• Keine Ergebnisse gefunden

Why homogenization should be the averaging method of choice in hydrodynamic lub...

N/A
N/A
Protected

Academic year: 2022

Aktie "Why homogenization should be the averaging method of choice in hydrodynamic lub..."

Copied!
11
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

ContentslistsavailableatScienceDirect

Applications in Engineering Science

journalhomepage:www.elsevier.com/locate/apples

Why homogenization should be the averaging method of choice in hydrodynamic lubrication

Michael Rom

a,

, Florian König

b

, Siegfried Müller

a

, Georg Jacobs

b

aInstitute for Geometry and Applied Mathematics, RWTH Aachen University, Templergraben 55, 52062 Aachen, Germany

bInstitute for Machine Elements and Systems Engineering, RWTH Aachen University, Schinkelstraße 10, 52062 Aachen, Germany

a r t i c le i n f o

Keywords:

Homogenization

Patir and Cheng’s average flow model Reynolds equation

Hydrodynamic lubrication Textured bearing Numerical investigation

a b s t r a ct

Forthecomputationofthehydrodynamicpressuredistributioninfluidfilmlubricationproblems,averaging techniquesarefrequentlyusedforapplicationswithroughortexturedsurfacestoreducethecomputationalcost.

Thepresentworkshowstheexcellentsuitabilityofhomogenizationforthispurpose.Forcomparison,themost frequentlyappliedaveragingmethodisused:theaverageflowmodelbyPatirandCheng.Thetwomodelsare brieflysummarizedtoshowthesimilarityoftheformulations,resultinginthesameimplementationeffort.By meansoftwotexturedapplications,thesuperiorityofthehomogenizationmethodisdemonstratedanderrors inherentintheaverageflowmodelarequantified.Homogenizationprovidesaveragedpressuredistributionswith asignificantlyhigheraccuracyandallowsforefficientlyrestoringlocalinformationtoresolvepressurepeaksby computingahigher-ordersolution.ThelatterisanaccurateapproximationofthesolutionoftheoriginalReynolds equationandcanbeobtainedwithnegligibleadditionalcomputationalcost.Furtherimportantadvantagesofthe homogenizationmethodsuchasitsextensibilitytosimultaneouslyaccountforroughnessandtexturesarepointed out.

1. Introduction

Forthecomputationofthehydrodynamicpressuredistributionin fluidfilm lubricationapplications, theReynolds equation (Reynolds, 1886)wasderivedfromtheNavier-Stokesequationsbyneglectingsev- eraltermsandaveragingoverthelubricatedgap.Thisreductionofthe dimensionfrom3Dto2DandthelinearityoftheReynoldsequationsig- nificantlyreducethecomputationalcost.However,adequatelycaptur- ingthesurfaceroughnessoftheapplicationrequiresahighly-resolved computationalmeshsuchthatevensolvingthe2Dlinearproblemcan betootime-consuming.Forthisreason,PatirandChengempiricallyde- velopedtheaverageflowmodel(PatirandCheng,1978;1979) which splitstheproblemintoonemacroscaleproblemwithoutroughnessand severalmicroscaleproblemswithroughnessandallowsforthecompu- tationofanaveragedpressuredistribution.Sinceamicroscaleproblem onlyrepresentsasmallpartofthewholeapplication,itcanbesolved efficientlydespitetheroughness.

ItquicklybecameclearthatPatirandCheng’saverageflowmodel onlyprovidesaccurateaveragedpressuredistributionsincaseswhere thesurfaceroughnessisisotropicorwherethemaindirectionsofthe roughnessanisotropyareparalleltothecoordinateaxes(orthotropic).

ThiswasforinstanceshownbyElrod(1979),whoconductedamultiple-

Correspondingauthor.

E-mail addresses: rom@igpm.rwth-aachen.de (M. Rom), florian.koenig@imse.rwth-aachen.de (F. König), mueller@igpm.rwth-aachen.de (S. Müller), georg.jacobs@imse.rwth-aachen.de(G.Jacobs).

scaleanalysisfortwo-dimensionalroughness,orbyTripp(1983),who extendedtheaverageflowmodelbyderivingthecorrecttensorialform ofthemodelusingastochasticapproach.Lo(1992)proposedtotrans- formperiodicanisotropicsurfacesintoisotropiconesbyusingamap- pingfunctionsuchthattheaverageflowmodelisapplicable.Worksby TealeandLebeck(1980)andLundeandTønder(1997)discussedsev- eralboundaryconditionsforthemicroscaleproblemstoimprovethe resultsoftheaverageflowmodel.

After Patir and Cheng’s average flow model, further averaging techniques with respect to the Reynolds equation were presented.

Phan-Thien (1982) used the mathematical concept of homogeniza- tion(Allaire,1992;PavliotisandStuart,2008)todevelophisso-called two-spacemethod.Mitsuyaetal.(1989)introducedanapproachbased ontheaveragingofthefilmthickness.Pratetal.(2002)appliedthe volumeaveragingmethod(Whitaker,1999)andobtainedamodelthat takesintoaccountcontactareasnotinvolvinganyfluidandcombines timeandspatialaveraging.

ApartfromPatirandCheng’saverageflowmodel,thehomogeniza- tiontechniquereceivedthemostattentionbecauseofitsaccurateav- eragingresults forsurfaceroughnessof arbitraryanisotropy. Several derivationsfortheReynoldsequationexistintheliterature,forinstance byBayadaandChambat(1988),BayadaandFaure(1989),Jai(1995),

https://doi.org/10.1016/j.apples.2021.100055

Received18December2020;Receivedinrevisedform1April2021;Accepted7May2021 Availableonline19May2021

2666-4968/© 2021TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)

(2)

Buscagliaetal.(2002),AlmqvistandDasht(2006)andAlmqvistetal.

(2007,2012).Thesederivationsdifferintheintendeduseofthehomog- enizedReynoldsequation,e.g.,forstationaryortransientproblems.For applicationswithisotropicororthotropicsurfaceroughness,thehomog- enizationmethodandtheaverageflowmodelprovidetheexactsame averagedpressuredistributions.Furthermore,homogenizationleadsto thesameaveragedpressuresasthevolumeaveragingapproachbyPrat etal.(2002)ifnocontactoccursandifonesurfaceissmoothandmov- ing,whereastheotheroneisroughandstationary.

Acomprehensiveliteraturereviewonaveragingtechniquesforthe Reynoldsequationisbeyondthescopeofthiswork,buttheoverview givenaboveshows thattheshortcomingsoftheaverage flowmodel arewellknownandthatnumerousimprovementsoralternativesexist.

