ML20134C328
| ML20134C328 | |
| Person / Time | |
|---|---|
| Site: | Farley  |
| Issue date: | 11/30/1996 |
| From: | Goa F, Heasler P Battelle Memorial Institute, PACIFIC NORTHWEST NATION |
| To: | NRC (Affiliation Not Assigned) |
| Shared Package | |
| ML20134C314 | List: |
| References | |
| NUDOCS 9701310362 | |
| Download: ML20134C328 (11) | |
Text
_ _ . _ . . _ _ _
November,1996 Technical Letter Report i on PNNL Review of Farley Unit 2 Submittal on EOC Voltage Distributions, Burst Pressure Probabilities and Leak Rate Feng Gao Patrick G. Heasler I
l November 1996 Prepared for Office of Reactor Regulation U.S. Nuclear Regulatory Commission under Contract DE-AC06-76RL01830 NRC JCN E2029 Pacific Northwest National Laboratory Richland, Washington 99352 Ehetosure 2 9701310362 970130 PDR ADOCK 05000348 P PDR
November,1996 I Abstract I 4 l This report evaluates an assessment completed by Southern Nuclear Operating Company, !
for an interim voltage-based steam generator tube repair criteria applied to the Farley Unit i 2 steam generators at the end of fuel cycle 10. The submittal evaluates steam generator i integrity in terms of tube leak rates and burst probabilities as specified by NRC's Interim i Plugging Criteria (IPC).
The submittal's leak rates, burst probabilities, and EOC ve!tage distribution were com-pared to results obtained from three computer programs developed by PNNL for the prob-i abilistic assessment of these quantities. In general, PNNL's projected EOC voltage distri-butions, burst pressure probabilities and leak rate agreed well with those projected in the j Farley Unit 2 report and were all well below the NRC approved thresholds.
l 1
1
November,1996 J
l This_ technical letter report reviews the Farley Unit 2 IPC submittal [1] for cycle 11 steam generator tube inspections. The cycle 11 inspections were used to predict leak rate
- and burst probabilities in the steam generators. Computer programs developed by PNNL for j the probabilistic assessment of tube burst and leak rate were used to evaluate the calculations done in the Farley Unit 2 submittal. The three computer programs are documented in [2].
This work was performed under Task Order No. 8 of JCN E-2029.
1 EOC Voltage Distribution The model for calculating EOC voltage distribution from beginning of cycle (BOC) inspec-tion results is YEoC = (Y oC B + YCROWTH) * (1 + UNDE) (1) where Vsoc is a random variable representing a BOC voltage at an inspected tube support plate (TSP), VanowrH a random variable representing the voltage growth experienced at this location, and Uxor a random variable representing the relative uncertainty due to NDE (UsoE is in relative units M*/P). The end of cycle voltage distribution is computed by convolving the distributions of the three input variables with each other (see [2] for more details.
In the Farley Unit 2 report, the input distributions for Vsoc were available for cycle 11 for all the SGs (Table 7-2 in [1]). The input Vanowrn was also tabulated for all SGs and for the combination of the 3 SGs for both cycle 9 and cycle 10 (Table 4-6 in [1]). Since the voltage growth distribution for cycle 9 for the combination of the 3 SGs is more limiting, the voltage growth for cycle 9 for the combination of the 3 SGs was used in the projection of EOC 11's voltage distributions.
The distribution for NDE uncertainties is generated from a normal distribution with mean 0 and relative standard deviation of 12.5%. The 12.5% value can be compared to the Cycle 9 and 10 voltage growth curves, which contain information about the actual sizing error experienced during the inspections. The portions of the growth distributions that are negative must be entirely due to sizing error and can be used to estimate the sizing i error distribution. There portions of the distribution indicate that the sizing error standard deviation is about 0.1245 Volt, or a relative standard deviation of about 15%, which compares ;
favorably to the postulated relative standard deviation. l The Farley Unit 2 projected EOC 11 bobbin voltage distributions were given in Tables 7-3 in [1]. PNNL's corresponding projections together with Farley Unit 2 projections are presented in Figure 1 and Table 1. As can be seen from the results, Farley Unit 2 calculations agree well with PNNL's even though all the PNNL distributions are slightly larger than the Farley II distributions.
1 It should also be noted that the length of cycle 9 is 462 effective full power days (EFPD)
~
while the length of cycle 11 is 479 EFPD. In the calculation of EOC voltage distributions,
~
these cycle lengths were considered to be so close to a full cycle that no cycle length adjust-ment was made to account for these variations.
