Alarm flood reduction using multiple data sources

95 Pages·2014·3.27 MB·English

Alarm flood reduction using
multiple data sources
Vicent Rodrigo Marco

Department of Automatic Control
MSc Thesis
ISRN LUTFD2/TFRT--5961--SE
ISSN 0280-5316
Department of Automatic Control
Lund University
Box 118
SE-221 00 LUND
Sweden
© 2014 by Vicent Rodrigo Marco. All rights reserved.
Printed in Sweden by Media-Tryck
Lund 2014
Abstract
Theintroductionofdistributedcontrolsystemsintheprocessindustryhasincreased
thenumberofalarmsperoperatorexponentially.Modernplantspresentahighlevel
of interconnectivity due to steam recirculation, heat integration and the complex
control systems installed in the plant. When there is a disturbance in the plant it
spreadsthroughitsmaterial,energyandinformationconnectionsaffectingthepro-
cess variables on the path. The alarms associated to these process variables are
triggered.Thealarmmessagesmayoverloadtheoperatorinthecontrolroom,who
willnotbeabletoproperlyinvestigateeachoneofthesealarms.Thisundesiredsit-
uationiscalledan“alarmflood”.Insuchsituationstheoperatormightnotbeableto
keeptheplantwithinsafeoperation.Theaimofthisthesisistoreducealarmflood
periodsinprocessplants.Consequentialalarmscomingfromthesameprocessab-
normality are isolated and a causal alarm suggestion is given. The causal alarm
in an alarm flood is the alarm associated to the asset originating the disturbance
thatcausedtheflood.Multipleinformationsourcesareused:analarmlogcontain-
ingallpastalarmsmessages,processdataandatopologymodeloftheplant.The
alarm flood reduction is achieved with a combination of alarm log analysis, pro-
cess data root-cause analysis and connectivity analysis. The research findings are
implemented in a software tool that guides the user through the different steps of
themethod.Finallytheapplicabilityofthemethodisprovedwithanindustrialcase
study.
5

Acknowledgements
I would like to use this oportunity to extend my sincere thanks to all the people
withoutwhomthisthesiswouldnothavebeenpossible.
Iwishtothank,firstandforemost,mysupervisorDrMoncefChioua(researcher
in ABB Corporate Research Center in Ladenburg, Germany) for his patient guid-
ance,encouragementandhiscriticalfeed-back.Hehelpedmetoshapemythoughts
intherightdirection,notonlyonthisworkbutalsoformyfutureasanengineer.I
considermyselfextremelyluckytohavesuchatalentedsupervisorthatwasalways
availabletorespondtomyquestionsandqueries.
Besidesmyadvisor,IwouldliketothankDrMartinHollenderforhisencour-
agementandinsightfulcomments.
Idedicatethisthesistomyparents,SalvaandRosa,andmybrotherCarlesfor
theirlove,advicesandunconditionalsupportthroughoutmylife.
7

Contents
1. Introduction 11
1.1 AlarmSystems . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 StateoftheArt . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Scope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2. PatternanalysisofAlarmsequences 19
2.1 ModifiedSmith-Watermanalgorithm . . . . . . . . . . . . . . . 19
2.2 AgglomerativeHierarchicalClustering . . . . . . . . . . . . . . 23
3. Root-causeanalysisbasedonProcessData 27
3.1 PDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4. ConnectivityanalysisbasedonPlantTopology 33
4.1 Graphsinfaultdiagnosis . . . . . . . . . . . . . . . . . . . . . 33
4.2 IntelligentP&IDs . . . . . . . . . . . . . . . . . . . . . . . . . 38
5. TheMethod 41
5.1 RemoveChatteringalarms. . . . . . . . . . . . . . . . . . . . . 42
5.2 Identificationofalarmfloodperiods. . . . . . . . . . . . . . . . 45
5.3 Clusteralarmfloodsequences . . . . . . . . . . . . . . . . . . . 46
5.4 Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.5 Root-causeanalysis . . . . . . . . . . . . . . . . . . . . . . . . 52
5.6 Plantareasselection . . . . . . . . . . . . . . . . . . . . . . . . 53
6. Softwaretool 55
6.1 Alarmloganalysis . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Processdataanalysis. . . . . . . . . . . . . . . . . . . . . . . . 61
6.3 Topologyanalysis . . . . . . . . . . . . . . . . . . . . . . . . . 67
7. Industrialcasestudy 72
7.1 Processdescription . . . . . . . . . . . . . . . . . . . . . . . . 72
7.2 Plantareaidentification . . . . . . . . . . . . . . . . . . . . . . 73
7.3 Analysisarea01 . . . . . . . . . . . . . . . . . . . . . . . . . . 76
9
Contents
8. Conclusions 92
8.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.2 FutureWork . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Bibliography 93
10
1
Introduction
This thesis presents the findings from the research conducted into establishing in-
novative ways to manipulate process plant data to facilitate the supervision and
monitoringofprocessplants.Anoff-linemethodtoreducealarmfloodinginalarm
monitoringsystemsisproposedandasoftwaretoolthatprovespracticalapplicabil-
ityoftheresearchfindingsisimplemented.
1.1 Alarm Systems
Theprofitintheprocessindustryishighlyrelatedtotheplantoperation.Overthe
last decades advanced operation and control approaches have been developed in
order to optimally operate plants. This is done by keeping process variables at a
specific value [Christofides et al., 2007]. A deviation from the optimal operation
pointisusuallytranslatedintoaneconomiclossorendangeringtheenvironmentor
thesafetyofthepersonnel.Theoperatorsworkinginthecontrolroomareincharge
ofkeepingtheplantoperatingatthispoint.
Analarmsystemisthemainelementthatinterfacesinmodernplantswiththe
operator to the plants. It is a crucial element since it monitors the plant operation
and alerts the operator when some undesired state that requires his assessment or
actionisreached.Itsmainobjectivesare:toassisttheoperatortocorrectpotentially
dangeroussituationsbeforetheemergencyshutdownsystemistriggered,toavoid
financial loss by identifying deviations from the optimal operating conditions and
to help the operator to better understand the process conditions that gave rise to
the upset. The primary function of an alarm system is defined by the EEMUA as
follows:
The purpose of an alarm system is to direct the operator’s attention towards
plantconditionsrequiringtimelyassessmentoraction[EEMUA-191,2007].
11
Chapter1. Introduction
1.1.1 Alarms in Process Industry
Analarmisasignalsenttotheoperatorsinordertodrawtheirattentionduringan
abnormalsituationintheplant.Analarmoccurrencecomesusuallytogetherwitha
sound,flashinglightandanindicator.Inaddition,alarmsusuallypresentamessage
withsomeinformationabouttheproblem.
Every alarm presented to the operator should be useful and relevant to the
operator [EEMUA-191,2007].Ideally,foreachabnormalsituationjustonealarm
israised.However,operatorsusuallyreceivealargeamountofalarmsinpractice.
Most of these alarms are either false alarms, i.e. alarms that alert of an abnormal
situation when there is none, or nuisance alarms, i.e. alarms that are redundant,
sinceotheralarmshavealreadyinformedoftheabnormality.
Traditionallyanalarminaprocessplantisassociatedtoaprocessvariablegoing
outofitsnormalrange(seeFigure1.1).
High alarm
High alarm trip point
Low alarm trip point
Low alarm
Figure1.1 Signalalarm
These type of alarms are called absolute alarms. However, one should bear in
mindthatmanyotheralarmdetectionmechanismsareusedinpractice,andnotall
12