Text

Summary of 980922 Meeting with EPRI in Rockville,Md Re EPRI Program for on-line Instrument Monitoring & TR That Will Be Docketed to Support Program.List of Attendees & Affiliations Encl
	ML20154D719
Person / Time
Issue date:	09/28/1998
From:	Joshua Wilson; NRC (Affiliation Not Assigned)
To:	Essig T; NRC (Affiliation Not Assigned)
References
	PROJECT-669 NUDOCS 9810070306
	Download: ML20154D719 (108)
	v • d • e

{{#Wiki_filter:_ September 28, 1998 a !EMORANDUM TO: Thomas H. Essig, Acting Chief Generic Issues and Environmental Projects Branch Division of Reactor Program Management, NRR FROM: James H. Wilson, Senior Project Manager Original Signed By: Standardization Project Directorate Division of Reactor Program Management, NRR

SUBJECT:

SUMMARY

OF MEETING HELD ON SEPTEMBER 22,1998, WITH EPRI CONCERNING ON-LINE INSTRUMENT MONITORING On September 22,1998, the staff hald a public meeting with the Electric Power Research Institute (EPRI) at the NRC headquarters in Rockville, Maryland to discuss EPRI's program for on-line instrument monitoring and the topical report that will be docketed to support it. A list of attendees and their affiliations is provided as Attachment 1. Staff from EPRI provided en overview describing its instrument calibration and monitoring program (ICMP). The slides used by EPRIin its presentation are provided as Attachment 2. On-line monitoring of instrumentation would allow licensees to extend calibratior, beyond 18-or 24-month intervals. This would minimize the activities required during outages, minimize the number of contract personnel, and reduce occupational exposure to maintenance personnel. A safety benefit would also accrue due to the ability to be more proactive and look at instrumentation more often than during refueling outages. EPRI requested that the staff issue a safety evaluation of its ICMP topical report within about 6 months. The staff obse. sed that, if it were to perform its safety review within five months and then allow one month for the necessary review by the Committee for the Review of Generic Requirements, a six-month review was possible. The staff also stated that requests for additional information may be necessary in order to get information that the staff will rely on for 2 conclusions regarding the acceptability of the ICMP topical report. Although no plant-specific license amendment requests have been docketed, EPRI stated that a number of licensees are interested in implementing EPRI's ICMP at their facilities. The staff and EPRI discussed how implementation of the ICMP could best be accomplished, including applications by individual licensees to amend their plant technical specifications (TS). The NRR Technical Specifications Branch will be involved early on in the review and able to make informed recommendations in that regard. Project No. 669 Attachments: As stated Distribution: Central FHes Public PGEB r/f RArchitzel JWermiel WBeckner CDoutt LSpessard JRoe DMatthews TEssig JHWilson JMauck CSchulten DOCUMENT NAME: g:\\jhwi\\meetsum.922 o - [' O OFFICE PGEB _ SC:PGEB C:QlCf3 ( (A)BC:PGEB / M JWfrmll TEssig M ~ / f Mll NAME JWilso[ RArchitzel l DATE 9/2f/98 9/ 8 8 9/Z /98 9/)[/98 /9 OFFICIAL RECORD COPY ~ 9810070306 980988' ' PDR TOPRP EXIEPRI 9f. p7 bh k!E $[ @M 10$P, ppg

w. pa %g a k UNITED STATES j j NUCLEAR REGULATORY COMMISSION 2 WAsHINoToN, D.C. 2056 50001 o +9.....,o September 28,1998 MEMORANDUM TO: Thomas H. Essig, Acting Chief Generic Issues and Environmental Projects Branch Division of Reactor Program Management, NRR FROM: James H. W;ison, Senior Project Manager N> Standardization Project Directorate Division of Reactor Program Management, N R

SUBJECT:

SUMMARY

OF MEETING HELD ON SEPTEMBER 22,1998, WITH EPRI CONCERNING ON-LINE INSTRUMENT MONITORING On September 22,1998, the staff held a public meeting with the Eiectric Power Research Institute (EPRI) at the NRC headquarters in Rockville, Maryland to discuss EPRl's program for on-line instrument monitoring and the topical report that will be docketed to support it. A list of attendees and their affiliations is provided as Attachment 1. Staff from EPRI provided an overview describing its instrument calibration and monitoring program (ICMP). The slides used by EPRIin its presentation are provided as Attachment 2. On-line monitoring of instrumentation would allow licensees to extend calibration beyond 18-or 24-month intervals. This would minimize the activities required during outages, minimize the number of contract personnel, and reduce occupational exposure to maintenance personnei. A safety benefit would also accrue due to the ability to be more proactive and look at instrumentation more often than during refueling outages. EPRI requested that the staff issue a safety evaluation of its ICMP topical report within about 6 months. The staff observed that, if it were to perform its safety review within five months and then allow one month for the necessary review by the Committee for(ne Review of Generic Requirements, a six-month review was possible. The staff also stated that requests for additional information may be necessary in order to get information that the staff will rely on for conclusions regarding the acceptability of the ICMP topical report. Although no plant-specific license amendment requests have been docketed, EPRI stated that a number of licensees are interested in implementing EPRI's ICMP at their facilities. The staff and EPRI discussed how implementation of the ICMP could best be accomplished, irciuding applications by individual licensees to amend their plant technical specifications (TS). The NRR Technical Specifications Branch will be involved early on in the review and able to make informed recommendations in that regard. Project No. 669 Attachments: As stated cc w/atts: See next page e

l l 1 1 LIST OF ATTENDEES AT MEETING WITH EPRI MELD IN ROCKVILLE, MARYLAND ON SEPTEMBER 22,1998 NAME AFFILIATION J.H. Wilson NRC L. Spessard NRC J. Wermiel NRC J. Mauck NRC C. Doutt NRC W. Beckner NRC M. Reinhart NRC C. Schulten NRC R. Shankar EPRI J. Butler NEl G. Taylor SCE&G P. Rose SCE&G R. Rusaw SCE&G E. Davis Edan D. Hooten Toledo Edison R. Singer ANL l Attachment. l

_ -. _ ~.... _ _ _ _ -. - _ - -.. _ _ _ Proj:ct No. 669 Electric Power Research Institute 1 - Mr. Kurt Yeager President and CEO Electric Power Research Institute 3412 Hillview Avenue Palo Alto, CA 94303 Robin Jones Vice Presidet and Chief Nuclear Officer Electric Power Research Institute 3412 Hillview Avenue Palo Alto, CA 94303 Mr. Raymond C. Torok Project Manager, Nuclear Power Group Electric Power Research Institute ' Post Office Box 10412 Palo Alto, CA 94303 Mr. Gary L. Vine Senior Washington Representative Electric Power Research Institute - 2000 L Street, N.W., Suite 805 Washington, DC 20036 Mr. Bindi Chexal Electric Power Research Institute Post Office Box 10412 Palo' Alto, CA 94303

4 NRC-Utility Meeting On-Line Monitoring Sep 221998 AGENDA 8:00 a.m. Introductions & Meeting Purpose Shankar, EPRI 8:10 a.m. Utility Perspective Taylor, SCE&G 8:30 a.m. Utility Experience SCE&G Rusaw First Energy Hooten 9:15 a.m. EPRI Support in Tech Transfer Shankar 9:30 a.m. Discussion & Feedback 10:00 a.m. Break 10:15 a.m. MSET Overview Singer, ANL 10:30 a.m. TR-104965 Davis, Edan NOON LUNCH 1:00 p.m. TR-104965 (cont.) 2:00 p.m. Discussions & NRC feedback '2:30 p.m. SER Schedule & Wrap Up 3:00 p.m. Adjourn 9

i i V /,

,m a

2 u 'Q >= H D* h M, r ummme l L / mama 4 Wk 4M v m< >m V

HISTORY 1990 VCS Committec to serve as Host Utility for the EPRI-ICRP Prof ect. Octo aer 199: - Tae VCS/EPRI System. went "OX-LINE" June 1992-Startec. T1e ICMP USERS GROUP Svit:1 S.O.N.G.S. MAY 1994-VCS.ae..c. an incustry wor 1 shop on ICMP

OBJECTIVES To im: grove P: ant Safety. To improve : plant Equipment Availa 3ility. Recuce cali~3 ration Incuced human errors l Support Incustry goals for development anc. integration 0:f new techno..ogies to enaance p.. ant 0;perationa: Per:formance. I l i

OBJECTIVES Provide a cost effective automated tool to En aance P: ant Engineering & Maintenance functions. Provic.e a means to support P:. ant & Industry ALARA Objectives. Provide a means to Rec uce Radioactive Waste. P f I i

t O 1 ICMP PERFORMANCE Operational t: 1rouga Five :itel cyc..es. Total of 77 months. Provic.ec. 98% System avai:. ability. Has provic.ec. Early Detection of Eig:1t Failing or Non Con: forming Concitions in Sa:Tety Re. atec. Equipment t:aat requirec. maintenance action. i

E- .A a -4 --h ed ide* -a 9 0 Il l l l l l l l l 1l I I l l 3 l I l l l l l l l l l l l l l l l l l l a h 'k O __. R e 5 E Esk D F \\ 1 a ~L \\ h .b l o bf1]$ 8 t i t e. 3> g c n 8 s ~ ~ 0l j 8 1 w g 3 a B ) r e g ~ q ~ k e a o b () J f g g 5 g i r ) N d hh 1,,,,,,,,,li,,,,o C ,,,,,,11 1, 0, e fal D

e e O _- ~___ _ _- _ - _ - ~ _ _ _ _ - I i o s4 8 s0 9 .o6 v / ma t o8 3 s 9 vV I I / t t 8 e/ / l i o9 i " e i I t q TI W A MR t P e e K y m. t I Mp t t aa 9 I n_ l T t D tmt t e o ol 7 Dnt nB m / i / 4 l 0 T i c I R A T I S i A T D I m I A 8 ^ 9 v7 /3 4 2 4 I ^ l / 4 5 0r I I P E l oe tc i E N i 1 T t 4 s 2 D aa E i I h R US i S 7 E 5 P I W 1 O t 4 i rP R 2 1 0 i E ^ l O tl U 1 I E l ~ SS RP I ~ I 1 ~ I 65 40 ~ I T 3 j P i I A I g_ 1 ~ I I I I 55 I 1 5 4 4 4 0 0 rT 1 1 PP _~ - ~.. - - ~ _ OV 79/ 4 / 1 1 0 0 0 0 0 0 0 0 0 0 0 0 6 2 0 0 2 2 2 1 2 2 2 2 EU l l sj \\llllji l .!ll: i' i,t 1

PERFORMANCE " possibilities " \\ [ Elimination of 2: 6 Calibrations z z-/ Outage Reduction of ~75 RCZ Dressouts I Recuction of :..z-Rems l I i

l l TODAY'S ACTIVITIES Continue TR-:.0L965 efforts ::nough the Working Grou:7 l U:?c ating VCS to the new EPRI/SAIC

program.

