ML20080B563
| ML20080B563 | |
| Person / Time | |
|---|---|
| Site: | Limerick, 05000000 |
| Issue date: | 08/12/1983 |
| From: | Bari R BROOKHAVEN NATIONAL LABORATORY |
| To: | Chelliah E Office of Nuclear Reactor Regulation |
| Shared Package | |
| ML20080B566 | List: |
| References | |
| CON-FIN-A-3393, FOIA-83-408 NUDOCS 8308160340 | |
| Download: ML20080B563 (1) | |
Text
.
BROOKHAVEN NATIONAL LABORATORY
~
'~
ASSOCIATED UNIVERSITIES, INC.
Upton, Long Island. New York 11973 (516) 282' 2629 FTS 666' go - 3 L' }
August 12, 1983 Dr. Erulappa Chelliah Reliability and Risk Assessment Branch U. S. Nuclear Regulatory Commission Washington, D. C.
20555
Dear Erul:
Enclosed herewith are five copies of the draft BNL report "A Prelimi-nary Review of the Limerick Generating Station Severe Accident Risk Assess-ment". This report satisfies the milestone for Task I of Project 3 on FIN A-3393.
We look forward to your comments on this report.
Warm regards, Yf Robert A. Bari Associate Chairman and Division Head Engineering and Risk Assessment RAB/nn Enclosures cc:
W. Y. Kato R. E. Hall I. A. Papazoglou A. Thadani F. Coffman XA Copy Has Been Sent.to.PDR 13%L%bf XA dS 2 1
~.
)
' 4 y NUREG/CR-BNL-NURE'i-Gy%$
A PRELIMINARY REVIEW OF THE LIi4ERICK GENERATING ST SEVERE ACCIDENT RISK ASSESSMENT VOLUME I:
Care Melt Frecuency M. A. Azam, R. A. Sari, J. L. Soccio, N. Manan, I. A. Papazoglou, C. Ruger, K. Sniu, J. Reec*, M. McCann*, A. Kafka **
Engineering and Risk Assessment Division Department of Nuclear Energy Brookhaven National Laboratory Upton, New York 11973 August 15, 1983 (DRAFT)
- Jack R. Benjamin Associates, Inc.
- Soston College 9
%)
a,E y P
f A
\\
y-e e-
-iii.
~
ABSTRACT A preliminary review is performed of tne Severe Accia' enc Risk Assessment for tne Limerick Generating Station.
... review consicers tne iacact on :ne core melt frequency of seismic and fire initiating events. An evaluation is performed of methodologies used for determining tne event frequencies and their impact on the plant components and structures.. Particular attention is given tc uncertainties and critical assumptions. Limited requantification is per-formed for selected core melt accident sequences in order to illustrate sensi-tivities of the results to the underlying assumptions.
k 4
m-
-n-e y
-.-------,ry
-,,-7-.
,m-r
--w w
- 3 r-e,-
8 4..
/)
-iv-TABLE OF CONTENTS Page ABSTRACT................................................................
iii LIST OF FIGURES.........................................................
L I S T O F TA B L E S..........................................................
SUMMARY
l'. 0 INTRODUCTION.......................................................
1-1
1.1 Background
1-1 1.2 Objective, Scope, and Approach to Review......................
1-1 1.3 Organization of Report........................................
1-2 2.0 EXTERNAL INITIATING EVENT CONTRIBUTORS.............................
2-1 2.1 Review of tne Seismic Hazard and Fragility Analyses...........
2-1 2.1.1 Introduction...........................................
2-1 2.1.1.1 Sensitivity Analysis for Seismic Effects......
2-4 2.1.1.2 Sei smi c Secti on Organi zati on..................
2-6 2.1.2 Seismic Hazard.........................................
2-6 2.1.2.1 Review Approach...............................
2-6 2.1.2.2 Seismic Hazard Metnodology....................
2-6 2.1.2.3 Seismogenic Zones.............................
2-9 2.1.2.4 Sei smi ci ty Pa rameters.........................
2-14 2.1.2.5 Grouna Motion Attenuation..................... 2-17 2.1.2.6 Comparison of the LGS Hazard Analysis with Hi sto ri c Sei smi ci ty........................... 2-18 2.1.2.7 S u mma ry.......................................
2-21 2.1.3 Sei s mi c Fra g i l i ty.......................................
2-23 2.1.3.1 Damage Factor.................................
2-24 2.1.3.2 Upper Bound Accelerations..................... 2-30 2.1.3.3 Reactor Enclosure and Control Structure.......
2-31 2.1.3.4 Reactor Pressure Vessel Capacities............ 3-32 2.1.3.5 Potential Impact Between Reactor Building and Containment...................................
2-33 2.1.3.6 Electri c and Cont rol Equi pment................
2-36
~
2.1.3.7 Review of Significant Components.............. 2-37 J
4 l
em t.
\\
-v-TABLE OF CONTENTS (Cont. )
Pace 2.1.3.8 General Fragili ty-Related Comments............
2*5 l
2.1.3.9 Closure....................................... 2 43 2.1.4 References to Section 2.1.............................. 2 19 2.2 Fire..........................................................
2-58 2.2.1 Deterministic Fire Growtn Modeling..................... 2-53 2.2.1.1 Introduction.................................. 2-58 2.2.1.2 Summary Evaluation of Deterministic Fire Growtn Modeling............................... 2-60 2.2.1.3 Detailed Evaluation of Deterministic Fire Growtn hadaling............................... 2-62 2.2.1.3.1 Fuel Burning Rate.................. 2-62 2.2.1.3.2 Fu el Element Igniti on.............. 2 64 2.2.1.3.3 Fire Near Enclosure Walls or Corners............................ 2-68 2.2.1.3.4 Stratified Ceiling Layer........... 2-68 2.2.1.4 Recommendations for Improving Fire Growtn Modeling......................................
2-69 2.2.2 Procaoi l i sti c Fi re Analysi s Revi ew..................... 2-70 2.2.2.1 Evaluation of Significant Fire Frequencies in l
General Locations............................. 2-71 l
2.2.2.1.1 Sel f-Ignited Cable Fi res........... 2-72 2.2.2.1.2 Transient Combusti bl e Fi res........ 2-73 2.2.2.1.3 Power Distribution Panel Fires..... 2-73 2.2.2.2 Screening Analysis............................ 2-73 2.2.2.3 Probabilistic Modeling of Detection and Suppression................................... 2-74 2.2.2.4 Procabilistic Modeling of Plant Damage State.. 2-75 i
l 2.2.2.4.1 Zone Speci fi c Comments............. 2-78 r
l 2.2.3 References to Section 2.2..............................
2-80 l
l 3.0 ACCIDENT SEQUENCE ANALYSIS.........................................
3-1 i
3.1 Seismic....................................................... 3-1 3.1.1 Plant Frontline Systems................................ 3-1 l
I l
(-
y 7
l
-vi-TABLE OF CONTENTS (Cont. )
Page 3.1.1.1 Overview of the LGS-SARA Approa:n in Frontline System Modeling...............................
3-1 3.1.1.2 BNL Revisicn and Review of Frontline System Fault Trees...................................
3-2 3.1.2 Accident Sequence Analysis.............................
3-13 3.1.2.1 Overview of LGS-SARA Accident Sequence Analysis...................................... 3-13 3.1.2.2 BNL Review of Accident Sequence Quantifica-tion..........................................
3-12 3.2 Fire..........................................................
3-32 3.2.1 Overview of tne LGS-SARA Accident Sequence Quantifica-tion................................................... 3-32 3.2.2 BNL Revisions in Quantification of Accident Sequences..
3-33 3.2.2.1 Fire Zone 2:
13 kV Switchgear Rocm.......... 3-34 3.2.2.2 Fire Zone 25: Auxiliary Equipment Room....... 3-37 3.2.2.3 Fire Growth Event Trees for Fire Zones 20, 22, 24, 44, 45, and 47............................ 3-39 3.2.3 Review Results......................................... 3-40 3.3 References to Section 3....................................... 3-47 4.0 SOME GENERAL ISSUES AND SPECIFIC RECOMMENDATIONS...................
4-1 4.1 Seismic Hazard and Fragility Recommendations..................
4-1 4.1.1 Introduction...........................................
4-1 4.1.2 Seismic Hazard.........................................
4-1 4.1.3 Seismic Fragility...................................... 4-2 4.2 Fire..........................................................
4-9 APPENDIX A: Detailed Review of the Quantification of the Fire-Growth Event Trees................................................ A-1 APPENDIX B: Report of Professor Alan L. Xafka: A Critique of " Seismic Ground Motion at Limerick Generating Station," by ERTEC Rocky Mountain, Inc........................................ B-1
~
S e
e
t.
\\>:!
t
SUMMARY
Overall, tne Severe Accident Risk Assessment (SARA) for the Limerick Gen-erating Station appears to use state-of-the-art methodologies for evaluation of
~
the core melt frequency due to seismic and fire initiating events. These re-sults are useful in a relative sense and should not be viewea as absolute nuccers. The'autnors of SARA are well aware of tne uncertaintias associatec witn analyses of these. events and provide discussions of the najor contributcrs to uncertainties. However, it should be noted that the prescriptions used for quantifying uncertainties in the SARA are themselves somewnat arbitrary.
t The proceaure used to quantify seismic risk is cased on simcle procabil-istic models wnic5 use some data, but currently rely neavily on engineering judgement. The analysis dces not include a comprenensive consiceration of design and construction errors and, hence, may be (conservatively or non-conservatively) biased.
The metnod used for estimating tne probacility distribution on frequency of exceedance for the seismic hazard is a well-established, straigntforward i
l approach and is considered appropriate. With regard to the application of tnis metnod, it is not well defined by the coarse sampling of parameter nypotneses used in SARA.
In addition, specific concerns are raised with regara to tne definition and selection of seismogenic zones and to the assignment of seis-micity parameters.
It was judged that the various issues raised witn regard to tne seismic hazard analysis would individually have a small impact (less than a factor of two) on the mean value of the seismic-induced core melt frequency, but that the total impact could be moderate (less than a factor of ten).
The seismic fragility analysis also was found to be reasonably witnin the state-of-the-art, but specific questions are raised with regard to the justi-fication for tne fragility values of various components and structures.
4 Simple audit calculations were performed'in an attempt to replicate the results given in the SARA for the mean frequency of seismic-induced core melt from dominant accident sequences. The simple calculations were generally in 1
e
., _ -.., _ - _. _ _, _ _ _.,. _ _.. _ _ _,.. ~ -
c lI
SUMMARY
(Cont.)
good. agreement with tne SARA results. However, a closer examination of the event trees suggests that tne Boolean algebra was not executed properly and
- that the seismic-SARA quantification procedure is not correct. While tne effer
- this apparent error is small (about a factor of two reduction) for the total core melt frequency due to seismic events, it aces alter tne relative contributions from the seismic failure of components and structures. Namely, SARA-determined that tne mean frequency of core melt is dominated ay five electrical components in series, wnich have nearly the same median capacities.
On the other hand, if corrections are made to the Boolean equations, tner it is found that tne mean frequency of core melt is dcminated by contributions feca structural failures ratner than electrical component failures.
~ ll resciution u
of tnese issues would require further analyses.
In tne analysis of the core melt frequency due to plant fires, tne SARA employs state-of-tne-art technology for the determination of fire growth, de-tection, and suppression.
In addition, tne impact of fires on plant systems is witnin the current state-of-the-art.
It was found that the analysis was con-servative in many aspects, but that tnis is in keeping with current methodol-ogies in tnis difficult area whicn is fraugnt with large uncertainties. Addi-tionally, it was found tnat some of the analysis, particularly tne determin-istic fire growth modeling, has unphysical aspects which may be either conser-vative or nonconservative. Based on the foregoing, the reviewers believe that it would be difficult to quan-ify the effect of these uncertainties, particu-larly as they relate to probabiltstic analyses.
The approach taken on the fire analysis to the identification of critical plant areas is sound and the analysis appears to have identified all of these areas. However, in some cases, critical components, cabling, and layout of panels were not properly identified. The data base adopted for estimating the fire frequency is appropriate, but in some cases the specific estimates appear to be incorrect. The cumulative fire suppression distribution function J
my-m
<---e r+
a
SUMMARY
(Cont. )
generated in tne SARA does not seem to agree with available data. BNL cotained a distribution fit (Weibull) to tne appropriate data base and thereby generated a cumulative distribution wnicn, for an" jiven time, yields a lower probability of fire suppression than the corresponding SARA results.
Based on the review of prooabilistic aspects of fire initiation, growta, and suppression, a limited recuantification was performed of the fire-incuced core melt frequency.~ An overall factor of (approximately) tsa increase in ne fire-incuced core melt frequency was estimated and tnis is attributed to dif-ferences in 1) the probability of fire suppression at any given time and 2) in tne frecuency of self-ignited cable-raceway fires.
A. major contrioution to the core melt frequency comes from tne stage of fire growtn in wnicn all safe-snutcown systems are assumed to be damaged and faulted by the firs. Sensitiv-ity studies were performed to examine each of tnese. contributors separately and tney were found to be equally important. Sensitivity studies were performed with regard to operator error and it was found that the fire-induced core melt frequency was not very sensitive to (one order of magnitude) cnanges in 1) the failure of tne acerator to depressurize tne reactor in a required, timely fasnion or 2) in :ne failure of tne operator to initiate required systems from a remote snutdown panel.
In tne main text, this report contains recommendations for furtner work and information requirements in the seis.nic and fire areas wnicn would oe help-ful in assessing these risks at tne Limerick plant.
e
1-1
1.0 INTRODUCTION
1.1 Background
In February 1983, Bracknaven National. Laboratory (3NL) issued a re-port (1) (NUREG/CR-3028) on its review of the probabilistic risk assess-ment (2) for tne Limerick Generating Station (LGS-PRA). Tne LGS-PRA exclucec seismic events, fires, tornadoes, nurricanes, floccs, and sabotage fecm tne s&t of initiating events (internal events) tnat it considered.
In April 1983, Philadelpnia Electric Company (PECo) completed a study wnica included the evaluation of risk due to seismic initiating events and to fires tnat mignt be initiated within :ne plant. Inis study, the Severe Accident Risx Assessment for the Limerick Generating Station (LGS-SARA), also included a revised an-alysis cf the offsite consequence analysis witn the CRAC2 computer ccde.
In June 1983, NRC requested tnat BNL undertake a preliminary, short term review of the LGS-SARA. The review will be contained in a two-volume report.
The present document is a draft of Volume I, whicn reports tne review of seis-mic and fire methodologies as they relate to the determination of tne core melt" frequency. Volume II will report the review of tne analysis of the core melt pnenomenology, fission product benavior, and offsite consequences and will be issued at a later date.
1.2 Objective, Scoce, and Acoroacn to Review The coJective of this work, as reported in Volume I, is to perform a pre-liminary review of the LGS-SARA including consideration of tne core melt fre-quency. Inis includes an evaluation of the appropriateness of the overall methodology used to identify structures and components damaged and faulted due to seismic events and fires and a comparison of PECo's methodology with current state-of-the-art approaches.
In particular, this work reviews PEco's estinates of: the occurrence frequency of ground motion acceleration and the fragility analysis of structures and components damaged during seismic events; and the frequency of significant fires and the conditional failure probacilities of e
- Tne concept of core melt frequency used here and in the LGS-SARA is equivalent to the concept of core damage frequency used in NUREG/CR-3028 (and in some places in tne present report).
=. -
I 1-2 mitigating systems damaged and. faulted during the fire.
Finally, a deter-mination is made of the influence of the findings of this review on the prediction of the core melt frequency as calculated in the LGS-SARA.
It is noted at tnis point, that the determination of the impact of the findings dn the core melt frequency is qualitative in scme places and, at best, se'ni-cualitative in otner places.
In general, major uncertainties in tne analysis are nignitgnted, subjective notions are identified, and limited re-calculations are done to focus concerns and indicate sensitivities. A more detailed, quantitative reevaluation of the core melt frequency due to seismic events and to fires would te a more time extensive, re' source intensive enter-pri se.
Tnis preliminary review was conducted over a two montn period by BriL with tne assistance of Jack R. Benjamin Associates, Inc. (JBA) for the seismic por-tions of tne review. The BNL reviewers included J. L. Boccio (overall fire nazard and vulnerability review), M. A. Azurm (probabilistic fire modeling),
C. Ruger (deterministic fire modeling), I. A. Papazoglou (overall systems / core melt review), N. Hanan (fire / core melt review), and K. Shiu (seismic / core melt review).- The JBA reviewers included J. Reed (overall seismic hazard and fragility review) and M. McCann (seismic hazard review). Finally, JBA subcon-tracted witn Professor A. Kafka of Boston College for a review of the seismic hazard analysis from a seismologists viewpoint. The overall review contained in Volumes I and II was coordinated by R. A. Bari of BNL.
The review process was facilitated by several discussions and meetings held between BNL, NRC and PECo and its consultants (notably NUS Corporation and Structural Mechanics Associates). BNL and JBA reviewers visited the Limerick site on July 15, 1983 in order to obtain direct plant configuration information for the seismic and fire reviews.
I 1.3 Organization of Report Section 2.1 contains a review of the seismic hazard and fragility analy-ses. Section 2.2 contains a review of both the deterministic and probabilistic aspects of fire growth and suppression analyses. Section 3.1 contains a review
)
_.e_,s.
- * ^ '
+v
+-
1
-ty,-.,
, y,.-*-
,.-,...--,e,w.4 e-,--,cw r~~+re,-
,-a
-*m.
- .,r e..- -,,.. - -
.__.._m___.
1-3 of tne core melt sequence analysis related to seismic events. Similarly, Sec-tion 3.2 contains a review of the core melt sequence analysis relating to fire events. Sections 3.1 and 3.2 rely on information developed in Sections 2.1 and 2.2, respectively. Section 4 contains a discussion of general issues and scecific recommendations based on tnis review.
- tote nat all references are proviced locally in tne corresconcing sections cr sucsections.
Y e
2.1 REVIEW OF THE SEISMIC HAZARD AND FRAGILITY ANALYSES 2.1.1 Introduction Jack R. Benjamin and Associates, Inc. (JBA) was retained by SNL to perform a preliminary review of the LGS-SARA for the effects of seismic
~
events. The following sections of the LGS-SARA were the principal focus of the review by JBA:
Appendix A:
Seismic Ground Motion Hazard at Limerick Generating Station Appendix B:
Conditional Probabilities of Seismic-Induced Failure for Structures and Components for the Limerick Gener-ating Station.
Also included in JBA's review was applicable information in Chapter 3 and Appendix C.
Jack R. Benjamin and Associates, Inc., has performed similar reviews of the Indian Point Probabilistic Safety Study (IPPSS)Ill and the Zion Probabilistic Safety Study (ZPSS).I2I (See Reference 3 for the Indian Point review. The Zion review has not been published.) The review of the LGS-SARA focused on the critical issues which may signifi-cantly impact the results. Based on the experience gained from the IPPSS and ZPSS reviews, a preliminary review 'of the LGS-SARA was con-ducted in a short time period in order to discover the critical issues and to make recommendations to address those issues which remain unre-solved. In contrast to the previous reviews which consisted of an in-depth evaluation of each section and subsection of the PRA reports, this 2-1 e,...-,
review focused primarily on critical areas which' may impact the results. Since both the hazard and fragility calculations for the LGS-SARA were periormed by the same engineers and were based on the identi-cal methodologies used for the IPPSS and ZPSS, many of the issues and concerns generic to all sites and plants already have been discussed and evaluated.(3) This review documents the important concerns applicable to the Limerick plant.
The reader is directed to Reference 3 which provides a general point-by-ocint discussion of the seismic risk rrethod-ologies used in PRA studies submitted to the NRC to date. Differences between the current study for Limerick and the IPPSS and ZPSS reports ire discussed in this report.
In the review of the LGS-SARA, JBA assumed that the Boolean equa-tions for the sequencies leading to core melt are correct. The review performed by the BNL reviewers addressed the adequacy of the event and fault trees, random equipment failures, operator errors, and resulting Boolean equations. The discussion concerning potential discrepancies i
for these issues is given in Section 3.1.2.
As part of the review a meeting was held at the Structural Mech-anics' Associates (SMA) office in Newport Beach, California, on 8 July 1983.
Dr. Robin McGuire of Dames and Moore, who performed the seismic hazard analysis while employed by Ertec Rocky Mountain, Inc.; SMA, who j
conducted the fragility analysis; and NUS met with Dr. John W. Reed and Dr. Martin W. McCann of JBA along with representatives from the NRC.
The purpose of the meeting was to discuss issues raised to date concern-ing the LGS-SARA and to focus the review effort on the critical compon-ents and issues. Subsequent to this meeting a tour of the Limerick plant was conducted on 15 July 1983. Based on these events, review of the LGS-SARA, and discussions with the NRC, Sections 2.1 and 4.1 of this report were prepared by JBA.
In the review, an attempt is made to look for both conservative and unconservative assumptions which could signficantly impact the 2-2 e
~
results.
In order to help the reader, an effort is made to indicate, where possible, the ultimate impact of the issues which have been raised.
Comments are primarily directed to the mean freouency of core melt or to the individual secuence's which contribute significantly to core melt. Where pass.fble, the imoact of the issues raised on the median frecuency of core melt is indicated.
The folicwing scale has been adopted to cuantify comments made in the review of the LGS-SARA:
Effect on Mean Freouency Comment of Coce Mel t Small Factor < 2 Moderate 2 < Factor < 10 Large Factor > 10 The methodology used in the LGS-SARA for seismic effects is appro-priate and adequate to obtain a rational measure of the probability distribution of the frequency of core melt.
The results from the LGS-SARA are useful in a relative sense and should not be viewed as absolute numbers. The procedure used to cuantify seismic risk is based on simple probabilictic models which use some data, but currently rely heavily on engineering judgment.
The analysis does not include a comprehensive consideration of design and construction errors and, hence, may be.
biased (note that errors may be either conservative or unconserva-tive). Because of the newness of these types of analyses and the limi-tations pointed out above, the results are useful only in making rela-tive comparisons. Although more sophisticated analytical models exist, the limitation of available data dictate that the simple models used in the LGS-SARA are in a practical sense at the level of the state-of-the-art.
l 2-3
2.1.1.1 Sensitivity Analysis for Seismic Effects The approach used by NUS to combine the hazard and fragility curves is different from the rethod used by Pichard, Lowe, and Garrick (PLG) for the IPPSS and ZPSS.
In the PLG method a discrete probability distribution (DPD) approach was used to systematically account for the variability (i.e., randomness and uncertainty) in the hazard and frag 11-ity parameters. Sequences were ccmbined to form the final Boolean equations for core melt and the various release categories. System fragility data for core melt or the release categories were obtained and provided in the PLG reports for Zion and Indian Point. The combination of the system frag'ility curves and the hazard curves were performed directly using numerical integration.
In contrast, the NUS approach differs frcm the PLG methodology in two respects. First, NUS included the potential for random equipment failures and operator errors in the seismic event fault trees. Second, they used Monte Carlo simulation instead of the DPD approach adopted by PLG.
It appears, based on a preliminary review, that random equipment failure and operator errors have a small effect on the mean frequency of core melt, but may have a moderate effect on the median frequency of core melt relative to the case where only seismic contributions are included.
As part of the preliminary review,'an attempt was made to replicate the results given by NUS for the mean frequency of core melt as contri-buted by the significant sequences. This exercise also provided a basis for detennining the possible changes which differences of opinion could produce on the mean frequency of core melt. The procedure used was based on the component fragility curves represented by their median values and combined variabilities (i.e., the randemness and uncertainty logarithmic standard deviations were combined).
In addition, mean values for the random equipment failure and operator error events were assumed. This approach is approximate, but gives reasonable results for mean frequency values.
7 2-4
~
The fragilities for the components in each of the sequences which were considered to contribute significantly to the mean frequency of core melt, were combined according to the Boolean equations and inte-grated with the hazard curves. Table 2.1 gives the comparison between the approximate values calculated as described above and the values reported in the LGS-SARA.
In general, the approximate results ccmpare reasonably well with the values given in the LGS-SARA. The calculated mean frequ'ency of core melt is 6.5-6* and is witnin 15 cercent of the LGS-SARA value of 5.7-6.
The maximum di fference for' individual sequences is 150 percent, which is a small effect. However, the differ-ence for sequence T RPV, which consists of a single component (i.e., the s
R?V), is aporoximately 50 percent.
It was surprising that the calculated value was relatively different as compared to the LGS-SARA reported value (i.e., 4.4-7 compared to 8.0-7).
Table 2.2 gives the breakdown of the mean frequency of core melt contributed by the various hazard curves. Over 86 percent is contri-buted by the Decollement and the Piedmont, Mmax = 6.3 hazard curves, with the Decollement contributing slightly more. The Northeast Tectonic hypotheses, which is weighted by a probability of 0.3 in the LGS-SARA, centributes only about 4 percent.
Table 2.3 considers the hypothetical case that only one hazard curve exists and gives the value for mean frequency of core melt assum-ing that only one hazard curve is possible (i.e., probability weight is 1.0).
This assumption is made independently for each of the six hazard hypotheses and the corresponding mean frequency of core melt values are given in Table 2.3 along with the ratios of values compared to the case where the curves are weighted as assumed in the LGS-SARA.
It is inter-e: ting to note that if the Decollement is the oaly hazard curve, the mean frequency of core melt will only increase by a factor of 4.6, which
- 6. 5-6= 6. 5x 10-6 2-5 i
r l
l L
The comparisons given in Tables 2.1, 2.2 and 2.3 give an indication of the potential sensitivity of the mean frequency of core melt to changes in the contributions from the different sequences and hazard curves.
2.1.1.2 Seismic Section Organization Section 2.1.2 presents the results of the review of the seismi hazard analysis, while Section 2.1.3 gives the review of the fragility analysis. Recommendations for actions to address the signif: int unre-solved issues are presented in Section 4.1 of Chapter 4.
2.1.2 Seismic Hazard 2.1.2.1 Review Aporoach A critical review of Appendix A of the LGS-SARA, which describes the methodology and analysis of the earthquake ground motion hazard at the Limerick site, was conducted. Section 3.3.1 of the LGS-SARA summar-izes the methodology and the results of the probabilistic seismic hazard analysis which is provided in Appendix A.
To assist in the review, the services of a consultant, Professor Alan L. Kafka, were retained by JBA to review Appendix A from the seismologist's viewpoint. Professor Kafka's report is provided in Appendix B to-this review, while important points are incorporated in this body of this report.
The review of the seismic hazard analysis in the LGS-SARA has concentrated on a number of issues. To begin, the adequacy and appro-priateness of the overall probabilistic methodology to estimate the frequency of ground motion is considered in Section 2.1.2.2.
-Individual elements of the seismic hazard analysis: seismogenic zones, seismicity parameters, and the ground motion attenuation are reviewed in Sections 2.1.2.3-5, respectively.
2-6 1
-p g.,
w
I In Section 2.1.2.6, a preliminary assessment of overall reasonable-ness and accuracy of the LGS-SARA hazard curves is made through a comparison with results derived from the historic site intensity data.
A qualitative summary of the preliminary review of the seismic hazard analysis is given in Section 2.1.2.7.
As discussed previously in Section 2.1.1, the impact of cccments on the mean frequency of core melt is assessed in a qualitative manner.
In addition the influence they have on the estimate of the site seismicity curves is also indicated where possible.
t 2.1.2.2 Seismic Hazard Methodoloay The approach used in the LGS-SARA seismic hazard analysis is well established and considered appropriate to estimate the frequency of ground shaking. W 5) The analysis consists of two basic elements. The first step involves establishing hypotheses, to model the seismicity in the tectonic vicinity of the site and the ground motion associated with seismic events. Hypotheses are established to consider reasonable models of seismogenic zones, estimates of seismicity parameters (i.e.,
maximum magnitudes, b-values, etc.) and ground motion attenuation. For the most part, expert opinion is the principal basis for establishing the hypotheses used in the LGS-SARA. Associated with each hypothesis is a probability value that expresses the degree-of-belief that a given set of parameters is the "true" representation of the site seismicity.
The second step in the analysis involves the calculation of the annual frequency that levels of ground motion will be exceeded at the site. This step is performed for each seismogenic zone hypothesis and the suite of likely parameter values (i.e., activity rates, b-values, maximum magnitudes, etc.). The final product of this analysis is a family of seismicity curves, each having a discrete probability value associated with it. The discrete probability values sum to one, imply-ing that a complete probability distribution on the annual frequency of exceedance'has been derived.
=
2-7 y a o
n
-s_.e._,,
2,.7.._
The application of this approach in the LGS-SARA is appropriata to estimate the seismic hazard at the plant site. With regard to adequacy, the application does not insure that the probability distribution on frequency has been completely defined.
In the LGS-SARA study, an impli-cit decision was made only to consider those hypotheses for seismogenic zones, sources parameters, etc., that had a major influence on the
~ estimate of the frequency of occurrence. That is, of the many reason-able hypotheses that could be considered to estimate the ground motion hazard at Limerick, a relatively small sample was selected.
In a sense, a filtering of the various parameter sets that could be included in the analysis was made. The consequences of this approach depend on the randem process being considered. However, the consequences include the possibility that the probability distribution on frequency is defined by a coarse set of discrete probability values. Further, depending on the manner in which the hypotheses are selected, the tails of the prob-ability distribution on the annual frequency of exceedance may be poorly defined.
The approach used in the I.GS-SARA presuppose *, that the analyst, in consultation with a seismologist, can adequately sample the space of alternate hypotheses, such that the probability distribution on frequen-cy is adequately defined. Although the influence of individual para-meters can be reasonably estimated prior to performing the analysis, it is generally not true that the analyst can select a set of hypotheses that will adequately define the probability distribution on frequency over its entire range.
In the LGS-SARA, six discrete probability values are used to define the distribution on frequency, which generally ranges over one or more orders of magnitude. This is not to suggest that a discrete representa-tion of such a wide distribution by 6-10 points is not adequate.
'Certainly, if the entire distribution were known and the points were selected in a prudent manner, this may be reassnable. However, in the 2-8 s
LGS-SARA, six hypotheses and their discrete probability values were selected beforehand wi.thout knowledge of their counterpart result on the probability distribution on frequency. The solution to this issue is simple; a more complete sampling of the possible model hypotheses and distributions of individual parameters is needed. Specific examples where this could be achieved in the LGS-SAAA are discussed in the sec-tiens which follow.
In regards to the importance of having an adequate representation of the probability distribution on the frequency of exceedance, one point should be considered. A reliable representation of the proba-bility distribution on seismic risk (i.e., core melt), is determined for the most part by the hazard analysis. That is, both the order of magnitude of the results and the uncertainty are dominated by the probability distribution on the frequency of ground motion. For new plants such as Limerick, this issue becomes more important because the tails of the seismic hazard curves, which are even more uncertain, determine the estimate of seismic risk.
If the seismic hazard analysis does not adequately represent the probability distribution on frequency, results based on it may be jeopardized.
It should also be pointed out that in terms of estimating the mean frequency of core melt, the LGS-SARA results may not be influenced by the above comments. However, if the entire distribution on the frequen-cy of core melt is of concern, then these comments are more important.
2.1.2.3 Seismogenic Zones To'model the seismic hazard at the LGS site, four hypotheses on the tectonic origin of earthquakes in the plant vicinity were defined. The definition of the different seismogenic zones is based in part on geo-logic, geophysical, and seismic data and expert opinion. Seismicity parameters are then estimated for each zone. On the basis of expert opinion, the Piedmont, Northeast Tectonic, and Crustal Block zones were
=
2-9 r ----
r w-
assigned probability weights of 0.30, and the Decollement hypothesis was assigned a 0.10 weight. Major concerns with the zonation used in the LGS-SARA are discussed.
As described in the LGS-SARA, the Crustal Block hypothesis attempts to account for the occurrence of earthquakes in the northeast by the movement along the boundaries of large blocks of the earth's crust.
It is assumed that earthquakes occur along block boundaries while the interior areas are relatively quiet.
In the LGS-SARA, eight zones maka up the Crustal Block hypothesis (see Figure 2.1).
Of these, Zone 8 is the dominant contributor to the hazard at the site. This hypothesis is questioned on two accounts. First, while the principle that large blocks of the earth's crust may control the seismicity in the region along their boundaries is reasonable, such a theory should correlate reasonably well with historic and instrumentally located seismicity.
In general, this is not the case (see Figure 2.1).
As stated previously, Zone 8 is reported to have the greatest contribution to the site hazard. A review of Figure 2.1 indicates that the closest proximity of Zone 8 to the LGS-SARA site is approximately -
30-40 miles. This fact alcrie explains to a large extent why the hazard curves derived for the Crustal Block hypothesis produced the lowest frequencies.
It is furth'er noted in Figure 2.1, which also shows the distribution of seismicity to 1980, that the northwest boundary of Zone 8 appears to be inconsistent with the pattern of earthquake occurrences in southern New York, New Jersey, and eastern Pennsylvania. At the meeting at SMA, it was learned that Zone 8 was modeled to represent the Triassic Basin. The inconsistent delineation of Zone 8, with respect to local seismicity patterns, may be attributed to two factors. The LGS-FSAR (Ref. 6) reports that Limerick is in the Triassic Lowlands, suggesting that the. northwest boundary of Zone 8 should be c.oved toward the plant. This would also be consistent with the distribution of seismic events in the region (see Figure 2.1).
2-10 O
Secondly, it is not apparent that the boundaries of seismogenic zones should be coincident with the perimeter of a large geologic struc-ture. If in fact these boundaries generate seismic events, it may not
~
be realistic to restrict their occurrence to the boundary itself.
Instead, events should be modeled as occurring in a volume of crust, defining a zone of weakness.. In one sense, this has been doce for Zone 8 towards the southeast.
A redefinition of Zone 8 in the Crustal 31cck hypothesis that places the LGS site within its boundaries is judged to have a moderate impact on the estimated hazard curves (i.e., at least a factor c' 2).
The consequences of this change on the mean frequency of core melt is estimated to be small (i.e., a factor of 2 or less). Hcwever, a moder-ate increase in the median core melt frequency is considered possible.
To consider the possibility that large magnitude events could occur in the northeast, the Decollement source zone was defined. A maximum magnitude of 6.8 was assumed, and a probability weight of 0.10 was assigned to this hypothesis. 'The selection of maximum event size is discussed in Section 2.1.2.4.
The Decollement hypothesis is one of a number of theories being considered by seismological experts to explain the possible occurrence of large magnitude events in the eastern U.S.
The physical basis of this hypothesis is the identification of a snal-low-dipping reflector beneath and along tne east coast that ha, been interpreted as a seismically distinct block of the earth's crust. U, 8)
A major concern with the Decollement hypothesis is the fact that patterns of instrumentally located seismicity do not correlate well.ith
.i t.
That is, fault plane solutions and source depths do not suggest that earthquakes in the region of Charleston, South Carolina, or any-where else along the eastern seaboard occur on a decollement surface.
In addition, since the evidence that a major decollement may exist generally applies to the southern Appalacians, it is not clear that a decollement seismogenic zone should extend to the northeast in the vicinity 'of the Lime. rick site.
2-11
s At the SMA meeting, discussions with Dr. McGuire revealed that the Decollement hypothesis was not selected solely on the basis of physical arguments that explains the seismicity in the east. A principal motiva-tion was its use as an all-inclusive hypothesis, in a probabilistic sense, in that it allows the possible occurrence of events as large as M6.8.
