ML20127C804
| ML20127C804 | |
| Person / Time | |
|---|---|
| Site: | 05200001 |
| Issue date: | 11/16/1992 |
| From: | Kelly G NRC |
| To: | Poslusny C NRC |
| References | |
| NUDOCS 9301140350 | |
| Download: ML20127C804 (8) | |
Text
.
D000 t 5~ e?- 0 0) !
flovember 16, 199?
NOTC 10: :ChetPoslu,ny) j ,,
FROM:
GiannKoly}'4,wgff i
SUBJECT:
PROPER ENCLOSURE FOR ABWR ISMIC MEMO J
friclosed is the correct enclosure to be sent to GE on how to do PRA-based seismic margins analyses.
s 1
I.
l l'
1 i
130070 if qk ,
. 9301140350.921116 e .
4 H PD91 ADOCK 05200001 a
eDn N p)Wpt[
&W
a ,
J r GUIDAf1CE ON PERf0RMANCE OF A PRA-BASED :
SEISMIC MARCINS ANALYSIS An analysis of the capability of an evolutionary or passive plant to withstand !
, seismic events beyond SSE is necessary to achieve a better understanding of ,
- (1) plant vulnerabilities to low return period, large magnitude carthquakes
! that can cause failure of structures, systems, and components (SSCs) and could lead to core damage and/or significant releases, (2) random failures and human ,
errors that in combination with seismic-related failures could lead to core damage, (3) seismic related additions to the reliability assurance program and a
llAAC, and (4) seismic +1nduced containment bypass and containment failure .
potential. The staf f, through the EPRI Evolutionary and Passive Requirements documents, the CESSAR 80+ DSER, and ongoing interaction with GE on the ABWR PRA, has indicated to the nuclear industry that a seismic PRA is not required ;
for new plant designs. IL is the staff's preference that a PRA-based margins '
approach be applied since it eliminates the need to deal with the uncertianty i of the seismic hazard curve for a site. Use of seismic hazard curves could introduce very large uncertainties, rendering seismic core damage frequency estimates (in absolute value) an obstacle to making safety decisions. No hazard curve is needed for the PkA-based margins approach, since it does not convolute hazard curves with fragilities.
In a PRA-based seismic margins analysis, a high confidence of low probability ,
of failure (HCLPf) estimate is essentially equivalent to 95% confidence that at a particular acceleration is less than a 5% probability of failure for particular structures, systems, or components (SSCs). Although the definition of HCLPr is given in mathematical terms, the concept of HCLPF is deterministic '
- and is used as a "go, no-go" criteria for a structure, system, or component.
At the HCLPF acceleration value, there is margin in the SSC's capacity to withstand a seismic event. However, beyond the HCLpf value the SSC is assumed
, to fail (i.e., its " margin" cannot be relied upon for regulatory decisions).
The concept of IICLPF was also chosen so that the HCLPF values would be sufficiently conservative, but not overly so, so that experts would not argue about whether SSCs would survive a particular seismic event given that the accelerations of the event were below the llCLPF values. While the staff believes that a remnant margin exists above the llCLPF level, it is valuable to
, continue to calculate HCLPFs in a manner that obviates arguments about the capability of equipment to survive at or below HCLPf values.
The staff intends to use plant and sequence HCLPF acceleration values in the following ways: (1) to estimate the robustness of a nuclear power plant to withstand seismic events'beyond the SSE, (?) to help identify plant seismic vulnerabilities to beyond design basis earthquakes, (3) to help: identify structures, systems, and components that should be added to the reliability assurance program, (4) to identify seismically induced . containment- bypass and l containment failure potential, and (5) to identify random f ailures and human '
errors that in conjunction with seismic-induced failures can lead to core damage. There are several steps necessary to attain these insights.
STEP 1 - Deciding Whether to Use Min / Max.or Convolution Methods to Analyze Seismic Sequences leading to Core Damage 1 :
l
- l. 3 L _ ._ _ . _ _ . _ _ . _ _ _ . _ _ _ . _ _ _ _ _ _ _ - _ . -
. _ . _ _ _ _ _ . . _ - _ _ _ _ . _ . . _ ____ .________._s Min /Mn f or the Min / Max approach, either the Conservative Deterministic f ailure Margin !
(CDfM) approach or the fragility Analysis (FA) approach can be used to !
