ML20045A953
| ML20045A953 | |
| Person / Time | |
|---|---|
| Issue date: | 03/30/1992 |
| From: | Boehnert P Advisory Committee on Reactor Safeguards |
| To: | Minners W NRC OFFICE OF NUCLEAR REGULATORY RESEARCH (RES) |
| Shared Package | |
| ML20042D089 | List:
|
| References | |
| FRN-57FR14514, REF-GTECI-B-56, REF-GTECI-EL, RULE-PR-50, TASK-B-56, TASK-OR AE06-1-036, AE6-1-36, NUDOCS 9306150382 | |
| Download: ML20045A953 (9) | |
Text
I.
AEo lo -l i
I
- pn necq o
UNITED STATES l-NUCLEAR REGULATOF.Y COMMISSION
~
n
- ,E ADVISORY COMMITTEE ON RE ACTOR SAFEGUARDS 0,
WASHINGTON, D. C. 20555
/g i
March 30, 1992 I
I i
i MEMORANDUM FOR:
W. Minners, Director, Division of Safety Issue Resolution, Office of Nuclear Regulatory Research i
FROM:
P. Boehnert, Senior Staf f Engineer, ACRS (/
SUBJECT:
TRANSMITTAL OF ACRS MEMBER H.
LEWIS' COMMENTS ON RELIABILITY OF DIESEL GENERATORS Attached are some comments provided by Dr. H. Lewis, ACRS Member, on the issue of emergency diesel generator reliability and the associated NRC proposed rule on same.
Please note that this document contains the personal views of Dr.
i Lewis and does not necessarily reflect those of the ACRS.
It,is requested that Dr. Lewis' comments be considered'in the development j
of the final version of the proposed diesel generator reliability j
rule.
Attachment:
As Stated cc:
ACRS Members l'
R. Fraley, ACRS R. Savio, ACRS ACRS Technical Staff
\\ A. Serkiz, RES J. Johnson, OCM/IS l
D. Trimble, OCM/JC 1
i 9306150392 930422 PDR PR 50 57FR14514 PDR k [2 ['
N v
l
{
DIESEL RELIABILITY
{
March 23,1992 As you all know, we have been trying for some time to seek out the statistical basis for the staff's proposed resolution of the diesel reliability question, taking into account the requirements of the station blackout rule. The staff has been struggling to find deterministic criteria for meeting a probabilistic reliability criterion in the rule, with as little success as might have been expected. (The number of failures of any given diesel are simply inadequate to determine a quantitative reliability in any reasonable length of time.) Anyway, our long search for real data ended when INPO generously provided us (through NUMARC) the i
actual start and load diesel failure data for the years 1988-1990, for the full complement of 195 diesels, asking only that the utilities not be identified by name. So, for each diesel for the three-year period, we have efforts to start, failures to start, efforts to assume load, and failures to assume load. INPO itself then divides failures by tries, to determine an "unreliability" for each diesel, and multiplies the start and load "reliabilities" together, to get an overall rating. That makes little mathematical sense, because of the small numbers of failures involved-that's been our point since the beginning. Blood from a turnip, and that sort of thing. (Of the 390 entries in the table, the largest number of failures for any diesel is 5, achieved by two diesels on start, which were started 1% and 271 times, respectively, and by one on load, on 118 tries.)
The problem, of course, is that the pure statistical fluctuations of a number like 5 are so lage that one can learn little about the actual reliability of the particular diesel, though the 93 failures to start and 128 failures to load of the whole population of diesels in that period begin to provide adequate population data. But the staff isn't interested in the population-it wants to identify the bad actors-so it has invented the various trigger ~ ~
criteria for regulatory attention, to say nothing of the infamous (and in my view arbitrary to the point of capriciousness) standard of seven consecutive starts before return to service of a suspect diesel. Anyway it seemed reasonable to stop playing Sisyphus for a while, and to ask what kind of statistical analysis is really possible with these data?
The first step in any analysis is to try the most trivial hypothesis, in this case that in fact ad one is seeing is the effect of statistical fluctuations, and that all diesels really have the same intrinsic unreliability, that for the population: 0.00432 for the start unreliability and 0.00867 for the load unreliability. (As a sanity check, if that were true, one would expect 124 diesels with perfect start records and 105 with perfect load records. The actual numbers are 133 and 111 resgectively, which isn't bad.) A further possible test of this hypothesis is 2
the x test, since x should be 194 30 for both start and load. (The 30 is approximate.)
