactconditions)an
experimen-talrelationshipbetweenthevolumeofremovedmaterialandthe
frictionalwork[6,7]hasbeenused.Particularlytherelationshipis
abletodirectlyevaluatethespecificvolumeofremovedmaterial
ı
Pwijk(t)(x,y)andıPrijk(t)(x,y)relatedtotheithcontactpointsP
jk wi(t)
andPrijk(t)onthejthwheelandrailpairduringthekthoftheNc
dynamicsimulations.
Thecalculationofı
Pjki(t)(x,y)requiresfirstofalltheevaluation
ofthefrictionpowerdevelopedbythetangentialcontactstresses;
tothispurposethewearindexIW(expressedinN/mm2)isdefined
asfollows:
IW=
pt·s
V . (29)
Thisindex, bymeansof appropriateexperimentaltests, canbe
correlated with the wear rate KW (expressed in g/(mmm2))
which representsthemassof removedmaterialforunit of
dis-tance traveled by the vehicle (expressed in m) and for unit
of surface (expressed in mm2). Wear tests carried out in the
caseofmetal–metalcontactwithdrysurfacesusingatwindisc
Fig.12.TrendofthewearrateKW.
relationship between KW and IW adopted for the wear model
describedinthisworkisthefollowing(seeFig.12):
KW(IW)=
5.3∗IW IW<10.4 55.0 10.4≤IW≤77.2 61.9∗IW IW>77.2. (30)OncethewearrateKW(IW)isknown(the samebothfor the
wheelandfortherail),thespecificvolumeofremovedmaterialon
thewheelandontherail(forunitofdistancetraveledbythe
vehi-cleandforunitofsurface)canbecalculatedasfollows(expressed
inmm3/(mmm2)): ı Pwijk(t)(x,y)=KW(IW) 1 (31) ı Prijk(t)(x,y)=KW(IW) 1 (32)
whereisthematerialdensity(expressedinkg/m3).
4.3. Profileupdate
Theprofileupdatestrategyisthesetofnumericalprocedures
thatallowsthecalculationofthenewprofiles ofwheelwn(yw)
andrailrn(yr)(theprofilesatthenextstep),startingfromtheold
profilesofwheelwo(yw)andrailro(yr)(i.e.theprofilesatthe
cur-rentstep)andallthedistributionsofremovedmaterialı
Pwijk(t)(x,y)
andı
Prijk(t)(x,y).Theupdatestrategy,besidesevaluatingthenew
profiles,isnecessaryfortwoadditionalreasons:
1.thenecessitytoremovethenumericalnoisethatcharacterizes
thedistributionsı
Pjki(t)(x,y)andthat,duetononphysical
alter-ationsof thenew profiles,can causeproblems totheglobal
contactmodel;
2.theneed to mediatethe distributions ı
Pijk(t)(x,y) in order to
obtainasingleprofilebothforthewheelandtherailasoutput
ofthewearmodel(asrequiredbythespecificationsofTrenitalia
andRFI).
Thefollowingmainstepscanbedistinguished:
• Longitudinalintegration: 1 2w(yjkwi)
+a(y) −a(y) ıPjk wi(t) (x,y)dx=ıtot Pjkwi(t)(y) (33) 1 ltrack +a(y) −a(y) ı Prijk(t)(x,y)dx=ı tot Prijk(t)(y) (34)wherew(ywijk)isthewheelradiusevaluatedinyjkwiandltrackisthe
lengthofthesimulatedtrack.Thisfirstintegrationsums,inthe
longitudinaldirection,allthewearcontributesinsidethe
con-tactpathandaveragesthisquantityoverthewholelongitudinal
Fig.13. Normalabscissaforthewheelandrailprofile.
developmentofthewheelandoftherail(bymeansofthe
fac-tors1/2w(yjkwi)and1/ltrack);inotherwordsitprovidesthemean
value ofremoved material(expressed inmm3/(mmm2)). The
differencebetweentheterms1/ltrackand1/2w(yjkwi)(thetrack
lengthismuchgreaterthanthewheelcircumferencelength)is
themaincausethatleadsthewheeltowearmuchfasterthanthe
railandconsequentlytoadifferentscaleofmagnitudeofthetwo
investigatedphenomena.Thisreflectsthephysicalphenomena
thatthelifeoftherailismuchgreaterthanthatofthewheel.