2
- November,1996 I
Figure 1: Distributions of projected EOC 11 voltages.
EOC11. SG A. POD =0.6 EOC11. SG 8. POD.0.6
. . /
. , /
l M:.
. M. :.
l B:. .
B:.
l 0.= l ..
l .~ = l (a) SG A (b) SG B EOC11, SG C. POD =0.6 U.
B :-
l = ::: _ l 4 ., 9 ,
3 lupuunst (c) SG C i
1 l
3 1
l 2
1 November,1996 i
Table 1: Projected Distributions for EOC 11 Voltages Quant SG A SG B SG C
-tile PNNL Farley PNNL Farley PNNL Farley 0.00 -2.40 -2.40 -2.30 -2.30 -2.30 -2.30 0.25 -0.42 -0.46 -0.52 -0.58 -0.30 -0.35 O.50 -0.08 -0.15 -0.21 -0.28 0.04 0.00 i
- 0.75 0.31 0.27 0.12 0.06 0.36 0.31 0.90 0.62 0.56 0.40 0.34 0.63 0.58 0.95 0.77 0.72 0.56 0.49 0.78 0.71
- 0.99 1.00 0.92 0.84 0.83 1.02 0.94
- .995 1.09 0.98 0.96 0.85 1.11 1.01 I 4
l 2 Burst Probability i The following relationship was used in PNNL's calculation of probability for one or more
- tube bursts for a SG at the EOC
- '
Si (oi + o2 ogio(F,)
l + E i) (2)
P,= S,,f where;
- 1. (oi,o2) are regression parameters describing the correlation between voltage and burst
] pressure data. The regression parameters have an uncertainty described by the covari-
- ance matrix Cor(oi,o2).
- 2. V iis a voltage obtained from the EOC distribution computed in the last section.
- 3. The error term, E, has a variance of o} with degrees of freedom dof, as determined by the burst pressure regression.
- 4. S, is the tube specific Bow stress and S,,f si the reference flow stress. S, is assumed to be normally distributed about the value Sm (average flow stress for this steam generator's tubes) with standard deviation mo . Sm, om, and S,,f are obtained from an industry-developed data-base.
The PNNL model simulates P., for each tube using Equation 2 and drawing from ap-propriate random distributions as described in [2]. The simulation accounts for random variation in S t, and V, as well as uncertainties in the regression that appear in or , o2 and 4
E . The value of the burst pressure is then compared to the steam line break (SLB) differ-4 ential pressure to determine if the tube would be likely to burst during a postulated SLB event.
The EPRI recommended database described in [3] was used to obtain the estimates of the regression parameters for burst pressure in Farley Unit 2 report. The database described 4
November,1996 in [4] was used to get estimates of the regression parameters for burst pressure in PNNL's Monte Carlo simulation. Therefore, PNNL's inputs were not identical to those used in Farley Unit 2 report. The input parameters used in PNNL's calculation are; (oi,0 2) = (8.220,-2.515) Cov(oi,02) = / 001975 -0 01080 S b.01080 0.b2535 / '
- dof = 64 or = 0.815 am = 3.500
- Since no flow stress parameters were available in the Farley Unit 2 report, the flow stress parameters for 7/8-inch tube listed in Table 4-1 of Westinghouse report (6} were used in PNNL's simulation. The SLB differential pressure used in PNNL's calculation of probability of burst is 2560 psi.
The numbers of indications at EOC 11 for SGs A, B and C at Farley Unit 2 were es-timated to be 73,114 and 252, respectively, (this included the POD =60% adjustment).
PNNL's projections of EOC 11 voltage distributions were obtained as discussed in the previ-ous section. Samples of 73,114 and 252 voltage values were obtained from the distributions presented in the last section for SGs A, B and C, respectively. Ten thousand simulations were performed on each set of the voltage values for SGs A, B and C to get 10,000 sets of simulated burst pressure values for each voltage value for each SG. Each set was compared to the SLB differential pressure, which was 2560 psi, to determine if any tube in that set would be likely to burst during a postulated SLB event. The results of the simulation are summarized in Table 2 and Table 3. To make comparison easier, the projected EOC 11 probabilities of burst from Table 8-1 of Farley Unit 2 report [1] were reproduced here in Table 2.
