+ All Categories
Home > Documents > JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... ·...

JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... ·...

Date post: 26-Sep-2020
Category:
Upload: others
View: 0 times
Download: 0 times
Share this document with a friend
25
.. . . - ...- .. I ... .. , .. .. .. . . . .. . .. . . . - .. . .. . AutomatedSoftumre Analysis ~ ofNuclearCoreDischargeData T W. L JtmesK.tfalbig johnkhuell ~m• George W. Eccle#on %“rkyF.Klbstedwer ..... . .. . . Lodilamos MATIO N A L 1 A-8ORA Tii R—V L A N M 8 L D 0 .. M n * OIH’R18UTIONOFTHISC)C)CWLENT ISUNLIMiTa ... .... —. ———————————————
Transcript
Page 1: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. . . . - ...- ..I. . . . .

,

. .

. .

,

. . .. .

.. .

.. .

. .

- .. . . . .

AutomatedSoftumreAnalysis ~of NuclearCoreDischargeData

T W.LJtmesK.tfalbigjohnkhuell “ ~m•GeorgeW.Eccle#on%“rkyF.Klbstedwer

.....

.

..

.

.

LodilamosMAT I O NAL 1A-8O RATiiR—V

L A N M 8

LD 0

..

M n*

OIH’R18UTIONOF THISC)C)CWLENTISUNLIMiTa—... .... —.——— ————————————

ABOUT THIS REPORT
This official electronic version was created by scanning the best available paper or microfiche copy of the original report at a 300 dpi resolution. Original color illustrations appear as black and white images. For additional information or comments, contact: Library Without Walls Project Los Alamos National Laboratory Research Library Los Alamos, NM 87544 Phone: (505)667-4448 E-mail: [email protected]
Page 2: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

..

.

,

. .

. .:,

. .

.

. .

. ..

. .

. .

. .,.. . . ~. .. . .. . - .. .... .. .. . ... . . ...... .- ... —.——.— —

. .,

. .:

. . ------ . - . . . . . . ........ ... - .

C “..

ABSTRACT●****...*, W * * . ””**e***w**” ““.,.... w**,,..,*.,..*................. 1

“I N T............w........,.......~....”..,.,,....”....”.,.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .,..00” 1

A US OA N●.., *.. w”.* ”..... *.. m....w..”*......,*.......”.. 00..” sm~~mc)hi OFAREAS(w NIERESTA F DISCHAK3E

“**,e.,*.w...e..e.”..*.ww””””“**nw*.””.**ebe***.....,.W**WWW”.*W**WWW”UW**W**OOW6

CORRELATIONOFEVEN’1$ “ “Wwwww..””**mw*.bo...W..***,W.*W”*.W**W*.W””W**W*.M.*”**7

MOMTORINGREAC’17)RFOWERLEVEL.W..-....”..wbb”w”“..”””..”..”00”...-7

STATISTICALPHENOMENACWCDMDATA●W*.W.*......W*”*.W..”..”..W**W..””..”8

NEm~mtim FoRs T F m. . . “ ”” ............................“ ” W*” ●* 14

NEURALNEXWORKSFORFUELBURNUPCOMPUTATION................18

RESULTSWNEuRALNETWORKMCDELSFORS(lVIIW. .GEOMETRYANDBURNUP●w”....”....wow””””””We.w..e,”..””..”.,”””..”..”..21

. . CONCIJJSIONS . . ...."w.."""".."w" www.. M . w "*e".***,* 22

.

,

. .

,.. “

., .

. .. .. .,

.. ..

..

,

. .

. .... . -J . .

. “.

. . .

v

Page 3: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

AUTOMATEDSOFIWAREANALYSISOFNUCLEARCOREDISCHARGEDATA

Td W.LarsowJamesK.HalM&JoAnnHowcuGeorgeW.EcckstcwandShirleyF.Kbterbm

.“ :

ABSTRACT

MonitdngtheMing pmccsso anon-loadnuclearmactaris afull-timejobfmnuckarsaf

-Fak==mmi@rs(CDM$)can~tor’s fiding d tyfw later, e dew bya safcguds-~. A@mNhivCdyfi thiscdkctcddatadptQbe a greatasset to inspectorsbecausemoreinformationcan beCxtrmedtklmthedataandtb analysistimecanbereducedamsider-

?Jg:VgG~F~&m&~A-~. Ncurdnctwakndcls

weredcvckpcdfm @cuhtingtheIv on on thercx%orp=hJ’.ti’3ti&*hlntmuP.e=

usingactualdatacollectedfkoma CDMW-at m ~~ Iemor kilityo Coktivdy, theseautamatedq~titative analysisprograms

!!%la.-&”&?.*:LzTs-3:c#@ve solutionfm autamat~‘&nitdng of odoad reactom

gdkamly Ieducingtimead Cffat9. .

. . . . .“ . .

INTRODUCTION ..

Nuckarpowerstatiomintheu s C I q w c ( b af a c usuallythetq, fuelcanonlybeacmssedwhatthcIemoris shutdown.b safe-guardsadvan~ tothistypeofIclKtaris thatitis dativelyeasyfmanuclearsafeguardingagmcytomonitorthefuelingp!Uccs&An“mpcctork a st@uding agencycanbesenttothesiteto~ thefuelingpmccdlmOn-badnuclearIemlx’sarcdifferentfim u s l it o pmayobtainaccessto theComfrombothCn@andtheycanbeCatthouslyfidulwithoutshuttingtimtlCbwms a ● o fx _ i p !a~

i challengefhnna safeguardsfmthedivcmiond nuclearmatcddCh40ad

Iemorsmeudl-suitedfocjmdodngplutoniumfiotntheirstadad fuelbtuxik Sabgwd&gamon-badIewmr- MP@ - affid asit ispushedtbugh * lwwtocwhata fieshfitdInlndkis pushedinme SkkOftb feactw,a spcntfid htndkisshbmudy ~sacdktion mcchdsm on thealter$i&eU*g thisfbdittgscheme,● typicalon-loadmmorwill

1

Page 4: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

.

discharge55to65fuelbundkperweekl%gurc1showsaamccptualdiagramofthisfuelingcyck.Becausethisis anongoingpraess, it is notsimplefm a safeguardingagencyto continuouslymonitorfuding;itisratherlak L%cnsivctohaveasafeguardsinspmtoron-siteallofthetime.