Agreed to be a Partner Utility wi:h ANL on MSET/SPDS signa: validation project.

? I FUTURE PLANS Draft Tech Spec Change for submi:tal after a:pproval of tae Re: port. Irr:egra:e MSET irr:0 ICMP for use on Non-Safe,y Instrumerr:ation. Imp:.emen:ICMP as an 0;pera:iona:. P:.an: Program I s

I I i On-Line Monitoring i i

~." :' + " i :b -:t se:.m s

ni]~ 1 w"*" ^ 'i_;_,. t Ramesh Shankar l 704-547-6127 l rshankar@epri.com EPRI Charlotte Energy Conversion Division l i i f NRC-Utility Meeting on On-Line Monitoring i Rockville, Md j September 221998 i EPE l i

Instrument Maintenance & l l Calibration Reduction f ':) W fg d Q *1 7 - M ae.A u,- 4Was7#tf*@p*Md9L+et rypy., t - Utility Concern on Reliability and Costs l s - Over 30 Products Developed by EPRI - Guidelines for Calibration Extension-TR-103335 t - IPASS - ICMP - Review of Technical Issues Related to Oil-Fill Loss TR-102908 l - Nonprocess Instrumentation Surveillance Reduction, l TR-103457 i i EEP121 I i .l

o = 4 %M hCV9) a M *QW y 9Moflg/ 7 Whym5 y v. p {anned 4Actualg ,A ; ; 6W hi+MAfW{ rg rg*M$1 y"gt,*vg'FNM%%D bF MN5M Compleid.CampletionNf% ~ N:.f ip :d@.WW.t hM'4 -T y 9(;, s. h JUfFA ($ " Fredust ! Ns N I DelifedbieMhNOldNedildMMM.MN M$dMNdNblei Mb?dMMWDatelMWDateWNanager i R-1034%VI INS I KUMLN r CALiliRAllON AND MONilORING PkOUP AM 12/31/93 12/31/93 Shmaar (K'MPP B AS13 FOR 11tE METITODOtI)GY, VOLUME I AP 106822 Instrument Cahtsanon and Monnormg Program, Release 2: Seitware $ 2/31M8 ShmAar and User's Manual IR.104965 On-Lene Morntormg ef instrument Channel Peiformance 41/30/98 Shma m FCAfr Techaefogy Pac Aage !R-103436-V2 INSTRUMENT CAllBRAlK)N AND MON 11ORING PROGRAM 12/31/93 12/31/93 Shmam (! CMP);lNSTRUMENT FAILURE MODES AND EITECTS AN ALYSIS, Vol.UME 2 N P-7207-(TMLV I INSTRUMENT CALIltM ATION RFDUCilON PROGR AM 3/IIMI 3/3 BMI shaedar NP-7207-COMLV2 INSIRUMENT cal lBR ATION RFDtKTION PROGR AM 3/31/91 3/31/98 Shad ar NP-7207-CCMLV3 INSTRUMEN! CAllBR ATM)N REDUClk)N PROGRAM 3/3 tMI 3/ 3 tMI Sha Aar IN-104181 INS TRUMENT CAUDR AllON R EDUCTION METilODOI.OGY 2/1/94 2/IM4 Shard ar AP-106752 INSlRUMENT PERFORMANCE ASSESSMLNT SO!TWARE 6/1/96 6/1/96 Shmaar SYSTFM ifPASS) iR-103335 GUIDELINES FOR INSTRUMENT CAllBRAllON 12/31/93 12/31/93 SharAm i EXTENSION /REDUCT10N PROGR AMS iPASS Techaefogy PacAage IN-103609 APPLICATION OF INSTRUMEN T CAUBRATION EXTENSION-14/30/95 11/30/95 SharAar REDUCTION GUIDEI INES AT FT. CAlllOUN A P-106752-R I D K IPASS, Release 2 (He's Venion) 1/31M8 3/11M8 Sh:ua ar Mi-Il0158 andiestry Cahbration Database 7/)1M9 Shank er 1 0-809339 CONDUCI PILOT TR AINING FOR USE OF IPASS 82tl2M7 12/12/97 Shmaar I R-103335-R 1 GUIDELINES I OR INSTRUMENT CALIBRATION 9/30M8 Shadas NEON EX1ENSION/ REDUCTION IT *)GR AMS-Revisio Maintenance & Calibration Reduction Guidelines SW-lob 752 R2DK iPASS, Release 2 7/31N9 hAar 10-109339 RISL Development of IPASS Iramire. Comse 7/3 tM9 hats !R Ii1348-51 Instrument Dnft Study:Unizio flydro Bruce Ninlear Generareng I1/30/98 SharAar Station instrument Caharutsentsarrestlance I R-loo 819 Instrumentatmn Cahbrueson Re lortion llalbook 12/31/00 ShasAar Reduction Tools N P-6067-V I INSIRUMFN E AllON SURVi ll LANCE:MEllK)DS 10/31/88 10/38/68 ha:s NP-6067-V2 SUR VEll1ANCli OF INSlR UMENTAT ION Cll ANNI l.S 12/11/89 12/31/89 Shavit ar IR 403437 NONPROCESS INSIRUMLNTAllON $URVLillANCE AND 12/31193 12/3 tM3 Shaeaar IEST R FDUCTION NP-7523 PRODIliDINGS:l&ClEST LEDUCIKJN WORKSilOP 12/31/89 12/31/89 Sluinbr NP-7243 INVESTIGAllON OF RESPONSE TIME TESTING $/l/91 $/1/91 Shardar REQUIREMENTS NP 7243 RI INVLSilGAi!ON OF RLSPONSE llME lES ilNG 1/27/93 1/27/95 Sha Aar I'RRATA REQUIREMENTS FRR ATA Response Time Terting Srmdies i B.NP.59.7.9 8 EPRI SlT.lDIES PRESSURE SENSOR RESPONSE llME lESIlNG 7/l/91 7/IMI shadas REQUIREME NP-7121 GUIDANCE FOR FAILURE DE1LLTION OF ROSEMOUNT ll/3IMO 12/31/90 Sha Aar IRANSMITFRS IR-102908 RiiVIEW OF iliCilNICAL ISSUES RLLAlED lO I IL1-Of L LOSS 8/IM4 8/lM4 ham INDUCED FAILURE OF ROSEMOUNT PRESSURE T R ANSMITTTP S Ka<traries Alemterrag Feels 1 R - 102644 CAllBRAllON OF RADIAlN)N MONilORS AT NUCLEAR 3/l/94 3tlN4 Shardar POWFR PLANTS I R-104081 UTILilY EXPERIENW Willl MAJOR RADIATION 12/20/94 12/20/94 Shasaar MONITORING SYSTTM UPGR ADES

l Technology Transfer __h._'. f. L j .t' i ' -Workshops & Training - Guidelines L'pdate - L~ser Groups - Software C pgrades j -Working Groups for Emerging Issues - Utility / Vendors j -Engagement with the NRC EPfiE!I

l L On-Line Monitoring l ... e -eta, & mimstenmwnr - -..:::: # c.:vma :<.u.am- - Formation of Worring Group - First Draf: Submittal of TR-10L965 (8/95) - RAIs (2/96) - Revised TR-104965 Reviewed with WG and NRC (5/98) - Advance Copy (9/98) i EEE P riE!l l I

i TR-104965 - Provides basis for extending calibration intervals - Contains licensing aspects ofimplementing OLM - Describes Benefits of OLM - Details Industry Experience with OLM - Edf; B&W Owners Group; CANDU - Change will require Tech Spec change approval. Guidance for submittalincluded - Detailed Answers to NRC RAIs (2/26/96) l t EEE P l21

f TR-104965 Purpose ,_m m,,m,-.. - - At least 1 redundant transmitter calibrated each fuel cycle; all calibrated at least once j every n outages - Transmitters identified out-of-calibration will be calibrated as necessary - No change in bistable functional checks - Periocic channel checks to continue i E E P fiE!l i

EPRI Follow-Up - - -,,na- =, -ICMP Software Commercialization l - Licensing MSET for E tility Application - E ser Group Formation - Training & Workshops - Update TR-104965 and/or develop new products as necessary I EEPliE!l' .t

t r o f-l t Toledo Edison /First Energy Experience Dave Hooten i 1 l f t }

On-Line Monitoring Methodology Evaluation B&WOG ICRWG o report comparing & evaluating i various techniques / approaches. o participation in DOE project i i o recommends ANL's MSET/SPRT l I l I

Collaborative Project with University of Cincinnati and Argonne National Lab MSET/SPRT application to feedwater flow = ~100 correlated signals. o detect slow degradation o GUI development (i.e., bridge software) l b i Demonstrate MSET/SPRT capability to plant staff j I l 1

Long-Term Intentions Use with non-TS instruments upon completion of UC/ANL project Submit LAR to implement OLM on safety-related instruments o coordinate with B&WOG Tech. Spec. Committee o appropriate relationship to ISTs t i b a

Expected Benefits Calibration reduction o hire fewer contract I&C techs during RFOs o minimize human errors during calibration Distinguish between process and instrument anomalies o RCP motor thrust bearing temperature example o identification of oil less in Rosemount transmitter example i Early warning of problems o proactive vs. reactive maintenance o support Maintenance rule activities l GL 91-04 monitoring program for instrument drift over 24 month calibration intervals o supplement to AF vs. AL analysis o provide insight into drift behavior vs. time. f t

l l THE MSET PROCESS MONITORING SYSTEM AN EXECUTIVE

SUMMARY

Ralph Singer, Kenny Gross and Stephan Wegerich Reactor Analysis Division Argonne National Laboratory >

THE MSET SURVEILLANCE SYSTEM SYSTEM OVERVIEW The MSET Surveillance System (Multivariate State Estimation Technique)is a software-based. highly sensitive and accurate tool for on-line monitoring of the health of any process that has at least one sensor. MSET can detect and identify any malfunction that may occur in process sensors, components and control systems as well as changes in process operational conditions utilizing an unique and patented integrated suite of statistically-based pattern recognition computer codes. These codes interact and operate in a conceptually simple manner to provide the user with information needed for the safe, reliable and economical operation of a process by detecting, locating and identifying subtle changes that co,uld lead to future problems well in advar.ce of actual degradation. To utilize the MSET Surveillance System, all that is necessary for the user to do is collect sensor-generated data from the process under consideration that bounds all normally expected operational states. These data are used by the MSET system to establish the domain of normal process operation (i.e., " train" MSET t.o recognize nonnal behavior) and will be used in the monitoring phase to determine malfunction incipience. During monitoring, sensor data are read by MSET, an estimate of the current state of the process is determined by comparing the measured sensor data with that obtained during training and the difference between this state estimate and the measurement is calculated. This difference or estimation error is then analyzed by a statistically-based hypothesis test (the sequential probability ratio test or SPRT) that determines if the process is operating normally or abnormally. If an abnormal condition is detected, the initial diagnostic step identifies the cause as either a sensor degradation or an operational change in the process. When a sensor degradation is identified, MSET utilizes the estimated value of the signal from this sensor to provide a " virtual sensor" that can be used to fully replace the function of the faulted sensor, i DESCRIPTION OF THE MSET STRUCTURE AND MODULE FUNCTIONS i, i A flow diagram illustrating the architecture and data flow of the MSET Surveillance System is shown in the figure below. All of the system modules located within the large l rectangle are represented by fixed coding, i.e., these modules are generically applicable to any monitored system and do not require modification for new applications. The only interfaces i f.'.ke.., 4 C.y, - _, :4.,. > i 27g ^ ;;.g g._ 7 ;, 7 g.. g [ y.