Table 2: Comparison of Pwjected EOC 11 Burst Probability Results. PNNL's Projections Were Based on 'c.),000 Simu'.ations. POD =0.6 was used.
Farley 2 Proj. Burst Prob. PNNL's Proj. Burst Prob.
SG 0 Tube Burst 21 Tube Burst 0 Tube Burst 21 Tube Burst A 1 - 2.4 x 10-5 2.4 x 10-6 1 0 B 1 - 7.3 x 10-8 7.3 x 10-8 1 0 C 1 - 1.05 x 10-4 1.05 x 10-4 1 0 __.
As one can see, no bursts occured in the 10,000 PNNE simulations, so this provides strong evidence that the actual probability of burst is below tht NRC-threshold of 1 x 10-2. The Farley II simulations all produced burst probability estimates that were on the order of 10-4 or lower, so the PNNL and Farley II results agree within sampling error. -
3 Leak Rate The total leak rate, T, is calculated by summing together the leak-rates from individual indications that have a positive voltage at EOC. Assume that L indications have a positive 5
l
- ._ . _ _ _ . . _ . _ _ - _ ~. _
i November,1996 Table 3: Summary Statistics from PNNL Projected Burst Pressure (ksi) Calculations at EOC 11 Based on 10,000 Simulations and POD =60E Statistic SG A (# ofind.=73) SG B (# ofind.=114) SG C (# ofind.=252) l
- Maximum 15.6642 15.5444 14.9713 Minimum 3.3777 2.8534 3.1335 Average 9.0806 9.2478 8.9795 Median 9.0647 9.2371 8.9615 l
l Std Deviation 1.2103 1.1641 1.1579 1
4 eddy current response voltage of 4V, i = 1,2,..., L. The total leak rate is given by:
l L !
j T = { RQ,, (3)
- i=>
) '
where & is a binary variable that describes whether or not indication i is a leaker, and Q, is the conditional leak rate of the indication. The individual indication leak rates, RQi are j assumed to be independent, and their distributions have been related to inspection results.
- The relationship between & and the inspection voltage F, is
1 Pr(& = 1) = logit(S i+ S2 logio(P)), i (4) i
- and the conditionalleak-rate Qi si determined by:
j logio(Q,) = $3 + #4 logio(Ug) + E (5) j i where
- 1. (di,#2) are regression parameters from a logistic regression of leak rate data. The l 1 estimated parameters have uncertainty described by the covariance matrix Cov(Si,32 ).
- 2. (#3,#4) are regression parameters from a regression of leak-rate on voltage, and their l uncertainty is Cov(S3, #4).
l j 3. E represents the variations of log-leak-rate about the mean. The regression produces l the standard deviation of E, or, with degrees of freedom dof.
- 4 These results are incorporated into a simulation that produces T according to the above equations. (See (2) for further details). The inputs used for the PNNL simulation are; j
I
/ 3.505 - 3.849 T (Si, #2) = (-6.872,8.325) Cov(#i, #2) = 1.
. g -3.849 4.583 j f 0.0265 0 3
(#3, #4) = (0.6555,0) Cov(#3, #4) =
0 0 j' i
dof=23 as = 0.7982 6
i
November,1996
~
The NRC database described in [3] were used to obtain the the regression parameters for both probability ofleak and conditionalleak rate in Farley Unit 2 report. The database described in [4] were used to get the deterministic estimates of the regression parameters for probability of leak under a SLB differential pressure of 2560 psi, and the database doc-umented in (5) were used to obtain the deterministic estimates of the regression parameters for conditionalleak rate in PNNL's simulation.
Since the SLB leak rate data for 7/8-inch tubes do not satisfy the requirement for applying a SLB leak rate / voltage correlation, the SLB leak rate estimate is based on an average of l
all leak rate data independent of voltage. The analysis method for applying this leak rate model is similar to the method described above, except that the slope A is assumed to be zero and the intercept & is estimated by the average of the common logarithm of the leak rate data. Afore detail of this method is given in Section 4.6 of WCAP-14277 [6).
Ten thousand simulations were performed. The result is summarized in Table 4. In order
- to compare PNNL's projections with those in Farley Unit 2 report, the projected leak rates reported in Table 8-1 of Farley Unit 2 report [1] is reproduced and included in Table 4. The
, numbers from Farley Unit 2 report [1] are assumed to be the 95% bound values of the SLB leak rates.
i Table 4: Summary and Comparison of Projected SLB Leak Rates (gpm) at EOC 11. PNNL's Projections Are Based on 10,000 Simulations. POD =0.6 Was Used.
