ML20212B179

1 SG ARC Analyses in Support of Full Cycle Operation

ML20212B179
Person / Time
Site:	Farley
Issue date:	10/31/1997
From:	WESTINGHOUSE ELECTRIC COMPANY, DIV OF CBS CORP.
To:
Shared Package
ML20212B177	List: ML20212B169 ML20212B179
References
SG-97-10-004, SG-97-10-4, NUDOCS 9710270277
Download: ML20212B179 (48)

Text

. - _ . . - . _ . _ . _ _ - _ - . _ _ _ . _ . _ _ _ _ - _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ . _ _ _

e i

. +

SG.9710 004 4

i Farley 1 SG ARC Analyses In Support of Full Cycle Operation 1

l October,1997 1

i i

Westinghouse Electric Corporation Nuclear Service Division 4

9710270277 971020 "

PDR ADOCK 05000349 P PDR.

. - _ _ _ _ _ _ _ _ _ _ . . _ - . _ . - . _ . _ _ _ _ _ _. . _ ~. . . - _ _

4 ,

. t i

Table of Contents Topic Page

1.0 INTRODUCTION

11 2.0

SUMMARY

AND CONCLUSIONS 2.I 3.0 DATABASES AND '.DRP. ELATIONS USED FOR SLB LEAK AND BURST ANALYSES 31 3.1 Summary of Parley Units 1 and 2 Pulled Tube Examination Results 31 3.2 Current NRC Approved Database 33 3.3 Current NRC Approved Database with Addition of Latest Farley Pulled Tubes 33 3.4 SLB Leakage Using Only Farley Pulled Tubes and Model Boiler Data 33 3.5 Current NRC Database with Exclusion of French Data 4 3.6 Revised EPRI Data Exclusion Criterion 3 Applied to Current and Latest Farley Database 34 4.0 ANALYSIS METHODS bvALUATED 4l 4.1 WCAP 14277: Blased Leak Rate Parameters, No Leak Rate Correlation 41 4.2 Unblased Leak Rate Parameters, No Leak Correlation 4 ,2 4.3 Empirical / Exponential Leak Rate Simulation, No Leak Correlation 43

. 4.4 SLB Voltage Dependent Leak Rate Correlation 46 4.5 Voltage Dependent Growth Rate for Farley 1 47 5.0 SLB LEAK RATE AND TUBE BURST PROBABILTI'Y ANALYSES 51 5.1 Comparison of EOC 14 and BOC 15 SLB Leak Rates for POD = 0.6 5-1 5.2 Analysis Results for EOC-14 As Measured Voltage Distribution 51 5.3 Projected Leak and Burst Results at EOC-15 52 5.4 Conclusions 5-3 6.0

SUMMARY

OF SUPPORTING BASES FOR VOLTAGE DEPENDENT SLB LEAK RATE CORRELATION 61

7.0 REFERENCES

71 1

e

FARLEY 1 SG ARC ANALYSES IN SUPPORT OF FULL CYCLE OPERATMN

1.0 INTRODUCTION

(n the Farley 1 1997 ARC 90 day repon (Reference 7.1), it was reponed that the projected EOC.

15 SLB leak rate slightly exceeded the current allowable leak rate limit. It was funher noted that inclusion of the 1997 Farley 1 pulled tube data in the ARC correlations could have a sipificant impact on the leak rate correlation and a modest impact on the burst pressure correlation. The 90 day report funher identified that the SLB leak rate would be below the acceptance limit if updated analysis methods currently under review by the NRC are accepted for ARC analyses.

This repon provides ARC analyses for the updated database including the recent 7/8 loch diameter pulled tube resuhs from Farley Units I and 2. SLB leak and burst probability analyses are reponed for both the current and updated ARC databases with sensitivity analyses for variations in the analysis methodology and data included in the databases. This report provides suppon for full cycle operation of the Farley 1 SGs based on demonstrating that SLB burst probabilities and leak rates within allowable limits are obtained by acceptance of one or more of the database or methodology changes currently being reviewed the NRC.