That is, an assumption is made in the LGS-SARA that all reason-able hypotheses which would consider the possibility that large-cagni-tude events could occur in the vicinity of the plant site are fully represented by the Decollement hypothesis. Although such an approach may provide a best estimate of the ground shaking hazard at the LGS site, it is not clear that it is appropriate or adequate for use in the L'3S-SARA. No basis is provided to support the belief that the Cecollement hypothesis in fact adequately represents, even in a best estimate sense, the hypothesis that large events can occur. Also, the variability ir key parameters was not considered in the Decollement hypothesis (i.e., b-values and Mmax). Neither is it clear that the Decollement source zone is the most appropriate way to model the occur-rence of large magnitude events in the eastern seaboard.
The use of decollement tectonics to explain the occurrence of large magnitude events in the east is one of many theories based in part on scientific evidence-and expert speculation. Although experts differ as to the validity of any theory to explain the 1886 Charleston, South Carolina, earthquake or the occurrence of future large events, the Decollement source zone is certainly one that could be used. However, in the LGS-SARA the Decollement zone serves as a single physical charac-terization of the process that generates large-magnitude events as well as a summary of a multitu'de of hypotheses that define other physical processes.
It is with this expanded role that a concern is raised.
A number of alternatives exist to model the occurrence of large-l magnitude events in the east. Among the possibilities is to allow the occurrence of M6.8 events in the other source zones defined in the i
o 7
2-12
.,l l
i u.o~,._-
...e,
-ee
~-o
....-e.ee..-e.
O
L GS-SARA. That is, an Mmax = 6.8 would be considered as one hypothesis on maxinum magnitude for each source zone. The basis for this aoproach is straight-forward. The occurrence of large-magnitude events in the east is considered possible on pre-existing zones of weakness in the earth's crust. What defines these zones as earthquake genera' tors vary.
In part a variety of such theories are the basis of the seismo-genic zone and hypotheses in the LGS-SARA (i.e., Piedmont, Northeast Tectonic zons, and Crustal Block). The concept of pre-existing zones of weakness 13 consistent with the thinking expressed by the four experts in Appendix B to the LGS seismic hazard analysis. Furthe rmore, a preference was given in the hazard analysis to the piedmont, Northeast Tectonic zones, and Crustal Block hypotheses. A combined probacility weignt of 0.90 was assigned to them. A 0.10 probability was given to the Ceco 11ement hypothesis. Consistent with this degree-of-belief and the consensus in Appendix B that large earthquakes can be expected on pre-existing zenes of weakness, the possibility of large-magnitude events in source hypotheses that define such zones, should be consid-ered. This approach was discussed at the SMA meeting with Dr. McGuire, and recognized by him to be a reasonable alternative to model the occur-rence of large magnitude earthouakes. However, it is the opinion of Dr.
McGuire (and but not necessarily the consensus of all the consultants) that the total probability weight assigned to any and all hypotheses is 0.10.
The question as to whether 0.10 probability is a reasonable value l
to be assigned to the hypothesis that large magnitude events (i.e.,
M6.8) can occur in the vicinity of the Limarick site is a difficult '
question and one that must be answered on the basis of expert opinion.
In Appendix B to the LGS seismic hazard analysis, the four experts interviewed agreed universally that such events could occur at the L GS. The degree-of-belief assigned to such a hypothesis variedt from i
zero to twenty-five or thirty percent. Presumably the value of zero is j
actually a very small number, otherwise there could not have been the l
l 2-13 l
I i
l r
aforementioned univers&l agreement. At this point in the preliminary review of the seismic hazard analysis, the value of 0.10 is not accepted by JBA nor all the experts retained in the LGS-SARA. Quali tatively, this value should be considered a lower bound.
The alternative approach suggested to model large-magnitude events would produce at least one additional hazard curve for each source zone. By virtue of the arguments on maximum acceleration, these addf-tional hazard urves would be unbounded as is the Decollement zone.
y Depending on the source considered, the impact on the frequency of ground motion varies. However, it is felt that in must cases the hazard curve associated with a large-magnitude event will be higher by a factor of 2 or less, compareo to the existing hazard curves. At higher accal-erations, these new curves will be unbounded and thus have nonzero occurrence frequencies, unlike the previous hypotheses.
With respect to their impact, the fact that these additional curves are unbounded means that they will make a greater contribution to the mean frequency of core melt than their counterparts for each source zone. Previously, the Piedmont, Mmax = 6.3 and Decollement hypotheses contributed 86 percent of the mean core melt frequency, since they only allowed accelerations greater than 0.80g to occur. All zones will have some contribution to the mean frequency of core melt. The overall influence of these additional curves is judged to result in a small increase in the mean core melt frequency.
2.1.2.4 Seismicity Parameters For a prescribed zone of seismicity, the random occurrence of earthquakes is defined by the seismic activity rate, the Richter b-value, and the maximum magnitude that can be generated by the source. Estimates of seismic activity are based on the nistoric record. However, the statement that seismic activity rates are well' determined in the eastern U.S. is in some ways an overstatement or at Y
2-14 9
least easily misinterpreted. For a prescribec area in the east, the catalog of earthquake occurrences is generally believed to be long enough and sufficiently complete that estimates of activity rates are reasonably well determined. That is, their uncertainty is low enougn that its impact on the frequency of exceedance of ground motion can be ignored. However, from the point of view of the rate of seismic activ-ity per unit area (i.e., say 104 km2) the variation can be larga. Frca Table 2 in the LGS-SARA hazard analysis, the rate of seismic for the four source hypotheses varies from 4.33 to 38.0 x 10-3 events par-year, 4
2 per-10 km. This effect is taken into account in the LGS-SARA, however this variation per se is not recognized as such.
In the LGS-SARA, the estimate of Richter b-values was basei solely on expert opinion as reported in Reference 9.
A best estimate of 0.90 was used for all source zones, and no uncertainty was considered.
In Reference 9, the experts came to a consensus that 0.90 was a realistic, albeit default value that can be used for all seismogenic zones in the eastern U.S.
However, it was further stated by many of the experts that it is believed that b-values for different seismogenic zones may vary from 0.90 as a best estimate.
This notion suggests that variability in the mean value of b exists. That is, a difference exists between the 0.90 global estimate, and the true best estimate for a given source zone.
In fact, some experts indicated a preference for a regional dependence for b-values. Furthermore, there is the contribution of statistical variability in b-value estimates derived from the data, which depends on the number of data points. Thus, as a minimum, two sources of variability exist in the estimate of Richter b-values:(1) a possible bias in the use of 0.90 best estimate value recommended by experts for all source zones, regardless of the actual distribution of the data and (2) the statistical variability due to limited sample size. The failure to account for the variability in b-values is an example of the inadequate degree to which parameter hypotheses have been sampled in the LGS-SARA.
It should be noted that the LGS-SARA did not 2-15
_,4 w
w
-v
'ww "4
directly estimate Richter b-values from the catalog of earthquake occur-rences.
In considering the estimate of b-values, PECo should consider the results obtained using the historic data.
The impact of a complete characterization of the variability in b-values on the mean core melt frequency is judged to be small.
The final seismicity parameter defined for a seismogenic zone is the uaximum magnitude.
In the previous section, the manner in which large magnitude events were modeled in the LGS-SARA was censicered.
Here, the matter of what the size of the largest events should be assumed is addressed.
The estimate of maximum magnitudes for the Piedmont source zone reflacted the issue of the 1982 New Brunswick, Canada event and the Cape Ann earthquakes. The magnitude 5.7 New Brunswick event is used as the basis for establishing the distribution on Mmax, while it was stated that the Cape Ann earthquakes do not belong in the Piedmont zone. The basis for limiting the occurrence of the Cape Ann events to New En gland is presumably related to the theory that a Boston-Ottawa seismic be it exists as discussed in References 10,11 and 12. However, the existence of such a trend does not correlate very well with results of recent studies questioning the existence of a such a trend.(12) Thus, no definitive basis exists to support the hypothesis of a Boston-Ottawa seismic belt and therefore no reason exists to exclude earthquakes near Cape Ann, from the Piedmont region. This is further supported by the arguments provided in the LGS-SARA that suggest the 1982 New Brunswick, Canada, earthquake belongs in this seismic province.
If the 1755 Cape Ann earthquake is considered to be a 6.0 event,Il4I the distribution on M would be modified to reflect the max fact that the largest observed event had a magnitude of 6.0 as opposed to 5.7.
If it is assumed that the two point distribution on M was max enanged fro'n 5.8 and 6.3 to 6.0 and 6.5, it is estimated that the effect on the frequency of exceedance curves and the mean core melt frequency would be small.
2-16
)
./
O
~
The hypothesis that a large-magnitude event, the size of the 1886 Charleston, South Carolina, event could occur on the eastern seaboard was considered in the Decollement source zone. A magnitude of 6.8 was assigned to this Modified Mercalli Intensity (MMI) X event. No basis is provided in the LGS-SARA to support the implicit assumption that the observed magnitude of the Charleston event is the maximum event that could occur. Should it for example be considered a lower bound on Mmax? This question and the uncertainty in M should be addressed max by PECo.
2.1.2.5 Ground Motion Attenuation To describe the attenuation of ground motion with magnitude and distance, Nuttli's relationship for sustained acceleration was used.(15) The uncertainty in ground motion predictions is described by a lognomal distribution with a standard deviation of 0.60.
This stand-ard deviation value corresponds to a factor of 1.8 times the median val ue.
The attenuation relationship was modified in the hazard analysis to predict sustained-based peak acceleration and to account for the random orientation of ground motion. This factor is magnitude dependent.
Above magnitude 6.0, sustained-based peak acceleration is 1.23 times sustained acceleration. The attenuation model used in the LGS-SARA' is appropriate and adequate to describe the ground motion at the plant si te.
The prediction of ground motion in the eastern U.S. is a difficult task due to the limited strong motion data available for that region.
However, a number of relationships have been developed and used in I
probabilistic hazard analyses.Il* 9I Results of sensitivity studies are available to compare the impact of various functions on the estimated hazard curves. A preliminary review of these studies suggests the attenuation for sustain-based peak accelerations used in the LGS-SARA is 2-17 i
-1
.---vt---srr-w--
+---e+w*.
2
---,-wes-w t-
.r
---t-
e e--r-w-v-w,ev~re-+--
Nm=---n---ve'--w-c-
generally on the conservative side (i.e., it gives higher accelerations at a given frequency of exceedance level).(9)
It is noted however that there can be considerable variation in the hazard analysis results for various attenuation relationships. This suggests that a more compre-hensive sampling of attenuation functions is appropriate, since it is generally believed that the capability to predict ground motion in the eastern U.S. is not well' established. The imoE.t of including alterna-tive attenuation hypotheses on the mean core melt frequency is consid-ered to be small.
2.1.2.6 Comparison of the LGS Hazard Analysis with the Historic Seismicity The accuracy of the LGS-SARA seismic hazard analysis might be compared with the historic distribution of earthquake ground motion experienced at the plant site. However, since a record of the ground shaking intensity at the LGS site is not available, another approach must be taken.
In the Limerick FSAR(6) the earthquakes that have occur-red since 1737 within 200 miles of the site (Table 2.5-2, Reference 6) are reported. These data provide a basis to estimate the distribution of historic ground motion. The approach used to do this is summarized below.
The catalog of earthquake occurrences provided in the FSAR describes event size in terms of Modified Mercalli Intensity. To estab-lish a distribution of the MM intensities experienced at the LGS, the reported epicentral intensities are attenuated to the plant. This is done using the intensity attenuation relation in Reference 16 for rock sites given by the following equation, Is*Io + 2.6 - 1.39 inR (2.1) 2-18
)
J where:
Is = site intensity In = epicentral intensity R = distance (mfies)
For each event and distance reported in the FSAR, a site intensity was estimated using Ecuation 2.1.
In establishing a record of the 'WI levei experienced at the LGS site, no attemot was made to verify the cataloo reported in the FSAR or to correct the record for inconsistencies.
Also, no uncertainty in the estimate of site intensities was consid-ered.
Intensities above MMI eaual to IV are considered.
To define the distribution of seismic intensities at the site, t"e Gutenburg-Richter relation that describes the number of events versus intensity is given as follows:
log 10 (I ) = a + bi (2'2)
N s
s where a and b are parameters fit to the data. The b term is known as the Richter b-value. The b-value on intensity is estimated to be
-0.72.
The seismic activity rate for events of MMI > IV is 0.0266 events per year based on a 226 year record.
An estimate of the historic ground motion in terms of ground accel-i eration can be obtained by a transformation of intensity to peak ground acceleration using an appropriate relation. To do this, the following equation was used:(17)
I log 10A = 0.014 + 0.30!s (2 3) 2 where A is peak ground acceleration in cm/sec. To account for the uncertainty in estimating A in Ecuation 2.3 and the uncertainty in attenuating intensity in Equation 2.1, a legnormal distribution on peak j
acceleration is assumed, with a logarithmic standard deviation of 0.28 (base 10), which corresponds to a factor of 1.9.
1 2-19 i
o The distribution on acceleration at the LGS is estimated according to v(A >a) = v f(I).a! P(A > a l I)
(2.4) where v(A > a) annual frecuency of peak acceleration A,
=
greater than a value a.
seismic activity rate for intensities greater v
=
than or ecual to IV.
f(I) a !
doubly truncated exponential distribution on
=
intensity I with parameter b'in 10 where al is the increment on intensity.
P( A > a l I)
= probability of peak acceleratinn A greater than a, given an intensity I.
This is described by a lognormal distribution whose median is defined by equation 2.3 with a logarithmic standard deviation, of 0.28 (base 10).
The result of this computation, using I of VI, is shown in Figure 2.2 max with selected curves from the LGS hazard analysis.
The historic seis-micity curve rances at accelerations around 0.10g from the results obtained for the Decollement and Piedmont zones to the lower frecuencies estimated by the Crustal Block zone. These observations suggest that the overall frecuency of events producing accelerations of 0.10g is r1asonably well described by the Decollement and Piedmont zones and Crustal Block zone, M = 6.0, to within a factor of 2.
Since the maximum intensity felt at the site is MMI VI, the historic frequency curve falls off sharply.
Equation 2.4 can also be used as a prediction tool by allowing the possibility of site intensities greater than VI to occur. To do this, an estimate of the maximum site intensity that can occur must be made.
2-20
)
,e.
ew ++
w-.
,f.
1 This is the same step that is taken in the probabilistic seismic hazard analysis. A maximum intensity of X is assumed, which corresponds to a large-magnitude (=M7.0) event occurring very near the site.
The result of estimating f(I) in equation 2.4 and calculating v( A > a) for a maximum intensity of X, is also shown in Figure 2.2.
This assumotion allows the possibility of high accelerations associated with large events to occur.
In general, the site intensity curve tracks the trend of the Piedmont and Cecollement seismicity curves cuite well.
As a final estimate based on the historic distribution of ground motion at the LGS, a seismicity curve is estimated assumino a Richter b-value of 0.d5 which corresponds to the 0.90 value used for the magnitude scale in the LGS-SARA. Also, a maximum intensity "MI X is assumed.
The hazard curve for this case is also shown in Figure 2.2.
The effect af assuming a b-value of 0.45 (equivalent to 0.90 for the magnitude scale) is a factor of four increase in the hazard.
The results based on the historic-31te intensity distribution agree reasonably well with the seismicity curves derived in the LGS-SARA.
From the point of view of prediction, f a maximum site intensity of X is postulated, the Piedmont and Decoll.3 ment zones acree most closely with the historically derived curve.
The same could be said for the Northeast Tectonic zone, expect that the truncation on peak acceleration produces a sharp fall-off at 0.30g.
2.1.2.7 Summa ry The previous sections provide the results of a preliminary review of the LGS-SARA seismic hazard analysis.
The adecuacy and appropriate-ness of the analysis approach were considered.
The appropriateness of individual technical aspects of the analysis were also reviewed.
The methodology used to estimate the probability distribution on frecuency of exceedance is considered appropriate to estimate the seismic risk due to nuclear facilities.
The method used in the LGS-SARA 2-21
~
1J is a well established straightforward approach to estimate the ground shaking hazard.
With regard to the adequacy of the way the method was applied, it is felt that in principle the estimation of the probability distribution on frequency is not necessarily well defined by the coarse sampling of parameter hypotheses used in the LGS-SARA. The approach used in the LGS-SARA was to select six hypotheses, each with an assigned probability weight.
It was then assumed that the six hazard curves generated, fully define the probability distribution en frecuency.
Although a best estimate can be obtained in such a manner, this approacn does not insure that the probability distribution on frequency will be adequately represented.
With regard to seismogenic zones, two major concerns were raised.
First, delineation of the boundaries of the Crustal Block hypothesis was questioned.
In particular, Zone 8 in this model was considered inappro-priately defined to be approximately 30 miles from the LGS at its closest point. The impact of redefining Zone 8 on the mean frequency of core melt was considered to be small. Secondly, the Decollement source was used as an all-inclusive model to consider the general hypothesis -
that large-magnitude events can occur in the east. This approach was not considered to be the most reasonable means of evaluating the hazard due to such hypotheses.
An alternative was recommended that allows the possible occurrence of large-magnitude events to occur on the other source zones as well. The impact of this alternative on the mean core melt frequency was considered to be small.
With regard to seismicity parameters, two issues were raised. The first deals with the assignment of Richter b-values. The LGS-SARA uses a single b-value for all source zones. The basis for this was expert opinion. No uncertainty in b-values was considered. This approach was not considered appropriate, rather, a distribution on b-values should be used since there exists a source of bias in the best estimate of the b-value for each source zone, as well as statistical uncertainty. The r
2-22 T
.J y
n
-r.
9
, - - - -7
,.m
~
f impact of not considering the uncertainty in this parameter is consid-ered to be small.
Particular concern was expressed with regard to the estimate of maximum magnitudes. For the Piedmont source, evidence was presented that questioned the basis for establishing the distribution on maximum magnitude.
Specifically, the Cape Ann events should be included in the Piedmont province and considered in the estimate of M Tne overall max.
impact on the mean core melt frequency is considered to be small.
The possible occurrence of large-magnitude events (=M7.0) was considered in the Decollement source hypothesis. The 1886 Charleston, South Carolina event was estimated to have a magnitude of M6.8 in the LGS-SARA and was used as the Dasis to estimate the largest event that could occur.
No uncertainty in this estimate was considered, neither was there any basis for this hypothesis.
In a preliminary assessment of the hazard analysis results, the frequency distribution of ground motion due to historic earthquakes was comouted. Generally, the results from the analysis of the historical data suggest that LGS-SARA' study results are reasonaM e.
Hazard curves that include the possibility of an intensity X event are consistent with the hazard curves estimated for the Piedmont, Decollement, and Northeast Tectonic zones at low accelerations.
The recommendations given in Section 4.1.2 are directed towards resolving the issues summarized above. Although the effect of the individual issues on the mein frequency of core melt is judged to'be small, the total effect could be moderate.
2.1.3 Seismic Fragility The oreliminary review of the seismic fragility parameter values focused on Appendix B of the LGS-SARA and included a review of those portions of Chapter 3 and Appendix C pertinent to the seismic risk 2-23
a analysis. As described in Section 2.1.1, the results of the meeting with SMA and the plant tour helped direct the review effort to the critical components and issues.
In addition, the calculations for the significant contributors in Table 3-1 of the LGS-SARA were obtained and studied. The fragilities for other components were considered in rela-tionship to their potential impact on the mean frequency of core melt.
For example, the median capacity of the batteries and ra:ks is reported to be 2.56g and, thus, was not included in the secuences. This compo ~
nent was inspected during the plant tour, and its capacity value is judged to be reasonable.
The ccmments concerning the seismic fragility analysis are organ-ized in a manner to highlight the concerns, which were either most potentially critical or which were the most controversial during the review. Sections 2.1.3.1 through 2.1.3.6 discuss this category of concerns. Section 2.1.3.7 presents the results of the review of the calculations for the significant components. Many of the concerns found during the review of the calculations are also discussed in detail in Sections 2.1.3.1 through 2.1.3.6.
Section 2.1.3.8. addresses general fragility-related issues which should not be overlooked, but which are philosophical in nature (i.e., do not have an immediate resolution) or which are unlikely to have a major impact on the results. Finally,
[
Section 2.1.3.9 gives final closing comments on the preliminary review l
of the seismic fragility analysis in the LGS-SARA.
Throughout the discussion recommendations are made for additional in formation. Section 4.1.3 sumarizes the recommendations for addf-tional actions required to resolve the fragility-related issues which have been raised but not answered.
I i
l 2.1.3.1 Damage Factor Three adjustment factors are used in the LGS-SARA to estimate capacity to resist earthquakes. The hazard analysis documented in 2-24 l
8 w.
e O
Appendix A of the LGS-SARA presents tne frequency of exceedance for seismic hazard in terms of a sustained-based peak acceleration para-meter. As explained in Section 3.3.1 of the LGS-SARA, the accelerations from the Appendix A hazard curves were scaled 'y a factor of 0.81 (i.e.,
o 1/1.23) to convert-the sustained-based peak accelerations to effective peak accelerations.to reflect the less damaging characteristics of 10w magnitude earthquakes. This adjustment is identical to the adjustments made in the IPPSS and the ZPSS. As explained in Reference 18 Reference 18 was provided to the reviewers by PECo to support the LGS-SARA), this factor was conservatively selected to account for smaller nonlinear res-ponse and, hence, damage caused by lower magnitude events.
It is im-plied in Reference 18 that thc. adjustment factor should be 0.5 for mag-nitudes less than M5 and distances less than 20 km. For magnitudes greater than M7 and distances greater than 40 km, the adjustment factor i s uni ty.
A second factor was introduced in the LGS-SARA which is discussed in Section 4.1.3 of Appendix B of that report. This factor is called an earthquake duration factor, which is used to increase the median capac-ity of structures by a factor of 1.4.
The justification for this factor as discussed in Section 4.1.3 is very similar to the justification for the hazard reduction factor (i.e.,1/1.23) described above; thus, it is concluded that these factors account for the same phemonena and only one factor should be used. Note that the duration factor of 1.4 was not included in the IPPSS and the ZPSS.
This apparent discrepancy was discussed at the meeting held at SMA, and it was explained by SMA that for future PRAs only the 1.'4 factor
~
will be used and no adjustment will be made to the seismic hazard curves.
In defense of the LGS-SARA analysis, SMA explained that very low ductility values had been used in the development of the ductility factors for Limerick (i.e., 2.0 for shear and 2.5 for flexural failure of concrete walls). The ductility factor is the third adjustment factor used in the LGS-SARA. More realistic values of 3 to 4 for the ductility 7
2-25
s ratio should have been used. The use of low ductility values compen-sated for the extra 1/1.23 factor used to adjust the hazard curves for structures. The 1.4 factor was not used for equipment which generally had realistic ductility values.
In conclusion, if only the 1.4 duration factor and realistic concrete ductility values had been used for the structures, the results would have been essentially the same. The reviewers concur with this explanation.
The justification for the duration factor of 1.4 was also reviewed. The underlying basis for the duration factor is recent work reported in Reference 19. As documented in this report, a series of analyses were conducted to investigate the response of single-degree-of-freedcm (500F) nonlinear oscillators to real earthquake motions. Earth-quakes which varied in magnitude from M4.3 to M7.7 were used. It was explained at the neeting at SMA that a ' duration factor is required to correct the capacity of SDOF systems when subjected to earthquakes less than M6 to obtain the same level of damage.
The ductility factor based on the approach developed by Riddell and Newmark,(20) which was used in the LGS-SARA, assumes earthquakes larger than M6.
Since this method is used to develop the ductility factors for structures, a duration factor (really a magnitude factor) was applied for events with magnitudes less than M6. An analysis was conducted by
'SMA using the data from Reference 19, where the response of the non-linear SDOF oscillators to earthquakes less than M6 to events greater than or equal to M6, were compared. By fitting a lognormal distribution to the ratios of the response factors for these two groups of events, the median adjustment factor of 1.4 us determined. In the LGS-SARA this factor was applied for all hazard curves, which implicitly assumes all earthquakes have magnitude less than M6.
In an effort to verify the earthquake duration factor used in the LGS-SARA fragility analysis, the data contained in Reference 19 was reviewed. As described above, arguments which support the use of an 2-26
)
s s._,.
~
~ ~ - = - ~ -
s
-m.-
... - ~
~
earthquake duration factor are based on the assumption that seismic events of magnitude less than 6 contribute to the likelihood of fail u re.
It was on this basis that the median value of 1.4 was derived for use in the LGS-SARA. As a check, the data in reported in Table 4-1(a) for u= 4.27 in Reference 19 were considered in two groups:
M<6 and M)6.
The artificial time history was included in the M>6 grouo.
From the histogram for each group the median response factor and logarithmic standard deviation were derived.
Then, the ratio of the response factors was determined and compared to the LGS-SARA values. A summary of the estimates made are given below.
Resconse Factor Data Grouc F,
S M<6 2.65 0.25 M)6 2.15 0.26 FED = F <6/F >6 1.23 0.36 M
M LGS-SARA 1.40 0.20 Sc 0.12 Sp 0.08 Su From this comparison, it appears that the median factor used in the LGS-SARA is over estimated by 14 percent (i.e.,1.23 compared to 1.40).
It should be noted that including the artificially generated time history in the M>6 group has a negligible effect on the median.
A second look at the scale factor data was taken by dividing the data in short and long duration (T ) groups.
The data were divided O
according to whether durations were less or greater than 2.5 seconds, as defined in Referance 19.
2-27 1-
In this case the artificial time history is in the T >2.5 second D
group. Basically all the records in the M)6 group were in the T >2.5 D
second data set with one exception.
The UCSB Goleta recordina of the M5.1 (M 5.6) 1978 Santa Barbara earthquake had a duration of 3.0 s
seconds, and thus was included in the long duration subgroup. The results for these data sets is given below.
Resoonse Factor Data Group F
S 7 42.5 sec.
2.85 0.51 0
T >2.5 sec.
2.05 0.26 D
FED = FTO'2.5/FTD>2.5 1.39 0.57 LGS-SARA 1.40 0.20 Sc 0.12 Sp 0.08 Su From this comparison, it would seem that in deriving the duration factor, that a duration, rather than a magnitude criteria was used.
This is inconsistent with the application in the LGS-SARA. Possibly of greater significance is the fact that a single earthquake record produced a variation in the estimated median duration factor from 1.4 to 1.23.
This would seem to point out that, although Reference 19 provides a clear indication of the duration effect of strong motion on structural damage, it is a study limited in its application because of the rela-tively small data base. As discussed at the meeting with SMA, the use of the M6 cutoff to establish the duration factor is a gross character-ization of a process that is continuous over magnitude and/or duration.
Thus, a median duration factor should preferably be a func-tion of magnitude. Data to establish such a function are not avail-abl e.
Furthermore, Reference 17 also suggests that the duration factor 2-28
= - ' -
y
---.yg-3-w yvyy
-v
,,-.-y 9
gr c
r r
has a frequency dependence. This is not taken into account in the LGS-SARA.
The estimate of the logarithmic standard deviation of tne duratien factor in the LGS-SARA appears low.
In carticular, due to the uncer-tainty in estimating FED and the limited data base, s
= 0.08 i s l ow,
u and in any case should not be lower then the randomness component.
Direct estimates of the variability in FE0 ranged frca 0.35 to 0.57.
'lalues of s of this size are considered core approoriate.
e In principle, incorporating the effects of duration in the estimate of seismic capacities is appropriate. And although the results reported in Reference 17 are consistent with engineering judgmant and observed earthquake damage, the approach used in the LGS-SARA is a si plification of a complicated issue.
The arguments leading to the 1.4 duration factor, when included with the ductility adjustr:ent factor based on Reference 19, are gener-ally reasonable for earthquakes with magnitudes less than M6; however, as discussed above, the 1.4 factor may be slightly high and the uncertainty estimate low. For events greater than M7 it was agreed by SMA that the duration factor should be unity. Between macnitude M6 and M7 events the data in Reference 17 do not support a duration factor of 1.4 in the opinion of the reviewers.
If the duration factor of 1.4 is changed to 1.0 for structures and equivalently the hazard curve adjustment of 1/1.23 for equipment is also changed to 1.0, for the region of peak-sustained accelerations corresponding to average magnitudes greater than M6.0, the frequency of core melt distribution will be affected.. Note that the ductility values used for equipment are generally realistic, hence the 1/1.23 hazard curve factor is analagous to the 1.4 duration factor used for structures.
Based on Reference 21, the hazard curve for the Decollement seismo-genic zone is the only curve which has average magnitudes equal to or greater than M6.0.
For sustained-based peak accelerations equal to or greater than 0.40g, the average magnitudes equal or exceed M6.0.
It is 2-29
estimated that if the duration factor is changed to unity for this region of the Decollement hazard curve the mean frequency of core melt will increase by a factor of app.oximately 1.4.
The effect of this adjustment will rot significantly affect the median core melt frequency.
2.1.3.2 Upper Bound Accelerations All the hazard curves, except the Decollement case, are truncated to reflect the belief that maximum ac~celerations are associated with each seismic hazard hypothesis. The argument leading to the limiting acceleration values is documented in Reference 18, which was provided to the reviewers by PECo to support the LGS-SARA. This is the same argu-ment which is given in the IPPSS and ZPSS reports (1,2) for limiting accelerations. The explanation for limiting upper-bound accelerations consists of two steps. The first step is the assumption that there is a maximum intensity associated with each source zone corresponding to the maximum magnitude for that zone. This is assumed to be true by seis-mologists. The second step relates the predicted accelerations for masonry structures with the qualitative descriptions of the MMI scale.
The basis for the argument leading to maximum acceleration values in the second step is as follows. Masonry structures are selected since they are the only engineered components for which damage is systemati-cally described in the MMI scale.
If the accelerations are higher than predicted, then a higher MMI value (corresponding to more damage) would occur. However, since the maximum MMI vaiues are limited by the seis-mologist, a higher acceleration is not possible. The problem with limiting accelerations for the Decollement hazard curve is the assigned maximum magnitude value of M6.8 which corresponds to a maximum intensity of approximately MMI X.
This intensity is associated with failure of most masonry structures; thus, the argument cannot be used since all higher MMI values also include failure of most (if not all) masonry structures. As explained at the meeting at SMA, it was conservatively decided not to truncate the Decollement hazard curve.
2-30 e
i It also follows directly that if upper bounds on intensity exist then upper bounds on damage exist since intensity is a scale which measures damage. Although it is believed by the reviewers that it is more appropriate not to truncate the hazard curves but to reflect a limit on damageability in the. fragility curves, the effect of modifying the hazard curves produces the same result. Thus if upper bounds exist for lower intensity values, similar limits should apply for higher intensity values for engineered concrete structures. Hcwever, it is difficult to quantify this belief at this time.
In conclusion, tne assumption not to truncate the Decollement hazard curve is on the con-servative side.
Based on the approximate analysis described in Section 2.1.1, the effects of truncating the Decollement hazard curve were investigated.
It was found that when truncating the curve at 1.0g (which represents a reasonable lower bound) the mean frequency of core melt will change by a factor of approximately 0.85.
The effect on the median frequency of core melt is expected to be very small. Thus, it is concluded that truncating or not truncating the Decollement hazard curve has a small effect on the results of the LGS-SARA.
2.1.3.3 Reactor Enclosure and Control Structure The median capacity of the reactor enclosure and control structure is reported in the LGS-SARA to be 1.05g (see Table 3-1 in the LGS-SARA). The structural calculations for this ccmponent were reviewed.
The reviewers believe that the capacity of the walls is rationally represented by 0.90g, which is based on the total capacity of the walls in the north-south direction between elevation 177 feet and 217 feet.
This capacity is based on the capability of the floor diaphragm at elevation 217 feet to redistribute forces. At the meeting with SMA, it was stated that the diaphragm capacity for the Susquehanna plant was checked in detail and since the Limerick plant is structurally the same, 2-31
l.
the diaphragm capacity is adequate to redistribute forces as the varicus wall sections yield.
~
Based on a median capacity of 0.90g, it is estimated that the mean frequency of core melt would increase by a factor of approximately 1.2.
2.1.3.4 Reactor Pressure Vessel Capacities Three of the significant. earthquake-induced failure components listed in Table 3-1 of the. LGS-SARA are associated with the reactor pressure vessel (RPV) which is located in the containment structure.
In the development of the median capacity values for the reactor internals, RPV, and the CRD guide tubes, it was assumed that the containment struc-ture had an effective damping value of 10 percent. Since the original analysis of the combined containment /NSSS was based on 5 percent damping for the concrete structure, a 1.3 factor, which increased the capacity of the RPV ccmponents, was developed from the ground spectral accelera-tions by SMA.
It is not obvious from the LGS-SARA or the calculations that the 1.3 factor is appropriate since the stresses in the containment struc--
ture may not be sufficiently high to warrant the assumed 10 percent damping value. The median capacities of the three RPV components range between 0.67 and 1.379, while the limiting median capacitics of the supporting containment structure components are as follows:
Sacrificial shield wall 1.6g Containment wall (shear failure) 3.49 RPV pedestal (flexural failure) 2.8g The upper portion of the RPV is resisted by a ring at the top of the shield wall which, in turn, is anchored to the containment wall by steel lateral braces. The relative stiffness of the lateral supports versus the stiffness of the sacrificial shield wall is not known.
If a 2-32 T-]
O
4 h
major portion of the resistance comes from the shield wall, then 10 percent damping is probably appropriate. On the other hand, if the input to the RPV is dominated by the support at the top of the shield wall,10 percent damping may be too large.
If the 1.3 damping response factor is changed to unity, which is the most conservative assumption for this factor, it is estimated that the mean frequency of core melt would increase by a factor of acproxi-mately 1.10, which is a small effect.
In the original analysis conducted for the design of the contain-ment and RPV components, a coupled model was used with a single input time history. An additional uncertainty for variation in response due to time history analysis should be included for the RPV-related ccm-ponent capacities. Also, the model used to develop the capacity of the RPV lateral support is approximate and, hence, additional uncertainty is present. It is believed that due to the SRSS operation for combining uncertainties, the effect of these additional uncertainties would have a small effect on the mean frequency of core melt.
2.1.3.5 Potential Impact Between Reactor Building and Containment The reactor building and containment are constructed on different foundations and are separated by a gap filled with crushable material.
The gap reportedly varies between one inch at the foundation level to three inches at the top of the structures. It is stated in Appendix B of the LGS-SARA that at 0.lg, the containment begins to uplift, and at 0.45g the two structures begin to impact at elevation 289 feet (it is believed that elevation 283 feet is the correct level). It is also stated that since the' reactor building shear walls are expected to fail between 0.74g and 1.0g no signficant additional damage due to impact is expected to occur.
This assumption was questioned during the review. Three possible effects were considered. First, the impact between the structures might 2-33
1 cause high frequency motions which could affect electrical and control equipment. Based on inspection of the plant, the gap between the reactor building and the containment appears to be irregular; thus, the transfer of energy during impact would occur over scme finite period of time which would soften the impact. The suddenness of impact would also be cushioned by local crushing of the concrete. Because of the large size of the walls and floor slabs, gross structural failure due to impact is not expected. As a minimum, the chatter and trip of relays would increase; however, NUS states that this is not a problem whether caused by either impact or just due to dynamic motions.
