SPECIALISSUEPAPER499Machinevisionconditionmonitoringofheavy-axleloadra - PDF document

SPECIALISSUEPAPER499Machinevisionconditionmonitoringofheavy-axleloadrailcarstructuralunderframecomponents BWSchlake,STodorovic,JREdwards,JMHart,NAhujaCPLBarkan Correspondingauthor:RailroadEngineeringProgram,Depart-mentofCivilandEnvironmentalEngineering,UniversityofIllinois, JRRT376Proc.IMechEVol.224PartF:J.RailandRapidTransit 500BWSchlake,STodorovic,JREdwards,JMHart,NAhuja,andCPLBarkan Asaresult,USrailwayshaveprogressivelymovedawayfromreactivemaintenancetoplannedandscheduledcomponentreplacementinordertoimproveef-ciencyandreducecosts[].Consequently,theAsso-ciationofAmericanRailroads(AAR)andtheTrans-portationTechnologyCenter,Inc.(TTCI)initiatedtheAdvancedTechnologySafetyInitiative(ATSI)andaprogrammecalledTechnologyDrivenTrainInspec-tion(TDTI)todevelopandimplementautomatedinspectiontechnologies[TheobjectiveofATSIandTDTIistoprovidesafer,moreefcient,andtraceablemeansofrollingstockinspectionbyautomatingthemechanicalinspectionprocessthroughavarietyoftechnologies.Exam-plesofthesetechnologiesincludewheelimpactloaddetectors(WILDs),truck(bogie)performancedetectors(TPDs),acousticbearingdetectors(ABDs),andmachinevisioninspectionoftruckcomponents,safetyappliances,andbrakeshoes[].TheplanforTDTIistoimplementanintegratednet-workofautomaticwayside(lineside)systemscapa-bleofinspectingandmonitoringtheNorthAmer-icanfreightcareetwithtwoprincipalobjectives.Therstgoalistocost-effectivelymaintaincom-pliancewithFRAregulationsandrailroad-specicmaintenanceandoperationalstandards,andthesec-ondistoimprovetheoveralleffectivenessandef-ciencyoftherailcarinspectionandmaintenanceprocess.InorderforTDTItoprovidesubstantialimprove-menttotheinspectionprocess,eachcomponentandsystemontherailcarmustbeaddressed.Ifnot,railcarsandtrainswillstillneedtostopinordertomanuallyinspectthecomponentsexcludedfromautomatedinspection.Anautomatedsystemthatonlyaddressesalimitedselectionofinspec-tiontasks,oronlyinspectscertaincars,wouldofferincrementalandqualitativebenets,butitmaynotprovidesufcientsavingstocostjustifytheinvest-mentintheseexpensivetechnologies[].Conse-quently,thenalwaysideinspectionsystemsshouldbecomprehensiveinscope,inspectingasmanyaspectsofthecaraspossible[].Inadditiontothewidevarietyofcomponentsrequiringinspec-tion,therearealsomanyvariationsinthenatureofcomponentdefects,symptomsofinterest,andtherequiredmeansofascertainingcomponentcondi-tion.Asaresult,therequisitetechnologiescapableofaddressingtheseinspectiondemandsmustvaryindesignandapplication.TheAARhasinitiatedresearchanddevelopmentonavarietyofdifferentautomatedinspectiontechnologiesthatwilladdressmanyaspectsofthefederallymandatedfreightcarinspections,includingbrakeapplicationandreleaseverication,brakeshoethicknessmeasure-ment,safetyapplianceconditioninspection,wheeldefectinspection,andwheelprolemeasurementement3,6,8].2BACKGROUND2.1StructuralunderframecomponentsOnecomponentoftheTDTIinitiativeisthedevelop-mentofasystemknownasAutomatedInspectionofStructuralComponents(AISC)thatwillusecamerasandcomputer-aidedimage-searchmethodstoinspectfreightcarunderframes.Steelstructuralunderframecomponentscontributetothestructuralintegrityoftherailcarbysupportingthecarbodyandladingandtransmittinglongitudinalbuffanddraftforces(slackaction)withinthetrain.Onmanytypesoffreightcarstheprincipalstructuralmemberoftheunderframeisthecentresill,extendinglongitudinallyalongthecen-trefromoneendofthecartotheother.Thecentresillisthelargestelementintheunderframestruc-ture,supportingverticalloadsandalsotransmittingthemajorityofbuffanddraftforcesthroughthecarcar15].Inadditiontothecentresill,severalothercom-ponentsareneededtosupportvertical,longitudinal,static,anddynamicloadswhilethecarisintran-sit.Thesecomponentsincludesidesills,bodybolsters,andcrossbearers.Somecarsalsohavesmallercrossmembers(sometimescalledcrossties)andsmallerlongitudinalmembersknownasstringersoroorsup-ports.Unlikeotherunderframestructuralmembers,neithercrosstiesnorstringersbearsubstantialloads.Thesidesillsarelongitudinalmembersrunningalongeachsideofthecar.Theyareconnectedtothecen-tresillbyvariouscrossmembersthatruntransverselyfromthesidesillstothecentresill.Thetwobodybol-stersarethelargestofthesetransversemembersandarelocatedneareachendofthecar.Besidestheirroleasmajortransversemembersoftheunderframestruc-ture,theysupportthecarbodyatopitstrucks(bogies).Crossbearers(andcrossties)areadditionaltransversemembersconnectingthesidesillstothecentresillandfurtherhelptodistributeandsupportverticalloads.Allofthesecomponentscombinetoformastructuralsystemthatmaintainsthecar’scamberandstructuralintegrity.2.1.1StructuralunderframedefectsFreightcarstructuralunderframecomponentsaresubjecttocyclicloading,shock,andvibrationwhileinservice.Cycliclongitudinalloading,alsocalledslackaction,occursintheformofbuffanddraftforces.Themajorityoftheseforcesareabsorbedbyacar’sdraftsystem,butextremeloadinginbuff(compression)ordraft(tension)mustbeabsorbedbythecentresillandtheaccompanyingstructuralcomponents.Lateralcyclicloadingoccursincurvesascentripetalforcesgeneratehigherloadsononesideoftherailcar.Aportionoftheseloadsarecarriedbythesidesillandtransferredtothecentresillthroughthecrossbear-ers.Damageandrepeatedloadingandunloadingcan Proc.IMechEVol.224PartF:J.RailandRapidTransitJRRT376 502BWSchlake,STodorovic,JREdwards,JMHart,NAhuja,andCPLBarkan aperpendicularviewofthetruckwithrespecttothetrack,andalgorithmsweredevelopedtobothdetectthelocationsofbrakecomponentsandspringsetsandidentifymissingbearingendcapbolts.ThisresearchprovidedabasisforsubsequentresearchontheAuto-matedSafetyApplianceInspectionSystem(ASAIS)(ASAIS)12].ASAISdetectsdeformedladders,handholds,andbrakewheelsandusesvisuallearningtechniquestodeterminethedifferencebetweenFRAdefectsneed-ingimmediateattentionanddeformationsthatarelesscritical.TheresultsandmethodsdevelopedintheseprojectshavebeenincorporatedintotheAAR’sTDTIprogrammeandarecurrentlybeingdevelopedbytechnologycompaniesandadoptedbyrailroadsforeldtesting.SubsequentUIUCresearchdemonstratedthefea-sibilityofusingmachinevisiontodetectdefectsandotheranomaliesontheunderbodiesofpassen-gercarsandlocomotives[].Imageacquisitionandmachinevisiontechniquesweredevelopedtorecordimagesandinspectrollingstockandloco-motiveundercarriages.Algorithmsusingimagescap-turedinboththevisibleandtheinfraredspectrademonstratedthatmissing,damaged