However,forthestudyoflubricationapplicationswithroughsurfaces, theoriginalmodelbyPatirandChengisstillthemostcommonlyused method(Fatuetal.,2012;Gropperetal.,2016).Thisalsoappliesto thestudyoftexturedapplications,whichareofincreasingimportance duetotheirpotentialtoimprovehydrodynamicperformance(Gropper etal.,2016).

Thepopularityof theaverageflowmodelhasledtothedevelop- mentofmanyextensions,e.g.,byHarpandSalant(2001)toaccountfor cavitation,byWuandZheng(1989)orWilsonandMarsault(1998)to incorporatesurfacecontactorbyMengetal.(2009)tocombineelas- ticdeformation,inter-asperitycavitationandthermaleffects.Sincethe methodofhomogenizationiscertainlynotascommoninthefieldoftri- bologyastheaverageflowmodel,onlyfewextensions,e.g.,byBayada etal.(2005)forincorporatingcavitation,orimplementationsasinthe mixedlubricationmodelbySahlinetal.(2010a,b)areavailableinthe literature.

Comparisonsofaveragingmethodsmainlyconcentrateontheeval- uationofthesolutionsof themicroscaleproblems,i.e.,theresulting pressureandshearflowfactors.Suchcomparisonswerepresentedby Sahlinetal.(2010a), Almqvistetal.(2011b)andFatuetal.(2012). Macroscaleresults,i.e.,averagedpressuredistributions,obtainedfrom thehomogenizationmethodandfromthefilmthicknessaveragingap- proachbyMitsuyaetal.(1989)werecomparedbyJaiandBou-Saïd (2002).

Mostoftheaforementionedworksdiscussingthederivationofthe homogenizedReynoldsequationorthecomparisonofhomogenization withotheraveragingtechniquesmissonemajoradvantageoftheho- mogenizationmethodoratleastdonotquantifyitsbenefit:byasim- pleupscalingoftheaveragedpressuresolution,whichhasnegligible computationalcost,local informationarerestoredsuchthatpressure peaksareresolvedwithahighaccuracy.Thisallowsforamoreaccu- ratepredictionofperformanceparameterssuchasmaximumpressure orfrictionforcethanpossiblewithotheraveragingtechniques.Theup- scalingwasstudiedbyBayadaandFaure(1989)forajournalbearing withdifferentgroovesetups.Theirinvestigationmainlyconcentrateson theevolutionofrelativeerrorsintheload-carryingcapacity,obtained fromthehomogenizedortheupscaledpressuresolution,forincreas- ingroughnessfrequency.Sincetheload-carryingcapacityiscomputed byintegratingtheaveraged(homogenized)ortheoscillating(upscaled) pressure,theresultingvalues usuallydeviateonly slightly.However, theauthorsstateintheirconclusionthat“theanalysisofsomeother bearingcharacteristicswhicharerelatedwiththefloworthegradient ofthepressuremustnotbecomputedfromthegradientof𝑝0”,where 𝑝0denotestheaveragedpressuresolution.Thegradientofthepressure solutionisforinstanceinvolvedinthecomputationofthefrictionforce.

AnotherworkbyBayadaandJai(1996)demonstratestheupscalingby showingpressuredistributionsforultra-thingasfilmbearingsobtained withandwithoutupscaling.

Themainobjectiveofthisworkistodemonstratetheadvantageof thepressureupscalingofthehomogenizationmethod.Wealsoaimat emphasizinghoweasilytheupscalingisobtained.Thehomogenization methodiscomparedwiththeaverageflowmodelbyPatirandCheng sincethelatter isthemostwidely usedaveragingtechnique. Onthe

onehand,atheoreticalcomparisonisintendedtounderlinethesim- ilarityofthetwomethodsregardingthegoverningequations,bound- aryconditionsandcomputationalprocedure,whichleadstothesame implementationeffort.Eventhoughtheformulasarewellknown,this overviewwill helppointoutthesimilarityof thetwomethodsand, inparticular,howtoimplementthehomogenizationmethodwithup- scaling.Ontheotherhand,acomparisonofnumericalresultsismeant toverifythesignificanceof theerrorsintroducedbyusing theaver- age flowmodelinstead ofthehomogenization method.These errors do notonly originatefrom themissingupscalingor fromtheuse of anisotropicroughnessortexturesbutalsofromthepressureaveraging.

Hence,eveninthecaseofisotropic/orthotropicroughnessortextures,a considerableerroriscausedbyapplyingtheaverageflowmodel,which isavoidablewiththehomogenizationmethodatvirtuallynoadditional cost.

Tokeepthetheoreticalcomparisonandthenumericalinvestigation simpleandcomprehensible,wefocusonstationaryhydrodynamiclubri- cationwithiso-viscousandincompressiblelubricantsanddonottake intoaccountcavitation.Thisallowsfortheunadulteratedidentification oftheerrorsinherentintheaverageflowmodel.

Inaddition,wediscussfurtheradvantagesofthehomogenization method overtheaverage flowmodel. These comprisetheavoidance of modelinguncertaintiesoftheaverageflow modelpresentforsur- faceswithdeepdimples(Grützmacheretal.,2018;Königetal.,2020a;

Tomaniketal., 2020) andtheextensibilityoftheconcept byreiter- atedhomogenization(AllaireandBriane,1996),whichallowsforthe efficientinvestigationofsurfacesthatareroughandtextured.Thepro- cedureofreiteratedhomogenizationfortheReynoldsequationwasde- rivedbyAlmqvistetal.(2008)andMartin(2008),thelatterincorpo- ratingacavitationmodel.

Forthecomparisonofnumericalresults,weinvestigatetwotextured applications,namelyajournalbearingandaplane-inclinedsliderbear- ing.Inbothcases,twodifferentsetupsregardingthetexturesarestud- ied:(𝑖)withsymmetrictextures(isotropic/orthotropiccase)and(𝑖𝑖)with nonsymmetrictextures(anisotropiccase).Thesolutionsoftheaverage flowmodelandthehomogenizationmethodarecomparedwithcorre- spondingreferencesolutionsobtainedbysolvingtheoriginalReynolds equationwithhighly-resolvedcomputationalmeshes.Inaddition,the performanceparametersmaximumpressure,load-carryingcapacityand frictionforcearecomputedfromthesolutionsforaquantitativeerror investigation.Tothebestoftheauthors’knowledge,suchaquantitative comparisonisnotavailableintheliteraturesofar.

InSect.2,thetwoaveragingtechniquesareintroducedintermsof theirgoverningequationsandboundaryconditions.Thisisdonewith particularfocusonthegeneralcomputationalprocedureandtohigh- lightthesimilaritiesanddifferencesbetweenthetwomethods.Thead- vantagesofthehomogenizationmethodovertheaverageflowmodel aresummarizedinSect.3.Theapplicationofthemethodstogetherwith aqualitativeandquantitativecomparisonisdemonstratedforthetest casesmentioned aboveinSect.4.Finally,thearticle isconcludedin Sect.5.