It is not clear whether failure of the electrical equipment located in the reactor building will be increased by impact between the two structures. The capacity of the electrical ccmponents located in the
~
reactor building (some of which are located at elevation 283 feet within 30 feet of the seismic joint) range between 1.46g and 1.56g. This is considerably higher than the motion level at which impact may occur; hence, these capacities may, in reality, be less.
The second potential problem is spalling of concrete which could fall and impact safety-related equipment.
It was learned during the tour of the plant that all electrical and control equipment are located away from the seismic joint. Thus, these types of components will not be affected. Various safety-related pipe lines cross between the two buildings. It is expected that the size of any spalled concrete pieces will be small since the reinforcing steel will tend to hold any frac-tured concrete pieces in place. In addition, the slope of the contain-ment wall will break the fall of spalled concrete pieces. The risk of a major rupture of a pipe or valve due to impact from spalled concrete is believed to be relatively small.
The final concern is the relative displacements caused by the movement of the two buil_. dings and their effects on safety-related piping. It was stated at the meeting with SMA that all piping which 2-34
=
i
-l w.,
, - -e e -
l contains hot water has sufficient flexibility to accommodate temperature changes to resist the potential relative displacements between the two structures due to earthquakes. Subsequent to the meeting at SMA, the question arose concerning whether piping with Icwer temperature require-ments could resist the potential relative displacements. During the tour of the Limerick plant, an 18-inch diameter line was identified and inspected.' The line number was obtained (GBB119) and the locations of lateral supports were found on the isemetric plans in tne plant engin-eering office.
It was confirmed that this line belongs to the RhR system and is a low temperature line. The first critical support was located approximately 10 feet horizontally and 12 feet vertically from the containment wall in the reactor building. The flexibility of tnis pipe was checked approximately and it appears to have sufficient flexi-bility to resist two-to-three inches of relative movement. A stress of approximately 10,000 psi would be caused by a three-inch relative dis-placement which, when added to other stresses, probably would not signi-ficantly affect the core melt frequency distribution.
Several small lines (probably control-related) were attached to a valve close to the containment wall. These lines were also attached to the reactor building close to the valve.
It is possible that these lines might fail during large relative motions; however, it was stated by NUS that small leakage in small lines is acceptable.
The concerns raised regarding impact between the containment and reactor building have not been entirely resolved. The effect of impact on the capacity of electrical and control equipment should be addressed by PECo.
In addition, all the safety-related piping which connects both buildings should be systematically reviewed to verify that sufficient flexibility is provided to accommodate relative. displacement between the two structures.
T 2-35
-,v-en
-e r
-n-----,.
--m
,,-,--,-,,r
,--------,e.----~~m-~~e--
-e
-w
--w w
~e.
n v
- --+-
2.1.3.6 Electrical and Control Eauipment The mean frequency of core melt reported in the LGS-SARA is 5.7x10-6 per year. About 60 percent of this value is contributed by sequence T E 0X, which includes the following five electrical or control ss components which arc in series:
e 440-V bus /5G breakers
- o. 440-V bus transformer breaker e 125/250-V dc bus e 4-KV bus /SG e Diesel-generator circuit breakers These components have median effective peak acceleration capacities which reportedly range from 1.46g to 1.56g (see LGS-SARA Table 3-1), and which contribute most of the mean frequency of core melt value of 3.15x10-6 reported in the LGS-SARA for sequence T E UX. A concern s3 raised in the review is the actual number of units which exist for each one of these five components. For an increase of one additional inde-pendent unit (e.g., if there are two independent switchgear breakers instead of only one), the mean frequency of core melt will increase by approximately 0.4x10-6 per year.
Several issues should be considered in determining whether addi-tional units should be added in series. First, the fragility values for these components are based primarily on generic data obtained from equipment tests for the Susquehanna nuclear power plant.
It Is not apparent from the documentation in Appendix B nor the LGS-SARA whether the test specimens used in the Susquehanna tests were for single or multiple units (i.e, was one switch gear breaker tested at a time, or were multiple units tested simultaneously?). Also, how similar are the components in the two plants?
2-36
.s
=
1
The second consideration is the question of independence between components. It can be argued that identical units have high capacity dependence (i.e., if two units of the same component are subjected to the same dynamic motion either they both will survive or they both will fail).
If two components are located next to each other and receive the same dynamic input, they also may have high resconse dependence. This is true even though they may be different types of components.
If multiple units of a particular component exist in series (e.g.,
440-V bus /SG breakers) but they are identical units located next to each other, they may be in a practical sense perfectly dependent, and the frequency of failure would be equal to the frequency of failure of one unit. On the other hand, if the units are constructed differently and/or placed at different locations, they may approach being indepen-dent which in the extreme case implies that the frequency of failure is approximately equal to the sum of the individual failure frequencies.
In order to evaluate the impact of this concern PECo should deter-mine the number, location,.and characteristics of the electrical and control equipment which are part of sequence T E 0X, and compare the s3 ccmponents to the generic test specimens from the Susquehanna tests. As suggested in Section 2.1.3.7, component-specific calculations should be performed +.o develop the fragility values for these components since they are significant contributors to the frequency of core melt.
2.1.3.7 Review of Significant components A copy of the calculations performed by SMA for the signficant components listed in Table 3-1 of the LGS-SARA were obtained and reviewed. Although the capacities of' other components were considered in the review, the effort focused on the significant components which affect the dominant sequences leading to core melt. As an aid in this phase of the review, equipment fragility values deveicped in the Seismic Safety Margins Research Program (SSMRP) were used as a guide. (22, 23) 2-37
The following comments are given for the 17 significant components.
Offsite Power (500/230-XV Switchyard) (S ) - The fragility for i
offsite power is based on the failure of porcelain ceramic insulators.
No specific calculations were given for this component. The capacity is based on historic data and is reasonable.
Condensate Storage Tank (S ) - This component is not a major con-2 tributor to the mean frequency of core melt. The capacity of the tank is based on the weakest failure made wnich is shell buckling. A small ductility value of 1.3 was assumed. This is probably reasonable but may not be conservative since a bc:kle could cause a leak in the tank. This assumption is also inconsistent with the analysis performed for the SLC tank where cuckling also controlled. For this case, no ductility was assumed.
No adjustment for soil structure interacticn was made which assumes that the tank is on rock.
It was not apparent from the tour of the Limerick site that the tank base is founded on rock; however, based on the fundarrental frequency of the tank given in the calculations, the effect of fill would increase the capacity.
In summary, the fragility parameters for the condensate storage tank appear to be reasonable.
Reactor Internals (5 ) - The capacity of this component is limited 3
by the strength of the shroud support. The exact failure location was not given in the calculations. The capacity factor was derived based on the calculated stresses obtained from the original design analysis. As discussed in Section 2.1.3.4, only one time history was used in the analysis. Although a randomness logarithmic standard deviation of 0.05 was used, this value is low for the amount of variability which coyld occur, if multiple time history analyses had been used. The total l
effect of increasing the logarithmic standard deviation for time history l
variability is s. mall.
l l
2-38
-.--n
_.,en n.
4
As discussed in Section 2.1.3.4, the factor of 1.3 which increases
~
the capacity of the reactor internals to reflect 10 percent damping expected for the containment (as opposed to 5 percent damping in the original design analysis) may be high.
It is estimated that the maximum impact, if this factor were 1.0, woul'd be an increase in the mean frequency of core melt by a factor of approximately 1.10.
Reactor inclosure and Control Structure (5 ) - The capacity of this 4
ccmponent is controlled by the failure of the lowest story shear walls and is based on adjusting the forces obtained frem tne original design analysis to median-centered values. As discussed in Section 2.1.3.3, the median capacity is better represented by 0.90g (as compared to 1.05g given in the LGS-SARA). This change would increase the mean frequency of core melt by approximately 20 percent.
It was noted that the uncertainty value for modeling was only 0.1.
Because of the approximate nature of the analysis which was con-ducted, a value of at least 0.20 is more appropriate. In comparison, a modeling uncertainty value of 0.17 was used for testing in developing the fragil,ity for equipment, which gives an indication of a value for this factor that is more reasonable.
As discussed in Section 2.1.3.1, a ductility value of 2.5 assumed for the case of shear wall flexural failure is low. However, the effect of this value is balanced by the extra factor assumed for earthquake size effects used to adjust the hazard curves frcm sustained-based peak acceleration to an effective peak acceleration parameter.
CRD Guide Tube (S ) - The capacity of a CR0 guide tube is con-S trolled by functional binding of the control rod due to bending. The fragility parameters are based on test results coupled with the response of the guide tube calculated during the plant design. The test capacity was increased about 20 percent based on judgment since failure was not observed in the tests.
2-39
Since. the CRD guide tubes are. attached to the reactor pressure vessel (RPV) the comments above for the reactor internals, pertaining to use of a one-time analysis history and containment damping, also apply to the CR0 guide tube analysis.
Reactor Pressure Vessel (5 ) - The capacity of the RPV is due to 6
the potential failure in the weld between the connections of the top supports for the RPV and the top of the shield wall. An approximate analysis was used to determine the median capacity factor, wherein the total capacity was assumed to be equal to tne sum of the capacities frcm the support skirt and failure in the weld at the top support. A 0.1 uncertainty value was included for modeling, which, in the opinion of the reviewers, is small. Similar to the comments made for the reactor enclosure and control structure above, a value of at least 0.20 is I
appropriate for this type of approximate analysis. The effect of this si::e of increase in variability would have a small effect on the mean frequency of core melt.
The coments given for the reactor intervals, pertaining to one-time history and containment damping, also apply to the RPV capacity.
Hydraulic Control Unit (S ) - The components of the hydraulic 7
control unit consist of valves, tanks, piping, and electrical control s.
The fragility parameters are based on tests and fragility calculations performed for the Susquehanna nuclear pcwer plant.
In essence, the median capacity from Sasquehanna was scaled by the ratio of the two SSE peak ground acceleration values (i.e., 0.10/0.15). It is not apparent from the documentation in either the I.GS-SARA nor the sup-porting calculations for this component whether the SSE scaling from Susquehanna is appropriate. The concerns include possible' differences in the foundation condition and, hence, the response of the reactor enclosure, locations of the hydraulic control units in the two plants (i.e., is one unit higher, therefore it has a higher response?) and, finally, construction and, hence, similarity of the two units. These
_ issues should be addressed by PECo.
2-40 i
.J t
O The uncertainty for the spectral shape factce for this ccmponent appears to be conservative. The logarithmic stano. ' deviation values are based on the range of ratios between the test response spectrum (TRS) and the required response spectrum (RRS) at different frequen-cies. The total range of values for different frequencies and for the.
two horizontal directions were used to calculate the uncertainty val ue. If the components have similar dynamic characteristics and capacities in the two horizontal directions, the range should be bas-d on the minimum of the largest ratio in the two horizontal directions and the maximum of the largest ratio.
If this approaeh"is used, the uncer-tainty value is approximately one-third (i.e., 0.09 compared to 0.29).
Even' if the revised value is doubled for mcdeling uncertainty, the value used in the LGS-SARA will still be conservative.
The median capacity value also appears to be conservative, but was developed using considerable judgment. The minimum ratio of the.TRS and RRS valuts at the frequencies considered in the analysis was used. This value was assumed to represent the 95 percent level of survival (i.e., 5 percent would fail above this level) along with a 0.40 logarithmic standard deviation value. These two assumptions lead to doubling the minimum ratio to produce the median value. The final median value is essentially equal to the average of all ratios of the TRS to RRS values.
It should be noted that the total uncertainty logarithmic standard deviation value for the hydraulic control unit is 0.52 which is the~
highest value for any of the significant components. Although the uncertainty value for the spectral shape factor may be high, the total uncertainty appears to be reasonable considering other uncertainties due to modeling which have not been included.
SLC Test Tank (S ) - The capacity for the SLC test tank is based on 8
generic calculations for rigid equipment. This tank is supported on four columns and is not rigid. Based on inspection of this component during the plant tour, it appears to be very strong; however, the analy-sis performed for this tank is not applicable to the actual component.
2-41
--4y,
-t
_-y
o The capacity of.the anchor bolts which attach the base of the four columns to the concrete floor should be analyzed. The response factor shoul'd be recalculated taking into account the flexbility of the tank and the actual charactertistic of the four columns.
If the tension force in the columns or anchor bolts control the capacity, the earth-quake component factor may be as low as 0.71 (as compared to 1.04 which was assumed in the generic component analysis). Since the capacity may be controlled by a ductile element, a ductility value greater than 1.0 may be appropriate.
In summary, a component-specific analysis should be conducted for the SLC test tank.
Nitrogen Accumulator (SLC) (5 ) - The nitrogen accumulator is 9
described in the calculations as an 18-inch diameter by 48-inch nign tank which is anchored to the floor with six bolts. After visiting the Limerick plant, the reviewers are uncertain if the nitrogen accumulator which they saw fits this description. Since the capacity of this com-ponent is based on extrapolating an analysis from Susquehanna to the Limerick site, the similarity between the nitrogen accumulators at the two plants should be verified.
SLC Tank (S10) - The capacity for this tank is based on the buck-ling of the shell,' which was the weakest mode of the various modes of failure which were checked. One other possible failure mode is tearing of the base plate flange through which the anchor bolts penetrate. This failure mode apparently was not checked. There are no stiffening ele-ments in the vicinity of the anchor bolts, which may mean that tearing of the base plate flange is the weakest capacity. This possibility should be checked.
The uncertainty value for modeling error was assumed to be 0.10 which is small. A value equal to 0.20 would be more appropriate; how-ever, this change would have a small effect on the frequency of core mel t.
2-42
\\-)
e
440-V Bus /SG Breakers (Sti) - The capacity of this component was developed in a similar manner to the capacity for the hydraulic control unit, which also was based on test data from the Susquehanna nuclear power plant. The calculations, which were based on the ratios of the TRS to the RRS at different frequency values, are not clearly stated.
The minimum ratio was assumed to represent the 95 percent level of survival along with a 0.40 logarithmic standard deviation va.ue. These two assumptions led to doubling the minimum ratio. The final value is close to the average ratio (however, calculations of the average ratio are not apparent).
It is interesting to note that the uncartainty value for the spectral shape factor is only 0.08 which is much less than the value of 0.29 obtained for the hydraulic control unit (see c:mments above for the hydraulic control unit).
In summary, the fragility parameter values for this component appear reasonable, but it was not possible to check all the calcula-tions. Since this component is a significant contributor to the mean frequency of core melt, a specific analysis should be conducted for this component.
440-V Bus Transformer breaker (S12)*
125/250-V DC Bus (S I
13 s 4-KV Bus /SG (S14) -The capacities for these three components are the same and are based on the fragility analysis of the diesel generator circuit breakers. The only difference between the capacities of these three components and the diesel generator circuit breaker capacity is that the former components are in the reactor enclosure, while the later component is in the diesel generator building. Comments concerning these three components are the same as given below for the diesel gener-ator circuit breakers.
Because these three components contribute signficiantly to the mean frequency of core melt, a specific component analysis should be con-ducted for each.
2-43
Diesel Generator Circuit Breakers (SIS) - The capacity of the diesel generator circuit breakers is based on an analysis of test data for the Susouehanna plant. The approach used to develop the capacity factor is identical to the approach used for the hydraulic control unit (see coments above). The same issues for that component also apply to the diesel generator circuit breakers (and also the three components above, i.e., S12' 313, and 514).
Since this component is a significant contributor to the mean frequency of core melt, a specific analysis should be conducted for this component.
Diesel Generator Heat and Vent (S16) - The capacity of the diesel generator heat and vent is supposedly based on the fragility of the exhaust fan supports wnich are assumed to be the critical -link. How-ever, the actual fragility parameters are based on generic passive flexible equipment. The calculations for this class of equipment were specifically formulated for tanks and heat exchangers.
It is stated in
~
the calculations that shock test data indicate the capacity is 9.5g for the handling units; thus, the values used are conservative. However, since this component is a significant contributor to the mean frequency of core melt, a specific analysis should be conducted.
RHR Heat Exchangers (St7) - The capacity of the RHR heat exchanger was obtained by scaling the capacity factor for the same component at the Susquehanna nuclear power plant.
It is assumed in the calculations that the response factors for Susquehanna and Limerick are the same.
The controlling element is the lower support bolts.
The earthquake combination factor is 0.93, appears to be high since the columns supporting the RHR heat exchanger are located at the four corners of a square pattern. Since tension in the bolts is sign.ficant, the factor will be somewhere between 0.71 and 0.93.
2-44 i
.)
)
~
(
This component does not appear to be a significant contributor to the mean frequency of core melt; hence, small changes in the values of the capacity factors for the RHR heat exchanger do not appear to be critical.
2.1.3.8 General Fragility-Related Connents The following comments are made in order to inform the reader of potential issues which because of their philosopnical nature may not be resolved in the near future. Also, minor issues and errors wnich were found during the review are documented for completeness. The reader is directed to Reference 3 which gives a more detailed discussion of scme of these general issues.
As discussed in the previous sections, there are cases wnere the uncertainty values seem to be low. In particular, modeling errors appear many times to be smaller than what was expected.
In Section 5.3.1.4 of the LGS-SARA, it is stated that the coefficient of variation for equipment response factors is about 0.15.
Since this factor is very sensitive to the relationship between the equipment fundamental frequency and the frequency corresponding to the peak of the floor response spectrum, it is easy to visualize cases where a slight shift in frequency could mean a factor of 2 or 3 (or even more) in the value of the spectral ordinate. Thus the logarithmic standard deviation for response should be developed on a case-by-case basis.
In general, the uncertainty in some of the parameters has been understated.
In particular, there is uncertainty in using a simplistic analysis to obtain the capacity of a component which' was not recognized in the LG3-SARA. On the other hand, the median capacity values are probably on the high side. These two effects likely are self-compensating.
No uncertainty was assigned to the ground response spectrum factor used in the analysis. By definition this implies that this is the 2-45 h
absolute best (within the context of the analytical model) that can be achieved; hence, there is no motivation ever to conduct site-specific studies to improve the estimate of the frequency content of the seismic input. Although Limerick is a rock site, therb is still uncertainty in the ground response spectrum which should be included in the analysis.
It is believed that a reasonable value for uncertainty, if included, would have a small effect on the frequency of core melt.
The documentation of the basis for the fragility values does not carefully distinguish between the categories of infomation which were used. The use of subjective or data-based infomation (either analysis or testing) should be specifically noted ta infom the reader.
In addition, sensitivity analyses should be perfomed to indicate the robustness of the assumptions. This is particularly applicable to Chapter 3 where the fragility, hazard, and systems infomation is ccm-bined to produce the core melt frequency distribution.
The issue of dependency and its affect on the core melt frequency distribution was considered in the review of the LGS-SARA. Except for sequence T E uX, it appears that any additional capacity or response-3s related dependency effect: would not have a significant impact on the mean frequency of core melt. For the case of T E 0X, Section 2.1.3.6 ss discusses the implications if additional components were added to the series expression. For the current Boolean expression for the T E UX s3 sequence, if any additional dependency exists, the frequency of core r
l melt would decrease. As discussed in Reference 3, there are potential dependency effects which could effect the fragility values for cable trays anc piping systems, although it is likely that the current capa-city values account for these effects.(3)
Another important issue is the use of ductility factors for one degree of freedom (SDOF) models -to represent multidegree of freedom (M00F) structures or equipment.I3) Research is required to resolve this issue. At the present, not enough uncertainty is generally assigned for this situation.
2-46 i
(
e As discussed in Section 2.1.1, design and construction errors are not systematically recognized and quantified in the LGS-SARA. This is a particularly important consideration for components in series which could lead to a major failure if only one of the ccmponents fails. At best, the results of a seismic PRA can only be used to make relative comparisons.
One concern which was raised is potential leakage through internal ccmponents caused by seismic motion, thus bycassing a closed value barrier. This probably is not a major problem but should be formally verified by PECo. The MSIV and purge and vent valves are important examples. Also, the type of SRV used at Limerick has a history of sticking randomly in the opened position (i.e., failing to close after the signal is received).
The possibility that seismic motions could increase the likelihood of this type of failure should be addressed.
The effects of soil pressures on the buried walls of the reactor enclosure were not explicitly addressed in the LGS-SARA. Because the walls have as much as 40 feet of fill against them, they should be investigated to determine if the fill reduces the capacity of the reactor enclosure walls.
The potential for secondary components failing, falling, and impac-ting primary safety-related components apparently has not been syste-matically addressed since the plant is still under construction. The potential effects of block walls failing has been considered. Other components could also be a potential hazard. At the completion of construction, secondary components should be reviewed and their capac-ities incorporated into the LGS-SARA f f they are weaker than the primary components already considered.
On page 5-15 of Appendix B of the LGS-SARA, the value 648 K in.
should be 648,000 K-in.
This is believed to be a typographical error.
2-47 w
3 On page 5-60, the damping factor for valves appears to have been included twice (ance for the piping and once for the valves).
It was explained by SMA that only one fac' tor was used for both piping and for valves and is based on adjusting the damping used in the original design analysis (i.e, 0.5 percent) to a median-centered value (i.e., 5 percent).(24)
Toward the ccmpletion of the preliminary review, Section 10.1.6.5 was brought to the attention of JBA (other parts of Chapter 10 were not reviewed by JBA).
In this section, the effect of earthquakes on the effectiveness of evacuation was quantified for the various accident classes. The argument for limting upper-bound accelerations on the hazard curves given in Reference 18 was incorrectly used to establis'h that below 0.61g effective peak acceleration evacuation will not be impeded. This value was then used to develop the percent of occurrence when evacuation would be affected by earthquake. Although the arguments in Reference 18 are appropriate for establishing upper-bound accelera-tion limits for the hazard curves, the rationale has been incorrectly reversed. The result of this error means that the percentages of affected evacuations are much higher than given in Table 10-7.
PECo should reexamine the percentages and establish more realistic values and incorporate them in the offsite consequence analysis.
2.1.3.9 Closure The LGS-SARA differs from the IPPSS and ZPSS in that the mean frequency of core melt is dominated primarily by five electrical compo-nents in series, which have nearly the same median capacities.
In contrast, nonelectrical components and structures controlled the results of the IPPSS and ZPSS.
(Note that in Section 3.1.2, discrepancies concerning the Boolean equations are discussed which, if correct, may mean that the electrical components do not dominate the results.)
2-48
=
s 4-l-
i The capacities for the LGS-SARA electrical components are based on generic tests and are not component specific. This approach is reason-
~
able as long as the components do not control the final results. Since the electrical components are significant contributors, a more detailed analysis should be conducted. The recommendations given in Section 4.1.3 are directed to this goal.
2.1.4 References to Section 2.1 1.
Pickard, Lowe, and Garrick, " Indian Point Probabilistic Safety Study," Prepared for Consolidated Edison Company of New York, Inc., and Power Authority of the State of New York, Copyrignt 1982.
2.
Pickard, Lowe, and Garrick, " Zion Probabilistic Safety Study,"
Prepared for Consolidated Edison, Co.,
)t dated.
3.
Kolb, G. J., et al., " Review and Evaluation of the Indian Point Probabilistic Safety Study," Prepared for U. S. Nuclear Regulatory Commission, NUREG/CR-2934, December, 1982.
4.
American Nuclear Society and the Institute of Electrical and Electronics Engineers, "PRA Procedures Guide," Vol.1 and 2, U.S.
Nuclear Regulatory Commission, NUREG/CR-2300,1983.
5.
Cornell, C.
A., "Probabilistic Seismic Hazard Analysis: A 1980 Assessment," Proceedings of the Joint U.S.-Yugoslavia Conference on Earthcuake Engineering, Skopje, Yugoslavia,1980.
6.
Philadelphia Electric Company, Limerick Generating Station Final Safety Analysis Report, 1983.
2-49
7.
Cook, F. A., D. Albaugh, L. Brown, S. Kaufman, J. Oliver, and R.
Hatcher, " Thin-Skilni d Tectonics in the Crystalline Southern Appalachians: CCCORF Seismic Reflection Profiling of the Blue Ridge and Piedmont," Geology, Vol. 7, pp. 563-567,1979.
8.
- Seeber, L., and J.G. Armbruster, "The 1886 Charleston, South Carolina Earthquake and the Appalachian Detachment," Journ.
Geochys. Res., Vol. 86, No. B9, pp. 7874-7894,1981.
9.
Tera Corporation, " Seismic Hazard Analysis," Prepared for U.S.
Nuclear Regulatory Commission, NUREG/CR-1582, Vols. 2-5,1980.
10.
Diment, W. G., T. C. Urban, and F. A. Revetta, "Some Geophysical Anomalies in the Eastern United States," in The Nature of the Solid Earth, Ed.
E. C. Robertson, pp. 544-572, 1972.
11.
Sbar, M.
L., and L. R. Sykes, " Contemporary Compressive Stress and Seismicity in Eastern North America: An Example of Intraplate Tectonics," Geol. Soc. Am. Bull., Vol. 84, pp.1861-1882,1973.
12.
Fletcher, J. B., M. L. Sbar, and L. R. Sykes, " Seismic Trends and Travel-Time Residuals in Eastern North America and Their Tectonic Implications," Geol. Soc. Am. Bull., Vol. 89, pp.1656-1976,1978.
13.
Yang, J. P. and Y. P. Aggarwal, "Seismotectonics of the North-eastern United States and Adjacent Canada," Journ. Geophys. Res.,
Vol. 86, No. 86, pp. 4981-4988,1981.
14.
Street, R. L., A. Lacroix, "An Empirical Study of New England Seismicity: 1727-1977," Bull. Seis. Soc. Am. Vol. 69, pp.159-175, 1979.
2-50
-/
e
~
e men o O
15.
Nuttli, O. W., "The Relation of Sustained Maximum Ground Accelera-tion and Velocity to Earthquake Intensity and Magnitude," Report 16, Misc. Paper S-7-1, U.S. Army Engineer Waterways Experiment Station, Vicksburg, Miss., 1979.
16.
Cornell, C. A. and H. Merz, " Seismic Risk Analysis of Boston,"
Journal of the Structural Division, American Society of Civil Engineers, ST10, 1975..
17.
Trifunac, M. D. and A. G. Brady, "On the Correlation of Seismic Intensity Scales with the Peaks' of Reccrded Strong Ground Motion,"
Sull. Seis. Soc. Am., Vol. 65, pp.139-162,1975.
18.
Kennedy, R. P., " Comments on Effective 3round Acceleration Estimates," SMA Report 12901.04R, February, 1981.
19.
R. P. Xennedy, et al., " Engineering Characterization of Ground Motion Effects of Charateristics of Free-Field Motion on Struc-tural Response," SMA 12702.01, prepared for Woodward-Clyde Consul tants, 1983.
20.
Riddell, R., and N. M. Newmark, " Statistical Analysis of the Response of Nonlinear Systems Subjected to Earthquakes," Depart-ment of Civil Engineering, Report UILU 79-2016, Urbana, Illinois, August, 1979.
21.
McGuire, R. K., " Transmittal to Mr. Howard Hansell of Philadelphia Electric Company," July 12, 1983.
22.
Kennedy, R. P., et al., " Subsystem Fragility," Lawrence Livermore National Laboratory Report prepared for the U. S. Nuclear Regula-tory Commission, NUREG/CR-2405, October,1981.
2-51 r.
23.
Bohn, M. P., " Interim Recommendations for a Simplified Seismic Probabilistic Risk Assessment Based on the Results of the Seismic Safety Margins Research Program," Lawrence Livermore National Laboratory, Draft, April 15, 1983.
24.
Letter frem R. D. -Campbell,to J. W. Reed, dated July 11, 1983.
O e
Y 2-52
.s e
4
M A
- Tabl e 2.1 Comparison of Mean Frequency of Core Melt Values Contribution to Mean Frecuency of Core Melt Seauence Accroximate Analysis LGS-SARA Values T E UX 4.0-6*
3.1-6 3s TR 9.5-7 9.6-7 3B T PPV 4.4-7 8.0-7 s
TECC 33m2 6.0-7 5.4-7 TRC 3.5-7 1.4-7 sSm TEW 1.1-7 1.1-7 ss Total 6.5-6 5.7-6 4.0-6 = 4.0x10-6 l
T 2-53 t
s
+. - ~a-n,--..--.-.
.a+----
se..,,-,,
,..,_,.-n,--.-
Table 2.2 Hazard Curve Contribution to Mean Frequency of Core Melt Contribution to Hazard Curve Mean Frecuency of Core Melt Percentace Decollement 3.0-6 26.2 Piedmont, M
= 6.3 2.6-6 40.1 max Piedmont, M
= 5.3 5.7-7 8.8 max Northeast Tectonic 2.4-7 3.7 Crustal Block, M
= 6.0 6.2-8 1.0 max Crustal Block, M
= 5.5 1.5-8 0.2 max Total 6.5-6 100.0 1
2-54
)
.s
l Table 2.3 Hypothetical Mean Frequency of Core Melt (Based on Individual Hazard Curves)
Individual Mean Frequency Ratio to Hazard Curve of Core f.'el t 6.5-5 val ue Deco 11ement 3.0-5 4.5 Piedment, M
= 6.3 1.7-5 2.6 max Piecmont, M
= 5.8 3.8-6 0.58 max Northeast Tectenic 8.0-7 0.12 Crustal Block, M
= 6.0 4.1-7 0.06 max Crustal Block, M
= 5.5 1.0-7 O.C2 max s
2-55
o 5E A
I E
53 ut c.
=
ma khiw M M 3 5 M5$hf t
3 1 *=
4 3 21 1
i
=
J 4
e e O 0 0 00 ***
y k
i p
w *pr{ih~ '
~
o i.
O.
FJ oi O
^$
p i
1
,2
- 0
).
r e I4
. [_0
,;;..%o g Lt 1
@o:
[~ ~ ^'~ ~ ~ ' -
at -
---~~~~~
o p._;, ? _: -. _.
3 o
)
m F"p EO %
v.
i.
y
- 0. rpe g 5'n i
v i
3 8
O.
^
t
?,
s
=
d.'c..e O,
a o
4,
-j e
e i,
__A__g,-
i.
49 s
'-~~T g
o O,.'
t8 s
I i
a o
o f""1 u
/
9
- C C
.c x
c.
(
I O
_ _ _ _ _ 1 4, _
- n e
____)
o o
i
-'~~~
~,~ 5 ~s~
s
=
c-O a
o,lh,
g g
e c
8 G,
3 e
e,1 g_9__
i.
a a
q
_,_c_ (--
=
o a
e o
I i
o i
N
.o.
o i
_i_
D._ _ _ _ _o_ _I o i
7,_
ry
___3 u
r 3
m M
3 1
O r==
e N
o Y
\\
.?
.J m
2-56
-a 10-2 Historic Seis-icity Cur /es 1.
bin = IV, La = X, b = 0.25 2.
bin
- IV, ha = X, b = 0.72 3.
Imin
- IV. In a = VI, b = 0.72
\\
LGI-S;PA Seismicitv Curies 1
Decoll e.
g g
- 5. Crustal Block, l1 = 5.5
?._
\\
s i
\\
N 5
- 2 \\
\\
\\
~
N
=
\\
N x
N E
\\
N k
5-
\\
9
% 4 C
10 \\
g
\\
2 E
\\
\\
\\
10-6_
\\
S 3
10-7 i
i i
i e
i 4
0 0.20 0.40 0.60 0.80 Peak Grouna Acceleration (g)
Figure 2.2 Comparison of various historical seismicity curves and the LGS-SARA seismicity curves from Appendix A for sustained-based peak acceleration for tne Decollement and Crustal Block, ti=5.5 seismogenic zones.
~
2-57
., \\
2-58 2.2 FIRE
' 2.2.1 Octeministic Fire Growth Modeling 2.2.1.1 Introduction A deterministic fire growth model is used in the Limerick SARA to provice fire growth times. These times then serve as input to the probabilistic mcdel from which the likelihood of a particular fire grcwth stage'is datermined given an initial size fire. The deterministic model contains the methocolcgy which explicitly incorporates the physics of enclosure fire develeccent.
The Limerick SARA uses the ccmputer code CCMPBRN[1,2] as its determinis-tic fire growth model. Briefly, this code is a synthesis of simplifieo, quasi-steady unit models resulting in what is commonly called a zone aporoach model.
A detailed evaluation of this code and its application in the Limerick SARA appears later in this review. There are many other ccmputer codes [3-C which use the unit-mcdel approach to model ccmpartment fire develeccent. Of particular interest is the DACFIR Code [83 developed at the University of Dayton Research Institute, which models the fire growth in an aircraft caDin as it progresses frca seat to seat. This is analogous to the problem of fire spreading frem cable tray to cable tray as analyzed in COMPBRN.
At this point some general thoughts are deemed warranted on the complexity of fire phencmena and the state of fire science with regard to enclosure fire development. Computer models of enclosure fire development appear capable of predicting quantities of practical importance to fire safety, provided the model is supplied with the fire-initiating item's empirical rate of fire growth and the effect of external radiation on this rate. As a science, how-ever, we cannot predict the initiating item's growth rate due to relatively poor understanding of basic combustion mechanisms. Questions and doubts have even been' raised regdrding the ability to predict the burning rate of a non-spreading, hazardous scale fire in tems of basic measurable fuel properties.
However, while awaiting development of meaningful standard flammability tests and/or more sound scientific predictions, realistic " standardized" fire test J
l 2-59 procedures should continue to be formulated for empirically measuring the rate of growth of isolated initiating items, the attendant fire plume, its develou-ment within an enclosure, and the convective and radiative heat loads to
" target" comoustibles, hus, in lieu of large-scale computer coces to assess the fire hazard in an enclosure, the unit-problem approach (as used in CCMPBRN) is about the best that can be taken at the present time.
.i However, due to the state of infancy of fire modeling alluded to acove, many judgemental assumptions in both modeling and pnysical data must be mace in arcer to model fire development in tne ccmoiex enclosures existing in
- nuclear power plants. Additional complexity is introduced when one considers the fuel as electrical cable insulation rather than the more commonly con-sidered fuels such as wood or plastic slacs, wnich may nave a more uniform ccn-position than cable insulation..
In fact, as discussed later, some of tne models used in CCMPSRN are non-p hys ic al. That is, while usually leading to results wn1cn are hignly con-servative, these models do not adequately reflect the dependence on the pnysi-cal parameters which are evidenced in experimental data. Other models, assumptions, and omissions in the application of CCMPBRN to the Limerick SARA are either conservative or non-conservative.
This conbination of non-physical models anc conservative as well as non-conservative assumptions leads to very large uncertainties in the detenninis-tic modeling process.
It is therefore also difficult to quantify the effects of these uncertainties on the'probabilistic analysis, since the latter uses the results of the detenninistic analysis as input.
Indeed, as a general con-ment, one wonders whether more is gained by making gross judgemental assump-tions, using them in an uncertain detenninistic methodology and " cranking" the results through a probabilistic analysis, than would be gained by making direct judgements on the risk of fire.
In any case, we will evaluate the modeling and assumptions of the COMPBRN code and its application in the Limerick SARA in following, sections. Section 2.2.1.2 briefly summarizes our concerns with the deterministic modeling, while Section 2.2.1.3 gives a more detailed discussion of each item. Some suggestions for reducing the uncertainties are given in Section 2.2.1.4.