The actual values are 258 and 239 respectively, which is a tad on the high side, but not l
bad for such an oversimplification. So, as a first cut, the approximation that all diesels are alike works pretty well, and is certainly superior to the hypothesis that the failure rate of each one, based on a few events, is a direct measure of its reliability. I would recormnend
~
l accepting that as a simple measure of the state of the diesel world, and treating special cases of gross misbehavior as such. Presumably any failure is analyzed for root cause and corrective action, as it should be. But if one feels the compulsion to squeeze the data hard.
can one do better?
l
E There is a class of non-Bayesian statistical methods one can use for this kind of problem.
though they are probably unfamiliar-I doubt that they are taught in a traditional quarter course in statistics. (I really believe that Bayesian statistics provide the only logically co.
herent statistical rrsthods around, but I'll swallow my pride here, and pander.) There will now be a bit of mathematics. For those few (if any) who may prefer to skip it, and simply trust me, the printouts at the end provide an answer. (It is not the only possible way to treat the data, but seems reasonable to me.) The essential point is that in a situation like this one, where the population has many events, but each individual has only a few the best estimate for the hidden parameters of an individual depend on a combination of what you have measured about him, and what you have meased about the population to which he belongs. The individuals regress toward the mean as tue data roll in. The method I'll use here is a version of the James-Stein estimation procedure, which has been applied. inter alia, to the question of estimating a baseball player's end-of-season batting average from his success in the first few weeks of the season. (As we all know, batters regress toward the norm as the season wears on, and the hotshots of April and May become part of the crowd in September.) The method is closely related to what has come to be known as " empirical Bayes," for reasons I do not pretend to understand.
In the tables at the end, the firct column is the number of the diesel, as assigned by INPO-they believe the table is in order of increasing reliability. The next four columns are efforts to start, failures to start, the ratio of these (which the statistically unsophisticated might call unreliability), and then the regressed best estimate of the "real" unreliability of that diesel.
The following four columns are the same as the previous four, except that they refer to efforts to load and run. The final column is the net estimated unreliability, assuming that success in starting is uncorrelated with success in loading. Everyone assumes this, possibly because it is hard to see how to do better. Note that the worst overall diesel in the gang of 195 has an estimated overall reliability of 0.976, by this method. That may not be right-we'll never know-but it's certainly lietkr than just dividing raw data for that diesel. Note that the 2
resulting x for this method is as much below expectations as the trivial assumption that all the diesels are alike is above. Neither is too far out of line. Anyway, the difference between the two models is small potatoes.
Now the method used. (We'll just describe the starting case. The loading case is identical.)
j Index the diesels with i, where i = 1..195, and call the number of tries for the i'h diesel Ng, and the failures ng. Call the actual (hidden from our mortal eyes) unreliability of the J^
diesel Si, and our estimate ofit og. Then the total number of efforts to start is N = E N.i and the total number of failures n = E ng,, where the quotient # = n/N is a pretty good estimate of the unreliability of the population. (When you add them all up, the numbers are large enough to be meaningful.) Finally we need two things: a way to estimate the unreliability of each diesel, dependent only on observed data, and a measure of how good the model is-a so-called " loss function." The procedure is to minimize the latter to make the model as good as possible, within its structure. In choosing each of these, one has to be guided by experience-this is non Bayesian statistics, which has no rules. A Bernoulli distribution is of course assumed for the underlying statistical process.
A reasonable choice for the loss function is some form of least-squares function, with which you try to make the sum of the squares of the differences between your estimate and the
I truth as small as possible. So we'll take for the loss function
~
L = { W(og - Bi)2 i
where W is some kind of weight function, chosen with what you guys call engineering i
judgment, but others might call omphaloskepsis. If W = 1, that would mean giving all i
the diesels equal weight in the sum, regardless of how often they were started. The choice W = Ng would give equal weight to each start, instead of to each diesel. One could make l
i an argument for either choice, and I ran it both ways. It made little difference in the end results, and I ended up choosing to give equal weight to each diesel.