‘herearcseveralapproachestotheprobkmof monitoringtheconstantfuelingprocessof on-kadreactors.Obviously,sometypeof amcasummntsystemfmprovidingoontimmus,unattdedmonitoringof fuelingis themostattractivealternativebecauseit is thekastMor-intensive.Coredischargemonitors(CDMS)lprovideaconvenientsystemfmsafeguardingon-loadrcacmrsbecausetheycanbeinstalledasanindepcnckn~tamper-msistantpackage.ACDMusesradiationdetectorstomonitorb movementof fid betweenthereactorcoreandthefuelstoragearea.Coredischargemonitoringcanbeperformedbyanekctronicspackagedkd GRAND(m myandneutrondetector)andassdatd detectorsdevelopedattheLosAlamosNationalLaboratory(LosAlamos).A typicalnxasurement statimconsistsof aGRAND,whichis thedata~quisitionekctronkqandfm demctorsmountedinoneenclosure.BwhGRANDcontinuouslyColkctsdatafmmitsassoci-ateddetectoratdiscmtctimeintervalsandtransmitsthedatatoanMS-DOScompatibkmmputcrfa moding. ‘Ilwfm detectorsineachcnclostueincludetwoncutmdetectors(f~sionc~)andtwogamma-raydetectors(ionchambers),oneshieldedandoneunshielded.Becauseof kw-enriohcdf a lowexposureof thespentf datively fewneutronsarcemittedbythespentfud ‘mencutmn&@ctorsarcencasedina containerof heavywater.IntIMidealsitua~ theneutrondetmorsaresensitivetoneutronsmatedby(’y@)mactknsinb deutaiumsumudqg“ thedetecmrs.To producea tmmm in the(y,n)rewtionrequiresa gamma-rayenergythresholdof22 McV.Tormnitorcare~ wemountedf- GRANDsaroundthenuckarcorGtwooneachremorface.Rx Im==of ~ Fig.1showsan cxampkofhowa typkalon-kmd

. .

. .

2

Page 5: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

I “ . .

rcactormightbelaidoutwiththedetectorsandGRANDsinstwvd.h F16a~a),thecomis nxntntcdinMCbuildingsucht t a facesarcI’”1tk castandwxtsidt Mth:rearer.~ fuelingtakesplacefromcasttowest.orwesttoeast.Becausethecoreis nunmte“ .hisfashion,each~-mentstationisdesignatedbyitslocauoninrelationshiptotk **em,eith~‘hex@teast(SE~north-=wt(NE),southwest(SW),ornarthwcst(NW)corner.

EachGRANDcolkctsntdcturadidondtxafiotntnedetu. Tenck ~, .dtcfiP.timestampsi$ andttmpmily storesi T d aa t hfa mtheco%ctionXnpu“ruponIeqtx%tformampumancntstorage.Aalatertime,data@oheofHoadcdfromtheccWctmncomputerfa off-linerwkw.Thedetectmdatafedfhm tk G- consistof fiveC d I ‘ T C

n a h a fohvx f- C dxrA,fissioudmnt)cr~,fissionC c i chmthtv~,andionchamber2.FksionC hALmq)ondstothefirstnmtrondctecto.: .nedw ‘*.wedo-SuleoFissionchamberBis anotherviewofthcfirstneutrondetcctw,whichcanbeusd fa possi~tamperdetectk Thesecondwutrondetectorineachdctecmrenclosureis labClfdoI*tksionoham-bcrc T n c& ti st i ~ m &‘ t t ‘ 0t o pf=. I%rcxampktheNEfissionchamberCis wiredintothelb.‘YGRAND,andtheNWfissionC hC is wiredintotk NECiRANDoThisprovidestk O tll ~@elllwithabackup,incasetheGRW fa we of thedetectorsfail~‘lldscrosswiringis showninFig.2{a)astk spliceboxbetweenthem GRANDSoneachsideof thereactorm. Fhlally,thetwog&rlma-raydetectmmmspondtotheionchamber1and2 Channckrespectively.Figure2(b)shokst%layoutof a detectorenclusim. Anin-depthdiscussionof tk detectorassemblimandtheGRANDekctmnl“CSpackagecanbef- in“TheDesignandInstallationof a CwcDischargeN&mitocfmCANDu-typeReac#Xs,”byLK Halbig,etal?

ThedatacolkctioncomputerssampktheGRANDsata pm-deter.inedinterval,wally wk~10or11seconds.‘Ihctotalnumberofdatapoin,~coilectedbyme detectorM oneGRANDis lim-idto7855 pointsperday.W all20detq thisworksoutto ●s about157100d p p

d a0 713~ ~month. Tostoreallthedatap f~one moi+ f a 1 816000bytcsdstoragc.Itis impmctidtoanalpethisamountofdata.BecatMof~ ~ ~age ~~:$ “

~* G~ U@OY ad-=we* tC&iWC tOdUCCti SUIKWda ~- ~lOW ~v@. ~ ~RA~ -S ~Y aIWC=~VC 2%of he homhg datatotk c tc o m p uThisreducesthedataW*considerably.Becausetk dataaretimes~”.mi>:d,?* -gap donotpresenta probleminansdysisordew of tk data.ShowninFig.3 areg of&t@fiUSDt ( bduring- -k dily.

A qualitativeanalysisof thedatabya safeguardsinspectorcanyielda considerableamottntofinformationonthefuelingactivityatthereactor.Eachlargespikeontk graphcomqmds toapawof fuelbundlesbeingdischargedfromthereactor.Smallerspikesordecaycurvesorbothnnthe@’a@!lltlycorrespondtootheractivitiessuchasthcrotationofthefuelingmachine,ortheradioac-tivedecay,calledcooling,ofthespentfbelfissionproductsbeingheldinthefbcling~hine duringa shufflingoperation.Thereactorpowerlevelcanalsobedctcmiwi fibmthe&la becausetheamountOfbackgroundthedetemrsamsensingcomspds tothecurrentpowerlevelofthereactor.Tk backgroundinthiscontextisconsideredtobetheamountof rdiationthercactorcdtsduringnormaloperationwhennofuelis presentoutsideof thecore.Currently,a s h

m qualitativejudgmentsaboutreactoractivityby visuallyeXaminingthegraphsof detectoractivityonbothftwcsOfthc~. we- aninspectmcancountb numberof spikesonthegraphand&ermineth8tmalnumbemffuelpushesthereactormadeinaparticularday.Thetotalnumberof fitclpashcscountedcouldthehbe compamdto f&Mtydeclarationsf= safeguards

ly, reviewingall the informationcollectedby all the detectorsis avdkation. Unf~

1-3

I

II

Page 6: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. . .. .. . . . .. . . . . .