2-60 2.2.1.2 Summary Evaluation of Deteministic Fire Growth Modelina The dcceministic methodology contained in the computer code CCMPSRN[1.23 is used in the Limerick SARA to evaluate the thermal hazards of postulated fires in tems of heat flux, temperature, and fire growth. This coce employs a unit-model approach which is acceptable given the current state of the art in enclosdre fire modeling as discussed in the previous section. However, we find some of the sub-models contained in the cede to be non-physical, scme as-sumptions highly overco'nservative, while other assumptions and applications yield non-conservative results. The uncertainties arising frca the coccina-
, tion of these counterbalancing models and assumptiens are difficult to 7
quantify, but if forced to draw a conclusion we feel the deteministic analy-sis as applied to the Lir arick plant is generally on the conservative side.
Mcwever, we also wish to restate that we do not feel that the countercalancing of a non-physical, non-conservative model or assumption with another ncn-physical model or assumption, no matter hcw conservative, leads to a result which is useful in quantitative terms.
Based on our initial review of the deteministic fire modeling in the Limerick SARA, we have identified the following items of concern, whicn will be discussed in more detail in the next section.
The burning rate model is probebly the most important source of uncer-tainty in the CCMPSRN code. The methodology employed is not realistic ano can lead to results which are dependent on the arbitrary choice of the size of
" fuel elements" into which the fuel bed is discretized. Instead the fuel burning rate should be dependent on the instantaneous size of the fire. Al so use has not been made of existing cable flammability data.[9,103 It is difficult to detemine if the cable insulation burning rates obtained by this method are conservative or non-conservative. For the postulated transient-combustible oil fire the burning rate considered appears overconservative with respect to that reported in the literature.[ll]
Another example of non-physical modeling is the fuel element ignition time relationship. This model yields a finite fuel ignition time even if the inci-dent heat flux is considerably below the critical value of 20 kW/m2 found necessary to initiate cable insulation damage in experiments.[12] The model m_
, 9
-rs,
._cy_.9. g
,,,__m.g-6,--*
r'W-F
- -. = = - '. -
t' m'+-u""*
=='-="*'=.-<---" awe
~
2-61 assumes a constant input heat flux even when cables in a convective plume are considered. Convective heat flux must be a function of the difference between the plume and target temperatures, and must therefore decrease as the target fuel heats up. A cable damageability criteria based on a critical heat flux and an accumulated energy, as discussed later and in Ref.12, would be more appropriate. The model used in COMPSRN leads to highly conservative cable ignition times.
The mcdel used to calculate the radiative heat transfer frem the flame to a target ocject is also overly conservative. The radiative heat flux cotainec frcm this model is much greater than that obtained frca a classical Stefan-Boltzmann model, wherein the heat flux is a function of the flame gas tempera-ture to the fourth power. The CCMPSRN acdel also neglected the attenuation of the heat flux with distance cue to intervening het gas cr smoke. The mocei neglects, too, tne partial reflection of the impinging ractative neat flux frcm a target fuel element, as well as re-radiation, convection, anc cther losses.
Additional conservatism is introduced by assumptions made concerning tne three stages of fire growth. The second stage considers fire growtn to adjacent cable raceways once an initial raceway is ignited. The analysis assumes adjacent cable raceways are separated from the initial fire by the minimum-separation criteria specified for redundant safety-related cable raceways (5 feet vertically and 3 feet horizontally).
In other words, only one calculation of fire spread time is made for this configuration and the results are applied to all plant areas considered. This will yield a highly conservative upper bound calculation. Growth stage three assumes damage to redundant cables separated by 20 feet and up to 40 feet and those protected by fire barriers. Redundant raceways separated by more than 20 feet from the initial fire were assumed to be damaged in a time interval equivalent to the damage time of a fire barrier taken as a 1 inch thick ceramic-fiber blanket.
This appears conservative since raceways separated by this distance would usually be damaged by convection in a stratified ceiling layer, and therefore there should be some dependence on the height of the raceway from the ceiling, those cicser to the ceiling failing earlier than those below.
Intermediate growth stages between stages two and three might be appropriate.
1 2-62
)
i Another area of uncertainty concerns the quantity and size of the assumed transient-combustible fires. The Limerick SARA assumes three possible trans-ient combustible configurations; 2 pounds of paper i foot in diameter,1 quart 1
of solvent 0.5 foot in diameter, and 1 gallon of oil 1 foot in diameter. No rationale is given for this selection.
It is certainly possible for larger quantities or ccabinations of these fuels to exist in nuclear power plants. A distribution of varying quantities would be more appropriate. Also, it is not clear that given 1 gallon of oil that a 1 foot diameter pool represents tne most severe hazard. A larger diameter pool will give a larger neat release al-thougn for a snorter duration. The damage sustainec by the target cable may
- be a function of this combination of heat flux level and duration of imposi-tion.
Scre consicerations omitted frcm the Limerick $?PA would tend to make the analysis non-conservative. These include the effects that enclosure walls and corners, in close proximity to the initiating fire, have on the convected heat flux and the possibility of cable damage cue to convection in a stratified ceiling layer.
2.2.1.3 Detailed Evaluation of Deterministic Fire Growth Modelino 2.2.1.3.1 Fuel Burnino Rate The CCMPSRN code [13 models the specific burning rate, 6", of the fuel, which is equivalent to the mass loss rate in combustion, for fuel surface con-trolled fires as h"=m"+Cs4" ext (2.1)
The tem h"o is defined as a specific burning rate constant and the second tem represents the effects of external radiation on the burning rate.
The specific burning rate constant is assumed to represent the effects of flame radiative heat flux to the surface, q"fl, r, and surface reradia-tion, q" loss S" = (q"f), p - q" loss)N where L is the heat required to generate a unit mass of vapor. Note that the use of H, the heat of combustion of the fuel, in Eqn. (4.4) of Ref.1 is F
incorrect. The correct fomulation is given by Eqn. (3) of Ref.13.
...N
.. ~ - _ _.
2-63 Note that if the exteraally applied heat flux, q" ext, is zero, the object will burn at a constant rate given by 6"=5"o.
The consideration of h"o as a constant for an element of fuel burning during the early growth stages of a fire is questionable. For non-charring combustibles, such as PMMA or Plexiglas, experimental data indicates that 5"o is indeed a constant.
However for complex solid fuels such as electrical cables,, this may not be tne case. Also, the burning rate is a function of the size of the fire tnrougn 6"fl,r anc q" loss. The mass loss rate of a small samcle of PE/PVC caole, suojected to a constant external neat flux, i s shown in Figure 4.4 of Ref.10.
The mass loss rate is certainly not constant with time as would be incicated by Eqn. (2.1) with 6"o and q" ext constant by definition.
In CCMPSRN, Ecn. (2.1) is apolied to each small square " fuel element" into which the individual cable trays (super modules) have been discretized. The fire is assumed to initiate in one element and spread to adjacent elements wnen their ignition criteria is reached due to the inc1 dent radiation from tne 2
i.itial fire. A constant value of A"o = 0.002 kg/m -sec. is chosen for each element. This methodology results in a non-physical condition when the couplete cable tray is considered since the specific burning rate becanes a function of the aroitrary number of elements into which the tray is divideo.
For instance, if a fuel element was burning in infinite space witn no externally applied heat flux, then according to Egn. (2.1) its burning rate would be 5" tot"b"o.
However, if this fuel element is divided into two contiguous subelelements (1) and (2) with equal areas A/2 and with tne flame of sucelement (1) supplying the external heat flux to subelement (2) and vice versa, then according to Eqn. (2.1) 5"totib"o"[5"o+Csk" ext]
(2.3) where we have tacitly assumed that k" ext,1*k" ext,2"4" ext Likewise if the element we.re divided into n subelements with eacn j-th element supplying an external heat flux to every other element, by definition, the progressive total burning rate when each of the j-subelements becane involved will not be equivalent to the total burning rate had all the sucelements been involved initially. This indicates that care must be exercised in using Eqn.
(2.1) to predict the ensuing development of a fire along an individual cable tray.
2-64 Intermediate scale data for the EPR/Hypalon cable used at Limerick is given in Fig. D-18 of Ref. 9.
The cable weight loss for the twelve trays considered increases with time and a steady burning rate of 6.7 kg/ min was eached after about 37 minutes. This translates into a specific steady state burning rate of 0.008 kg/m2.sec. Use of such data and that of Ref.10 could remove some of the uncertainty of the present model.
For transient ccmbustibles, the fuel is not discretized and the specific burning rate is assumed to be the constant steady state value, 5"a.
Tacle D-4 of the Limerick SARA gives values of $", for pacer and oil of abcut
,0.061 kg/m2-sec. Hopefully, the value for paper is a misprint and should be 0.0062 kg/m2-sec. The value for oil seems scmewhat conservative since Ref.
11 gives a value of 0.04 kg/m2-sec.
2.2.1.3.2 Fuel Element Icnition In the CCMPBRN code, a fuel element is considered ignited simoly if its surface temperature exceeds a critical ignition temperature, T*.
Acdition-ally, the fuel elements are modeled as semi-infinite slabs and the losses frca the fuel to the environment due to re-radiation and convection are neglected.
An expression for the ignition time, t*, is obtained by solving the heat conduction equation, following page 75, Ref. 14, for the condition of a constant imposed surface heat flux, d"o.
t* = (3/4a)[k(T*-T )/q"o]2 (2.4) o This expression is not physically correct since it implies that an ig-nition time will be reached no matter how small a value of heat flux is ap-plied. Cable flammability test data [12] shows that cables are generally not damaged unless the heat flux is above a critical value of about 20kW/m2 due to heat losses at the surface.
Also, the assumption of constant imposed heat flux is overly con::erva-tive since the heat flux received by an object is a function of the object surface temperature, Ts, which increases with time as the object is exposed to the external flux.
~
~ ^
2-65 For instance, in the case of an oil fire 10 feet beneath a cable tray considered in the Limerick SARA,, the convective heat flux at the caole surface wilI be 6"o=h[T;-T3 (2.5) p s
where Tpt is the plume temperature at the cable height, T3 is the cacle surface temperature, and h the surface heat transfer coefficient. There fore,
the surface heat flux will decrease substantially as the temperature of the cable surface aoproaches the plume temperature. The CCMPSRN coce assumes tne surface temperature remains at its initial value for the duration of the fire.
For the i foot diameter oil pool fire considered in the Limerick SARA, we estimated the plume temperature at 10 feet above the fire using three retnocs.
These include two correlations of convective heat flux by Aloert,[15,15]
(one of wnich was used in CCMPSRNE13) and a more recent plume correlation by Stavrianidi s[173 The plume temperatures tnus cotained range between 370 K and 450*K.
These low values of plume temoerature indicate tnat caoles witnin the convective plume and located 10 feet above the fire, would never reacn their designated critical ignition temperature of 340*K.
Tnis indicates tne overconservativeness of Limerick SARA which precicts caole igntion in 4 minutes for this target / fire source configuration.
Of course, one must also consider the radiative neat transfer fecm tne flame to the target (the electrical caoles) in order to predict the time required for the cables to achieve this critical ignition temperature.
In this regard, audit calculations, using the method described in Ref.18, yields 2
a radiative heat flux, q"r, of 0.42 kW/m. This is based upon use of the following equation:
4 2
q"p = (oTf1 /s) (A /1 ) c (2.6) p where e is the Stefan-boltzmann constant; Tf1 is the flame temperature (1255*K)[173; 1 is the distane.e of the target from the radiating body (with a flame height of 5 ft[163. and a cable height of 10 ft,1 is equal to 5 ft, and A is the flame projected surface-area. The emissivity,c, was assumed p
~
>r m.,
,c y
2-66.
to be 0.3 (the sum of a gaseous value of 0.2 and a luminous scot value of 0.1)..This value of radiative heat flux, when added to the previcusly calculated convective heat flux, then yields a value of ignition time, t*,
_ (via Eqn. 2.4) markedly higher than the 4 minutes stated in the Limerick SARA.
Even using tne radiative heat flux model, as described in CCMPSR i, yielcs a similar value of_ radiative heat flux which is lower than that required cc achieve the critical ignition temperature of 840*X within a minutes. In CCMPSRil, the radiative flux is given by q"r=Fo_f1hr/Afl (2.7) where F _fl is the shape factor between the object and the flame, Af1 is o
the flame surface area, and is the heat radiated by the fire wnicn is r
expressed as hr * '(b (2 3)
In the above expression, y reflects the radiant output fraction (Y=0.4 as assumed in Ref. 1) and Q represents the total heat release rate of the fire.
In order to reconcile this wide disparity between ignition times reported and those calculated by the methods described above, "back" calculations using Eq'n. 2.4 indicated that an imposed surface heat flux, q"o, of approximately 12 kW/m2 is required to achi. eve a t* of roughly 4 minutes. This value is obtainable using the COMPSRti model, if Af1 in Eqn. 2.7 reqpresents the projected flame area (or pool area in this case) and not the flame surface This is clearly inconsistent with the methodology used to derive Eqn.
area.
2.7.
These audit calculations clearly point out that the results of the Limerick SARA are based upon an overconservative estimate of critical times to reach cable ignition.
)
~
o.
l 2-67 Even in the event that the radiative heat flux dominates the convective heat flux, the target will not absoro the total flux since significant amounts will be convected away. If a proper model for convective heat transfer, Een.
(2.5) is used, once the surface temperature increases above the plume tempera-ture, heat will be convected away from the target. reducing the effects of radiation.
The selection of 840 K as the spontaneous ignition tamcerature for EPR/
Hypalen cable is also somewnat conservative since Table 3-1 of Ref. 9 pre-
.sents experimental data showing that the critical temperature at or below which ignition cannot be achieved is 893*K for piloted ignition and is con-siderably higher for spontaneous ignition. Actually, as stated by Siu,[13 the concept of a threshold ignition temperature is scmewnat imprecise.
Ex-perimental data generally exhibit significant variations with further uncer-tainties arising if ill-defined caole insulation ccmpositions are involved.
The crucial issue is not wnether the fuel surface reaches a certain tempera-ture level, but rather if the heat gains by the pyrolyzing gases are great enough to overccme the losses and trigger the cccoustion reacticns, and if the resulting heat of gaseous comoustion is great enough to sustain the reaction.
Lee (12] has developed a set of. caole damageability criteria along these lines. For an applied heat flux, the time for spontaneous ignition is defined in terms of a critical heat flux, q"cr, at or below which ignition cannot be initiated and an accumulated energy, E, required for sustaining ignition.
t=E/(q" ext-6"cr}
(2.9)
Fig. 2.1 (attached) shows test data [12] for the inverse ~ of time to piloted ignition plotted vs. external heat flux for EPR/Hypalon cable. The slope of the straight line is 1/E. Also plotted is the ignition time model, Eqn. (2.4), using a critical spontaneous ignition temperature of 840*K.
The CCMPSRN codel is more conservative than even the piloted ignition data especially for low levels of external heat flux, i.e., a given external heat flux will give an earlier time to ignition than the data. Also, while the data shows no ignition below a heat flux of about 20kW/m2, the model pre-dicts an ignition time for all values of heat flux. The 10 minute ignition time for stage two, self-ignited cable raceway fires is indicated for reference.
e, ve1
,i%
g y----ws
.9--
,.y-w-
- ----.*w-t+,
w----,.-,
y_..y.g..
-,w-r-4 e----ty 9
9va--
,--g-s
,p wee-w-e y
3m+-+eg,-,im---,.--.
2-68 2.2.1.3.3 - Fires Near Enclosure Walls or Corners The COMPBRN code does not consider the effects that the close proximity'of walls or corners of an enclosure can have on the temperature distribution in the convective plume of fires. The gas temperature at an elevation above.tne fire will be increased by the presence of walls by a magnitude that can be theoretically estimated by considering initiating fires having " equivalent" heat release rates of 2 and 4 times the actual neat release rate for walls and corners.respectively. The neglect of this effect will have a non-conservative effect en fire growtn calculations, especially in Fire Zone 2 where cacie
' trays are stacked against the "J" wall.
Evidence of the increased gas temperatures at a given elevation accve a fire is available in the literature.
In Ref. 16, Epns. (3) and (4) illustrata tne concept of equivalent neat release rates mentioned above. Fig. 5 of the same reference shows test data of tne fire positioning effects on ceiling temperature. On page 119 of Ref.19, tne average plume temperature rise is found to increase by factors of 1.75 and 2.5 for fires adjacent to walls or corners respectiv' ely. Finally, Table A-1 of Ref._.20 shows the upper layer gas temperature is likewise affected by burner locations near walls and corners.
The increased gas temperatures in tne presence of walls results from tne effects of reduced cool air entrainment, which results in higher flames due to the additional distance needed for fuel vapor / air mixing. We are concerned with the distribution of energy, not just the maximizing of the overall energy. Even though the code considers ccmplete ccmbustion, which maximizes the heat release rate and the temperatues near the fire, the wall effect causes local temperature increases which must be considered to yield a con-servative result.
2.2.1.3.4 Stratified Ceilino layer The application of the.CCMPBRN code in the Limerick SARA failed to con-sider the stratified hot gas layer near the ceiling of enclosures even though such a model is included in the code. This assumption that enclosure effects are minimal may be valid since the fires considered are small with respect to
2-69 the size of the enclosure. However, in small fire zones, as the static inver-ter room.the hot gas layer near the ceiling could preheat the non-burning fuel elements and reduce their time to ignition. Some substantiation of the neglect of this effect should be included in the analysis.
' The consideration of therral stratification might also effect the cefini-tion of fire growth stages in the Limerick SARA. It is conceivable that unprotected cables near the ceiling, but horizontally separated by more tnan 20 feet frcm an initiating fire, could ignite prior to a cable closer tnan 20 feet but considerably below the ceiling.
This would tenc to have portidns cf
' fire growth stage 3 ahead of fire growth stage 2.
t The ceiling gas layer model in COMPBRN is based on a simplified steady gross heat balance. A uniform gas tempe-ature is assumec thrcugnout the nacer hot layer. Alpert[15] inoicates that the ceiling gas temperature cecreases with distance frca the ceiling, as well as witn radial distance fecm tne plume l4 axis. More recently, Newman and HillE213 nave developed a transient cor-relation for the heat flux below the ceiling of an enclosure containing a pool fire, which includes the effects of forced ventilation. This correlation shows a decrease in heat flux with distance below the ceiling, but contraryLtc -
Alpert, it indicates very little depencence on lateral separation. These s
works indicate that consideration in the Limerick SARA of all unprotected trays witn greater than 20 feet horizontal separation as equivalent in damage rating to a fire barrier as being an oversimplification.
2.2.1.4 Recommendations for Imorovina Fire Growth Modelino The previous sections have detailed some of the concerns we.have regarding the sometines non-physical, usually over-conservative, deterministic fire growth modeling in the Limerick SARA. There are four major areas where we feel the modeling can be made more realistic, thereby reducing the resulting uncertainties. These are the cable burning rate model, the fuel element ignition time model, the flame radiant heat transfer model, and the surface -
temperature dependance of the convective heat transfer model.
e
2-70 Incorporation of recent test data [9,10] on cable flammability into the determination of the burning rate of the EPR/Hypalon cables should give a more realistic representation of fire growth. Similarly, the use of a cable ignition /damageability criteria,[123 based on a critical heat flux and an accumulated energy, would yield cable ignition times more consistent with test data. An improvement of the model for calculating the radiated heat flux re-
,:eived by a fuel element, by using an appropriate flame araa and by con-sidering attenuation due to hot gases and soot, will resu't in more realistic
^
fire growth scenarios and establish a mot e correct proportionality between convective and radiative heating. Final:y, the convective heat transfer =ccel
- should take into account the instantaneous temperature of the surface of the object being heated. This will reduce the convective heat absorbed as the object heats up and will allow for convective cooling i.f tne coject temcera-ture exceeds the temperature of the local fire plume.
2.2.2 Probabilistic Fire Analysis Review For the Limerick Generating Station (LGS) L' nit 1, tne Severe Accident Risk Assessment-(SARA) study reports that fire accident sequences constitute a sig-I nificant portion of the overall public risk. Based on our review of the docu-ment, we have found no evidence wilich contradicts the conclusion that the risk -
of fire is significant. However, based on our understancing of the state of the art in fire PRA, and the existing inadequacies in botn pnysical and proba-bilistic modeling in this area, we would like to avoid any judgement based on the quantitative results presented in the LGS report.
In addition, the expec-ted large uncertainties associated with the quantitative results would suggest tnat less importance be given to the numbers. Hence, the scope of our review i s two-fold: first to identify the existing inadequacies in physical and probabilistic modeling in fire PRAs in general; and second to review and com-ment on the existing LGS report for the fire risk assessment.
The generic comments associated with the physical modeling of fire growth have been discussed in Section 2.2.1.
The level of conservatism used in the detenninistic ~ analysis has also been discussed.
In addition, fire growth modeling during the suppression phase will be described in the following sections *ich basically indicate that the LGS approach is again highly conservative. Concerning the specific approach and data implemented in LGS fire risk assessment, we have concluded that:
- e. om m
2-71 1.
The approach taken for systematic identification of critical plant areas is sound, and the LGS fire hazards analysis appears to have identified all these areas.
2.
The LGS fire analysis has adopted an appropriate data base for es-timating the frequency of fire in Nuclear Power Plants (f.PPs).
3.
The LGS analysis has generated plant-specific fire frecuencies using the data base and has taken into account the specific features of.the
' plant. In a few cases these estimates are unconservative.
4 The LGS analysis appears to have icentifiec all important safety ccm-ponents and cabling which are located in the critical fire areas ex-cept for Zones 44 and 47.
5.
The event trees for panel fires generated by the LG3 analysis snould be modified to take into account the layout of the panels witn respect to the critical portion of tne zone.
6.
The cumulative suppression distribution function generated in the LGS report does not seem to agree with available data.
7.
Suppression probabilistic modeling seems to be very conservative and is not representative of the actual case.
8.
The LGS analysis does not quantify the uncertainty of the final re-suits. The uncertainty bounds generated are merely judgemental.
Consistent with these conclusions, the following section discusses each item in detail.
2.2.2.1 Evaluation of Significant Fire Frecuencies in General Lccations In this part of the LGS analisis, the frequencies of fires in general locations were estimated based on historical fire occurrence data in HPPs.
The 7eneral locatio,ns for LGS were identified from the Fire Protection and Evaluation Report (FPER). The data base adopted appears to be suitable for estimating the frequencies of fires in NPPs. The point estimate frequencies calculated for the general locations.seem to be reasonable, but the uncer-tainty bounds were not determined. The fr'equency of fires for the individual fire zones was then calculated using the ratio of the weight of concustible i
material contained within a zone to the total weight of ccmbustible material I
a I '
,.. ~.,.,, -..
m
._m,
,,y.
,.,_.,_m__,,,____-m._,-
2-72 in the general location. Tnere is no justification for using tnis ratio for estimating the specific zone fire frequer.cy. However, the results of tnese estimations were used for the systematic identification of critical fire zo.;es tnrough screening analysis, rather than the detailed fire risk assessment.
For the detailed fire risk assessment, the fire occurrence frequency witnin eacn zone was estimated based on tnree cifferent mechanisms of fire initiation. These are: self-ignited cable fires, transient cc bustible firas, and distribution panel fires. Following are comments regarcing eacn type of fire occurrence frequency estimation.
2.2.2.1.1 Self-Icnited Cable Fires inree incidents of caole raceway fires nave been reported in tne cata base for NPPs. Two of them spreac beyond one cable tray and were estimatec to burn for 30 minutes before being extinguisned. The LGS report indicates tnat all these caele fires were attributable to bad caole splices and uncerrated caoles.
dased on a review of tne LGS data given in Tables D-1 and D-2 of tneir submittal, incicent 43 (Table 0-1) does not, seem to nave been caused by uncerrated cables or bad splices. Hence, we cannot agree witn the five-fold recuction of self-ignited caole raceway fire frequencies as indicated in tne LGS report based on :ne Limerick protection measures and flame retarcant cables.
It appears to us tnat a tnree-fold reduction should have ceen implemented for cable raceway, self-ignited fire frequencies in the Limerick plant.
In orcer to estimate tne frequency of fires witnin tne individual fire zones, the frequency per reactor year was weignted according to tne fraction of cable insulation weight in that zone to the total cable insulation weight in the control structure and reactor building. We cannot follow the logic behind this fractional weignting factor.
In our view, tne numcer of conductors and splices, the voltage / power ratings, geometric factors, etc.,
may De more suitable for weighting tne frequency of fire in each fire zone, ratner tnan simply tne insulation weignt. This matter of concern indicates that large uncertainties are present in the fire frequency estimates of various zones, e
1
l 1
2-73 2.2.2.1.2 Transient Comeustible Fires Three types of transient combusticle fires were included in the analysis.
The quantity and the area of eacn type of transient comcustible were con-sicered to be fixec. Tne state of tne art for fire risk analysis is to con-sider various quantities of transient camoustibles eacn with an assignec probability distribution. Hence, the effective camageaoility area and tne critical propagation time for transient ccmoustible fires are exoec ed to be in ene form of a districution. Considering that no cata are availaole, the
. frequency of fires for transient comoustibles estimatec in the LGS report 4
seems to be reasonable.
2.2.2.1.3 Power Distribution Panel Fires Tne estimated frequency of fires occurring in power cistribution panels was estimated cased on five reportea fires tnat nave occurred during 564 years of reviewed U.S. LWR experience. The point estimate of fire frequency witnin i
a power distribution panel was derived from these cata and seems to be reason-aDie.
2.2.2.2 Screening Analysis A systematic approacn is usec in tne LGS report for icentifying the critical fire areas.
In this approacn it is assumea that upon tne occurrence of a fire in a zone, all the equipment and cables in tnat zone will ce l
cisaoled. The core melt probability was tnen recalculatec and was multiplied 1
by the frequency of fire occurrence in that zone to provide a measure for screening analysis. Using this approach, the LGS fire analysis appears to have identified all the critical areas in the plant. The quantitative reassessment of their results are beyona the scope of this review. Based on our review of the FPER and the use of engineering judgement, tne critical fire areas identified by the LGS report seem to be reasonaole.
I 1
l 2-74 2.2.2.3 Probabilistic Modeling of Detection and Sucoression The probabilistic suppression / detection model used in the LGS stucy in the fom of a cumulative probability distribution to predict tne procacility of fail f ng to extinguish the fire within a time interval is based on actual plant data for automatic detection and manual suppression.
It is indicated that the data base for cable insulation fires reported by Fleming, et al.[22] was used to construct the suppression probability distribution.
This document was reviewed and the cumulative suppression / detection was reconstructed based on our interpretation of the data. A comparison of the curve constructed by 3.'<L with the curve given in the LGS report is made in Figure 2-2.
Tacle 2-1 presents the data used by BNL.
It is our understanding that in the LGS esti-mate of the suscression success probability, the self-extinguished cabinet fire incicents were included.
In our opinion, the LGS repor snould not take credit for the cata on self-extinguisned cabinet fires wnen estimating tne suppression success procability for the cable raceway fires.
In accition, the LGS report constructed the cumulative suppression probability distribution with the assumption that the longest suppression period is 1.3 hrs (based on the longest suppression period observed in the data base). We feel it is more appropriate to obtain a distribution fit to the data rather than the " eyeball fitting" used' by the LGS report.
In our analysis, the legnomal, exponential,
and Weibull PDFs were considered as the likely candidates. The cni-squarea goocness of fit for both the BNL and the LGS data indicates that the para-metric Weibull distribution is the best choice. A cumulative Weibull distri-bution F(x) can be defined by two parameters, n and o, and is given by F(x) = 1 - exp (-x/a)G (2.10) l The estimated (o,n) values for the BNL and the LGS data ce (0.615,13.5) and (0.458,6.83) respectively. A comparison of the original LGS curve with the modified LGS curve and the BNL curve is given in Figure 2.2.
Between the time interval of 30 to 75 minutes, the curve I obtained by the Weibull fit to the l
LGS data is essentially the same as the curve II, obtained by the " eyeball fit" in the LGS report. Outside the above interval, the difference observed is not expected to result in any significant change in the final fire PRA results. However, the curve III obtained by the Weibull fit to the BNL data l
shows that the LGS estimate of suppression success probabilities is higher at all times than curve III.
)
- s l
l
- 6 w,
- ~,,, -
,g 7
,,,,,c.g
,q,
+,-
l 2-75 In the LGS report, similar to other conventional probabilistic risk as-sessments, it is assumed that fire growth and suppression are two independent processes, and they are treated separately. This is one of the most important deficiencies of existing fire risk analyses which usually results in very con-servative values for fire-induced risk. The interaction between the fire growth and suppression will be discussed qualitatively in Secticn 2.2.2.4 The probability calculated by the LGS recort for fire prcpagation cut of a distribution panc1 was considered to be 1/25 = 0.04 This estimaticn was mace cased on the cata base whicn indicates that all of the five recortec distribu-
- ion panel fires were self-extinguished and none of tnem propagated out of the t
panel. It was conservatively assumed that one of these fires had the poten-tial to propagate.
In addition, a five-fold reduction was considered, cased on engineering judgement, to give credit to the IEEE 383 qualifiec filme-retardant cable insulations. This reduction may not be justified. The ccm-bustibility of cacle insulation can best be described through the sensitivity of the cables to various thermal environments, expressed as the change in generation rate of ccmbustible vapor per unit change in the flux receivec by the combustible. This value is usually denoted by "S".
The value of S is.
C.17(g/k ) for EPR/Hypalon ana 0.22 (g/ka) for PE/PVC cable insulation [23,10)
J Hence, 3 maximum factor of 2 may be credited because of flame retarcant caDie insulations.
Aaditionally, during a visit to the plant, it was noted that scme of the panels are air-tight. For these panels, we feel the probability of fire propagation is negligible, therefore the value used in the LGS report is conservative. For panels with louvers or openings, the value used in the LGS report may be unconservative.
In. general, we do not expect the impact of panel fires to change appreciably if more detailed analyses were performed.
2.2.2.4 Probabilistic Modelino of Plant Damace State Generally, three stages 'of fire growth and corresponding sta'tes of shut-down equipment damage were evaluated in the analysis. The first stage con-sidered is damage to components in the immediate vicinity of the source of fi re. The second stage considered is fire growth to adjacent unprotected
o 2-76 cable raceways separated frem the initial fire by minimum separation criteria (5 ft. vertically and 3 ft. horizontally). The third stage of fire growtn re-presents fire of sufficient severity and duration to damage the mutually re-dundant shutdown methods which may have cabling with a separation cistance of at least' 20 feet or protected by fire barriers. There are certain inherent assumptions in the analysis. These are:
1.
The rate of fire growth is not dependent on the suppression.
2.
A 20 ft, separation is considered to be equivalent to 1/2 hour fire barriers (1 in. thick ceramic olanket).
3.
Cable raceways separated by a distance of 40 ft. or more frem the fire source were considered undamaged by the fire.
4 It was assumed that long-term heat receval systems not recuirec until 20 hours2.314815e-4 days <br />0.00556 hours <br />3.306878e-5 weeks <br />7.61e-6 months <br /> into the fire-induced transient could be recovered by oper-ating valves manually and operating pumps locally. The precacility of failure of the operator to perform tnese recovery actions was con-sidered to be 10 times greater than human errors ascribed to internal events.
Given these assumptions, the LGS report analyzed the impact of fire in various critical zones as identified through the screening analysis.
Identification of various equipment damaged in different fire growth stages could not be verified by the BNL review group due to lack of information ano time limitations. However, based on a limited ' identification of various critical components and systems in different fire zones by means of the information gathered from LGS-FPER and the plant visit, we concluded that in most cases, the LGS report identified the components properly. There are two l
exceptions th'at are given as follows:
i 1.
In Zone 44, BNL has identified seven distribution panels and motor control centers. These are distribution panels 10D201,100202,100203 and motor control centers 10B211, 10B212, 10BV215 and 108216. We have also concluded that a fire in distribution panels 10D202 and 100203 1
.,)
b
eb 2-77 would affect the operation of the HPCIS, and a fire in distribution panel 100201 would affect the operation of the RCICS. Hence, tnere are three critical panels in this area. The LGS report indicates tnat there are six distribution panels in this area and only two of them are critical (100201 and 100203).
2.
During the plant visit, it was noted that in Zone 47, General Ecuip-ment Area, there is a booster fuel pool cooling purp in the vicinity of the northeast corner, unich is the critical area in this cce.
This pump was not identified in the LGS rescrt. Therefore, its poten-tial for intitiation and progression of fire causing an adverse effect on the cables in this area was not considered.
Sefore presenting our ccmments on eacn critical fire zone, it is aparap-riate to discuss further the inherent assumptions used in :ne LGS report as mentioned earlier in this section. More specifically, we would like to dis-cuss the interacting nature between fire growth and suppression activities. In the LGS report, it was assumed that a fire can progress regardless of sup-pression initiation, but teminates with scme probability after an expected time which is required for successful suppression. The lack of physical mcd-eling for the suppression phase of a fire scenario appears to be one of the weakest links in the analysis. We are aware that this deficiency exists in other fire PRAs and it seems to be a conventional practice, usually resulting in very conservative estimates for fire impact on equipment and cabling.
While reevaluation of the results given in tt:e LGS report, taking into account l
proper detection and suppression modeling, is beyond the scope of this re-i view, it seems necessary to discuss the basis for such analysis.
l In trie analysis of a fire scenario, initiation time for detection and sup-pression is of great importance. Detection and suppression can be achieved j
either manually or automatically. In a detailed fire PRA, both detection time and suppression initiation time should be expressed in the fem of probability distribution function (pdf); For the automatic suppression and detection re-sponse, sore design charts are available which graphically, or through some equations, detemine the response time vs. the spacing, ceiling height, and heat release rate. [24,25,26] If detailed fire growth modeling, with the as-sociated uncertainties of various fire parameters, is available for a specific
2-78 secnario, the detection and suppression response may be directly estimated in the form of pdfs.
If detailed fire growth modeling is not available, a
- generic response can be considered by assuming the two extreme fire growtns (slow, fast) as defined in Ref. [24].
In this case, the lower and upper bounds for responte time may be determined assuming fast or slow fire growth, respectively. These bounds may be used to define a pcf for'the response. The response time for the inititiation of the manual sucpression may be estimated by means of available data on response time during fire drills and some en-gineering judgement. The moceling of a fire grcwtn during the suppressicn j
phase can be very cceplicated depending on the governing mechanism of the process (heat removal, chemical reaction, oxygen removal.) However, for the purpose of fire PRAs, a ccabination of simplistic models, coupled with em-i pirical correlations, may be usec. For example, the' effect of sprinkler sys-tems en fire growth may simply be ecceled in the form of gicbal energy cal-acce.[27]
In conclusion, the time in which fire can reach various stages of greuth is dependent on suppression initiation time. There is a strong belief that fire cannot grow'significantly once the suppression has begun.
In the LG'S.re-port, it is conservatively assumed that probabilities of various stages of growth can be determined using the time period for the completion of success-ful suppression, rather than the initiation of suppression. This is a very conservative assumption and at present the effect of this conservatism on the final results cannot be evaluated.
2.2.2.4.1 Zone Soecific Comments In addition to the generic comments made in previous sections, there are additional zone. specific ccaments that may impact the results of the fire PRAs given in the LGS report. These ccoments are mostly associated with the layout l
of different ccaponents in various critical zones and they are based on the j
review of the FPER and the plant visit.
e y---
.yv-
.-m y+
-m
-w-y=m_..
ywe,-
vvwv.