The final decision to be made is a form for og, the estimated unreliabili,ty, which has to be a function of the observables. Given the philosophy expreued above, I chose a linear combination of the raw unreliability ni/N and the population unreliability n/N, with the i
coefficients for the linear combination chosen to minimize the loss function L, and to add up to one.
oi = a. (h) + (1 - a)- (),
where a is what we have called the regression coefEcient in the table. That finally makes the model explicit, and the rest is arithmetic. The " regression" coefficients shown at the head of the table are the contribution of the population mean unreliability n/N to the estimate for each diesel, with its raw unreliability ni/N contributing the rest. These coefficients are i
pretty near unity (the population mean) because the statistical fluctuations expected for such small numbers of individual failures are much larger than the observed spread in the population. That makes perfect sense. The staff (and INPO) procedure is equivalent to choosing these coefficients as zero, which is a very bad approximation. (Actually, the kind i
of procedure used here was invented by Stein in the late 1950s. He actually proved that the conventional procedure is never right when there are more than two members of the l
population. Three's a crowd.195 is a mob scene.)
My conclusion has three parts:
. There is no defensible statistical evidence that there are any problem diesels out there.
There may be other evidence, but it has never been mentioned.
. If there were bad diesels, they could not be identified though individual diesel failure i
data, as the staff proposes to do. (Except, of course, in egregious cases, but then you only obfuscate the issues if you bring in bad statistics.)
. The proposed Rule should be held in abeyance, according to the persuasive principle that it helps to know what is broken before starting repairs. One can probably find a l
way to grant relief from ill-considered diesel testing requirements without adopting a dumb Rule.
I don't pretend for a moment that this is the only or even the best way to analyze these data, but at least it has -ome statistical rationale going for it, and can hold its own until something better comes along. I also think (where have you heard this before?) that the data should be scrubbed by some professional statisticians, and, to the extent that this is an example, the whole package of failure analysis and risk-based maintenance might benetit from more professional attention.
(
HAL LEWIS
6G55631066 p.or 4
E ADDENDUM Since writing this and shipping it off, I have learned that the procedure used here does indeed have a venerable (thirty-five years) statistical history, and is indeed a form of " Empirical Bayes." I am delighted that that gives it a credibility that I could not myself have bestowed upon it, but apologize for not having been sure of the right terminology before now.
It reminds me of the time, many years ago, when the great theoretical physicist Paul Dirac gave a talk at some international meeting. In the question period the comparably great mathematician Hermann Weyl is said to have remarked, "You said that you would derive your results without the use of group theory, yet everything you have said follows directly from the principles of group theory." Dirac is alleged to have answered, "That isn't what I promised. I said I would derive all my results without prior knowledge of group theory."
Anyway, this is (with only a few questionable innovations) an Empirical Bayesian procedure, and such methods not only have a history, but have been successful on an impressive number of problems. In addition, I tried it out on a couple of contrived problems, to which (as the contriver) I knew the answer, and it worked far better than anything the staff has suggested.
I don't know yet why it took so long for someone to think of using I'.-it is precisely geared
.to the treatment of this kind of problem.
As it says in the. main text, this may not be the best way, but it is a step forward. I see no reason why the NRC should resist using the best analytical methods available.