2. Averageflowmodelandhomogenizationmethod

Inthiswork,we focusonhydrodynamic lubricationproblemsfor which thelubricatedgap isboundedby onestationarytexturedsur- faceandonemovingsmoothsurface.TheReynoldslubricationequa- tion(Reynolds,1886)describingsuchaproblemisbrieflyintroducedin Sect.2.1.Itisthestartingpointforthetwoaveragingtechniquesstudied inthefollowing.Thegeneralcomputationalprocedureforbothaverag- ingmethodsisdemonstratedinSect.2.2.Afterintroducingthedifferent filmthicknessrepresentationsusedbythetwomethodsinSect.2.3and somenotationfortheparameterdependencyregardingthemacro-and microscalecoordinatesinSect.2.4,wesummarizetheempiricallyde- rived averageflow modelbyPatir andCheng (1978,1979) andthe mathematically derived homogenizedReynolds equation, see for in-

(3)

Fig. 1. Illustration of the macro- and mi- croscaledomainsandthecomputationalpro- cedurefortheaverageflowmodelandtheho- mogenizationmethod.

stance(AlmqvistandDasht,2006;Almqvistetal.,2007;2012;Bayada andChambat,1988;BayadaandFaure,1989;Buscagliaetal., 2002;

Jai,1995),inSects.2.5and2.6,respectively.Thesimilaritiesanddif- ferencesbetweenthetwomethodsareemphasizedinSect.2.7. 2.1. Startingpoint:Reynoldsequation

Thetwo-dimensionalReynoldsequation(Reynolds,1886)iswidely used for fluid film lubrication problems. Its derivation from the three-dimensionalNavier-Stokesequationscan be foundforinstance inKhonsariandBooser(2017).Duetounderlyingassumptionssuchas negligibleinertiaforces,thevalidityoftheReynoldsequationislimited.

ItdependsontheReynoldsnumberoftheinvestigatedproblemandin caseofatexturedsurfaceontheaspectratioofthetextures(Dobrica andFillon,2009).Inthiswork,weonlyconsiderproblemsforwhich theReynoldsequationisvalid.

Asmentionedabove,westudyhydrodynamiclubricationproblems withonestationarytexturedsurfaceandonemovingsmoothsurface.

Neglectingcavitation,thestationaryincompressibleReynoldsequation foraniso-viscousandincompressiblelubricantreads

∇⋅(

(𝒙)3𝑝(𝒙))

=6𝜇𝒖𝑎⋅∇(𝒙) (1)

andissolvedforthehydrodynamicpressure𝑝(𝒙).Thelocalfilmthick- ness(𝒙)isgiven,e.g.,byafunction.Thedynamicviscosity𝜇andthe velocityvector𝒖𝑎=(𝑢𝑎,1,𝑢𝑎,2)𝑇 ofthemovingsurfaceareconstantand alsogiven.

2.2. Generalcomputationalprocedurefortheaveragingtechniques

Theobjective of theaveraging methodsis to avoid theneed for highly-resolvedcomputationalmeshesand,hence,toreducethecom- putationtimewhilepreservingahighsolutionaccuracy.Thisisdone bydividingtheprobleminto onemacroscaleproblemfor whichthe roughnessorthetexturesdonothavetoberesolvedandseveralsmall microscaleproblemswhichcanbesolvedefficientlyandprovidecoef- ficientsenteringthemacroscaleproblem.

ThecoefficientscomputedbytheaverageflowmodelbyPatirand Chengarecalledpressureandshearflowfactors.Thesearedetermined by solvinga reference problem on the microscale being representa- tivefortheroughness/texturesoftheproblemandoccurringperiodi- cally.Theflowfactorsdescribethedeviationoftheflowthroughthe rough/textured gap from the flow through a gap with smooth sur- faces.Incontrasttothis,thehomogenizationmethodis basedonan

asymptoticexpansionofthepressurewhichispluggedintotheoriginal Reynoldsequation(1).Scaleseparationthenleadstosimilarreference problemscalledcellproblems.Thedifferencesinthemethodsstemfrom theirderivations:theaverageflowmodelwasempiricallyderivedby investigatingtheflowbehavior,whereashomogenizationisavalidated mathematicalconcept.

Figure 1 exemplarily illustrates how an averaged pressure 𝑝avg

can becomputed withbothmethods. Thetwo-dimensionalCartesian macroscaledomainΩwithcoordinates𝒙=(𝑥1,𝑥2)𝑇 isoflength𝑙and width𝑤.Itcontains𝑛𝑡=8squaretextures(𝑛𝑡,1=4in𝑥1-directionand 𝑛𝑡,2=2in𝑥2-direction)oflength𝑙𝑡andwidth𝑤𝑡=𝑙𝑡.Thedashedlines indicate the distribution into eight elements of the same size with length 𝑙𝑡𝑒andwidth𝑤𝑡𝑒=𝑙𝑡𝑒,which wecalltexture elements.These areidentifiedbyindices(𝑖,𝑗)asgiveninFig.1.Themicroscaleproblem correspondstoonetextureelementandisrepresentedbythedomain withcoordinates

𝒚= (𝑦1

𝑦2

)

∶=

⎛⎜

⎜⎜

𝑥1𝑖𝑙𝑡𝑒

𝜀 𝑥2𝑗𝑤𝑡𝑒

𝜀

⎞⎟

⎟⎟

. (2)

Thescale𝜀oftheroughness/texturesisdefinedby𝜀∶=𝑙𝑡𝑒.Hence,the microscaledomainisscaledsuchthatitisoflengthoneandwidth 𝑤𝑡𝑒𝜀.Forsquaretextureelementswith𝑤𝑡𝑒=𝑙𝑡𝑒,thedomainisthe unitsquaresince𝑤𝑡𝑒𝜀=1.

Fortheaverageflowmodel,thevelocitymustbealignedtothe𝑥1- direction,i.e.,thecomponent𝑢𝑎,2 in𝑥2-directionhastobezerosuch that𝒖𝑎=(𝑢𝑎,1,0)𝑇.Thereisnosuchrestrictionforthehomogenization method.Thevelocityofthestationarysurfaceisgivenby𝒖𝑏=𝟎.