.

. .4

i

i}IiI

.1i1L

I.-—-.. .- -—1

Fig.2(8). stm@e@mdyalJ#Cd9w@%drdmwr0“

IkkdOr Enckxmm

II ——

— IonChambe$l—Ia91cbmnbe$2—* CbmlkA/B—

Fig.2(b).A@cd~encAmw

4

Page 7: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. .

. .

. .. .

. .

.

. .

mau SRFfsi&nChhmbarA

n

,

F

,. . ..

“ s e l o

. .

. . . ..

022#9”

acMTOn2

Fig.3.Samp&CDM&a@mana4badreacW.

tbcmvkwt andmakeitcask fw asafe investigatortowa&throughtbcvuiumcsofdataavailaldcfillmanopcmhal on-load-tar. ThispapercxambcsthcfcasiM@of&ek@n!gaD

=W= --a~ datatohelpasaf6gumb

IAUTOMATEDSOFTWAREANALYSIS

Analysiso Ming datafklmanOn40adfcwtol i acompkxtk Asa~ wcdmignedtbeammatcdanalysissystemtoinvestigatethefeadbilityof8cvcralo&cdve%d -W -diffidtoimpkmcnLTbcscobjectivesamasf-:

; , 1.

,. ‘“z., “3*......“

Icknm areasof intcmstin W datafa a safeguds invcdgamrto cxamineingmW!#

5

Page 8: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

4. COmlatcaneventf= one&tcctorchannelwithallother(ktcctmchannels.ThiscouldbeusedtodetectpossibletamperingandtoCnsumthatauthechannelsamoperatingCm’ectlyo

s Idcmi&thefuelingchannel-which thespentfuelwasdischarged.

Qti~uPof* spentfudbutuues.

Ed o theseO@jcctksCxtrwtsf af t e MO fd T f ct h a safbgudinginspectorassessallthedatacolkctcdfw a prtkuhr on-badreactor.Tofdlitatc theseobjectives,apmotypcanalysistoolwasdevelopedcalledCDManalysis.The(DManalysispackageaccomplishes*tivcs otwthroughf- reasonablywell.Wedevelopedthistooltoexplorethepossibilityof developinga pmdmion-gradeanalysistool.objectivesfiveandsixwereattemptedusinga neuralnetworkmodelingparadigmdescribedlater.TodevelopandtestCDManalysisascompletelyaspossiblew neededaconsiderableamountof data.Unfatwwtdy,onlyabout30daysofdatawasavaiJabkfmanalysis.WeconcludedthatalthoughthetotalamountOfdatausedfa testingtheanalysissoftwarewas~ thesoftwarestillperformedveryWcIL

IDENTIFICATIONOFAREASOFUWERESTANDFUELDISCHARGEEVENTS “

A s i g n ipmhktnwithanautomatedanaIysisof CDMdatais idcntifjdngareasinthedatathata safcguantsinspectorwouldb intcrcstodincxamining.Identifyingareasof highactivitycan.condcmNyreducetheamountof- the_ l t p t d Atsuchan~y sta@

dcveAoptncntit is impatantthatinspcmm~ti7. . do notuscthissystemas a substitutefaamtnatmof allthedaa butmthcrasanaidinthereview_ whenthcmactotisrun-

ningataconstantpowerkve~a baselinecanbecsmNisMasanawrageof thebackgroundnoisecollectedbythe&tccmrs.~~t ~~ -- ~ ti~ = b c~w ~ ~ * oftheaveragesig@ aboveorbelowthebaschtmThisis ale techniquew usedtoidentifyareasof●

(!DManalysismakestwopassesovertheCDMdataduringitsseamhfmareasofintemstInthefirst- itslidesana~alongthc signalbokingfwsignifkantchanges.whenthesbpcoftb osignaljumpsabovew belowtheslidingaveragebymomthan10%,thedatapointsamflaggedasscmmhingtolookintolater.Inthefirst~ a hugeamountOfdatapointsmaybeflaggedasintcr-cstin~Toreducetheclutter,asecondpassis ma&ovcrjttsttheareasthatw flaggedAreasneareachOthcrinthetimesales aleClusteredtoguhcrwith* maximumdatapointbeingmarkedasthemiddleof theevent.Fmmtheresultinglistof areasof_ ●rcpxtcanbegcncmtdtoalertthesafeguardsinvestigatortospccifkareasofthcdataDuringthis&vclopmuwCDManalysiswasnotexpandedtoexplaintotheinvestigatorwhatis actuallyOccuuingintheUnda’IyingsystemItcur-rwulytells* inv “Cmgatortocheckoutcertainareasfixauivity.Inthef-a dctaikdanalysisofanwtualreactorinOpcmtkmalongwiththeColbctddatacouklbeusedtodevelopa systemthatcouldgcnwatcsuchareport.Finding- therefuelingspikesis aseasyasfindingallothercvca*CDManalysisjustneedsto bemomddddng, If tbetltmhoklill9ctveryhiglhatW%*two-passtechniquewillyieldaso ua t f wS ph t g d a ‘ f

e v as p iw ib m a rb t C c no h p o b C vc d nM a

- * - h h i m pt r e at t a di r aa ub i

a s a n n e s t hw a e rd u r i n

6

Page 9: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. -

. .