Forthisreason,aswillbebetterexplainedinthefollowing,it
isnecessarytodevelopadifferentstrategyfortheupdateofthe
wheelandrailprofilerespectively.Inthisresearchthefollowing
strategieshavebeenadopted:
1.forthewheelupdatethemileagetraveledbyvehicleis
consid-ered.Thetotalmileagekmtot(derivedfromtheexperimental
dataprovidedbyTrenitaliaandRFI)issubdividedintoconstant
stepsoflengthequaltokmstep;
2.fortherailupdate,thetotaltonnageburdenonthetrack[18]is
considered.ThevehiclenumberNtotcorrespondingtothetotal
consideredtonnageissubdividedintoconstantstepsequalto
Nstep. • Trackintegration:
Tend Tin ıtot Pwijk(t)(y)V (t)dt≈ Tend Tin ıtot Pwijk(t)(sw−s cjk wi(t))V (t)dt= Pwijk(sw) (35) Tend Tin ıtot Pjk ri(t) (y)V (t)dt≈ Tend Tin ıtot Pjk ri(t) (sr−scjkri (t))V (t)dt= Pjk ri (sr); (36)thetrackintegrationsumsallthewearcontributescomingfrom
thedynamicsimulationtoobtainthedepthofremovedmaterial
forwheel
Pwijk(sw)andrail Prijk(sr)expressedinmm=mm
3/mm2.
Inordertohaveabetteraccuracyinthecalculationoftheworn
profiles,thenaturalabscissasswandsrofthecurvesw(yw)and
r(yr)havebeenintroduced.Inparticularthefollowingrelations
locallyhold(seeFig.13):
y≈sw−scjkwi(t) y≈sr−scjkri (t) (37)
w(yw)=w(yw(sw))= ˜w(sw) r(yr)=r(yr(sr))= ˜r(sr) (38)
wherethenaturalabscissasofthecontactpointsscjkwi andscjkwi can
beevaluatedfromtheirpositionsPwijkandPrijk.
• Sumonthecontactpoints:
NPDC
i=1 Pjkwi(sw)= w jk(sw) (39) NPDC i=1 Pjk ri (sr)= rjk(sr) (40)Fig.14.Discretizationofthetotalmileage.
whereNPDCisthemaximumnumberofcontactpointsofeach
singlewheel(andrespectivelyofeachsinglerail);sincethe
num-berofcontactpointsonthewheel–railpairisusuallylessthan
NPDCandchangesinthetimeduringthedynamicsimulation,it
hasbeenassumedthatthewearcontributionassociatedtothe
fictitiouspointsiszero.
• Averageonthevehiclewheelsandonthedynamicsimulations:
Nc
k=1 pk 1 Nw Nw j=1 w jk(sw)= w (sw) (41) Nc k=1 pk 1 Nw Nw j=1 r jk(sr)= r (sr) (42)whereNwisthenumberofvehiclewheelswhilethepk,1≤k≤Nc,
NCk=1pk=1arethestatisticalweightsassociatedtothevarious
dynamic simulationsderived fromthestatisticalanalysis.The
averageonthenumberofwheel–railpairshastobeevaluated
inordertoobtainasoutputofthewearmodelasingleaverage
profilebothforthewheelandfortherail(asrequiredbyTrenitalia
andRFI).
• Scaling:
theaimofthescalingprocedureistoamplifythesmallquantity
ofmaterialremovedduringtheNcdynamicsimulationsand,at
thesametime,tolimitthecomputationalload;usingthealmost
linearityofthewearmodelwiththetraveleddistance,itis
possi-bletoamplifytheremovedmaterialbymeansofascalingfactor
whichincreasesthedistancetraveledbythevehicle.
Thealmostlinearityofthewearmodelinsidethediscretesteps
kmstepin whichthetotalmileage traveledkmtot issubdivided
isaworkinghypothesiscomingfromthediscreteapproachof
themodel.Itisbasedontheideathatthewearrateinsidethe
simulateddistance(kmprove)remainsthesamealsoinsidethe
dis-cretestepkmstep,sincetheconsideredvehiclealwayscoversthe
sametracksofthestatisticalanalysisbothduringthesimulated
distance(kmprove)andduringthediscretestep(kmstep).
Inthisworkaconstantdiscretestephasbeenchosentoupdate
thewheelandrailprofiles(seeFig.14):infactthismethodwell
adaptstothealmostlinearcharacteristicofthewearevolution.
Furthermoreitrequireslimitedcomputationalloadwithout
los-ingaccuracyifcomparedwithdifferentsuitablestrategiesasthe
adaptivestep[7].
Theevaluationofthediscretestep,withtheconsequentscaling
of w(sw)and
r
(sr),representsthemajordifferencebetween
wheelupdateandrailupdate:
1. theremovedmaterialonthewheelduetowearisproportional
tothedistancetraveledbythevehicle;infactapointofthe
wheelisfrequentlyincontactwiththerailinanumberoftimes
proportionaltothedistance.Thefollowingnomenclaturecan
beenintroduced(seeFig.14):
–kmtotisthetotalmileagetraveledbytheconsidered
vehi-cle(kmtotcanbechosendependingonthepurposeofthe
simulations,forexampleequaltothere-profilingintervals);
– kmstepisthelengthofthediscretestepinwhichthetotal
mileagekmtotissubdivided;
–kmprove=ltrackistheoverallmileagetraveledbythevehicle
duringtheNcdynamicsimulations;thenecessityof
accept-ablecomputationaltimeforthemultibodysimulationsleads
toadoptsmallvalueofthekmprovelengthandforthis
rea-sontherelativeremovedmaterialhastobescaledwitha
multiplicativefactor.
