Neural network based framework for fault diagnosis in batch chemical plants.pdf

《Neural network based framework for fault diagnosis in batch chemical plants.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Computers&ChemicalEngineeringELSEVIERComputersandChemicalEngineering24(2000)777-784www.elsevier.com/locate/compchemengNeuralnetworkbasedframeworkforfaultdiagnosisinbatchchemicalplantsDiegoRuiza,Jos~MariaNougu~sa,ZulyCalder6nb,AntonioEspufiaa,LuisPuigjanera,,aChemicalEngineeringDepartment

2、,UniversitatPolit~cnicadeCatalunya,Av.Diagonal647,E-08028Barcelona,SpainbUniversidadIndustrialdeSantander,EscueladelnginerladePetrdleos,AA.678,Bucaramanga,ColombiaAbstractInthiswork,anartificialneuralnetwork(ANN)basedframeworkforfaultdiagnosisinbatchchemicalplantsispresented.TheproposedF

3、DSconsistsofanANNstructuresupplementedwithaknowledgebasedexpertsystem(KBES)inablock-orientedconfiguration.ThesystemcombinestheadaptivelearningdiagnosticprocedureoftheANNandthetransparentdeepknowledgerepresentationoftheKBES.TheinformationneededtoimplementtheFDSincludesahistoricaldatabaseo

4、fpastbatches,ahazardandoperability(HAZOP)analysisandamodelofthebatchplant.ThehistoricaldatabasethatincludesinformationrelatedtonormalandabnormaloperatingconditionsisusedtotraintheANNstructure.Thedeviationsoftheon-linemeasurementsfromareferenceprofileareprocessedbyamulti-scalewaveletinord

5、ertodeterminethesingularitiesofthetransientsandtoreducethedimensionalityofthedata.TheprocessedsignalsaretheinputsofanANN.TheANNsoutputsarethesignalsofthedifferentsuspectedfaults.TheHAZOPanalysisisusefultobuildtheprocessdeepknowledgebase(KB)oftheplant.Thisbasereliesontheknowledgeoftheoper

6、atorsandengineersabouttheprocessandallowstheformulationofartificialintelligencealgorithms.Thecasestudycorrespondstoabatchreactor.TheFDSperformanceisdemonstratedthroughthesimulationofdifferentprocessfaults.TheFDSproposedisalsocomparedwithotherapproachesbasedonmulti-wayprincipalcomponentan

7、alysis.©2000ElsevierScienceLtd.Allrightsreserved.Keywords:Faultdiagnosis;Batchplants;Artificialneuralnetworks1.Introductionquencesofabnormalsituationscanbeanticipated.Arobustfaultdiagnosissystem(FDS),thattimelypro-videsthefaultinformationtoboththesupervisoryandSupervision