CDManalysiscangenerategraphsof collecteddata.Figure4 is m exampkof anoutputgraphfromCDManalysis.Twochannelsare_ whichamtypicaiof alleventsoccmringonbothfacesof the~. Theverticallinesrepresentmarkedfueldischargeeventsorpowerlevelexcur-Sionsoftheremarorbath.The&l dischargeeventsandthepowerkvelexcursionslu’cailtnarkedindifferentcolorsonthecomputersawn tomakeiteasytodifferentiatebetweenthem.Thethmh-oldvaluetncntiomdearlieris markedbyahorizontalline.inthiscasethetbrcsholdis hightolocateonlyf &chargeandpowerl e vCORRELATIONOFEVENTS

Onceallthef dischargeeventsami&ntifiedandclustered,eacheventis thenlocatedonthedatafromalltheotherchannels.Toaccomplishthiscorrelation,wchadtoovcmotnetheproblemthattheclockson theGRANDsarenotsynchronizeCorrelationis difficultbecauseaneventmarkedwithone timestampon one GRANDmaynotbe foundat thesametimeon anothaGRAND.FindingeventsonGRANDsthatarconthesamereactorf= is nottoodiffiicukCDManalysissimplyfd thespikeheightmaximumIwamsttothetimeatwhichtheeventwasrecoded.Theproblemof findingeventsthatoccurredontheoppositef= inoneparticularGRAND’s~canberesolvedbyusingthecrosswiringof theCftin chambers.CDManalysisfd be eventon fissionchamberC andusesthespiketo performa pseudo-synchronizationof theGRAND’sclock7MsallowsCDManalysistofindwherethepeakshouldbeona&tectorviewinganeventon~ -g *C*

MONITORINGREACTORPOWERLEVEL .

o thea o famkkmifi~ powerleveludtoring israthersimple.whenm Cvattsamoccwrin~backgnntndmdiatknissensedbydwktwtm Theaveaagcofthcbdcgmnd canbe

. .

. .

.

..

. .

. .“

.. . . . . to. . . ., . .,

. , , . . . , , ,,, ●. . . , ,,, 0 ,. # o , ,., , 0. . . 0 00. . ., . ,,

, , 0 ,0 0 ,0

i : !: .0 . ,.

*

. , . B n i!! ! !:mwtlsd. 0 M. . ;; : ~;, ~, , .. - . . . . . . . . . . .. . . . . . . . . J--. --. 0. * ● .

. .. ,,

. ., 00 0

.

,,

“ , . : ●0 ,0 0

,,

,*

,0 ,, .

! SE ; “ .. ,0 ,

.0 ,0 0

M ;,. ., .,, ,. 0

.0 0. 0

.0

., . . . .

,. ., . .

●0 ,, 0

,, ,, ●

9* ● *. . . . . . . . . . . . . . . . ,.. - .“. . . . . . . . . . . . . : . . . . . . . . . . . . . . . . . . .

. . . . . . . . . 2!.... ....... ,*“

F h

Ri&4.&w@bOku#jhM8tkcmtadyds$rognau,

— —

. .

7

Page 10: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. . .— .I. .. . .

. .,

usedtoc a m pthepowerkvcl byetabbhbg a basdinereadingof whatis comidmdm. m baselineis Clompmdby

U)bfilllcxadningdata&mareactorthatisopcmtingatafixedpower

w i t h*1 outsidethecore.Theavengevaluerecordedbyeachdetectoris usedasthebaseline.Thisbasdineis markedon thegraphin Fig.4 bya Iutrizontalline.If theaveragevalueof thebackgroundmovesfromthisbaseline,thenthepowerkvel is changing.‘Ilwdatahassknvnthatm powerchangesoccurredinastcpwisefashion.CDManalysiscvaluaeesthepowerexcursknsinthefollowingmanner.Ifthelvactor_ ~ ~ ~ = ti ~ Ofthcaverage~~startstobecomeverysteep.Thisis markedasthebginningof apowercxcumimwhcnthisslqmflattensoutagain,theendof thepowerexcursionis mwkcdTIMnewvalueatwhkhtheaveragebackgroundC Ot i “c ot n p l o r T a b

a a pcmcntageof thepwldincd baselineis thepaccntageof fullpoweratwhichthereactorismnning.(hmmly, C!DManalysis&es notexaminem than~ channelonme &tcctorwhenmakingitspowakvc!computations.Inaproduction-kvelanalysispackage,thispcmcattagcshouldbeanavcxagcof allthepemcntagcscomputdfiwmallchannelsonalldetectors.Bytakingpmwrlevelmcaswmcntsb all sidesof thereactorcomandaveragingtlwm,a momaccumtcpowerlevelreadingcouldbeobtained.Eventhoughcxammin“ g justonechannelgivesa ftiy accumtcreading,within5%,examiningall channelsis a muchbetterstrategybecauseit provides●redundancycheckFigure5 is anexampkof thepowerlevelof a reactorbeingraisedfiunnmicX9-powcrto ~~. NoticethattheC Xo i m s CDManalysisis alsocapableofprintinga reportthat&tailseachstepof thepowerlevelchangeandwhatpowerkvel therewtormovedto.AnexampkofaxepottforthedatagraphedinFig.Sis showninFig.6

I . STATISTICALPHENOMENAOFCDMDATA “: . . .

Tobcttcrundemadthcmactorf@mll=-~~ yanalyzedtheavailabkdamTheh e iOfthcradkionspikesr qM dischargeeventsc beComlatdtothefuelburnupandthe&cationOfthcf c hf w thefuelisbeingdiscbgd ThebumupContributiontothespikeheightis a mult of fissionproductbddupinthespentbl bundks.Ifoncknowstkspikeheightsfromall20Channdsfmapdcular evcn~oneshouldbeti tocomputethcbcatknandpossiblytheburnup.Dcmnmmn“ “ g burnupis a diffkul~multivariatcpmbkmthatwillbedia-cusscdingfcatcrdetaillater.Doall20channckmakcavalidcodbuth tocomputingthelocatknof theM bundks?Themo6t“~-~ whcthcrarnotdetectmonom * Oftheluac-torseeeventsO ct -- ~ $ Cnoughsothatthcemarinm ismalLF@e 7 demmmmsthatdetcc$mononekc&M smeventsonthe!oppositefs whataneventis occumingm onefwxof thereactor,tbtspkeappmingonthcdetectmOnthcoppmingfaceis insignificant.Thismeansthatoneshouldmat each*of the~ separatelyWh- m ptodeterminethekcationo fuelbundlesbeingdischarged.A comlationdoesexktbttwccnthetuospikeheightsonOppodng&tectmonthcsamcrcactorfaceRxa givenbumpof* fid bundksbeingdisc- andas thesome is closerto one&tcctorandfarthcrtianothcf,thesignalwillbelaqp onthenemer&tcctOrandsmalkmnthcfartltcrdewctoc.ThcgraphOfthcsignalstrengthfkxnmmoppositlgncwon&tectomm * same* is showninF@8.