9 v--ww, 7
myyw-a.--
9-w,-,.c--_
e---y-9-wy,-mec.w-g
-=w--
-wt ty
,iu---ee- - = -i-=i-
2-79 Zone 44, Saf eguard Access Area (CH=36f t., A=8930 f t.2, ASD=357.2, a.
S=M).*
In this zone, there are a total of seven notar control centers (MCC) and distribution panels. Four of these panels are located cicsa to the critical corners. These are districution panels 100202 in SW, 10D203 in NE, 100201 in SW, and MCC-1CB211 in SW (Drawing M119, Rev.)
The event tree associated with the panel fires should be modifiec.
Zone 45, CRD Hydraulic Equipment Area (CH=25f t., A=12860 f t.2, b.
ASD=676. Ef t, S=M/A) The only critical panel wnicn is located in tne NE corner is the t'CC-1C3224 The other panels are not locatec in ne vicinity of the NE corner (Drawing M119 Rev.19). The event tree as-sociated with the panel fires should be modified.
Zone 47, General Ecuipment Area (CH=?, A=98CO f t.2, 233.ago f:,, s,
c.
M/A). Basec on tne drawing M-120, Rev.18, ncne cf ne cistricutien panels, load centers, or motor control centers are located in the vicinity tf the critical NE corner.
Theref ore, tne event tree as-sociated with panel fires in this zone should be modifiec.
The only component that may result in a fire hazard and whicn'is located in tne NE corner cf this zone, is a booster f uel pool cooling pump.
i
The "S=M" repres5nts manual suppression where "S=A" represents automatic suppression.
r N
w 0
4 9-._.,,--
..y
,r m,
-_g e-
,-y_._.. - _
e,.-, _ _.
,__,-9
2-80
2.2 REFERENCES
1.
Siu, N.0., "Probabilistic Models for the Behavior of Compartment Fires,"
School of Engineering and Applied Science, University of California, Los Angeles, Ca., NUREG/CR-2269, August 1981.
2.
Siu, N.0., "CCMPBRN - A Computer Code for Modeling Ccapartment Fires,"
School of Engineering and Applied Science, University of California, Los Angeles, Ca., UCLA-ENG-8257, August 1982.
3.
Mitler, Henri E. ar.d Emmons, Howard W., "Occumentation for CFC7, the Fifth Harvard Cc=puter Fire Ccde," Harvard University, Camaricge, Ma.,
October 1981.
4 Quintiere, J.G., " Growth of Fire in Building Cc=partments," ASTM Special Technical Pub. 614, 1977.
5.
Tatem, P.A., et al, " Liquid Pool Fires in a Ccmolete Enclosure," 1982 Tecnnical Meeting, the Eastern Section of.the Combustion Institute, Atlantic City, N.J., Cecember 14-16, 1982.
6.
Zukowski, E.E. and Kubota, T., "Two Layer Modeling of Smoke Movement in Building Fires," Fire and Material, 4, 17, 1980.
7.
Delichatsios, M.A., et al., " Computer Modeling of Aircraft Cabin Fire Phenomena," FMRC J.I. OGONI.BU, Factory Mutual Researcn Corp., Norwood, Ma., Cecemoer 1982.
8.
MacArthur, C.D., " Dayton Aircraft Cabin Fire Model Version 3," Vols. I and II, University of Dayton Research Institute, 1981.
9.
Sumitra, P.S., " Categorization of Cable Flammability, Intermediate-Scale Fire Tests of Cable Tray Installations," EPRI NP-1881, Electric Power Re-search Institute, Palo Alto, Ca., August 1982.
- 10. Tewarson, A., Lee, J.L and Pion, R.F., " Categorization of Cable-Flammability, Part 1: Laboratory Evaluation of Cable Flammaoility Para-meters," EPRI NP-1200, Electric Power Research Institute, Palo Alto, Ca.,
October 1979.
- 11. Tewarson, A., " Fire Behavior of Transformer Dielectric Insulating Fluids," 00T-TSG-1703, prepared for U.S. Dept. of Transportation, Trans-portation Systems Center, by Factory Mutual Research Corp., Norwood, Ma.,
September 1979.
- 12. Lee' J.L., "A Study of Damageability of Electrical Cables in Simulated Fire Environments," EPRI HP-1767, Electric Power Research Institute, Palo Alto, Ca., March 1981.
- 13. Tewarson, A., "Physico-Chemical and Combustion / Pyrolysis of Polymeric Materials," NBS-GCR-80-295, prepared for U.S. Dept. of Commerce, National Bureau of Standards, Center for Fire Research by Factory Mutual Research Corp., Norwood, Ma., November 1980.
.j 9
~
2-81 14., Carslaw, H.S. and Jaeger, J.C., " Conduction of Heat in Solids, 2nd Ed.,"
0xford Clarendon Press, 1959.
- 15. Alpert, R.L., " Calculation of Response Time of Ceiling-Mounted Fire Ce-tectors," Fire Technology, Vol. 8,1972, pp.181-195.
- 16. Alpert, R.L. and Ward, E.J., " Evaluating Unsprinklered Fire Hazards,"
FMRC J. I. No. 01836.20, Factory Nutual Research Corp., Norwood, Ma.,
August 1982.
- 17. Stavrianidis, P., "The Behavice of Plumes Above Pool Fires." a thesis presented to the Faculty of the Cepartment of Mechanical E.6gineering of Northear;ern University, Boston, Ma., August 1980.
- 18. Orloff, L., " Simplified Radiation Modeling of Pool Fires," FMRC J.I. No.
OE1EO.BU-1, Factory Mutual Research Corp., Norwood, Ma., April 1980.:
- 19. Zukoski, E.E., Kubata, T., and Categen, B., "Entrai nment in Fi re Plumes,"
Fire Safety Jour.ial, Vol. 3, 1980/81, pp. 107-121.
- 20. Steckler, K.D., Quintiere, J.G., and Rinkinen, W.J., " Flow Incuced by Fire in a Compartment," National Bureau of Stancarcs, NSSIR 82-2520, Septemoer 1982.
- 21. Newman, J.S. and Hill, J.P., " Assessment of Exposure Fire Hazards to Cable Trays," EPRI-NP-1675, Electric Power Researen Institute, Palo Alto, Ca., January 1981.
- 22. Fleming, K., Houghton, N.J., and Scaletta, F.P., "A "etnodology for Risk Assessment of Major Fires and its Application to an HTGR Plant,"
GA-A15402, General Atcai-Company, San Diego, Ca., 1979.
- 23. Tewarson, A.,
"Damageability and Combustibility of Electrical Cables,"
paper presented at FMRC/EPRI Seminar, Factory Mutual Conference Center, Norwood, Ma., December 1981.
24 Benjamin, I., et al, "An Analysis of the Report on Environments of Fire Detectors," Ad Hoc Ccemittee of the Fire Detection Institute, 1979.
- 25. Newman, J.S., " Fire Tests in Ventilated Rooms - Detection of Cable Tray and Exposure Fires," EPRI NP-2751, February 1983.
- 26. Hill, J.P., " Fire Tests in Ventilated Rooms - Extinguishment of Fire in Grouped Cable Trays," EPRI NP-2660, December 1982.
- 27. Levinson, S.H., " Methods and Criteria for Evaluation of Nuclear Fire Protection Alternatives and Modifications," Ph.D. thesis, Rensselaer Polytechnic Institute, Troy, N.Y., December 1982.
-.gw-._
,7.-
+sm,~
ow--
2-82 TABLE 2-1 Suppression Data and Calculations Performed f or Suppression Success Probacility Self-Ignited Cable Raceway Fires Index* P.lant Name Time :: Bring Fire Type of Type cf Ur.cer Control (hr)
Detection Sucoression i
58t Browns Ferry
- 7. 0 Autcmatic Manual I
23 Zion 2 1.3 Manual /Autcmatic ManuaT7 Automatic 25 San Cnofre 1
- 0. 7 Manual Manual 24 San Onof re 1
- 0. 5 Manual Manual 8
Kewaunee 3.5 Automatic / Manual Automatic / Manual 23 Three Mile Is. 2
- 0. 5 '
Manual Manual 37
- Vermont Yankee 0.5 Automatic Panual 42
.1ine Mile Pt. 1 0.05 Manual Manual 46 Oyster Creek 0.05 Manual Manual 27 Trojan 0.05 Manual "anual
- Indices are the same as those in Fleming's report.[22]
tThe fire occurrences during the construction phase or those that were self-extinguished and confined to a caoinet were not included.
In addition, the Brown's Ferry fire indicated above is not included in our analysis.
5 l
.)
se
+-r 9,ii 9
,.--,,,_,g---p, p.,_q.--Aw--,------g r
g.
y,,,,7,-,+,-.-+,.,-,r9-rw,--,,-
-.w,w.3,
,,- w
,w,,,
.w.
l 2-83 13 i
i i
i i
i i?
II
~
COMP 8RN AUTO-lGNITION 10 Ts* = 8 40 *X
\\
u i
9 9
x
'a 8
z9 7 t-z2 6
c w
Ho 5 d
c.
$4 w
2P 3 i
h l
2 l
l t =lO min.
O O
10 20 30 40 50 60 70 80 2
EXTERNAL HEAT FLUX (kW/m )
Fig. 2.1.
Piloted ignition of EPR/Hypalon cable under various external heat flux.
l l
l l
l L
2-84 1.0 m
e I'
g II T
I
.t
~
E III
.5 a
0 0.5 C
.a.
O o
a 40
'o I
eL 1
.?
l u
2 o
l Y
0 15 30 45 60 75 9O
~
Time Available to Suppress Fire (minutes)
Fig. 2-2: Fire suppression model.
i
.)
G
' ~ ~ ~
O
~
I 3-1 3.1 Seismic The objectives of this section are to provice: (1) a brief' description of the methodology and assumptions adopted in the LGS-SARA (1) report in the quantification of seismic accident sequences, (ii) BNL review ccmments of particular critical areas of tne LG!-SARA accument. Results based on SNL mocifications are also presented wnenever simplified estimations can ce made to illustrate the effects of tne modifications.
Inis section is civiced into
- two parts. Section 3.1.1 addresses tnose plant frontline systems which are identified in tne LGS-SARA report and the method of quantification by wnicn system unav.ailabilities, including the seismic contributions, are evaluated.
Section 3.1.2 summarizes tne seismic event tree approacn and tne seismic accident sequence analysis.
3.1.1 Plant Frontline Systems Tnis section is comprised of two subsections. Subsection 3.1.1.1 pre-sents an overview of the LGS-SARA ap' roac5 in modeling frontline systems.
It p
also summarizes tne assumptions made pertaining to systems and components of the systems in the evaluation of the seismic contribution to tne system un-availability. Subsection 3.1.1.2 provides the BNL revisions to the frontline system mocels and the results thereof. A discussion of the assumptions and tne LGS-SARA approacn to system fault trees are also included.
3.1.1.1 Overview Of the SARA Approach in Frontline System Modeling The system analysis part of the LGS-SARA effort is based extensively on the structure and contents of the LGS-PRA(2). This includes using the LGS-PRA frontline system fault trees in tne description of the randem failure of tne various systems.
In addition, these fault trees also provide the basis for the development of the seismic related failures. Finally, the components that appear in the LGS-PRA system fault trees constitute, in part, the group of components for which fragility evaluations were conducted.
t
1, a
3-2 LGS-SARA purported to nave examined fragility of two groups of cca-ponents: tnose tnat are contained in the LGS-PRA system fault trees and tnose wnich are identified to nave tne potential of significantly influencing tne
-likelihood of' core damage from seismic events, suc's as the reactor vessel and other related structure. A detailed discussion of iomponent fragility is presented in Chapter 2.1.
These comconents are th a ran<ed according to tne acceleration capacity of eacn item; those with a median ground acceleration capacity of greater than 1.56 g were not considered since they.are deemed to have a far nigher ground acceleration capacity than tnose predicted for the reactor site. Based on this criterion, a final list of 17 components are selected to be used in tne LGS-SARA evaluation, Taole 3.1.1.
Eacn seismic frontline system fault tree developed in tne LGS-SARA an-alysis is made up of two parts: the first part, wnicn leads to the failure of tne system, consists of tne random independent failures evaluated in the LGS-PRA; the second part includes all the pertinent seismic-related failures as determined using a specified criterion. This criterion for inclusion as a seismic-related failure requires that the component appears in Table 3.1.1.
The random independent failures for each system, as calculated in tne LGS-PRA, are trec*.ed as a basic event in the seismic system fault tree. For botn the HPCI and tne RCIC system, failure of the condensate storage tank (CST) neces-sitates the transfer of the water source from the CST to the suppression pool and is included in the fault trees. A total of eight seismic fault trees were developed for the LGS-SARA study and they include the following: nign pressure coolant injection (HPCI), reactor core isolation cooling (RCIC), low pressure coolant injection (LPCI), low pressure core spray (LPCS), residual heat re-moval (RHR), standby liquid control (SLC), automatic dapressurization system (ADS) and emergency power. An example of HPCI seismic system fault tree is given in Figure 3.1.1.
3.1.1.2 BNL Revisior, and Review of Frontline System Fault Trees Fault Tree Aporoach Tne inclusion of random independent failures into tne seismic fault trees represents a more realistic approacn than those that focus solely on seismic-related failure events.
In some circumstances, tnese random independent
)
O
- 6 O
O 3-3 failures wnen coupled with a substantial reduction in tne operatior's ability to follow procedures due to high stress conditions resulting from an earthquake may contribute significantly to core damage.
BNL reviewed the LGS-SARA modularized system f ault trees developed for the seismic analysis. These fault trees were prepared based on tne list of 17 components identified to be more susceptiole to seismic event related failure.
Oh p. 3-1 of tne LGS-SARA, it is stated tnat tne internal system fault trees provide, in part, tne list of comp 0nents for wnicn fragility functions were ceveloped and that additional items were included when tney were deemed to nave tne potential for significantly influencing the likelinood of core camage from seismic events. BNL agrees that consideration of only tnose components tnat are identified in tne internal event system fault trees does not ensure tnat all important seismic sensitive components have been included. Since in the construction of the internal event system fault trees, depending on tne levels of details at which the trees are developed, approximations may have been made to reduce tne canplexity of the trees. For instance, in modeling the faults of an injection train, the piping faults could have been excluded
'in the fault tree. Consequently, wnen it is used in the seismic assessment, dependence on piping failure would not have been properly evaluated.
It is not clear from tne report that a systematic search was conducted and wnat criteria were used to select,those components for fragility evaluation to ensure tnat all ccaponents sensitve to seismic events are included in tne analysis.
'In the modularized system fault tree approach, inter-system and support system dependences are not explicitly modeled. LGS-SARA did include the com-mon mode failure of the diesel generators as a means of failing the systems.
Preliminary review of the fault trees appears to indicate that common mode diesel failure is one of the dominant scenarios to core damage. However, by not explicitly modeling the dependences, other contributions to core damage may be lost in this approximation process. The BNL review of Limerick internal event report (7) assessed that contributions from incluaing the support system dependence constitutes a 60% increase.
It is judged in the context of a seismic event that these dependence contributions will be quite insignificant.
%-tm.-
,,3--
~--w p
,,e-
,,,-y.
---,wr, 7-:--
,,,---my
,----.,,-,,g
=,,- -
i 3-4 In addition to those depencences discussed earlier, tnere is one de-pendence wnich involves failure to transfer water source-fecm CST to sup-pression pool. This operation is required for both the HPCI and the RCIC sys-tem wnenever there is a low CST level. An operator failure to transfer given that a low CST level is likely to affect both nign pressure systems. Inis ce-pencence snould be incluced to properly reflect its impact on the final re-suits.
Electric Power Ine failure of the electric power system'is modeled with tne failures of seven comoonents; namely, two faults leading to tne loss of tne 440V power supply, tnree faults resulting in the loss of the diesel generators, one lea-ding to losing the 4KV bus, and one in the loss of DC power.
For both the HPCI and RCIC systems, loss of control power due to failure of tne DC bus, is assumed to disable the systems.
In principle, it is pos-sible to operate the two high pressure systems in a total blackout condition.
for an extended period of time with the operator manually providing the con-trols necessary. Nevertheless, in the event of an earthquake, BNL concurs tnat the LGA-SARA assumption may be more realistic.
In tna LGS-SARA Appendix B, it is estimated that at 1.09, no significant damage to the diesel generator fuel oil tanks is expecsted and that at ac-celerations somewnat in excess of 1.0, failure of attachments would be 9
likely.
It was identified in the Indian Point external event PRA review (3) (p, 2.7.1-15) that t?.a diesel generator fuel oil tanks are major contributors to core damage frequency. The data reported in the Indian Point Safety Study 4 for the diesel generator oil tank is of a generic nature and the median ground acceleration capacity is estimated to 1.15 g.
In lignt of this information, it is pertinent that the LGS-SARA report includes a more detailed analysis on the diesel generator fuel oil tank to show that they have the capacity much greater than 1.0g to justify their exclusion from the system fault trees.
=
O O
s l
3-5 Human Error
~
In the seismic part of LGS-SARA, it was reported that in estimating tne error rates for operator actions required during seismic accident sequences, the probability of failure witnin a given time scale was increased by a factor of 10 suojected to a maximum probability of 1.0.
Tne ba' sis to this selection of a factor of 10 comes from the fact enat an earthquake of sufficient intens-ity to damage reactor systems will initially disturb the performance of tne ope'rators and raise doubts in their minds about the performance of instru-mentation and controls.
The eartnquake may also lead to component failures that are not normally encountered in plant operations and therefore, may require innovative actions on the part of the operators.
It is BNL's judgment tnat during and suosequent to an eartnquake, tne operators ability to fol. low procedures, to diagnose proolems or to taxe cor-rective actions depends on tne intensity of the earthquaxe. Given tne limited information that is available in this area, it is often difficult to quantify tne likelinood of fail re under these unusual circumstances. However, one would expect tnat an increase in the numan failure probability is warranted.
Moreover, tnere are three factors wnich also play a significant part in determining the numan failure probability. One of them is tne avail'ablity of reliable instrumentation.
Subsequent to an earthquake, with alarms and an-nunciators sounding, it may be difficult for an operator to adequately assess tne true plant condition given tnat some instrumentation might give erroneous information. This is by far a more challenging situation and it significantly increases the complexity of the situation tnat confronts the operator. This type of scenario may cause two types of human failure in addition to those normally considered in the LGS-PRA; they are namely: 1) as a result of false
~
instrument readings, tne operator is misled and follows the wrong procedure in securing tne plant, 2) as a result of wrong and confusing information, tne operator may be misled to err in an error of commission.
i 2
m,
., o 3-6 The second contributing element to the subject of human failure tnat was not addressed in LGS-SARA is tne impact of aftersnock upon tne performance of tne operator to discharge nis responsibility. Based on seismic data, the probability of occurrence of aftershock decreases following an exponential type of. pattern; in otner words, the aftersnock is most likely to occur rignt after tne first quake and that likelinood decreases as a function of time in an exponential type manner.
If an aftersnock occurs witnin the time frame wnen operator action is critical, it may furtner impede his ability to respond to the demands of the plant.
Tne tnird area entails tne subject of display instrumentation, wnich is intended to provide tne operator pertinent information to nelp nim to under-stand the status of tne plant.
Display instrumentation could be in tne form of lignts, enart recorder, annunciators, alarms, etc.
Intheevenh.ofan eartnquake or an aftersnock, the failure modes of tne display instrumentation coulo be: 1) display information which is inconsistent of other indicators, 2) loss of display function.
LGS-SARA should furnisn a discussion on this sub-
~
ject to ensure tnat failure of display instrunentation has been investigated and is deemed not to impact significantly on final results.
BNL concludes that the factor of 10 increase in-tne human failure probability in some instances may be reasonable while it may also prove to be conservative or non-conservative in otner instances depending on the situa-tion wnicn confronts the operator. An example of how the absence of readily available and reliable informa' tion can affect the operator's ability to pursue the proper course of actions is given for the CST.
The CST is calculated to nave a median ground acceleration capacity of 0.24 g wnich is comparatively low -in light of the other component values.
It constitutes one of the two water sources from wnich the HPCI and the RCIC take suction. Failure of the CST would necessitate a transfer of the suction from the CST to the sup-pression pool. As for the HPCI, this transfer process is automatic, i.e.,
given that tnere is a low CST tank level, an automatic switch over will be initiated; newever, for tne RCIC, this transfer is a manual operation.
The J
e
l 3 -7 failure mode of tne CST water level sensors given that tne CST is failed was not addressed in LGS-SARA. Nevertneless, one could postulate tne following:
- 1) tnat despite tne failure of the CST, whetner it be rupture or toppled over, tne level sensors give a low level reading; 2) that in the failure of tne CST, the level sensors are damaged and that erroneous or misleading information re-sults. Preclusion of one or the other would require a more detailea 1,nvestigation of the failure modes of botn tne CST and the level sensors.
i If scenario 2 occurs, it implies that tne information given to tne oper-ator is misleading, anc nence in tnat case, tne failure probability for tne operator to respond properly snould be close to unity and snculd not be based on an aroitrary rule of tnumo - a factor of 10.
It so nappens that wnen tnis factor of 10 is applied to tne HPCI and tn.e RCIC transfer from CST to sup-pression pool, the human failure probability is unity. But there are other numan operations within these trees as well as other system fault trees wnich snould be examined on a case by case basis to deter: pine tne respective numan failure probability, for instance, manual failure to re-start system, failure to transfer service water, etc. A detailed discussion of this impact upon system unavailabilities is deferred to the next section.
Finally, it is important to note that LGS-SARA did not convey to the re-viewers tnat the increase in human error was applied consistently to all the pertinent basic numan events. BNL reviewed the LGS-pRA system fault trees and identified a number of manual operations which are omitted in tne seismic sys-l tem fault tree consideration, for instance, the manual failure to initiate HPCI, failure to manually initiate the LPCS and others. A more detailed investigation of the system fault trees should be conducted and pertinent j
findings on manual errors should be included in the modularized system fault trees.
Relay Chatter It is reported in LGS-SARA that low accelerations cause a momentary inter-ruption of control circuits and. power supplies (typically from relay-contact.
=
{
chatter); however, relay-cnatter is dismissed as a means of leading to system failure based on the fact that the operator can intervene and reset tne cir-cuit, hence restoring the system to its initial state.
[
l
t 8
O 3-8 It appears that tne question at issue here is not wnetner tne relays will chatter or at wnat acceleration will they begin to chatter, but wnat credit snould be given to the opeator to reset them if relay cnatter occurs.
If in one part of LGS-SARA, it is maintained that human error, in the event of an eartnquake, should be modified by a factor of 10 to reflect the increase in stressful conditions, it would seem consistent that tnese numan resconses to reset relays be given ene similar treatment in assessing tneir failures. 5NL is of tne opinion tnat given there is relay cnatter, failure on the part of tne operator to reset would result in tne equivalent of a relay failure.
If one wants to quantify the impact of relay-cnatter upon the system failure, tnen one would have to ascertain relay fragility information for tne various kinds of relays.
In tne Indian Point study (4), it is stated tnat relay-cnatter occurs at 1.2 g and that it presents no major difficulty. Tne 3SMRP data (5) snow tnat cnatter occurs at as low as 0.75g (spectral accel-eration).
Moreover, for certain relay cnatter which results in a breaker trip, reset of the system may be readily possible at tne control room; nowever, there are those relay trips wnich may require resetting at local panels and this causes a substantial increase in the failure probability of numan to reset.
It is important that LGS-SARA provid.es additional analysis on the fragility of relay cnatter and its impact upon various systems. Failure of numan action required to reset relay, hence leading to relay failure, should also be considered.
Finally, there is the underlying question that, in view of the aifferent relay trips, the operator is presented with a scenario for which he has not been trained and for wnicn no procedure has been written, what is the probabil-ity tnat ne will perform adequately to reset tne relays. Attempts to answer this question should be furnisned in LGS-SARA to support the premise that the operator can indeed in a reasonable time reset the relays and restore the system.
An example to illustrate these points can be found in the SLC fault tree.
There are two relays per SLC pump, for example, K4A and KSA for train A, X48
~
and K5B for train B, etc.
If chatter causes these relays to terminate tne
- j
i
?
3-9 operation of the tnree SLC pumps, tnis will lead to a direct failure of tne SLC system. Furtnermore, in tne redundant reactivity control system, relay chatter may cause the failure of all APRM channels wnien in turn will result in failure to initiate the SLC explosive valves and the SLC pumps.
In an ATWS accident event, the time available to an operator to respond to tnese cnal-lenges is also significantly reduced to tne order of minutes.
In lignt of t,nis information, tne impact of relay chatter upon :ne SLC system shoul: De evaluated in more detail.
Transients A list of tne LGS-SARA mean random fall'ure values and tne nomenclature is given in Table 3.1.2; the first column of values are tnose given in tne LGS-SARA report. Tne second column tabulates the values used in tne internal ev-ent risk assessment study, LGS-PRA. The third column denoted by NUREG/CR-3028 enumerates tnose values generated by BNL in tne review of the LGS-PRA. The last column represents values that BNL believes snould be used in the LGS-SARA study.
It is quite obvious tnat between tne first and second column, differ-ences in values can ce noted. Despite the fact tnat little explanatiens are furnisned in LGS-SARA to address the differences for botn nigh pressure sys-tees, these differences are miniscule. But for the low pressure system (V) and tne manual depressurization (X) function a more detailed discussion is warrantad.
As stated repeatedly in LGS-SARA, the seismic evaluation was done based extensively on tne LGS-PRA; therefore, it is reasonable to assume unless noted otnerwise that the nomenclature used would also correspond to that of LGS-PRA.
The manual depressurization function, X, denotes the failure on the part of the operator to depressurize the reactor in a timely manner using the Auto-matic Depressurization System (ADS). The low pressure injection function (V) represents eitner the failure of the ADS nardware or a simultaneous failure of tne'LPCI system and the LPCS system.
In the LGS-PRA, the X function unevail-ability is calculated to be 2 x 10-3; the V function value is estimated by BNL based on the LGS-PRA unavailability of the LPCI and LPCS systems given tnat tnere is a loss of offsite power and a failure to recover offsite power to be 2.65 x 10 4 According to the information providea on p. C-15 of
i 3-10 Table C-6 of the LGS-SARA, it appears enat tne V function defined in tne report only consists of the LPCI and the LPCS system; this notion is furtner confirmed in the Boolean expression of X
= X +a shown on p. C-14 of Tacle R
C-5.
Xg is defined in the report as the random failure of X and A, as the loss of electric control and motive power. since the manual action to cecres-surize the re. actor does not require electric control or motive cower, nence, it is possiole to argue tnat the nardware failure of :ne ADS is lum;ec witn tile X function witnout much impact on the function unavailacility. However, tnis is not consistent witn what nas been presented in the LGS-PRA, and may resuit in inisleading conclusions of dominant sequences' The imcact of properly including the ADS hardware failure witnin the 7 function for various accident sequences will be addressed in Section 3.1.2.
Quantification of the RHR system with the loss of offsite power and no recovery was not performed by BNL nor by PeCo and hence no value is reported in LGS-PRA and NUREG/CR-3028. The most substantial increase between the LGS-SARf and NUREG/CR-3028 internal event values occurs witn the V function -
a factor of 3.7 followed by a factor of 3.0 increase for tne X function.
The common made diesel generator failure probability was reported in tne earlier revisions of the LGS-PRA as 1.88 x 10-3 and it is for this reason that tnis value was used in the NUREG/CR-3028. Revisions received subsequent to tne report at NUREG/CR-3028 modified the unavailability to 1.08 x 10-3 claiming that it was a typographical error in the earlier versions.
In LGS-SARA, a diesel generator common mode failure mean value of 1.25 x 10-3 was reported. BNL agreed that the 1.88 x 10-3 value is overly conservative.
Recsently, there have been studies (6) conducted to attempt to better evalu-ate tne diesel common made unavailability, and values lower tnan 1.0 x 10-3 nave been suggested. As for this review, BNL will use the 1.25 x 10-3 for comparison purposes.
The increase in numerical values for the HINIA and RIN3 is due to the followi ng: HINIA and RIN3 represent failure to provide flow from suppression pool given that tne CST water is unavailable.
The major difference between tne two events lies in the manual action required to perform tne operation for I
s e
s s
3-11 the RCIC system, FSAR p. 7.48.
Consequently, if a factor of 10 increase is assumed, tne manual error for failure to transfer becomes unity and dominates the failure of the RCIC system. - Because of tne automatic transfer function in the HPCI, a similar increase in the manual error results only in minimal in-crease in the system unavailability.
If, instead, a numan factor of 7.5 is used, tne RIN3 will be 0.75, wnereas, HINIA would remain uncnangec.
Anotner major cnange tnat is evicent based on :ne factor of ten increase is in the manual dep.'essurization function. This increase results merely from applying the numan error factor of 10 to the NUREG/CR-3028 value of 6x10-3 It appears tnat for HINIA, RIN3 and X, the increase cue to tne numan error factor was not incluced in tneir values as it snoula be.
Anticioated Transients Witnout Scram I
In the event tnat an earthquake occurs which results in an ATWS, LGS-SARA analysed tne sequence using a loss of offsite power ATWS event tree. A set of mean failure values that was used in the LGS-SARA analysis is shown on Taele 3.1.3.
There are four columa.s of values in tne Table, the first three of them present values used in the LGS-SARA study, the LGS-PRA and the BNL internal event review, NUREG/CR-3028, respectively. The last column represents values wnich BNL believes should be used in the LGS-SARA analysis.
It snould be pointea out tnat the values presented in Table 3.1.3 are representative numoers; one snould refer to tnese reports for more detail information.
LGS-SARA values for both the HPCI and RCIC failure values are in general lower tnan those of t'ne LGS-PRA and NUREG/CR-3028. The increase for RCIC, RR is about a factor of 6.6 times. As for the ADS inhibit function, the LGS-SARA value is 8.0 x 10-3 versus 2.0 x 10-2 from NUREG/CR-3028; another factor of 10 increase due to the intense st'ress level for the operator brings f
tne final value (last column) to 2.0 x 10-1 There is no disagreement on I
the values of Up as assessed by BNL and LGS-SARA. Little increase is noted between the LGS-5 RA and BNL values for the SLC system, nowever, LGS-SARA value is about a factor of 10 larger than the LGS-PRA value. W2 was re-ported in LGS-SARA to be 0.1 rather than the 0.14 used in tne LGS-PRA. The
r 3-12 ciesel generator common mode failure, HINIA and RIN3 failure values are 3
described in the previous paragraphs.
The va.ae selectea for tne mecnanical failure of the scram system increased to 1.5 x 10-5 Tne variable PCR is defined in tne text of LGS-SARA'to nave'a value of 0.2.
No description is providea as to wnat PCg is.
Failure to scram is cefined in LGS-SARA as:
CM = (1-PC ) CR + PCg (S3+SS+S)
R 7
A telepnene conversation with PeCo revealed tnat tne PCg is a jucgment factor applied to tne seismic f ailure of tne reactor interna:s anc CRD guice tuces, see Figure 3.1.2.
PeCo stated that in the event of an earthquake, only 20". of the time would tne failure of the CRD guide tubes or tne reactor inter-nals cause a failure to scram. Since there is insufficient information on how tnis PCg value is obtained, it is difficult for BNL to render some sort of judgment on its valtaity.
Another area of concern is in the treatment of random failure to scram; BNL believes that, given that there is a challenge to the scram system, the
. failure to scram probability snould no't be weignted by a factor of (1-PCg)=
0.8.
Also in the telephone conversation, PeCo explained that they attempted to preserve tne scram failure probability from the 0.8 reduction by increasing tne scram failure probability from 1.0 x 10-5 to 1.5 x 10-5 It is sug-l gested tnat a more detailed cocumentation of these points by PeCo be provided l
in LGS-SARA l
For the purposes of sequence quantification to be presented in tha next section, failure to scram is cefined by BNL as follows:
Cg = CR*S3*SS+5-7 Finally, it is suggested tnat a detailed discussion be provided in LGS-SARA to identify and reconcile differences in the randem failure values usea in LGS-SARA ar.c LGS-PRA.
l
--e e--,
--,.--.n.--n
i s
3-13 3.1.2 Accident Sequence Analysis Tnis section addresses the deffnition of accident sequences and the quantification of core damage probability given tnat there is an earthquake.
4
.Section 3.1.2.1 briefly d' scribes the approach and methodolocy used in e
^
LGS-SARA for accident sequence definition and core damage quantification.
Section 3.1.2.2 contains results of the BNL review.
3:1.2.1 Overview of LGS-SARA Accident Sequence Analysts LGS-SARA examined various fragility estimates (provided in Appendix B) and concluded tnat tne offsite power system was most sus _ceptible to an eartnquake wnicn, wnen failed, would result in an initiating event. Failure of pipes and vai/es causing an initicting event is dismissed as highly improo-able in light of the significantly greater capacities of tnese components.
It is for tais reason that LGS-SARA maintains that the frequency of a seismically induced LOCA (be it large, medium, or small) is quite insignificant. The simultaneous occurence of an earthquake and a random LOCA event is also estimated to De a few orders of magnitude smaller than tne loss of offsite power event. Thereforg, only the seismic-induced loss of offsite power was investigated as a credible initiating event.
~
Tne event tree method was used to define the accident sequences. A total of tnree event trees were developed: tne first event tree depicts tne suc-cess or failure of a number of critical functions wnose operation or inopera-tion greatly affects the analysis to be followed, see Figure 3.1.3.
This tree is made up of five functions, namely, the seismic event initiating frequency, reactor pressure vessel, reactor and control ouilding, and reactor scram.
Failure of the reactor pressure vessel given that there is an earthquake leads directly to core damage. The nature of the failure was identified to ce initially the failure of the vessel supports which, in turn, results in severing of all four steam pipes. To mitigate such a breach of the reactor coolant boundary is far beyond the capability of the ECCS.
0
~ - -. -,.. - -
e 3-14 Given that the reactor pressure vessel stays intact, failure of the re-actor and control building will result in core damage regardless of wnether tnere is a s,uccessful reactor scram or not, Sequences 4 and 5.
If, however,
- the reactor building aces not fail, tnen failure of offsite power coupled witn eitner successful or unsuccessful scram would lead to transfers to Figure 3.1.4 and Figure 3.1.5 respectively.
The event tree presented in Figure 3.1.4 is icentical in structure to tnat of the internal loss of offsite power event. Systems wnicn are re-quired to nitigate the event are assessed and accident secuences are cefinec.
In Figure 3.1.5, tn' mitigation of an ATWS event is preeentec.
It is again identical in structure to tne one given in tne LGS-PRA for loss of off-site power.
Inputs to these event trees for individual systems are based upon tne modularized system fault trees and a discussion of tnese trees is provided in Section 3.1.1.
The quantification of these event trees were performed using.
the computer code SEISMIC. The Monte Carlo metnod is used in tne code to simulate the failure probability of seismic and random failure of components and accident sequence frequency is then calculated based on the Boolean expression inputed for that particular sequence. Median and mean values, and conficence' levels of the sequences are also evaluated and tnose for the dominant sequences are reported.