.9;*, %..~.$.v-l h ( M :l h h l B EL-2

" M,*o.:~ ggg(g sma k" [ p-FMT t aza.. :. 5 ga,p. { pgg ClG0$fC l SY31EM S gg. sm g3 j / 'q eut L

S%1EM.

~I ~ ~. L4 l .f t=4 MODEL',A ,a. Wsa m q.ds) l b)- ':S g: s . s _, < gm,v. q 9,, l .. : s. ~

)

.1, l necessary are those that supply the signals from the monitored system and a standard user display j i screen which can be customized if desired. Additional detailed diagnostics can also be supplied that address specific process issues. 1 To operate MSET, the first step is to train the model to recognize normal behavior of the ] monitored system (process) and its sensors and components. To do this, it is necessary to operate 4 I

the process over the full range of conditions that are expected to be seen. During this phase, data from all sensors are collected and stored in a file; MSET then selects an optimized minimum set of these data that are sufficient to determine the state of the process during the subsequent monitoring phase and places these data in the model's memory. Once this step is completed, monitoring can be started in which signals from the same sensors used in training are fed into the MSET system model. In this step, the measured data are continuously compared to the data in the model memory to determine the best match between the current process conditions and that learned as normal conditions. From this comparison, which utilizes a pattern recognition algorithm optimized to minimize error, MSET predicts estimated values of all the sensor signals. The difference between these estimated values and those measured is calculated and this error residual is provided to the FAULT DETECTION module which contains the SPRT fault detection and identification algorithms. Based upon the statistical characteristics of the error residual, SPRT determiner, if a fault is starting to develop in any of the sensors or process equipment or if the operating state of the process is starting to deviate from that known to be normal or desired. If no incipient faults are detected, monitoring continues. If incipient faults are detected, logic in the LEVEL 1 FAULT DIAGNOSIS module utilizes the output from SPRT to determine the location of the fault;i.e., identifies the specific sensor or component that is indicating probable future degradation or annunciates the beginning of an abnormal operating state of the process. More detailed diagnostics are reserved for the LEVEL 2 FAULT DIAGNOSTICS module in which process-specific data, such as flow charts, component design information, FMEA ( failure mode and effects analysis) results, etc. are utilized. The MSET Surveillance System has a number of important and unique features that provide capabilities beyond that claimed for other monitoring systems, including the following: L MSET automatically selects the optimum set of sensor signals needed for the determination of the process operational health during monitoring;

2. MSET analyses and preprocesses the signals being monitored to optimize their informational content for use in fault detection;

3. MSET is trained to recognize desired operational conditions with a one step, deterministic calculation; 4

4. The signal value estimate provided by MSET is extremely accurate with errors between estimate and measurement ranging between 0.1 to 0.5 %;

5. MSET provides accurate signal estimates even when a large fraction of sensors are providing erroneous information or are totally failed;

6. The MSET state estimate module predicts not only the mean value of the signals but also the " deterministic noise" riding on these signals (critical for early fault detection);

7. The SPRT module utilizes the characteristics of the signal noise to identify incipient faults with the theoretically minimum number of observations;

8. SPRT permits the user to specify false-alarm and missed-alarm probabilities, allowing the coatrol of the likelihood of missed detection or false alarms;

9. Faulted sensors may be replaced with virtual sensors generated by the state estimate;

10. The kernel of the code is only a few hundred lines and for most practical applications runs in real time.

EXAMPLE APPLICA TIONS OF FA ULT DETECTION USING MSET The MSET process monitoring system has been successfully applied for the solution of numerous industrial problems involving issues ranging from product quality control to detection of process abnormalities. In one application involving a high-speed automated production line,it was necessary in real time to identify products that did not meet certain specifications and to separate those so identified from the acceptable product. Only one sensor was available and due to its poor signal to-noise ratio, unnecessarily large numbers of product were being rejected. Following the installation of MSET, the quality decision was substantially enhanced and the product waste (good product being rejected) was reduced by approximately 25%. j At an electric power generating station that utilized several thousand sensors for control, safety and monitoring, the task of determining sensor health and identifying recalibration and i replacement needs was expensive, time-consuming and inefficient. MSET was installed on the station's central data acquisition system to monitor 88 of these signals in a demonstration beta- -

i test. During this test period, the plant engineers were able to continuously validate sensor operability and avoid unnecessary maintenance as well as detect a,id identify subtle faults in several plant subsystems months before they were observable using the pre-existing methods. One of the regulated conditions of automotive internal combustion engine operation is that of misfire rate. When a misfire occurs in an engine cylinder, unbumt fuel is released to the environment and the catalytic converter is degraded. Thus, an important issue is the detection and identification of engine misfires in near real-time (millisecund time scale) utilizing existing engine sensors with an extremely low error rate. MSET was applied to this problem and was able to identify specific cylinder misfires occurring during both steady and highly transient engine operation with error rates less than 0.1%. In the following sections, two specific examples of MSET capabilities in detecting - abnormal behavior in advance of physical damage will be presented, including the type of information that is provided to the user. SUBTLE DEGRADhTION OF A CENTRIFUGAL PUMP One of the most useful attributes of MSET is its ability to detect subtle changes in the operation of sensors and equipment that are indicative of future degradation wellin advance of actual malfunctions. This capability is illustrated by the following example of a long-term monitoring of a liquid centrifugal pump. In this situation, the pump was artificially degraded by reducing its output flow at a rate of 0.2% over 50 days (i.e.,0.004%/ day). MSET was trained to recognize the normal behavior of the pump prior to the imposed degradation and then used to monitor the pump's performance. After 15 days of normal operation, the degradation was ' initiated at time zero on the plot shown below. The upper portion of the figure shows the actual measured pump flow rate superimposed upon the estimated flow rate generated by MSET. The " center portion shows the difference between the measured and estimated flow rates. In neither case can any deviant behavicr be noted - the usual monitoring technique of data trending would predict completely normal operation of this pump. However, as can be seen from the lower portion of this figure, MSET is able to extract sufficient information from the difference between 6-

l l the measured and estimated flow rates to conclude that a problem is starting to develop as early 6 Flow Meter with O.2% Drift Over SO Days 3s.e 36.4 Flow Segnal 35.6 MSET Estam.ste - 35.4 -10 -5 0 5 10 15 20 25 30 i MSET Estimation Error 1 O.5 g ""Y '.* * "'K* 2-W* " ^ '^ - *f4TT ^"" '*& # "'SWS^' HE ^ ~~T ^, ^ n L ^* M O MI:"^* ^ -0. 5 -10 -5 O 5 10 15 20 25 30 SPRT Indices Showing Degradation to-r- 1 , i; c, f ;; }7 j~i r.,h g ;-l 5 g- ! P r. } ..hM 1-.

,l.i o>

-~. ~ ,-t : ; 9 J. -5 .) ;' -~?.'i$T:

2

-10 -5 0 5 10 15 20 25 30 Days After Start of Sensor Drift as 13 days after the start of the degradation when the flow rate has changed only about 0.05%. If maintenance of this pump is required when its flow output drops 1% (which in this example occurs after 250 days), MSET has provided almost 8 months advance notice for planning this work. FAILURE OF A PRESSURE SENSOR The previous example was one of the detection of a very slow degradation of a liquid pump; this example will be of a more rapidly developing fault. In this case, a pressure sensor, which has been operating normally for an extended period of time, fails (i.e., its output drops by about 5% in several hours). Again, MSET was trained to recognize the normal behavior of the system in which this pressure sensor was located and then used to monitor the system. As shown in the figure below, the measured and estimated pressure level are plotted in the upper portion and they agree quite well for the first 6 cr so hours of the monitoring period. This is also indicated by the center portion of this figure which shows the difference between the measured and estimated pressure. As can be seen in the upper plot, the measured and estimated pressure values clearly diverge after about 7.5 to 8 hours (a 1.7% difference) and if this signal was being closely watched, this failure would probably be detected. However, at about 6.3 hours, more than a full hour before the fault could have been normally detected, MSET starts to alarm (the lower ponion of the figure). If this had been a operationally or safety critical sensor, the process could have been shut down prior to the loss of this sensor. However, it would also have been possible to utilize the estimated sensor reading from MSET to replace this faulted sensor and to continue operation of the process. I -

Pressure Transtnitter Failure Detection 3.2 3.1 Sensor w MSET =:> First SPRT Alarm i g 1 y 2.9 - -~ ~. xm 2.e 2.7 1 2 3 4 5 6 7 8 9 10 1 MSET Estimation Error ./ i 0.5 n)r,e),pfl-y r:: e O ~ Wm ~~ '" ~~ "

w&'" ~% W' * *N* <!~+V~+d*
N ^

T -0. 5 First SPRT Alarm l 10 1 2 3 4 5 6 7 8 b 1 i SPRT Indices Showing the Onset of a Failure 10 l s

p s

g 0 -+ -,,:<- y - ~ - * - ~ ' ' ' ' ' * - M .}: e i ~5 l _10 3 4 5 6 7 8 9 to 4 1 Hou rs ) 1 1 > 1 I I

~- Chicago Tribune l Business News Page 1 of 6 Olitago@ibunt $@ DEEM m GLEANING NUGGETS BUSINESS IN THE NOISE NEWS BUSINESS NEWS By Jon Van HOME Tribune Staff Writer August 28,1998 se Antici ating when a bridge needs ' A VI Superior P Rate. replacing, when a nuclear power $. q* y ~ 5?7,5%W, and, perhaps, when the stock market is @/ 1 g g generatmg plant is gomg out of kilter g6 MonthfDf due to take a dive may all be possible ~ 4 with variations of the same basic 1 computer program. gucjeagn atn arner g Stephen Wegerich at Argonne Indeed, researchers at Argonne National National Laboratory. (Photo for Laboratory in Chicago's westem @',,7"ft,"3ne photo by Stacey c suburbs have developed software that can analyze pattems from just about any ) kind of complex signal and detect subtle variations associated with big changes @ EMAILTHIS STORY before they happen. Send the text of this story to sorneone's ernait address The software was developed to help spon aed oy mstreet.com engineers discern whether warnings triggered by sensors monitoring steam pressure, flow and temperature in h SEARCH ARCHIVES nuclear plants are legitimate causes for Browse the Tribune arenive for concern or mere false alarms. ther articles In an industry where safety concems dictate three and four levels of sensor redundancy, false alarms from failing sensors can cause costly, unneeded equipment shutdowms. Telling real alarms from false ones is a huge issue, one important enough for the federal j Energy Department to sponsor a j national contest to see if anyone could solve it. Researchers were given 18 months worth of digitized signals from a nuclear plant in Florida, where 3,400 sensors fed information about the state of processing equipment. Buried within this mountain of data were 10 simulated sensor failures, and the contestants were invited to find them. Argonne's team not only spotted all 10 http://chicagotrib une.com/ business /businessnews/ article /0,1051, ART-13855,00.html 9/21/98