,oroverheatedcomponentscouldbedetected,aswellasincipientfailuresandforeignobjectsbeneaththecars.Videosoftrainswererecordedastheymovedoverastation-arycameramountedbetweentherailsinarepairpitbeneaththetracks.Thecombinationofinformationfromboththethermalandthevisiblespectraiden-tiedcertaindefectsthatmighthaveotherwisegoneunnoticedbyhumaninspectorsinthecourseofrou-tinevisualinspections.Thisresearchaddressedsomeoftheproblemsassociatedwithacquiringimagesfrombeneatharailcar:aninherentlychallengingloca-tionduetolightingrequirements,spaceconstraints,anddifcultiesinvolvedwithkeepingtheequipmentcleanandprotected.Hardware,algorithms,andtech-nicalmethodologiesforimageacquisitiondevelopedintheseearlierprojectswereadaptedandexpandedtodevelopasystemformachinevisioninspectionoffreightcarunderframesintheAISCprojectdescribedinthisarticle.2.4RegulatorycomplianceTheFRAregulationsforfreightcarinspectionformedthebasisfordeterminingwhichcomponentswouldbeinspectedbyAISC.Section215.121ofTitle49intheUSCodeofFederalRegulations(CFR)governstheinspectionoffreightcarbodiesandtwoofthesixpartsinthissectionpertaintotheinspectionofstructuralcomponents[].AccordingtotheFRAregulations,thecentresillmaynotbebroken,crackedmorethan15.24cm(6in),orbent/buckledmorethan6.35cm(2.5in)inany1.83m(6ft)length.Specicparametersareestablishedfortheallowablemagnitudeofcracksandbucklingbecausethesedefectsmayunderminetheintegrityofthecentresill,resultinginafailure[Therefore,theseregulationsareintendedtoidentifypotentiallyhazardouscarssothattheywillberepairedbeforeanin-servicefailure.DuringFRAmotivepowerandequipment(MP&E)inspections,inspectorshavemultipleenforcementoptions.Theinspectormaytakeexceptiontotheconditionofastructuralcomponentandissueawarningtotheoperatingrailroadofpos-siblemonetarypenaltiesifthedefectisnotrepairedimmediately.Whendeemednecessary,inspectorscanalsoissueviolationshavingmonetarypenaltiesrang-ingfrom$2500to$6000dependingonthetype,severity,andlocationofthedefect[2.5ResearchfocusInordertodeterminewhichstructuralelementsshouldhavethehighestpriorityamongAISCinspec-tiontasks,railcarinspectiondatafromtheFRAOfceofSafetywereanalysedforthetimeperiod2000–2007.Inspectiondatapertainingonlytorailcarunderframecomponentswereconsideredinthisanalysis.Fifty-ninepercentofallstructuralcomponentdefectsiden-tiedbyFRAMP&Einspectorswerebroken,cracked,bent,orbuckledcentresills,whereastheremaining41percentweredefectivesidesills,bodybolsters,orcrossbearers(Fig.1).Thesedatasuggestthatdefectsinthecentresillarethemostfrequenttypeamongfreightcarstructuralunderframecomponents.ThisisconsistentwithFRAtrainaccidentdatafrom1999to2008.Overthis10-yearperiod,bentorbrokencentresillswereresponsiblefor75trainaccidentsonUSClassIrailroadsincompar-isontoonly31accidentsduetobrokensidesillsandonlyoneaccidentduetoadefectivebodybolster[Giventheimportanceofcentresillsinprovidingloadbearingcapacityandtheirroleintransmittingbuff Fig.1Averagenumberofannualstructuralunder-framedefectsrecordedbyFRAMP&Einspectors Proc.IMechEVol.224PartF:J.RailandRapidTransitJRRT376 504BWSchlake,STodorovic,JREdwards,JMHart,NAhuja,andCPLBarkan