The only quantity which enters the microscale problems with macroscaleinformationisthelocalnominalfilmthicknessnom(𝒙)(av- erageflowmodel)orthelocalgapheight0(𝒙)(homogenization),see Sect.2.3forthedefinitions.Onthemicroscale,nomor0isconstant, whichresultsinaconstantgapheightforthecorrespondingmicroscale problem.Forthebest-possibleaccuracyofthemacroscalesolution,one microscaleproblemhastobesolvedforeachvalueofnom(𝒙)or0(𝒙) whichoccursinthediscretizedmacroscaledomain.Alternatively,the microscale problemsareonlysolvedforaselectionoftheheightsin whichcasethemissingvaluescanbeinterpolatedfromthecomputed ones.Onthemicroscale,thecoefficients(see𝜙𝑠(𝒙),etc.inFig.1)are computed bysolvingthemicroscale problemwithvarying boundary conditionsandsubsequentlyaveragingtheparticularsolutionoverthe reference/celldomain.Thesecoefficientsenterthemacroscaleproblem

(4)

Fig.2. Localfilmthicknessasdefinedintheaverageflowmodel(a)andthe homogenizationmethod(b).

whichisfinallysolvedwithzeroboundaryconditions,i.e.,𝑝avg=0on allfourboundaries.

2.3. Filmthicknessrepresentations

Theaverageflowmodelandthehomogenizationmethodusediffer- entrepresentationsofthefilmthickness.Thedifferencesaredepicted inFig.2foronetextureelementonthemicroscale.Notethat,forthe purposeofillustration,thefigurecontainsthe𝑧-coordinatewhichdoes notexistintheactual2D-domain,cf.Fig.1.Intheaverageflowmodel, seeFig.2(a),thenominalfilmthicknessnomisdefinedasthedistance betweenthetwocenterlinesoftheroughness/textures.Inoursetting, thelowersurfaceissmoothsuchthatthesurfaceitselfdefinestheend- pointofnom.Thelocaldistancefromtheroughness/texturecenterline isdenotedby𝛿.Incontrast,inthecaseofthehomogenizationmethod, thefilmthicknessiscomposedoftheactuallocalgapheight0andthe localroughness/textureheight1,seeFig.2(b).Insummary,thefilm thicknessesaredefinedby

(𝒙,𝒚)=nom(𝒙)+𝛿(𝒚), (averageflowmodel) (3)

(𝒙,𝒚)=0(𝒙)+1(𝒚). (homogenization) (4) Inbothcases,thelocalfilmthickness(𝒙,𝒚)isdividedintoamacroscale (𝒙)andamicroscale(𝒚)part.Notethatthenominalfilmthicknessnom

changesiftheroughness/textureheightchangesand,hence,formally alsodependson themicroscale(𝒚).However,aslongasno asperity contactsoccur,wehave

nom(𝒙,𝒚)=0(𝒙)+const.≡nom(𝒙). (5) 2.4. Notationforparameterdependency

Sections 2.5 and 2.6, respectively, summarize the average flow modelandthehomogenizationmethodin theorderof theircompu- tationalsteps.Theformulasofthetwomethodsaresimilarandgivenin awaysuchthattheyareeasilycomparable.Thedependencies(𝒙,𝒚)and (𝒚;𝒙)canbeexplainedbymeansofthefilmthickness:basically,the filmthicknessisinfluencedbyboththemacroscaleandthemicroscale, i.e.,(𝒙,𝒚).However,asdescribedabove,nom(averageflowmodel)or 0(homogenization)isaconstantvalueinthemicroscaleproblemsuch thatisvariedonlyby𝛿(averageflowmodel)or1(homogenization), seealso(3)and(4).Consequently,thenotation(𝒚;𝒙)indicatesthat isafunctionin𝒚dependingonaparticularfixedparameter𝒙. 2.5. AverageflowmodelbyPatirandCheng

1. Foreachlocalnominalfilmthicknessnom(𝒙): (a) Solvethethreereferenceproblems(microscale)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜓0(𝒚;𝒙))

=6𝜇 (𝑢𝑎,1

0 )

⋅∇𝒚(𝒚;𝒙), (6)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜓1(𝒚;𝒙))

=0, (7)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜓2(𝒚;𝒙))

=0 (8)

withDirichletandNeumannboundaryconditions

In(6),theright-handsideisgiveninitsgeneralformulation.In ourcasewithnoasperitycontacts,thegradient∇𝒚(𝒚;𝒙)canbe replacedby∇𝒚𝛿(𝒚).

(b) Computethethreeflowfactors 𝜙𝑠(𝒙)=− 1

6𝜎𝜇𝑢𝑎,1(𝒚;𝒙)3𝜕𝜓0(𝒚;𝒙)

𝜕𝑦1

𝑑𝒚, (shearflowfactor) (9)

𝜙𝑝,1(𝒙)= 1

nom(𝒙)3(𝒚;𝒙)3𝜕𝜓1(𝒚;𝒙)

𝜕𝑦1 𝑑𝒚,

(pressureflowfactor1) (10)

𝜙𝑝,2(𝒙)= 1

nom(𝒙)3(𝒚;𝒙)3𝜕𝜓2(𝒚;𝒙)

𝜕𝑦2 𝑑𝒚

(pressureflowfactor2) (11)

fromthereferenceproblemsolutions𝜓0,𝜓1,𝜓2,where𝜎 isthe standarddeviationoftheroughness.

2. SolvethemodifiedReynoldsequation(macroscale)

𝒙⋅ ((

𝜙𝑝,1(𝒙) 0 0 𝜙𝑝,2(𝒙)

)

nom(𝒙)3𝒙̄𝑝(𝒙) )

=6𝜎𝜇𝑢𝑎,1𝒙⋅ (𝜙𝑠(𝒙)

0 )

+6𝜇 (𝑢𝑎,1

0 )

⋅∇𝒙̄ℎ(𝒙) (12)

with ̄ℎ(𝒙)=∫(𝒚;𝒙)𝑑𝒚 and boundary conditions as shown in Fig.1with𝑝avg∶= ̄𝑝.Notethat̄ℎ(𝒙)=nom(𝒙)inthecaseofnoas- peritycontacts.

