Research

2005PetroleumScienceV.01.2No.3

    ;ResearchonPick-upFeatureofESPRecordingChartsBasedon

    ;BitmapImages

    ;FengDing一,TuYiliu,ZhuHongwuandXueDunsong

    ;flCollegeofMechanicalEngineering,YangtzeUniversity,Jingzhou434023,China) ;Dept.ofMechanicalandManufacturingEngineering,UniversityofCalgary,CalgaryT2N1N4,Canada) ;f3CollegeofMechanicalandElectricalEngineerin~ChinaUniversityofPetroleum.Beijing102249,China)

    ;ReceivedApril3,2005

    ;Abstract:Duringtheoilproduction,inordertomonitortheworkingconditionsofanelectricalsubmersiblepump(ESP),

    ;anelectricalcurrentrecorderisusedtomonitortheelectricmotorcurrentofanESETherecorderchartsindicatevarious

    ;workingconditionsoftheESPSubtlemalfunctionsorabnormalproblemsoftheESPcanbedetectedandfurther

    ;analyzedfromvariousfeaturesofthesecurrentcurvesontherecordingcharts.Presently,thesecurrentchartsare

    ;manuallyreadandanalyzedinoilfields.Inthispaper,adiagnosisexpertsystemispresentedforautomaticallyanalyzing

    ;thesecurrentrecordingchartsandidentifyingtheworkingconditionoftheESPThisexpertsystemincludesanopen

    ;knowledgebase,whichcanbeupdatedorenrichedaccordingtotheidentifiedfeaturesofthecurrentcurvesonthe

    ;recordingcharts.andaconditionmonitoringandfailurepa~emrecognitionmethod,whichiscalled’’pick-upmethodof

    ;featureoftherecordingchart”.andcanbecorrectlyappliedinmostcases.Ithasbeenshownthatthisexpert

    systemcan

    ;effectivelyimprovetheaccuracyandefficiencyoffailurediagnosisandworkingconditionmonitoringofESPs.

    ;Keywords:Electricalsubmersiblepump,expertsystem,faultdiagnosis,machineconditionmonitoring,pa~em

    ;identificationorrecognition

    ;1.IntrOductiOn

    ;Asaneffectivetool,theelectricalsubmersiblepump

    ;(ESP)hasbeenincreasinglyemployedinoilproduction.

    ;ESPisamultistagecentrifugalpumpdrivenbyan

    ;electricmotorThepumpandthemotorarenormally

    ;suspendedfromtheproductiontubing,withthemotor

    ;positionedbelowthepump,andthepumpdischarging

    ;directlyintotheproductiontubing.Powerissuppliedto ;themotorthroughathree.conductorcable,whichisrun ;downtothemotorfromthesurfacethroughthe ;tubing/casingannulus.

    ;Thevelocitythatisimpartedtothefluidbythe ;rotatingimpellerisconvertedintopressureenergyby ;thediffuserinacentrifugalpump.Eachimpeller/diffuser ;setiscalleda”stage”.Thedischargeofonestageisthe

    ;inputofthenextstageinamultistagepumpandhence ;higherpressures,whichwouldbeimpossiblewitha ;singleimpeller.canbeachieved.V撕ablespeeddrivers

    ;(VSDs)changethefrequencyofthepowersuppliedto ;themotoLthuschangingthepumpspeedandthe ;operatingenvelopeaccordingtotheamnitylaws.Since ;shafthorsepowerislinearlyproportionaltomotor ;frequency,smallerpumpsandthemotorsthatrunata ;higherfrequencycangiveequivalentperformance.An ;electricalcurrentrecorderinstrumentcanbeinstaliedto ;monitorandrecordtheelectricalcurrentcurvesofthe ;pumpandthemotordailyorweekly.Recordingshould ;bemadewhilethewellisperformingproperly,sothat ;comparisonscanbemadeifanyfaultdevelops.The ;recordingchartscanbeusedtocontinuouslymonitor ;anddiagnosetheelectricalcurrentofthedrivingmotor ;ofanESP.