Finallythematerialremovedonthewheelhastobescaled
accordingtothefollowinglaw:
w(sw)
kmstep
kmprove = w sc
(sw). (43)
Thechoiceofthespatialstepmustbeagoodcompromise
betweennumericalefficiencyandtheaccuracyrequired by
thewearmodel.Akmsteptoosmallcomparedtokmtotwould
provideaccurateresultsbutexcessivecalculationtimes;the
contraryhappenswithkmsteptoobigcomparedtokmtot.
2.thedepthofrailwearisnotproportionaltothedistance
trav-eledbythevehicle;infacttherailtendstowearoutonlyin
thezonewhereitiscrossedbythevehicleand,increasingthe
traveleddistance,thedepthofremovedmaterialremainsthe
same.Ontheotherhandtherailwearisproportionaltothe
totaltonnageMtotburdenontherailandthustothetotal
vehi-clenumberNtotmovingonthetrack.Therefore,ifNstepisthe
vehiclenumbermovinginadiscretestep,thequantityofrail
removedmaterialateachstepwillbe:
r(sr)∗Nstep=
r sc
(sr) (44)
whereNstepiscalculatedsubdividinginconstantstepthe
vehi-clenumberNtotcorrespondingtothetotaltonnagethathasto
besimulated;Ntotcanbeobtainedstartingfromthevehicle
massMv:
Ntot=Mtot
Mv . (45)
• Smoothingoftheremovedmaterial:
I
w sc(sw) = w scsm(sw) (46) I r sc(sr) = r scsm(sr); (47)thesmoothingoftheremovedmaterialfunctionisnecessaryto
removethenumericalnoisethataffectsthisquantityandthat
wouldbepassedtothenewprofiles ˜wn(sw)and ˜rn(sr)ofwheel
andrailcausingproblemstotheglobalcontactmodel.Tothis
end,adiscretefilter(i.e.amovingaveragefilterwithwindow
sizeequalto1%÷5%ofthetotalnumberofpointsinwhichthe
profilesarediscretized) hasbeenused;obviously thediscrete
filterhastoconservethemass.
• Profileupdate:
yw(sw) ˜ wo(sw) − w scsm(sw)nrw re−parameterization −→ yw(sw) ˜ wn(sw) yr(sr) ˜ro(sr) − r scsm(sr)nrr re−parameterization −→ yr(sr) ˜rn(sr) ; (48)thelaststepconsistsin theupdateoftheoldprofiles ˜wo(s)=
wo(y)and ˜ro(sr)=ro(yr)toobtainthenewprofiles ˜wn(s)=wn(y)
and ˜rn(sr)=rn(yr);sincetheremovalofmaterialoccursinthe
normaldirectiontotheprofiles(nr
wandnrraretheoutgoingunit
vectorforthewheelandrailprofilerespectively),onceremoved
thequantities w scsm(sw)and
r sc
sm(sr),are-parameterizationof
theprofilesisneededinordertoobtainagaincurves
5. Wearmodelvalidation
Inthis sectionthewearmodelvalidation phasewillbe
pre-sented. Initially, the set of Nc curvilinear tracks,on which the
dynamicsimulationsof theDMU Aln501Minuettovehiclehave
beenperformed,willbeintroduced(tracksextractedstartingfrom
thestatisticalanalysisoftheAosta-PreSaintDidiertrack,thedata
of which has been provided by RFI); moreover the wear
con-trolparametersforthewheelandrailwillbedefined(theflange
heightFH,theflangethicknessFT,theflangesteepnessQRand
thequotaQMfortherail).Thentheexperimentaldata(provided
byTrenitalia)measuredontheAosta-PreSaintDidiertrackand
theirprocessingwillbeintroduced.Finally,thesimulation
strat-egyusedtoanalyzethewearbothonthewheelandontherailwill
bedescribedandtheresultsobtainedwiththewearmodelwillbe
analyzedandcomparedwiththeexperimentaldata.
5.1. StatisticalanalysisoftheAosta-PreSaintDidiertrack
StartingfromthedataofthewholeAosta-PreSaintDidiertrack
(providedbyRFI),thestatisticalanalysishasbeenperformedby
dividingthelinebothinradiusclasses(determinedbyRmin and
Rmax)andinsuperelevationclasses(determinedbyhminandhmax)
[11].Moreparticularlyfivesuperelevationsubclassesaredefined
foreachradiusclass.Thesubclassesthatdonotincludecurvehave
notbeentakenintoaccountinthedefinitionofthesetofNctracks.
AlltheNccurvedtracksareshowninTable3.
