Data-Driven Fault Detection for Industrial Processes: Canonical Correlation Analysis and Projection Based Methods

124 Pages·2017·3.847 MB·English

by  Zhiwen Chen (auth.)

Zhiwen Chen
Data-Driven
Fault Detection for
Industrial Processes
Canonical Correlation Analysis
and Projection Based Methods
Data-Driven Fault Detection for Industrial
Processes
Zhiwen Chen
Data-Driven
Fault Detection for
Industrial Processes
Canonical Correlation Analysis
and Projection Based Methods
Zhiwen Chen
Duisburg, Germany
Dissertation, University of Duisburg-Essen, 2016
ISBN 978-3-658-16755-4 ISBN 978-3-658-16756-1 (eBook)
DOI 10.1007/978-3-658-16756-1
Library of Congress Control Number: 2016961279
Springer Vieweg
© Springer Fachmedien Wiesbaden GmbH 2017
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer Vieweg imprint is published by Springer Nature
The registered company is Springer Fachmedien Wiesbaden GmbH
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany
Tomyparents,myloveCaiwenandmyson
Preface
Inordertomaximizethecustomersatisfactionandprofitaswellastoobeygovernment
regulations, the complexity and automation degree of modern industrial processes are
significantly growing. To ensure the safety and overall reliability of such complicated
processes,automatizedfaultdetectionisofgreatimportance. Althoughthemodel-based
faultdetectiontheoryhasbeenwellstudiedinthepastdecades,itsapplicationsarestill
limitedforlarge-scaleindustrialprocessesbecauseitisdifficulttoestablishaccuratemodel
bymeansoffirstprinciples. Ontheotherhand,sufficient(real-time)dataarecollected
andrecordedduringprocessoperations,andhigh-speedcomputationisavailabledueto
therapidimprovementinsensorandcomputertechnologies. Therefore,themainobjective
of this work is to develop advanced data-driven fault detection methods for different
applicationscopes.
Thisworkisfirstlydedicatedtoevaluatebasicfaultdetectionstatisticswithanalter-
nativeperformanceindex. Thecommonlyused T2 andQstatisticsarecomparedwith
respecttothegeometricrelationshipandperformanceindices.
Thefurtherstudyfocusesondevelopingeffectivefaultdetectionmethodsforstaticand
steady-statedynamicprocesseswithavailableprocessinputandoutputdata. Different
fromthewell-establishedmethodsbasedonmultivariateanalysistechniques,thecoreof
theproposedmethodsistobuildresidualsignalsbymeansofthecanonicalcorrelation
analysistechniqueforthefaultdetectionpurpose. However,theproposedmethodsareless
powerfultodealwithincipientmultiplicativefaults. Toimprovetheirfaultdetectability
ofsuchfaults,thestatisticallocalapproachisintegratedintotheoriginalmethods.
For dynamic processes, an alternative fault detection method is proposed, in which
residualsignalsaregeneratedbymeansofaprojectionofprocessinputdata. Thisway
ofresidualgenerationcircumventstheparameteridentificationprocedure. Ontheother
hand,italsoallowsustoaddressdeterministicdisturbances,whichareoftennottaken
intoaccountbytheexistingdata-drivenfaultdetectionmethods. Finally,theproposed
faultdetectionmethodsareverifiedbyfourbenchmarkcases,i.e. thealuminaevaporation
process, theTennesseeEastmanprocess, thepilotscalecontinuousstirredtankheater
andtheinvertedpendulumsystem. Theapplicationresultsshowstheeffectivenessofthe
proposedmethods.
ThisworkhasbeendoneattheInstituteforAutomaticControlandComplexSystems
(AKS)intheFacultyofEngineeringattheUniversityofDuisburg-Essen,Germany. First
VIII Preface
ofall,Iowethedeepestgratitudetomysupervisor,Prof. Dr.-Ing. StevenX.Ding,forall
theinspirationandhelpheprovidedduringthecourseofthelastfouryears. Iamheartily
thankfulforhisguidanceonmyscientificresearchwork. Mysincerethanksmustgoto
Prof. Dr. PingZhangforthevaluablecomments,whichimprovedthequalityofthiswork.
IwouldliketoexpressmysincerethankstomygroupmatesDr.-Ing. KaiZhang,Dr.
ZhangmingHeandDr.-Ing. HaiyangHaoforthevaluablediscussionsandsuggestions;
Prof. YuriShardtforsharinghisrichandvaluableexperienceonacademicresearchand
scientificwriting.