    ;Thepick-upmethOdOffeaturesoftherecording ;chartisakeystepforusingacomputertOautomatically ;recognizetherecordingcharts.V撕OUSfeaturesonthe

    ;recordingchartsreflectlotsofpossiblefaultsofan ;electricalsubmersiblewellsystem(HuangandShi. ;2000;Coben.1997).Pick.upfeaturebecomesoneofthe ;difficulttaskstobuildapaaemrecognitionsystemfor ;automaticallyidentifyingtherecordingcharts.Some ;importantfeaturescannotbeeasilyfoundoridentified ;becauseofbeingrestrictedbythecomplicated ;environmentandworkingconditionsOfESPinthewel1. ;Therefore,moreandmorepeopleintheworldare ;pa~ngattentiontothepick-upfeatureproblem. ;Pick.upfeatureisessentialtoidenfifyingthe ;distinctivecurvesfromtherecordingcardsbyusing ;mathematicmethodstodescribesomephysical ;characteristiesofacurrentcurve.Fordi虢rentcurves.

    ;featuredistinctionmaybelittleinthesamekindbut ;greatindifferentkinds.Featureshouldneverchange

;

    ;6PetroleumScience2005

    ;alongwithparallelmoving,rotatingandscalingacurve. ;Therearemanymethodsforfeaturepick.up,e.g. ;unchangedmatrix,Fourierdescriptorandtimeseries ;model(Bian,1998;Xiong,1999),etc.Thedrawbackof ;theunchangedmatrixmethodisthatitcannotreflect ;enoughlocalinformationofafigure(orcurrentcurve) ;onarecordingchartthroughlimitedinvariables.A ;Fourierdescriptorhasenoughnumbersofcharactersto ;reflectsmallchangesofthelpealfigure.butitneedsa ;lotofworkforfurtherpaRernrecognition.Furthermore. ;byusingtheFourierdescriptorthecharactercannotbe ;processedbyparallelmoving,rotationorscalingofthe ;currentcurve.Bothmethodsaresensitivetounwanted ;noiseonthechart.Inrecentyears,thetechniqueofa ;timeseriesmodelhasbeenappliedinthefieldof ;patternrecognition.

    ;Muchattentionhasbeenpaidtothedevelopmentof ;therecordingchartsrecognition,butfewresultshave ;beenachievedinconcernwiththeconfidenceofthe ;predictionsbyusingtherecordingchartsandofthe ;circumstancesunderwhichthepredictionsarevalid. ;Thisstatementcanbefurtheriusfifiedthroughthe ;followingliteraturereview.

    ;2.Literaturereview

    ;Theuseofelectricalsubmersiblepumpsisthe ;predominantartificialliffmethodintheoilindustry ;andhasbeenwidelyusedindeepwells.arctic ;locations,o凰horeanddirectionalwellstoprovide ;artificialliffpower.Everyyear,alotofESPfailures ;occurandadverselyaecttheliftingcostsandrig ;utilityandproductivity.TheoccurrenceofESPfailures ;causedbyspinningdiflusersistolerablyfrequentin ;moderateoperating-costcircumstances.Alongitudinal ;compressiveforceisimposedonthediffuserstackof ;ESPduringassemblytopreventdifluserrotation.The ;mechanicsofthespinning.diffuserfailuremodethat ;demonstrateswhysomepumpswithaninadequate ;compresswelbrcecanpasscommonpumptestsbutfail ;inthismodeisanalyzedinreference(Powers.2002). ;stacisticalmethodsbasedontheWleibulldistribution ;andonthebathtubreliabilitymodelhavebeenusedto ;analyzeESPreliabilitydata.Theanalysessuggested

    ;threemainfactorsaffectingESPreliabilityintheMilne ;Pointfield,i.e.reservoiL

    ;ESPisthefirstinstalled

    ;sandcontrol,andwhetherthe

    ;inawellorisareplacement.

