ML19211C556
| ML19211C556 | |
| Person / Time | |
|---|---|
| Issue date: | 02/27/1979 |
| From: | Thadani A Office of Nuclear Reactor Regulation |
| To: | Cottrell W Oak Ridge |
| Shared Package | |
| ML18025B195 | List:
|
| References | |
| FOIA-80-587, REF-GTECI-A-09, REF-GTECI-SY, TASK-A-09, TASK-A-9, TASK-OR NUDOCS 8001110678 | |
| Download: ML19211C556 (7) | |
Text
.
UNITED STATES
.c
[*'QTf. t NUCLEAR REGULATORY COMMisslON
[QQ s
j wAsnincron. o. c. 2cn:
s.w..7.f
- 5. -:
FEB 2 7 S79 c-g William B. Cottrell Nuclear Safety Journal Oak Ridge National Lab.
P. O. Box Y Oak Ridge, Tennessee 37830
Dear Mr. Cottrell:
Thank you for affording me an opportunity to review the manuscript, "An Assessment of the Frequency of Failure to Scram," submitted for publication in Nuclear Safety.
After a careful review of the proposed article, I have concluded that:
1.
The article has not considered some of the recent publications which have a bearing on the subject of this paper.
Examples: NUREG-0460, Vol. 3, December 1978.
"A First Approach of the Rare Event Problem by the Study of the Reliability of the Protective System of the Fessenheim 1 PWR Reactor," A. Carnino, et al.,
~
January 5., 1979.
In my detailed comments (see enclosure) I have identified sections of Vol. 3 to NUREGr0460 which could influence the char cterization of the NRC analyses.
2.
The presentation of the issues is adequate but incomplete and
~
possibly biased.
Again, the enclosure provides a number of recommendations which, if followed, would enhance the quality of the presentation.
3.
Finally, I consider the manuscript to be acceptable if a,
the authors present the sensitivity of the posterior to a range of priors (see General Comment and comments 9, 13 thru 19,and21);
b, the authors reasonably represent NRC studies (see coments 4 thru 8);
1736 261 800112,0
r.;r.f
'G Uilliam B. Cottrell
- FEB 2 7 E73 m
the authors represent the scram system more cocpletely in the c.
development of the prior (see coments 11 and 12).
I hope these coments are helpful in improving the quality of the proposed publication.
[
Sincerely yours, n W.ao ing srs Ashok C. Thadani
~
bcc:
R. Mattson Distribution:
RSB File t
R. Tedesco A.Thadani i
T. Novak paaannchron3 L. Abramson t
1 l
1 e
9 1736 262 gp.m.w e
w emium o
ee
Enclosure General Comments yi:
.r:,
The ma~in v'irtue of this paper is its description of.f.fie steps involved in the calculation of the posterior distribution of the frequency of RPS failure.
However, because of the arbitrary and unjustified nature of so many of the numbers which went into the calculated posterior, the reader is left with serious questions as to the validity and meaning of the calculated posterior. Except in the rather unlikely circumstance of his agreeing completely with all of the author's subjective probability estimates and judgments, the reader cannot judge the extent to which his posterior would differ from the author's.
In order to place the author's posterior into proper perspective, it is essential that the sensitivity of the posterior to the subjective probabilities and judg-ments be investigated.
Otherwise, the reader may attribute far more validity
- to the author's posterior than is warranted.
Another reason for carrying out a sensitivity study stems from the possibility that tht. choice of a " prior" has been influenced by the author's knowledge of the safety record of nuclear reactors, thus giving the " experiential data" more weight than the Bayesian analysis calls for. The range of priors considered in the sensitivity analysis should be broad enough to encompass all reasonable priors which might have been chosen if this analysis had been carried out before any nuclear
}
reactors were put into operation.
.,, Detailed Conrnents
'1.
Page 5:
~
Equations 1 and 2 are incorrectly characterized.
a.
Equation 1 is based on time-dependent model.
b.
Equation 2 is based on a combination demand-dependent model and a time-dependent model.
It leads to the nonsensical j
result that P depends on N.
c.
NUREG-0460 uses the time-dependent model.
Reconinendation: The authors should review p. II-32 of Vol. 2 of NUREG-0460 and correctly characterize the NUREG-0460 approach.
- For any reader, validity refers to the closeness of agreement of the author's posterior with his posterior.
1736 263
(" (-
[.,
2.
Page 6 above Eq. 31 to'the bottom of page 8:
U The NRC analysis was based on the theqry of confidence bounds for a' Poisson failure rate..The author's derivation is equivalent L-%
to this, Recommendation: Repsace these pages with a simple statement that the NRC results are classical confidence bounds for a Poisson failure rate.
3.
Page 9:
This explanation of the EPRI analysis is tantamount to saying that, while EPRI claimed that they were using a demand model, in reality they were using a time-dependent model.
4.
Pages 10 and 11, last paragraph on page 10:
1 It is important to note that the use of the rectification concept in itself does not result in significantly different estimates by NRC and EPRI.
In fact, NRC has accounted for rectification, testing, and improved designs to modify the scram unreliability to 3 x 10-5 (see pages 7, and D-5 thru D-8, Vol. 3, NUREG-0460).