. Chicago Tribune l Business News Page 2 of 6 simulated s:nsor faults, but its program also found two subtle disturbances in the data, suggesting potential failures that the Florida plant's engineers didn't know about. None of the other contestants discovered any of the planted faults. "We won that competition in 1996, and it really sparked a lot ofinterest from the industry," said Kenny Gross, an Argonne researcher who is one of the developers of the mathematical model on which the software is based. The approach taken by Gross and his Argonne colleagues, including Ralph Singer and Stephan Wegerich, is to not only look at signals from monitors, but also to study patterns in the so-called " noise," the part of the data thought to be meaningless variations in the signal associated with vagaries in the apparatus. Gross and his colleagues have come to believe that this " noise" harbors usefL information about the relationships among components in systems under scrutiny. According to this theory, if one valve in a nuclear plant becomes sticky and is likely to fail eventually, subtle pressure changes measured directly by monitors near the valve may not spot the trend. But when combined wit other subtle changes throughout 'he system that stem from the sticky vr' , he problem may - be diagnosed. Called the Multivariate State Estimation Technique, or MSET, the software begins by looking at the signal patterns generated when everything in a process is working nonnally, and it then keeps looking at the patterns over time. Using its mathematical formulas, the computer can decide whether an alann produced by one sensor is genuine or if the sensor itselfis faulty, not the valve it's monitoring. When MSET spots a faulty sensor, the computer can generate a virtual sensor signal to replace the faulty one, keeping http /chicagotribune.com/ business /businessnews/ article /0,1051, ART-13855,00.html 9/21/98 1

_._ _ _ _. _ _ _ _. - - _ _ _.. _ ~. _.. _. _ _... _ Chicigo Tribune l Business News Page 3 of 6 r e the system running smoothly. MSETs value to nuclear plant operations is why it was developed at ' Argonne, which has a mission of a providing technology to the nuclear powerindustry. Once they proved MSETs value, the researchers decided their sophisticated signal analyzer probably had applications elsewhere. Othus agree. The National Aeronautics and Space Administration has provided funding to MSET's developers to produce a version of their soRware that may help NASA avoid costly delays when it launches the space shuttle. Developers believe even enterprises with lower profiles may benefit from MSET, and they've found backers to provide capital to start a private firm, Smart Signal Corp., to exploit applications outside of nuclear plants. The first target market has undertakings similar to nuclear plants, said Alan Wilks, vice president of development and operations for Smart Signal, which 'is based in Mt. Prospect. "The chemical processing industry has lots of sensors and complex processes," . Wilks said, "so that's a natural. At - petroleum refineries, for instance, accurate monitoring is a life-and-death matter." Some fatal refinery explosions have been caused by pressure sensor failures that went unnoticed because the sensor continued to register in the normal range when, in reality, pressure was building to explosive levels. By scrutinizing noise in the signal, MSET can detect such failures. "What the work at Argonne has really shown," said Wilks, "is that 50 to 80 percent of what people used to think of as noise is really predictive if you know how to use it." ( This insight makes MSET so powerful '. http://chicagotribune.com/ business /businessnews/ article /0,1051, ART-13855,00.html 9/21/98

' Chicago Tribune l Business News Page 4 of 6_ .-that Wilks said the technology miy have. applications over a wide range of areas. Alan Schriescheim, director emeritus of Argonne and a member of Smart Signal's board of directors, said that "this technology has been fully vetted and there's no doubt that it works." "The real question before Smart Signal ' is where to focus our attention. We have limited resources, so we need to decide how best to use them to advance the company." Argonne's Gross said he has worked with a financial firm that wishes to remain anonymous to explore the software's ability to predict stock market ' trends. The goal was to train MSET to - look at patterns generated by price fluctuations and note subtle connections in the noise. J As those patterns change, the program made predictions. "We had data for five years," Gross said. "The first four years were analyzed and i then a prediction made for the next day. Then we did it again for the day after that and so on." Someone following the computer's predictions to make investments could have doubled his money in a year, the initial research suggested. Such analysis ' is possible using a personal computer, Gross said. "It takes about half an hour to do the analysis, but since you have a lot of time after the markets close, that's fast enough for you to make investment decisions for the next day," he said. MSET also may be employed to extend the useful life of bridges. Steve Copley, chief executive of the Packer Group, a Naperville-based engineering firm, said that as bridges become older, state highway departments face a dilemma because they don't want to repair or rebuild bridges before it's necessary, but they don't want to wait until bridges fail, either. http://chicagotribune.com/ business /businessnews/ article /0,1051, ART-13855,00.html 9/21/98 1 \\

Chicago Tribune l Busin:ss News Page 5 of 6 Putting strain gauges at key points on a bridge and transmitting the signals to a central monitoring point is seemingly a good way to solve the problem, but until now there has been no easy to way to analyze the signals to determine quickly when each bridge needs detailed human inspection. - "MSET looks like the missing link here," said Copley, whose firm has become a minority owner of Smart Signal. John Fildes, a senior scientist at Packer, said that officials in the Illinois Department of Transportation have been receptive to trying out the technology on some bridges in Illinois, perhaps later this year. Packer also has discussed the notion with other states and with federal highway authorities. When a crack begins to propagate in a bridge's infrastructure, the strains borne by key supports begin to shift, he said. This is a pattern shift that Smart Signal's software should detect quickly. The alternative has been to post older bridges with regulations limiting the speed and weight of vehicles using the bridge. About 70 Illinois bridges currently are posted, with a couple hundred more close to it, Fildes said. "There's a real economic cost to posting bridges," Fildes r aid, "and bridge monitoring probebly is a less costly alternative." In a variation on this theme, Gross said that he has even undertaken some research for people who operate amusementparks to determine if MSET technology might improve the safety of roller coasters. "This isn't just the tracks, but the coaster cars, too," he said. "You might be able to put microphones at different points along the route to monitor the sounds and look for changes that might mean bearings were wearing out in a wheel." http://chicagotrib une.com/ business /bu sinessnews/ article /0,1051, ART-13855,00.html 9/21/98 4 4 r

Chicago Tribune l Business News Page 6 of 6 RETURN TO TOP l HOME w,b sit. to wcach in 199a= DQ av,.n u w..k u t 2/es SECTIONS DragDEributte fdf i5 SPORTS BUSli4ESS LEISURE COMMJfJJi;ES f/ARKETPLACES INTERACT TOOLS C U S TG:/4ZE HELP SLAV.e EITE MAP ADVERTISING OUESTIONS/ COMMENTS l COPYRIGHT NOTICE AND TERMS OF SERVICE l http ://chicagotribun e.com/ business /b usinessnews/ article /0,1051, ART-13855,00.html 9/21/98 1

i l intellig nt System Applicitbn to Tower Systerns(ISAP'97) July 6-10.19% Seoul. Korea MODEL-BASED NUCLEAR POWER PLANT MONITORING AND FAUI T 4 DETECTION: THEORETICAL FOUNDATIONS Ralph M. Singer, Kenny C. Gross, James P. Herzog, Ronald W. King, Stephan Wegerich Argonne National Laboratory Argortne, IL, USA and Idaho Falls, ID, US A l ABSTRACT - The theoretical basis and validation industrial processes. This paper will describe the i studies of a real-time, model-based process monitoring theoretical basis of the resulting monitoring and fault and fault detection system is presented. Through use of detection system and summarize the supporting a non-linear state estimation technique coupled with a validation studies. A companion paper in this probablistically-based statistical hypothesis test, it is Conference [2] will present results obtained from an possible to detect and identify sensor, component and experimental monitoring program performed at a process faults at extremely early times from changes in commercial nuclear power station. the stochastic characterist cs of measured signals. Date i from an experimental fast reactor and a commercial OVERVIEW OF APPROACH i PWR are used to demonstrate functional capabilities of the monitoring system. Numerous objectives can be ascribed to a modern process monitoring system, but perhaps one pf the most KEY WORDS important ones, at least in terms of reducing operational and maintenance costs is the following: to detect and State estimation, fault detection, hypothesis test, identify incipient disturbances sufficiently early to surveillance. avoid forced shutdowns and physical damage. In other words,it would be highly desirable to be able to detect INTRODUCTION AND BACKGROUND disturbances so that necessary repairs can be scheduled and performed at a time convenient to plant operational From thett ear:lest conception nuclear power plants requirements. Bis objective immediately leads to the have included monitoring and faelt detection systems following two needs: the detection of disturbances as part of their basic design. Howeser. in many cases prior to changes in signal mean values (when mean these systems were of fairly elementary character and values change, physical damage has likely occurred and usually consisted of nothing more than upper and lower shutdown may not be far olT) and the minimization of limits placed upon important sensor measurements that both false and missed alarms (false alarms ultimately when exceeded would alert plant operators. Althougra result in operator indifference and missed alarms can this approach has proved to be reasonably successful, permit continued damage and probable shutdown). considerable advances in methods of signal and Accordingly, since the initial indication of abnormal information processing have permitted the development operation of a piece of equipment usually appears as i of new techniques that not only are much more reliable changes in the characteristics of the noise on sensors y and real. time operable, but also provide additional associated with this equipment, the monitoring method f must utilize such information to detect and identify { validation, extremely early fault detection and disturbances. To achieve this functionality, a model-j functional capabilities such as on-line sensor identification, provision of " virtual" sensors for based monitoring method must utilize a technique that 3 replacement of failed sensors, diagnosis and prognosis not only has a high precision (to minimize error rates) 4 f of disturbances, fault tolerant component operation, but also be easily adaptable to the inevitable insight into the plant operational state. etc. At Argonne configuration and operational changes in the plant. In National Laboratory. such methods were initially addition, the fault detection ponion of the method must developed to address critical operational issues at the identify disturbances based upon changes in statistical lj i Experimental Breeder Reactor No. : Power Plant (such characteristics of the signals as well as the mean values f j as loss ofineplaceable sensors and forced shutdowns with prescribable missed and false alarm probabilities. due to increased pump friction)[1] and then expanded These are challenging requirements. i for use in commercial power plants as well as many ) l l l s 60

i e I J l O

et Now, if a new observation is made and the sensor 3.