ThesetconsistsinNc=18distinctelements(17realcurvesand
thestraightline)characterizedbytheradiusvalueRc,the
superel-evationvalueH,thetravelingspeedVandthestatisticalweightpk
(with1≤k≤Nc)thatrepresentsthefrequency withwhich each
curveappearsonthe consideredrailwaytrack (Aosta-PreSaint
Didierline).TheradiiRcarecalculatedbymeansoftheweighted
meanonallthecurveradiiincludedinthecorresponding
superel-evationsubclass(theweightedfactoristhelengthofthecurvesin
therealtrack).Foreachsubclass,thevalueHisthemostfrequent
superelevationvalueamongthestandardvaluesthatcharacterize
thecurvesoftheconsideredsuperelevationsubclass.Thetraveling
speedsVarecalculatedimposingathresholdvalueonthe
uncom-pensatedaccelerationalim
nc=0.6m/s2: ˜ V2 Rc − H sg=a lim nc (49)
wheresistherailwaygaugeandgisthegravityacceleration.The
estimatedspeed ˜Vhasbeenthencomparedwiththemaximum
velocityVmaxonthelinetogetthedesiredtravelingspeed V=
min( ˜V ,Vmax).
5.2. Wearcontrolparameters
ThereferencequotasFH,FTandQRareintroducedinorderto
estimatethewheelprofileevolutionduetothewearwithout
nec-essarilyknowingthewholeprofileshape(seeFig.15).According
tothesequotastheuserwillbeablebothtoestablishedwhenthe
wornwheelprofilewillhavetobere-profiledandtodetectifthe
wearcompromisesthedynamicalstabilityofthevehicle[19].
Theproceduretodefinethereferencequotasisthefollowing:
1.firstofallthepointP0isdefinedontheprofile,at70mmfrom
theinternalverticalfaceofthewheel;
2.thenthepointP1isintroducedontheprofile,2mmunderthe
flangevertex;
3.finallythepointP2isdeterminedontheprofile,10mmunder
thepointP0;
4. thewear control parameters are then calculated as follows:
theflangethicknessFTisthehorizontaldistancebetweenthe
Fig.15.Definitionofthewheelwearcontrolparameters.
internalverticalfaceandthepointP2;theflangesteepnessQR
isthehorizontaldistancebetweenthepointsP1andP2,while
theflangeheightFHistheverticaldistancebetweenP0andthe
flangevertex(allthedistancesareconsideredpositive).
Anadditionalcontrolparameteristhenintroducedtoevaluate
theevolutionofrailwear.ParticularlytheQMquotaisdefinedas
therailheadheightinthepointyr=760mmwithrespecttothe
centerlineofthetrack:thisyrvaluedependsontherailwaygauge
(equalto1435mmintheAosta-PreSaintDidierline)andonthe
layingangle˛pofthetrack(equalto1/20rad).PhysicallytheQM
quotagivesinformationontherailheadwear(seeFig.16).
5.3. Experimentaldataandtheirprocessing
TheexperimentaldataprovidedbyTrenitaliaandRFIarerelated
onlytothewheelwearandconsistsinthewearcontrolparameters
measuredasafunctionofthetotaldistancetraveledbythe
con-sideredvehicleDMUAln501Minuetto;particularly,thedatahave
beenmeasuredonthreedifferentvehiclesoperatingonthesame
trackthatareconventionallycalledDM061,DM068,DM082.
AscanbeseenbyexampleinTable4forthevehicleMD061,
thereferencequotavalueshavebeenmeasuredforallthevehicle
wheels(eachvehiclehaseightwheelsetsasspecifiedinSection3.1).
Howeverthefollowingdataprocessinghasbeennecessaryinorder
toobtainasinglewheelprofilethatcouldbeeffectivelycompared
withtheprofileextractedfromthenumericalsimulationandto
reducethemeasurementerrors:
1.initiallythearithmeticmeanonallthesixteenvehiclewheelshas
beenevaluated;themeanisnecessarytoobtainasinglewheel
profileand,atthesametime,toreducethemeasurementerrors
affectingtheexperimentaldata;
2.thenascalingofthequotavalueshasbeencarriedoutinorder
todeletetheoffsetontheinitialvalueoftheconsidered
quanti-ties:thisprocedureimposesthatallthewearcontrolparameters
startfromtheirnominalvalues(thestandardvaluesfortheORE
S1002profilehavebeenused)inordertoremovetheinitial
differencesamongthevehiclesduetomeasurementerrors;
Table3
Dataofthecurvilineartracksofthestatisticalanalysis.