The@atiodu“pis coneshapedbecausethecrrocindatafium●channel- greateras●

functkmof the “~of~~~~~~.~~ e~m 8isnotlargeenoughtoprohibitdwckpinga mxkl fw bating fuelbttndksonthemactmfu%EdlpointonthcpkcinFig.8mneqmdsmapositknon thcmacmhccasa fimcthtoftwoOppOdng

Page 11: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

... .. .. . . .

...

. .

. .. .

. .

. .

. .

. .

.. .. .. .

.. .: .

.. .. .

. .

. .

SSndon ChanbuA r... .

. .

00 . . ~“2 “

. .. .

. . . .. . . .

PbwctE%curdm~

T- PercultafFullPower

90.11.02 134XM)5 14:1225 L 9.89 014:12:48 14:%15 9.8 21.3!N).11.02 15m28 15:31:47 213 33.1$M).11.02 15:32:W 15:S2:18 33.1 ~ 50.9m.11.02 M4M:14 1656S4 . 5(k9 - iO&O

. -. . . . .

. .

,

. .

,

. .,

. . .

Page 12: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

.

. :.

. .

.

10

Page 13: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

dctcmrs.TheneutronChanndstheeast

-~~~----9-*-*~seactorfalcwxcqmnmthc-ofthc UmhkkkdionChatnbachannd

Unfi)mnatdy,theVadanceintheknchanlbcrdataillvcfypmmmced‘nDcrcam●cUMkbbkmlmbcrOfoutliminthedatawtbwaaseOfthcapmtknOfthcdata~-~-ChambcrdataamintcgmtedOvathcentheSan@ingtim$whereastheionChambcritl8pdntSaInpkOvwabO@SoxnsBccawHtb kmchamberdataamnacintegmedWcrtheCntkdataacqdaitkmk* =@@ dataamnatasqmentab ofti ~ MingewntIntegmknCatssesan- *-to occamovatheattiresampktimeyieldingmuchmommpescntativedate.‘mlcalmslessVaIimcccooccurinthenuluundetectanwmmmcntthantb ionchambermcaslm-mcntWeattmptd tousethema ofthccntimMing eventtominimizedance duringtheanal-* butit didnotyieldbettermsalt&Thisoccurredbecausethehe) spikesdo nothavea well-dcllncdshapeonCithathencuaonortlwgamma-myChaMcl&As●mlallbitwasdMkolttodeciu-mincb timeintervalto integratem. q themam&caycurvesf*g Ming -whichcorrupttheintegralvalue.A fbd intend analysiswastried,butitdidnotmdoceh vati-amxin thedataenoughto [email protected] to a highersamplingratewouldmate dM datasetfmtmalysislikethisinthefbtum.

Qometricallyspeaking,* dctmonlfromwhichthesedatawewtakenamsetup on eachreactorfacewiththeSouth-llidc&tcccasbeinghigherthanthenorth.If* dkmluxs%cmte$chfdC kt mh of thetwocktccmnarccompatc4w canddvc an equationfi(mlphysks** - Cm@atcsthcl’atimofthwcdistanmtothcrat&60f&tcctaSpibhcigh&Figaro10showshowthedmctcmm setupandgivessampledabks fmderivinga filnukd fGrmof tbWtancesasafimctkmoftheqikehdgk

Ancquatkncanbe&rived**spike heightacsignalstrength(SglwS@ m●fimctknafthesoulVetam s~and* dstancetotheM channc~RI aR2 Usinga~ ~ fimctkdf6cmfa eachdctccmr,forWhkheachM bundlepairis a.pointmtnce,the - -as-“

“ ’ 2 = ’

. .

11

. . .. . .

Page 14: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

... . . .. .. ..

. .

. .

.

,

.. ‘

.

. . . . .

...I

.

m

6

4

2

00

8

. . . . .

. . . . . .

● m

1 1

. ..

S. . -

. . . .

, I?ig.lo.

. ,.

. ... . .

. .

Page 15: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

(

Anodncar OpthnidonW pa’fbnnedtoobtaintbeVaht86k.detmdnd tobethef-~

q~ =33.82

. .. .

ql = CiOl “,

. .. .. .

.4%#3.~9 “

. . . .alz = MO

I. .

eqaut&m’rhesevalue8wa8e

. .

. .

~~ mnotia tbatthel/Rcmponcmmakes●minimalCOnmibUtimtotheqtathmafterthbfit.ThegraphinF@ 11SboUmtbcld@OnMpbetweenthefitmioaandtbcacatal$gnalVahmf= fision ChadMrA.Ifthefitmm * OIEUNUMUcpectthep&ttobeh. ultfdU-nately,themk Vdanccia thcdatafimthiscxact, fitdmal fm’1’bbtmalls thattbcfmmxbalfmcannot beuscclfm&$Udaingexactbcatiomal tbemactarfaccof fuel&chargeevent&sane typeOfemmdmnt motblb llcukdtonlakcthi6&tedMtb MUmlm M f- mchapplkdma The~

lmwodcmodebtmSblmlinF@ 11canbeCoqemtd farby

- a~ve m~~ * - ~ totbcl/R2datksb@ tomakeamdkematkallmdddtbe & t a a c e - t o -6 yI t m h f I t a f

,:

,

,

. .

. .: .

,!

:,. .

. .. .

- O ) a n h rMBW@MIMO. . .. . . .

. .. . .

●✎ ✎ .. .

. . :’ .. . 98

●0 “

. . .

● mo. . . . . .. #.

● - ● .,. . ,m-9.Le!&’:8° ..“.

. .

0 1 2 3. “ 4 6 . 6’ 7A 8~ ,

. . .. .

. . “, Fig.41.AcauisfgaallnAtacum$mdOl?jk

. .

. .