if the observation vector is identical to one of the ] measurements from this process represent correlated column vectors in D, then the estimation vector tq must be identical to the observation vector; phenomena, then one can assume an estimate of this vector of measurements can be represented by a linear 1. the error vector (difference between the combination of the column vectors in the data observation and estimation vectors) must be l[ f collection matrix. By minimizing the Euclidean norm minimized. between the estimated and measured data vectors,it is Two such operators that fulfill these conditions have h been found, but cannot be presented here due to patent 4 possible to _obtain the following linearly optimal ] estimate of X: and proprietary issues. Seouential Probability Ratio Test (SPRT)

l-X = D -(67 - I*'A..

o ne sequenti l pr bability ratio test t ]is a statistical f m 5 Howe <er, this result has numerous limitations, hypothesis test that differs from the standard fixed including but not restricted to the requirement that [Dr Sample test in the way,n which statistical observations i l D) be non-singular, an inability to accommodate are employed. In the ftxed sample test, a given number 3 random uncertainties, non-random defects, and very f bservations are used to select one hypothesis from g large databases. These characteristics of real data tw or more alternatives. The SPRT, however, strongly diminish the range of applicability of this sequentially examines one observation at a time, and at 1 classical (linear) estimation technique. s me p int makes a decision and selects a hypothesit j Due to these limitations, it is necessary to examine He basic approach taken by the SPRT technique is i alternative approaches that can be used in non-linear to analyze successive observatics of a discrete systems as weli have robust properties in terms of physical process by a comparison (difference) of the j handling all types of data characteristics. However, the stochastic components of signals generated by two formalism of the linear approach lead to a relationship redundant sensors monitoring the process, or of signals i I between an estimate of the svstem state, a current generated by an actual and simulated sensor menitoring measurement and the system l$istory that has several the process. Details of the use of SPRT in fault very useful features. For example, the model detection were provisd in [6] and are briefly " memory" can be easily expanded bv simply addino summarized here. SPRT solves the problem of q new observation column vectors to the matrix D and deciding between two possible hypotheses: H,, where p the only aspect of the relationship that might be the difference set forTns a Gaussian probability density y computational intensive. the matrix inversion, can be function with mean M (a system disturbance performed "off-line" prior to repeated or on line magnitude) and variance c ; or H., where the 2 estimations. Rese features are of sufficient value and difference set form a Gaussian probability density a utility that, if possible, thev should be retained in any function with mean 0 and variance cr'. If the SPRT d new non linear approuh. 'With this in mind, one can accepts H,, then we declare that one of the two signals assume that the form of the linear estimation equation is degraded. Additional SPRT testing on other pairs of p derived earlier can be used. but with a modification or signals provides the information needed to logically the linear matrix operators to a non-linear form: deduce the specific sensor failure. The index and logic used for deciding upon hypothesis H, or H, is A = 6-(6# a 6I' d6' a A,) a m L = Q [" I Nt ' N1 y, Here, the non-linear operator is at present unknown and j U"k must be chosen so as to preserse the desirable features j of the linear operator and in addition to have the If Lsp/(1-a) then accept hypothesis H, as j following properties: i

true, 1.

the matrix [Dr s Dl must be non-singular; if p/(1-a)sLs(1-p)/a, then continue sampling, j'l 2. if some elements in the observation vector are not if LHl-p)/c). then accept hypothesis H, as j true. j within the ranges of the same elements of the 1( column vectors in the memors matrix D, the where a is the probability of accepting H when H,i-i estimation vector must still represent an optimum true (i.e. the false alarm probabilitv) and p is t: 3 estimation; probability of accepting H, when H, is true (i.e., the d ' missed alarm probability). sEy 62 q

) f 1 J correct value of a failed sensor, although not shown feedwater flowmeter is shown. A total of 60 sensors, j i here, correct values of signals from unfailed sensors all physically associated in some way with the 1 were also made. This is an important characteristic for feedwater flow, were measured at 13 times and used to i fault detection since it reduces the possibility of false construct the memory matrix. These 60 signals were alarms. then monitored over a period of 45 days during which no actual failures occurred; the test was performed by 'u superimposing an artificial drift in only the feedwater flow sensor starting at time zero. The rate of drift was i C 6-0.2% over 50 days which would result in just a 1.0% i s: [ ^j i i change in sensor output after 250 days had passed, a O degradation that would just possibly show up in a 3i! 104 ,} trending analysis and certainly not exceed operational ij .usET g limits. However, the monitoring code detected this drift q initially at 13 days after its start and clearly confirmed j 5 l! SENSOR the fault after about 3 weeks. At this time, the signal U had degraded just 0.084%. Accordingly, if this had E '35 been a real fault, the operators would have had many [ 0 120 240 months to perform any additional hands-ort diagnostics i and repairs and almost certainly not interfered with TIME (m,nutes) plant operation i l Figure 2. Demonstration of Method's Capability l I to Operate with Multiple Failures SUNDIARY AND CONCLUSIONS [ The third type of validation test involved studies of A highly accurate and sensitive model-based the sensitivity of the overall method (state estimation monitoring and fault detection system has been coupled with the SPRT hypothesis test) to detect faults; devel0 Ped and validated with data from an operating this is necessary if sufficient warning time can be sodium-cooled fast reactor and a commercial PWR, provided to permit non forced corrective action. Again, Tne model is based upon a state estimation technique numerc us tests were performed, but a typical result is that utilizes nonlinear optimal filtering pattem shoe.n Figure 3 where data from a commercial PWR ,..ognition and an extended version of the sequential probability ratio test. Extensive validation tests were xs o ss ena o so om j performed in which the prediction accuracy, tolerance j -t of multiple faults and fault detection sensitivity were g d, g demonstrated; selected results from these tests were t I I l--;*gspg l reported here. Additional operational testing of this I method has been perfortned in a PWR and are reported is ia .s o s ie is to 25 n a companion paper in this Conference (2]. t } .\\tSET Estimation Error REFERENCES { ] e s.

1. R. M. Singer, K. C. Gross and R. W. King.

l Je - " Applications of Pattern Recognition Techniques to ? _ __l Online Fault Detection," Proc. First Intl. Machinery 2"" ] N1onitoring and Diagnostics Conf., Las Vegas (1 l- ? 2.s IJ Sept l989)SD 363 5 a s io 5 n as so

2. Kenny C. Gross, et al. " Nuclear Power Plant N1 nit ring and Fault Detection at a PWR," this j

SPRT indices Showing Deersdation ~ conference. i; _ 3

1. P. Herzog, personal communication (Jan.1995).

-{-- ll 4 E. Turkean and o. Cificioglu " Neural Network i 5' Benchmark for S\\ TORN-Vilf' Overview Proc. 7th j -s Symp. Reactor Surveillance and Diagnostics (SNIORN Vili.NENN1SC' DOC 96(l7 Ntay,1996) } e .s a s

o is to 25 r

Dau Aher Surt of Sensor D ift 29-54. 4 3 A Wald, " Sequential Analssis " John Wilev & E .rigure 3. Sensitivity of Fault Detection Ntethod Sons. New York (1947). h to Subtle Disturbances m s 64 k.i

intenig:nt Syst:m Appliettion ta Powsr Syst:ms(ISAP'97) July 6-10,19hf. Mui, KorIa 1a , -ll Application of a Model-based Fault Detection System to Nuclear Plant !l Signals y uR K. C. Gross, R. M. Singer, S. W. Wegerich and J. P. Herzog Argonne National Laboratory f t and [ i R. VanAlstine, F. Bockhorst l Florida Power Corporation 3 U check. The SPRT provides a superior surveillance y ABSTRACT - To assure the continued safe and tool because it is sensitive not only to disturbances in 1 I reliable operation of a nuclear power station, it is signal mean, but also to very subtle changes in the essential that accurate online information on the statistical quality (variance, skewness, bias) of the k current state of the Aire system be available to the monitored signals. I operators. Such information is needed to determine the For slowly evolving degradadon modes (gradual f operability of safety and control systems, the condi. decalibration bias in a sensor, wearout or buildup of g tion of active components, the necessity of preventa. a radial rub in rotating machinery, build-in of a F tive maintenance, and tne status of sensory systems. radiation source in the presence of a noisy background To this end, ANL has developed a new Multivariate signal, loss of time response in a pressure transmitter. State Estimation Tec%nique (MSET) which utilizes etc), the SPRT can provide annunciation of the advanced pattern recognidon methods to enhance incipience or onset of the disturbance long before it g 3 sensor and component operational validation for would be apparent to visual inspection of CRT signal commercial nuclear reactors. Operational data from traces, and well before conventional threshold limit 9 the Crystal River 3 (CR 3) nuclear power plant are checks would be tripped. This permits the operator to ( used to illustrate the high sensitivity, accuracy, and terminate or avoid events that could otherwise result y the rapid response time of MSET for annunciation of in challenges to plant safety margins or system I a variety signal disturbances. availability goals / and, in many cases, to schedule E correedve actions (sensor replacement or recalibration; I Key Words: component adjustment, alignment, or rebalancing; etc.) { Fault detection, signal validation, plant monitor. to be performed during a scheduled plant outage. ing surseillar.ce A companion paper in this Conference (1) reports the theoretical foundations and algorithmic design of a Introduction and Background the MSET method. The purpose of this paper is to / The essence of the MSET method relies upon an report the results of a collaborative effort between } examination of the totality of information available ANL and the Florida Power Corp. to configure and from the array of sensors used to monitor the system test ANL's new MSET system using signals from j and a comparison of these data as a whole to similar Florida Power's Crystal Riser-3 (CR-3) PWR. sets of data collected from the same system operated at various conditions in the past. Based upon this Application to PWR Plant Signals comparison of the current condition of the system ANL and FPC have been engaged in a collabora-l with it's past history. an optimal estimate of the tive study since 1992 to investigate the use of MSET current state of the system is obtained even if there for a variety of surveillance applications that include are errors in the data curTently collected. i.e. some of signal validation. instrument calibration monitoring. the sensors have malfunctioned. and early detecuon of component operability degrada-tion. For the examples reported here, actual plant Hasing an estimate of the true current state of the l l system. the E.fferences between this estimate and the signals were taken from archive opdell disks from the 18 month operating cycle spanning from June of 1992 l current measurements are analyzed using an ex. i tremely sensitive pattern recognition technique, the through Dec. of 1993. sequential probability ratio test (SPRT). for automade annunciation of discrepant signals or the onset of Venturi Flowmeter Surseillance with MSET l degradation in sensors or reactor components. Con. One of the primary objectives of nuclear powet sentional parameter surveillance schemes are sensitive plants is the efficient operation of plant systems. only to gross changes in the process mean. or to large thereby reducing the cost of electricity. Accuratt .~ steps or spikes that exceed some threshold limit determination of the thermal power of the plant ir j i 66 ? lL ta

_n t em.w n iu. t. e. i-,4 v-m. w-o. ca*-i ,r:4 cm. w - w.aca u w m.= = na c m. l i w: "[ ] ii '.U'l ~ I .b i (hll{ ~ j_ W