Rmin(m) Rmax(m) Superelevationhmin−hmax(mm) Rc(m) H(mm) V(km/h) pk(%)
147.1 156.3 0 – 10–40 – 60–80 – 90–120 150 120 55 0.77 130–160 – 156.3 166.7 0 – 10–40 – 60–80 – 90–120 160 110 55 0.48 130–160 165 140 55 0.56 166.7 178.6 0 – 10–40 – 60–80 – 90–120 170 110 55 0.82 130–160 175 130 55 1.55 178.6 192.3 0 – 10–40 – 60–80 – 90–120 190 100 55 8.37 130–160 180 130 55 0.45 192.3 208.3 0 – 10–40 – 60–80 – 90–120 200 90 55 20.64 130–160 200 130 60 4.00 208.3 227.3 0 – 10–40 – 60–80 220 80 55 0.70 90–120 220 100 55 3.76 130–160 – 227.3 250.0 0 – 10–40 – 60–80 240 80 55 7.26 90–120 240 110 60 5.28 130–160 – 250.0 312.5 0 – 10–40 – 60–80 270 70 55 3.91 90–120 270 90 60 5.29 130–160 – 312.5 416.7 0 – 10–40 – 60–80 370 60 55 2.26 90–120 345 100 70 1.63 130–160 – 416.7 ∞ 0 ∞ 0 70 32.27
3.the arithmetic mean on the three vehicle MD061, MD068, MD082hasnotbeencarriedout,inordertomaintaina disper-sionrangefortheexperimentaldata.
Theexperimentaldata,properlyprocessed,aresummarizedin Table5.Ascanbeseen,theflangeheightFHremainsapproximately
constantbecauseofthelowmileagetraveledbythevehicles,while
theflangethicknessFTandtheflangesteepnessQRdecreasealmost
linearlyandhighlight,accordingtothecharacteristicsofthetrack,
thewearconcentrationinthewheelflange.
5.4. Simulationstrategy
Inthissectionthesimulationcampaignperformedtostudythe
wearonwheelandrailisdescribed.AsexplainedinSection4.3,
thetwophenomenaevolveaccordingtodifferenttimescales
(sev-eralordersof magnitude)and afullysimulation ofsuchevents
wouldrequirea tooheavycomputationaleffort.For thisreason
thefollowingspecificalgorithmhasbeenadoptedforupdatingthe
profiles:
1.both for the wheel and for the rail five discrete stepshave
beenchosen,nsw=5andnsr=5,sotohaveagoodcompromise
betweencalculationtimesandresultaccuracy:
(a)thechoiceof thewheelkmstep (seeSection4.3)hasbeen
made considering the whole distance traveled equal to
kmtot≈3500km(gotfromexperimentaldata);thusthe
sin-glesteplengthwillbe:
kmstep=
kmtot
nsw ≈
700(km). (50)
(b) toestimatethevehiclenumberNtot(seeSection4.3)a
crite-rionpresentinliterature(basedonthetotaltonnageburden
onthetrack)hasbeenused[18].Particularlythereisa
pro-portionalityrelationshipbetweentonnageandwear:arail
wearof 1mmontherailheadheightevery100Mt
Table 4 Experimental data of the DMU Aln 501 Minuetto DM061. km quotas 1r 1l 2r 2l 3r 3l 4r 4l 5r 5l 6r 6l 7r 7l 8r 8l Wheel diameter Wheel diameter Wheel diameter Wheel diameter Wheel diameter Wheel diameter Wheel diameter Wheel diameter 816 mm 815 mm 824 mm 823 mm 823 mm 823 mm 819 mm 820 mm 0 F T 30.953 30.944 30.983 30.784 31.099 30.957 30.938 31.076 30.401 30.367 30.830 30.987 30.437 30.717 30.852 30.933 FH 27.970 27.894 28.141 28.043 27.969 28.187 28.030 28.271 28.245 27.918 28.141 27.982 28.013 27.937 28.333 27.883 QR 10.208 10.140 10.424 10.457 10.220 10.306 10.279 10.833 10.332 10.445 10.364 10.219 10.421 10.500 10.338 10.396 1426 FT 29.855 28.977 30.283 29.317 30.118 29.383 30.152 29.450 29.796 29.799 30.288 29.483 29.802 29.085 30.267 29.316 FH 28.010 27.923 28.104 28.108 28.000 28.249 28.095 28.278 28.248 28.284 28.247 28.030 28.997 28.003 30.383 27.919 QR 9.297 8.226 9.822 8.956 9.344 8.749 9.551 9.072 9.635 9.767 9.773 8.763 9.593 8.883 9.675 8.762 2001 FT 29.056 28.498 29.722 28.878 29.441 28.667 29.629 28.717 29.153 28.101 29.739 28.841 29.066 28.447 29.625 28.777 FH 27.990 27.880 28.161 28.080 29.998 28.248 28.128 28.283 28.290 27.994 28.273 28.022 28.027 28.014 28.362 27.957 QR 8.404 7.558 9.233 8.637 8.702 7.950 8.873 8.436 9.144 8.141 9.235 8.086 9.038 8.152 9.248 8.373 2575 FT 28.259 27.096 29.333 28.045 28.972 28.385 29.029 28.124 29.053 27.600 29.095 28.505 28.553 27.866 29.205 28.473 FH 28.009 27.089 28.173 28.020 28.063 28.243 28.090 28.241 28.285 27.963 28.244 28.085 28.030 28.018 28.352 27.968 QR 7.198 7.024 8.853 8.163 8.123 7.598 8.438 7.791 8.868 7.395 8.559 7.840 8.372 7.340 8.777 7.900
appreciablerailwear,amaximumvalueofremovedmaterial
depthof2mmontherailheadheighthasbeenhypothesized
(naturallythisvaluecanbechangedaccordingtothe
require-mentsofthesimulation).Thenumberofvehicles,ofknown
massMv(seeTable1),whichshouldevolveonthetrackto
reachthe200Mt,hasbeenthereforecalculated:
Ntot= Mtot Mv ≈ 2,000,000 (51) andthen: Nstep= Ntot nsr ≈ 400,000. (52)
2.the wear evolution on wheel and rail has been decoupled
becauseofthedifferentscalesofmagnitude:
(a)whilethewheel wearevolves,therail issupposed tobe
constant:infact,inthetimescaleconsidered,therailwear
variationisnegligible.