- determine the llCLPfs of individual SSCs. In the Min / Max approach, a sequence i HCLPF is cqual to the lowest HCLpf of any cut set, where the liCLPF of each cut set is the highest HCLPF of any cut set element. In a Doolean sense, this simply means that the llCLPT of any "Af10" function is the highest HCLPF of any !
"ANDod" element and the HCLPF of any "0R" function is the lowest HCLPF of any *
"0 Red" element. for example, let there be a sequence that requires the failure of two independent systems for the plant to go to core molt. We will ,
assume that system A has a HCLPF of 0.409 and system B has a llCLPF of 0.559 .
In this exampie, (1) both systems must fail to result in core damage, (2) up !
to 0.409, both systems have available margin, (3) above 0.409, system A has l "used up" its margin, and is no longer credited, but system B still has margin i up to 0.559, (4) above 0.559 , system B has "used up" its margin, thus there is '
no margin left and the sequence liCLPF has been reached. This also directly '
provides insights into what controls the margin and how the margin can be ;
increased. The Min / Max approach assumes that SSCs with the same llCLPF are fully correlated, d
With regard to reporting combinations of seismic and random failures / human i errors, the appropriate way to report the sequence llCLPF is to provide the .
seismic only portion in combination with the random / human error portion. This ,
would be done in the following form:
("x"g) * (random probability / human error probability) 1 Separating the seismic f ailures from the random / human errors makes it possible ,
to identify SSCs that should be added to the reliability assurance program,- !
human errors that should be added to the COL applicantis training program, and l SSCs that should receive extra attention during seismic walkdowns. This i method of reporting also helps to identify potential vulnerabilities to seismic events and equipment to be added to ITAAC.
When reporting seismic / random combinations, do not report combinations involving random failure of cut set elements that dominate the seismic only HCLPF. For example, take the following case: .t Cut set - A
- B Element "A" - 0.5g HCLPF or 0.01 random i Element "B" 0.39 HCLPF or 0.01 random -j There are three combinations involving seismic failure, with the resulting HCLPF:
A,
- B,* - max (0. 5g , 0. 39 ) - O . 59 A, B, - (0.59) * (0.01)-
B, .
- A; - (0.39) * (0.01) 2 ag -9 T v- p TMr'- 'TP-*E'*rW1* M 897Y Y '*8' * * *
- 7Fd*d P9 "*Pf* W3*#r--t'1 E"""'*?'9 PY'4'T'WT
't"#C"8'W P9'"M'T"M'#" * ~~*''*#"*-**W"W^~'""T' ' ' ' ' ' * * *WW 'W* '
)
Oniy the third combination is meaningful in addition to the first (seismic l only) combination. It suggests that a lower llCLpf is possible if a certain i random event occurs. The second combination has the same seismic HCt.Pf as the seismic only combinatlon, and is thus "non-minimal". Only report 1 seismic / random / human error combinatlons where the random or human contrilution ;
is 0.001 or greater. ;
Ctn W M iDu i When we speak of convolution, we are referring to the mathematical combination of fragility curves representing several systems, structures, or components.
In particular, we are Interested in combining the fragility curves (actually -
the entire family of fragility curves for each particular SSC) for elements of sequences that lead to core melt, in addition, we wish to determine the overall plant fragility curve. As with the Min / Max approach, design-specific information should be used to determine the individual fragility curves for SSCs. Note: the CDfM methodology cannot be used to input information to ;
convolution analyses.
From the pRA-based event trees and fault trees, the analyst will have identified sequences (including random failures and human errors) that lead to core damage following a seismic event, lhe analyst must identify and justify dependent failures / correlated failures, lhe sequences leading to core damage are represented in the form of Boolean expressions.
Using the symbols "U" and "n" for *0R" and "AND" operations, a typical core melt accident sequence is given as follows:
CM - 3 0 4 U l n [6 U 1 0 8 U ((2 U 15) n ($ V 16)} U (12 n 14)) 1 where the numbers represent different components.
Although Boolean equations can consist of both seismic induced component failures and non-seismic unavailabilities of components as a result of random failures, maintenance outages, or operator errors, the staff wants non-seismic unavailabilities reported separately. If these unavailabilities are commingled with the seismic-induced failures,.it_will be difficult if not '
impossible to use the results to provide insights for the reliability ,
assurance program, operator training, seismic walkdowns, and ITAAC. '
Therefore, when reporting the HCLPT for a sequence or plant, it should be reported in the following form:
("x"g) * (random probability / human error probability)
Once the. core damage accident sequences are identified, the family of sequence level fragility curves is evaluated by cnmbining the component fragility curves according to the Boolean expression for-the accident sequence.
Assuming that each component.has "n" fragility curves, with'specified-probabilities, the procedure consists of performing the required _ operation-(union or intersection) on-two components at a time, for each of the n l 3
=-- - . - - - . . . _ = - = . - - - - . - =- --
fragility curves. When the uncertainties in componentsareindependent,thisresultsinn}hemedianfragilitiesoftwo fragility curves. If the i
. median fragility uncertainties are perfectly correlated, only n fragility curves result. In either case, the final n fragility curves of the combined l event are then combined with the n curves of another component. lhis process is continued until all the component fragilities have been combined as '
specified by the Boolean expression, finally resulting in n sequence level curves. Assumptions about the independence or correlation of median fragility uncertainties should be stated and justified. The convolved sequence level fragility should also be reported.
Consider two components A and B, each with n fragility curves and respective probabilities p, (1 - 1, . . . .n) and q) (j - 1,...,n).
For the union, C - A V B, the fragility Feg(a) is given by fco(a) - fu(a) + f,j(a) - [f,i(a)
- f,;(a)] (1.0) where subscripts i and j indicate one of the n fragility curves for the components. The probability p g associated with fragility curve f c a) is t given by p, qj if the median capacities of A and B are independent.g(For !
perfectly correlated median capacities of A and B, P y is O for i/J and is max The result of the intersection term in equation 1.0 is f[pi(,a )q;*) for i-j.
(fiu(a) *F,3(a) t,)(a)] whenwhen theit randomness in the two events is independent and min is perfectly correlated. 1 For the intersection, D A n B, of two components A and B, the fragility .
F,q(a) is given by f ou(a) - Fu(a)
- F,3(a) (1.1) and is evaluated as described earlier, lhe probability p g is given by p q if the median capacities of A and B are independent and min [pi, q;) if the,se) are perfectly correlated for i-j and 0 otherwise.
For the independent case, the ni curves are condensed to n curves by sorting the failure frequencies feu(a), at each acceleration level considered, in ascending order [Kaplan and Lin, 1987). Starting with the smallest fcu(a).
it is multiplied by its associated probability, p g. This product is then summed with the product of the probability of the next larger fcg(a) and its associated pi This is continued until the summation of the pq's is equal to the first prob. ability level pi, in general, the summation of p s will'not In such cases an interpolation is performed.g'The procedure exactlyequalp[.
is continued un il all the n failure frequencies with associated probabilities-pi (i = 1, . . . .n) are computed. The entire procedure is applied to all acceleration levels considered, finally resulting in n fragility curves. >
STEP 2 - Deciding Whether to Use fragility Analysis or C0FM Methods to Estimate Component / Structure HCLPfs 4
1 .. _ . _ . - . _ c_ _.- .f. u n ._ - .. _ _ _ _ . _.L._.__ - - . _ - - . - - M
,, .- _ _. m _ _. _ _ _ __ _ _ _._ _ . _ - - _ . _ _ _ .
5 i . j
{ ,
I i
If'a vendor intends to convolute its fragility curves when estimating sequence ,
or plant HCLPTs, it must use the fragility Analysis approach. If the Min / Mar ,
approach is used, either fragility Analysis or CDfM may be used.
f raciliidnalysis Approach. '
HCLPF values can be cbtained directly from either individual SSC fragility ,
curves or composite curves representing SSCs modeled in core damage sequences.
A particular HCtPF value is obtained from the 5% probability point on the 95%
confidence fragility curve for the SSC (or from the 95% confidence curve for 3'
the composite fragility curve). However, to get a true measure of the robustness of the plant, it is necessary to use fragility curves that.are representative of the appropriate input acceleration and floor spectra. The :
staff is concerned that wide spread use of generic fragilities, based on equipment at plants that have SSEs significantly lower than 0.39, can lead to unwarranted conservatism, incorrect identification of seismic vulnerabilities, and, in general, clouds the insights from the margins analysis. Therefore, to the extent practicable at the Design Certification stage, vendors should develop design-specific fragility curves for safety significant equipment. At t the COL stage, applicants should use procurement specifications and engineering workups of actual equipment designs to determine the appropriate fragility curves, at least for components within those sequences that have the lowest calculated HCLPF values (including non-seismic failures). Details and methods for fragility and HCLPF calculations are discussed in a number of-references, such as NUREG/CR-4334, NUREG/CR-4482, NUREG/CR-5076, and NUREG/CR-5270.
