ML20058K045
| ML20058K045 | |
| Person / Time | |
|---|---|
| Issue date: | 12/29/1988 |
| From: | Bagchi G Office of Nuclear Reactor Regulation |
| To: | Kniel K NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| Shared Package | |
| ML20058K048 | List: |
| References | |
| FOIA-93-285, TASK-082, TASK-82, TASK-OR NUDOCS 8901230023 | |
| Download: ML20058K045 (6) | |
Text
-
^
[
,,j NUCLEAR REGULATORY COMMISSION e
W AsmNGTON. D. C. 20555
'ff V s
'.. [.. '
Decersber 29,198n HEMORANDUM FOR:
Karl Kniel, Chief Reactor and Plant Safety Issues Branch Division of Safety. Resolution 2
Office of Nuclear Regulatory Research FROM:
Goutam Bagchi, Chief Structural and Geosciences Branch Division of Engineering and Systems Technology
SUBJECT:
COMMENTS ON DPJFT RESOLUTION FOR GENERIC ISSUE 82 "BEYOND DESIGN BASIS ACCIDENTS IN SPENT FUEL POOL 5" ESGB has reviewed the above referenced Draft Resolution of GI-82. Our corc.ents relate primarily to seismic hazard assumptions. The Draft Regulatory Analysis relied upon an evaluation carried out by Lawrence Livermore National Laboratory (LLNL) and their consultants as reported in NUREG/CR-5176. This evaluation utilized as input seismic hazard calculations for the Yennont Yankee and Robinson sites carried out by LLNL as a part of a larger seismic hazard study funded by NRR and monitored by our branch. These calculations are still being reviewed by ESGB and we requested that prior to.their final use in the resolution of GI-82 we have the opportunity to provide comments and assure that the application has been appropriate. Our cocraents follow:
1.
The NUREG/CR-5176 evaluation and the regulatory analysis place heavy emphasis on the estimated mean annual frequency of seismically induced failure of the spent fuel pools and, therefore, the estimated mean annual frequency of exceeding different earthquake ground motion levels (seismic hazard). Ground motion was parameterized as peak ground acceleration and the mean was determined by fitting the 5th, 15th, 50th, 85th and 95th percentile hazard curves provided by the seismic hazard study to a log-normal distribution, making a judgement as to the appropriate cutoff level (99th percentile) and determining the mean of that truncated distribution (see, for example. Table 4.6.8 of the Draft Regulatory Analysis). While such an approach may be considered when the distribution (of expert input) is unknown, the seismic hazard study.
estimates a true mean of the input data without making eny such assumptions. Such a true mean was actually shown graphically in NUREG/CR-5176 in Figure 2.1 but was not used. We contacted LLNL and their consultants, informed them of our concerns and supplied them with numerical values associated with the true means for the sites in question. Their subsequent calculations are attached to this memorandum. We also supplied them to Ed Throm, the GI-82 project manager, as soon as they were received. The appropriate mean annual frequencies of failure of the Spent Fuel Pool Structure (SPP) based upon the estimated true means of the hazard study are 3.7 E-05 for Vermont
/
Yankee and 4.7 E-06 for Robinson. They are labelled in the attachment as 3p "5 G-Experts". An additional lower estimate labelled "4 G-Experts" is S*
also shown. This is based upon a sensitivity study which is discussed 3
below. The new estimates of the mean annual frequency of failure of SPP w
I are a factor of 5.5 higher for Yemont Yankee and a factor of 2.6 higher 8
for Robinson.
m gQg g
.w n m o:
>a;,,
. ~. ~
~
2-2.
According to NUREG/CR-5176 the ground motion frequencies of interest in the analyses of seismic failure of the SFP are 25 Hz vertical for Yemont Yankee and 3.8 Hz horizontal fr.? Robinson. The ground motion at'these frequencies was estimated by assuming a response spectrum consistent with the NUREG/CR-0098 median shape. The vertical was also assumed to be two thirds of the horizontal. We were subsequently infomed by LLNL that the true frtquency of interest for Vermont Yankee was 13 Hz vertical. The seismic hszard study aside from estimating peak acceleration also directly calculates the response spectral values associated with different probabilities of exceedance. These response spectral values are based upon the recorr:endations of seismic experts who are familiar with the latest ground motion models. These response spectral values are about the same at the frequency of interest for Vermont Yankee and lower at the frequency of interest for Robinson than that assumed in the evaluation of the SFP. While it is difficult to assess the impact upon the estimated frequency of. failure of the SFP without taking into account the uncertainty assumed in response spectral values in the analysis. Our initial estimate is that the frequency of failure at the SFP at Robinson may be conservative.
3.
In the Seismic Hazard study i.11 calculations were, among other inputs, r
based upon the input of 5 seismic ground motion experts. The sensitivity test shcwn in the attachment indicates that if one expert input was eliminated, the estimated mean annual frequency of f ailure of the SFP structure would '
decrease by factor of 8.8 for Vermont Yankee and 2.8 for Robinson.
In the April 1989 we will have additional estimates of seismic hazard from an ongoing utility sponscred study being carried out by EPRI. Previous comparisons at other sites suggest that while there may be some agreement betwean the LLNL and EPRI studies at the median level, based upon the difference in methoriologies and input there will be a larger difference between the mean estimates of seismic hazard. EPRI means will most likely be lower than LLNL means. The reason.for the great sensitivity of means to differences in methodology and input is due to the skewed nature of the distribution of seismic expert judgement.
