ML19310A255

Responds to NRC Request for Assistance in Review of GE Probability Studies for Facility Site.Forwards Benjamin & Associates Comments & Lll 800421 Memo.Tera Corp Rept Seismic Rupture Hazard at Getr:Review & Analysis Encl

ML19310A255
Person / Time
Site:	Vallecitos File:GEH Hitachi icon.png
Issue date:	05/08/1980
From:	Bernreuter D, Tokarz F LAWRENCE LIVERMORE NATIONAL LABORATORY
To:	Eisenhut D Office of Nuclear Reactor Regulation
Shared Package
ML19310A248	List: ML19310A247 ML19310A249 ML19310A250 ML19310A251 ML19310A252 ML19310A253 ML19310A254 ML19310A255 ML19310A256
References
CON-FIN-A-0233, CON-FIN-A-233 EM80-185, NUDOCS 8006060481
Download: ML19310A255 (17)

Text

.

O.

~~

~~

~~

~R

'~

?

lPPE.*QiX ~

Ev.80-185 May 8, 1980 Mr. Darrell G. Eisenhut, Director Division of Project Management Office of Nuclear Reactor Regulation U.S. Nuclear Regulatory Ccemission Washingten, D. C. 20555

Subject:

Review of the Probability Studies Regarding the Probability of Surf ace Ruptum Beneath GETR Submitted by GC Docketed under License TR-1, Docket 50-70.

Dear Mr. Eisenhut:

Under Task 12 of the Site Specific Spectra Project (tac FIN A0233) we were requested to assist the NRC staff review GE's probability studies for the GETR site and provide independent quantification of this likelihood. The enclosed m ports:

(1) Comments en the Report by Benjamin and Associates and Reviews of their Report by D. Bernreuter.

(2) Seismic Rupture Hazard at the General Electric Test Reactor: A Review and Analysis, by TERA Corp.

(3) Memo frem R. Mensing to D. L. Bernreuter.

document our current werk under Task 12 and provides our input to your SER on GETR.

In addition, we have had several meetings with members of the Geosciences Branch to discuss these findings and other related matters.

Future effort under Task 12 is restricted to preparing for th AC'S meeting in June. However, we would be happy to provide your staff with any dditional assistance that they might require.

Sincerely,

~

Don L. Bernreuter Principal Inve tigator, AO~ (

7 T4.,N NSSP/ Reactor Sa ety Progr (

Frank J. Tokarz, Leader DLB:s ac enclosures as noted m

, 5-

.b :, y,

.m :,

.-.u :.

._-n 5g g

...g,.5 ::.. ::.,_, 3 3, 3

_,..p 80 060 6 0'f T l,

l CCMETS ON THE REPORT BY BENJA\\lIN AND ASSOCIATES i

~

AND REVIEXS OF 'mEIR REPORT j

1 by D. L. Bernreuter, Leader Engineering Geosciences Lawrence Livennore National Laboratory CVERALL CGtDTS In our review of the report,(1) Additional Probability Analyses of Surface Offset Beneath Reactor Building General Electric Test Reactor, we have fcund several areas where we Pcve disagreements with what wcs done.

First, as noted by several of the attached reviews,(24) the analyses are difficult to follow.

In many ccses, 3

little or no notavation is given - particularly with regard to the cases studied 4

and the range of values given to the parameters of their model. Secondly, there i'

are several places where the test did not agree with what was done. For examnle, i

as noted by Davis ( ) in his review (pp 3-6), setting Pg= 1 1 - (1-C)1/D*

(1)

P

~ =

BSl0N appears incorrect. Davis showed that if Pg was taken much smaller than unit in fact, equal to n/t* (t* = age of soil beneath reactor building) and a corrected expression was taken for PBS % that the net result would be the sue l

as Eq. (1). However, if in fact Pg = 1 is used then P = probability of surface 1 - (1-C)"*1 P

which is much offset beneath GETR would actually be P %

BSj0X E3!83 i

larger than the expression (1) used in the report.

For example, if n = 200 Another ex:urple is discussed on p-12 of Davis's review.( ) In this P s 41 PRBjBS.

case, the model in the Report (1) does not result in a probability that increases with increasing to (t = time since last movement on the shears).

g 4

e 2

---gg.,,

-g-

,.w,-,

7 y

3 I:'

^ -

carious raviews o f 3tatistical : eicl:Ics,'

.it:1e at:

m

.3

i'.en to the choice o f parametars used in the mcdel.