PNNL SG A SG B SG C Projection (# of ind.=73) (# ofind.=114) (# ofind.=252) hiaximum 187.5517 52.6613 109.5373 hiinimum 0.0000 0.0000- 0.0000 Average 0.2009 0.1381 0.4195 Aledian 0.0028 0.0000 0.0385 Std Dev. 2.7628 0.9484 1.9793 95% Bound 0.5980 0.5570 1.8667 Farley 2 Proj. 0.26 0.35 1.23 1
(95% Bound) 4 The PNNL prrjections are all larger than the Farley II leak rate projections, by about 50% However, when these results are compared against the plant-specific allowable limit, which is 11.4 gpm, it can be seen that the answers are approximately equal.
4 Comparison of Projected and Actual EOC Voltage Distribution for SG C -
This submittal also compared the projected EOC distribution with the observed distribution.
Such a comparison allows one to determine if the components of this inspection model are 7
i November,1996 working correctly. In this report, the projected EOC voltage distribution at EOC 10 can 3 be compared to the actual EOC 10 inspection results for SG C to determine whether the
- voltage growth modeling is correct.
i Figure 2 presents a comparison of the projected with actual EOC 10 Voltage distribution
- for SG C. These results were extracted from Table 7-1 in [1]. The actual distribution in this plot has been surrounded by 95% confidence bounds, which describe the amount of variability j one would expect in an estimate such as this when it is being estimated from a sample of this j size. These particular con 6dence bounds do not describe the sizing and detection inspectiva i errors that might be present within the actual distribution. Nevertheless, the confidence
- bounds give a gauge that can be used to compare the two distributions.
l As one can see from the Figure, the projected distribution generally falls within the 95%
! bounds, indicating general agreement between the projected and actual distributions. These j results show that the strategy of using the voltage growth rates from the previous cycle gives j a good prediction for the next cycle.
i j Figure 2: Comparison of Projected and Actual EOC 10 Voltage Distributions for SG C 1
I t
l 'i j O / j Adal j 8 i -
Projected i i
e' $ ---- 95% Bounds s' \
e \ n
% l 's l ',
~
l . *fs..
\l \
,i\
- n. ,:s ,o'\
-e o
' '-r s ,
h_
o
/l jl N,,,g --
s------
, s,',,,'
- s et _
O
_ _8 ..___."-S h u
, , i . . .
0.5 1.0 1.5 2.0 2.5 3.0 (V) Volts 8
November,1996 5 Comments on Voltage Dependent POD The NRC Interim Plugging Criteria requires the use of a 60% POD in the leak rate and burst probability calculations, unless a different value can be justified. In Section 4.3 of the Farley submittal, evidence is presented to justify a POD that is related to voltage.
POD is usually considered to be a function of flaw size (as flaws get bigger, they are easier to detect), and experiments have been conducted to determine this relationship. However, in the case of the IPC inspections, no flaw size is produced, se historical POD curves cannot be applied to IPC results.
The submittal presents two sets of data that are relevant to POD; One set of data originates from cycle 9/10 inspections of Farley 2, and the other from cycle 8/9. The data is used to estimate a quantity called probability of prior cycle detection (POPCD). The POPCD statistic originates from EPRI, and EPRI has produce a POPCD curve using data from several plants. Both data sets show that POPCD is related to voltage, and for "large" voltages (above 2 volts), POPCD is close to one. Specifically, the submittal comes to the conclusion; "In summary, the Farley Unit 2 EOC-8 and EOC-9 POPCD's support a voltage dependent POD higher than the NRC POD =0.6 and approaching unity above 1.8 volts. It is concluded that the POD applied for IPC leak and burst projections needs to be upgraded from the POD =0.6 to a voltage dependent POD."
Based on a review of the POD assessment in reference [1], the these conclusions are not fully supported by the data. The data does indeed show that POPCD is voltage dependent, and that POPCD is close to unity above 1.8 volts. But the submittal treats POD and POPCD as if they were equivalent, and they are not. An examination of the definition of the two statistics should clearly illustrate the differences. POD is defined as; POD = Total # of defects in inspected tubes (6) while the definition of POPCD (ignoring plugged tubes) is given by; POPCD = # f defects detected in first & second inspections
- of defects detected in second inspection POPCD can be best thought of as a POD calculated under the assumption that the second inspection produces a "true-state" description of the tubes. If the second inspection had a very high POD, so that it found almost all the flaws, then POPCD would provide a good estimate for POD. However, no evidence is presented to show that the second inspection POD's are near 100% In fact, if evidence existed to show that the second inspection POD was high, this evidence would be sufficient to replace the current POD =60% value and there would be no need to eslculate POPCD.