start Is'ression coeff= 0.814 load regression coeff= 0.828 start load total 4
starts fail ratio unrel loads fail ratio unrel unrel 1
77 1
0 913 0.006 49 3
0.061 0.018 0.024 2
149 2
0.V13 0.006 44 2
0.045 0.015 0.021 3
124 2
0.016 0.007 94 4
0.043 0.015 0.021 4
61 1
0.016 0.007 55 2
0.036 0.013 0.020 5
145 1
0.007 0.005 118 5
0.042 0.014 G.019 6
101 1
0.010 0.005 54 2
0.037 0.014 0.019 7
102 3
0.029 0.009 119 2
0.017 0.010 0.019 8
114 0
0.000 0.004 44 2
0.045 0.015 0.018 9
153 0
0.000 0.004 45 2
0.044 0.015 0.018 10 69 3
0.043 0.012 59 0
0.000 0.007 0.019 11 77 1
0.013 0.006 67 2
0.030 0.012 0.018 t
12 144 1
0.007 0.005 58 2
0.034 0.013 0.018 13 58 0
0.000 0.004 50 2
0.040 0.014 0.018 14 95 0
0.000 0.004 75 3
0.040 0.014 0.018 15 65 0
0.000 0.004 51 2
0.039 0.014 0.017 1
16 187 4
0.021 0.007 172 3
0.017 0.010 0.018 17 196 5
0.026 0.008 151 2
0.013 0.009 0.018 18 97 2
0.021 0.007 63 1
0.016 0.010 0.017 19 249 1
0.004 0.004 62 2
0.032 0.013 0.017 20 85 2
0.024 0.008 79 1
0.013 0.009 0.017 21 103 0
0.000 0.004 84 3
0.036 0.013 0.017
't 59 1
0.017 0.007 53 1
0.019 0.010 0.017 87 2
0.023 0.008 79 1
0.013 0.009 0.017 135 0
0.000 0.004 57 2
0.035 0.013 0.017 25 71 1
0.014 0.006 47 1
0.021 0.011 0.017 26 123 0
0.000 0.004 58 2
0.034 0.013 0.017 27 139 0
0.000 0.004 88 3
0.034 0.013 0.017 28 104 2
0.019 0.007 69 1
0.014 0.010 0.017 29 122 2
0.016 0.007 60 1
0.017 0.010 0.017 30 271 5
0.018 0.007 70 1
0.014 0.010 0.017 31 86 1
0.012 0.006 48 1
0.021 0.011 0.016 32 67 1
0.015 0.006 57 1
0.018 0.010 0.016 33 95 2
0.021 0.007 90 1
0.011 0.009 0.016 34 91 0
0.000 0.004 63 2
0.032 0.013 0.016 35 299 0
0.000 0.004 129 4
0.031 0.013 0.016 36 148 0
0.000 0.004 99 3
0.030 0.012 0.016 37 81 1
0.012 0.006 63 1
0.016 0.010 0.016 38 94 0
0.000 0.004 39 1
0.026 0.012 0.015 39 84 1
0.012 0.006 76 1
0.013 0.009 0.015 40 82 1
0.012 0.006 79 1
0.013 0.009 0.015 41 106 1
0.009 0.005 67 1
0.015 0.010 0.015 42 124 0
0.000 0.004 83 2
0.024 0.011 0.015 1
43 122 0
0.000 0.004 42 1
0.024 0.011 0.015 44 88 1
0.011 0.006 83 1
0.012 0.009 0.015 45 87 2
0.023 0.008 74 0
0.000 0.007 0.015 46 144 2
0.014 0.006 110 1
0.009 0.009 0.015 47 110 1
0.009 0.005 72 1
0.014 0.010 0.015 283 0
0.000 0.004 132 3
0.023 0.011 0.015 149 1
0.007 0.005 128 2
0.016 0.010 0.015 sd 45 1
0.022 0.008 43 0
0.000 0.007 0.015 51 182 4
0.022 0.008 51 0
0.000 0.007 0.015 52 156 1
0.006 0.005 64 1
0.016 0.010 0.015 53 149 2
0.013 0.006 123 1
0.008 0.009 0.015 54 59 0
0.000 0.004 48 1
0.021 0.011 0.014
4 55 106 1
0.009 0.005 87 1
0.011 0.009 0.014 56 54 0
0.000 0.004 49 1
0.020 0.011 0.014
'7 155 1
0.006 0.005 147 2
0.014 0.010 0.014 1