J.s noted abcre, it is (1) e0!. 0 1.._,,< - >

to tollow :, e analysis ;;rien oy Eenja.nin a_., sociates.

n u

...c.

citticult no variatica is given to the offsets of 12 and S feet used to calculate strain rate. The so-called t;per bounds on T are clearly not upper bounds nor are the x

lower bounds on t* lower bounds. Tne variations of parameters used in the model for the different cases (Tables 4-15 4-3) are also c,uesticnable - or at least it is not clear as to the choices ;rade and how they really bound the variatica in parameters because the variations in the key parr.eters sugested by the U5GS are much larger.

It may be (see below) that these changes have only minor impact - or it may tum out (as it does) that only one or two para.eters control the analysis.

It is had to detemine what is indeed important fmm Tables 4-1 a 4-3.

I=ortant Parameters coglexity of the analysis developed by Benjamin and /tssociates(1) nakes it T-difficu!: :o focus on the important asstmptions, li:aitations and meaning of the analysis.

Davis, in his review,(-) developed a siglified model which provides simpler relations between the probability of offset under GETR and the important geologic inputs than the Benja:ain 5 Associates model yet obtains the same results for the first approach uced in Ref. (1).

Davis shcwed that starting with the same relation for P given by 2

o fi).p(SlCN R3lSS i) p (3) p=

p p

+

ON BS j._

'CS B

CN w

i=1 that.

(a) for both approaches used by Benjamin 5 Associates in Ref. (1). The i)

C.N.P.,S[CN,P(BS!ON and P stay the sa e and are approxi-ated by:

terns P o

(

R3lSS

3 1

1

('

(";. ) (1 - (1-C.1 ' "* l) ; "

",,' I ~ 1 1

Pw P3.25 ; c,.

(3)

N L..

i 1

(i) 1 (l-C)

^

P (4)

BSlCN N

2. + b (5)

P

=

RBlES L-b (b)

The tems Pf,,i) change for the t'<o different approaches and are approxi.mted by:

(i) 1 3_..

(for the first approach)

(6)

P S

t and T'

(i) t (for the second approach).

(7)

P t7. '

t-s For practical purposes, the model can be made even simpler because the various parameters are inter-related.

For ex=ple, as a first si.ple approximation we can take T

x Np

- = average h.splacement % r t per event r

x S

N%T)i

.t.

2 ;r the lir.'t :1pprO;ld.7 5^1 ? r2f,

• t i_:.t' C ?!*.St n e'f r'. 5 ls "3i'".~r*d t.d:'.

N s T / (r I) x

thus, 1

1 1n (1-c)

T

- 1 (1-c) 1 t

x n

t s

s -

1 (1-c)

P,,, U. epa. se (t )

T t

n

'I.

(r t) s x

s m

thus, 2

P b+

b (i(1))

1 (1-c) p 1

n P3! Ec, t"

7,

S i=1 1

r 1_

1 1

1 er t*

(1) t(2) n'l-c) p ul e

p

+

1 P3!ES L

c s s

By sirtilar approximations it is possible to reduce the second approach to this same result.

For the second approach, ' is asst =ed, thus (2) becomes t(i) v2 -

"i 1

1 o

1 1 (1-c)

P (9)

P m

- +

t(ij-RBl BS t*

5 n

1 i=1 s

t (I) S' t N 1

o t

o s

tne tera 7

'i

. (i)"

t,_-

.s E."

but

t. N tN thus NS --

3 t

t N t t t /E o

o s o

2 2

s t

t :s s

~

i Fo r cos t cases, t /I o 1 or raller and it u uniihely that t till ;rew g

g si;;nificantly larger than E. Tc.us, (9) ceccmes approxi=t:21:. the sa e as (3).

Tnere are of course, variations.,hich could significantly alter the above analysis, however, most logical choices are going to maintain a coupling between the variables si::tilar to the above approximaticas.

These results indicate that P is insensitive to cost parameter variations -

provided that the parameters of the redel are inter-related. Tais is seen to be the case frora Tables 4-1 and 4-3.

Lcoking over the range of the variation of the key parameters used in Tables 4-1 and 4-3, it appears that many parameters were not given a vere wide variation. However, unless scmewhat arbitrary variaticas are introducel to study the effect of each para eter then the resulting P will always be abeur -he same for the codels developed in Ref. (1). The siglified rodel developed by Davis in his review could be used to study how important each parameter is.

