ML19309G202
| ML19309G202 | |
| Person / Time | |
|---|---|
| Site: | Vallecitos File:GEH Hitachi icon.png |
| Issue date: | 04/16/1980 |
| From: | Reid R Office of Nuclear Reactor Regulation |
| To: | Darmitzel R GENERAL ELECTRIC CO. |
| References | |
| NUDOCS 8005050125 | |
| Download: ML19309G202 (3) | |
Text
80 05050 [ $
/gP
.m.
,,[
y h >.
'o, UNITED STATES g
NUCLEAR REGULATORY COMMISSION 5
C WASHINGTON, D. C. 20555 t
o
'%, '...../
April 16,1980 Docket No. 50-70 Mr. R. W. Darmitzel, Manager Irradiation Processing Product Section Vallecitos Nuclear Center General Electric Company P. O. Box 460 Pleasanton, California 94566
Dear fir. Darmitzel:
We have reviewed your March 12 and 17,1980 reports regarding the Probability Analysis for GETR and have determined that the additional information requested in the enclosure is necessary to complete our review. Please provide your written discussion of each of these items as soon as possible.
Sincerely, 1.Q
) <d Ir/' din l Robert W. Reid, Chief Operating Reactors Branch #4 Division of Operating Reactors
Enclosure:
Request for Additional Information C
L
Enclosure QUESTIONS AND C0!HENTS ON " ADDITIONAL PROBABILITY ANALYSES OF SURFACE RUPTURE OFFSET BENEATH REACTOR BUILDING GENERAL ELECTRIC TEST REACTOR" Pace Comment If the first term P P
P is essentially 1-1 thesameasinthehev$5dgNan$$hkswhyisn'tthefinal answer the previously reported result plus the second tem ?
l-1
" Simultaneous" is a wrong word, "related" or
" casually-related" would be better.
1-1 The " conservative" effect of simply adding the probabilities is negligible.
2-3 Why isn't PBS/07{ modelled by the Poisson?
2-4 The nomal distribution can of course always be truncated so that only non-negative values are represented.
3-1 What are the associated confidence intervals for the best estimate ofrf obtained from Sieh's trench evaluations?
Ithought(gwasassumed?
4-2 4-3 Instead of t why not use some other given time period and idok at T amd N occurring during this x
time period?
4-4 Why don't you treat N as a parameter instead of using only one value? Approach 1 seems to be a hodgepodge of class'ical and Bayesian approaches-why not do the problem in a consistent manner using first classical then Bayesian techniques.
Fo r example in the classical, N would be treated as a parameter - in the Bayesian a prior muld be assigned
~
as done. By using only one value for N it seems as if you're trying to hide the potentially large variations possible for N.
4-4 The geometric distribution is kind of a strange distribution to use as a prior for N, for example the most probable value is zem.
Why not use a negative binominal, binominal (if we can bound N), or some other family which allows J-sheped plus unimodal behavior?
2 Enclosure 4-5 For Approach 2, why not also develop consistent Bayesian and classical approaches and compare?
For example in the classical, sensitivity studies would be performed treating t as a parameter.
4-6 What does " maximum" bounds mean?
4-8 You really should do more sensitivity studies -
letting parameters vary more than shown, varying additional parameters (particularly N) and varying parameters two at a time.
4-9 You again should really increase the amount of sensitivity studies performed.
O
%