    ;Furthermore,theESPfailurerateisdynamically ;lnfluencedbythedeliveryofnewwells,andthetime ;lagbetweenthefailureoccursandthereplacementis ;made.Thisobservationhasbeenincorporatedintoa ;computer-basedfailuresimulatorthatusestheMonte ;CarlotechniquetoSimulatethefailurerates.Numerical ;valuesforthestatisticalvariablesareincludedwithin ;thedetailsofthesimulator(Pattersonl993;Upchurch. ;1990;Oliveira,eta1.,1997;Sawaryn,etaL,l999; ;GramesandSawaryn,2002).Successfulmanagementof ;apopulationofESPworkingconditionsandfailures ;requiresanaccurateestimateoftheexpectednumberof ;failuresasafunctionoftimeunderconsiderationof ;associateduncertaintiesinthewellenvironment.An ;effectiveriskassessment,whichisthetacticsdealing ;withoverworkandtheimpactofthesetacticsonthe ;productioneconomics,mustbemadefortheuncertainty. ;StawarynandZiegelf2001)providedthedetailsof ;modelselections.parameterestimatesandvariation ;withrespectofawell’seconomiccriteriaandrisk

    ;assessments.Theirworksuggestedthatthe

    ;uncertaintiesinherentinthecontinuousreplacementof ;ESPsandtheestimationoftheiroperating.1ifehelp ;maketheestimatesofalargenumberoffailuresacross ;anentirefield.Stawaryn(2000)presentedthe ;analyticalexpressionsofthegrowth.renewaland ;declineofasingleESPpopulationanditstransitionto ;dualESPsystems.DualESPsareconsideredtobea ;meansofimprovingoperating.1ife.Horizontalsurface ;insta11ationsforwaterinjectionhavebeenmade. ;Consequently,themachineoperatingconditionscanbe ;monitoredandtheequipmentperformanceanalyzed. ;However,Stawaryn’sworkindicatedthatindustryhas

    ;offentriedtoovercomethesymptomsratherthanthe ;causeoffailureproblems.Unbalancedpumpvibration ;hasbeenprovedtobethemaincauseofmanyseal ;sectionproblems,whichcallultimatelycausemotor ;failuresaswellaspumpfailures.Somevibration ;problemsarecausedbywellfluidconditions(Durham

;andWillliams.1987).

    ;Athoroughsearchforpapersandreports,which ;mightcontainthediscussionsofrecordingchartfeature ;recognitionandmethodsforestimatingtheexpected ;numberofESPfailures.foundthattherewasverylittle ;publisheddiscussionaboutthepredicationand ;identificationoftheESPfailuresusingrecordingcharts. ;3.Featureanalysis

    ;Traditionally,tomonitortheworkingconditionof ;anESP.theelectroniccurrentofthedrivingmotorof ;theESPismeasured.However,ifonlytheelectric ;currentofamotorismonitored,diagnosedand ;analyzed.itisverylikelythatanumberofaspects ;affectingthesystem’soperationmaybeomitted.To

    ;improvetheeffectivenessofthediagnosisandanalysis, ;anewrepresentationofthe

    ;systemsisrequired.This

    ;processingofvariablesto

    ;operatingconditionofESP

    ;canbedonebyrelevant

    ;obtainananalogywitha

    ;

    ;V.01.2No.3ResearchonPick-upFeatureofESPRecordingChartsBasedonBitmapImages7

    ;well-knowndynamicgraphchartofasuckerrodpump ;(Feng.1996).Tlledynamicgraphshowsmoreinformation ;thantheelectriccurrent.Followingthesameidea, ;thispaperreportsapick-upfeaturemethodtofully ;analyzetheelectriccurrentgraphonarecording ;chart.