3.1.2.2 BNL Review of Accident Secuence Quantification SNL reviewed the event trees and assumptions which enter into the de-velopment of these trees. Review comments are presented in this section. A number of areas were identified which warrant further discussions ind.tney are also presented in tnis section. As a result of tne revisions made to tne modularized system fault trees, estimates of their impacts on respective accident sequence core damage frequencies are described.
f b
i
.J
e 3-15 Metnodology The event tree-fault tree methodology employed in tne LGS-SARA represents a widely practiced approacn used winnin the nuclear industry today to assess accident sequences and core damage frequencies. BNL agrees that it is adequate in evaluating risk indices witnin the co7 text and recuirements of today's risk assessment studies.
Ine LGS-SARA analysis based extensively on the approacn and results of tne LGS-PRA. Two event trees from the LGS-PRA were adopted to analyse the seismic initiating event. They are tne transient and ATWS loss of offsite power trees. While BNL agrees that these trees will model tne loss of offsite power event adequately if caution is exercised in addressing the dependent failure of components due to an eartnquake, additional information snould be included in LGS-SARA to estaolisn the basis wny the seismic event evaluation can be based exter.sively on the internal event analysis.
In other words, Justification snould be presented to snow tnat external event accidents do not warrant separate event trees to model the diffe. rent scenarios. Rationale on wny tne LGS PRA event trees were used should reflect these concerns.
Initiating Events As described in Section 3.1.2.1 of this review and in Chapter 3 of LGS-SARA, tne loss of offsite power due to failure of the switchyard ceramic insulators (median ground acceleration capacity of.20 g) was identified to be the major initiating event contributor. Failure of the reactor and control building and of tne reactor pressure vessel nas also been included in the consideration of initiating an accident event. BNL agrees that these are important initiating scenarios that should be investigated.
Nonetheless, it is not clear from what is reported in LGS-SARA that the searcn for initiating events went beyond those components and some structural members.
In particular, it is not obvious that effort was devoted to ex-amining the nnn-safety related equipment or equipment whicn is not important for a safe snutdown of the plant to determine if they could become initiating
=
event contributors given that there was an eartnquake. These two types of equipment are not subjected to the same rigorous seismic qualification
=
3-16 standarcs as other seismically qualified components. Cepending on the capacities of tnese non-safety components, an earthquaxe witn low ground accelerations may cause a reactor trip without failing the switchyard ceramic insulators. Such an event will initiate a transient which ;hould be evaluatec by event trees similar to tnose presented in Figures 3.1.4 and 3.1.5.
Tne difference between the event trees is tnat there is offsite power in tnis In the event that a transient coes not occur given an sartnqua<a, tnen case.
the sequence is a success event.
An example to illustrate these points is the feedwater system.
It is not a system tnat is required for a safe snutdown of the plant nor is it a safety-related equipment, nowever, if an eartnquake occurs control, relays anc otner components of the feedwater system may generate a trip of the system resulting in an a reactor transient.
In Figure 3.1.3, the event T, sequence number 1, was treated as an OK 3
sequence. A note at the bottom of the figure states that a seismic event that does not lead to the loss of offsite power is considered to be benign and is acequately accunted for in the turbine trip initiating event.
Given tnat there is an earthquake, if offsite power is still available, the event tree presented in Figure 3.1.3 does not model tne plant response beyond that point.
In principle according to the event tree, the reactor is not even scrammed and therefore, there is no need for it to be transferred to the turbine trip event tree. However, if there is failure of non-safety equipment or tripping of the equipment offline which results in a plant transient, such as the loss of feedwater due to seismic, then the event tree snould be further developed to define the accident sequences. The internal event turbine-trip event tree is not appropriate since the mitigation system considered will not include the necessary seismic failures. The new event tree will be similar to the one wnich is in Figure 3.1.4 with certain random failure values modified to reflect the availability of offsite power. BNL estimated that by transferring T3 to this new event tree, the only sequence wnicn may contribute to the overall core damage would be T UX.
The core 3
damage probability is estimated to be in the order of 10-7 to 10-8
.~
3-17 In the event that given the reactor transient, there is a failure to scram, an event tree similar to Figure 3.1.5 should be developed.
It is conceivable tnat tne contribution to risk due to class V sequences may not be negligible. It is reccmmended that these consicerations of additional acci-dent sequences sno'uld be accressed in tne LGS-SARA.
Not Event Guantification LuS-SARA statea that non-failure states are includea in tne Boolean expression of the accident sequences and therefore in the quantification process.
BNL performed some preliminary estimates of tne core damage procacility for tne six dcminant sequences as identified in Taole 3.1.4 and tne results are also prdvided in the table. The values under tne LGS-SARA column come directly from Table 3.2 of the LGS-SARA report, wnereas tne SNL estimates reflect two assumptions made to assess the core damage procacility of the respective sequences based on the Boolean expression given on Table C-3 to C-5 in tne LGS-SARA. One of tnese estimates, (tne secono column), did not incluce tne NOT events that appeared ir. the Figure 3.1.3.
They are namely tne NOT of R, Cg and RPV. Good agreement was obtained between the LGS-SARA S
results and the second column BNL preliminary estimate values. The final column delineates BNL results wnen the NOT e' ents were included. These v
estimates show a major reduction for two sequences: an orcer of magnituce change for the sequence T E UX and a factor of 2 decrease for sequence S3 T
R.
As far as the remaining sequences are concerned, little impact is 3
g observed.
Attempts were made to identify the cause of this apparent difference in results. A number of possibilities have been identified; 1) the preliminary estimates performed by BNL was not accurate; the agreement between LGS-SARA values and the BNL estimates witnout NOT events (second column on Table 3.1.4) is fortuitous, 2) the NOT events were not properly included in the PeCo quan-tification, 3) the computer code SEISMIC did not perform these quantifications properly.
If indeed the NOT events were not included in the PeCo quantifica-tion, tnen BNL results (last column on Table 3.1.3) indicates a substantial impact reduction for the T E UX and T RS g sequences. Furtnermore, tne 33 i
~-
w
--.n..
m-
9 3-18 core damage frequency would be more evenly distributed over tne six sequences as indicated in the last column.
ADS Seismic Failure A discussion of how LGS-SARA modeled the ADS in the seismic system faul:
tree is given in Section 3.1.1.
It is inferred tnat the failure of tne ADS nardware is incluced in tne cefinition of X whicn is the manual cepressuriza-tion function, X=A+X-R XR represents the ranccm failure of the manual depressurization function; A comprises seven different types of electric failures. Tney incluce tne loss of tne 440 V power supply, the 4KV supply, the diesel generators and tne DC power. LGS-SARA conservatively assumed that the failure of all these events would lead to a failure of tne ADS hardware.
In essence, only tne failure of tne DC power supply would lead directly to an ADS failure.
It is, of course, obvious that the availability of AC power provides added assurance of the reliability of tne DC power supply; however, failure of tne 440V bus does not result in failure of the ADS.
It is for this reason that LGS-SARA is conser-vative wnen it assumed that X = A + Xg.
Since NOT events are considered to be important in sequence quantifica-tion, they snould be included in the sequence evaluation. However, as a re-l suit of a conservative definition of X, tnis may lead to non-conservatism in j
other sequences. This effect may not necessarily manifest itself in the i
change of the core damage frequency, but it may well have substantial impacts l
on tne risk evaluation.
For instance, if accident sequence T E UX, sequence No. 6 in Figure 33 3.1.4, is calculated by assuming either the 440V, the 4KV, the diesels or the DC power will fail tne function, then a NOT-X event will imply that these various types of power supplies are available. This represents a non-con-1 servative departure from the system modeling, since tne operation of ADS can only imply that DC power is available. This will tend to underestimate sequences T E 0, T E UW and T E UV.
The impact may reside in 33 3S 33 underestimating the contribution to accident class IS, whereas the cnange in
'T l
)
l l
l
3-19 core damage probability may be inconsequential. Other risk indices, sucn as, latent and acute fatalities may be affected differently.
One of the approaches to address this concern is to integrate the ADS nardware with the low pressure injection function, V consistent witn tne LGS-PRA definitions.
Secuence Quantif4 cation Ine focus of tnis discus'sion will be primarily on tne six dominant sequences identified by LGS-SARA and on other sequences which BNL believes will reflect scme impact on the risk indices.
(I) Ocminant Sequences if tne modifications in Section 3.1.1.2 and this section are included in the sequence quantification, only three of the six dominant sequences are significantly affected. Table 3.1.5 enumerates tne changes in core damage frequency given a modification in system unavailability for eacn of the dominant accident sequences. The core damage frequencies tabulated on Table 3.1.5 are preliminary estimates only. The first column identifies the six ccminant accident sequences. Two of them are ATWS events: TECC 3S32 and T R C. The value in parenthesis following each sequence name is 3B3 the core damage frequency as calculated in LGS-SARA. The second column depicts the system which is modified when the sequence is re-quantified. The l
value in parantnesis denotes the revised system unavailability. The last column is the core damage frequency as a result of the requantification. The sequence T E UX is calculated in LGS-SARA to have a core damage frequency 33 of 3.1 x 10-6 and if the manual depressurization function random failure is modified to the new BNL value of 6.0 x 10-2, BNL estimated that tne core damage will increase to about 4.0 x 10-6 It is assumed in the calculation tnat beside X, all other components retain their values as suggested in LGS-SARA. Similarly, if only the U function is modified, an increase frcm 3.1 l
to 3.8 x.10-6 is observed.
Increases in the failure to transfer from the CST to the suppression pool produce similar results - 3.8'x 10-6 If all of e
these modifications are integrated into tne accident sequence T E UX, the 33
~
3-20 total core damage frequency is about 5.2 x 10-6; approximately a factor of 1.7 increase.
If one assumes that both the HINIA and tne RIN3 become unity and tne otner system values are tnose of the LGS-SARA, tnen the core damage frequency for tne accident sequence T E UX becomes 4.0 x 10-6 In other words, in 33 tne event tnat tnere is a total fail'ure to transfer from tne CST to tne suppression pool, for causes wnich may be numan dependence failure or failure.
of all CST level sensors, the core damage frequency increases by a factor of about 1.3.
The other two sequences that are affected are the ATWS sequences. Stil revised the definition of C3 to reflect a more prudent approacn in view of the lack of information in LGS-SARA on the definition of mechanical failure to scram. Tne BNL definition'is given as follows:
Cg=CR+S3+SS+S-7 This definition leads to an increase of about a factor of 5 for botn tne TECC3 3 3 2 and tne T R C3 g g accident sequences.
There is no impact for tne remaining three dominant sequences as a re-sult of the modifications in Table 3.1.2.
The total core damage frequency is increased by slightly less than a factor of 2.
This increase does not include' the contribution from considering tne NOT events.
T E UV Accident Sequence
~
(II) 33 The core damage frequency of the accident sequence T E 0V is 33 calculated in LGS-SARA to be 5.9 x 10-9
.The Boolean expression of tnis se-quence can be written as follows:
T E 0V = T N E GUTV 33 3
g3
= T T I S C IUV-364IM If one uses the definitions of V and X provided in LGS-SARA, the following exp.ession will result:
T E 0V = T T I 5 IIS S 3i 2 17 + I IS V H R1RRR 33 364M g R
(3.1)
+ T I5 5 1 17 R g + T IS 3 V G HR R
R 12R O
1 3-21 where Sj, i = 1,2,.. 17 are tne seismic-induced component failures; a ce-tailed listing is given in Table 3.1.1.
The bar above eacn variable denotes a NOT event. C is the mechanical failure to scram; A is seismic failure of 3
tne electric power system; the subscript R denotes random failures. H and R represents tne HPCI and tne RCIC system respectively. G cenotes tne coccina-tion of transfer and high pressure system failures, and is definea as follcws:
G = HINIA
- RIN3 + HINIA
- Rg + RIN3
- Hg However, if one uses the BNL definitions of X and V, namely X = XR ind V =-
- LPCS + AUS, where LPCS and LPCI are tne same as those defined in LGS-SARA, anc wnere the accea term ADS is the sum of tne ADS narcware randem failure Ag and the electric power A, the following Boolean expression is ob-tained:
T E 0V = T T I I IgS A + I S 3 SR 1 2 17 33 364M t
+ISRHVRIgRR+ISRHSR 1 R R 17 (3.2)
+TSRHAR1ggR+IS3R12 SV R
+IS3 R12 GAR Vg represents tne random failure of the LPCS and LPCI systems. Comparison of the two expressions in Equations 3.1 and 3.2 indicates that except for NOT-A, Equation 3.2 contains all tne terms of Equation 3.1 and, in aadition, there are tnree more terms wnich are not in Equation 3.1.
These terms contain a failure of tne electric power system and failure of the ADS narcware given the loss of hign pressure injection. BNL did not estimate the contribution of this sequence as a result of the modifications made. It is suggested a more detailed analysis be provided in LGS-SARA to better identify the contribution of T E UV to core damage and to tne final risk.
3S (III) Otner ATWS Sequences BNL reviewed the LGS-SARA ATWS event tree and found that, beside those two dominant ATWS sequences, T E C C3 3 g 2 and T R C, the contribu-3gg tion to core damage from otner ATWS sequences defined in Figure 3.1.5 is relatively small. However, with the BNL definition of C, there will be g
about a factor of 5 increase for all the ATWS sequences in Figure 3.1.5.
The
,__..r.
=
3-22 total ATWS core-damage frequency reported in LGS-SARA is B.1 x 10-7; by eliminating tne PCR and using the BNL C3 definition, the total ATWS core damage becomes approximately 4.0 x 10-6 This does indicate that the ATWS results are quite sensitive to tne parameters used to define the failure of scram. PECo believed that, based on the failure modes defined for tne reactor internals and the CR0 guide tubes, it will Oe conservative if one assumes that fa,ilure of tnese components would cause directly a failure to scram. SNL tends to agree that' there may be conservatism innerent in the definition of failure modes of these components and would encourage additio.1al analysis be provided to support the LGS-SARA assumptions. A refined analysis in this area is needed since it will nave significant impact on tne acute and latent fatalities.
Examination of ATWS function unavailabilities provided in Table 3.1.3, reveals a number of major increases in the random failure probabilities: a factor of about 1.6 for the HPCI; a factor of approximately 6.6 for the RCIC; a factor of 2$ for the ADS inhibit function; and a factor of 1.4 for the W2 functions.
BNL did not perform any re-assessment of those accident sequences wnicn are affected by these modifications, however, due to the change in magnitudes of some of these functions, and the fact that significant contri-bution to risks comes from the Class IV events, it will be prudent to evaluate the effects of these cnanges upon the results on core damage as well as the final risks. Sensitivity analysis would also provide nelpful insight in the evaluation of these accident scenarios.
(IV) Summary SNL did not reassess the final core damage frequency as a result of all the proposed changes. A few of the areas identified requires more detailed analysis whereas others need additional information to substantiate tne as-sumptions. Re-quantification of some changes were made wnerever it was possible and results are discussed earlier in this section.
It appears that for these modifications investigated, at most a factor of two cnanges to the core damage frequency is observed.
In view of tne large uncertainty associ-ated with tne seismic accident sequences, tnese changes in magnitude do not constitute any significant impact on tne core damage frequency, but their
.j effects on the acute and latent fatalities may be significant.
1
3-23 Table 3.1.1 Significant Earthquake-induced Failures Mecian ground Failure cause acceleration tio..
Cccponent or mode capacity S S
g g
c S3 Cffsite power (500/230-kV Ceramic f asulator 0.20 0.20 0.25 switchyard) breakage S2 Condensate storage tank Tank-wall rupture 0.24 0.23 0.31 53 Reactor internals Loss of shroua support 0.67 0.28 0.32 54 Reactor enclosure and control structure Shear-wall collapse 1.05 0.31 0.25 SS CRD guide tube Excess bending 1.37 0.28 0.35 S6 Reactor pressure-vessel Loss of upper support 1.25 0.28 C. 22 bracket S7 Hydraulic control unit Loss of function 1.24 0.36 0.52 S8 SLC test tank Loss of support 0.71 0.27 0.37 Sg Nitrogen accumulator (SLC) Anchor-bolt shearing 0.80 0.27 0.20 S10 SLC tank Wall buckle 1.33 0.27 0.19 S11 440-V bus /SG breakers Power circuit 1.46 0.38 0.44 S12 440-V bus transformer Loss of function 1.49 0.36 0.43
. breaker f
S13 125/250-V dc bus Loss of function 1.49 0.36 0.43 S14 4-kV bus /SG Breaker trip 1,49 0.36 0.43 SIS Diesel-generator circuit Loss of function 1.56 0.32 0.41-l S16 Diesel-generator heat and Structural 1.55 0.28 0.43 i
vent Sy RHR heat exchangers Loss of lower support 1.09 0.32 0.34 l
(anchor bolts) l l
3-24 Table 3.1.2 Mean Values for Random System or Function Failures used in Transient Events LGS-SARA LGS-PRA NUREG/CR-3023 SNL f:his review)
HPCI 8.8X10-2 0.07 0.1157 0.1157 RCIC 7.6x10-2 0.07 0.07 0.07 y
- 1. 0x10-4 ***
2.7x10-4 3.7x10-4*
3.7x10-4" W
2.6x10-4 2.6x10-4**
HINIA
- 1. 0x10-2 1,0xio-2 1.0x10-2 1.0x10-2 RIN3 7.0x10-3 7.0x10-3 7.0x10-3 1,0 OG' 1.25x10-3 1.08x10-3 1.88x10-3 1.25x10-3 C
X 2.0x10-3 2.0x10-3 6.0x10-3 6.0x10-2 With ADS hardware and no offsite power.
- With no offsite power and only RHR.
HPCI - High Pressure Coolant Injection RCIC - Reactor Core Isolation Cooling V
- Low Pressure Injection Function W
- Containment Heat Removal Function HINIA - Failure to Transfer from CST to Suppression Pool in HPCI RIN3 - Failure to Transfer from CST to Suppression Pool in RCIC DGC
- Diesel Generator Common Mode Failure X
- Manual Depressurization.
,s
3-25 Table 3.1.3 ATWS Mean Random Failure Values SARA LGS-PRA NUREG/CR-3023 SNL (this review)
H,q 3.8x10-2 0.1 0.14 -
0.14 RR 7.6x10-2 0.5 0.5 0.5 0
'JH 2.0x10-3 20x10-4 2.0x10-4 2.0x10-3 C12 1.6x10-2 1.5x10-3 1.4x10-2 1,2x10-2 C
1.5x10-5 1.0x10-5 1,0xio-5 1.0x10-2 3
HINIA 1.0x10-2 1.0x10-2 1.Cx10-2 1.0x10-2 RIN3
_7.0x10-3 7.0x10-3 7,0xio-3 1.0 W2 0.1 0.14 0.14 0.14 PCg 0.2 1.0 1.25x10-3 1.08x10-3 1.88x10-3 1.25x10-3
'DGC HR hPCI random failure R R RCIC random failure 0
ADS inhibit failure l
UH Failure to control reactor vessel level 8 l
C12 Failure of two or three SLC pumps Cg Scram failure-mechanical W2 Failure of both RHR PC Fraction of events that lead to scram f ailure*
R
" Definition not given in LGS-SARA, inferred from modularized system fault tree.
I i
e l
3-26 Table 3.1.4 Ocminant Seismic Core Damage Sequences BNL Estimates Sequence-Class LGS-SARA W/o NOT Event W/NOT Event T E 0X I
3.1x10-6 4,0xio-6 4x10-7 33 TR33 IS 9.6x10-7 9.5x10-7 5x10-7
' T RPV S/III 8.0x10-7 4.4x10-7 4.4x10-7 S
7 E CgC2 III/IV 5.4x10-7 6.0x10-7 4.5x10-7 33 T R Cg IS 1.4x10-7 3.5x10-7 3.5x10-7 SB TEW II/IS 1.1x10-7 1.1x10-7 1.1x10-7 33 Total 5.7x10-6 6.5x10-6 2.3x10-6 Table 3.1.5 Dominant Seismic Sequences with BNL Changes Sequence (Core System Modified Core Damage
- Damace Probability)
(Unavail ability)
Frecuency T E 0X (3.1x10-0)
X (6x10-2) 4.0x10-6 33 U (8.1x10-3) 3.8x10-6 HINIA, RIN3 (1x10-2,1,0) 3,gxio-6 All combined 5.2x10-6 T E CgC2 (5.4x10-7)
CM 3.0x10-6 33 TRC3 B g'(1.4x10-7)
CM 1.8x10-6 T PRV (8.0x10-7) 8.0x10-7 3
T E W (1.1x10-7) 1.1x10-7 33 TRS g (9.6x10-7) 9.6x10-7 Total 5.7x10-6 1.2x10-5*,
- Based on LGS-SARA sequence values.'
~
- Sum total of T E 0X (combined) and the other 5 sequences.
33
)
J
~e
-m=
y rre m---
c-y.,-
.m
.r
e 8
,e 3-27 l
1 FAILURE TO P9CVDE ADEOUATE FLC*J FRCH HPC:
l l
HMI I
1 I
I i
..I:m2E Cr sci:MI:
w R wecM RANOOM AND DOEPDOW SCOMIO j
FAILURES IAILUEES FA m ts I
w2 ll i i i
n A
i I
FAILURE TO SCEMI:
i SUPPt.Y WATER g37 FRCM SUPPRESSION i
PCCL RUPTURC l
l l M
l C
0 I
I l
LOSS CF FAILURE TO SUPP9ESSICN PCCL DUE TO SCSMIC TRANSFER CN FAILURE WATER LC$$
CV irHR Htt i
II wrNIs l
l l
Figure 3.1.1 Re uced fault tree for the HPCI system.
d i
i o
__._y,,,,
a
,,_m
_..____-..+,,,c
___-..._m.
c.
3-28 FAILURE TO SCRAM i
i N
I i
SCRAM SISNAL
'dC SG A" ogggg3r egg gg S!ONat doESENT SC2An pgcy, SEFCRE SCRAN ran "EOt. FAIL RANCCM FAILURE SCSMIC CF SCRAM
, SYSTEM FAILURES I
l I
i I
l i
SCSMIC FAILURE SCSMIC SCSMIO CF REACTCR FAI:.URE CF FAILUKE CF j
DTERNALS CRD GUI:E MYORAULIO CSHRCUD SURPCRT)
TUSES CCNiaCL UNIT I
l l 1I i
Figure 3.1.2 Reduced fault tree for failure to scram.
=.
I
=
4 wI
-de s
d e l
a s s l
r ol I
S a
S gcc I
I/
I e
S D
y e c s
8 l
c n e
~
7 7
7 na ne f
a u r s K
3 e u on e n u q a
O 6
4 0
0 0
0 Mnqc a e i t
ig r
sf 9
1 F
t M
C's V,
'D '
'3 3
h 8g th i
C 13.
t 4" '
f.
1 C
83 83 3.
c>
1 1
1 1
1
-4e 3
2 4
5 6
' i, 1
i t
c C
s I'
.ee r
t t
n e
i v
i e
8 o
C i
c S
i' f
O E
i 0P' m
s i
n e
o s
c t
e h
- t T
- i-p -
r g
t d, g ss g
r o
i 3
c i
,d a, gis I
i 1
g s
l c
H g a o
u.
t
- a. 3
.w e i u.
o ru n
g owi o.F i
- a. in 8
e s
us t
s l
V ic is e
o c ut a s t
P s s e l
o I
o v
f n a..
l p
c.
....o a.
, n i.
i o,
op 1
y c a.
cc e
n v
ne r
uu S
n a.
cq T
i 3
Se 1
4 s
i
8 I
i 3-30 4Je m
45 3 5
W E
)* 7
=
Q
~
1BIII
~
=
- i v i
v w
a a
a w
w S. ' aI o
O
~
C C
2
=
a n
=
2 l
t w
I I. 2
=
M O
C' C'
2 i
3,
-3 a
s s
h 6
f
.a s
a a
a a
A A
A e
^
y' d'
a.
1 a
".A
-A A
.A A
y' A
p i
b S.e n
q dd n
e, e
a e
m e
g a5 k
6 l
V 1
3 s
?-
U
.S..*
toa
.=
a es-3 e
2e i i*
L
. 's C
w15 3O O.
O a
W Ea l
?4 1& T Q
=4j w
Q m
MC
- ul L
M3G1 3
o 5 4 4-0 0L
'M
.e C
GJ 8
6 Ng G
=
('
U E
W
.e=
h M
.>e zv 2
33 W
e 7
j Q
a7 b
3 n
m O
3.
=^
, x e
.?
Li-A*
i i
i r
I f
I
3 O
e e
s 3-31 9 3
$I DgDU l i l E 8 1 3 I I
I I
I O
s '1 j !.
e 1
gg3 3...
m.
8 *,2
{
a.... -
o-a o....
o o
.-,oo
- 2
~
a 2 *
, ( ( { 3, 3, A ( ( % ( ( '
=
=
a i E89i.22222 : JJJJJJJJJJJu
>}.
( <~2s' a o 1
2 -
2 2 2 2 2
- 2. 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3
% % % % % % % % % % % % % % % % % % % % u, % % *, % %, %, %
...-.. s: : : :
e: 2 2 4: : : : 4 4: 4A2 JJ a
il l
l l
I t
a !
l l
5 !. I '
rm
- j E
e e
o a
m um m um um me s
.i 2
=
4 Z
3
- s.
,* I i
I i
I g
l 4 1 w
A J
l 3
'd e
11 3 3' t.
l r
i.
2 llgi!!
3 5-i.
c) 1'
, ! $..f
,1 3o I
j 3
C i
e i
l w
as-C v
}
I h
C i
I i
4 8
Mo 1u s.
O w
0 CJ L
M 3I}
Jt!
ue o
e G>G s*3 U 2 s lie <n m
- W-sn LO.
u u j M M 3 O 5 e Q } L 3 5 Q W au 5 1..1 2 1 '* l l-u3
3-32 3.2 Fire The objectives of this section are to give a brief presentation of the LGS-SARA approach to quantification of the accident sequences generated as a consequence of fires in tne different critical zones along with the cor-responding results, to descrite the BNL modifications to the quantifica-tion, and to present the revised results. This section is organizec as fol-lows. Section J.2.1 summarizes tne LGS-SARA approacn to quantification of ac-cident sequences and presents the mean values for tne frequency of core-camage for tne different fire zones. Section 3.2.2 presents the detailed BNL review of the different fire types for two fire zones: Fire zone 2, for which the fire growth event tree is similar in structure to all other fire zones with tne exception of the second fire zone described here; i.e., fire zone 25. In this section the fire growth event trees for all other fire zones are also presented, but the details are given in Appendix A. In Section 3.2.3 a sum. mary of the review results is presented. 3.2.1 Overview of the LGS-SARA Accident Secuence Quantification For each critical zone the LGS-SARA (1) report identified the following steps used in the quantification of accident sequences: 1. Identification of potential initiating fires within the fire zone; the following types of fire were considered: a. self-ignited cable raceway fires, b. self-ignited fires in power distribution panels, and c. transient combustible fires. 2. Evaluation of tne frequency of each of the above types of fires witnin the fire zone. . 3. Subdivision of the growth of fires into several intermediate stages between ignition and damage to all safe shutdown systems served by ca-bling or components loccted within the fire zone. ~ s.) -m -~
s' 3-33 4. Evaluation of each fire growth stage in terms of (a) the probability of failing to suppress the fire before reacning each stage, and (b) the shutdown systems tnat remain undamaged at each stage. 5. Evaluation of the conditional probability of core melt at each stage of fire growtn, taking credit only for the reliability of systems not already damaged by tne fire. This was acnieved by modifying tne fault and event trees developed in the LGS-PRA(2), 6. Evaluation of the core damage frequency associated with individual fire growtn stages by comoining the frequency of failing to suporass tne fire at eacn stage of growth and the associated probaoilities of core damage from random failures of the undamaged systems. 7. Summation of tne core damage frequencies associated witn each damage stage for all types of fires to obtain the overall fire-induced core damage frequency for the fire zone. Based on the above desc.'ibed steps a fire-induced core melt frequency of 2.3x10-5/yr was obtained in the LGS-SARA report; the breakdown of the con-tribution of_ the different fire types for eacn fire zone is given in Table 3.2.1 (LGS-SARA Table 4.6, modified to correct some typographical errors). 3.2.2 BNL Revisions in Quantification of Accident Sequences The BNL review of the LGS accident quantification considered eacn of tne steps identified in Section 3.2.1. Review of steps 1 through 4 is described in detail in Section 2.2, and the main disagreements found in this review are surrenarized below. a. A reduction factor of five in the frequency of self-ignited cable raceway fires was used in the LGS-SARA report. As described in Sub-section 2.2.2.1.1, the BNL review indicates that a reduction factor of three is more appropriate, based on the existing data base. b. It is the BNL judgement that the probability of fire suppression suc-cess is overestimated in the LGS-SARA report. Based on the discus-sions in Section 2.2.2.3, the following probabilities of failure to extinguisn a fire in t minutes, P(t), are used in tne BNL review:
3-34 P(10) = 0.43 P(30) = 0.195 P(60)=0.08 The following values were used in the LGS-SARA report: 0.40, 0.15, and 0.04, respectively. Review of step 5, evaluation of conditional procacility of core melt at e'ach stage of fire growtn, is based on the BNL review of the LGS-pRA(7) (NUREG/CR-3028). It is noted that a computer reevaluation of system unavailaoility or core damage fault trees was not made; only nand calculations were performed. The approacn usea in tne reevaluation of steps 6 and 7 is essentially the some as used in the LGS-SARA report; tne results of the review of steps 1 tnrougn 5 are used in the BNL review. In the following sections, a detailed review of accident sequences for fire zones 2 and 25 is described, along with the respective fire growth event trees for the otner zones. In this review, the following will be presented for each fire type: frequency of fire, fire-induced transient, undamaged mitigating systems, and dominant sequences for each fire growth stage. 3.2.2.1 Fire Zone 2: 13kV Switchgear Room a. Quantification of Fire-Growth Event Tree for Self-Ignited Cable-Raceway Fires. The fire growth event tree for fire zone 2 is snown in Figure 3.2.1, and the evaluation of the branch point probabilities is discussed below. Event A: Frequency of Cable-Raceway Fires The frequency of cable-raceway fires is computed by multiplying two quantities: (1) the ratio between the weignt of cable insulation in this zone (8736 pounds) and the total weight of cable in tne reactor enclosure and control structure (172,799 pounds) and (2) the frequency of cable fires per reactor year:
o ~ 3-35 172,799(1b)x (S.3x10-3 8,736(1b) ) = 8.9x10-5/yr, 3 where the frequency of cable fires per reactor year is 5.3x10-3 and, the reduction factor of 3 is based on the BNL analysis of tne data base as discussed'above. Event B: Undamaged Systems Mitigate Accicent Given Fire-Growtn Stage 1 Since most of tne cabling in this fire zone is associated with balance-of-plant (80P) equipment, loss of the power conversion system for inventory makeup and long-term neat removal was assumed. At tnis stage all safety-related equipment is undamaged, and the ccminant accident sequences anc tneir conditional probabilities, based on the SNL review of tne LGS-PRA (NUREG/CR-3028), are as folicws: Class I QUX = 4.9x10-5 QUV = 1.5x10-6 Class II QW = 9.4x10-6 Total (Event B) = 6.0x10-5 Event C: Fire Suppressed Before Damaging Unprotected Raceways It is considered unlikely tnat a cable-tray fire would be suppressed before damaging cables in conduits that are not protected by a ceramic-fiber blanket. A failure probability of 1.0 is assigned to this event (the same as in the LGS-SARA report). Event D: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 2 This stage represents damage to all safety-related equipment except that associated with snutdown metnods A and B (Table 4.1 of LGS-SARA), wnich are served by protected cabling. The dominant accident sequences and their conditional probabilities are as follows: -+ g y--yy- - ~- ,,9 --gyp.i.e-~-y.---.yw w wq-w a .m-.* w y- ,&y +a-g
o t 3-36 Class I QUX = 4.9x10-5 QUV = 6.6x10-6 Class II QW = 4.5x10-3 PQW = 4.5x10-6 Total 4.7x10-3 Event E: Fire Suppressed Before Damaging Protected Raceways This event is concerned with the probability of failing to suopress tnis fire cefore protected cables serving snutdown metnods A and B are damaged. This is equivalent to failure to-suppress the fire witnin one nour after the fi re. This probability is equal to 8.0x10-2, using the BNL curve given in Section 2.2.2.3; the LGS-SARA uses a value of 0.04. Event F: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 3 Fire growth stage 3 represents damage to all safe-snutdown systems served by tne equipment in the fire zone. From the description of this zone it is clear that sucn damage would result in a loss of all systems required for safe shutdown and the resulting conditional probability of core melt is thus 1.0. o. Quantification of the Fire-Growth Event Tree for Equipment-Panel Fires. Tne fire-growth event tree for panel fires is also snown in Figure 3.2.1 and the evaluation of the branch probabilities follows. Event A: Frequency of Panel Fires The BNL review agrees with the frequency of panel fires as calculated in LGS-SARA, i.e., 1.8x10-3/yr. Event B: Undamaged Systems Mitigate Accident Given Fire Growth Stage 1 Since the panels in this zone serve BOP equipment, the initiating event is loss of the oower conversion system and the quantification of tnis event is identical with that described for Event B in Section a. - i
f-3-37 Event C: Fire Suppressed Before Damaging Unprotected ~ Raceways The probability for fire propagation out of a distribution panel was considered to be equal to 0.04 in tne LGS-SARA report. In Section 2.2.2.3 of tnis report there are some qualitative ccmments about how this value was ootained. However, tne SNL review does not cnange this value. Events D, E, and -F Given that a fire has propagated from the panel in wnicn it originated to aajacent cable raceways, the quantification of the conditional probabilities associated witn events D. E, and F is identical witn that described in ~ Section a. c. quantification of tne Fire-Growtn Event Tree for Transient-Ccmcustible Fires. The fire growtn event tree for transient-combustible fires is also presented in Figure 3.2.1, and the evaluation of the branch probabilities follows. Event A The BNL review concludes that tne frequency of transient-ccmbustible fires given in the LGS-SARA report seems to be reasonable; this probability is equal to 1.3x10-6/yr. Events B, C, D, E, and F The evaluation of the conditional probabilities associated with Events 3, C, D, E, and F is identical with that described in Section a. 3.2.2.2 Fire Zone 25: Auxiliary Equipment Room a. Self-Ignited Cable Fires. The frequency of self ignited cable fires in the raceways of the auxiliary equipment room was determined in the same way as described for Event A in Section 3.2.2.1.a. This frequency is given by: %-+ s m -=
O 3-33 l 172,799(lb) x (5.3x10-3 4,400(1b) ) = 4.5x10-5/yr 3 In the LGS-SARA report it is argued tnat based on fire analysis of raised floor sections, a fire initiated in one section will neitner propagate througn installed combustible material (cable insulation), nor cause any camage to caoling in adjacent floor sections. Thus, One maximum fire damage tnat coula rbsult is the loss of one division of safe-snutcown equipment, and assuming the most demanding transient, MSIV closure, tne dominant accident sequences and their conditinal probabilities are: Class I ~QUX = 5.6x10-4 QUV = 9.2x10 5 Class II OW = 7.8x10-7 Total = 6.5x10-4 Using tnese conditional probabilities, the resulting frequency of core melt is: (4.5x10-5) x (6.5x10-4) = 2.9x10-8/yr. b. Self-Ignited Cable Fires. The frequency of cabinet fires in the auxiliary equipment room is estimated as 1.75x10-4/ cabinet year This auxiliary equipment room has four cabinets wnere fires may cause significant damage to safe-snutdown systems. Assuming a fire in any of those cabinets would destroy the contents of the cabinet, the following equipment woulo still remain undamaged: 1. The RCIC or HPCI System 2. Means of Reactor Depressurization 3. The LPCI System (Two Trains) 4 The Core Spray System (One Train) 5. The RHR System 4
l ~ 3-39 Assuming the initiating event is a transient with isolation frcm tne pcwer conversion system (LGS-SARA assumption), the following are the ocminant accident sequences witn tneir conditional core melt prooabilities: Class I QUX = 5.6x10-4 QUV = 2.9x10-5 Total 5.9x10-4 The core melt frequency resulting from self-ignited panel fires is tnerefore; 4 x (1.75x10-4) x (5.9x10-4) = 4.1x10-7/yr. c. Transient-Combustible Fires. Tne frequency of transients-ccmbustible fires were estimated as follows: 3.4x10 4/yr Trasn-can Fire = 3.4x10-4/yr Solvent-can Fire = 3.4x10-5fyr Oil Fires = Heat transfer analysis was used to evaluate cable temperatures resulting from external-exposure fires, and based on this analysis, locations within the fire zone where fires may be significant contributors to core melt were identified, and the area associated with each location is given in Table 3.2.2 (Table 4.4 of LGS-SARA). Using the results in Table 3.2.2 and tne concept of critical location probability (the ratio of the area of tne fire location and the total free area of the auxiliary equipment room associated with Unit 1, excluding the area taken up by cabinets), the core melt frequency is calcu-lated and given in Table 3.2.3. It should be pointed out that the dominant sequences for each fire location are QUX and QUV. 3.2.2.3 Fire Growth Event Trees for Fire Zones 20, 22, 24, 44, 45, and 47 The detailed description of each event in the fire-growtn event trees for zones 20, 22, 24, 44, 45, and 47, as well as their branch probability is given in Appendix A. In the following section, the review results for core damge frequency are presented. 9 ,,,w - - _ ~ - - -- e n. n.., , =
3 40 3.2.3 Review Resul~ts The core damage frequency for each fire zone and for each type of fire as obtained in tnis review is presented in Table 3.2.4. The most important re-suits are: 1. The total core damage frequency resulting from fire-induced transients obtained in tne BNL review is 5.2x10-5/yr, as compared to 2.3x10-5/yr reported in tn'e LGS-SARA report. 2. The difference between the BNL review and the LGS-SARA core damage frequency can De attributed to two factors: (a) tne protability of fire suppression in any given time, and (b) the reduction factor used in the calculation of self-ignited cable-raceway fires (see Section 2.2). 3. Most of tne core damage frequency comes from the fire growth stage 3 (about 85% the both BNL review, and about 81% in LGS-SARA). At this fire growth stage, in almost all zones,all safe-snutdown systems are assumed to be damaged oy the fire. Thus, tne core damage frequency is determined by the initiator frequency and the probability of failing tu suppress the fire within a given time interval. This indicates that the changes made by BNL in the accident sequence quantification (relative to the LGS-PRA quantification) have a small impact upon the total fire-induced core damage frequency. 4 In the BNL review, about 67% of the total core damage frequency comes from the self-ignited cable-raceway fires (about 57% in LGS-SARA). 5. In the BNL review, about 93% of the total core damage frequency comes from fire zones 2, 44, 45, and 47 (about 91% in LGS-SARA). 6. In the BNL review, about 97% of core damage is binned in the Class I category (see LGS-PRA); tne other 3% is Class II. The results presented in Items 1, 3 and 4 show tnat the total core damage frequency is very~ dependent upon the modification made by BNL (Item 2 above).