EREtlppmonLgLljCLH from ths_Cangr._yAtive Deterministig
[ailure_ Margin _LCDflu_Anpraath !
As an alternative vendors can use the Conservative Deterministic failure Margin (CDfM) method of calculating HCLPF accelerations. In the CDFM method, a set of deterministic guidelines (e.g., ground response spectra,' damping, .
material strength, and ductility) have been recommended that are used to '
calculate the HCLPF directly. The method is very similar to the design-procedures followed in the industry, except. that some of the parameter values have been liberalized. As with the fragility Analysis method, design-specific component and structure information should be used in estimating HCLPF values.-
The details of the CDFM method are given in EPRI NP-6041. The basic approach ,
is to judiciously select the parameter values of different variables (e.g., '
strength, damping, ductility,. load combination, and response analysis
-methods), taking into account the margins and uncertainties. The object is to obtain a conservative yet somewhat realistic assessment of the capacity. In Table 1 (from NUREG/CR-5270) we have summarized the CDfM approach.
l 5
STEP 3: Evaluate Whether Scismic Events Around Twice the Design SSE (g-level)
Hould Lead to Large Releases lhe proposed cont ainment designs f or evolutionary plants are similar in many ways to ; hose of current operating PWRs and OWRs. Seismic PRAs performed to date have shown that containment structures are quite rugged, often capable of "
withstanding accelerations in excess of ?.0 9. Areas where cont ainnient integrity has been a concern for earthquakes exceeding the design basis involve the potent ial for containment bypass and containnent isolation failure. lo address these issues, GE should evaluate the isolation capabilities of the ASWR for seismic events at twice the SSE using a PRA-based margins approach.
PRA-lMed.llE9 ids Analnh. for Evalualing ,
ConultliertL11glation_pLRyNss far t h.e Alj}@ -
lhe following added steps should te used to address whether seismic accelerations at around twice the SSE (design spectrum curve) have the potential, in conjunction with seismic-induced core damage, to lead to a large release. Use of t his method will provide somewhat anservative results that are more realistic than those provided by deterministic evaluations.
A. Identify each cutset whose HELPF capacity is less than twice the SSE (i.e., less than 0.6 9) considering random failures and using the min / max approach. If there are none, the methodology stops.
B. Given there are cutsets identified in Step A, examine active aad passive ,
systems and components irrportant to containment isolation whose failure '
would lead to an unscrubbed release in order to identify if any have HCLPF values below 0.69 .
C Given there are cut sets identified in Step A, exnine the ruggedness of potential containn'ent bypass paths to identify if any have HCLPF values below 0.69 . -
D. Report any systems or components identified in Steps B or C and discuss the potential effects associated with the combination of sequences identified in Step A with potential containmer.t bypass / isolation failure areas identified in Steps B and C.
Reference:
Kaplan, S. and J.C. tin, "An Improved Condensation Procedure in Discrete Probability Dist ribut ion Calculations", Risk Analysis, Vol . 7, No.1,1987.
6
\
- e 4
l l
1
Table !
GUh0.tARY OF CONSERVATIVE DETERMINISTIC FAILURE MARGIN APPROACH Load Combination: Normal + Seismic Margin Earthquake Ground Response Spectrum: Conservatively speelfled (preferably 84% Non.
Exceedance Probability Site Specific Spectrum, j ,
if available)
Damping: Conservative estimate of median damping 5tructural Model: Best Estimate (Median) + Uncertainty Variation in frequency So't. Structure Interaction: Best Estimate (Median) + Parameter Verlation Material Strength: Code Speelfied minimum strength or 95%
caccedance actual strength jf test data are kvallable.
Static Capacity Equations: Code ultimate strength (ACl), maximum strength (AISC), Service Level D ( ASME), or functional limits. If test data are available to demonstrate excessive conservatism of code equations, then use 84% exceedance of test data for capacity eq u a tion.
Inelastic Energy Absorption: For non brittle fallu.e modes and linear (ductility) "
analysis, use 80% <( computed seismic stress In ,
capacity evaluar:on to account for ductility benefits, or pr. form nonlinear analysis and go
- to 95% cxceedance ductility levels, in structure (Floor) Spectra Use frequency shifting rather than peak Generation: broadening to account for uncertainty plus use median demping.
t 9
m
, . -- ,- - , , - . , . - , - - - .