IwouldliketothankDr.-Ing. LinlinLi,M.Sc. ChangchenXiang,Dr.-Ing. DongmeiXu,
Dr. YongZhang,M.Sc. SihanYu,Dr.-Ing. HaoLuo,M.Sc. MinjiaChang,Dr. ShashaLi,
Dr.-Ing. YingWangfortheirhelpduringmystayatAKS.Mythanksshouldalsogoto
allmyotherAKScolleagues,Dr.-Ing. TimKo¨nings,Dr. ZhiqiangGe,Dr. MingzhuTang,
Dr.-Ing. ChrisLouen,Mrs. SabineBay,Dr.-Ing. ShaneDominic,Dr.-Ing. Ko¨ppen-Seliger,
Dipl.-Ing. EberhardGoldschmidt,Dr. QingchaoJiang,M.Sc. LuQian,M.Sc. Changsheng
Hua, M.Sc. Yunsong Xu, M.Sc. Zhengen Zhao, M.Sc. Tim Daszenies, M.Sc. Judith
Minten,M.Sc. SvenjaSiewers,Dipl.-Ing. KlausGo¨belandMr. UlrichJanzen,aswellas
theformercolleaguesM.Sc. PingLiu,Prof. YaguoLei,Prof. BoShen,Prof. HongliDong,
Prof. XuYang, Prof. JianbinQiu, Dr. ShouchaoZhai, Prof. Dr.-Ing. ShenYin, Prof.
KaixiangPengandProf. YingYangfortheirvaluablediscussionsandhelpfulsuggestions.
Withouttheirhelpthisworkwouldnothavebeencompletedatthislevel.
Iwillforeverbeindebtedtomybigfamily, especiallymyparentsandsisters, forall
theirsupportandlove. Inparticularly,Iwouldliketothankmywife,CaiwenLi,forall
hersupport,encouragement,patience,andforbeingbymyside.
MyspecialthanksmustgotoProf. ZhikunHu,awonderfulmentor,whointroduced
meto‘faultdiagnosis’andpassedawayduringthewritingofthework. Hewillbedeeply
missed.
ZhiwenChen
Contents
Preface VII
List of Figures XIV
List of Tables XV
List of Notations XVII
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 BasicConceptsandMotivationoftheWork . . . . . . . . . . . . . . . . . 1
1.2.1 BasicConceptsofFaultDetection . . . . . . . . . . . . . . . . . . . 1
1.2.2 MotivationfortheWork . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 ObjectivesoftheDissertation . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 OutlineoftheDissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 The Basics of Fault Detection 13
2.1 MathematicalDescriptionsofIndustrialProcesses . . . . . . . . . . . . . . 13
2.1.1 RepresentationofStaticProcesses. . . . . . . . . . . . . . . . . . . 13
2.1.2 RepresentationofDynamicProcesses . . . . . . . . . . . . . . . . . 14
2.2 BasicPrincipleofFaultDetection . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 BasicStatisticalFaultDetectionMethods . . . . . . . . . . . . . . . . . . 16
2.3.1 T2 andQTestStatistics . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 PrincipalComponentAnalysis-basedMethod . . . . . . . . . . . . 18
2.3.3 PartialLeastSquares-basedMethod . . . . . . . . . . . . . . . . . 21
2.3.4 DynamicalPCA-basedMethod . . . . . . . . . . . . . . . . . . . . 22
2.4 QuantitativeModel-basedResidualGeneration. . . . . . . . . . . . . . . . 23
2.4.1 FDF-basedResidualGeneration . . . . . . . . . . . . . . . . . . . . 23
2.4.2 DO-basedResidualGeneration . . . . . . . . . . . . . . . . . . . . 24
2.4.3 PS-basedResidualGeneration . . . . . . . . . . . . . . . . . . . . . 24
2.4.4 KernelRepresentation . . . . . . . . . . . . . . . . . . . . . . . . . 25
X Contents
2.5 Data-DrivenKernelRepresentation-basedFDMethods . . . . . . . . . . . 26
2.5.1 Data-DrivenRealizationoftheKernelRepresentation . . . . . . . . 26
2.5.2 Data-DrivenPS-basedMethod. . . . . . . . . . . . . . . . . . . . . 27
2.5.3 Data-DrivenDO-basedMethod . . . . . . . . . . . . . . . . . . . . 28
2.6 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Evaluation and Comparison of T2 and Q Statistics for Fault Detection 31
3.1 PerformanceEvaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 FARandFDR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.2 MeanTimetoAFalseAlarm . . . . . . . . . . . . . . . . . . . . . 33
3.2 GeometricRelationshipbetweenT2 andQStatistics . . . . . . . . . . . . 34
3.2.1 DistributionofQisApproximatedbyλ χ2(m). . . . . . . . . . . . 36
1
3.2.2 DistributionofQisApproximatedbytr(Σ )χ2(1) . . . . . . . . . . 37
y
3.2.3 DistributionofQisApproximatedbygχ2(h). . . . . . . . . . . . . 37