In these distributions the most extreme opinion weighs very heavily when a mean is calculated. The mean, which is an arithmetic average of all the inputs, frequently exceeds the 85th percentile of all the inputs. The problem exists whether ore uses the actual mean (see the comparison between 4 and 5 expert input) or whether one assumes a legnormal distribution (see the sensitivity tests in Table 4.5.6 of the Draft Regulatory Analysis). The mean is a highly unstable estimate of seismic hazard and its use subjects the analyst to the possibility of mean based decisions being viewed as incorrect depending upon who does the analysis and when it is carried out.
It has been pointed-out that it is inappropriate to base risk decisions on other measures such as the median which are inherently more stable. Until that time when mean seismic hazard estimates become more stable or until sufficient justification is provided to allow the use of inother measure we recommend that regulatory I
aralysis greatly deemphasize the role of these estimates in the decision making process.
I l
w. u. a.c t
4 3
4.
A specific recommendation with respect to the resolution of GI-82 could be:
a.
Utilizationofallthemeanestimates(LLNLandEPRI)showsthatthe bottom line conclusions of the regulatory analysis would not be changed. A specific statement is made recognizing the wide range of uncertainty in mean estimates indicating that in this case, the bottom line conclusions are not affected by the uncertainty existing in the best available estimates of seismic hazard in the central and eastern United States, b.
Utilization of all the mean estimates (LLNL and EPRI) shows that the bottom line conclusions of the regulatory analysis depend upon which estimate of the mean seismic hazard is used. Uncertainty in seismic hazard is discussed pointing out the weaknesses in the use of the me a'n.
Probabilistic analyses based upon seismic hazard should be greatly deemphasized with a major increase in the emphasis upon engineering analysis and judgement. This is actually the preferred method irregardless of how the mean estimates effect the results.
It avoids setting a precedent that may be hard to live with in future regulatory analyses.
S.
The relationship of seismic safety to plant vintage is not very precise.
We recommend that you modify the last sentence of the third paragraph on
- p. 2-1 to read "In general the older plants are more vulnerable to seismic induced failures".
6.
The rationale behind the last paragraph on p 4-32 is not clear.
It would be helpful to see Referer:e 26 so that we may provide a more appropriate wording.
We would be happy to discuss this memorandum and its recommendations with you in greater detail.
Please contact Leon Reiter (X-20841). We understand that the Plant Systems Branch has some reservations about the conditional probability of zirconium fire.
It is suggested that a joint meeting be held to discuss all the staff coceents.
Ic j
}l s
Goutam Bagchi, Chief Structural and Geosciences Branch Division of Engineering and Systems Technology cc: w/ attachment DISTRIBUTION:
J. Craig E5GB R/F C. Thomas
-Central File e L. Shao GBagchi i
J. Richardson A. Murphy ESGE: EST ESGB: E E
LR(f(er DJen Bagchi 12/#/88 12$88 129/88
m._..
4
~
FACSIMILE 12['5/_0 8 10 h
T!Mi AND Of1E.
k L_ [h, A
FROM:
NA4C:
f.0F rngineering, Inc..
3250 Dristol Street, suito 350 Costa Mesa California 92626 USA Fax # (714) 050-9267 e
,.........-- - - ~ ~....,,,
LON O '~
70; NAMt. :
' N UE.$w,,.,,n.ysemg7 C0HPANY.
FAX A:
S o l - 4.9 2 - 0 2 5 9 NUMBEP. OT PAG [S INCLUDING COVCR PAut:
3 I
I f y o u h e v e a p ro bl em, pl e a s e c al l,__,..
at {714) 650 9299.
Co m/nt::
t 100nb/ fax 1
x.
P 8 PENT FUEL POOL RISK ANALY313 BASED ON THE USE OF MEAN St!&MIC HAZARD CURVE In the seismic risk analysis, a family of seismic hazard surves are convolved with a family of plant level seismic fragility curves to obtain a probability distributton of the frequency of occurrence of the seismic inttfated accident under study. The family of seismic hazard curves is developed in a seismic hazard study by postulating different hypotheses regarding the seismic sources, paranters of the sources and ground notion characteristics. The fastly reans a set of seismic hazard curvest each curve is developed on a particular hypothesis and a subjective probability is associated with that curve expressing the degree of belief in the hypothesis.
In the spent fuel pool risk analysis performed by LLNL for the NRC, the complete fastly of selsmic hazard curves were not available other than the hatard curves at some selected percentile values. When the selsmic hazard curves are grouped in this fashion, the specific features of the individual h424rd curves (e.g., they could intersect each other) are lost. For the purposes of obtaining approximate risk estisates, e lognorral distribution was assumed over the seismic hazard at any peak ground acceleration. Using the median and 95 percentile hazard curves, a set of discrete seismic hazard curves were developed. The lognormal distributton was cutoff at different percentile values to judge the sensitivity of the results.
Table 6.2 of HUREG/CR 5176 shows the esan seismic failure frequency for Vermont Yankee spent fuel pool structure varies from 3.8 E 06 to 3.8E 05 depending on the cutoff level chosen.
It is judged that the same kind of insight on the failure frequency could be obtained by just using the mean seismic hazard curve in the analysis. The arithmetic esan hazard curves for Yersont Yankee and H.S.
Robinson sites were provided by the NRC. They were convolved with the
+
is............
e 6.
l l
s 1
\\
l fully of seismic fragility curves for the spent fuel pool structure at t
l tach site to obtain the following results:
han Annual Freqasney of A11ure of SFP StrLcture 5 G-Experts 4 G Experts Vermont Yankse 3.7 E -05 4.2 E 06 Robinson 4.7 E -06 1.7 E -06 l
i e
f
-