    ;Fig.1illustratesthefeaturebaseofrecordingcharts. ;whichincludes16differentkindsofelectriccurrent ;curvesontherecordingchartsunderstandardworking ;conditions.AsshowninFigurel,A1:ordinary ;recordingchart;Bl:recordingchartwhenthepower ;volmgefluctuates;Cl:recordingchartwhenairslugsof ;pump;D1:recordingchartwhenthepumptakesoutnot ;somuchoil;A2:recordingchartwhenthepumpstarts ;abnormally;B2:recordingchartwhenthepumpruns ;continuallyandshutsperiodically;C2:recordingchart ;whenthepumprunsinagas-bearingwell;D2:

    ;recordingchartwhenthepumpstopsduetoalow ;current;A3:recordingchartwhenthepump’sprotector

    ;doesnotwork:B3:recordingchartwhenpump’sdelay

    ;timeisveryshort;C3:recordingchartwhenthepump

    ;stopsduetoanormaloverload;D3:recordingchart ;whenpumprunsinanimpuritywell;A4:recording ;chartwhenpumpstartscompulsivelybyhand;B4: ;recordingchartwhenthepumpfluctuatesduetoan ;overload;C4:recordingchartwhenthepeninthe ;recordercannotdrawapitcharc:D4:recordingchart ;whenthesafetypressureisnottested.

    ;ABCD

    ;I

    ;?

    ;!

    ;?

    ;,

    ;..

    ;:f.

    ;毫?’j:?|’.

    ;‘；’i

    ;.0

    ;2,

    ;3

    ;潦:

    ;4:

    ;托

    ;4.Establishingacoordinate

    ;Fig.1Featureoftherecordingchart

    ;Inthecourseofdiagnosinganelectricalsubmersible ;pumpbyusingrecordingcharts,thepick-upfeatureof ;electricalcurrentcurvesontherecordingchartdecides ;thecategoryofthefaultsandtheaccuracyofthe ;diagnosticresults.

    ;Discussingthepick.upmethodoffeaturesofa ;standardrecordingchartcandrawtheconclusionas ;follows:Underdifferentworkingconditions,thevalue ;andfluctuationofanelectricalcurrentaredifferent.A ;coordinateisestablishedfirstonarecordingchart.and ;thenanumberofpointsonacurrentcurvecanbe ;sampledaccordingtothepre.determinedorrequired ;precision.Thetotalvaluesofthecurrentsonarecording ;chartreflectadifferentworkingconditionofthedriving ;motorinanESP.Therefore,thesum(recordedasof ;thesquareofcurrentvaluesortheareaofaclosedloop ;currentcurvecanbeusedtoidentifydifferentgeometric ;shapesofcurrentcurvesorworkingconditionsofan ;ESP.TheMvaluecanbeusedforcurrentcurve

    ;patternrecognition,whichcanbecalculatedby ;Equation1(KoesnihadiandLea,2002).

    ;=

    ;?,,(f_1,…,,1)(1)

    ;Thecurrentcurveonarecordingchartillustratesa ;twenty-fourhoursworkingcurrentofthedrivingmotor ;ofanESP,whichisnormallyaclosed.1oopcircleas ;showninFig.2.Inthecoordinate.acurrentvaluecan ;bereadeasilyalongthecircle.Everypointonit ;correspondstoacurrentvalue,whichsrecordedasIj ;f==1.2.…n.

    ;

    ;PetroleumScience2oo5

    ;5.Pick-upfeaturesoftherecordingchart

    ;Thescannedrecordingchartcanbekeptinabitmap ;imageformatandstoredinacomputer~Thenthe ;computersystemcanprocessthesebitmapimagesand ;automaticallypickupfeaturesfromthecurrentcurves ;onthestoredbitmapimages.T0beprocessedbythe ;computer,acontinuouscurveonastoredbitmapimage ;mustbeconvertedintoasetofdiscretedata.The ;digitalizedprocessingofabitmapimageiscarriedout ;byreadingthedifferentcolorsofpixelsoftheimageon ;acomputerscreen.Differentcolorsofthepixelhave ;differentRGB(red.greenandblue)values.Asshownin ;Fig.2.thecurrentcurveisdistinguishedbycolorfrom ;itsbackgroundonabitmap

    ;selectingproperRGBvalue,a

    ;image.Therefore,by

    ;currentcurvecanbe

    ;automaticallyreadbyacomputersystemandrecorded ;asasetofdiscretedatainthecomputersystem. ;ThecomputerProgramforautomaticallV

    ;pickingupfeaturesfromarecordingchartisdeveloped ;inVisualC++.Theindexvalueofpixelcolorsranges ;from0to255fora256.colorbitmap.Forexample,the ;redcolorhasanindexvalueof249.AsshowninFig.2. ;thecurrentcurveisplottedbyredcolor.Therefore,the ;computersystemcanpickupthiscurrentcurveby ;settingthecolorindexvalueas249fromitsbackground, ;whichconsistsofpixelswithothercolors.Toerasethe ;background,thecomputer

    ;indexvalueofotherpixels

    ;color.

