ML19340E337
| ML19340E337 | |
| Person / Time | |
|---|---|
| Issue date: | 12/10/1980 |
| From: | Sege G NRC OFFICE OF POLICY EVALUATIONS (OPE) |
| To: | NRC OFFICE OF POLICY EVALUATIONS (OPE) |
| References | |
| FRN-45FR71023, RULE-PR-50 NUDOCS 8101140074 | |
| Download: ML19340E337 (3) | |
Text
'
'lW
(
08
,T Y
.nnea.
g cCWU
',\\
UNITED STATES OD
'WC NUG11 EAR REGULATORY COMMISSION 2
e y Qrt 10180 > pj r.
wismacroa. o. c.2ases
\\, e,,,<,d 0'Ruf Se Eg,7 NRC PUBLIC DOCUMEWT RWM 5/
December 10, 1980 DMn s l
3xch C
b i j -a MEMORANDUM FOR:
File Y'
Geor;;e Sege [
]
FROM:
SUBJECT:
DISCUSSION WITH PROFESSOR NORMAN RASMUSSEN CONCERNING NRC SAFETY GOAL Professor Norman C. Rasmussen, Department of Nuclear Engineer, M.I.T., was interviewed on November 17, 1980, by Edward Hanrahan and George Sege, to gain the benefit of his current personal thoughts and views.
Dr. Rasmussen noted three problems that he considered particularly important for NRC to confront in establishment of safety goals:
. Fix A of Problem A may have side effects that could detract from the safety gain, or even create a worse problem in some other respect. The pluses and minuses must both be considered.
. The demand for probabilistic risk assessment (PRA) exceeds the available supply of people who can do it.
If quality is allowed to suffer as a result, that may discredit the technique.
. A criterion is needed for disregarding a potential accident if its prob-abilty is low enough.
It would be quite consistent and proper for the NRC to make both of the following sorts of statement:
"There is no accident we won't look at."
"The probability of Accident X is low enough that we don't need to do anything about it."
The following are some highlights of other views expressed by Dr. Rasmussen:
. A general quantitative safety criterion in the form of a curve is not likely work.
. There may be merit in a criterion that calls for demonstration that some standard (previously found practical) is in fact met.
(E.g.,"Showus you're not worse than what we've licensed already.")
The British have a criterion of 10-7 failure likelihood per use for aircraft automatic landing systems.
810llooo M
. The Dutch have a 10-4 annual failure probability criterion for dikes.
(This is better than before, and 10-5 would be too expensive.)
. Criteria are worthless without a way to determine if they have been achieved.
Dr. Rasmussen saw difficulty in this regard with the NRC staff's originally proposed ATWS criterion.
It was set so low that the NRC would not believe any analysis to show it had been met.
Provability could be obtained by setting criteria on individual systems, where annual failure probabilities can often be in the 10-3 to 10-2 range. Balanced goals should be established for the most important systems. There should be reasonable restraints on ratcheting.
Any overall quantitative goal should be set only tentatively; one should not be too rigid about applying it initially.
. The connon wisdom that relative risks are better known than absolute ones is only partly true and may often be false. There is a tendency to over-state the greater dependability of relative figures. Assumptions in the cases compared do not necessarily cancel.
. To deal with uncertainties, we should make the best assessment we can, and then realistically insert uncertainties. These should be required to be met at the 2-sigma level.
Realistic estimates are often difficult to get.
(E.g., fuel temperatures for ECCS purposes.)
One can put an uncertainty on any number. The high end of the prob-ability range is not hard to bound.
. Grandfathering should be justified in terms of limiting repetition of exposure to a risk, rather than on economic grounds.
. Safety goals should reflect control of both individual and aggregated social risks.
.' Regulation should be based on degree-of-safety result, rather than uniformity of requirements.
A plant closer to population should be accepted if it has compensating safety features.
. Stability of requirements in going from a construction permit to an operating license should be stressed.
However, improvements should be required when indicated by new problems; conversely, requirements should be relaxed when justified by new analyses.
Backfitting requirements should be kept to reasonable limits, with assistance of rational reliability-analysis techniques, to sort out the important items. With some 200 items to change, some significant fraction are likely to make the plant less safe, rather than safer.
3-
. The " single-failure" criterion should be abandoned:
counting failures is too crude in light of today's PRA techniques.
Professor Rasmussen called to our attention a recent M.I.T. thesis -
"A Risk Comparison Methodology for the Assessment of Acceptable Risk," by Dan Litai, dated January 1980.
1
..-----r
,-,,,.,._.n,..,.,-,.
,n.,
,,.---n,.,. -,,,.
.e
,,,,, - -,.. - -.