If, however, the parameter variations are counted together then even-for hrge changes in the parameters P 1

~ + :-.

t t

S The largest contribution to the ha ard comes frc:a an undiscovered This occurs because Benja-dn and Associates (1) asstned that for ts ears shear.

y there.have been no offats off the existing shears coucled with covements on the two existing shears. This see:as incorrect. The only thing that can be said is that for t* years there have been no offsets associated with the two known shears other than on these two shears.

In other words, the N in Eq. (4) should be associated with the nu:rber of events that have occurred in t* years and not t, d

years as was used in the report. Case (3) in Table J-1 is closest to the correct value. Using the strain rates and offsets given for Case (S) we find thn: Sar shear 3-1/3-3 the effective time t_ associated with the N used in the analysis is a

~

~

6

--. - '30 ---)3, -*

- L, ~n

-ri, s

-4

-,/,

=

"3.,>

SS/1.43 x 1[ =

607000 caars t,s Using these values into Eq. ( 3 ),,;e get

-6

-6 P s 2.3 x 10 which is very close to the value 2.3 x 10 given

-6 in Table 4-1.

If t = t* = 160000 is used then P s 4.3 x 10 s

Conclusions and Recomendations

't is difficult to interpret the probability results provided in the Benjanin 5 Associates report. The report ( ) is flawed by a number of errors, both in the way the ecuations were developed and in the parameter values used to obtain :he probability values given in Tables 4-1 and 4-3. The above analysis shc.is that the combined model had several self-cancelling errors, hence, the probabi' i:y values calculated were core-or-less correct provided that the proper input values for the various parameters were used. It was also ccacluded that improper values of t and/or N etc., were used giving a probability of 3

offset lower than it sheuld be by a factor of 2 or more, depending upon the case considered.

1 The net result of the above review is that for both models P s n */nich, m

as discussed by Baecher in his reviews,( '*' is not a very useful result. There may be no way out of this trap; however, one possible approach which ni;ht shed some useful light on the problem, yet incorporate all the geological facters, is provided in outline for n in the report by TEM.Il Because geologic data indicates that the GETR sita is located vithin :-

ene of active faulting, it is difficult to disaiss the need to consider scre surfaca

7 ruptura ccupied with a : ugh, level in the safny assessmen: cf CETR.

'. 2 patica is hc.. cuch off3at and shat g value?

n approach si

ailar to the ene suggested by TEP.1f#) c:uld be used to davelop a

~

hazard curve, which would account for all the models (CE's, USG3, etc.) at the site. The best way to use such a ha:ard curve is to develop a failure curve for GETR, conditional on the ha:ard curve then properly cc bine the two to obtain the risk cf failure at the site for either the current design or modified desi;n.

It is re y difficult to pick a single number off the ha:ard curve and use it for the analysis unless that value is on the same order as what is deemed an acceptable risk number.

It could be that the ha:ard (for all geological tcdels) is small enough so that it would be possible to arrive at an acceptable offsat/ value without resorting to a risk study, cut considerin; the differences in di : rent g2ciogical medels this might not be pass bie.

If a ri,k analysis is not be be performed, then scte value for the offset will have ta be selected.

tais is difficult to do because typical hazard curves are con: int.cus rather than step functions.

Appendix A to 10CFR100 is the only official guidance available to deal with this question. The spirit of Appendix A is to try to rea.anably bound the worse reasonable case.

In this spirit we note that cost geologic models of the site have the Vercna fault as active, henca, it is c1carly possible for surface offset to occur near GETR. Given this and the uncertainty in all paraueters, it is possible that the surface offset could occur beneath GETR. The analysis (Table 3-3 of Ref. (7)) tends to show (even in the worse case) that this is a low likelihcod event, thus, it seems reasonable i

ta select the offset and g value that should be used for the safety assessment of GETR near the mean of the data for a magnitude 6 to 6.5 eardquake. Typical

3

1ues for mean fault di plaennu ;iven in Tale 3-: af'bf. 6 s:m ; -ha:

F. 5 t o 1. 4 me t e r.,ould c e :t reascnaale range of v11ues of offset to selec:.

c:e that even if it is 23streu that all of the activity en de '.'enna faul;

~

is asstred ta occur on a shear beneath GT'l (thus no credit is tRen far activity on the other shears or for t* years of no offset between shears 31/33 and B2) the probability of one meter of offset is still on the order of

-5 5 x 10 Finally, one meter of offset seems consistent with what has been observed at the GTR site.