1

Page 16: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

NEURALNETWORKS10R SOLVINGTHEFUELGEOMETRYPROBLEM

Ncud nctwofkrmddsamtnathanatd● lyderivedfmnenfeccll biology.TheyamanalMrtw-tionof a collectionof fidamentalunitsorganizedina hiemrchkalfashion.Inbiologicalsystemseachunitwouldrepresenta ncwecell.Learning,andthusreactingtostimuli,isawomjWd inaM&k 60Xsclmw. The blackboxis moldedwhenpatternsof stimuliandtheircorrespondingmspmsesarepresentedina repetitivefashion.Giventheinputandtheexpectedoutput,theMaclcboxkrns themapping-input tooutput.Thelearninginsidetheblackboxtakesplacebyupdat-ingaseriesof weightedmdmmtwd● fimctionsaftcrcachpattcmpresentation.Thegoalis tomini-mizetheglobalerrorovertheentiretrainingset.Aftertraining,themodelis presentedwithnewstimuli.Ifthemtxklhasbeentrainedtoaminimalamountof error,itshouldgeneralizeandpredicttheappmprktcoutputfa anarbitraryinputpattern.Furtherdetails~gardingtheinternal,technicalWOfti neuralnetworkmodelingparadigmmaybef- inParailelDistributedProcessing,byl)widE.Rumdhart andJamesL McCklland.sTheneuralnctwotksusedinthisproof-of-prin-cipkwerecreatedusingNeuralWorksProfessionalWPlus!acommercialneuralnetworkdevelop-menttoolmanufactmdbyNeuralWare,Inc.A SunSPARCstationcomputerwasusedtoperformtheanalysisanddeveloptheneuralnetwork

Neuralnetworksseemwell-suitedfm automationof continuousprocessesbecauseof theircapabilitytogeneralizegivenonlyarepresentativesampleofdata Forexampk,NipponSteelhasutilizedneuralnetwds toachieve_ reliabilityintheircontinmtscastingprocess(Hidetaka1991).SNeuralnetworksarcalsomovingintothenuclearpowerarenabecauseof theirpdctivecapability.RohandCheonJwe developedseveralneuralnetworkmodelsto attempttopedictthermalloadmquimmcntsof a nuclearpowerstationandhaveachievedverypromisingm~~7BartkttandUhrighavealsoappliedneuralnetworkstonuckarpowerbycreatinga mmkltoauto-matestatus~08 w -neural nerwork applkations described inthispaperarecon-sickrablydifferentfrompreviousapplicationstonuckarpowerinthatweareperforming nuckarsafeguardsrathcrthanopdmhingthepowergenemdmcapbilitiesofthe~.

The30daysof availabledataykkkd around170e of f d e Unf&tu-nately,theseeventscmlyused90outof the460availabkfhelchannelsintherectorcore.Asweuallthe170eventsoccurred*• resultofM shufflingopemths duringinitialreactor- achalknscbecausedata* dtmg

startup.ThisOpemtknsarcnot~Y ~ “ Ofnamal

reactorfuelingactivity.R~ wethoughtthatanappropriatelytrainedneuralnetworkmodelcouldc-&l eventsintodifkremregionsonthe~ fm. Thistypeofc&Mkadoncouldbeapotentialb t s b ea i nC thenv t eventagainstfti-itydeclamtkns. Regionsm the~facemust bcusedbccauseof the- Ofthegeometryin theproblem.Itis possibkthatthemare&l C tforwhkhthepmpomonof thedismcesfmneachdctccWis theSam&Figure12showsh pmpdons Ofthcdistanmforautheeventsontheeastbceofthereactur●8sfbeld&char&evenm

TheWdappingpointsrqm?entexamplesof thegcomctnc“ Symmeuyproblem.TldSprobkmisthattheratioof thedimncesfiumthetwoopposingdetectmtoa givenM channelcouldbetheSalmformomthanonechannelon thercwtmface Thisproblemcanbe solvedeitherbyusing-of~ ~k Whkhilwludcsymmaicpointsinb same* w bydiminahgtheSylmmrkpohltsduringcreationOfthemodelandalkowingtheneuralMtwrktoinf’’theregkninwhichtheyk Thelattermethodwasusedinthertmdckpmentedh * ~, - ~ *thatusingsymmetrkregionswouldbeagoodsolutionto tbepmblernaswell.Tbepotedal capa-bility to extraphte the&l dischargeIocab basedononlya fm exampimis oneof thekey

14

Page 17: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. .

.. . . ..... . .

.

M

woo1

M

l

)

D&anwh’mDsmw’stoxbstibcendcMmds

18..

u

t ●

““o.8.=,..: .● # ●,9 . 8

● c● ,0 8. . *00

.. ● m●

9●

masonswecboseamuralnetworkmodeltosalvethtspmbb Weesumate0 ‘h’ tminingtbeoet-wwkononlyaf- examiplesofetwbscgion,theotkevcws inthat@on cmddix nfti IMswouldallmvtbemodeltouseeventsthathave~ @ * et w o

T f il m n ’ am ad i vh c hI i eightmgkms.ThischannelmapandthedghtregionsamshmuninFig.13.Alamtalltbemgionswmchosenbecauseoftbedstrkbutimoftbepointsbhavaihbk-

Because&tectmm em * & WtreliablyseeeventsontheOppos&gfat%only1 Cf h n tO u t c fw u mt nwal netwaknxxJeLAlthat@the~ c- c-k ~ Hy ~k b b c~~J~t h) %add ~ hhg ~k ~-tocm Tbeionchambers=’@=” ●

m asnoisedufingthe-E m=~ ~ w -the inputvectorsintoappmpme categories.Back-pmpagatioflwasCbosmas tbe mddiagparadgmi%causcof itsabilitytouseleal-valusdinput&9m Iesb,..llgneural- d w- of 10m 2- IayefseachWitb9 nodeS#and3 Oul,!&‘riletlm!eqm ~usedtoperfbrmabinarymsppingoftbe eightpssibie _ Figw. 14showsagmphicd~-SentathOfthee i g h t -m o

B s c at p eb d go t C o n& aa U a i t t n nt i ascalingpmbkm.Thenumsfkvaluesibtbe heightscanbeverylargentmherstoM toaneurd~To memomethisproblemtbeinputsmeDmmliadtowithinthe Iqpof”lomo1*QTbeSesultof thisMmndbkm is ●vay tightdustsdnsof tbeMxmatbntrainingPmb&mforthenetWodL-9- WoUklusethe

, w ~ * ●

n r O t L

l M a ai c r aw I D

1

Page 18: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. .