3.3 NumericalExampleStudy . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.1 ExampleforGeometricRelationship . . . . . . . . . . . . . . . . . 39
3.3.2 ExampleforPerformanceEvaluation . . . . . . . . . . . . . . . . . 39
3.4 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Canonical Correlation Analysis-based Fault Detection Methods 43
4.1 BackgroundandProblemFormulation . . . . . . . . . . . . . . . . . . . . 43
4.2 TheBasicofCCATechnique . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3 CCA-basedFDMethodforStaticProcesses . . . . . . . . . . . . . . . . . 46
4.3.1 CCA-basedFDMethod . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3.2 AnIllustrativeExample . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 DCCA-basedFDMethodforDynamicProcesses. . . . . . . . . . . . . . . 50
4.4.1 ModelingofInputandOutputDataSets . . . . . . . . . . . . . . . 50
4.4.2 DCCA-basedFDMethod . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.3 AnIllustrativeExample . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.6 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5 Improved CCA-based Fault Detection Methods 59
5.1 BackgroundandProblemFormulation . . . . . . . . . . . . . . . . . . . . 59
5.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1.2 AMotivationExample . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2 IntegratingtheStatisticalLocalApproachintoCCAforFD . . . . . . . . 61
5.2.1 IntroductionoftheStatisticalLocalApproach . . . . . . . . . . . . 61
5.2.2 DerivationofthePrimaryResidual . . . . . . . . . . . . . . . . . . 62
5.2.3 TheFaultDetectionMethod . . . . . . . . . . . . . . . . . . . . . . 63
Contents XI
5.3 OnDetectingIncipientMultiplicativeFaultsUsingtheT2 Statistic . . . . 65
5.3.1 T2 StatisticforDetectingMultiplicativeFaults. . . . . . . . . . . . 66
5.3.2 Statistical Local Approach for Detecting Incipient Multiplicative
Faults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6 A Projection-based FD method for Dynamic Processes with determin-
istic disturbances 71
6.1 BackgroundandProblemFormulation . . . . . . . . . . . . . . . . . . . . 71
6.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.1.2 ProblemFormulation . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.2 I/ODataModelwithDeterministicDisturbancesandFaults . . . . . . . . 72
6.3 TheProjection-basedData-DrivenFDMethod. . . . . . . . . . . . . . . . 75
6.3.1 On-lineUpdateOrthogonalProjectionMatrix . . . . . . . . . . . . 76
6.3.2 AlgorithmfortheProposedApproach . . . . . . . . . . . . . . . . 77
6.3.3 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.4 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7 Benchmark Studies 79
7.1 CaseStudyontheAluminaEvaporationBenchmarkProcess . . . . . . . . 79
7.1.1 ProcessDescription . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.1.2 ApplicationoftheCCA-basedMethods. . . . . . . . . . . . . . . . 80
7.2 CaseStudyontheCSTHBenchmarkProcess . . . . . . . . . . . . . . . . 85
7.2.1 ProcessDescription . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2.2 ApplicationofImprovedStaticCCA-basedMethod . . . . . . . . . 86
7.3 CaseStudiesontheTEBenchmarkProcess . . . . . . . . . . . . . . . . . 88
7.3.1 ProcessDescription . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.3.2 ApplicationofImprovedDynamicCCA-basedMethod . . . . . . . 89
7.4 CaseStudyontheInvertedPendulumBenchmark . . . . . . . . . . . . . . 92
7.4.1 BenchmarkDescription. . . . . . . . . . . . . . . . . . . . . . . . . 93
7.4.2 SimulationSetting . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.4.3 ResultsandDiscussion . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.5 ConcludingRemarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8 Conclusions and Future Work 99
8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
8.2 FutureWork. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Bibliography 103