(b) becauseofthetimescalecharacteristicoftherailwear,each
discrete railprofilecomes incontact, withthesame
fre-quency,witheachpossiblewheelprofile.Forthisreason,for
eachrailprofile,thewholewheelwearevolution(fromthe
originalprofiletothefinalprofile)hasbeensimulated.
Basedonthetwoprevioushypotheses,thesimulationshave
beencarriedoutaccordingtothefollowingstrategy:
Wheelprofileevolutionatfirstrailstep:w0
i p1,1
⎧
⎨
⎩
w0 0 r0 → w0 1 r0 →···→ w0 4 r0 →w0 5Averageontherailsr(i1+1)forthecalculationofthesecondrail
step:r1 p1,2
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
w0 0 r0 w0 1 r0 . . . ... w0 4 r0⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
→⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
r1(1) r1(2) . . . r1(5)⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
→r1 . . .Wheelprofileevolutionatfourthrailstep:w4
i p5,1
⎧
⎨
⎩
w4 0 r4 → w4 1 r4 →···→ w4 4 r4 →w4 5Averageontherailsr(i5+1)forthecalculationofthefifthrail
step:r5 p5,2
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩
⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎜
⎝
w4 0 r4 w4 1 r4 . . . ... w4 4 r4⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎟
⎠
→⎛
⎜
⎜
⎜
⎜
⎜
⎜
⎝
r5(1) r5(2) . . . r5(5)⎞
⎟
⎟
⎟
⎟
⎟
⎟
⎠
→r5 (53)wherewjiindicatestheithstepofthewheelprofilethatevolves
Table5
Experimentaldataprocessed.
Vehicle Distancetraveled(km) FH(mm) FT(mm) QR(mm)
DM061 0 28.0 32.5 10.8 1426 28.2 31.5 9.8 2001 28.1 30.8 9.1 2575 28.0 30.2 8.6 DM068 0 28.0 32.5 10.8 1050 28.0 31.8 10.0 2253 28.0 30.2 8.5 2576 28.0 30.0 8.4 DM082 0 28.0 32.5 10.8 852 28.0 32.3 10.6 1800 28.0 31.3 9.6 2802 28.0 30.3 8.7 3537 27.6 30.0 8.3
thesameforeachjandcorrespondtotheunwornwheelprofile (ORES1002).
Initially the wheel (starting from the unworn profile w0 0)
evolvesontheunwornrailprofiler0inordertoproducethe
dis-cretewheelprofilesw0
0,w01,...,w50(stepp1,1).Thenthevirtual
railprofilesr(i+1)1 ,obtainedbymeansofthesimulations(w0 i,r0)
with0≤i≤4,arearithmeticallyaveragedsoastogettheupdate railprofiler1(stepp1,2).Thisprocedurecanberepeatednsrtimes
inordertoperformalltheraildiscretesteps(uptothestepp5,2).
Thecomputationaleffortrequiredbythesimulationstrategy isthefollowing:
(a)inthewheelwearstudy,foreachupdateoftherailprofile rj,thewholewheelwearloopwijwith0≤i≤4(nswstepsof
simulation)issimulated.Thecomputationaleffortresultsof nsw×nsr=25stepsbothforthedynamicanalysis(in
Sim-packRail) andfor thewearmodel necessarytocalculate theremovedmaterialonthewheel(inMatlab).Sothetotal numberofsimulationstepsare2(nsw×nsr)=50.
(b)intherailwearstudythedynamicanalysesarethesame asthepreviouscasebecauseforeach railstepthewheel profileswij(0≤i≤4)aresimulatedonrjinordertoobtain
rj(i+1)andthustheupdatedrailprofilerj+1bymeansofan
arithmeticmean.Therefore,noadditionaldynamical anal-yses are needed. In this caseonly the wearmodel steps mustbesimulatedsoastogettheremovedmaterialonthe rail.Consequentlythetotalnumberofsimulationstepsis nsw×nsr=25.
Thecharacteristicsoftheprocessorusedinthesimulationsand themeancomputationaltimesrelativetoeachdiscretestepofthe wholemodelloop(dynamicalsimulationandwearsimulation)are brieflysummarizedinTable6.