2.6. Homogenization

1. Foreachlocalgapheight0(𝒙):

(a) Solvethethreecellproblems(microscale)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜒0(𝒚;𝒙))

=6𝜇 (𝑢𝑎,1

𝑢𝑎,2

)

⋅∇𝒚1(𝒚), (13)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜒1(𝒚;𝒙))

=−∇𝒚⋅( (𝒚;𝒙)3𝒆1

), (14)

𝒚⋅(

(𝒚;𝒙)3𝒚𝜒2(𝒚;𝒙))

=−∇𝒚⋅( (𝒚;𝒙)3𝒆2

), (15)

(5)

where𝒆1=(1,0)𝑇 and𝒆2=(0,1)𝑇,withperiodicboundarycon- ditions

(b)Computethesixcoefficients 𝑏1(𝒙)=

(𝒚;𝒙)3𝜕𝜒0(𝒚;𝒙)

𝜕𝑦1

𝑑𝒚, 𝑏2(𝒙)=

(𝒚;𝒙)3𝜕𝜒0(𝒚;𝒙)

𝜕𝑦2

𝑑𝒚, (16) 𝑎11(𝒙)=

(𝒚;𝒙)3

(𝜕𝜒1(𝒚;𝒙)

𝜕𝑦1

+1 )

𝑑𝒚, 𝑎12(𝒙)=

(𝒚;𝒙)3𝜕𝜒2(𝒚;𝒙)

𝜕𝑦1 𝑑𝒚, (17)

𝑎21(𝒙)=

(𝒚;𝒙)3𝜕𝜒1(𝒚;𝒙)

𝜕𝑦2

𝑑𝒚, 𝑎22(𝒙)=

(𝒚;𝒙)3

(𝜕𝜒2(𝒚;𝒙)

𝜕𝑦2

+1 )

𝑑𝒚 (18)

fromthecellproblemsolutions𝜒0,𝜒1,𝜒2.

2. SolvethehomogenizedReynoldsequation(macroscale)

𝒙⋅ ((

𝑎11(𝒙) 𝑎12(𝒙) 𝑎21(𝒙) 𝑎22(𝒙) )

𝒙𝑝0(𝒙) )

= −∇𝒙⋅ (𝑏1(𝒙)

𝑏2(𝒙) )

+6𝜇 (𝑢𝑎,1

𝑢𝑎,2

)

⋅∇𝒙0(𝒙) (19) withboundaryconditionsasshowninFig.1with𝑝avg∶=𝑝0. 3. Computetheupscaledpressure

̃𝑝(𝒙,𝒚)=𝑝0(𝒙)+𝜀𝑝1(𝒙,𝒚) (20)

=𝑝0(𝒙)+𝜀 (

𝜒0(𝒚;𝒙)+𝜕𝑝0(𝒙)

𝜕𝑥1 𝜒1(𝒚;𝒙)+𝜕𝑝0(𝒙)

𝜕𝑥2 𝜒2(𝒚;𝒙) )

(20)

inapost-processingbysimplycombiningthecellproblemsolutions andthegradientoftheaveragedpressuresolution.Asdescribedin Sect.2.2,𝜀istheroughness/texturescalewhichinoursettingisthe lengthofonetextureelement,i.e.,𝜀=𝑙𝑡𝑒.Theupscaling(20)orig- inates from the asymptotic expansion 𝑝(𝒙,𝒚)=𝑝0(𝒙)+𝜀𝑝1(𝒙,𝒚)+ 𝜀2𝑝2(𝒙,𝒚)+…forthepressure,whichisinsertedintotheReynolds equation(1)toderivethehomogenizedReynoldsequation.Sincethe termsstartingfrom𝜀2𝑝2(𝒙,𝒚)aresmallcomparedwiththeleading terms,(20)providesaveryaccurateapproximationofthepressure𝑝 computedfromtheReynoldsequation(1)whileallowingforsignifi- cantlycoarsercomputationalmeshes.ThisisshowninSects.4.2and 4.3.

2.7. Similaritiesanddifferencesbetweenthemethods

Table1summarizesthedifferencesintheformulasoftheaverage flowmodelandthehomogenizationmethod.Theaveragedpressureso- lutions ̄𝑝and𝑝0 canonlybe equaliftheroughness/textureelements aresymmetric(Almqvistetal., 2011b) andiftheflowisattachedto

Table1

Differencesbetweentheaverageflowmodelandthehomogenizationmethod.

average flow model homogenization local film thickness = nom+ 𝛿 = 0+ 1

surface velocity 𝒖 𝑎= ( 𝑢 𝑎,1, 0) 𝑇 𝒖 𝑎= ( 𝑢 𝑎,1, 𝑢 𝑎,2) 𝑇 reference problem solutions 𝜓 0, 𝜓 1, 𝜓 2 𝜒0, 𝜒1, 𝜒2

coefficients 𝜙𝑝,1, 𝜙𝑝,2, 𝜙𝑠(flow factors) 𝑎 11, 𝑎 12, 𝑎 21, 𝑎 22, 𝑏 1, 𝑏 2

averaged pressure ̄𝑝 𝑝 0

upscaled pressure - ̃𝑝 = 𝑝 0+ 𝜀 𝑝 1

the𝑥1-coordinate.Then,thecoefficients𝑎12,𝑎21and𝑏2andtheveloc- itycomponent𝑢𝑎,2arezero.Aroughness/textureelementissymmetric ifmirroringtheelementatthecoordinateaxesresultsinthesameel- ement.With𝑎12=𝑎21=𝑏2=𝑢𝑎,2=0,thehomogenizedReynoldsequa- tion(19)reducesto

𝒙⋅ ((

𝑎11(𝒙) 0 0 𝑎22(𝒙)

)

𝒙𝑝0(𝒙) )

= −∇𝒙⋅ (𝑏1(𝒙)

0 )

+6𝜇 (𝑢𝑎,1

0 )

⋅∇𝒙0(𝒙). (21) AcomparisonofthemodifiedReynoldsequation(12)of theaverage flowmodelwith(21)revealsthatfor̄𝑝(𝒙)=𝑝0(𝒙)thefollowingrelations musthold:

𝜙𝑝,1(𝒙)nom(𝒙)3=𝑎11(𝒙), 𝜙𝑝,2(𝒙)nom(𝒙)3=𝑎22(𝒙), 6𝜎𝜇𝑢𝑎,1𝜙𝑠(𝒙)=−𝑏1(𝒙),

̄ℎ(𝒙)=nom(𝒙)=0(𝒙)+const. (22) 3. AdvantagesofthehomogenizedReynoldsequation

Regarding the applicability and accuracy, the homogenization method has several considerable advantages over the average flow modelbyPatirandCheng.Thesearesummarizedinthefollowinglist.

Homogenizationisamathematicalconcept, whichis validatedby two-scaleconvergence(Allaire,1992; PavliotisandStuart,2008), whereastheaverageflowmodelwasempiricallyderivedbyinvesti- gatingpressureandshearflowforreferenceelementswithsymmet- ricroughness.Hence,incontrasttothemodifiedReynoldsequation (12)oftheaverageflowmodel,thehomogenizedReynoldsequa- tion(19) alwaysprovides anaccurateaveragingof theoriginal Reynoldsequation(1).

