On-LineDiagnosisofDynamicSystemsbasedon Qualitative Models andDependency-recording DiagnosisEngines
Oskar Dressler · 2000
Dynamicsystemsimposeaninherent complexityonsimulation anddiagnosistasks .Inparticular, dependency-based diagnosis enginesperformextremelybadonmodelsofsuchsystems.Puttingthesediagnosers on-lineisanecessity fromtheapplication pointofview,buta quiteambitiousgoalinviewofprevious attempts inthisdirection . Understanding dynamicsystemsintermsusedbythequalitative reasoning communityandmakinguseofthedistinction between intea-state andtruer-state constraints allowsustomake amajor steptowardsasatisfying solution . Inter-state constraints aremotivatedbythecontinuity ofphysicalvariables andthedirection of(the)time(arrow).Previous authors, e.g.[131,[141, and[161,haveunconditionally assumed thatpredictions acrosstime,i.e. inter-state constraints, are necessaryfordiagnosingdynamicsystems.The key insight, however,thatenablesustodiagnosealargeclassofdynamic systemsefficiently isthatinter-state constraints oftenneednot becheckedforthepurposeofconsistency-based diagnosis atall! We identify sufficient conditions aboutmodels,observability, andfaults forthisresult . Basedon thisideawe employeda techniqueforcachingand temporallygeneralizing inferences tobuilda new diagnostic engine,Magellan-MT.Itiscapableofdiagnosingdynamic systemsatmultiple timesbutatthecomputational costofstatic systems. Magellan-MT requires qualitativemodelsto fully exploit thesecomputational advantages . We presentempirical results fromanapplication onballast tank systemsthatareusedonshipsandoffshore platforms .