外文翻译--集成设计与制造中的的智能综合系统 英文版.pdf
JournalofMaterialsProcessingTechnology76(1998)2328IntelligenthybridsystemforintegrationindesignandmanufactureDaizhongSu*,MarkWakelamDepartmentofMechanicalandManufacturingEngineering,TheNottinghamTrentUni6ersity,BurtonStreet,Nottingham,NG14BU,UKReceived23May1997;receivedinrevisedform11June1997AbstractAnintelligenthybridsystem(IHS)approachhasbeendevelopedtointegratevariousstagesintheprocessofdesignandmanufacture,includingproductdesignspecification,conceptualdesign,detaildesign,processplanning,costingandCNCmanufacture.TheIHSapproachblendsarule-basedsystem(RBS),artificialneuralnetworks(ANNs),geneticalgorithm(GA),hypermedia(HM)andCAD:CAE:CAMpackagesintoasingleenvironment.ThedesignandmanufactureexpertiseiscapturedbytheRBSandANNs,whiletheothertasks,suchasnumericalanalysis,designoptimisation,engineeringdrawinganddataprocessing,areperformedusingGA,HMandCAD:CAE:CAM.TheRBSworksasaco-ordinator,controllingthewholeprocess.Asanapplicationoftheapproach,anIHSformechanicalpowertransmissionsystemshasbeendeveloped.©1998ElsevierScienceS.A.Allrightsreserved.Keywords:Artificialintelligence;Hybridsystem;CAD:CAE:CAM;Mechanicaltransmission;Design;Manufacture1.IntroductionDuetovariousstagesandcomplexactivitiesin-volved,theprocessofproductdesignandmanufactureistediousandtime-consuming.Inordertoreduceproductioncostsandtime-to-market,itishighlydesir-abletointegratethewholeprocessusingadvancedcomputingtechniques.Inthisresearch,anintelligenthybridsystem(IHS)approachhasbeendevelopedtointegratevariousactivitiesinvolvedindesignandman-ufacture.Thisisachievedbyblendingarule-basedsystem(RBS),artificialneuralnetworks(ANNs),ge-neticalgorithms(GA),hypermedia(HM)andCAD:CAE:CAMpackagesintoasingleenvironment.Bothintegrationindesignandmanufactureandap-plicationofartificialintelligence(AI)inengineeringhaveattractedresearchersattentions,forexample13,however,integrationofmultipleAItechniques,asmentionedabove,intothewholeprocessofdesignandmanufacturehasnotbeenfoundintheliterature.Hence,theIHSisanovelresearch.EachindividualtypeofAIsystemhasadvantagesanddisadvantages.TheIHScombinesthepowerofeachtechnique,avoidstheirweakaspects,andthus,providesamorepowerfultoolforAIbasedintegration.Nowadays,manyCAD:CAM:CAEpackagesareap-pliedinindustries,however,inmostcases,theyarestand-alonetypes.Asoneoftheimportantfeaturesofthisapproach,itprovidesameanstointegratesuchexistingpackagesintoasingleenvironment.Thisisobviouslycosteffectiveforindustrialusers,particularlyformediumandsmallsizecompanies.Inthefollowingsections,theapproachisbrieflydescribedfirst,togiveanoverviewofthemainstruc-ture,relationshipsofthesoftwaretechniquesinvolved,designandmanufacturingactivitiesconsidered,andtheintegratedprocess.Then,theimportantfeaturesoftheIHSarepresentedinmoredetail,includingtheRBS,ANNs,GA,HM,andCAD:CAE:CAMintegration.Theapproachhasbeenappliedintothedesignandmanufactureofmechanicalpowertransmissionsys-tems.Thekeyfeaturesarepresentedinconjunctionwiththeapplication.*Correspondingauthor.Tel.:441159418418;fax:441159486506;e-mail:mec3sudntu.ac.uk0924-0136:98:$19.00©1998ElsevierScienceS.A.Allrightsreserved.PIIS0924-0136(97)00310-5D.Su,M.Wakelam:JournalofMaterialsProcessingTechnology76(1998)232824Fig.1.Softwareintegration.2.OverviewoftheapproachThesoftwaretechniquesandpackages:programmesinvolvedintheIHSandtheirrelationshipareshowninFig.1.Integrationofthevariouselementsintoasingleenvironmentisachievedviathecontrol,whichisaRBS.ThedesignandmanufactureexpertiseiscapturedbytheRBSandANNs,whiletheothertaskswithinthedesignandmanufacturingprocess,suchasnumericalanalysis,engineeringdrawinganddataprocessing,areperformedusingrelevantCAD:CAE:CAMsoftwarepackages.GAandHMareusedtoconductdesignoptimisationandtoprovideeffectivemeansforuserinterfacesanddatatransfer.TheRBScommunicateswiththeothersandworksasaco-ordinator,con-trollingthewholeprocess.UsingtheIHS,thefollowingactivitiesareintegratedthroughoutthewholeprocessofdesignandmanufac-ture:conceptualdesign:formulationofproductdesignspe-cification,conceptgenerationandevaluationdetaildesign:analysis,sub-assembly:componentde-signandassembly,engineeringdrawings,designre-trievalmanufacture:processplanning,costingandCNCmanufacture.TheintegratedprocessisshowninFig.2,whichindi-cateshowthewholeprocessiscontrolledbytheRBSandhowrelevantsoftwaretechniques:packagesarein-volvedateachstage.Asanexampletoillustratetheapproach,anIHSforthedesignandmanufactureofmechanicalpowertransmissionsystemshasbeendevel-opedtoperformthefollowingtasks:(a)Formulationofproductdesignspecification(PDS).12PDSitems,suchasorientationofinput:outputshafts,manufacturecost,etc.areconsidered.Itispossi-bletoextendthesystemtoaidmorePDSitems.(b)Conceptualdesign.TheconceptstobeconstructedbytheprototypeIISfallintothefollowingrange:stagesofthetransmission:one,twoorthreeorientationofinput:outputshafts:parallel,crossandperpendicularcomponentsateachstageofthetransmission:seventypesofcomponentsincludinggears,beltsandchains(c)Detaildesign.TheIIScanconductthedetaildesignofgearboxes,beltdrivesoracombinationofboth,includingthefollowing:gearstrengthanalysis,bearingselection,shaftdesign,casedesign,beltandpulleyselection,designoptimisation,componentandassemblydrawingsandparametricdesignofcomponents.Fig.2.Theintegratedprocess.D.Su,M.Wakelam:JournalofMaterialsProcessingTechnology76(1998)232825(d)Manufacture.TheIIScanperformthefollowingtasks:costanalysis,processplanningforthemanufac-tureofmajorcomponentsandCNCprogrammingformanufactureofshafts.3.SystemcontrolAsshowninFig.2,withintheintegratedprocess,thePDSareformulatedfirst,thentheIHScreatesallpossibleconceptstomeetthespecificationsandselectsthebestconceptfordetaildesign.Atthedetaildesignstage,twotypesofsub-systemsareconsidered,beltdrivesandgearboxes,fromwhichapowertransmissionsystemisbuilt;andthedesignofcomponentssuchasgears,shaftsandbearingsarealsocarriedout.Aftercompletionofthedetaildesign,theIHSmovestothemanufacturestagewhereCNC,planningandcostingaretobeconducted.TheIHScanalsocarryoutanyredesigntaskwheneveritisnecessary.Theprocessiscontrolledbyasystemcontroller(SC)andsub-controllers(SubC).TheSCcontrolstheoverallprocessbycommunicationwiththeSubCs,whileaSubCcontrolstheactivitieswithinaparticularstage.TheSCandSubCsareRBScomprisingtwotypesofrules:controlrulesandinformationrules.Thecontrolrulesholdtheknowledgerelatingtothedesignsprogress.Theserulesstructurethedesignpro-cessandcontrolitsdevelopment,formingtheinferenceengineandactivatingtheappropriateinformationrulesdependingonthecircumstancesappliedandthedesignstage.Forexample:facewidthratioisnotdefinedbyuserIFacti6atefacewidthinformationrulesTHENThecontrolrulesalsocontroltheactivationofothermoduleswithinthedesignsystems.Forexample:IFgearisthroughhardenedacti6ateANNmodule1THENFig.4.Fullyconnectednetworks.Theinformationrulescontaininformationintheformofnumericalvaluesanddesignfeaturesoractivateanequationthatencapsulatesthedesigninformation.Forexample:IFthegearisdoublehelicalanditsheattreat-mentisnitridedanditsmountingissymmet-ricTHENmaximumfacewidthratiois1.4heattreatmentiscasecarburisedIFTHENmaximumnumberofteeth119gearratio37.334.Artificialneuralnetworks(ANNs)withintheIHSANNsareappliedattwostageswithintheintegratedprocess:(1)attheconceptualdesignstage,fourANNsaredevelopedforcreationofconcepts,and(2)atthedetaildesignstage,fourANNsareusedtoobtaindesignfactorsintheprocessofinitialgearsizing.AlltheANNsareoffeed-forwardmulti-layerperceptronstrainedwiththeback-propagationtechnique,i.e.com-monlyknownasback-propagationnetworks.Thistypeofnetworkhasbeenselectedfortwomainreasons:fixedstructureenablingastandardstructuretobeestablishedwithinthedesignsystemforeasymodifi-cationandupdateofknowledge.successinpatternrecognitionforsimilarapplications,increasingtheprobabilityofsuccessfortheirappli-cationwithintheproject.4.1.ANNsforconceptualdesignAnexampleoftheconceptsforthemechanicaltrans-missionsystemisshowninFig.3,indicatingthataconceptisformedbyassemblingacomponent(s)intotheposition(s)ofanarrangement.Fournetworksareusedforthetaskofconceptgeneration:threeforcomponentselectionandoneforthearrangement.4.1.1.ComponentnetworksThreenetworksarerequiredtoincorporatethehier-archy4.EachnetworkistrainedwithvariationsoftheFig.3.Conceptconstruction.D.Su,M.Wakelam:JournalofMaterialsProcessingTechnology76(1998)232826Fig.5.Combinationofgenestoformachromosome.samedatawitheithernone,oneortwooftheoutputspermanentlysettozero.Settingtheelementstozeroeffectivelyremovestheelementfromthenetworkwhilemaintainingastandardnetworkstructure.4.1.2.ArrangementnetworkThearrangementnetworkincorporatesthePDS,ori-entationandnumberofstagesinitsdecisionmakingprocess.Thenumberofstagesthatarerequiredisfirstlydefinedbyaseriesofproductionrules.However,adegreeofambiguityisintroducedwhentheparame-tersoftheproductionruleconditionsoverlapgivingmorethanoneresultantsolution.ThestructureofthenetworksissuchthatthedesignspecificationformtheANNsinputwhiletheconceptsrepresenttheoutput.Fig.4illustratesthisrelationship.TrainingdatahasbeenpreparedforthenetworksusinganExcelprogrammewhichperformsthesuitabil-itycalculations,thenscalestheresultsbetween0and1.TheresultsarematchedagainstthePDS,formingthetrainingdatasimilartotheformbelowwhichrepresentfourinputstothenetworkandthreeoutputsfromthenetwork.Fordetailedinformationregardingthetrainingmethod,see4.4.2.ANNsforinitialgearsizingInthedetaildesignofgears,informationabouttherequireddesign,suchaspower,inputspeed,gearratioandcentredistance,isspecifiedfirst.Oncethegearspecificationshavebeenobtained,theIHSgoesthroughtheprocessofgearsizingtodeterminethegeometricparametersincludingmodule,numberofteeth,diameters,helicalangle,etc.ThentheIHSrunsaprogrammeforthegearratingofcontactandbendingstrength.Withinthegearsizingprocess,thevaluesofsafetyfactors,pitchaccuracy,leadaccuracyandloadfactorKHbhavetobedetermined.Thosevaluesareinitiallyprovidedbydesigngraphs,whichareforuseinthemanualdesignprocess,andnormally,theoriginaldatathatthegraphwasconstructedfromareunobtainable,whichmakesitdifficulttoincludethemintoacom-puterintegratedprogramme.However,theapplicationofANNssolvedthisproblem.Inthiscircumstance,fourANNshavebeentrainedusingthedataobtaineddirectlyfromthegraphs,whichprovidesadesirablesolutiontoencodethedesignprocessintothesystem.Forfurtherdetails,see5,6.5.Applicationofgeneticalgorithms(GA)intheIHSTherehavebeenseveraloptimisationsearchtech-niquesavailablesuchashillclimbingandNeutronRaphsonmethods,however,duetothemagnitudeofthesearchareaofacontinuoussearchinvolved,theyaretootimeconsuming.Anadaptivesearchtechniqueappearstoprovideasuitablesolution,asitiscapableofcoveringthesearchspacewithoutanalysingeverypoint.TwoGAprogrammesareimplementedwithintheIHS:onefortheoptimisationofANNarchitecture,theotherforsearchingthebestcombinationofgeardesignparameters.Duetotherestrictiononthelengthofthepaper,onlytheformerisdescribedinthissection,whilethelattercanbefoundin7.Trainingofaback-propagationnetworkhasprovedtobeadifficultprocess,duetothelackofeffectiverulesandguides.Toovercomethisproblem,aGAhasbeendeveloped.Basedupontheperformance(fitness)ofthenetwork,theGAisusedtooptimisethetopol-ogy,transferfunctionandtrainingperiod,whicharethemajorfactorsaffectingnetworkperformance.Opti-misationofthesefactorsisperformedsimultaneously,consideringtheircombinedeffectsuponperformanceandconvergence.Thisobviouslymakestheoptimisa-tionmoreeffective.Theinformationrelatingtothefactorsthataffecttheperformanceofanetworkduringtrainingareencodedintothegenesinbinaryformwithinchromosomes.Fig.5showsanexample,whichcorrespondstoasigmoidtransferfunction,19elementsinthefirsthiddenlayerand9inthesecondandatrainingfactorof5.TheoptimisationprocessisshowninFig.6.Uponinitiation,thevaluescontainedwithinthegenesarerandomlysetfromvalueswithinthesearchspace.Thenetworkscorrespondingtotheinformationcontainedwithinallthechromosomesofthepopulationaredeter-minedandsortedintoorderofdescendingfitness.