3-41 Thus, calculations were performed to snow the impact of these two modifica-tions, and the results are as follows: a. If the LGS-SARA probability of failing to suppress the fire within 60 min. (0.04) is used instead of tne BNL value (0.08), the total fire-induced core damage frequency would be equal to 3.6x10-5 reactor / year. b. If tne LGS-SARA redaction factor (RF=5), used in tne calculation of self-ignited cable-raceway fires is used, instead of the BNL Value (RF=3), tne total fire-induced core damage frequency would te equal to 3.Sx10-5/ reactor year. Anotner area where some sensitivity study is warranted is in the evalua-tion of human errors in case of fire-induced transients. Since 97% of the total fire-induced core damage frequency is due to failure of injection, two cases were analyzed here: a. Operator fails to depressurize tne reactor (X in tne accident sequences). The results presented in Table 3.2.4 are based upon the value of X given in tne BNL review of the LGS-PRA(7); i.e., X=6.0x10-3 If this value is increased by a factor of 10, the total fire-induced core melt frequency would be equal to 7.6x10-5 (an increase of 45%). b. Operator fails to initiate required systems from remote shutdown panel (pertinent to fire zones 22 and 24). The results presented in Table 3.2.4 are calculated using a valus of 1.0x10-3 for this error. If a human error probability equal to 1.0x10-2 was used, the total fire-induced core damage frequency would increase to 5.6x10-5 (an increase of 7.7%). m g--r. y- -r*%-'- r +-w++ =* e .-e s r-
i ~ 1 A B C D E F Undama9ed Fus suppsessed Undamaged Fire suppsetied tindanaged Fire in 3,3 3 c,0
- tuloss danaging systeens i
calJ:/TC/ panel nutigate tactose dena9 rn sy stems unpsosected sness94:e peutected siistigate accidsnt given s aceway acculcast givest sxcways dcC8'lcat 9'ven FGSI FGSI IFailure gives FGS 2) FGS2 Ifd"'" 985 fGSll IGS3 h'd
- Asuusal sarysence
{'~ cal.fe 1C gianel isesysency OK ~ e OK OK 8.0-2 8.9-5/1.3-5 / 1.0/1.0/4.0-2 1.0 Ct.t 7.1-6 1.0-6 5.8-6 1.8-3 (se imte) (see not e) 4.7-3 cf.1 4.2-7 4.1-8 1.4-7 6.0-5 cf.1 5.3-9 /.8-It 1.1-7 7.s-6 L,1-6 ii. ?-6 FGS = Fire growth stage 1C - Tsansient comtmastit,te Tata an.iisA cuse snett frequency - 1. 5-5 FJose: Decause of the evaluation of event E, the prob.tahiv of event C it inO! included in the evaluelson of the lequcinCe les;quency. Figure 3.2.1 Fire growth event tree for fina zoito 2 Qi al
. _ ~. -. _ _ _. _ _ - _ _ _ l 3-43 Table 3.2.1 Summarj of Fire-Analysis Results Annual Contribution to Core-Melt Frequency 3 Self-Ignitec iransient-Caole Sel f-Ignited Ccmousticles Fire Zone Raceway Fire Panel Fire Fire Total 2 12-kV switchgear room 2.4-6D 3.2-6 5.9-7 6.2-6 20 Static inverter rcom 5.0-8 3.5-8 1.5-8 1.0-7 22 Cable-spreacing room 6.1-8 NAC 1.9-7 2.5-7 24 Control room Negligible 1.6-7 1.0-7 2.6-7 25 Auxiliary equipment room Negligible 1.0 2.6-7 3.6-7 44 Safeguard access 1 area 4.2-6 1.5-6 4.1-7 6.1-6 45 CR0 hydraulic equipment area 4.7-6 1.0-6 6.6-7 6.4-6 47 General equipment area 1.2-6 5.0-7 1.8-7 1.9-6 1.3-5 6.5-6 2.4-6 2.2-5 Contribution from all other fire zones' 1.0-6 Total annual core-melt frequency from fires 2.3-5 a Point estimates b 2.4-6 = 2.4 x 10-6 2 Not applicable s I c .,y.-.m .,y.._. --,,-,---,--.--,-,-,,,-y., ,,w. ,,,...-_3
t 3 44 Table 3.2.2 Critical Locations of Transient Combustible Materials in the Auxiliary Equipment Room
- 2 Area of location (m )
Systems assumed to ce un-Solvent-Can uil camaged anc capaole of RPV Fire Location Fi re Fire inventory makeup Intersection of floor areas 0 2.4 LPCI train D, means of 1UU792 (a) and 100791 depressurization Intersection of floor (b) areas 7.7 12 LPCI train D, means of 10U791 anc 10U793 cepressurization Floor area 10U795 (c) 0.6 2.3 LPCI trains B and C, means of depressurization Floor area 100789 (d) 0.6 2.3 LPCI trains C and D means of depressurization
- Table 4.4 of LGS-SARA Y
a
w 6re w 3-45 Table 3.2.3 Evaluation of Sequence Frequencies of 011 Fires (Transient Cosibustibles) b Probability' Criticala of Rancom Annuala Location Equipment Core-Pelt 0 Fire Locationa Frequency Probacility Failure Frequency OIL FIRE Location a 3.4-5 0.01 0.022 7.5-9 Location b 3.4-5 0.05 0.022 3.7-8 Location c 3.4-5 0.01 0.014 4.3-9 Location d 3.4-5 0.01 0.014 4.8-9 SOLVENT FIRE Location a 3.4-4 0 0 0 Location b 3.4-4 0.03 0.022 2.2-7 Location c 3.4-4 0.003 0.014 1.4-9 Location d 3.4-4 0.003 0.014 1.4-9 Total 2.8-7C a From LGS-SARA Table 4.5 b BNL Review c The corresponding LGS-SARA value is 2.6-7. i ? 9 _m
t 3-46 Taole 3.2.4 Summary of Fire-Analysis Results BNL Review Annual Contribution to Core-Mel Frequencya Sel f-Ignited Transient-Cable Self-Ignitec Comousticles Fire Zone Raceway Fire Panel Fire Fire Total 2 12-kV swi tengear room 7.5-6 6.2-6 1.1-6 1.5-5 20 Static inverter room 2.4-7 7.5-8 4.3-8 3.6-7 22 Cable-spreading room 3.7-7 UAb 7.4-7 1.1-6 24 Control room nab 4.8-7 2.2-7 7.0-7 25 Auxiliary equipment room 2.9-8 4.1-7 2.8-7 7.2-7 44 Safeguard access area 1.3-5 3.3-6 7.8-7 1.7-5 45 CRD hydraulic equipment area 9.6-6 1.8-6 8.6-7 1.2-5 47 General equipment area 3.9-6 1.7-7 3.7-7 4.4-6 3.5-5 1.2-5 4.4-6 5.1-5 Contribution from all otner fire zones 1.0-6 Total annual core-melt frequency from fires 5.2-5 a Point estimates D Not applicable N
s e 6 3-47 REFERENCES 1. Philadelphia Electric Ccmpany " Limerick Generating Station, Severe Ac-cident Risk Assessment," April 1983. 2. Pniladelphia Electric Company, " Limerick Generating Station, Pecoacilistic Risk Assessment," Maren 1981. 3. Kolb, D. L., et al., " Review and Evaluation of the Indian Point Probabilistic Safety Study," NUREG/CR-2934, SAND 82-2929, Decemoer 1982. 4. Pickard, Lowe, and Garrick, " Indian Point Probabilistic Safety Study," Prepared for Consolidated Ecison Company of New York, Inc., and Power Authority of tne State of New York, 1982. 5. Kennedy, R. P., et al., " Subsystem Fragility," SSMRP-Phase 1, NUREG/CR-2405, UCRL-15407, February 1982. 6. Battel, R. E. and Campbell, D. J., " Reliability of Emergency AC Power Sys-tems at Nuclear Power Plants," NUREG/CR-2989, ORNL/TM-8545, July 1983. 7. Papazoglou, I. A., et al., " Review of tne Limerick Generating Station ~ Probabilistic Risk Assessment," NUREG/CR-3028, BNL-NUREG-51600, February 1983. 1 7 4 -- wr e - t, a -+--,.,-er ~~* .w-.--,-- -- M ,a-
4.1 Seismic Nhzard and Fragility Recommendations 4.1.1 Introduction Many concerns have been raised in Section 2.1 in recard to the seismic hazard and fragility analysis. Recommendations for resolving these concerns are given in this section. These recommendations are primarily directed to PECo and are based on discussions already ore-sented in Section 2.1. Rather than repeatino the background, each recommendation is presented and followed by the applicable subsection in Section 2.1 which can be referred to for additional information. Al so, recommendations are made to the NRC to perfo'rm additional review tasks to complete the review of the LGS-SARA. Section 4.1.2 gives the recommendations for the hazard analysis anc Section 4.1.3 of ves the recommendations for the fragility and associated system analysis concerns. 4.1.2 Seismic Hazard The following recommendations should be addressed by PECo. The numbers in parentheses at the end of each recommendation refer to the subsection of Section 2.1 which gives background information. 1. The delineation of zone boundaries in the Crustal Block hypo-thesis should be reconsidered. Soecifically, a redefinition of Zone 8 is recommended that is better correlated to the pattern of seismicity in the vicinity of Limerick and the geologic structure of the Triassic Basin. (See Section 2.1.2.3.) 2. The possible occurrence of large-magnitude events (i.e., =M7.0) should be considered as an alternative hypothesis on maximum magnitude for each seismogenic zone. The distribution should be selected in consideration of recommendation 4, below. (See Section 2.1.2.3. ) 4-1
3. The uncertainty in Richter b-values should be considered in the seismic hazard analysis. Consideration should be given to the distribution of earthquake magnitudes based on the historical ^ record in each seismogenic zone and expert opinion. (See Section 2.1.2.4. ) 4. Justification should be provided for the estimate of the large-magnitude (i.e., M = 6.8) events considered in the hazard analysis. Specifically, the basis for assumming that the magnitude ' estimated for the 1886 Charleston, South Carolina earthquake is the largest event that can occur should be pro-vided. Also, the basis for not considering uncertainty in this parameter should be justified. (See Section 2.1.2.4. ) 5. The implication of including the Cape Ann events in the Pied-mont source zone should be addressed. Consideration should include recent work that rejects the notion of a Boston-Ottawa seismic belt and the fact that the 1982 New Brunswick Canada event is included in the piedmont province. (See Section 2.1.2.4.) The following recommendation is addressed to the NRC. 1. An independent analysis should be conducted to verify the hazard analysis results. Also, an independent quantitative evaluation of the impact of comments raised in this review should be performed. 4.1.3 Seismic Fragility The following recommendations should be addressed by PECo. The numbers in parentheses at the end of each recommendation refer to the subsection of Section 2.1 which gives background information. l 4-2 j
1. .lustification for using the 1.4 duration factcr to increase tne capacity of structures and the 1.23 factor to shift the hazard curves frcm a sustained-based peak acceleration to an effective peak acceleration should be provided. Speci fically, the con-cern is the region of the Decollement hazard curve at and above 0.40g effective peak ground acceleration (i.e., in the region where the average magnitude is M6.0 or larger). (See Section 2.1.3.1.) 2. Justification should be provided for the median duration factor. Specifically, the median value of 1.4 and the varia-bility associated with this factor should be addressed. A median value which is magnitude dependent (as used in the LGS-SARA) should be developed. Also the uncertainty components of variability of 0.08 should be increased. 3. The revised median capacity value of 0.90g for the reactor enclosure and control structure should be verified. (See Section 2.1.3.3. ) 4. The assumption that the containment building will have an effective damping value of 10 percent at the acceleration levels corresponding to the failure of the reactor internals, CRD guide t'abe, and reactor pressure vessel should be justi-fied. Bot; the damping values for the individual containment components (i.e., containment wall, pedestal, lateral support, and RPV components) and the combined system damping value should be addressed. For the latter concern, either a weighted model damping calculation or a time history reanalysis of the l containment /NSSS model should be conducted. (See Section I 2.1.3.4.) Note that this recommendation has a icwer priority 4-3
since the mean frequency of core melt would increase by only 10 percent for this effect. 5. The implications of impact between the containment building and the reactor enclosure should be addressed for the following concerns: Failure of safety-related electrical and control equipment a. located in the reactor enclosure. b. Failure of safety-related piping which crosses between the two buildings due to relative displacements. In addition, it should be verified that no safety-related components will be damaged by spalled concrete caused by impact of the two structures. (See Section 2.1.3.5.) Finally, it should be verified that failure of small lines attached to the safety-related piping near the junction of the two structures and anchored to the reactor enclosure will not contribute to the frequency of core melt. 6. In regard to the safety-related electrical components which significantly affect the frequency of core melt including, but [ not limited to: o 440-Y bus /SG breakers (Sti) e 440-V bus transformer breaker (S I 12 e 125/250-V dc bus (S13) I 4-4 .] t
~ 4-KV bus /SG (Sg4) e e Diesel-generator circuit breakers (S15) identify the number of actual components, their locations, and their characteristics relative to the generic tests at Susque-hanna which were used to derive their capacities. J usti fica-tion should be provided for the number of each ccmponent type which should be included in the Boolean equation for sequence ~.'s sE VX. Consideration should be given to the possible effects of capacity and response dependencies which exist. (See Section 2.1.3.6.) 7. Justification should be provided that the test results for the Susquehanna components can be directly scaled by the ratio of the design SSE values for the two plants (i.e., Limerick and Susquehanna) and used to develop capacity values for the fol-lowing Limerick components: e Hydraulic control unit (5 ) 7 e 440-V bus /SG breakers (Sit) e 440-V bus transformer breaker (S12) e 125/250-V de bus (S13) e 4-XV bus /SG (S14) e Diesel-generator circuit breakers (SISI 4-5 y ,r, , - -,.- - w " ~
Consideration should be given to the location of the ccmponents in the two plants, foundation conditions, and construction similarities. It is recommended thtt fragility parameter values specifically calculated for each of the above components at Limerick be developed. (See Sections 2.1.3.6 and 2.1.3.7. ) 8. The capacity parameters for the SLC test tank should be based on a component-specific analysis which includes tne dynamic characteristics of the tank and the actual geometric configura-tion. The capacity of the anchor bolts should be checked and the earthquake component factors derived based on the actual response and capacity characteristics. (See Section 2.1.3.7, Component S.) g 9. The similarity between the nitrogen accumulators at the Limer-ick and Susquehanna plants should be verified since the anal-ysis from Susqeuhanna was used as the basis for the capacity of the nitrogen accumulator at Limerick. (See Section 2.1.3.7, Component Sg.)
- 10. The possible failure of the SLC tank due to tearing of the base plate flange near the anchor bolts should be checked to verify that it is not the weakest capacity. (See Section 2.1.3.7, Component S10 I
- 11. A specific analysis should be conducted for the diesel ganera-tor heat and vent which is based specifically on the character-istics of this component.
(See Section 2.1.3.7 Component S16-)
- 12. Verification should be provided to document that the fragility values for valves include consideration of potential leakage through the internal components bypassing the closed valve 4-6 sJ
barrier. Specific consideration should be given to the MSIV and the purge and vent valves. Verification should also be provided to document that seismic motions will not cause SRVs to stick open. (See Section 2.1.3.8.)
- 13. Verification should be provided to document tnat soil pressures on the embedded portions of the reactor enclosure walls do not reduce the capacity of these walls and thus decrease the capa-city to resist in-plane lateral loads. (See Section 2.1.3.3.)
- 14. After construction of the plant is ccmpleted, a systematic review of the plant, including walkthroughs, should be con-ducted to locate secondary components which could fail, fall, and impact primary safety-related ccmponents. Analyses of potential failures should be conducted to determine whether the secondary components are weaker than the primary components already considered. (See Section 2.1.3.8.)
- 15. The percentages of occurrences when evacuation would be affected by earthquakes should be recalculated using realistic relationships between damage to civil structures and ground acceleration.
(See Section 2.1.3.8) The following recommendations are addressed to the NRC. 1. A followup rev1t., should be conducted to independently verify the capacity values used for the electrical components. A coordinated task between nuclear systems and structural engin-eers should be performed since these components are major contributors to the mean frequency of core melt. 4-7
2. Other significant nonelectrical components are based on generic capacities. Independent, specific calculations should be performed for the following' components since they are important to the final risk. e Hydraulic Control Unit (S ) 7 e Nitrogen Accumulator (S ) 9 e e Diesel Generator Heat and Vent (316) t i f l-l I 4-8 3
-~~ - 4-9 4.2 FIRE The methods to evaluate the risk due to a fire in a nuclear power plant (NPP), as described within the Limerick SARA, and as reviewed herein, can ce divided into three categories for the development of ignition, detection, suppression and propagation models: physical models, point probability models, and probabilistic models. The Limerick SARA attempts, and.in our judgement rightly so, to use a hybrid of all three. A hybrid approacn is inceed war-ranted. Physical models suffer frcm the ccaplexity of the large numoer of variables and relationships required to calculate a fire history. Point probability models suffer frcm small and ir. adequate data bases. While a com-pletely probabilistic approach also suffers from data base inadequacy, it, more imcortantly, suffers from an inability to accurately model certain phases of fire development. To put the issues of fire-development modeling in proper perspective, let us censider those conpanents of the fire wnich are relevant in assessing fire growth: the burning object, the flame, the hot layer, the cold layer, the vents within an enclosure, target objects (other camoustibles), and inert surfaces (walls and ceilings). As Friedman [13 points out rather simplisti-cally, 20 interaction vectors involving heat and material flux exist between these seven components. Several of these interactions have multiple elenents with positive feedback as a critical part of the fire growth phenomena. Adequate knowledge of the various feedback loops should suffice, in prin-ciple, to permit description of the growth rate of the fire. However, in order to make scfety assessments, it is also mandatory to have additional in-formation such as carbon monoxide and smoke content, for its impact on plant personnel safety. More importantly, from a public risk viewpoint, it is necessary to have information on the plant damage states as a function of fire growth. l Indeed, assessing fire risk is a highly coupled, nonlinear, dynamic pro-cess. We at BNL are of the opinion that the state-of-the-art in fire - modeling, coupled with such canplex issues as systems interaction frca automatic / manual suppression and human error, is such that probabilistic analyses which purport to quantify the safety of NPPs in the event of a fire have a wide range of uncertainty. p,w, e t -e+-,,w - - ~ -. - - ,w m-----
r-ev
4-10 Furthermore, the very conservative assumptions used in the Limerick SARA fire analysis (in most respects) asy, without proper context, lead to a distortion of perspective for fire risk relative to other risks at the plant. In some respects, assumptions and submodels that are touted to be con-servative are tantamount to gross violations in physical realities. Several cases in point have been discussed in the previous sections - not linking a suppression model directly to the fire growth model; a mass-loss rate model that does not truly reflect the positive feedback of the various fire grow n stages; an ignition-timemodel that does not adequately reflect tne various ' heat-exchange mechanisms are some of the modeling inadequacies wnich have been addressed directly. The Limerick SARA on fire a1alysis has only consicered intracone fire prcpagation. ' A true assessment of fire risk must consider inter:ane fire propagation and all aspects pertaining thereto including the decilitating effect of smoke mig' ration. While this latter facet has no immediate bearing on component reliability, it should have immediate ramifications with regard to manual suppression effectiveness. Hence, smoke propagation is one aspect that should have been considered even if its level of sognistication can be construed to be only on a par with the physical models used in ascertaining the thermal history. In this connection, the mechanisms by which fire suppression systems (automatic and/or manual) can cause the failure of redundant or diverse safety systems should be considered in the assessment, again to a level of detail conmensurate with the probabilistic/ deterministic analysis that is applied to assess fire risk. The foundation on which the fire propagation model, basically a one-room fire model, rests is sound. Various compartment fire modelsZ have been developed and COMPSRN can be considered as one which lies within their spectrum of sophistication. COMPBRN, along with other fire models, uses a control volume, or " zone approach," in lieu of those models which discretize the governing differential equations directly, the so-called " field-model approach."
4-11 This zonal approach has several important advantages: (1) canputational simplicity, (2) ease of decoupling zones for independent investigation; (3) simpler conparison of theory and experiment for individual zones, and (J) easier conceptualization of the interacticn between zones. Field models, however, in the long run should provide the most general, accurate, and detailed prediction of fire development. However, at present field models: (1) are limited by computer capacity, (2) do not yet :rcoerly treat action-at-a-distance radiative energy transfers, and (3) are still awaiting a more rigorous treatment of buoyancy driven turbulence. Botn the zone and fiela . approach should, in BNL's jucgement, be pursued with the field approach used as a basis for " fine-tuning" the unit models that are built into the zone-model aoproach. Zone mcdels, like CCMPSRN, represent a nearer tern engineering accroacn which is closely tied to experimental ocservations. However, a basic philo-sachical limitation in zone-model structure is in its emphasis in predicting roam flashover. For assessing nuclear power plant risk, predicting the onset of flashover is not as crucial as predicting the effects of in-place component vulnerability during the earlier fire-growtn stages. As such, for canplete-ness a larger spectrum of initiating fire sizes must be incorporated into the analysi s. Accordingly, several of the unit-models employed in the zone approacn re-quire improvement.[2] Other aspects of fire growth that are lacking in existing models (like COMPBRN) are needed. For direct application in assessing nuclear power plant fire risk, these additional models snould reflect the possibility of (1) the effects of walls, corners, and costacles on fire plume and thermal plume development, (2) the possibility of combustion of excess pyrolyzate within the. stratified layer, (3) the effects of turbulent-induced buoyancy on plume development (4) intra zone mass and er.ergy exchange, (5) and implementation of existing knowledge 'and correlation of fuel-flammability characteristics, specifically current cable flammability and damageability indices. I i l i l- '~ ~~~ .1
4-12 Another keypoint regarding the practical use of a zone model in general, and COMPSRN in particular, is that the structure of the numerical code is not " user friendly". Before one can use a coce employing a series of unit mocels,, one must have an awareness of the assumptions that are built into the analy-sis, the key physical parameters and their sensitivities, and finally a working knowledge of the state-of-the-art in fire phenomena and modeling. 1 e e
4-13
4.2 REFERENCES
1. Friedman, Raymond, " Status of Mathematical Modeling of Fires," Factory Mutual Research Corporation, FMRC RC 81-BT-5, April 1981. 2. Jones, Walter J., "A Review of Ccmpartment Fire Mocels," NSSIR 33-2682, April 1983. 1 i 7 m
I t A-1 APPENDIX A DETAILED REVIEW OF THE QUANTIFICATION OF THE FIRE-GROWTH EVENT TREES In this appendix tne detailed review of tne fire-growtn event trees for tne following fire zones is described: 1. Fire Zone 20: Static Inverter Room 2. Fire Zone 22: Cable-Spreading Room 3. Fire Zone 24: Control Room 4 Fire Zone 44: Safeguard Access Area 5. Fire Zone 45: CRD Hydraulic Equipment Area 6. Fire Zone 47: General Equipment Area A.1 Fire Zone 20: Static' Inverter Room Tne fire growtn event-tree for all types of fires in zone 20 is snown in Figure A.1. A.l.a Uuantification of the Fire-Growth Event Tree for Self-Icnited Cable-Raceway Fires. Event A: Frequency of Self-Ignited Cable-Raceway Fires. Tnis frequency is calculated in the same way as for Event A in Section 3.2.2.1.a. i.e., 9,558(lb) 5.3x10-3 172,799(10)x ( ) = 9.8x10-5fyr, 3 Event B: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 1. Based on the locality of the initial fire a reactor trip transient is assumed, with the loss of one division of safety related equipment. The dominant seqences and their conditional probabilities are as follows: Class I QUX = 8.4x10-6 1 -QUV = 1.1x10-5 ,y y ,, _, -,,.,, ~ r
- __g,
-m
e A-2 -Class II PW = 4.5x10-6 QW = 5.4x10-6 Total = 7.0x10-5 Event C: Fire Suopressea Before Damaging Unprotected Raceways. ~ The procability of this event is given by tne probability of failing to suppress tne fire within 10 min. (estimated time before damage to unprotected raceways). BNL value for this event is 0.43. Event 0: Uncamaged Systems Mitigate Accident Given Fire-Growtn Stage 2. Fire growtn Stage 2 represents damage to all safety-related equipment except that associated with shutdown method A wnicn is served by cable raceways protected with ceramic-fiber fire blankets; also unaffected is equipment associatea witn tne power conversion system. Tne dominant sequences and their conditional probabilities are as follows: Class I QUX = 8.4x10-6 QUV = 2.6x10-5 Class II PW = 6.6x10-6 QW = 7.9x10-6 Total = 1.1x10-4 Event E: Fire Suppressed Before Damaging Protected Raceways. The probability of this event is given by the probability of failing to suppress the fire within 1 hour ' estimated time before damage ta protected raceways). BNL value for this probability is 8.0x10-2, Event F: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 3. This stage represents damage to all safe-snutdown systems served by equipment in this zone. Only the power conversion system would remain undamaged to mitigate tne accident. The dominant sequences with their j e
~ A-3 conditional probabilities are: Class ! QUV = 2.0x10-2 i Class II P'd = 1.0x10-2 Total = 3.0x10-2 A.1.D Quantification of the Fire-Growth Event Tree for Panel Fires. Event A: Frequency of Panel Fires. BNL agrees witn tne frequency of panel fires given in LGS-SARA, i.a., 4.4x10 4/yr. Event B: Uncamaged Systems Mitigate Accident Given Fire-Growtn Stage 1. Based on the panels located in this zone, reactor trip transient is assumed, and the following equipment is assumed to have failed: HPCI, RHR Trains B and D and Train B of LPCS. The dominant accicent sequences, and tneir conditional procabilities are: Class ! QUX = 8.4x10-6 QUV = 1.1x10-5 Class II PW = 4.5x10-5 QW = 5.4x10-6 Total = 7.0x10 5 Event C: Fire Suppressed Before Damaging Unprotected Raceways. The quantification of this event is identical with that for Event C in Panel Fires for Fire Zone 2 (see Section 3.2.2.1.b). Y r
- w
( ~ A-4 Events 0. E, and F The quantificatio of the conditional probabilities associated with those events is identical witn that described for self-ignited cable-raceway fires in Section A.1.a. A.1.c Quantification of the Fire Grewtn Event Tree for Transient-Combustible Fires. Event A: Frequency of Transient-Comeustible Fires. SNL agrees witn the frequency of transient-comoustible fires as calculatec i n LGS-SARA, i.e., 1. 7x10-5/yr. Events B, C, 0, E. and F The evaluation of the conditional probability associated with Events 8, C, 0, E, and F is identical with that described in Section A.1.a. A.2 Fire Zone 22: Cable-Spreading Room The fire growth event tree for all types of fires in Zone 22 is shown in Figure A.2. A.2.a Quantification of the Fire Growth Event Tree for Self-Ignited Caole Raceway Fires. Event A: Frequency of Self-Ignited Cable-Raceway Fires. This frequency is calculated in the same~way as for Event A in Section 3.2.2.1.a. i.e., 172,799(ib)x (5.3x10-3 35,526(1b) ) = 3.6x10-4/yr. 3 Events B and C Since all fires are capable of damaging adjacent cable raceways, except those protected by a ceramic-fiber blanket, svent B is effectively omitted and Event C is assigned a probability of 1.0.
A-5 Event D: Undamaged Systems Mitigate Accident Given ' Fire Grcwtn Stage 2. The initiating event is a transient witn isolation from the power-conversion system and tne only equipment potentially operable is tnat associated with shutdown metnods A and B. The acminant accicent sequences anc their conditional procaDilities are: l Class I QUX = 4.9x10-5 QUV = 6.6x10-5 Class II QW = 2.7x10 4 PQW = 4.5x10-5 Total = 4.3x10-4 Event E: Fire Suppressed Before Damaging Protected Raceways. The protected raceways (serving shutdown metnods A and B) consist of caole trays protected by a 1" thick ceramic-fiber blanket wnich is equivalent to a 1/2 hour fire rating. Thus, Event E is assigned a probacility of 1.95x10-1 wnich is the prcDanility of failing to suppress a fire witnin 1/2 hour. Event F: Undamaged Systems Mitigate Accident Given Fire Growth Stage 3. At this stage all safe-shutdown equipment dependent on cabling within this zone is considered to be damaged. The only equipment that is potentially operable is that served by tne remote shutdown panel. Therefore, the cominant accident sequences and their conditional probabilities are: Class I QUX = 4.2x10 4 QUV = 2.2x10-3 Class II QW = 4.0x10-4 PQW = 6.6x10-5 Total = 3.1x10-3 0 v,
e A-6 A.2.b Quantification of Fire Growth Event Tree for Transient Cemeusticle Fi re s_._ Event A: Frequency of Transient-Comoustible Fires. BNL agrees with the frequency of transient-combustible fires presentec in LGS-SARA, f.e., 7.2x10-4/yr. Events 8. C, D. E, and F The quantification of all those events is icentical witn that discussed in the previous section (see Section A.2.a). A.3 Fire Zone 24: The Control Rocm Since there_is no exposed cable. insulation in the control rocm, the only types of fires analyzed in tnis section are: Self-Ignited Panel Fires and Transient-Combustible Fires. A.3.a Quantification of Fire-Growth Event Tree for Self-Icnited Panel Fires. The fire-growth event tree for self-ignited panel fires is snown in Figure A.3. Event A: Frequency of Self-Ignited Panel Fires. The frequency of significant panel fires in the control room was estimated to be 1.8x10-3, Event B: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 1. This stage represents damage that is confined to the cabinet in wnich the fire starts. There are 17 separste cabinets in the control room. However, only fires in 3 cabinets can cause significant damage. Fires in one of those cabinets may disable all systems required for reactor shutdown with the exception of e,quipment that may be controlled from the remote shutdown panel. Fires in the other two cabinets will only disable the power-conversion system. r J O
9 e A-7 The transient resulting from any of tnese fires is Loss of Feecwatae or hSIV Closure and tne dominant accident sequences, witn tneir condition 45 probabilities are: Class ! QUX = 3.1x10-5 quV = 1.3x10 4-Class II QW = 2.5x10-5 PyW = 3.9x10-6 Total = 1.9x10 4 Event C: Fire Suppressed Before Propagating Beyond tne Confinement of tne Caoinet. BNL agrees with the evaluation of the probability of a cabinet fire propagating beyond the confinement of the cabinet as given in LGS-SARA, i.e., 2.5x10-2, Event _D: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 2. In tnis stage it is considered that only the equipment which can be oper-ated from the remote shutdown panel is potentially operable. The dominant ac-cident sequences and their conditional probabilities are identical with those calculated for Event F in Section A.2.a. i.e., the total conditional core dam-age probability is equal to 3.1x10-3, l A.3.6 Quantification of Core Damage Probability for Transient-Combustible l Fires. BNL agrees with the quantification of the frequency of transient-l combustible fires which can damage safe-shutdown equipment in the control f room. This frequency is edual to 7.2x10-5/yr. Given the occurrence of a transient-combustible fire it is assumed that the only potentially operable equipment is that whicn can be operated from the remote snutdown panel. In this case the dominant accident sequences and their l l i I
A-8 conditional probabilities are identical witn those calculated for Event F in Section A.2.a; i.e., the total conditional core damage frequency is equal to 3.1x10-3 So, the total contribution of transient-combustible fires to the core dmage frequency is given by: (7.2x10-5) x (3.1x10-3) = 2.2x10-7/yr. A.4 Fire Zone 44: Safeguarc Access Area l The fire-growtn event tree for all types of fires in Zone 44 is shown in j Figure A.4. 1 A.4.a Quantification of tne Fire-Growth Event Tree for Self-Ignitec Cable Raceway Fires. Event A: Frequency of Self-Ignited Cable-Raceway Fires. This frequency is calculated in the same way as for Event A in Section
- 3. 2. 2.1.1, i.e.,
28,290(lb) .5 3x10-3 172,799(LB )' 3 Event 8: Undamagec Systems Mitigate Accident Given Fire-Growtn Stage 1. The accicent initiating event was taken to be a transient witn MSIV closure, and at this stage of tne fire the following systems would remain l potentially operable: RCIC or HPCI system, tne ADS, the RHR system (tnree trains), and the LPCS (one train). The dominant accident sequences and their conditional probabilities are: t-Class I l QUX = 5.6x10-4 l QUV = 9.2x10-5 Class II QW = 7.8x10 7 Total = 6.5x10-4 t ( ~. I i I
A-9 Event C: Fire Suppressed Before Damaging Unprotected Raceways. The probability of this event is given by the probability of failing to suppress tne fire within 10 minutes (estimated time before damage to unprotected racewyas). BNL value for the prooability of this event is 0.43. Event D: Undamaged Systems Mitigate Accident Given Fire-Growtn Stage 2. Given fire-growtn stage 2, the following equipment would remain potentially operable: the ADS and the RHR system (2 trains). Tne dominant accident sequences and tneir conditional probabilities are as follows: Class I QUX = 6.0x10-3 quV = 8.2x10-3 Total = 1.4x10-2 Event E: Fire Suppressed Before Damaging Protected Racewtys. The probability of tnis event is given by the probability of failing to suppress the fire witnin 1 hour (estimatsd time before damage to protected raceways). However, since only fires in two quadrants can grow to this stage, tne probability of Event E is given by: P(Event E) = 0.5 x Probability of Failing to Suppress tne Fire Within 1 nr. = 0.5 x 8.0 x 10-2, 4.0 x 10-2 Event F: Undamaged Systems Mitigate Accident Fire-Growth Stage 3. In this zone, fire growtn Stage 3 repr'esents damage to all shutdown methods, and consequently the conditional failure probability of Evrt
- is 1.0.