Thus NRC did not ignore the benefits of rectification.
This value should then be compared with the EPRI estimates for one failure and for no failures (for example, see page 24 of Part 1, EPRI NP-251, August 1976). Major differences between NRC and EPRI are a.
Frequency of us,eful testing b.
Use of Naval data Synthesismodeis.
i c.
Recommendation:
Review pages II-45, Vol. 2; pages 7 and D5-D8, Vol. 3, of NUREG-0460.
Correctly represent the NRC study which was not "too pessimistic" and which did not " ignore the learning curve effect." NRC did ignore the learning curve effect as suggested by EPRI.
5.
Page 11, paragraph above Section 4.2:
If Kahl failure is excluded, then experience prior to this failure should have been excluded.
Authors should make this point for consistency.
1736 264
/
\\. (.
6.
Page 11 Section 4.2:
/
LWR-experience in their data base. Authors should n except USSR) 7.
Page 12, top paragraph:
This important difference is inadequately treated to Vol. 3 of NUREG-0460 to provide perspecti ppendix D point.
n this important 8.
~
Page 12, Section 4.3, first paragraph:
NRC describes the EPRI " Synthesis Model" on pages II 51 NRC(NRR on.
See pages) has emphatically rejected the " Square Root M 11-55 to II-58.
j
_ Recommendation:
" Fault trees" are useful, but the model i J,
i j
" fault trees."
j questionable of this paper,.at best.
(See lines
)
11, 12, 13, and 14 on page 13 f
9.
Page 15:
)
I I
be included in formulating the "The authors make the poi lL_
ata" will not not included in a f6' rmal sense, prior" distribution.
While it is authors' extensive background in the operation of nuclthe poss l
influenced their choice 'of the prior, so that th ear reactors l
data" had a double influence on the posterior e " experiential might have been dispelled by a careful discussionThis possibility of the authors' subjective estimates of the reliabilit of the source j'
various subsystems, but no such discussion was forthc y of the i
i oming.
Recommendation:
for the authors to demonstrate that their prioIt would be e by the " experiential data."
The possibility of this contaminationr has not b determine the effect of changes in the prior on s
v ty stud conclusions.
e authors'y to 10.
Pages 18 and top of 19:
should not disregard the design features, e.gP c
ve; however, one 1
most sensors are displayed in the control room, indications for i
transients separate, redundant and diverse (i In a BWR for most sensors are available.
n principle of operation) 1736 265 l
.=m-m
. = =
~
Authors should acknowledge difficulty with anc Recomendation_:
the arbitrariness of the assumptions in this section.
l.
y-The authors have omitted the signal conditioning and bistables This is a most serious error since these h,g 11.
portions of scram. portions may well be quite susceptible to common
- 12. Page 20:
Inconsistency between the statement on Scram Discharge Volume (Section 5) and the Sensors (Section 1) is hard to und Operator inadvertently-leaving the discharge volu in failure to insert any rods.
The authors should carefully study the BWR scram system and give better engineering design considerations (as Recomendation_:
claimed at the top of_page 15).
- 13. Page 20, last paragraph:
s instead of 10-57 Also, this If 10-3 is too high, why not take 10 4 was not a seems to be a clear case where the chosen va j
Examine the sensitivity of the posterior to all values of the frequency of unidentified failure modes which are Recomendation:
consistent with experience.
- 14. Page 21, first paragraph:
The authors' The conclusion does not follow from the premise.
assertion that faildre frequencies smaller than 3 x 10-4b are very 4
difficult to justify (It certainly does not imply that it is an upper is a lower bound.Accordingly, the derived value of 10-5 is a lower bound i
i bound.)
and is therefore non_ conservative.
- 15. Page 21, last line:
The fact that the lognormal is used extensively does not justify its choice in this case.
Study sensitivity to the choice of distributions (While some of the reliability data in Recomendation_:
~ her than the lognormal.
WASH-1400 is consistent with a lognormal distribut ot 1736 266 ~
( i s
16.
Page 22, second paragraph:
]..m
~
arbitrary choice.The choice of 1.5 x 10' for the 95th percentile app
- O Furthermore the phrase "in the absence of
. evidence to the contrary" sugge,sts that this choice is not a " prio
~,
scram systems have operated.one but was strongly influenced
~
17.
Page 23, last sentence in Section 5:
i The large s analysis." pread of the prior does not " result from our preceding the values for the mean and 95th percentile given b
' (10).
. (9) and
- 18. Page 25, line 4:
What does it mean to say that the number of failure lies between zero and one?
should not be glossed over by some arbitrary weightin all the study. posteriors should be exhibited as part of the sensitivity
- Instead, I
- 19. Page 25, line 6:
i The final distribution introduces still more arbitrarines this procedure, and the weights assigned to the posteriors in Fig. 3 are not even stated.
- 20. Page 26, line 5:
The definition of scram failure as the failure of five adjacent i
rods to insert has nof been used in the derivation of the prior except to show (pp.16-17) that the probability of a random independent failure is dominated by the probability of dependen
~
~
j failures.
- 21. Page 28, line 8:~
~
but what about a wider distribution or one skew the right?
1736 267 I
l
__