Theboundaryconditionsforthecellproblemson themicroscale (periodic)arealwayscorrect(Almqvistetal.,2011b),whiletheyare notforthereferenceproblemsoftheaverageflowmodel(Dirichlet andNeumann).Inparticular,theperiodicboundaryconditionsallow fornonsymmetricroughness/textures.Then,thecoefficients𝑎12, 𝑎21and𝑏2,see(16),(17)and(18),arenonzero.

Theupscaling(20)providesahigher-ordersolutionandapprox- imatesthesolutionoftheoriginalReynolds equation(1)well.It allowsforresolvinglocalpressurepeakswhichisimportantfor thedetectionofcriticalstatessuchaslubricationbreakdownorfail- ure(ZhuandWang,2011)orforthedeterminationofaccurateval- uesofthefrictionforce.ThelatterisdemonstratedinSect.4.3.The computationofthehigher-ordersolution ̃𝑝onlyinvolvestheeval- uationof knownquantities,namelythegradient oftheaveraged pressuresolution𝑝0 andthecellproblemsolutions𝜒0,𝜒1and𝜒2. Hence, ̃𝑝canbeevaluatedinapost-processingstepwithnegligible computationalcost.

Byapplyingtheconceptofreiteratedhomogenization,seeforin- stance(AllaireandBriane,1996),furtherscalescanbetakeninto account.Hence,therangeofapplicationsregardingthestudyoflu- bricationproblemswithanaveragingmethodisextendedfromsys- temswithroughortexturedsurfacestoacombinationofboth.This isdonebydividingtheproblemathandintomacroscale,mesoscale

(6)

(textures)andmicroscale(roughness)problems.Aderivationforthe Reynoldsequation(1)withoutandwithacavitationmodelcanbe foundinAlmqvistetal.(2008)andMartin(2008),respectively.In additiontothefirstlocalscaleforthetextureswith𝒚=𝒙𝜀,asec- ondlocalscaleisusedfortheroughnesswith𝒛=𝒙𝜀2.Notethat forsimplicitytheserelationsaregivenfor(𝑖,𝑗)=(0,0)here,cf.(2). Applyingtheasymptoticexpansionthenleadstocellproblemsfor eachofthetwolocalscales.

Insummary,reiteratedhomogenizationallowsfortheefficientin- vestigationofapplications withtextured surfaceswhile alsocon- sideringthe naturalsurfaceroughness.Up tonow,usually semi- deterministicsimulationsareconductedforthispurpose,seeforin- stance(Maetal., 2017;Profitoetal.,2017),forwhichoftenthe averageflowmodelisusedforthesurfaceroughness,whilethetex- turesarediscretizedbyhighly-resolvedcomputationalmeshes.The computationtimeforsuchsimulationscanbesignificantlyreduced byusingtheconceptofreiteratedhomogenizationinstead.

AsdescribedinTomaniketal.(2020),anotherproblemoccurswhen applyingtheaverageflowmodeltosurfaceswithdeepdimplessuch assuperficialporesinathermalspraycoatingforplainbearings.The deepdimplesresultinadistortionofthenominalfilmthickness.The nominalfilmthickness,whichistheprimaryreferenceplaneforall calculationswiththeaverageflowmodel,affectsthehydrodynamic filmthickness,thehydrodynamicpressure,theshearrateandthehy- drodynamicfriction.Furthermore,thedistortionofthenominalfilm thicknessleadstoawrongestimationoftheasperitycontactpressure inthecaseofmixed-frictionsimulations,forinstanceinjournalbear- ingswithlowslidingspeed(Grützmacheretal.,2018;Königetal., 2019;2020a)orinstart-stopoperation(Königetal.,2020b).Similar observationsweremadeformicro-coinedandmulti-scaletextured surfaces,viz.acombinationofmicro-coininganddirectlaserinter- ferencepatterning(Königetal., 2020a).Incontrast,byapplying theconceptofhomogenization,neitherforthecellproblemnorfor thehomogenizedproblemthenominalfilmthicknessnomorthe standarddeviationof theroughness𝜎 arerequired.Toconclude, thehomogenizationtechniqueavoidsthemodelinguncertainties andthelimitedinterpretabilityoftexturedsurfaceswiththeaverage flowmodel.

Despitetheseadvantagesofthehomogenizationmethodandthepo- tentiallackofaccuracyoftheaverageflowmodel,thelatteriswidely usedintribologicalstudies,mostlyinitsoriginalformulationbyPatir andCheng(1978,1979),eventhoughnumerousmodificationsforits improvementwereproposed(Fatuetal.,2012;Gropperetal.,2016).

4. Applicationofthemethodsandcomparisonofresults

Forthecomparisonofthetwoaveragingmethods,wenumerically studytwoapplications,namelyajournalbearingandaplane-inclined sliderbearing.ThesesetupsaresummarizedinSect.4.1.Afterabrief meshconvergencestudyinSect.4.2,thenumericalresultsarepresented inSect.4.3.Sincebothaveragingmethodsarederivedfromtheoriginal Reynoldsequation(1),thesolutionof(1)isusedasreferencesolution formeasuringtheaccuracyoftheaverageflowmodelandthehomoge- nizationmethod.

4.1. Numericalsetups

ThesimulationparametersarelistedinTab.2.Testcase1isbasedon thejournalbearingapplicationdescribedinGrützmacheretal.(2018), whereastestcase2forasliderbearingissetuptoverifythegeneral- ityofthefindings.Forbothcases,weinvestigatedifferenttextures:(a) symmetricduetosquare/rectangularsurfacearea(isotropic/orthotropic case)and(b)nonsymmetricduetoadiagonallayout(anisotropiccase).

Theparticulartexture elementsareillustratedin Fig.3.Thetexture height𝑡isconstantforbothtestcasessuchthateither1(𝒚)=𝑡or

Table2

Parameters for the simulations of the journal bearing andtheplane-inclinedsliderbearing.TheReynoldsnum- ber𝑅𝑒iscomputedwith=𝑐fortestcase1and= 0.5(0,min+0,max)fortestcase2.

test case 1 2

bearing type journal slider

bearing parameters

diameter 𝑑[mm] 80∕ 𝜋 -

(unfolded) length 𝑙[mm] 80 6

width 𝑤 [mm] 20 2

number of textures 𝑛 𝑡,1[-] 160 60 number of textures 𝑛 𝑡,2[-] 40 50 total number of textures 𝑛 𝑡[-] 6 , 400 3 , 000 lubricant parameters

density 𝜌[kg/m 3] 820 820

dynamic viscosity 𝜇[Pa s] 0.014 0.014 operating conditions

radial clearance 𝑐[ μm] 17.5 - minimum gap height ℎ 0,min[ μm] 3.5 5 maximum gap height ℎ 0,max[ μm] 31.5 10 relative eccentricity 𝑒 rel[-] 0.8 - convergence ratio 𝑘 [-] - 1 surface velocity 𝑢 𝑎,1[m/s] 0.2 4 Reynolds number 𝑅𝑒 = 𝜌𝑢 𝑎,1𝜇[-] 0.205 1.757

1(𝒚)=0inthemicroscaledomain.Theratio𝐴𝑡𝐴𝑡𝑒ofthetexture areatotheareaofthewholetextureelementissimilarforallfourtex- tures.AsdiscussedinSects.2and3,wecanonlyexpectaccurateresults oftheaverageflowmodelincaseofthesymmetrictextures,whereasthe homogenizationmethodshouldprovideaccurateresultsforalltextures.