It should also be noted that the peak acceleratica value used to define the seismic design criteria will not occur e.t the same tine as the ma.dntzt o f fset. The peak acceleration value should be associated with a lesser value of offset. The problem is that it is very difficult to decenaine what this value of offset should be.

If it is an important design censideratien, i.e., if the affxted stn:ctures cannot be shown te he safe for a combined loading of the peak ground mot cn mu. max =m o:: set or a.cout 1 neter t,.,.en ccnsiderable additien analysis will be required to atterpt to determine what value of offser..itould be associated with the peak ground acceleration.

l l

,:--.,..,.~.: =.- =.u = 3 1.

Jack R. Eenja;: tin 5.usoc., !ac., ".idditional Probability Malysis of Surface Rt:pture Offset Eeneath Reactor Building General Electric Test Reactor," J3A-111-013-01, March 12,1930.

2.

B. J. Da'ris, " Review of the Benjamin and Asscciates Report," Memo to L.11. Wight, April 14,1930. (Appendix C of Ref. (7)).

3.

G. Baecher, Memo to L. Wight, January 24,1930, (Jppendix A of Ref. (7)).

4 G. Baecher, Letter to L. Night, April 7,1980 (Appendix B of Ref. (7)).

3.

R. W. Mensing, Memo to D. L. Bernreuter April 21, 1980.

6.

D. G. Herd and E. E. Brabb, " Faults at the General Electric Test Reactor Site Vallecitos Nuclear Center, Pleasanton, California," U.S.G.S. Admin.

Report, April 1930.

7.

TERA Corp., " Seismic Rupture Hazard at the General Electric Test Reactor:

Review and Analysis," :.'ay 1,1980.

t 1

'.I 90 20305 80-3S-119 April 21, 1930 MEMORANDUM TC:

D. L. Bernreuter PROM:

R.

• d. Mensing SU3 u'ECT : Cc ments en "Additicnal Probability Analyses of Surface Rupture Offset Beneath Reactor Building, General Electric Test Reacter" The probability, P, of a future surface rupture offset occurring beneath the reacter building during a specified ti;;e period, can be modeled as P=

P (exactly cne offset occurs and it occurs under the reactor building) +

P (exactly two offsets occur and at least ene occurs under the reacter building)

P (exactly three offsets ccour and at least one occurs under

+

building) +.

(1)

The first term en the right hand side of Eq. (1) can be written as P (offset en an undiscovered shear) ? (offset on shear between B-1/P-and 3-2/cn undiscovered shear) P (offset under the reactor building / offset on undiscovered shear between 3-1/B-3 and 3-2)

(2)

This probability was analyced in the earlier probabilistic analysis and is not discussed in detail in this report.

.].

{Q LAVA: ENC 3 L:VEPMCPE LAECP.aTCPY

_w

80-SS-119 April 21, 19SO The second term en the right hand side of Eq. (1) can be written as P (cne offset occurs on B-1/B-3 and one occurs under reactor building) + P (one offset occurs on B-2 and one occurs under reacter building) + P (two offsets occur en undiscovered shears and at least one occurs under reactor building)

(3)

The present report only analyzes the two first terms in Eq. (3).

The third term,as well as all terms (in Eq. (1) involving three er mere offsets are not considered.

It is likely that all these probabilities are negligible and hence can be ignored.

t NOIE:

I disagree with the statement on pg. 2-2 about Eq. (2-1) being conservative since the "

product of the probabilities for offsets l

occurring on both existing shears simultaneously is not subtracted."

I believe all the computations involve the probability of exactiv one offset occurring en a shear and cne offset cccurring under the reactor building. Thus the computations in Eq. (2-1) is not conservative.

Concentrating en only the first term of Eq. (3) the second term is treated exactly the same,

P (one offset on B-1/B-3 and one offset under reactor buildirg) = P (one i

offset en B-1/B-3) P (one offset between shears / offset on B-1/B-3) P (offset under reactor building / offset between shears and one en B-1/B-3)

(

=PCN

• BS/0N

• RB/BS where the latter expression in Eq. (4) is the notation used in the report.

Consider first the P If an offset occurs between the shears, a RB/BS.

i t

l i

__. _~

t 80-SS-119 April 21, 1980

+

reasonable model for the location of tne offset is a bimodal distribution over the area between the shears with high likelihood near the shears (Fig. 1).

l f(t)

/^

e i

8 I Shear B-1/3-3 I

l Shear t

B-2 l

f i

A more censervative medel is to assume a uniform distribution over the 4

)

range between the two shears. Using this model,

]

PRB/BS = 0.058 (See pg. 2-5 of report) i With regard to the term, PBS/0N, the event that an offset occurs between the shears given an offset occurs en a shear is modeled as a Bernoulli event with probability p = P Since no such events have occurred given N BS/CN.

offsets on shear B-1/B-3, an upper c 1005 confidence limit for p in given by

) = 1 - (1-c)1/N (5) 4 This estimate is used for the value of P Thus, the value of BS/0N.