_\\

y 1 l i h e j"" "3 -lg 1 1 } -*[ ti:

j

+ 1 1 .- [ i l c. ~3,- 3.---- He MSET analysis has shown that the degrada-tion of venturi flow meter measurements due to r a oe.4-w - m--w v.- n iu o.ca*i, fouling of the venturi surface can be detected. He di analysis of the Crystal River-3 data revealed a slow and steady degradation of the flow meter measure-I ments during the 1992/1993 cycle. i* Failure of a Flow Sensor j-... A s second example is of a rapidly failing now "3 -- sensor in CR 3 that was taken from signal archives. De upper subplot in Fig. 5 shows data from sensor 1 R237, the primary loop B flowrate, superimposed j :i 12 } upon the MSET estimate for that signal. His flow "( ,f, i sensor failed (i.e., its output dropped by about 5% in a several hour period) on day 167 in the reactor cycle. '=, .a m - - =

~""

MSET was trained to recognize the normal behavior By the end of the first 6 months of the testing of the system in which this flow sensor was located period, the measured flowTate is about 0.5% larger and then used to monitor the system. It can be seen than the calculated flowrate. A discontinuity is in Fig. 5 that the actual now signal and the MSET evident between the response of the difference signal estimate agreed quite well during the inidal portion of from before the period when the data archival system the monitoring period. This is also indicated by the was taken off line and the response from after the middle subplot. which shows the estimate error.-or the period. Immediately before the MUX was replaced, difference between the measured and estimated the measured Gowrate is about 30 ldbm/hr greater Dowrates. As observed in the upper subplot, the than the calculated flowTate, while immediately after measured and esdmated now values clearly diverge the MUX was replaced, the measured flowrate is after about T=167.5 d (a few percent difference). about 40 kibm/hr less than the calculated flowrate. Indeed, if this signal was being closely watched, titis The discontinuity is due to the recalibration of some failure would likely be detected by visual observadon of the sensors in the feedwater system. When the at this point. However at T=167.1 d. a full 9 hours sensors used as inputs to the model are recalibrated, before the fault would become evident to visual . the model must be retrained since the earlier training observation. MSET starts to alarm (the lower subplot pattems are no longer valid. To illustrate this point, in Fig. 5), indicating that sufficient information has a second calculation was performed in which only been obtained to conarm that a malfunction has data recorded after the MUX replacement were used. occurred. A short time later. the fault is obvious Data from the eleventh month of operadon (i.e., days where the rnessured signal decreases from its initial 300 through 330) were used to train the model in the value of about 72 to about 70 while the estimated { second calculation. In Fig. -1, the difference between signal remains at irs initial value. The signi0cance of the measured and calculated Gowrates for the second the estimated value remaining whe e it does is quite calculation is showTt. He second calculation reveals important; this directly implies that despite the large o a dnft of about 0.7% in the Dow meter measurements change in the signal from the pressure sensor, the dunng the later half of the operating cycle. model has determined from the total collection of { 68 i

4 ig. 7. (!!) RPS-Es Rosem.una Transnuact Time Constant railure l.* N h ~ W b g ri,, s,R1 R:co-R:oi ss in.:2,ms t, s,ibic r.iu,e or Raco o, R o i ( y i= === g 3 5 0 at an:' - - i 4... i.e tie imo izi 122 izs 12 12s ire m Residual Between R:oD and R202 sPRT R2tXbRao2: showing alarms (p ss.ble raslure or Ratt) or R202) h p i ) .f iis m ini a m i2 ins 32. m srRT R:oi_R:oz: showing no alarms (conclude that R2CX) has fated) sPRTIndi:es Showing Failure l d j. ~

q Summary and Conclusions s

s In this paper, a multivariate method (h! SET) has been presented that is capable of monitoring complex systems, detecting and idendfying malfunctions and Fig s. (!!!) RPS-ES Rosemouns Transmazes Time Const. ant Fadun analytically replacing faulted sensors. The method is '( based upon the integration of a data. based model of f ',- /r 5. ~,. the process (utilizing pattern recognition techniques) g * ^ j,,y 'i }; ',;v.N.l [.'i," W '.,,;-Q.l;, fil af ss Vif with a sensitive statistically. based fault de:ection T'[ algorithm that uses hypothesis testing (the SPRT). ~ -'i l2001 Previous work on the modeling aspect of this method l has demonstrated it's robustness to sensor fa.ilures and }l 2 2 s e this paper has presented for the first time a demon-3 E Residual Between R:Ot and RaO2 both sudden and subtle rnalfunctions in sensors f-om i an operating nuclear plant. 3 *! 'l d l References t

d 'I 1.

R. St. Singer. K. C. Gross, J. P. Herzog. S. 3 ~ i' .i, Wegerich, and R. W. King. ".\\todel. based Nuclear Power Plant Ntonitoring and Fault Detection. Theoret. ical Foundations." (this conference). '[ SPRT Ind.c:s Showing no Failars :C:nc:usion. F 7)0 is Failmg)

2. K. Kavaklioglu and B. R. Upadhyaya. "Stonitoring l

Feedwater Flowrate and Component Thermal Perfor. 3 ( l [e-( mance of Pressurized Water Reactors by Steans of Attiticial Neural Networks." Nuclear Technology. Vol. l I g 107. July 1994. pp 112 123. j I-l "[ 3 "Feedwater Row Steasurements in U. S. Nuclear i h j ad r'ower Generation Stations.' TR.101383. Electric A h 8 8 Power Research Institute. Nov.1992. { I x 4 "Rowrate Sti> measurement Causes Unneeded 1 Derating." Nuclear News. Feb.1993. pp. 39-40. j cr -b Ilr 70

L .MSET Applications

1) EBR2 Primary pump monitoring

- Virtual sensor to replace failing flowmeter sensor .. Virtual sensors to drive feedwater control system - 2) Florida Power / Crystal River #3 Off-line sensor validation - ' Off-line venturi flowmeter monitoring On-line complete plant sensor validation (frontend to plant computer) On-line SPDS sensor validation

3) First Energy / Davis Besse Venturi flowmeter monitoring

4) V.C. Summer On-line SPDS sensor validation

5) Private companies using MSET for Cogeneration system optimization Internal combustion engine misfire detection (Ford) l

-. Prediction of S&P 500 average movement Long-term radioactive material monitoring (Lockheed) . 6) NASA experimentally'using MSET to monitor space shuttle main engine l perfonnance during launch

7) ~ ANL using MSET internally for monitoring computer networks

8) University Chicago /ANL has spin off private company to use MSET in non-electrical energy applications (Smart Signals)

9) IIT using MSET as part of effort to abate aircraft engine noise 10)B&W Owns' group selected MSET as their preferred technology for monitoring

11) Negotiation underway with EPRI and AMS with ANL to license MSET i-

.12) MSET selected by R&D Magazine for 1998 R&D 100 award

f On-Line Monitoring of Instrument Channel Performance EPRI TR-104965 Overview of Topical Report September 22,1998 Topical Report Scope- + What is our request? + Technical basis for on-line monitoring + Plant-specific implementation 1

Summary of Request + TR-104965 forms basis for the use of on-line monitoring for extended calibration intervals + For n redundant channels, the channels will be calibrated at a staggered test frequency so that all channels are calibrated every n fuel cycles Summary of Request L + Technical specification change will be l proposed to apply on-line monitoring + Topical report applies to two types of on-line monitoring: - Redundant channel averaging method, such as EPRI ICMP - Pattern recognition method, such as MSET l 2

Technical Basis Topics Summary

Implementation Strategy

+ Functional Requirements + Resolution of Single-Point Monitoring + Transmitter Failure Modes + On-Line Monitoring Uncertainty + Proven Applications ofOn-Line Monitoring Plant-Specific Implementation Summary + Technical Specifications + Impact of Plant Procedures and Documents + Actions Upon Detection of a Drifted Channel + Ongoing Calibration Monitoring Program 3

4 i 1 1 Technical Basis Topics l i i. i Implementation Strategy

Calibration Extension + The use of on-line monitoring is intended to allow calibration extension of safety-related sensors + An unconditional replacement of Technical Specification periodic time-directed calibrations with on-line monitoring only is not proposed by this topical report j l Calibration Frequency + At least one redundant sensor will be calibrated each fuel cycle + Ifidentified as in need of calibration by on-line monitoring, other redundant sensors will also be calibrated + All n redundant safety-related channels for a given parameter will require calibration at least once within n fuel cycles 4 4

t Calibration Frequency + The maximum allowed interval between calibrations is 8 years, regardless of the number of redundant channels + A Technical Specification change will be necessary to extend the calibration interval to the new frequency l l Reason for Calibration Extension Versus Calibration Replacement + At least one redundant sensor will be calibrated each fuel cycle. The purpose of at least one calibration is as follows: - To provide an additional type of assurance that common-mode failure mechanisms do not exist - To ensure that each sensor continues to be periodically calibrated by a method traceable back to a reference standard 6

t Quarterly Verification + On a quarterly basis, a formal surveillance check will be performed to verify that no channels are outside the prescribed alarm limits + A quarterly frequency was established on the basis of engineeringjudgement and is consistent with the Maintenance Rule evaluation frequency On-Line Monitoring Functional Requirements 7

On-Line Monitoring Implementation Supply ne Morntonng Equpment Bounda m h >~~ 2 n'. Transmmer To 250 0 & stable p np Types of On-Line Monitoring + The following options are possible - An automated system that performs data acquisition and analysis essentially continuously at the system specified sample rate - An automated system that performs data acquisition and analysis at discrete specified intervals l l 8

Types of On-Line Monitoring + The following options are possible - An automated system that is normally off and is activated on at least a quarterly interval to perform data acquisition and analysis - A manual system in which data is acquired manually on at least a quarterly interval and entered manually into a computer program for the purpose of analysis SignalIsolation + For safety-related circuits, signals transmitted to the on-line monitoring equipment must be isolated in accordance with the plant's design basis 1 9

Reliability + System reliability is achieved in several phases: - System design and testing - Site acceptance testing - Continued monitoring and verification of operation l Software Validation + The on-line monitoring software is not I considered safety-related, but it is l_ considered quality related l + Software will require formal evaluation in accordance with plant software acceptance procedures l 10 j.