Thegapheightsforthejournalbearing(testcase1)andtheslider bearing(testcase2)areprescribedby

0(𝒙)=𝑐 (

1+𝑒relcos (2𝑥1

𝑑 ))

(23)

and

0(𝒙)=0,min0,max

𝑙 𝑥1+0,max, (24)

respectively.

All numericalsimulations areperformedon quadrilateral meshes with finiteelement solvers we implemented usingthe C++ library deal.II(Bangerthetal.,2007).Detailsontheweakformulationsforthe cellproblems(13)-(15)andthehomogenizedproblem(19)aswellas informationontheirimplementationcanbefoundinRomandMüller (2018).

4.2. Meshconvergence

Forareliablecomparisonof thenumericalresults,weperformed a thorough meshconvergence study for both test cases for each of thetexturesetups.Thiscomprisesthemeshesforsolvingtheoriginal Reynoldsequation(1),whichservesasreferencesolution,themeshes forthereference/cellproblemsonthemicroscaleandthemeshesforthe macroscaleproblems.Exemplarily,wepresenttheresultsofthestudy fortestcase1withsquaretexturesinthissection.

Figure 4 (a) shows thesolution 𝑝 of the Reynolds equation (1), Fig.4(b)thesolution𝜒0ofthecellproblem(13)for0=31.5μmand Fig.4(c)theupscaledsolutioñ𝑝,see(20),ofthehomogenizedproblem (19),eachintheareaofthemaximumoftheparticularsolutionandfor anincreasingnumberofmeshcells.IncaseoftheReynoldsequation(a) andthecellproblem(b),themaximumincreaseswithincreasingmesh resolution, butthedistancesbetween thecurves decrease.Themesh resolutionof10,240× 2,560fortheReynoldsproblemcorrespondstoa resolutionof64× 64pertextureelement.Foragoodcomparabilitywith thesolutionofthehomogenizedproblem,wechoose64× 64alsoasfinal resolutionforthecellproblems.Thisisdoneanalogouslyforthethree

(7)

Fig.3. Textureelementsforthetwotestcases.

Fig.4. Meshconvergencefortestcase1with squaretextures.

othersetups,i.e.,fortestcase1diagonalandtestcase2rectangularand diagonal.AsFig.4(c)shows,thehomogenizedproblemislessmesh- dependentthantheReynoldsandthecellproblem,andaresolutionof 1,280× 320isclearlysufficient.RelatedtotheReynoldsproblem,thisis areductioninthenumberofmeshcellsof98.4%.Themeshconvergence behavioroftheaverageflowmodelisanalogoustothatofthehomog- enizationmethod.Notethatthediagonaltexturesrequireafinermesh toadequatelyresolvetheedgesofthetextures.Togiveanexample,the numberofmeshcellsweusefortheReynoldsproblemfortestcase1 withdiagonaltexturesis16,000× 4,000,whichcorrespondsto100× 100 pertextureelement.

4.3. Numericalresults

Inthefollowing,wecomparethesolutionsoftheaverageflowmodel andthehomogenizationmethodwiththesolutionoftheReynoldsequa- tion.Foraquantitativeinvestigationoftheaccuracyoftheaveraging methods,weconsiderthemaximumpressure𝑝maxandtheperformance parametersload-carryingcapacity𝑊 andfrictionforce𝐹. Thelatter twoaredefinedby

𝑊 ∶=

Ω

[𝑝]+𝑑𝒙, (25)

whereonlypositivevalues[𝑝]+=max(0,𝑝)ofthepressurearetakeninto account,and

𝐹∶=

Ω𝜏 𝑑𝒙=

Ω

( 2

𝜕𝑝

𝜕𝑥1

+𝜇𝑢𝑎,1

)

𝑑𝒙, (26)

respectively,seeforinstance(WenandHuang,2018).Here,𝑝represents thesolutionoftheReynoldsequation.Itcanbereplacedbȳ𝑝(average flowmodel),𝑝0(homogenization)or̃𝑝(homogenizationwithupscaling).

4.3.1. Testcase1(journalbearing)

Figure5(a)showsapartofthedifferentpressuresolutionsforthe journalbearing oftestcase 1withsymmetric (square)textures.The curveisextractedattheposition𝑥2=10.25mm.Asexpected,theav- eragedpressures ̄𝑝and𝑝0 oftheaverageflowmodelandthehomog- enizationmethod,respectively,areidenticalandprovideanaccurate averagingoftheReynoldssolution𝑝.Theupscaledpressurẽ𝑝oftheho- mogenizationmethodagreeswellwiththeReynoldssolution𝑝.Inthe caseofthenonsymmetric(diagonal)texturesinFig.5(b),thesolution̄𝑝 oftheaverageflowmodelshowsalargedeviationfromtheReynolds solution𝑝,whilethehomogenizationmethodagainleadstoaccurate results.

These findings aresubstantiated by aquantitative comparison of 𝑝max, 𝑊 and 𝐹. The values are listedin Tab. 3. Inthe symmetric case (a), 𝑝max and𝑊 arealmost thesamefor all foursolutions. In contrast,𝑝max=0.4139MPaand𝑊 =73.27Ncomputedbytheaverage flowmodelforthenonsymmetrictextures(b)deviatefromthecorre- spondingReynoldsvaluesby9.8%and7.1%,respectively.Thefriction forces𝐹 whicharecomputedfromtheaveragedsolutions ̄𝑝and𝑝0are alwaysinaccuratefortexturedapplications,regardlessofthesymme- try.Thisisduetonotresolvingthepressurepeakswhichleadstoan errorinthepressuregradiententeringthecomputationof𝐹,see(26). ThedeviationsfromtheReynoldsvaluesliebetween8.4%(symmetric case,𝐹=0.4068N)and10.9%(nonsymmetriccase,averageflowmodel,

(8)

Fig.5.Pressuresolutionsat𝑥2=10.25mmfor testcase1.

Table3

Maximumpressure𝑝max[MPa],load-carryingcapacity𝑊[N]andfrictionforce𝐹[N]computedfrom thedifferentpressuresolutionsfortestcase1.

