外文資料與中文翻譯外文資料:EvaluationofHonedCylinderBoresF.PuenteLeonDesignofSystemsonSilicon(DS2),ParqueTecnologicode

Valencia,C./CharlesRobertDarwin2,E-46980Paterna(Valencia),Spain

SubmittedbyG..Spur(1),Berlin,GermanyAbstractThequalityofthehoningtextureoncylinderboresofcombustionenginesplaysanimportantrolewithrespecttooilconsumption,noxiousemissions,andrunningperformance.Toevaluatehonedsurfacesobjectively,featuresdescribingthesurfacetextureareextractedfrom2-Ddataofthesurface.Thepaperfocusesontwocrucialstagesofthedataanalysis:thepreprocessing,whichaimsatsuppressingirrele-vantcomponentsandenhancingtheinformationofinterest,andthefeatureextraction,whichyieldsreliablenumericalestimatesofthesurfacecharacteristicsofinterest,likethehoningangle,groovepa-rameters,surfacedefectsetc.Theassessmentresultscaneasilybeadaptedtouser-specificratings.Keywords:Honing,Surfacetexture,AutomatedvisualinspectionINTRODUCTIONCylinderboresofcombustionenginesarefinishedbyhoning.Theresultingsurfacetexturemainlyconsistsoftwobandsofhelicalgroovesplacedstochasticallyandappearingatdifferentanglestothecylinderaxis.Thetexturequalityishighlyimportantfordryoperationproperties,oilconsumption,noxiousemissions,andrunningperformance.Uptonow,expertsarestillratinghonedsurfacesvisuallybasedonmicroscopicimages.Thismethodistedious,subjective,andtimeconsuming.Togetobjectiveandreproducibleresults,anautomatedmethodofinspectionisnecessary.INSPECTIONAPPROACHSurfacedataTherearebasicallydifferentwaystomeasurethetextureofahonedsurface;seeTable1.Typically,amechanicalstylusonlyperformsaI-Dmeasurementofthesurfaceprofile.Incontrasttothis,greylevelimagesandopticalprofilometersprovide2-Ddatainareasonableamountoftime.Becausethelateral-geometricfeaturesofhoningtexturescanonlybeanalysedwith2-Ddata,inthefollowingwewillconcentrateonsuchkindsofdata.Othercharacteris-ticsrelatedtothedifferentmeasurementprinciplesinvestigatedarealsoincludedinthistable.Asignalmodeldescribingtheessentialcharacteristicsofahoningtextureconstitutesthebasisoftheevaluationapproachpresentedinthispaper.Basedonthismodel,clearandmathematicallywell-definedfeaturesareintroduced,whichenableareproducibleandobjectiveassessmentofthetexture.Thisstrategydiffersfrommanypopularmethods-suchasthoserelyingonneuralnetworks-,whichareoftentreatedasa'blackbox'[I].ThefeatureschosenareinspiredbytheHoningAtlas[2],bymanyopinionsofexperts,andhavealsobeenex-tendedbyaddingnewvolumetricparametersforthecaseofanalysingprofiledata.Thisresultsinanextensivesetoffeaturesthatcanbecustomizedtomatchtheneedsofindividualusers.PropertiesofhoningtexturesFigure1showssomeofthepropertiesofhoningtex-tures,baseduponwhichfeaturesaretobedefined.Themostpopularonesaretheroughnessparameters,suchasthosebasedontheBearingRatioCurve(AbbottCurve)[3],andR,,R,andR,,[4].However,dealingwithhonedsurfaces,itisimportanttodefinefeaturesthattakethelateralgeometryintoaccount.Thisway,mostrelevanttexturepeculiaritiescanbedescribed,suchasthehoningangle,materialsmearings,grooveinterrupts,straygrooves,holes,foreignbodies,andflakes,asshowninFigure1.Inaddition,featuresdescribingthebalanceofgrooves,presenceofplateaus,shapeofgrooves,cracks,residualturninggrooves,andchattermarksarealsoneeded.AutomatedinspectionFigure2showsanoverviewoftheabilitiesandaimsofautomatedinspectioninqualitycontrolappliedtothehoningprocess.A2-Dor3-Dsensorprovidesdatag(x)ofthehonedsurface,where

x=(~,yE)R~2denotesthelateralspatialcoordinates.Thegreycolouredblocksofthediagramarepartofthesensordataprocessing.Theoutputsofthesystemcanbeusedsimplyasastatementaboutsurfacequality,togivealarmscausinganinterruptofthemachiningprocess,oritcanbefedbackviaacontrollertoregulatethehoningprocess,becausethehoningtexturecontainsinformationaboutbothfunctional-ityandalsomachiningprocesdndependentlyofthefactwhetherapost-honingbrushingisperformedornot.Inthefollowingsections,wewillfocusontwocrucialstagesoftheautomatedinspection:thepreprocessingofthesensordataandthefeatureextraction,andwewillgivesomeexamplestothesesteps.PREPROCESSINGThegoalofthepreprocessingistosuppressirrelevantcomponents,namelytheinhomogeneitiesi(x)andthedisturbancesb(x),whileenhancingtheinformationofinterest,i.e.thetexturet(x).Inthecaseofacquisitionofimagedata,theinhomogeneitiesi(x)couldbeduetoMeehanicalstylusGreyleveHimageOpticalprofilanbetryMeasurementregionVD△口2-DDepthinformeflorYsbNaYesLaberalgeometrioinfbrrrisiiDnNdYesYesCoveringtheenlireEurfaDBVerytimH-cansumiing眥怖hreasennoieeffortVeryijmE-cansumingComputationalprocessingexpenseTowHighHighNnn-cflntacrmeasurempntNaYesYesStandardizedparametersYesNdYesTable1:Comparisonbetweenmechanicalstylusdevices,greylevelimages,andopticalprofilometersgreyironcylinderMdrillingiturningm日chin巳settings由區(qū)dim