I Single-Point Monitoring Drift Behavior and Characteristics + Explain traditional drift types and compare to observed drift + Show observed proportions of each drift type + Ifinstrument is in calibration at one point, review probability of drift elsewhere in span + Apply results to on-line monitoring s il

Study Scope + Data was acquired from a large number of nuclear plants: - Plants: 18 -Instruments: 1139 - Calibrations: 6700 - AFAL data pairs: 33,890 - Time frame covered: May 75-November 96 Coverage by NSSS Type i:- p.n / \\,

l}fGn

,s rame / \\ l / \\.

i$$

j 3 i \\ ~ 12

NRC Peer Group Sort

==.e ,., -.. ~ wa I

x w! & L-se;y;

== n km ,44 :g,g, e 1 r x \\ \\ / \\ / __=~ yL'- / Instrument Drift Types + Zero shift + Span shift + Nonlinear + Single outlier + Each type can occur alone or in combination with other types 13 t

Traditional Zero Shift As Found Conddion at Recabbration ~ 20 s ,e ,s' W / (enA) p' l s' s' / e / Ongwiel Cahbration y N 1 Pressure input 2 1 PO -OnginalZero PS Onomet Upper Span Lwne j j PO -Zero at Recalibratior. F$g. Upper Span Lwnd f y at Recabbration i Observed Zero Shift ^_ 3% 2% rn 4 1% b k o' a.1% 2% 0% 25 % $0% 75% 100% Cahbrabon Check Posnt (Percent of Span) 14

Forward Span Shift A>Found Contfibnn at Recettambon 20 N :y / / O2put y (mA) / j gg N N 1 1 Pressure IrpA FO, OrgnalZero PS,-Orgnal Upper Span Limit PS -Upper SpiLimit 2 at Recantration Reverse Span Shift Cahbration 20 N.N s' / / Output ,',s (mA) / As-Found Cahbration 4 ..f.'.._ at Recahbrabon PO. PS, PS, Pressure Input PO. - 06ginal Zero PS. Ongnal Span Upper Lmt PS, - Lower Span Lrnit at Reca: brabon 13

Observed Forward Span Shift sur 2% 1% s,- -1% i -2% 3% 0% 25 % 50 % 75 % 100% Cahbraton Check Port (Percert of Span) Observed Reverse Span Shift 4% ' 3% 2% 1% ~ i j g o% s% -1 / a.- l 3% ha 50% 75 % 100 % Cahton Check Pomt (Percent of Span) 16

Forward Span Shift With Zero Shift s% 4% t j2%-- f 4 = i I-N a{ N i 0% 25 % 50 % 75 % 100 % Calerabon Check Pomt (Percent of Span) Reverse Span Shift With Zero Shift 4% 1% 3 0% i $.1% E E g.2% 0% 25% 50 % 75 % 100% Calerabon Check Pomt (Percent of Span) 17 .i

Nonlinear Drift 3% ^- V" 2% 1% g h s 1% 2% .s M 0% 25 % 50 % 75 % 100 % Caltranon Check Poet (Percent of Span) 4 Single Outlier - One Point Out / N, q.i g' N N /

k..

.... tr Io - ~ ~ w 's s e O M M M M Cheek Point Perseen et Spea$ 18

\\ Observed Proportions of Fmach Drift Type + Zero shift and span shift were the most common types of drift + Nonlinear and single outlier cases were uncommon, but did occur Drift Proportions for Out-of-Cals >2% 5% oe ~ ..s.,. _ey a j$ - N NI$ihf$h'k $5 i % i%:n i cwpmu; :- SSi$r[psWkNd$ 'D emwe \\ .- QCghtw \\ ';' 9 = =a d$1hhk

y Forward Sean Shift 19

i Drift Proportions for Out-of-Cals >5% Nonhnear DrtR 5% f' ~~j1ll.+. neverse Seen Wuh -w s. gem sme vn. euww - - -n gl<gjg.9O(M. UdlMihiM/?ylA l =r W ' * )%+ l.> / si:hgCO2.b y,;d,2*~um ,.;p.2,+...w, 4%.-;;;, ,.w-u% i I h Bac. [ MNn$- hi E.fSTh2$c,2.f$. ~'v." i j 4. I s g g .g.g7fy.@.Qq$ / Forware Span wnh \\ 2** 5hN' \\ 'm;Q2..:K'&%$$ 29 % i A.n,. 4,,s m 1 t a. N R.kl( x y u% Drift Proportions for Out-of-Cals >15% One Pows Oes henhneer DNR n 5% ,/ m, *:3.,,k. g

,,~

,p. au" e('#*'!.~he, kI2

T

';(1,hh O$:'NW' 3'% / 9%.iwg;- J S 2.'"". k' I* ,k$h $?,$';?s$f.5 '.y %rg:n.:n;d 23 % l t f-ey:%q w. :t:}1 { / w gy:M; f,i 6.jj%7 'l ~ ;t lt: ' \\

t.i

.r;%.,s.r.n. i, - / \\ /' f .f ,/ y' 1 s N loro sn.n l ' ~ ~.. 8% Fenmerd Spen Sh. 22 % i l l 20

Additional Observations + Zero shift is a contributor to drift in about 75% of the data + Span shift (forward and reverse) is a contributor to drift in about 45% of the data + Nonlinear drift occurred in about 5% of the ) data + Single outlier drift may well be transcription errors Single-Point Monitoring + What is the likelihood of being in calibration at one point in the span (the monitored point) and yet be out of calibration elsewhere in the span? + Zero shift is clearly preferred as the predominant drift type .?] l

Single-Point Monitoring + Span shift means that an instrument is in calibration at one end of span and out-of-cal at the other end of span + Nonlinear and single outlier cases are unpredictable o Analysis Approach + All data was reviewed to establish probabilities of performance based on actual plant data + Data was set up to evaluate 5 calibration check points: 0%,25%,50%,75%, & 100% of span i 1 22

Analysis Approach + Each calibration check point was analyzed separately + The as-found minus as-left (AFAL) setting was used to quantify the drift + A drift limit was defined for the check point + AFAL values outside the drift limit were excluded Analysis Approach + Other check points were compared to the drift limit to determine if they were larger than the specified drift limit by some amount + Probabilities were established based on the frequency of occurrence 23 j

1 4 Results at the 1% Level X' / I! p /l, 5.o%/ 40% ~ 30% \\ 2.0% ' 1.o0% M MIh 1.0% / Calibraton O 5% Check Point oM 25 % o 50 % sog l Calibration Check Point ' 00.0%-10% e 1.052 0% C 2 043 0% 03 044 0% ud 055 0% ! l Explanation of Graph l l + Graph is a surface plot of probability that other check points will be larger than drift limit by 1% or more + X-axis is for each check point + Z-axis is drift limit for check point l + Y-axis is probability of exceeding drift limit l by specified amount l 24

Observations + Note that probabilities drop rapidly as drift limit goes from 0.5% to 1.5% + A drift limit of 1% or larger typically yields very acceptable results + This level of drift is typically less than the assumed drift for a transmitter Results at the 0.5% Level xW G2 M.,,;

v n

N ' I i _FC u- -v = 00% Check Point 25% Om g Caubration Check Point [C00520% e20%d 0% C4 0560% C60540% e8.05100%; 25

\\ l i Results at the 0.75% Level i /gT ! i l ~j j ' Ci j -+--~i 70%r / I / - p. f i j

W W.

I 60% g \\/ 60% }/ \\ ['N ',/' M 5.N '1m Probabinty 3.0%+{, /N q ! x 1.25 % } 2.0%Y -C / 4 1 00% Detft Umit for !/ / ,,j Calibration j 1.0% T O.75% Check Point gak ' i 25% 0.50% m 100 % Calibration Check Point .0 ~ -.,m =2 m c,m... ~ -cs _.. ~ -i E f i Results at the 1.25% Level /./ I I l .*(yI d6 '~ i 1 g 4 I I '\\ 'N f 33 m Probabihty 2 0% ! ..Ydk.4 j Sie w 1.50 % l / h t. ,y, 0% l l Drift Umit for ( 1.00% Calibration 00% i 0 75 % Check Point 25 % . 0 50 % g D 100% Cailbration Check Point ' D0 041.0%. t 0%2 0% C2 0%3.0% 33 064 0% : 26

Results at the 1.5% Level f i : ,,g i y [q \\ l 'r' 20%

i$5; I

Probatsty

q. - g f ~. '

1 50 % 1.0% - O <,_ i$h' 125% g AMJ ' 1.00% Drtft Umit for i...... Cahbrabon o og ' 0.75 % Check Point 75 % Calibradon Check Point L 00510% "$ 0528% C2 0530% ! Application to On-Line i Monitoring l + Span shift occurs often enough that it should be considered + Previous graphs clearly show that l monitoring higher in span has bett-probabilities than monitoring lower in span + Span shift does affect performance at all levels of drift 27

4 Recommended Uncertainty Allowance 14% b ' <25% of Span 2 225%-(50% of Span E O8* 250% - 100% of Span l*" / 04% y o2% I o 50 % o 7s% 100% 125% i 50 % 1 75% 2 00% Drtft Umit for Monitored Channel l l Conclusions + Percentage effect of span shift is small, but - not insignificant + A small uncertainty allowance in the on-line monitoring system acceptance criteria can account for span shift l l + Recommended allowance depends on location in span as well as the specified l drift limit 1 28

4 Transmitter Failure Modes Drift Versus Failure + In terms of on-line monitoring, the reason for the sensor failure is not of particular interest + What is ofinterest to on-line monitoring is what happens to the sensor output signal when the sensor fails 29

1 4 4 Drift Versus Failure + Any failure mode that causes a shift in the sensor's output signal would be detectable just as drift is detectable-the sensor's deviation from the parameter estimate increases + The failures of potential concern are those in which the output signal does not significantly change after failure Drift Versus Failure + Three such cases have been identified for consideration: - Process parameter is at or near the low end of span and the sensor fails low - Process parameter is near the high end of span and the sensor fails high - Process parameter is somewhere between the low and high span limits and the sensor fails as is. 30

t Drift Versus Failure + Sensor failure that causes the output to remain constant somewhere between the 0% and 100% span points regardless of the input was not observed in the calibration ~ data evaluated for this project i + Rosemount loss of fill oil events have been detected by monitoring programs 4 Drift Versus Failure + Failure in which the signal fails high, e.g., l >20 mA, regardless of the input is considered less likely than loss of signal + Even so, few instruments operate at the 100% span point; if they opemte high in the span, there is generally some remaing span in which a high signal failure would be detected as drift 31 4

Drift Versus Failure + Fail low events are considered possible + Following types of applications are potentially susceptible to loss of signal failures: - Auxiliary feedwater flow-there is usually no l flow and the signal is at the bottom of span, such as 4 mA, corresponding to no flow Drift Versus Failure - Engineered safeguards system actuation equipment-the equipment is usually off and the associated pressure or flow indication will be at or near 0% of span - Containment pressure,$epending on the calibrated span, the signal might be about 0% ofspan + On-line monitoring should not be used as a calibration assessment tool for these l applications L 32

Drift Versus Failure + On-line monitoring is proposed as a method to allow calibration extension; it is not proposed here as an unconditional replacement for safety-related calibrations + Periodic calibrations will continue to validate sensor performance and will also identify any unusual sensor failures Drift Versus Failure + Ongoing calibration monitoring program will ensure that sensor performance is evaluated on an ongoing basis 33