129 0
0.000 0.004 106 2
0.019 0.010 0.014 a9 66 0
0.000 0.004 53 1
0.019 0.010 0.014 60 118 0
0.000 0.004 53 1
0.019 0.010 0.014 61 165 1
O eD06 0.005 79 1
0.013 0.009 0.014 62 57 1
0.018 0.007 32 0
0.000 0.007 0.014 63 58 1
0.017 0.007 57 0
0.000 0.007 0.014 64 67 0
0.000 0.004 59 1
0.017 0.010 0.014 65 131 1
0.008 0.005 108 1
0.009 0.009 0.014 66 61 0
0.000 0.004 60 1
0.017 0.010 0.014 67 134 0
0.000 0.004 60 1
0.017 0.010 0.014 68 61 1
0.016 0.007 43 0
0.000 0.007 0.014 69 138 1
0.007 0.005 110 1
0.009 0.009 0.014 70 127 2
0.016 0.006 102 0
0.000 0.007 0.014 1
71 109 0
0.000 0.004 64 1
0.016 0.010 0.013 72 212 1
0.005 0.004 190 2
0.011 0.009 0.013 73 66 1
0.015 0.006 65 0
0.000 0.007 0.013 74 77 0
0.000 0.004 67 1
0.015 0.010 0.013 75 161 0
0.000 0.004 67 1
- 0. 0.* 5 0.010 0.013 76 88 0
0.000 0.004 67 1
0.015 0.010 0.013 77 71 1
0.014 0.006 67 0
0.000 0.007 0.013 78 97 0
0.000 0.004 74 1
0.014 0.010 0.013 79 150 2
0.013 0.006 64 0
0.000 0.007 0.013 80 152 0
0.000 0.004 75 1
0.013 0.009 0.013 81 77 0
0.000 0.004 76 1
0.013 0.009 0.013 163 1
0.006 0.005 158 1
0.006 0.008 0.013 95 0
0.000 0.004 81 1
0.012 0.009 0.013
.4 247 1
0.004 0.004 126 1
0.008 0.009 0.013 l
85 223 0
0.000 0.004 87 1
0.011 0.009 0.013 86 154 0
0.000 0.004 88 1
0.011 0.009 0.013 87 90 1
0.011 0.006 59 0
0.000 0.007 0.013 88 102 0
0.000 0.004 92 1
0.011 0.009 0.013 89 126 0
0.000 0.004 100 1
0.010 0.009 0.012 l
90 102 1
0.010 0.005 96 0
0.000 0.007 0.012 91 144 0
0.000 0.004 110 1
0.009 0.009 0.012 92 136 0
0.000' O.004 111 1
0.009 0.009 0.012 93 119 0
0.000 0.004 112 1
0.009 0.009 0.012 l
94 145 0
0.000 0.004 123 1
0.008 0.009 0.012 l
95 124 1
0.008 0.005 112 0
0.000 0.007 0.012 i
96 124 1
0.008 0.005 89 0
0.000 0.007 0.012 j
97 126 1
0.008 0.005 96 0
0.000 0.007 0.012 98 126 1
0.008 0.005 95 0
0.000 0.007 0.012 99 259 0
0.000 0.004 129 1
0.008 0.009 0.012 100 153 0
0.000 0.004 133 1
0.008 0.008 0.012 101 153 1
0.007 0.005 148 0
0.000 0.007 0.012 102 156 1
0.006 0.005 127 0
0.000 0.007 0.012 103 215 0
0.000 0.004 194 1
0.005 0.008 0.012 104 227 1
0.004 0.004 62 0
0.000 0.007 0.011 105 305 1
0.003 0.004 133 0
0.000 0.007 0.011 i
106 74 0
0.000 0.004 60 0
0.000 0.007 0:011 107 121 0
0.000 0.004 100 0
0.000 0.007 0.011 74 0
0.000 0.004 74 0
0.000 0.007 0.011 96 0
0.000 0.004 82 0
0.000 0.007 0.011 110 245 0
0.000 0.004 127 0
0.000 0.007 0.011 111 67 0
0.000 0.004 60 0
0.000 0.007 0.011 112 52 0
0.000 0.004 45 0
0.000 0.007 0.011 113 203 0
0.000 0.004 133 0
0.000 0.007 0.011 114 63 0
0.000 0.004 60 0
0.000 0.007 0.011
115 120 0
0.000 0.004 64 0
0.000 0.007 0.011 116 71 0
0.000 0.004 57 0