.. . . .... . . . . . .. .. .. . .

. .

A

:o●

&

..

92

. .

. . .

. . .“

!“ o

. .:

:. .

,

t

. . . .

:

. .

. .

. .

,

. .

. .. . .

16

Page 19: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. .

.

*

,

. .

. .

i d de n2

iddenl

. .

,1“n. . .. . “ . “ .,

“ F

.“ .

- @. .

,

,

. .

Page 20: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

— _ .

t a n gfunctionwasusedasa transferfunctionbecauseitswidsru~width stretchestheattenua-tion.Thebypwbohc“ tangentequationusedforthetransfafunuion”

x-e-xf(x)”- ●

A a r y l t l b ~ ~f t l oe f a particularno& at level n asa functionof the.

& ~ * - * y i t o O a b M ~ * = ~ ~? l

. g t t n t c Oit is ~ to uscslightvariationsof thestandadback-propagationalgorithm’sdelta-ink.Insteadof thestandarddeha-ruk,thecumulative,gencrit!-izcddelta-rubwasusutloInsteadof updatingtheweightsaftereachpatternPrcscnq acum@a-tivcweightupdatewastalliedandthenappliedattheendof thetrainingepoch.Thishelpsspeedupthetrainingof thenetworkandovcmomcanyproblemsthatmayexistwithorderinthetrainingdata.Themo&lalsousedabiastoassistinspeedinguptheconvergenceofthcnetwork

A fiwr-mgionneuralnetworkwasalsocreatedwith10inputs,1 hid&nlayercontaining15nodes,and2 outputs.~ internalS s a t t f a t k ruk,amthesameasf= h eight-regionmodel.‘JIMtwooutputsof thef--tegion mo&lalsoperformeda- mappingtothefw regionsal themmorfm. Theregionsin thisCawWmchosenran-domlytobequadrantsonthefaceOfthcreactor.m f--rwlionchannclmaDisshowninFig.15. “ , ‘

h the o bothtb eight”andfw”rcgion- &ary mappin~”of h regionnitmbcrswereusedinsteadof bin-ti outputsbecauseit sccmcdtoimprovethe~ SW of thtmo&LBothmodelswereabk toachieveconvergencein underSO000 trainingepochs,withanepochsizeof 12 Weusedtwohiddenlayersintheeight-regionmodelinthehopethatit wouldimprovethegeneralizationof thenetwork,butwc fti thatbothone-andtwo-layernetworksScunaltogenerab equallywell.RXthisprobkm,thew hidden-layernetwdtstwsemorediffi-cultto trainandrequiteda gruatcrnumberof trainingepochs.Thefu-region nctwoskshownin~& 16wascrcatcdusingdy onehid&nlaycrbccauseofthisfinding.

tNEURALNETWORKSFORFUELBURNUPCOMPUTATION

Computingfuelburnupis themostdifficultof theoriginalobjectivesbecause~ contdnsthe_ numberofvariables.Thef- @ctittg&l UQommndtberadktknlevek(ktededby$hcCDMm asf-:

1. . . .

‘ f uwherethebtmdkis loca&i(1of460)2- Ofthcbtmdkinskbqch f@lchannel(l-13)

‘. “j.” “.Tbe%’&chpositkm

~%l’%$%tim” “ “ “ “ “..,. “.“s:”““mUqMNvhk$thefbd. . .: .:. . “.

1

Page 21: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

.

. . ,

. .

B a~eacto~ IFace -

1 2 9 4 S 6 7 0 9 10 11 t2 13 44 1S 36171819 20 21 22 26 Mm 1 m , ■ m.. n m m m

:

. .

. .

.

. .

Page 22: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. . . ..

. . ---

. .

. .

ii ddenl

Fig.16. GrqAkdrepaniajh @Hmgitm namlnuwrk

Forsafe- itmaybe_ tobeableto&tamincifthcfacilityisdischarginglow-btlr-nupfuelfromthereactor.As in thefuel~ -w thinkthatthefwl burnupproblemcanalsobesolvedusinga neuralnetwotktoclassif$theburnupintodiff&aentcatc~ onebeingb W -h i d t ~ ~ n* fw thebump ofcachimlivkiualW bundkbccauscthespikerecodedbytheCDMis anadditiveV o b b & st aT m ca cf w t b c ois ma&alsodependsontherecentimdiationhistoryof the&l bundksbeingdschqcd Fornamdmctoropmtiom**.W wouldneedtobetakeni nx x oB ct hu s cm so incltdalshuf-flingopcrati~ theimdiationhistorybccomcSimkvant.Figure17showsthedistributionof theknown,a&iitivebumupsfmthe&char@ fuelbuldlcsontheCastIWmorf-* Itis~tolmticcthatthebump fallintome Offmdistimt I’cg&nS

Becausetheavailabledataonlycontainedfd eventsduringtheinitial@==-~~-flingoperation,the&l in therectoronlyoccupiedthelastfm positbs ineachof the90 fbdchannelsfmwhichthemwcmrecodedevents.Thefild bundks “&sChar@fimlthcleactorwmneveratdffcmntpositionsinthefuelingchanncL‘Iltismeansthehkposure-&pcn&nt vadab&shavea negligibleeffecton thebump ~ ~ it is possibktomate a neuralnetworkmodelfw classifyingthebump intooneofthcfmcatcgmb basalupontheCDMdata.

‘rhencufalnctUmkmdd usedfa thebump Cdcdath Cmtaiml1 ~ 1h lw 1 n ta 2 outpuWThetwooutputsm usedas8 binaryqmcn@onofthcfmpo6-+cmg!on&Thcinterndstmtureofthc- Suchastbctransfidbncthand karningntle,-~-=--~b ~-- - ~

m

Page 23: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

. .

. .

..-. “

#

# I S

0 I 2 aoe so 6 m 8 9 l 1 l 1 l W lA

. . . .

,

..

