Knowledge-based Inference Methods for Modeling

KILaborfürKünstlicheIntelligenzForschung,Entwicklung,Beratung,AusbildungKnowledge-basedInferenceMethodsforModelingTechnicalSystemsGerdKamp,BerndNeumannOctober1996LKI-M-96/04Laborf¨urK¨unstlicheIntelligenzFachbereichInformatik,Universit¨atHamburgVogt-K¨olln-Str.30,22527Hamburg,GermanycLaborf¨urK¨unstlicheIntelligenz1996Thisworkmaynotbecopiedorreproducedinwholeofpartforanycommercialpurpose.Permissiontocopyinwholeorpartwithoutpaymentoffeeisgrantedfornonprofiteducationalandresearchpurposesprovidedthatallsuchwholeorpartialcopiesincludethefollowing:anoticethatsuchcopyingisbypermissionoftheLaborf¨urK¨unstlicheIntelligenz,Hamburg,Germany;anacknowledgementoftheauthorsandindividualcontributorstothework;allapplicableportionsofthiscopyrightnotice.Copying,reproducing,orrepublishingforanyotherpurposeshallrequireanagreementwithLaborf¨urK¨unstlicheIntelligenz.Knowledge-basedInferenceMethodsforModelingTechnicalSystemsGerdKamp,BerndNeumannLKI-M-96/04Appearsin:30thHawaiianInternationalConferenceonSystemsSciences,Wailea,HI,Com-puterSciencePressThisresearchhaspartiallybeensupportedbytheDFGunderthegrantDFGNe/279Knowledge-basedInferenceMethodsforModelingTechnicalSystemsGerdKampBerndNeumannUniversityofHamburg,Vogt-Klln-Str.30,D-22527HamburgEmail:{kamp,neumann}@informatik.uni-hamburg.deAbstractDescriptionLogicswithconcretedomainspresentanapproachtorealizeagen-eralengineeringworkbench.Theyprovidearepresentationlanguagethatenablesustodescribeinauniformwaydevices,assembliesandcomponentsalongwiththeirstructure,constraintsonattributesandphysicallawsaswellasmodelsoftheircor-rectandfaultybehavior.Furthermore,soundandcompletealgorithmscanbegivenforasetofbasicinferences.Thesebasicinferencesrenderitpossibletosimulatethebehaviorofthedevicesandprovidethebasicbuildingblocksforconsistency-baseddiagnosis.Inadditiontheyenableustodeviseproceduresforndingerrors,omissionsandinconsistenciesinmodellibraries.1IntroductionDuringthedesignandengineeringphaseofanewtechnicaldeviceitisnecessarytodevelopsolutionsforseveraltasks,suchascongurationfromcomponentsandassemblies,determinationofparametervalues,simulationofbehavior,intelligentselectionfrompartscatalogsanddiagnosisofprototypes.Traditionallytheseproblemsaretackledwithanumberofspecictools.Thesetoolsmostoftenbestand-alonesystemsandarenotdesignedtocommunicatewitheachother.Inotherwords,everytoolrealizesitsownlanguagetodescribethedevice,therebypreventingresultsobtainedbyonetooltobeemployedinothertools.Inthispaperwepresentanapproachforanintegratedengineeringworkbenchbasedondescriptionlogicsextendedwithconcretedomains.Theknowledgerepresentationlanguagethatissuppliedbythisapproachenablesustodescribeimportantmodelfea-turesinauniformlanguage,e.g.:componenttypes,componentstructures,physicallawsandmodelsofcorrectandfaultybehavior.Anumberofbasicinferenceservicescanbedened,andsoundandcompletealgorithmsfortheseinferencescanbegivenforthedescriptionlanguageused.Firstresultsshowthatbymakinguseofthesebasicinference1servicesitisnotonlypossibletosimulatethebehaviorandsupportthediagnosisofthesystem,butalsotodetecterrors,inconsistenciesandlacunaeinmodellibraries.Theremainderofthispaperisorganizedasfollows:Firstwegiveabriefintroductiontodescriptionlogicsandthefollowingenhancementsneededintechnicaldomains:concretedomains,modelgenerationandcalculationofadmissibleparameterranges.Wethenfocusonhowdierentengineeringtaskscanbetackledwiththisapproach.InSection2weshowhowthebasicconceptsneededtodescribegearwheelmechanismscanberepresentedwithinthedescriptionlanguage.Section3showshowsimulationbehaviorcanbeaccomplishedviamodelgenerationandcalculationofadmissibleparametervalues.Section4presentsanapproachtodiagnosisbasedonobjectclassication.Section5showshowtheinferenceserviceofconceptclassicationisusefulforbuildingandmaintaininglargemodellibraries.Asummaryandanoutlookconcludethepaper.1.1DescriptionLogicsDescriptionlogics(DL)havealongtraditioninorganizinginformationwithapowerfulobject-orientedrepresentationschemeandclearlydenedsemantics.Descriptionlogicssystemsmainlyconsistoftwoparts:aTerminologicalBox(TBox)andanAssertionalBox(ABox)1:1.1.1TBoxAtthecoreofdescriptionlogicsliesthenotionofconceptterms.Startingwithprimitiveconceptandroleterms,newconcepttermscanbeconstructedfromothersbyasetofconceptformingoperators.Therearemainlythefollowingcategoriesofsuchoperators[6]:1.Booleanoperators(andCD:::),(orCD:::),(notC),allowingforthecombinationofconceptswithoutareferencetotheirinternalstructure.2.Roleformingoperatorsthatallownewrolestobedened,e.g.compositionofroles(composers).3.Operatorsonrolellersthatoperateontheinternalstructureoftheconceptterms,e.g.providequanticationoverrolellers(somerC).Terminologicalaxiomsoftheform(dene-conceptCNC)associateaconceptnameCNwithaconcepttermCandareusedtodenetherelevantconceptsofanapplication.Ter-minologicalaxiomsareroughlycomparabletotheclassdenitionsofanobject-orientedrepresentationlanguage2.FinallyaTBoxTisanitesetofterminologicalaxioms.1Wecannotelaborateonthebasicsofdescriptionlogicsandreferthereadert