0.000 0.007 0.011
7 49 0
0.000 0.004 46 0
0.000 0.007 0.011 1
70 0
0.000 0.004 64 0
0.000 0.007 0.011
.9 55 0
0.000 0.004 44 0
0.000 0.007 0.011 120 57 0
0.000 0.004 49 0
0.000 0.007 0.011-121 119 0
0 400 0.004 44 0
0.000 0.007 0.011 122 71 0
0.000 0.004 44 0
0.000 0.007 0.011 123 86 0
0.000 0.004 45 0
0.000 0.007 0.011 124 188 0
0.000 0.004 58 0
0.000 0.007 0.011 125 74 0
0.000 0.004 44 0
0.000 0.007 0.011 126 138 0
0.000 0.004 54 0
0.000 0.007 0.011 127 70 0
0.000 0.004 38 0
0.000 0.007 0.011 128 134 0
0.000 0.004 51 0
0.000 0.007 0.011 129 128 0
0.000 0.004 87 0
0.000 0.007 0.011 130 106 0
0.000 0.004 86 0
0.000 0.007 0.011 131 52 0
0.000 0.004 52 0
0.000 0.007 0.011 132 69 0
0.000 0.004 56 0
0.000 0.007 0.011 133 51 0
0.000 0.004 41 0
0.000 0.007 0.011 134 86 0
0.000 0,004 66 0
0.000 0.007 0.011 135 114 0
0.000 0.004 87 0
0.000 0.007 0.011 136 147 0
0.000 0.004 76 0
0.000 0.007 0.011 137 53 0
0.000 0.004 47 0
0.000 0.007 0.011 138 62 0
0.000 0.004 57 0
0.000 0.007 0.011 139 46 0
0.000 0.004 43 0
0.000 0.007 0.011 140 67 0
0.000 0.004 57 0
0.000 0.007 0.011 141 66 0
0.000 0.004 55 0
0.000 0.007 0.011
~7 38 0
0.000 0.004 37 0
0.000 0.007 0.011 36 0
0.000 0.004 34 0
0.000 0.007 0.011
.4 58 0
0.000 0.004 50 0
0.000 0.007 0.011 145 67 0
'O.000 0.004 59 0
0.000 0.007 0.011 146 166 0
0.000 0.004 43 0
0.000 0.007 0.011 1
147 95 0
0.000 0.004 63 0
0.000 0.007 0.011-148 119 0
0.000 0.004 47 0
0.000 0.007 0.011 149 125 0
0.000 0.004 110 0
0.000 0.007 0.011 r
l 150 45 0
0.000 0.004 44 0
0.000 0.007 0.011 151 76 0
0.000 0.004 69 0
0.000 0.007 0.011 l
152 82 0
0.000 0.004 78 0
0.000 0.007 0.011
! 153 62 0
0.000 0.004 32 0
0.000 0.007 0.011 1 154 71 0
0.000 0.004 60 0
0.000 0.007 0.011 155 63 0
0.000 0.004 34 0
0.000 0.007 0.011 156 53 0
0.000 0.004 35 0
0.000 0.007 0.011 l 157 63 0
0.000 0.004 56 0
0.000
-0.007 0.011
! 158 110 0
0.000 0.004 58 0
0.000 0.007 0.011 159 81 0
0.000 0.004 77 0
0.000 0.007 0.011 160 107 0
0.000 0.004 39 0
0.000 0.007 0.011 161 66 0
0.000 0.004 34 0
0.000 0.007 0.011 162 87 0
0.000 0.004 86 0
0.000 0.007 0.011 163 56 0
0.000 0.004 53 0
0.000 0.007 0.011 164 53 0
0.000 0.004 53 0
0.000 0.007 0.011 165 111 0
0.000 0.004-106 0
0.000 0.007 0.011 166 115 0
0.000 0.004 110 0
0.000 0.007 0.011 167 135 0
0.000 0.004 70 0
0.000 0.007 0.011 49 0
0.000 0.004 39 0
0.000 0.007 0.011 52 0
0.000 0.004 42 0
0.000 0.007 0.011 1/0 103 0
0.000 0.004 55 0
0.000 0.007 0.011 171 95 0
0.000 0.004 60 0
0.000 0.007 0.011 172 142 0
0.000 0.004 112 0
0.000 0.007 0.011
! 173 91 0
0.000 0.004 88 0
0.000 0.007 0.011-l 174 157 0
0.000 0.004 150 0
0.000 0.007 0.011 l
~ '.