ofun^anljed

pmcessstatesatermsMdrillingiturningm日chin巳settings由區(qū)dim

ofun^anljed

pmcessstatesaterms-g'LlGVCn-i：hatterirqetc——>honingprocesspr叩fk跪ipffeatu'Bestoction—leoinngmachirBsettingsquarrttative2-D/3-Dwfiser一」「SPCtimeseriesrahonirgdesiredvaluesCD^trolle.r伯Mln白cescrictcrstrend-,：iJpj'LmeterjFigure2:Automatedinspectionofhonedsurfaces.2-DgreyI敢elmage2-DSEMmmgeFigure1:Honingtexturesshowinglateralfeaturesanddefects:(a)materialsmearings,groove

interrupts;2-DgreyI敢elmage2-DSEMmmgesignalofinterestinversetransformseparationtransformIrrelevantComponentsFigure3:Principleofthepreprocessing.spatialvariationsofsurfaceillumination.Othercompo-nentsassignedtothedisturbancesb(x)includee.g.deviationsfromtheidealcourseofthegroovesanddefects,suchasmaterialsmearings,flakesetc.Weuseasignalmodelthatdescribesthesensordatag(x)asacombinationofthetexturet(x)andtheirrele-vantcomponentsi(x)andb(x):Tobeabletorecovertheinformationofinterestt(x),anassumptionisnecessary:thedifferentcomponentshavetobemathematicallydistinguishable.AsshowninFigure3,atransformmapstherawdatag(x)suchthatastrictseparationoftheircomponentsisobtained.Then,theundesiredcomponentsaresup-pressed,andfinallyaninversetransformisperformedthatyieldstheresultsofthepreprocessing.Thebenefitsofthisprocedureincludeasimplificationofthefeatureextraction,andamorerobustimageprocess-ing,asshowninthefollowingexamples.3.1HomogenizationWhenagroovetextureisdegradedbyanintensityinhomogeneityi(x)duetothedataacquisitionprocess,e.g.duetoaninhomogeneouslighting,ahomogenizationcanbeperformedtosuppressthisunwantedcomponent[6].Figure4showsanexampleofthisoperationforaplaningtexture.Ontheleftsideofthefigure,theoriginaltextureisshown.Thecentralimagerepresentstheresultofastandardhomogenizationmethod-thehomomorphicFigure4:Homogenization:(left)planningtexture;(centre)homomorphicfiltering;(right)homogenizationresult.Figure5:Texturedecomposition:(left)honingtexture;(centre)groovetexture;(right)backgroundtexture.—profileJpcqfiIp+—Icvz-passUlberlineFigure6:Referencesurface:problemswithconventional

low-passfilters.filtering,whichassumesamultiplicativecombinationoftextureandinhomogeneity.Especiallyintheupperleftcorner,thisimageshowsaverypoorcontrast.Theimageontherightresultsfromthemodel-basedapproachaccordingtoFigure3.Inthiscase,ahomogenizationofthelocalmeanvalueandthelocalcontrasthasbeenperformedbasedonamodelthatconsidersamixedadditiveandmultiplicativecombinationofbothsignalofinterestanddisturbinginhomogeneity[6].Theresultisclearlymorehomogeneousthantheformeroneandenablesamorerobustanalysisofthetexture.3.2TexturedecompositionThenextexampleconcernsthedecompositionofthehoningtexturetoeasethefeatureextraction.Duetothecomplexityofthehoningtexture,theextractionofrele-vantfeaturesneededfortheinspectiontaskcouldbesimplifiedconsiderably,ifthepartialtexturesconstitutingthesignalg(x)accordingtoEq.(1)wereavailable.Thus,itwouldbeadvantageoustodevelopamethodtosepa-ratethetextureg(x)intoacomponentt(x)containingthestraightstructures(i.e.thegrooves)andanotheroneb(x)showingtheisotropiccomponents(i.e.thebackground,includingdefectsandobjects).Inthiscase,ahomogene-oustexturewillbeassumed.Fortunately,averyefficientalgorithmtoperformthisseparationalreadyexists[7].TheleftsideofFigure5showsanoriginalhoningtexture;theothertwoimagesrepresenttheresultsoftheadaptivetexturedecomposi-tioncomputedwiththisalgorithm.Inthegroovetexture,onlytheidealgroovescanbeseen,whereastheback-groundimagecontainsalldeviationsfromtheidealgroovecourseaswellasdefectsandotherobjects.Foramorecomprehensivediscussionoftheseparationalgo-rithm,interestedreadersarereferredto[7]Figure7:Originalhonedsurfaceandreferencesurface.greylevelimageperiodogramisdalprojectionP:P2F^gurs8.Eslirnartionofhoningangle.ReferencesurfaceFinally,thedefinitionofareferencesurfacetoeliminatetheshapecomponentwillbepresented.ThegraphinFigure6representsatracethroughtheprofileofahonedsurface.Thesmoothlinedescribestheshapecomponenttobesuppressed.However,conventionallow-passfiltersleadtodistortionsintheareaofthegrooves,asshowninthecaseofthedashedline.Wehavefacedthisproblembydevelopinganiterative2-Dfilter-amodifiedGaus-sia