arrangementwasusedduringtestingattheNorfolkSouthern(NS)locomotiverepairfacilityinDecatur,Illinois(Fig.3(b)).Thetestset-upincludedtwocamerasplacedbelowtherails.Inanarrangementsimilartotheset-upusedatMonticello,camera1waslocated1.65m(65in)belowthetopofrail,centredbetweenthetworailsandaimedstraightupward,90fromhorizontal(Fig.3(a)).Thevideo-imagecollectionsystemforthiscameraviewwasaDragony2camerarecordingat15framespersecond,witha4.8mmlensandanf/1.8aperture.Illuminationwasprovidedbyeight575W(115V)halo-genlights,eachwithparabolicreectorsandmediumoodlenses.Thelightswereorientedontheoorofthepitinacirclearoundthecamera,andtheintensityofeachlightcouldbeindividuallyadjusted.Usingahandheldlightmeter,themaximumluminousinten-sityofeachlightwasdeterminedtobe40900lx(3880footcandlesatadistanceof3ft).Thelightswereaimedupwardbutadjustedinwardlyatvariousanglestopro-videevenilluminationofthecamera’sentireeldofview.Halfofthelightswereorientedtoilluminatethecentresill,whereastheotherfourlightswereposi-tionedathigherangles(closerto90vertically)inordertoilluminatethemoredistantportionsoftheunderbody(i.e.thetopsectionsofthehoppers).Camera2waspositioned1.22m(48in)fromtheeldsideoftherail,0.76m(30in)belowthetopofrail,orientedperpendiculartothetrackandaimedupward45abovehorizontal(Fig.3(a)).EquipmentforthiscameraviewincludedaMarlincamerarecordingat15framespersecondanda6mmlenswithanf/1.4aperture.Fourindividuallyadjustablehalogenlights(withthesamespecicationsasthoseusedforcamera1)providedilluminationforcamera2,eachorientedatapproximately45abovehorizontal.TestswererunusingtwoNSgondolacarsandoneNScoveredhop-percarbyrollingthempastthecamerasat5–8km/h(3–5mile/h).Fourteendifferentvideoswererecordedduringtesting,andtheimagedatawereconvertedintopanoramicimages(Fig.4).Sincethetwogondolacarswerealmostidenticaltoeachother,imagesofonlyoneofthecarsareshown.ResultsfromthetestingatDecaturshowmuchmoreevenilluminationforthecoveredhoppercarimage(Fig.4(a))comparedtothepreviouslyrecordedhoppercarpanorama(Fig.2).Thepanoramaofthegondolaunderbody(Fig.4(b))providesaclearviewofthecentresill,crossbearers,andcrossties(thinnerlateralmembersbetweenthecrossbearers).Othercompo-nentsofthegondolathatareclearlyvisibleincludethebrakereservoir,brakecylinder,andtheentirefoun-dationofthebrakingsystem.Inordertodeterminetheresolutionofthepanoramicimages,engineeringdrawingswereacquiredfromNSforeachofthecarsthatweretested.Bymeasuringcertaincomponentsinthepanoramicimageinpixelsanddividingbytheactuallengthsofthosecomponents,thepixel-to-cmratio(i.e.theimageresolution)wasdeterminedtobe3.03pixelspercentimetre(7.8pixelsperinch).Eachpanoramicimagewasdevelopedbycombiningthecentrestripsofover400videoframes,witheachstriphavingameanstripsizeofapproximately12pixels.Thepanoramasfromcamera2aremuchlongerduetothefactthatthiscamerawaslocatedclosertothetrackthancamera1.Asaresult,theimagesinFigs4(c)and(d)onlyrepresentonehalfofeachrailcar.Theimagesfromthiscameraviewarevaluablebecausecracksorbreaksinthesideofthecentresillwouldbevisiblefromthisangle.Thisviewcanalsobeusedtoinspectthecamberofthecar,todeterminewhetherthecentresillissaggingordeformed. Fig.4(aandd)Panoramicimagesofhoppercarand(bandc)gondolacarfromcamera1(aandb)andcamera2(candd) Proc.IMechEVol.224PartF:J.RailandRapidTransitJRRT376