[email protected]. Fowtxwgodat#wonkd lmJdn=wPkd=

R OFNEURALNETWORKMODELSFORSOLVINGGEOMETRYANDBURNUP

n n fw solvingtilegeomnryproblemw trabd awltestedondatafiumtbccastf- of tilerctwtor,altbougbtbcwestfm couldhavebeenusedjustas* Bccau8cofthclimitedamountof~ afqxescntativcnumberof fiid eventswe chosenastbctrainingsetinanattempttoexuaphtc M eventsoncbanndsthenetworkhadIMvcrseenbcfm Forb eight”* -63-- -~ thet rs o o 8 t M e ~ ~ ~t facethatwereusedasthetes;set.Asarcsul~72eventswereclasdfkdhy into1of the8regionsontheIeXXorftwc.Thismans ~~~~~~~bfa ninecvcn~ andby chssifkdtbefc@on82%of tbctime.Bcca9Setbe@on wascorlectfa9 at &2Spo6siMccx&@atd cvcnt&thenctwotkcmmpold

Thefm-regionmodelpcrf’ormai-=ofti-

slightlybttcrthantheeight-regionmodel,withotdy32pat-mnsintbc trainingsetoutof thetotal88.Themodelchsifkd 68eventsCaroctlyinm1of * 4legionsonthecastreactorfa Althoughtbisis only77%~, itgcndkcd andcxqdatdtbcpositionf 3 eventsoutofapossibleS6.ThismeansitU@a@atdcomedy68%Oftbetime.‘rhisiscoddmbly betterthantheeight-regionmodelWitbonlynilEoarlectUuqoMm&

Theneuralnetworkforcomputbgbump pcrfomd cspdally well.Itis impwtamthatpartofthispcrformanccis explainedbythe* tbatthe~missingflomthecquatkm‘l’bebump neuralnetworkwastrainedoa48_ outofthc88tiindwtestsa ‘1’b_ C bb b c i 1 a 4 c8tc@cs 81times= isanxmcy of92%,witb33jmuns beingemapdatdOatofa Thisism~ ~-mssof82%.

21

— —

Page 24: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

.. . . .

CONCLUSIONS

TheCDManalysistoolmated herewillactas a pmtotypfu studyingandcreatinga momrobustCDMdataanalysistooLThepotentialof thistoolas a bundkcounteranda power-kvelatoidtwhasbeendemonstrated.Neuralnetworkimplementationsfordeterminingthema of thereactorf= frotnwhichthe&l wasdischargedandtheadditiveburttopof thetwofuelbundleshavebeensuccessfulenoughtowarrantfurtherresemh.Wethinkthatwithamorecampktesetof_n@vc datafromanoperadngon-loadreactor,neuralnetworkl couldbecreatedtorenderupto 1(M%accuracyin positionandbump. Thedataneededto achievethiscapabilityshouldinclude&l pushes%Xnall460*ls Ofthereactorandacompletecyckoff~l thtwghall 13positionsinevery*10 TheresultsobtaindinthisMeat’chaleC%tmnelypremisingm“sideringthelimitedatnountof dataavailable.Webelievethattheresultscanonlygetbetterwithmomdata.

Futureworkshouldincludedevisingamoteaccumtetechniquef= dctmhing areasof intelestintheCDMdata,ratherthanaslidingaverage.Pbwerleveltnonitaittgusinganaverageoverall20channelswillalsoyielda mommcuratepowefkvel cakulah. Deficienciesin thecolkctionofquantitativedatashouldbecormctd Weneedmomsamplesof dataperunittimeanda gamma-ChannelreadingmantqresentativeOfthcmmummentPerio&In_ difkmnttypes&neuralnetwoalcmodelsshouldbetrkdonceampmsentativeamountOfdatahasbeenobtainedThepata-bilityofneuralnetwolkmodelstoothcrmactonlOfthcs typeShouIdalsobeinvestigated.Neuralnetworkmodelshold@eatpromisefa futureworkin theamaof comdischargemonitoringanda um dof largevolumesofcondmmly colkmddatafmbcttctnuclearsafegttmds.Wefirdy believethata commemw~ toolfw monitoringpowerandcountingfuelbundlesfmtnCDMdatashouldbeddopcd.

REFERENCES

1.

2

3.

4.

5.

4

%:

J. K.H aandA.c. Montic~“Pmof-of-pt’incipleMeasurementsfw anNDA-Based(kmDischargeMonitor,’’Nd.Mater.Manage.XIX(Proc.Issue),847-8S2(1990).

J. K.Halbig,A.C.Montkone,L._ andV.S~ “TheDesignandInstaMmofa CoreDischargeMonitorfm CANDU-typeK~,” NUCLMater.Manage.XIX(Pmc.Issue),839-846(1990J

DavidE RummclhartandJamesL ParalkfDistrikedProcessing:E@&n a t4 #C “g 1NeuralComUtht: NeuralWotksIbftional U/PlusUsersManual(NettralWareInc.,K_P~ 1981).

~ ~ ‘- NehMxksP N iS T~fa BreakoutmdicdonincontinuousCasting

Repatr@(April 1991).

M.S.M&S.W.~ andS.H.Chan~”PkmNSPredk&minNuckarIknwrPlantsUsing●Bdt- Pmpa@onLearningNeuralNetwxlL”NmLT~hwL94,270(1991).

2

Page 25: JtmesK.tfalbig johnkhuell “ ~m• George Eccle#on %“rkyF ... · discharge55to65fuelbundkperweekl%gurc 1showsaamccptualdiagramofthisfuelingcyck. Becausethisis anongoingpraess,

... . . . . . - - . . “.. . . .

8. BrieB.BanlcttandRobatB.Uh@ “NuckarPbwerPlantStatusD@noaks usinganArtifi-cialNeuralNctwok,’’Nuc/.TechnoL97,272(Mamh1992).

9. PbilipD.W~ NeiuulCanp~ T- mdfmxiee, (VanNmtrandRe- WYak city,NY,1991).

100 NelmlwaicsPmb&malIvPlusRe&mOCuGuide(Nemlw81e,InC.,~ p~ lgll).. . .

. .

. .,, :

! .

. .

. .

. .. .. . .

1

. .

. .

. . .. .

? .’ .

. . .. . ..

. ... .. . . ..

. .

. .

.,. . .,. . . .

. .:. . .

: . .

. .

, “. ..

. . .

. . . . ,..“. . . . “. . . .

: . .. . . “ . .. . .

. .

. . .

. .

:’. .

. .

. .

. .

23

. .

.. :. .

,

..

. .


Recommended