175 152 0
0.000 0.004 143 0
0.000 0.007 0.011 l
176 148 0
0.000 0.004 144 0
0.000 0.007 0.011 l
177 74 0
0.000 0.004 46 0
0.000 0.007 0.011 l
1 79 0
0.000 0.004 48 0
0.000 0.007 0.011 l
74 0
0.000 0.004 46 0
0.000 0.007 0.011 1
180 79 0
0.000 0.004 48 0
0.000 0.007 0.011 181 85 0
0.000 0.004 61 0
0.000 0.007 0.011 182 103 0
0.V00 0.004 85 0
0.000 0.007 0.011 183 79 0
0.000 0.004 65 0
0.000 0.007 0.011 184 109 0
0.000 0.004 93 0
0.000 0.007 0.011 185 137 0
0.000 0.004 130 0
0.000 0.007 0.011 186 127 0
0.000 0.004 122 0
0.000 0.007 0.011 187 65 0
0.000 0.004 51 0
0.000 0.007 0.011 188 96 0
0.000 0.004 53 0
0.000 0.007 0.011 189 85 0
0.000 0.004 81 0
0.000 0.007 0.011 190 106 0
0.000 0.004 55 0
0.000 0.007 0.011 191 95 0
0.000 0.004 70 0
0.000 0.007 0.011 192 115 0
0.000 0.004 85 0
0.000 0.007 0.011 193 133 0
0.000 0.004 90 0
0.000 0.007 0.011 194 97 0
0.000 0.004 74 0
0.000 0.007 0.011 l
195 98 0
0.000 0.004 57 0
0.000 0.007 0.011 chisq(start) = 125.0 chisq(load) 132.5
=
each chisq should be 194, with a standard deviation of about 30 1
i 9
c
-~
n e.-
w
,ea -. --.-,, - - -
w e
v-m,~-+
/?fo&-I
.'s m o
' NITED ST;.TES
(. NUC'_E 3R REGULATORY COMMISSION 3
-2 5
\\y$
j usnmcTo9 o c. cos~ s
~
e March 6, 1992 w nct x
-E
- E C T4 E T M '-
MEMORANDUM FOR:
James M. Taylor Executive Director for Operations 9
FROM:
Samuel J.
Chilk, Secretr 3
C
SUBJECT:
SECY-92-025 - RESOLUTION 07 GENERIC SAFETY
" DIESEL GENElyR RELIABILITY ISSUE B-56 The Commission (with all Commissioners agreeing) has approved issuance of the proposed amendment to the Station Blackout Rule and proposed revision of Regulatory Guide 1.9 (RG).
The Federal Recister Notice (FRN) and RG should be revised as indicated in the attachment.
Additionally, the FRN should solicit the views of the public on how the following concerns might be addressed:
1.
Recent NRC staff evaluations of operational events indicate that the unavailability of an emergency diesel generator (EDG) due to maintenance is about.04 as opposed to the.007 estimate in Regulatory Guide 1.155.
This is a significant contribution to EDG unavailability.
The trigger values were developed under the assumption that EDG unavailability from testing and maintenance is small compared to the maximum EDG failure probabilities of 0.05 or 0.025.
In recognition of the maintenance rule and the higher than anticipated unavailability contributions to EDG unavailability from maintenance, what if any provisions are required to ensure an acceptable overall reliability for the emergency AC power systen?
2.
The selected trigger values provide an unevenness between the treatment of a.95 EDG and a.975 EDG.
This unevenness can result in significantly different waiting times for detecting degradation in performance.
3.
Dr. Lewis suggested several alternative statistical approaches (e.a., Bayesian method, James-Stein SEC['CTE:
THIS SRM, SECY-92-025, AND THE VOTE SHEETS CF THE CHAIRMAN, AND CCMMISSIONERS ROGERS, CURTISS A::D 1
de PLANQUE WILL BE MADE PUBLICLY AVAILABLc
(
WORKING DAYS FROM THE DATE OF THIS SRM
~ ~ ~ ^ ~
- % 3 D0 2 87 A
af P-D'
l i
P l I
estimator, and the " Jackknife" approach) on which public comment should be obtained.
The revised FRN should be forwarded for publication in the Federal Register.
(EDO)
(SECY Suspense:
3/27/92)
The staff should present its conclusions regarding the alternatives proposed by Dr. Lewis and its responses to the questions posed for public comment in the FRN at the time that
-the proposed final rule is forwarded to the Commission for consideration.
(EDO)
(SECY Suspense:
10/30/92) l
Attachment:
l As stated l
l cc:
The Chairman i
Commissioner Rogers Commissioner Curtiss Commissioner Remick l
Commissioner de Planque l
I
~
J l
1 i
=. -
,,,..,.