5.5. Evolutionofwearcontrolparameters
In this section the evolutionof thewheel reference quotas
numericallyevaluatedbymeansofthewearmodel(flange
thick-nessFT,flangeheightFHandflangesteepnessQR)willbecompared
withtheexperimentaldataconcerningthethree DMUsAln 501
Table6
Computationaltime.
Processor Computationaltime
Dynamicsimulation
(SimpackRail)
Wearsimulation
(Matlab)
INTELXeonCPUE5430
2.66GHz8GBRAM
2h2s 31s
Fig.17.FTdimensionprogress.
Fig.18.FHdimensionprogress.
Minuettovehicles.FurthermoretherailreferencequotaQM evo-lutionwillbeshownandcomparedwiththecriterionpresentin literaturebasedonthetotaltonnageburdenonthetrack[18].
TheprogressofFTdimension,forthensrdiscretestepoftherail,
isshowninFig.17asafunctionofthemileage;asitcanbeseen,
thedecreaseofthedimensionisalmostlinearwiththetraveled
distanceexceptinthefirstphases,wheretheprofilesarestillnot
conformalenough.TheFHquotaprogressisrepresentedinFig.18
andshowsthat,duetothepresenceofmanysharpcurvesinthe
statisticalanalysisofthetrackandtothefewkilometerstraveled,
thewheelwearismainlylocalizedontheflangeratherthanonthe
treadandtheflangeheightremainsnearconstantinagreement
withexperimentaldata.TheQRtrendisshowninFig.19:alsothe
flangesteepnessdecreasesalmostlinearly,leadingtoanincreaseof
Fig.20.QMdimensionprogress.
theconicityoftheflange.Finally,theevolutionofthewheelcontrol
parametersremainsquantitativelyandqualitativelysimilarasthe
railwearraises.
Althoughthesimulatedmileageisquiteshortconsideringthe
meantraveleddistancebetweentwoturningsofthewheelsina
standardscenery(infacttheFHquotaremainsalmostconstant),
thevariationsoftheFTandQRdimensionsareremarkableandit
highlightsthewearproblemsaffectingthevehicleDMUAln501
Minuettorunningalongtherailwaylinetakenintoaccount.
Inconclusion,thecomparisonsshowthattheoutputsof the
wearmodelareconsistentwiththeexperimentaldata,bothfor
theflangedimension(FH,FT)andfortheconicity(QR);theslightly
steeperdevelopmentoftheexperimentaldatathanthesimulation
canbeexplainedwiththedispersionoftheexperimentaldataand
withwearmechanisms,likeplasticandpittingwear,not
consid-eredinthedevelopedwearmodel.
Finally,theQMevolutionfortheanalysisoftherailwearis
pre-sentedinFig.20andshowsthealmostlineardependencebetween
therailwearandthetotaltonnageburdenonthetrack:theamount
ofremovedmaterialontherailprofileisinagreementwiththe
cri-terionpresentin literature(1mmontherailheadheightevery
100Mtofaccumulatedtonnage).
5.6. Evolutionofthewheelandrailprofiles
Thewearevolutiononthewheelprofilesevolvingoneachrail
rj(with 0≤j≤nsrand nsr=5)is presentedin thefollowing(see
Figs.21and23–27).Asstatedpreviously,thewheelprofile
evolu-tionisdescribedbymeansofnsw=5stepsandthespatialstepkmstep
hasbeenchosenequalto700km,sincethetotalmileagekmtotis
3500km.Fig.22showsthecumulativedistributionsofremoved
materialinverticaldirectionzwonthewheelprofileatfirstrailstep
wKw(yw)=
Kwi=1iw(yw)asafunctionofyw(1≤Kw≤nsw),where
w
i (yw)istheremovedmaterialbetweentwosubsequentdiscrete
stepsofthewheelprofileevolution.
Thequitelimiteddistancetraveledbythevehiclejustifiesthe
lowwearonthewheeltreadandentailsasmallreductionofthe
rollingradius.Howeverthehightortuosityoftheconsideredtrack
leadstoappreciablewearonthewheelflange.InFigs.21and23–27,
focusingontheflangezone,thehigherwearrateduringthefirst
Fig.21.Evolutionofthewheelprofileonther0rail.
Fig.22. Cumulativedistributionsoftheremovedwheelmaterial.
Fig.23.Evolutionofthewheelprofileonther0railintheflangezone.
Fig.24.Evolutionofthewheelprofileonther1rail.
stepscanbeobservedbecauseoftheinitialnon-conformalcontact
thatcharacterizesthecouplingbetweentheORES1002wheel
pro-fileandtheUIC60railprofilewithaninclinationof˛p=1/20rad;
thentheratedecreasesbecomingmoreregularandconstantinthe
laststeps,whenthecontactismoreandmoreconformal.
Alsoasregardsthewheelprofileevolution(asforthereference
quotas)thetrendremainsquantitativelyandqualitativelythesame
astherailwearraises(seeFigs.24–27).