;:’

    ;x

    ;1

    ;systemcansimplysetthe

    ;as255,whichmeanswhite

    ;Fig.2Establishingacoordinateonarecordingchart ;Thevalueofarecordingchartcannotbedirectly ;read.becausethevalueisseria1.Sotheserialvalue ;mustbesampledintodiscretedata.Inordertoensure ;therequiredprecision,thecomputersystemreads360 ;pointsfromthecircleasshowninFig.2.Itstarttoread ;pointsrandomlyfromthecircle.Thesumofthesquare ;ofcurrentvaluesfromthesepointsiscalculated ;accordingtoEquation2,whichisanMvalueto ;representaworkingconditionpatternofanESE ;360

    ;M:

    ;二.一(2)

    ;Fig.3showsabitmapimageofarecordingchart ;beforethecomputersystemprocessesitandFig.4 ;showsthebitmapimageaftertheprocessing.Thepix. ;mapcoordinatecorrespondingcurrentcurveisrecorded ;inthecomputersystem.Accordingtoequation2.the ;chartfeatureMisdeterminedasfollowsintermsof ;thefeaturebaseasshowninFig.1:Al:=12960; ;Bl:M2=13039;Cl:M3=8740;Dl:M4--65l3:A2:Ms=3268; ;B2:M6=1750;C2:M7=10890;D2:Ms=1429;A3:M9=6435; ;B3:M10=11448;C3:M11=3187;D3:M12=5703;A4:Mf2232; ;B4:M14=3l35;CA:Mls=3359;I34:M16~5412.

    ;Fig.3Recordingchartbeforethecomputersystemprocessing ;p

    ;Fig.4Recordingchartaftertheprocessing

    ;FromtheMvaluesaslistedabove,itcanbeseenthat ;someMvaluesareveryclose.Tofurthertakethestatistic ;errorsintoconsideration.itwillbedifficulttodistinguish ;someworkingconditionssimplybvvalues.Inour ;work,astandardstatisticerrorissetas4XIo2=-144and,n ;isthecurrentvalueinanordinaryworkingstatus.which ;normallyhasavalueof6inthepolarcoordinate.In ;termsofthediscussionsmadeabove,theworking ;conditionsareclassifiedasfollowsintermsofMvalues: ;A1orB1.whenM1M2>12204

    ;

    ;V.01.2No.3ResearchonPick—upFeatureofESPRecordingCha~sBasedonBitmapImages9

;B3,when12204>n>11169

    ;C2.whenlll69>M7>98l5

    ;C1.when9815M3>7642

    ;D1orA3.when7642>M4,M9>6069

    ;D3,when6069>MI2>5558

    ;D4,when5558>MI6>4520

    ;A2.when452M5>3494

    ;C4,when3494>MI>3273

    ;C3orB4,when3273>MI1,4>2684

    ;B2orD2orA4,when2684_

    ;>M63

    ;Itisobviousfromtheworkingconditionsasclassified ;abovethatsomeoftheworkingconditions.suchasAland ;B1.orC3andB4.needtobefurtherdistinguishedby ;usingotherpatternrecognitionmethods,suchasthe ;supplementarymethodasproposedinSection7. ;Accordingtotheimplementationresultsinseveraloil ;fields.theworkingconditionidentificationmethodby ;usingMvaluesaspresentedinthispaperhasarelatively ;hiaccuracy,whichisupto97%.Inpractice,mostM ;values,whicharecalculatedfromEquation2,arelocated ;inthemiddledomainofthevaluesaslistedabove. ;6.AboutIB

    ;InapOlarcoordinate,/0equals6fo?