P depends on (1) the estimate of N, the number of offsets observed and BS/0N (2) the confidence level c.

Also, it can shown that i

p=-hin(1-c)

(6) 1

]

so that the value of P is inversely proportiened to N so that if N BS/0N decreases by a factor of, e.g.,

4, P nereases by a factor of 4.

i BS/0N j

l The remaininc, term on the right hand side of Eq. (4), P s evaluated CN, based on a model of the distribution of time, T, between offsets. In particular, if Q(t) = P(T > t), then P (T < t + T/T > t) = 1 - P (T > t + T/T > t)

(7) 1

= 1 Oft

• T)

Q(t) l

80-33-119 4-April 21, 1980 It is this conditional probability that is used to estimate P,3, i.e.,

t PON = P ( T < t + : /T > to) o where t is the estimated time since the last offset.

o NOTE:

Although it is unclear frem the report just what value of T is used, I hypothesize T =1 (yr) because the probabilisties are discussed as if they are yearly rates. Both a Weibull and a normal model for T are considered.

With regard to the probabilistic analyses, I have the following comments:

1.

The strain rate, r, used to estinate N has a significant influence on the resulting probabiliti es.

For example, (a)

If r = 1.34 x 10-f t/yr (value used in repor-t)

Using Weibull model, T = 210 ft, T = 1,567,164 and N = 221 3

then PB3/CN =.0104 (b)

If r = 5.3 x 10- f t/yr (value based on best estimate) 388,889 and N = 55 Using Weibull model, T = 210, T

=

3 Then PBS/0N *

• 19 The value of r used in the report is based on the lower bound of the total accumulated eff ect on existing shears (Table 3 2 pg. 3 5) which is the i

least conservative, i.e.,

provides the minimum strain rate. The value of r used in (b) above, is based on the best estimate of total accumulative offset (Table 3.2, p. 3.5), thus resulting in a better estimate of the strain rate.

This is one place I feel the report took an optimistic rather than conservative view.

2.

The level of confidence in estimating p(c =.9) also aff ects the results.

l A more conservative estimate of an upper cound on p would be baseo on l

=

30-SS-119 April 21, 1980 e =.99 or 99% confidence upper bound.

Ihis results in P

• UE BS/0N doubled. Thus, for r = 5.4 x 10 c =.99 PBS/0N =.0837 Combining the results of (1) and (2) gives a probability contributicn due

-6 to an offset en B-1/B-3 of 0.975 x 10 3

Another point of lack of conservatism in the report occurs in Eq. (7) where the probability of an offset ccourring en a shear is based on only one year in the future. This would be fine if a " random" occurrence (Poissen) model is proposed for the occurrence of offsets.

But the report tries to allow for an increasing probability of occurrence given no occurrence fcr t years (i.e., increasing hazard function).

If cne projects a 40-year remaining life for the reactor, then an estimate of is 1 x 10-2 instead of 2.5 x 10-" (which I believe is used in the PON re port). Combining the consequences of (1) - (3), a worst case probability of an offset en B-1/B-3 and under the reactor building over the remaining life (40 years) of the reactor is 3.9 x 10-3 4

Similar results hold for the contribution from an offset on shear B-2.

In all, the overall probability, P, can be considerably greater than indicated in the report.

5.

In general, the methods and models used throughout the report tended to be conservative, thus, if the inputs (strain rate, confidence, etc.) are reasonable, the results can be expected to be reasonable. My major question about the results tend to center around the values of the inputs.

l l

I l

80-30-119 April 21, 1980 6.

There is not a lot of data available to use in the probabilistic ar.alysis. Thus, it must be recognized that (a) the models selected (Weibull, normal for ttre between offsets, etc. ) are untested and (b) any estimates (e.g. the estimates of N and P are based on very limited BS/0N data. This has to limit the value of the resulte somewhat even if a degree of conservatism has been included in the analysis.

I bcpe these cc=:ents reflect my review of the ref erenced report in a meaningful way.

If you have any questions, please call me.

R WM/DLB/sa 1406j

.