In Figs.28 and 30theevolutionoftherailprofileis shown,
described bymeansofnsr=5discrete stepand withNstepequal
to400,000(the vehiclenumber Ntot,correspondingtothetotal
studied tonnage Mtot, is 2,000,000). In Fig. 29 the cumulative
distribution of the removed material on the rail proflie in zr
Fig.26.Evolutionofthewheelprofileonther3rail.
Fig.27.Evolutionofthewheelprofileonther4rail.
Fig.28.Evolutionoftherailprofile.
Fig.29.Cumulativedistributionsoftheremovedrailmaterial.
directionrKr(yr)=
Kri=1ri(yr)expressedasafunctionofyr(with
1≤Kr≤nsr) is shown, where ir(yr) is the removed material
betweentwosubsequentdiscretestepsoftherailprofile
evolu-tion.Thevalueoftotaltonnagetakenintoaccount(Mtot=200Mt)
causesanappreciablewearontherailhead,whileitisnotsufficient
toproduceanhighwearalsoontherailshouder(Fig.30).
Fig.30. Evolutionoftherailprofileinheadzone.
6. Conclusions
InthisworktheAuthorspresentedacompletemodelforthe
wheelandrailwearpredictioninrailwayapplication,developed
thanks tothe collaboration withTrenitalia S.p.A and Rete
Fer-roviariaItaliana(RFI),whichprovidedthenecessarytechnicaland
experimentaldataforthemodelvalidation. Thewholemodelis
madeupoftwomutuallyinteractiveparts.Thefirstoneevaluates
thevehicledynamicsandcomprisesboththemultibodymodelof
thevehicleimplementedinSimpackRailandaglobalwheel–rail
contactmodel(developedbytheAuthorsinpreviousworks)for
thecalculationofthecontactpointsandofthecontactforces.The
secondoneisthewearmodelwhich,startingfromtheoutputsof
themultibodysimulations,evaluatestheamountofmaterialtobe
removedduetowear.Theinteractionbetweenthetwopartsisnota
continuoustimeprocessbutoccursatdiscretesteps;consequently
theevolutionofthewheelandrailgeometryisdescribedthrough
severalintermediateprofiles.
Inparticularasuitableupdatealgorithmhasbeendevelopedin
ordertoconsiderthedifferenttimescalecharacterizingthewheel
andrailwearevolution:thewheelwearhasbeenstudiedbasing
onthedistancetraveledbyvehicle,whiletherailwearhasbeen
evaluatedbasingonthetotaltonnageburdenontherailwaytrack.
The whole model has been validated on a critical scenario
in termsof wearinItalian railways:theALSTOM DMU Aln501
MinuettocirculatingontheAosta-PreSaintDidierrailwayline.A
statisticalapproachtodescribethetrackhasbeenusedtoreduce
thetotalcomputationaleffort.
As regards the wheel wear, the results obtained from the
dynamic simulationshavebeencompared withthe
experimen-taldataprovidedbyTrenitaliawhile,concerningtherailwear,the
comparisonwithexperimentalcriteriabasedonthetotaltonnage
burdenontherailwaytrackhavebeenconsidered.Thedeveloped
modelreproducesquitegoodtheevolutionofalltheprofile
charac-teristicdimensionsdescribinginsatisfyingwaythewearprogress
bothonthewheelandontherail.
Theresultsobtainedforthewheelprofileevolutionhighlights
how,intheparticularoperatingconditionsofthestudiedrailway
line,thewearisquitesevereandstronglylocalizedonthewheel
flange,leadingtofrequentmaintenanceinterventions.Asregards
therailprofileevolution,thewearismainlyfocusedontherail
headduetotheinitiallowconformityofthecouplingORES
1002-UIC60with˛p=1/20radwhilethetotaltonnageconsideredisnot
sufficienttoproduceanhighwearalsoontheshouderoftherail.
Futuredevelopmentswillbebasedonfurtherexperimentaldata
(relative tootherrailway trackwitha higher mileagethan the
Aosta-PreSaintDidierline)alwaysprovidedbyTrenitaliaandRFI
and referredtoadvancedwearonthewheel(especiallyonthe
wheeltread)and ontherail. Inthis wayotheranalysiswillbe
carriedoutinordertofurthervalidatethewholemodel.
Moreoverotherwearmechanismsinadditiontoabrasivewear
(plasticwear,pitting wearetc.),thatmaybecomequite
impor-tantespeciallyinconditionsofadvancedwear(inparticularonthe
wheelflange),willbeconsideredwithinthewearmodel.Finally,
acodeoptimization,concerningthevehiclemodel(inparticular
theglobalcontactmodel),thewearmodelandthewholeloop,is
scheduledforthefutureinordertoreducethesimulationtimes.
Acknowledgements
AuthorswouldliketothankEngg.R.CheliandG.Grandeof
Tren-italiaS.p.A.forprovidingandgivingthepermissiontoeditthedata
relativebothtothevehicleDMUAln501Minuettoandtothewheel