Another assumption that could be used to relate POPCD to POD would be the assump-tion of independence. If the two inspection results were independent, then POPCD would
~
9
November,1996 produce a conservative estimate for POD. Unfortunately, there is much evidence to indicate inspection results are not independent, so evaluating the POPCD data from this perspective would require further justification.
6 Tube Pulls and Destructive Examinations l During this outage, one tube (R27C54) was pulled, nondestructively evaluated and then subjected to leak and burst tests. Three locations from this tube (TSP 1, 2, and 3) were subjected to testing. which resulted in the data points listed in Table 3-6 of reference [1]. l These data points were evaluated against the EPRI exclusion criteria, and one data point
, was excluded (TSP 3), because ofits crack morphology. Also, TSP 2 was classified as a NDD during field and lab inspections, so no voltage is associated with this flaw and it cannot t
be directly included in the burst or leak-rate regressions. However, the flaw at TSP 2 is indirectly relevant to these regressions because it is an example of a " missed" flaw that the 3
POD =60% correction is meant to account for. ;
Only one flaw (at TSP 1) provides burst and leak rate information from these destructive tests. The burst pressure and leak-rate information (the flaw did not leak) from TSP 1 conform very well with burst and POL regression results (see Figure 3-4 and 3-5 of the submittal). Since the data fits the regression models well, it causes a very small change when included in the regression fits (see tables 3-8 and 3-9). This data point is not significantly
' different than the other regression data and one can conclude that the Farley 2 cracking fits the burst and leak rate correlations derived from the EPRI data set.
l 7 Conclusions
! In this review, PNNL has produced independent calculations, which have been compared to the licensee's EOC 11 burst and leakage calculations. The following summarizes the findings of this review:
- 1. PNNL's projected EOC 11 voltage distributions for SGs A, B and C at Farley Unit 2 l agreed with those projected by the licensee.
i
) SGs A, B and C at Farley Unit 2 were consistent with those projected by the licensee.
l and were below the NRC approved threshold of 1 x 10-2, 4
- 3. The 95% bound values of PNNL's projected EOC 11 SLB leak rates and the Farley Unit 2 projected EOC 11 SLB leak rates were all well below the plant-specific allowable SLB leakage limit, which is 11.4 gpm. The difference between the two projections is likely caused by small differences in the inputs.
- 4. The projected and actual EOC 10 voltage distributions agreed quite well.
10
m i e
- November,1996
- 5. The data presented shows that POPCD is strongly related to voltage. However, it is not clear that POPCD is a good estimate of POD.
- 6. Farley II data from pulled tubes supports the EPRI leak-rate and burst models, References
[1] Westinghouse Electric Corporation Nuclear Services Division, "Farley Unit 21995 In- )
terim Plugging Criteria 90 Day Report," July 1995.
[2] Gao, F. and Heasier, P. G., "PNL Documentation of Computer Programs to Calcu.
late EOC Voltage Distribution, Burst Pressure Probabilities and Leak-Rate," Pacific Northwest Laboratories, December 1995.
[3] WCAP-14123 (SG-94-07-009), " Beaver Valley Unit 1 Steam Generator Tube Plugging Criteria for Indications at Tube Support Plates July 1994"
[4] Committee for Alternate Repair Limits for OCSCC at TSPs, "PWR Steam Genera-tor Tube Repair Limits - Technical Support Document for Outside Diameter Stress Corrosion Cracking at Tube Support Plates," TR-100407, Revision 2A, EPRI, January 1995.
[5] EPRI Project S404-29, " Steam Generator Tubing Outside Diameter Stress Corrosion Cracking at Tube Support Plates - Database for Alternate Repair Limits, Volume 1:
7/8 Inch Diameter Tubing," NP-7480-L, Volume 1, Revision 1, September 1993.
[6] Westinghouse Electric Corporation Nuclear Services Division, "SLB Leak Rate and Tube Burst Probability Analysis Methods for ODSCC at TSP Intersections," Westing-house, WCAP-14277, January 1995.
11