Development of the Task-Based Expert System for Ma

DevelopmentoftheTask-BasedExpertSystemforMachineFaultDiagnosisMaBo1,JiangZhi-nong1andWeiZhong-qing21DiagnosisandSelf-RecoveryEngineeringResearchCenter,BeijingUniversityofChemicalTechnology,Beijing,China2ResearchandDevelopmentCenterofMachineryandElectricityEquipment,GeneralResearchInstituteforNon-FerrousMetals,Beijing,ChinaE-mail:[e-mail]xpertsystemswidelyusedinfaultdiagnosisistoformulateasetofdiagnosticrules,accordingtothemechanismandsymptomsoffaults,inordertoinstructthefaultdiagnosisordirectlygivediagnosticresults.Inpractice,duetodifferencesexistinginsuchaspectsasproductiontechnology,drivers,etc.,acertainfaultmayderivefromdifferentcauses,whichwillleadtoalowerdiagnosticaccuracyofexpertsystems.Besides,avarietyofexpertsystemsnowavailablehaveadualproblemoflowgeneralityandlowexpandability,ofwhichtheformercanleadtotherepeateddevelopmentofexpertsystemsfordifferentmachines,whilethelatterrestrictsusersfromexpandingthesystem.Aimedattheseproblems,atypeoftask-basedsoftwarearchitectureofexpertsystemisproposedinthispaper,whichpermitsaspecificoptimizationbasedonasetofcommonrules,andallowsuserstoaddormodifyrulesonaman-machinedialogsoastokeeponabsorbingandimprovingtheexpertknowledge.Finally,theintegrationoftheexpertsystemwiththeconditionmonitoringsystemtoimplementtheautomaticandsemi-automaticdiagnosisisintroduced.1．IntroductionAsabranchofartificialintelligence,expertsystemisanewappliedscienceproducedanddevelopedintheearly1960s.Asanintelligentprogram,itcanadoptacomputermodelofhumanexpertreasoningtodealwithcomplexissuesneededtobeexplainedintherealworld.SincetheMYCINexpertsystemappearedinthefieldofdiagnosisin1974,peoplehavealwaysbeenfocusingonproblemsintheresearchofknowledge-baseddiagnosisinartificialintelligence[1].Recently,inthefieldoffaultdiagnosis,diagnosisexpertsystemaimedtothedifferentmachinehasbeenstudied[2-4].Knowledgereasoningtechnologiesusedinthesestudiesmainlyincluderule-basedreasoning,25thInternationalCongressonConditionMonitoringandDiagnosticEngineeringIOPPublishingJournalofPhysics:ConferenceSeries364(2012)012043doi:10.1088/1742-6596/364/1/012043PublishedunderlicencebyIOPPublishingLtd1case-basedreasoning[5],reasoningbasedonartificialneuralnetwork[6],reasoningbasedonroughsettheory[7],model-basedreasoninganddiagnosis[8,9],reasoninganddiagnosisbasedonstatisticalmethods[10],combiningrule-basedandcase-basedreasoninganddiagnosis[11,12].Objectivesofthesestudiesareonlysomespecificmachines,oracertainkindofmachines.Compressorsarekeymachinesinoilrefining,chemical,powerandotherprocessindustries.Onceafaultoccurs,itwillcausethedowntimeofsystem,disruptioninproduction,evencatastrophicaccidents,whichcanresultinagreateconomiclossfortheproducingenterprise.Inordertomakecompressorsoperatesafely,real-timemachineconditionmonitoringsystemsaredeployedonmostofthem,whichcancaptureabnormalsymptomsbytheearlywarningtechnology,reducetheeconomicandhumanlosses,andimprovetheefficiencyofmachine[13].Currently,monitoringisgenerallystrongerthandiagnosisanalysisinmostconditionmonitoringsystems,someofthemhaveintegratedthefunctionofdiagnosisexpertsystem,thoughnotobviouslyeffectiveandshortofgenerality[14,15].Especially,expertsystemsthatareusuallyforaspecificmachineoracertainkindofmachinescannotbeappliedinthevariousotherkindsofcompressors.Forexample,whenreciprocatingcompressors,centrifugalcompressors,steamturbines,gasturbinesandotherkindsofmachinesaremonitoredbyconditionmonitoringsystemsimultaneously,anexpertsystemwillnotbeobviouslysatisfiable.Inthispaper,aarchitectureofTask-BasedDiagnosisExpertSystem(TBDES)formachinefaultisproposed,itcanbeflexiblyintegratedintothemachineconditionmonitoringsystem,accordingtothedifferentkindsofmachine,anddifferentdiagnostictaskscanbechosenforexpertdiagnosis.Inordertoescalatetheabilityoffaultdiagnosis,object-orientedknowledgerepresentationmethodisproposed,thatistosay,therulesofspecificmachinecanbecustomizedflexiblyonthebasisofgeneralrules.Multipleinteractingmodesofknowledgearedesignedaswelltoimprovetheusabilityofsystem.2．ExpertSystemDesign2.1．SystemArchitectureDesignInaccordancewiththeapplicationmodelofconditionmonitoringsystemandcharacteristicsoffaultdiagnosis,Task-basedfaultdiagnosisexpertsystemadoptstheframeworkofforwardreasoningofproductionrules,inwhichafactisdefinedbyatriplestructureofobject-attribute-value.Forexample,forthefact‘dominantfrequency’,thepropertyis‘frequency’,andthevalueis‘workingfrequency’.SchematicoverviewofTBDESisshowninFigure1,andthekernelofTBDESisinthedashedboxofFigure1.TheKernelconsistsoftwomodules,includingrulereasoningmoduleandinterpretationmodule,andsixknowledgebases,includingrulebase,faultbase,factbase,attributebase,maintenanceadvicebaseandtaskbase.Therulereasoningmoduleistheinferenceengineoftheexpertsystem,whichadoptstheforwardreasoningtechnique.Theprincipleofrulereasoningistocombinefactsandrulesstoredintheworkingmemorytoestablishareasoningmodel,sothatnewinformationcanbedrawn.Theinterpretationmoduleistoexplaintheresultoffaultdiagnosisandthusgivemaint