外文翻译--端铣削自适应切削力的模糊控制策略 英文版.pdf
JournalofMaterialsProcessingTechnologyxxx(2005)xxxxxxAbstractoperations.metal-cuttingforfeed-rateconducted©K1.theofaremeragearederEvanthemachiningnecessarysatisfytioperatingAtainconditions.de0924-0136/$doi:10.1016/j.jmatprotec.2005.02.143Fuzzycontrolstrategyforanadaptiveforcecontrolinend-millingU.Zuperl,F.Cus,M.MilfelnerFacultyofMechanicalEngineering,UniversityofMaribor,Smetanova17,2000Maribor,SloveniaThispaperdiscussestheapplicationoffuzzyadaptivecontrolstrategytotheproblemofcuttingforcecontrolinhighspeedend-millingTheresearchisconcernedwithintegratingadaptivecontrolwithastandardcomputernumericalcontroller(CNC)foroptimisingaprocess.Itisdesignedtoadaptivelymaximisethefeed-ratesubjecttoallowablecuttingforceonthetool,whichisverybeneficialatimeconsumingcomplexshapemachining.Thepurposeistopresentareliable,robustneuralcontrolleraimedatadaptivelyadjustingtopreventexcessivetoolwear,toolbreakageandmaintainahighchipremovalrate.Numeroussimulationsandexperimentsaretoconfirmtheefficiencyofthisarchitecture.2005ElsevierB.V.Allrightsreserved.eywords:End-milling;Adaptiveforcecontrol;FuzzyIntroductionAremainingdrawbackofmodernCNCsystemsisthatmachiningparameters,suchasfeed-rate,speedanddepthcut,areprogrammedoff-line.Themachiningparameterssimulationswiththefuzzycontrolstrategyarecarriedout.Theresultsdemonstratetheabilityoftheproposedsystemtoeffectivelyregulatepeakforcesforcuttingconditionscom-monlyencounteredinend-millingoperations.Forcecontrolalgorithmshavebeendevelopedandeval-usuallyselectedbeforemachiningaccordingtoprogram-sexperienceandmachininghandbooks.Topreventdam-andtoavoidmachiningfailuretheoperatingconditionsusuallysetextremelyconservative.Asaresult,manyCNCsystemsareinefficientandrunun-theoperatingconditionsthatarefarfromoptimalcriteria.enifthemachiningparametersareoptimisedoff-linebyoptimisationalgorithm5theycannotbeadjustedduringmachiningprocess.Toensurethequalityofmachiningproducts,toreducethecostsandincreasethemachiningefficiency,itistoadjustthemachiningparametersinreal-time,totheoptimalmachiningcriteria.Forthisreason,adap-vecontrol(AC),whichprovideson-lineadjustmentoftheconditions,isbeingstudiedwithinterest3.InourCsystem,thefeed-rateisadjustedon-lineinordertomain-aconstantcuttingforceinspiteofvariationsincuttingInthispaper,asimplefuzzycontrolstrategyisvelopedintheintelligentsystemandsomeexperimentalCorrespondingauthor.Tel.:+38622207623;fax:+38622207990.E-mailaddress:uros.zuperluni-mb.si(U.Zuperl).uatedisnallyantrollerditions.trollerandatedallthetems,bysentedtesystems3controlhasmotion.forseefrontmatter©2005ElsevierB.V.Allrightsreserved.bynumerousresearchers.Amongthemostcommonthefixedgainproportionalintegral(PI)controllerorigi-proposedformillingby4.Kimetal.4proposedadjustablegainPIcontrollerwherethegainofthecon-isadjustedinresponsetovariationsincuttingcon-Thepurelyadaptivemodelreferenceadaptivecon-(MRAC)approachwasoriginallyinvestigatedbyCusBalic2.Thesecontrollersweresimulatedandevalu-andphysicallyimplementedby1.Bothstudiesfoundthree-parameteradaptivecontrollertoperformbetterthanfixedgainPIcontroller.Asregardsfuzzycontrolsys-anintroductorysurveyofpioneeringactivitiesisgivenHuangandLin3,andamoresystematicviewispre-byin4.Comparisonsoffuzzywithproportionalin-gralderivative(PID)controlandstabilityanalysisoffuzzyandsupervisoryfuzzycontrolareaddressedinRef.Muchworkhasbeendoneontheadaptivecuttingforceformilling2.However,mostofthepreviousworksimplifiedtheproblemofmillingintoone-dimensionalInthiscontribution,wewillconsiderforcecontrolthree-dimensionalmilling.2Processingscribesthesimulation/eposedimentalresearch.2.fuzzyseteThewhichplementcontrolmoreTherateasthecomparedcontrolFuzzyratecuttingcreasesrates,productionarebreakage.callytheforbelo1.2.3.4.signingonactualcentagemisationalcorrectcontrolleraplepro2.1.aaboutinputoperatorthroughU.Zuperletal./JournalofMaterialsThepaperisorganisedasfollows.Section2brieflyde-theoverallforcecontrolstrategy.Section3coversCNCmachiningprocessmodel.Section5describesthexperimentsandimplementationmethodofpro-controlscheme.Finally,Sections6and7presentexper-results,conclusions,andrecommendationsforfutureAdaptivefuzzycontrollerstructureAnewon-linecontrolschemewhichiscalledadaptivecontrol(AFC)(Fig.1)isdevelopedbyusingthefuzzytheory.Thebasicideaofthisapproachistoincorporatethexperienceofahumanoperatorindesignofthecontroller.controlstrategiesareformulatedasanumberofrulesaresimpletocarryoutmanuallybutdifficulttoim-byusingconventionalalgorithm.Basedonthisnewstrategy,verycomplicatedprocesscanbecontrolledeasilyandaccuratelycomparedtostandardapproaches.objectiveoffuzzycontroliskeepingthemetalremoval(MRR)ashighaspossibleandmaintainingcuttingforcecloseaspossibletoagivenreferencevalue.Furthermore,amountofcomputationtaskandtimecanbereducedastoclassicalormoderncontroltheory.Schematicrulesareconstructedbyusingrealexperimentaldata.adaptivecontrolensurescontinuousoptimisingfeedcontrolthatisautomaticallyadjustedtoeachparticularsituation.Whenspindleloadsarelow,thesystemin-cuttingfeedsaboveandbeyondpre-programmedfeedresultinginconsiderablereductionsincycletimesandcosts.Whenspindleloadsarehighthefeedrateslowered,safeguardingmachinetoolsfromdamagefromWhensystemdetectsextremeforces,itautomati-stopsthemachinetoprotectthecuttingtool.Itreducesneedforconstantoperatorsupervision.Sequenceofstepson-lineoptimisationofthemillingprocessarepresentedw.namicstheasvofcuttinglated,Delta1forceFig.1.ComparisonofactualTechnologyxxx(2005)xxxxxxThepre-programmedfeedratesaresenttoCNCcontrollerofthemillingmachine.Themeasuredcuttingforcesaresenttothefuzzycon-troller.Fuzzycontrollerusestheenteredrulestofind(adjust)theoptimalfeed-ratesandsendsitbacktothemachine.Steps1and3arerepeateduntilterminationofmachining.Theadaptiveforcecontrolleradjuststhefeed-ratebyas-afeed-rateoverridepercentagetotheCNCcontrollerafour-axisHeller,basedonameasuredpeakforce.Thefeed-rateistheproductofthefeed-rateoverrideper-andtheprogrammedfeed-rate.Ifthefeed-rateopti-modelswereperfect,theoptimisedfeed-ratewouldwaysbeequaltothereferencepeakforce.Inthiscasetheoverridepercentagewouldbe100%.Inorderforthetoregulatepeakforce,forceinformationmustbevailabletothecontrolalgorithmateverycontrollersam-time.Adataacquisitionsoftware(Labview)isusedtovidethisinformation.StructureofafuzzycontrollerInfuzzyprocesscontrol,expertiseisencapsulatedintosystemintermsoflinguisticdescriptionsofknowledgehumanoperatingcriteria,andknowledgeaboutthe±outputrelationships.Thealgorithmisbasedonthesknowledge,butitalsoincludescontroltheory,theerrorderivative,takingintoconsiderationthedy-oftheprocess.Thus,thecontrollerhasasitsinputs,cuttingforceerrorDelta1FanditsfirstdifferenceDelta12F,andoutputs,thevariationinfeedrateDelta1f.Thefuzzycontrolariablesfuzzification(seeFig.2)aswellasthecreationtherulesbaseweretakenfromtheexpertoperator.Theforceerrorandfirstdifferenceoftheerrorarecalcu-ateachsamplinginstantk,as:Delta1F(k)=FrefF(k)and2F(k)=Delta1F(k)Delta1F(k1),whereFismeasuredcuttingandFrefisforcesetpoint.andmodelfeed-rate.3.etalandforcesscribedmachinefeedingfitquencefromformcommandedtingmodel.mentalfeed-rateU.Zuperletal./JournalofMaterialsProcessingFig.2.StructureofafuzzyCNCmachiningprocessmodelACNCmachiningprocessmodelsimulatorisusedtovaluatethecontrollerdesignbeforeconductingexperimen-tests.Theprocessmodelconsistsofaneuralforcemodelfeeddrivemodel.Theneuralmodelestimatescuttingbasedoncuttingconditionsandcutgeometryasde-byZuperl1.Thefeeddrivemodelsimulatestheresponsetochangesincommandedfeed-rate.Thedrivemodelwasdeterminedexperimentallybyexamin-stepchangesinthecommandedvelocity.Thebestmodelwasfoundtobeasecond-ordersystemwithanaturalfre-yof3Hzandasettlingtimeof0.4s.Comparisonofxperimentalandsimulationresultsofavelocitystepchange7to22mm/sisshownonFig.3.ThefeeddriveandneuralforcemodelarecombinedtotheCNCmachiningprocessmodel.Modelinputisthefeed-rateandtheoutputistheX,Yresultantcut-force.ThecutgeometryisdefinedintheneuralforceThesimulatorisverifiedbycomparisonofexperi-andmodelsimulationresults.Avarietyofcutswithchangesweremadeforvalidation.changeFig.resultsTechnologyxxx(2005)xxxxxx3controller.Theexperimentalandsimulationresultantforceforastepinfeed-ratefrom0.05to2mm/toothispresentedin4.Theexperimentalresultscorrelatewellwithmodelintermsofaverageandpeakforce.TheexperimentalFig.3.Comparisonofactualandmodelfederate.4resultsandthe3.1.dardlarimentsforceusedfederatedialforcesaryU.Zuperletal./JournalofMaterialsProcessingFig.4.Structureofafuzzycorrelatewellwithmodelresultsintermsofaveragepeakforce.Theobviousdiscrepancymaybeduetoinaccuraciesinneuralmodel,andunmodeledsystemdynamics.CuttingforcemodelingTorealisetheon-linemodellingofcuttingforces,astan-BPneuralnetwork(NN)isproposedbasedonthepopu-backpropagationleeringrule.Duringpreliminaryexper-itprovedtobesufficientlycapableofextractingthemodeldirectlyfromexperimentalmachiningdata.Itistosimulatethecuttingprocess.TheNNformodellingneedsfourinputneuronsformilling(f),cuttingspeed(vc)axialdepthofcut(AD)andra-depthofcut(RD).TheoutputfromtheNNarecuttingcomponents,thereforetwooutputneuronsareneces-.ThedetailedtopologyoftheusedNNwithoptimaltrain-ingalso73.2.modelingferentberefnetwysed.difandtheinputconclusionsTechnologyxxx(2005)xxxxxxcontroller.parametersandmathematicalprincipleoftheneuronisshowninFig.5.BestNNconfigurationcontains5,3andhiddenneuronsinhiddenlayers.TopologyofneuralnetworkanditsadaptationtoproblemTheeffectoftopologyisalsostudiedbyconsideringdif-cases.Thetopologiesarevariedbyvaryingthenum-ofneuronsinhiddenlayers.Toevaluatetheindividualfectsoftrainingparametersontheperformanceofneuralork40differentnetworksweretrained,testedandanal-Thenetworkperformanceswereevaluatedusingfourferentcriteria5:ETstMax,ETst,ETrn,andETrnMaxthenumberoftrainingcycles.Thenumberofneuronsininputandoutputlayersaredeterminedbythenumberofandoutputparameters.Fromtheresultsthefollowingcanbedrawn.