4 On-Line Monitoring Uncertainty On-Line Monitoring Uncertainty + Two on-line monitoring approaches are covered by this topical report: - Redundant channel averaging method, such as EPRI ICMP - Pattern recognition method, such as MSET

On-Line Monitoring Uncertainty + For a redundant parameter type averaging algorithm, the uncertainty depends on- - Accuracy of the redundant channels from which the parameter estimate is determined - Ntunber of redundant channels-uncertainty decreases as the number of channels increases On-Line Monitoring Uncertainty + Redundant channel uncertainty (continued): - On-line monitoring algorithm and the method by which it excludes outlying measurements in its calculation of the parameter estimate 35

On-Line Monitoring Uncertainty + MSET Uncertainty: - Accuracy of the measurements used to establish the pattern recognition training set-depends on the calibrated state of the various sensors used to make up the training set - Coverage of plant operating modes by the training set i On-Line Monitoring Applications 1 36 l

Electricit6 de France (EdF) + Has implemented on-line monitoring at all 54 of their nuclear stations as a basis for extending calibration intervals + Implementation has received regulatory approval by the France Safety Authority A i Electricit6 de France (EdF) + At least one redundant channel continues to be calibrated each outage + 8 fuel cycles or 12 years is the maximum allowed time that a sensor can operate without having a traditional calibration + Calibration of at least one channel each outage is intended to assure that common-mode drift effects are not present s +

CANDU Research + On-line monitoring identified drifting channels + Correlated on-line monitoring to calibration results P l CANDU Research 51.0 ,7 N l l i 0.5 ejl ) i,' ,.,t' 5 00 I s I i e i i i

a 45 e

I l * ', + l { O j j,,' 31.0 l l l l I I $ -1.5 l E ,-'l 8 -20 _*Q ' I l I o t a j 'l l l l I l I I $ -2 5 i -2.5 -20 1.5 -10 0.5 00 0.5 1.0 f As-Left Minus As-Found Deviaton From Cahbraton Records l l l 38

EPRI Experience + ICMP has been installed at the following nuclear plants: - Millstone Unit 2 - V. C. Summer - South Texas Project + Upgraded ICMP software is scheduled for installation at Catawba and V. C. Summer MSET Experience + NASA has awarded a grant to adapt MSET for surveillance ofinstrumentation on space shuttle main launch vehicles + Recent application of MSET to safety-of-flight monitoring for the space shuttle main engine (SSME) demonstrated that MSET can significantly enhance the capabilities of the SSME engine control and monitoring system 39

MSET Experience + Private company (under nondisclosure) licensed MSET for energy optimization of cogeneration technologies + License was granted to the Illinois Institute of Technology for use in a collaborative IIT/MIT project for commercial aircraft engine noise abatement l MSET Experience + B&W Owner's Group has selected MSET as the preferred on-line monitoring technology l + Real-time version of MSET has been l l installed in Lockheed's Integrated Testing l and Equipment Laboratory as part of a l demonstration project for long term surveillance of radioactive materials 40

MSET Experience + R&D Journal recently awarded MSET the 1998 R&D-100 Award for one of the top 100 technological inventions in the world for the past year NRC-Funded Research + NRC funded a project to evaluate on-line monitoring methods + Project was conducted over a three-year period and involved both experimental and theoretical work -11

NRC-Funded Research + NUREG/CR-6343, On-Line Testing of Calibration ofProcess Instrumentation Channels in Nuclear Power Plants + Project conclusion: normal outputs of instrument channels can be monitored over a fuel cycle while the plant is operating to determine calibration drift in the field sensors and associated signal conversion and signal conditioning equipment Cross-Calibration of Resistance Temperature Detectors + NRC Branch Technical Position (BTP) H1CB-13, Guidance on Cross-Calibration ofProtection System Resistance Temperature Detectors, provides calibration evaluation guidance that is consistent with the on-line monitoring approach proposed in this topical report 42

Cross-Calibration of Resistance Temperature Detectors + A minimum of one RTD is calibrated each fue! cycle + The approach proposed for on-line monitoring includes a commitment to calibrate at least one redundant sensor each fuel cycle Cross-Calibration of Resistance Temperature Detectors + RTDs included in the cross-correlation program must be shown "to be subject to sufficiently similar temperature and flow conditions in the reactor coolant system" + Proposed on-line monitoring calibration approach will calibrate one redundant sensor for each parameter 43

Cross-Calibration of Resistance Temperature Detectors + Cross-correlated RTDs do not require calibration provided that the cross-correlation results are acceptable + Similarly, on-line monitoring will not require calibration of the remaining redundant transmitters, provided that they i meet the on-line monitoring acceptance criteria Cross-Calibration of Resistance Temperature Detectors + Calibration of one redundant channel each fuel cycle as part of on-line monitoring is similar to RTD approach + But, on-line monitoring version of cross-correlation is performed more frequently. than specified in the BTP for RTDs because quarterly verifications are also performed l l l { l 44

Plant-Specific Implementation 1 i Technical Specifications ) 45

4 Technical Specifications + Proposed approach works with the Standard Technical Specifications + New definition of on-line monitoring is added + New surveillances added for the quarterly verification and the 1 of n calibration each fuel cycle i On-Line Monitoring Definition + ON-LINE MONITORING is the assessment of channel performance and calibration while the channel is operatin;;. ON-LINE MONITORING I differs from CHANNEL CALIBRATION in that i the channel is not adjusted by the process of ON-l LINE MONITORING. Instead, ON-LINE MONITORING compares channel performance to established acceptance criteria to determine if a j CHANNEL CALIBRATION is necessary. 46 i

Staggered Test Basis

A STAGGERED TEST BASIS shall consist of the testing of one of the systems, subsystems, channels, or other designated components during the interval specified by the Surveillance Frequency, so that all systems, subsystems, channels, or other designated components are tested during n Surveillance Frequency intervals, where n is the total number of systems, subsystems, channels, or other designated components in the associated function.

Addition to Staggered Test Basis + Furthermore, for systems, subsystems, channels, or other designated components that are tested by ON-LINE MONITORING, all n systems, subsystems, channels, or other designated components will be tested at a frequency not to exceed 8 years, regardless of the size ofn. -17 i

On-Line Monitoring Surveillances + Define new surveillances: SURVEILLANCE FREQUENCY SR 3.3.1.17 Perform ON LINE MONITORING evaluation. [92) days SR 33.1.18 Perform CHANNEL CALIBRATION [18) months on a STAGGERED TEST BASIS Typical Surveillance Change + Application to Westinghouse STS: Tabic 33.11 Reactor Trip System instrumentauon Appbcable Modes or Other Smitied Required Survedlance Allowale Tnp Functon Conditnins h.ls Cariditsorts %uiremerits Vdue Myt hmm 1(g) [4] M SR 331.1 gl1886) 2l1900} 5R 33.1.7 psig psig sR u m SR 33.1.16 SR 33.1.17 SR 3.11.18 48

Example Implementation + A plant on an 18-month fuel cycle with three redundant instruments for a given parameter would, as a minimum, calibrate at the following frequency: - First channel: 18 months Second channel: 36 months Third channel: 54 months + All redundant channels are calibrated within 41/2 years in this case Example Implementation + A plant on a 24-month fuel cycle with five redundant instruments for a given parameter would, as a minimum, calibrate at the following frequency: - First channel: 2 years Second channel: 4 years Third channel: 6 years Fourth channel: 8 years Fifth channel: 8 years 19

a Plant Procedures and Documents Impact PlantDocuments Affected + Technical Specifications-approval will be necessary to allow longer calibration intervals for specified sensors + Calibration documents-the routine calibration frequency for redundant i channels will be changed from once per i fuel cycle to once per n fuel cycles, where n refers to the number of redundant channels 50

l Plant Documents Affected + Surveillance procedure-a formal procedure will be developed for the quarterly surveillance evaluation by on-line monitoring Plant Documents Affected + Surveillance procedure guidance: - Verify that on-line monitoring is functional - Verify that no monitored channels are operating outside alarm limits-required actions, such as notification of operations or an operability evaluation, should be addressed in the event that alarm limits have been exceeded 31

e 8 Plant Documents Affected + Surveillance procedure guidance: 4 - Verify that current plant conditions are appropriate for the surveillance - For example, plant conditions should not be outside the MSET training limits and process conditions should be stable for the parameters ofinterest - Document completion of the surveillance Plant Documents Affected + Setpoint documents depending on the implementation strategy, setpoint documents might be affected by the on-line monitoring acceptance criteria + On-line monitoring operating procedure, operating manual, or other type of users' guidance will be needed to ensure that users can operate the system $2

Plant Documents Affected + Miscellaneous other plant documents will be affected by the existence and implementation of on-line monitoring + Number of documents will vary based on plant-specific document control systems Actions Upon Detection of Drifted Channel 33

.... _ _ ~ ............-.__.m. .. _ _. _... _ _ _, _ _. ~ _~.._.m.._m.. 1 1-s I e i Evaluation Process /. _ _ _ _ _ _f_ _ Evaluate for OperatzmyCakbrate Devoten 1 Schedule Roubne Cahbraten ______'L______...__.______._ Parameter i Acceptable Region Devtabon i Estwnste Acceptable Regen .__.__L____.___..___.__..__ Schedule Roubne Cahbraten f4egabve a-a ._____1____________________ { Evaluate for OperatuktyCalibrate Acceptable Region + If a given channel remains within the l acceptance band, no calibration action is necessary for the monitored sensor unless that channel was already scheduled for its periodic calibration 1

Q 1 Schedule Routine Calibration + If a channel's deviation exceeds a certain pre-defined limit, calibration will be necessary + Provided that the deviation does not exceed channel operability limits, the urgency of calibration might not be critical + Routine calibration can be scheduled Operability Assessment + If a channel's deviation exceeds a pre-defined acceptance limit, the channel has to be evaluated for operability and corrective actions taken as directed by the Technical Specifications l 1 i r

r O i Ongoing Calibration Monitoring Program l-l Ongoing Monitoring Program + Concept is similar in some respects to the ongoing monitoring program for 2-year fuel cycles as discussed in NRC Generic Letter 91-04, Changes in Technical Specification l Surveillance Intervals to Accommodate a 24-Month Fuel Cycle 1 l 56 l

o Ongoing Monitoring Program + The aspects of an ongoing monitoring program that are ofimportance to on-line monitoring include the following: - Does sensor drift exceed allowable tolerances at the longer calibration interval? - Does the periodic calibration of redundant sensors identify calibration errors that were not detected by on-line monitoring? i On-Line Monitoring Users Group + EPRI will continue to support on-line monitoring + EPRI/ utility working group will continue i 57 l t -}}

ML20154D719

Contents

Text

SUBJECT:

SUMMARY

SUBJECT:

SUMMARY

SUMMARY

Navigation menu

ML20154D719

Text

SUBJECT:

SUMMARY

SUBJECT:

SUMMARY

SUMMARY

Navigation menu

Search