A.4.b Quantification of the Fire-Growth Event Tree for Fires in Power-Distribution Panels. Event A: Frequency of Fires. The frequency of panel fi ns 3, l imiined from the number of panels s )
A-10 multiplied by the frequency.of fires for panel-yeas. As described in Section 2.2.4.1, seven panels are located in this zone. Thus, the frequency of panel fires is: 7.x (2.2x10 4) = 1.5x10-3/yr. Event B: Undamaged Systems Mitigate Accicent Given Fire Growtn Stage 1. In this zone only fires in three panels are capable of causing initiating e' vents (turbine trip transient) and damaging mitigating systems. Such fires cause, at tnis stage, the loss of either tne RCIC of the HPCI system. The doninant acciaent sequences are: Class I Q'J X = 5.2x10-6 Class II QW = 1.1x10-8 PW ='9.4x10-8 Total = 5.3x10-6 Event C: Fire Suppressed Before Damaging Unprotected Raceways. The evaluation of the probability of this event is identical with tnat for Event C in Section 3.2.2.1.b. Event D, E, and F Once the fire has propagated to cable raceways, the quantification of Events D, E, and F is identical with that given in Section A.4.a for self-ignited cable Raceway Fires. A.4.c Quantification of the Fire-Growth Event Tree for Transient-Combustible Fires Event A: Frequency of Fires. BNL agrees with the frequency of fires calculated in LGS-SARA, i.e., 1.7x10-5fyr, G e
r ' I A-11 Events B, C, D, E, and F The quantification of these events is identical with tnat given in Section A.4.a fo' Self-Ignited Cable Raceway Fires. A.5 Fire Zone 45: CRD Hydraulic Ecuicment Area The fire-growen event tree for all types of fires in Zone 45 is sncan in Figure A.S. A.5.a Quantification of the Fire-Growth Event Tree for Self-Ionited Cable-Raceway Fi res. Event A: Frequency of Self-Ignited Cable Reaceway Fires. This frequency is calculated in the same way as for Event A in Section 3.2.2.1.a, i.e., 18,637(1b) L.3x10-3 x( ) = 1.9x10 4/yr. 7 7g Events B and C Tne quantification of tnese events is identical with tnat for tne same events in Section A.4.a. Event 0: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 2. .This stage represents damage to all safety-related equipment except that served by cable raceways or components protected by horizontal separation or ceramic-fiber fire blankets. The only quipment potentially operable is tnat served by shutdown method A or B (but not both). The dominant accident sequences and their conditional probabilities are: Class I QUX = 5.6x10 4 QUV = 1.9x10-3 Class II QW = 4.0x10-4 QUW = 3.7x10-5 PW = 6.6x10-5 Total = 3.0x10-3 + y n .g ,,m. n. ,, h ,e.
.___...e A-12 f Event E: Fire Suppressed Before Damaging Protected Raceways. The probability of this event would be given by tne probability of failing to suppress the fire within 30 minutes (time to damage to protected raceways). However, only fires in one quadrant (northeastern) are capabl,e of damaging equipment associated with botn snutcown metnods. So, tne probanility of Event E is given cy: 1.95x10-1/4 =a. 875 x10-2 Event F: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 3. This third stage of fire-growth represents damage to all safe-shutdown equipment, and tne failure procability associated witn tne event is 1.0. A.5.b Ouantification of the Fire-Growth Event Tree for Self-Ignited Panel Fires. Event A: Frequency of Panel Fires. ' BNL agrees witn LGS-SARA evaluation of panel fires in this zone, i.e., 3 x (2.2x10-4) = 6.6x10-4/yr. Event B: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 2., This stage represents damage that is confined to the panel in which the fire starts. Fires in two of the three panels can cause a turbine trip transient-and at,the same time disable one high pressure injection system (HPCI or RCIC) and one RHR train. The dominant accicent sequences and tneir conditional probabilities are as follows: i 1 ass I w QUX = 1.1x10-5 QUV'= 1.8x10-6 Class II PW = 1.3x10-7 QW = 1.6x10-8 i Total = 1.3x10-5 0 N 4 e. 7 =rm-9 -ay s w e - +.. -- --+w-- mye..--.y%,-- -g- -++ >my-y. p -e q-. g-b-wh
s e A-13 Since only two of the three panels can contribute to the accident sequences, the probacility of Event B is: 1.3x10-5 x 2/3 = 9.0x10-6, Event C: Fire Suppressed Before Damaging Unprotected Raceways. The probability of this event is icentical with tnat for tne same event in Section 3.2.2.1.b. Event 0: Undamaged Systems Mitigate Accident-Given Fire-Growtn Stage 2. The quantification of this event is identical with that of Event 0 in 'Section A.S.a. Event E: Fire Suppressed Before Damaging Protected Raceways. Only fires in one of the tnree panels are capable of damaging protected raceways. So, the probability of this event is given by: 1 -- x Probability of failing to suppress the fire witnin 30 minutes 3 (time to damage to protected raceways) = 1-- x 1.95x10-1 = 6.5x10-2, 3 Event F: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 3. This stage represents damage to all safe-snutdown equipment, and the failure probability associated with this event in 1.0. A.S.c Quantification of the Fire-Growth Event Tree for Transient-Combustible Fires. Event A: Frequency of Transient-Combustible Fires. BNL agrees with the frequency given in LGS-SARA, i.e.,1.7x10-5/yr. Events B. C, D, E, and F Given a transient-combustible fire that causes the ignition of cable l
4 A-14 trays, the evaluation of all tnese events is ider.tical with tnat for the same events in Section A.5.c. A.6 Fire Zone 47: General Equipment Area Tne fire-growtn event tree for all types of fires in Zone 47 is snown in Figure A.6. A,6.a Quantification of tne Fire-Growtn Event Tree for Self-Ignited Cable Raceways. Event A: Frequency of Self-Ignited Cable Raceway Fires. Tais frequency is calculated la tne same way as for Event A in Secticn
- 3. 2. 2.1.a i. e.,
172,799(1b) x (5.3x10-2 17,791(lb) ) = 1.8x10-4/yr. 3 Events B, C, and D The quantification of these events is identical with tnat for tne same event sin Section A.S.a. Events E: Fire Suppressed Before Damaging Protected Raceways. The probability of this event would be given by the probability of failing to suppress the fire within 1 hour (time to damage to protected raceways). However,.only fires in one quadrant (NE) are capable of damaging equipment associated with both snutdown methods. So, the probability of Event E is given by: 8.0x10-2/4 = 2.0x10-2, Event F: Undamaged Systems Mitigate Accident Given Fire Growth Stage 3. This stage of fire represents damage to all safe-shutdown equipment, and the probability associated with this event is 1.0. A.6.b Quantification of the Fire-Growth Event Tree for Self-Ignited Panel Fires. Event A: Frequency of Self-Ignited Panel Fires. ) l l~
ss A-15 8Ni. agrees witn the frequency of panel fires given in LGS-SARA, i.e., 5 x (2.2x10-4) = 1.1x10-3/yr. Event B: Undamaged Systems Mitigate Accident Given ' Fire-Growth Stage 1. Fires in three of the five panels in tnis zone may be capaole of causing an initiating event and disable one RHR train anc one core spray train. Tre initiating transient was assumed to be an MSIV closure, anc tne dominant xcident sequences and tneir conditional probabilities are: Class I QUX = 4.9x10-5 QUV = 8.1x10 6 Total = 5.7x10-5 Since only fires in t'hree of the five panels are contributors to those sequences, the probability of Event B is given by: 3-- x 5.7x10-5 = 3.4x10-5 o Event C: Fire Suppressed Before Damaging Unprotected Raceways. The quantification o'f this event is identical with that for Event C in panel-fires for Zone 2 (Section 3.2.2.1.a). Event 0: Undamaged Systems Mitigate Accident Given Fire-Growth Stage 2. Given that a fire has propagated from the panel in which it originated to adjacent raceways, the quantification of this event is identical to Event 0 in Section A.6.a. Events E and F It is BNL judgement that, since none of the existing panels are located in the NE quadrant, the progression of the fire to fire growth Stage 3 is not possible in this zone. ,m
e A-16 A.6.c Quantification of the Fire-Grcwth Event Tree for Transient-Comcusticle Fires. Event A: Frequency of Transient-Combustible Fires. BNL agrees with the frequency of transient-ccmbustible fires given in LGS-SARA, i.e., 1.7x10-5, Events B, C. D, E, and F Given a transient-combustible fire that ignites cable trays, the quantification of these events is identical with that in Section A.6.a. 1 S t + U
'. ~ A B C D E F Undamaged Fue suppressed Undamaged Fire suppressed tJndamaged Fire in sys tema tielore danaging systems tactose damagina 8V"'*8 cdeldTC/ panel mitigate unproiccied accident given metagate proseceed nutisate s aceway accident given raceways accident given FGS1 FGSI (Failure gives FGS 2) FGS2 IF *'"'" 9*5 FOS 3 3 FGS3 l 1 (f Annual sequence frequency rafale 1C panet OK f OK OK 6.0-2 9.8-5/1.7-5/ ~4 ! '4 !4* ' 'O' CM 2.3-7 4.1-8 4.2-8 4.4-4 (see (see note) note 1.1-4 CM 4.6-9 8. 0-I s l.9 ' 7.0-5 CM l.0-9 l.2-9 3.1-8 FGS = Fire growth staue 1 4-7
- 4. b8 7.5-H, 1C Transicna comtaussittle Tot.: annu. cues melt frequency - 3.6-7 flote: Because of the cveluation of event E. ihe prota.t2ileiy of eunt C is not encluded in the evaluatiori of trne sequssics frwqistricy.
Figure A.1 Firegrowtli cuent tree for fire zono 20 l
A B C D E F Undemaged F ase suppsessed Undamaged Fise suppressed undanaged II'8 In systems g,elose danagmg syssesns belote damagiswJ 8Vl! cms cable /TC/ panel nutigale unpsosecicd managate proscceed smisgate accident given s aceway xcedent geven saceways accident given FGSI FGS1 IFailuse gives FGS '2) FGS2 IFdi'"'c ueves FGSll FGS3 k"' Annual sequence {- cable TC panel frequency OK l l OK I OK i 1.95-1 l 1.0 3.1-3 i ].6-4/7.2-4 ct.t 2.2-7 4.3-7 4.3-4 l Cl1 1.5-7 3.1-7 l NA Ct.1 1,7-7 7.4-7 FGS = Fire growth stage l 1C - Tsansient combustible Total annual cose melt fesquency *1.1-6 fJote: Decause of the evaluation of event E. une psobatelity of evens C is not encluded in the evaluation of the sequcuce fsuquency. Figure A.2 Fire-growth event tree for fire zone 22 l O 'l
e r s r 7_ g, 7 7 7 a le e ) y n 4, g2 4 2 ( a , s p e 1 3 7 e p c s n c e si iy f pcn re C y e s s / c p T n lar e ue n u s i nI a g a c r r A s f i M t C i tn u a t e n ru 1 i o t 4s K e f t, i k" 8"l O c c t I. i s a l to u t T nu C 42 enoz e n i l ro f ee r t tneve h t wo rg n e d e i 7. r ys v ei ant g F 2 nia D ot t9. t S e nG ds e yi F 3 ns ml 3 n u c x A 1 3 e r I. u t iu ) n 2 F e S gm G d n e ei n F s di f s s a n e e e r r o v C p pc ,c, ig ps u d ;, s e ni se r a u e o o cl s t y i i c e, ta a g Ft t oI le ba t es g* 8 a n ts n d e d iv e o gseg ht c a nt I at wt ne B r egn S un i e G e e atit d F 4 9i s l' d ymi s ns n. c e U c i 9 r is a FT 1
S i GC c n. F1 i I 3 A P S inG 8 F e 1 s i F l;
- t l.
A 8 C D E F Undamaged Fase suppsessed Undamaged Fise suppressed Undsnaged I'8 In system % in: lose danag' g systems belose damagin2 ' Y * 8 m cat,le/TC/ panel nusegale unpsosected mingaie prutected nungate accident given g a(cway acculent given saceways acchtent given FGSI FGSI (Failure gives FGS 2) FGS2 (Failure gaves FGS1) F GS1 k"# Annual sequence d_, fsequeswy cable TC panel OK OK OK 4.0-2 0.43/0.43/0.04 1.0 2.9-y 1.7-5 1.5-3 (see (see note) iote) l 1.4-2 o CLI 1.7-6 l.0-7 1.5-7 6.5-4/6.5-4/5.3-6 Cf_1 1.9-7 l.1-8 /.9-9 l.1-S /.11-7 1.1-6 C auc at tible ''"""""'I'I9"*"'Y ~l.7-S = flote: Because of the evaluation of event E, the psc.b tulety o: event C is not encluded in the evaluation of the sequence iscquency. j Figure A.4 Fire-growth event tree for fire zono 44 ( l
l ~,. A B C D E F Undamaged Fue suppsessed tJadamaged Fire suppressed undarnaged Fire in systems wgo,e denagmg systems belose denagma systems 3 cable /TC/ panel nutigate unproincied snetagase psutected nutigate accident given s aceway accederit given saceways acculent given FGSI FGSI (Failure gives FGS 2) FGS2 (Failure gives FGS3) FGS3 Annual sequence 'g,, fa s tracocy catala 1C Inasiel [, OK 4 OK OK i '. 875-2/4.875-2/6.50-2 0.43/0.43/0.04 1.9-4/l.7-5/ l.0 C" 6.6-4 9.2-6 3.3-7 1.7-6 i (see (see no t. e) acte) 3.0-3 'M 2.4-7 2.2-8 1.9-8 i 6.5-4/6.5-4/9.0-6 CM 1.2-7 1.1-8 5.9-9 9.6-6 3.6-7 1.8-6 FGS = Fire growth staue i 1C Tsasisient coentassiible Tuid annua conc melt frequency g flote. Because of the evaluation of event E. tric probability of event C is not socluded in the evaluation of the sequence facquency, i Figure A. S. Fire-growtli event tree for fire zone 45 dl
A B C D E F Undamaged Fue suppsessed Undamaged Fire suppressed undamaged EI'8 In systems tulose d. unaging systems ticfore damagisq systems cable /TC/ panel nutigate unpsotected mitigate protected smtigate accident given saceway accident given saceways accailent given FGSI FGSI (Failure gives FGS
- 2)
FGS2 (Failuse gives FGS3) FGS3 k"' '* Annual seqinence g,~ catile TC panel frequency UK OK OK 2.0-2/2.0-2/NA 3, g _4 g, 7_ g 0.43/0.43/0.04 1.0 gg 1.1-3 (se 2 (see note) note) 3.0-3 N 2.3-7 2.2-8 1.3-7 6.5-4/6.5-4/3.4-5 ~. 3.9-6 3.7-7 1.7-7 FGS = Fire osowth stage lc = Transient comtmastitile Tosa annual cose melt frequency - 4,4_6 l FJote: Because of the evaluation of event E, the prob.tuhty of event C is no included in the evalualson of the sequeence fra:que:ncy. Figure A.6 Fir 60rowth event tree for fire zone 47 q 1
e.- t B-1 APPENDIX B: Report of Professor Alan L. Kafka: A Critique of " Seismic Ground Motion at Limerick Generating Station," by ERTEC Rocky Mountain, Inc. INTRODUCTION Although no theory has yet been developed that ' explains the cause of earthquakes in the eastern United States, seismologists and engineers. are still called upon to assess earthquake hazards in,this region. As the trends of urbanization and industrial 12ation spread threughout the East, the number of requests for earthquake hazards assessments increases. Seismologists must, therefore, respond to the need for a technical evaluation of the current state of knowledge of earthquake processes at a given site, while also te=pering hazard assessments with-clearly expressed ad=issions of their interent limitations. Thus, in the assessment of earthquake hazards at sites located in the East, two key issues emerge: (1) A realistic assessment must emphasize that there is no deterministic model that describes the cause of earthquakes in the Eastern United States in general, or (certainly in most cases) at the site in particular. (2) It is nevertheless incumbent upon seismologists to provide a practical guide for siting critical facilities that incorporates the ~ present state of knowledge in the field. " Seismic Ground Motion at Limerick Generating Station," a report prepared by ERTEC Rocky Mountain, Inc., can be evaluated in the light of these two issues. ' On the one hand, the report fails to state explicitly that very little is known about the cause of earthquakas in the East in general or at the Limerick site in particular. On the other hand, despite this significant
- =...,
i-B-2 omission plus a number of technical problems, the results contained within the 4. report can still be of practical value in the assessment of the seismic ha:ard at the Limerick Generating Station. In particular, the results shown in Figure 9 of the ERTIC report for the " Decollement" hypothesis probably yield a conservative esti= ate of seismic ground motion at the site. This conclusion is ironic, since " Decollement" is ,possibly the ar t speculative.of the feur hypotheses considered. Nccetheless, the practical. application of " Decollement" is ultimately useful, since its-essential feature.(as far as the calculated seismic hazard is concerned) is 3 that it creats the entire eastern seaboard as one seismogenic zene. This allows for the possibility that large earthquakes (M=7) could occur anywhere in that area. The inclusion of calculations of seismic hazard resulting from the other three hyyotheses on seismogenic zonation (Piedmont, Northeast Tectonic Zones, and Ceustal Blocks) also provides insight into the seismic hazard at the Limerick site. The peak ground acceleration curves shown in Figure 9 for all four zonatio:r models illustrate that a very wide range of hazard assessments l results from the lack of knowledge of the cause of earthquakes in this region. Nonetheless' it is useful, from a practical point of view, to know how sensitive the esulting hazard evaluation is to changes in the geometry of seismogenic zones. While these practical results can be gleaned from the ERTEC report, Section.3 (Seismogenic Zones) and Section 4 (Seismicity Parameters) contain a number of technical problems. Also, there is insufficient information in the report regarding the earthquake catalogues used in the study. These issues are discussed below.
e,.. ? B-3 SEISMCGENIC ZONES Section 3 of the ERTEC report describes the seismogenic zones used in the hazard analysis. In this section, seismogenic zone is defined as "[a :ene]... within which earthquakes are considered to be of sintlar tectonic ori.;in so that future seismic events can be modelled by a single function describing earthquake occurrences in time, space, and size." It is important to note that since the tectonic origin of all earthquakes along the entire eastern seaboard is at present unkncvn, all of the hypothe;ized seismogenic zones discussed in the ERTEC report are highly speculative. The report does not mention this fact. Some fundamental problems with the two more recently proposed hypotheses are discussed below. Decollement: This hypothesis is based on an analysis of intensities reported fer the 1886 earthquake in Charleston, SC (Seeber and Armbruster, 1981) coupled with results of seismic reflection studies of the deep crustal structure of the southern Appalachians (Cook et al., 1979). The seismic reflection profiles have revealed a continous shallow-dipping reflector beneath the southern . Appalachians that has been interpreted to be a maj or decollement. The inferred decollement has been proposed as the boundary of a seismically distinct block of the earth's crust, i.e. the " Appalachian Detachment" (Seeber and Armbruster, 1981). Historical eartNuake catalogues for the eastern United States (e.g. Barstow at al., 1980) show a rather low level of seismicity in the Charleston area, and the recent monitoring of the area cith a dense seismograph network 7 w 7-e sp- --F
- --*-~?'r
B-4 has also revealed a relatively low level of activity. Thus, studies of microearthquake distribution, fault-plane solutions, and earthquake depth have not been very abundant in this region (Hamilton, 1981). The hypothesis that .the current seismicity in the vicinity of Charleston, SC is occurring along a major decollement surface is, therefore, not well supported by quantitative . geophysical studies. The existence of an " Appalachian Detachment" should thus , be considered as interesting speculation, but speculation ncnetheless. Furthermore, although preliminary results from deep seismic reflection profiles in the northern Appalachians (e.g. Ando et al., 1981; Brown et al., 1982). have also revealed shallow-dipping reflectors, the lateral extent of these surfaces in the northeast does not appear to be as great as in the southern Appalachians. Thus, even if " decollement tectonics" were applicable to earthoiakes in the southern Appalachians, I have seen no convincing evidence to oggest that this hypothesis should be applicable in the northern Appalachians in general or in the vicinity of the Limerick site in particular. Figure 6 of the ERTEC report shows the northern boundary of the Decollement zone at about 41*N. No reason for choosing this boundary was given in the report. Crustal Blocks: According to this hypothesi', the occurrence of earthquakes in the s eastern United States is controlled by large crustal blocks. Supposedly, the boundaries of these blocks are seismically active and the interiors are relatively inactive. While this hypothesis seems reasonable in principle, and may eventually predict the locations of future large earthquakes, none of the crustal block models that have been proposed (e.g. Diment et al., 1979) correlate very well with historical or instrumentally located seismicity. mme
- 6
.j-t B-5 Lacking any definitive correlation with the only existing records of actual earthquakes, this hypothesis should be considered as interesting geophysical speculation _ worthy of further investigation, but - like the " Decollement" hypothesis - speculation nonetheless. SEISMICITY PAR.U!ETERS Seismic Activity Rate: The ERTIC report overstates to some extent the conclusions found in McGuire (1977). This is an example of how the report implies (at least in style,. if not in fact) that more is known about eastern earthquakes than really is known. My interpretation of the results of McGuire (1977) and the further studies on this topic by McGuire (1979) and McGuire and Barnhard (1981) is nor that the historical rate of activity is well determined. Rather, the value of these studies is that they show that even though the rates of activity in the East are poorly determined, a reasonable approach to hazard analysis for exposure times of chout 50 years in this region is to l assume a stationary model of the rate of seismic activity. This approach is i useful only in light of the current lack of knowledge of the cause of earthquakes in this region. Perhaps this approach should be referred to as being " reasonable" rather than " realistic" (see Table 1 of ERTEC report). i The ultimate test of such an approach to hazard assessments is, simply, the causative mechanism of earthquakes in the eastern United States. Perhaps the historical earthquake activity in China studied by McGuire (1979), for comparison with the eastern United States, was anomalously stationary due to a process that is at present unknown. Future investigators may discover that
B-6 the rate of activity in the eastern United States during the past two centuries was anomalously low or high by an order of magnitude or perhaps even more. If, for example, seismic gap theory (proposed for seismic ha:ard studies in the vicinity of plate boundaries; e.g. McCann et al., 1979) is found to be applicable to intraplate earthquakes, then there might be long periods of seismic quiesence premonitory to impending large earthquakes in ,this region. Does the rate of activity observed for the past 200 years in the East represent an intraplate variation of a seismic gap, or is this rate a result of many years of af tershocks of a large earthquake such as the New Madrid event of 1811? Such questions can not be answered without a deter =inistic model of the cause of earthquakes in the East. Maximum Magnitude: It is not clear which hypothese.s are being referred to in the ERTEC report. that restrict the recurrence of Cape Ann. Massachusetts type earthquakes to areas in New England; the author should have cited some references. I suspect, however, that the author is referring to an apparent association between the northwest-southeast trend of seismicity in this region, and a landward extension of the New England seamounts that was discussed by Diment et al. (1972), Sbar and Sykes (1973), and Fletcher et al. (1973). This trend crosses tha Ottava-Bonnechere graben and Mesozoic intrusions. that postdace 'the initial separation of North America from Africa (Cykes, 1978). The association between the trend of seismicity (the so-called " Boston-Ottawa seismic belt") and these tectonic features (possible candidates for ancient zones of weakness reactivated by the present-day stress field) has e
~
- l. -
f B-7 been analyzed in detail by Sykes (1978). Further analysis of the correlation by Yang and Aggarwal (1981) showed that there are a nu=ber of reasons to question the existence of such a seismic belt. The monitoring of earthquakes by a dense microcarthquake network in the northeastern United States reveals a gap in the Boston-Ottawa trend that goes through Vermont (Yang and Aggarwal, 1981). This gap (although not as distinct) can also be seen in the historical record of seismicity (e.g. Chiburts, 1981). .In addition, the pattern of crustal stress in this region appears to be different to the southeast of Ver=ent than to the northwest (e.g. Yang and Aggarwal, 1981). This observation suggests that earthquake processes =af be different in the cluster of seismicity that lies to the southeast of Ver=ent than it is in the northwestern part of the Boston-Ottawa trend. There is, therefore, no convincing geophysical evidence to support the existence of a Boston-Octava seismic belt within which earthquakes are of similar tectonic or'igin. Hence, I.;ee no reason to exclude earthquakes near Cape Ann Massachusetts from the Piedmont region. If the 1982 earthquake in New Brunswick, Canada is, to be included in this province, as stated in the ERTEC report, then certainly earthquakes that occurred near Cape Ann should be. LARGE EARTEQUAKES Nf.AR THE LIMERICK SITE Appendix B of the ERTEC report discusses the credibility of hypotheses that allow an earthquake of the size of the 1886 Charleston event to occur in the vicinity of the Limerick generating station. As stated in Appendix B, calculations of the hazard at the sic 2 are sensitive to the subjective I
~ r B-8 probability assigned to such hypotheses. In the main report a subjective probability of ten percent was assigned to the " Decollement" hypothesis, and this hypothesis can be considered to be representative of any hypothesis that treats the entire eastern seaboard as one seiscogenic zone, thus allowing for an earthquake the si=e of the Charleston event to occur at the L1:erick site. Since no explanation has been found for the cause of the 1886 Charlesten earthquake, there is no particular reason to excluda such an event fro: anywhere along the eastern seaboard. Thus, a probability of ten percent may be an underesti= ate for the credibility of tectonic hypotheses which wculd allow a large earthquake (v.= 7 ) in the next 50 years in eastern Pennsylvania. Perhaps the twenty-five to thirty percent probability for the scientific credibility of such an hypothesis (as suggested by at least one of the esperts consulted in Appendix B) is not unreasonable. Also, in evaluating Appendix 3, it would be useful to know the distribution of responses on this issue: '.e. i how many of the experts assigned a high probability (25-30%), and how =any a low probability (0%) to the credibility of such an hypothesis? EARTHQUAKE CATALOGUES There is no mention in the ERTEC report of the fact that there may be a bias in the distribution of seismicity shown in Figure 1 due to ince=plete reporting and/or recording of events. While the lower bound of g=4.5 0 01 intensity V-VI) that was used for the part of the study estimating seistic ground motion seems appropriate, it is not clear to what extent the w
.o e B-9 incompleteness of catalogues for smaller events could effect other parts of the study. ' Incomplete reporting could, for example, have an effect on the various studies' of determination of seismogenic zones. The report states : hat, consistent with the level of effort available for this study, it relies heavily on the work of others (p.1). This approach is justified, and a serious evaluation of the completeness of the catalogues used is justifiably beyond the scope of the study. Nonetheless, the report should state that completeness of catalogues could be a problem. This omission, again, creates an impression that the phenomenon of eastern United States earthquakes is better understood than it really is.
SUMMARY
AND CONCLUSIONS The general. writing style of " Seismic Ground Motion at the Limerick Generating Station," a report prepared by ERTEC Rocky Mountain, Inc., gives an unrealistic impression that more is known about earthquakes in the eastern United States than really is known. For example, the report relies heavily on the concept of seismogeni~c zones "within which earthquakes are considered to be of similar tectonic origin," but fails to state explicitly that the " tectonic origin" of all earthquakes along the entire eastern seaboard remains a mystery. Also, the following technical problems have been found with the report:
- The conclusion derived from. studies by McGuire (1977), McGuire (1979),
.-..n
% g, B-10 and McGuire and Barnhard (1981) that the rate of seismic activity in the eastern United States is well determined is, at least to some extent, overstated. ' ^ Earthquakes near Cape Ann, Massachusetts are assumed to be excluded from the " Piedmont" seismogenic zone, and there is no convincing geophysical evidence to support this assu=ption. A subjective probability of ten percent was assigned to the credibility of any hypothesis that allows an earthquake the size of the 1886 Charleston event to occur in eastern Pennsylvania. This probability is rather low, and a twenty-five to thirty percent prc.bability - suggested by at least one of the experts consulted in Appendix B - is not unreasonable. There is no mention in the report of the fact that there may be a bias in the distribution of seismicity shown in Figure 1 due to incomplete reporting and/or recording of earthquakes. Despite these significant problems, the results contained in the ERTEC report can still be of practical value. The peak ground motion curves (shewn in Figure 9 of the report) for all seismogenic zonation models are of practical value since they illustrate the very wide range of hazard assessments that result from the lack of kaowledge of the cause of earthquakes
- in the East.
In assessing the seismic hazard it is useful to know how sensitive the resulting hazard evaluation is to changes in the geometry of I _s O
f B-11 seismogenic zones. This is 'particularly true in cases like the East, where all zonation models are very speculative. ~ The results shown in Figure 9 for the " Deco 11ement" hypothesis probably yield a conservative estimate of the seismic ground motion at the Limerick site. This conclusion is, ironic, since " Decollement" is possibly the cost speculative of the four hypotheses considered. Nonetheless, the practical application of " Decollement" is ultimately useful, since its essential feature i (as far as calculated seismic hazard is concerned) is that it treats the entire eastern seaboard as one seismogenic :one. This allows for the - possibility that large earthquakes - such as the 1886 event near Charleston, SC - could occur anywhere in that area, thus resulting in a rather conservative estimate of the seismic hazard at the Limerick generating station. I i o l l l t { i i 4 I l_ ^~ +
n B-12 REFERENCES Ando, C.J., Cook, F. A., Oliver, J. E., Brown, L. S., Kauf=an, S., Klemperer, S., C:uchra, B., and Walsh, T., C0 CORP seismic reflection profiling in the New England Appalachians and i=plications for crustal geo=etry of the Appalachian Orcgen, ECS. Trans.. Am. Geophysical Union, 62, no. 45, p. 1046, 1931.
- Barstow, N.L.,
3 rill, K.G.,
- Nuttli, 0.W., and Pc=eroy, P.W, An approach to seismic zonation for siting nuclear electric pcwer generati:z facilities in the eastern U.S., USNRC, NUREG/CR-1577, 1950.
Brown, L., Ando, C., Klemperer, S., Oliver, J.E., Kauf=an, S., C:uchra. B., Walsh, T., and Isachsen, Y., Adirondack-Appalachian crustal structure: The C0 CORP Northeast Traverse, EOS. Trans.. A=. Geochvs. Union. 63, no. 18, p. 433, 1982. 1 Chiburis, E., Seismicity, recurrence rates, and regionalization of the northeastern United States and adjacent southeastern Canada, USNRC. NUREG/CR-2309, 76 pp., 1981.
- Cook, F.A., Albaugh, D.,
- Brevn, L., Kaufman, S.,
- Oliver, J., and
- Hatcher, R., Thin-skinned tectonics in the crystalline southern Appalachians: C0 CORP seismic reflection profiling of the Blue Ridge and Pied =ent, Geoloey, 7, p. 563-567, 1979.
Di=ent. W.H.,
- Muller, 0.G.,
and Lavin, P.M., Basement cectonics of New York and Pennsylvania as revealed by gravity and =agnetic studies, in Caledonides on the USA, published by Virginia Poly. Inst. and State Univ., Blacksburg, VA, 1979. Diment. W.G., Urban. T.C., and Revetta, F.A., Some geophysical anomalies i in the eastern United States, in The Nature of the Solid Earth, Ed. E.C. Robertson, p. 544-572, 1972. Fletcher, J.B., Sbar, M.L., and Sykes, L.R., Seismic trends and travel-time residuals in eastern North America and their tectonic i implications, Geol. Soc. Am. Bull., 89, p. 1656-1676, 1978.
- Hamilton, R., Geological origin of eastern U.S. seismicity, in Earthquakes and Earthouake Engineering, Eastern United States.
vol. 1, Ed. J.E. Beavers, p.3-24, 1981.
- McCann, W.R., Nishenko, S.P., Sykes, L.R.,
and Krause, J., Seismic gaps and plate tectonics: Seismic potential for major plate boundaries, l Pure and Appl. Geophys., vol. 117, p.1083-1147, 1979.
- McGuire, R.K., Effects of uncertainty in seismicity on estimates of seismic hazard for the east coast of the United States, Eull. Seis.
Soc. Am., vol. 67. no. 3,, p. 827-848, 1977. ? -] l l .L
m. \\ B-13 McGuire, R.K., Adequacy of simple probability models for calculating felt-shaking hazard using the Chinese earthquake catalog, Bull. Seis. Soc. Am., vol. 69, p.877-892, 1979. McGuire R.K., and Barnhard, T.P., Effects of temporal variation in seismicity on seismic hazard, Bull. Seis. Soc. Am. vol. 71, p. 321-334, 1981. Sbar, M.L., and Sykes, L.R., Conte =porary compressive stress and seismicity in eastern North America: An example of intraplate tectonics, Geol. Soc. Am. Bull., vol. 84, p.1861-1882, 1973. Seeber, L. and Ar=bruster, J.G., The 1886 Charleston, South Carolina earthquake and the Appalachian detachment, Journ. Geoohvs. Res.,vol.36, no. 39, p. 7874-7894, 1981.
- Sykes, L.R., Intraplate seismicity, reactivation for preexisting zones of weakness, alkaline magmatism, and other tectonism postdating continental fragmentation, Rev. Geophys. & Space Phys.. vol.16, p.621-688, 1978.
Yang, J.P. and Aggarwal, Y.P., Seismotectonics of the northeastern United States and adjacent Canada, Journ. Geophys. Res., vol.86, no. 36, p.4981-4988, 1981. ) 1 1 ...}}