(a) symmetric textures (square) (b) nonsymmetric textures (diagonal) Reynolds avg. flow homog. upscaled Reynolds avg. flow homog. upscaled 𝑝 max 0.5990 0.5987 0.5987 0.5991 0.4591 0.4139 0.4586 0.4604 𝑊 102.51 102.45 102.45 102.48 78.85 73.27 79.20 79.23 𝐹 0.4443 0.4068 0.4068 0.4443 0.4231 0.3771 0.3804 0.4222

Fig.6. Pressuresolutionsfortestcase1withnonsymmetric(diagonal)textures.

𝐹=0.3771N).Notethateveninthenonsymmetriccase,thevalueofthe averageflowmodel(𝐹=0.3771N)isalmostequaltothevalueofthe homogenizationmethod(𝐹 =0.3804N).Thisindicatesthattheslopesof theaveragedpressuresolutionsin𝑥1-directionand,hence,thepressure gradientsaresimilar.

Figure6visualizesthereasonfortheinaccurateaveragingoftheav- erageflowmodelincaseofthenonsymmetrictextures.Thesetextures leadtoanasymmetryofthepressuresolution,seeFig.6(a).Thisbe- haviorcannotbecapturedbytheaverageflowmodel,seeFig.6(b).In

contrast,thehomogenizationmethodisabletoreproducetheasymme- try,seeFig.6(c).Thisisduetotheadditionalcoefficients𝑎12,𝑎21and 𝑏2,cf.(12)and(19),whicharenonzerointhiscase.

4.3.2. Testcase2(plane-inclinedsliderbearing)

ThepressureplotsinFig.7fortestcase2,extractedat𝑥2=1.02mm, revealthattheinfluenceofthetexturesonthepressureislargerthanin testcase1.Apartfromthat,thebehaviorissimilar.Thehomogenization methodprovidesanaccurateaveragingandaverygoodupscalingfor boththesymmetric(rectangular)andthenonsymmetric(diagonal)tex- tures.Incontrast,theapplicationoftheaverageflowmodelleadstoa strongdeviationfromtheReynoldssolutioninthenonsymmetriccase.

Duetothelargerinfluenceofthetextures,themaximumpressures 𝑝max resultingfromtheaveragingtechniquesaretoolow,cf.Tab.4, by4.3%(symmetriccase),4.6%(nonsymmetriccase,homogenization withoutupscaling)andeven17.2%(nonsymmetriccase,averageflow model).Similarly,thereisonlyonelargedeviationregardingtheload- carryingcapacity 𝑊: inthenonsymmetric case,thevaluecomputed fromtheaverageflowmodel(𝑊 =1.4208N)deviatesfromtheReynolds value(𝑊 =1.6178N)by12.2%.Thedeviationsofthefrictionforces𝐹 resultingfromtheaveragingtechniquesfromtheReynoldsvaluesare slightlylargerthanfortestcase1andliebetween10.8%(nonsymmet- riccase,homogenizationwithoutupscaling,𝐹=0.07615N)and12.9% (symmetriccase,𝐹=0.07489N).

Asfortestcase1inFig.6,apressureasymmetryisalsovisibleinthe contourplotsfortestcase2inFig.8.Whilethereishardlyanydifference visiblebetweentheReynoldssolution(a)andtheupscaledsolution(c) ofthehomogenizationmethod,theasymmetryagaincannotbecaptured bytheaverageflowmodel(b).

Finally, Fig. 9 demonstrates how the upscaled pressure solution (20)is composedoftheaveragedmacroscalesolution𝑝0 andthemi- croscale solution 𝜀𝑝1. The pressure distribution in Fig. 8 (c) is the sumofthepressuresdepictedinFig.9(a)and(b),i.e., ̃𝑝=𝑝0+𝜀𝑝1. The texture-inducedasymmetry is visiblein both theaveragedsolu- tion 𝑝0 and themicroscale solution 𝜀𝑝1. The absolute values of 𝜀𝑝1

reachuptoapproximately14%oftheaveragedpressurevalues𝑝0with (𝜀𝑝1)max=0.047MPaand𝑝0,max=0.332MPa.

(9)

Fig.7.Pressuresolutionsat𝑥2=1.02mmfor testcase2.

Table4

Maximumpressure𝑝max[MPa],load-carryingcapacity𝑊[N]andfrictionforce𝐹[N]computedfromthe differentpressuresolutionsfortestcase2.

(a) symmetric textures (rectangular) (b) nonsymmetric textures (diagonal) Reynolds avg. flow homog. upscaled Reynolds avg. flow homog. upscaled 𝑝 max 0.3550 0.3399 0.3399 0.3551 0.3483 0.2885 0.3322 0.3501 𝑊 1.6378 1.6359 1.6359 1.6383 1.6178 1.4208 1.6300 1.6317 𝐹 0.08594 0.07489 0.07489 0.08595 0.08534 0.07606 0.07615 0.08531

Fig.8. Pressuresolutionsfortestcase2withnonsymmetric(diagonal)textures.

5. Conclusion

This work demonstrates the advantages of the homogenization methodovertheaverageflowmodelbyPatirandCheng.Thequali-

Fig.9. Componentsoftheupscaledpressurẽ𝑝from(20)fortestcase2with nonsymmetric(diagonal)textures.

tativeandquantitativecomparisonsofnumericalresultsforajournal bearingandaplane-inclinedsliderbearingclearlyshowthathomoge- nizationshouldbethemethodofchoice.

Theaverageflowmodelandthehomogenizationmethodarevery similarintermsoftheirgoverningequationsandcomputationalproce- dure.Theymainlydifferintheboundaryconditionsofthemicroscale

Referenzen

ÄHNLICHE DOKUMENTE

[ 7 ] The strength of this homogenization approach is in its ability (1) to detect an unknown number of nondocumented breakpoints in a candidate station by using a dynamic

in this study of Conrad’s representation of Otherness conducted through an examination of three of his works – The Nigger of the Narcissus, Lord Jim, and Under Western

The purpose of this bachelor´s thesis is to investigate the German movement Sturm und Drang (Storm and Stress) and its significance in the literary, historical

THE AVERAGING ~lliTHOD APPLIED TO THE INVESTIGATION OF SUBSTANTIAL TIME VARYING SYSTEMS OF A HIGHER

[r]

At its heart is the Global Change Experimental Facility (GCEF) – an experimental facility where researchers can investigate the effects of climate change on land use

In summary, six kinds of new nonclassical potential symmetry generators of the Burgers equation are deter- mined in this paper, and three classes of new explicit solutions are

In this work, the path integration method and the energy-based stochastic averaging method are introduced in order to study the stochastic responses of ship roll motion in random