The following database, methodology and POD variations are evaluated in this repon:

. , , Database Variations

• Current database

• Updated database based on current database with addition of Farley SG pulled tube results'-

revises ARC correlations for burst pressure, probab!!ity of leakage and leak rate Updated database modified to include only the five Farley SG pulled tube indications with measured leak rates (all domestic data - excludes French data from leak rate correlation) and the model boiler data - used only for SLB leak rate correlation Updated database modified to exclude the French data as recommended by EPRI in Reference 7.3 and currently under review by the NRC Updated database modified by application of the revised EPRI Data Exclusion Criterion 3 (Reference 7.6) which leads to the exclusion of one high voltage model boiler data point from the leak rate correlation referenced report to be submitted to the NRC Methodology Variations Current NRC approved SLB leak rate methodology based on no correlation ofleak rate with voltage and biased leak rate parameters SLB leak rate methodology based on no correlation of leak rate with voltage and use of

- unbiased leak rate parameters as recommended by EPRI in Reference 7.3 SLB leak rate methodology based on no correlation of leak rate with voltage and use of empirical leak rate parameters with the high leak rate tail of the distribution adjusted by an exponential function - methodology described in Section 4 of this report SLB leak rate methodology based on leak rate correlation with voltage that satisfies NRC statisilcal requirements for a correlation as obtained by including only the Farley pulled tube

. . *w ,ir.iu, ,,, , ia i ,

11

' , results and the model boiler specimens (excludes French data)  !

• SLB teak rate methodology based on leak rate correlation with voltage that satisfies NRC

' statistical requitements for a correlation as obtained by applying the revised EPRI Data Exclusion Criterion 3 Probability of Detection (POD) Variations

• Current NRC OL 95-05 requirement for use of POD = 0.6 independent of voltage

+ Minor POD modification to apply POD = 0.6 for indications < 10 volts and POD = 1.0 for indications > 10 volts

+ Voltage dependent POD based on EPRI POPCD (Probability of Prior Cycle Detection) as recommended in Reference 7.3 The databases and resulting ARC correlations applied in this report are described in Section 3.

Section 3 also includes a summary of the recent Farley Units I and 2 pulled tube examination results that provide the data for updating the database. Section 4 describes the alternate analysis methods. Results of SLB leak rate and burst probability analyses are given in Section 5. The summary and conclusions for this report are given in Section 2.

t m=va, .%., m - 12

,2.0 3UMMARY AND CONCLUSIONS

' The analy:;es of this report support full cycle operation for the Farley-1 SGs based on the h condition monitoring and operational W.essments summarized below. This report was prepared to supplement the Farley 1 1997 ARC 90 day report which identified that recent pulled tube results from Parley l could significantly impact the ARC correlations, particularly for leakage analyses.

Farley-1 pulled tube R2C85 had a burst pressure which satisfies the RG 1.121 burst margin guideline of 1.4AP3t, and the resulting burst pressure has a minimal impact on the ARC correlation.

The measured SLB leak rate for this indication was 0.72 gpm for which incorporation in the SLB leak rate correlatiun results in increased calculated leak rates. The field bobbin voltage for R2C85 was 13.55 volts .vhich increased to 28.5 volts after the tube pulling operations. Pressure Pulse Cleaning (PPC) was applied to the Farley 1 SGs at EOC 14 prior to 3

the eddy current inspection. The destructive examination results show some fatigue crack propagation within a cellular corrosion patch near the major axial crack but no evidence of fatigue was found on the faces of the axial burst crack. It is expected that the fatigue propagation resulted from the PPC operations. The PPC operations likely contributed to the large voltage growth (12.4 volts) found at the EOC-14 inspection for P2C85 due to opening of the cellular .

corrosion cracks and possibly some impact on the major cual crack. However, the influence of the PPC operations on the R2C85 EOC 14 voltage cannot be quantified and the indication is t

included in the ARC database for the analyses of this repon. An updated database and associated ARC correlations were prepared that include the R2C85 results and 1996 Farley 2 pulled tube results. These data ar the only pulled tube data obtained for 7/8" diameter tubing since the current ARC database smd correlations were prepared.

Tube R2C85 was in situ leak tested at normal operating conditions prior to the tube removal from the SG. No leakage m s found in this test even though destructive examination identified a throughwall crack length of 0.42" and leakage at normal operation was found in the laboratory leak test. No leakage for the in situ test confirms that the packed crevices restrict crack opening and leakage flow as also found for tubes / TSP with non dented intersections th?t were removed from the Dampierre SGs (Reirence 7.11). Pull force tests performed on the Dampierre tube and TSP intersections, which were removed intact from the SG, demonstrated large forces required to displace the TSPs such that the TSPs would not displace in a SLB event (Reference 7.11).

The Farley-1 TSPs would similarly prevent TSP displacement in a SLB event (which precludes a tube burst) and limit leakage to negligible levels as demonstrated by the in situ test. However, the tube integrity analyses of this report very conservatively assume the plates are not constrained in a SLB cvent and leakage would be under free span conditions consistent with the intent of NRC Generic Letter 95-05.

EOC-N Condition Moni*orine Assessment For the EOC-14 condition monitoring assessment, SLB tube burst probabilities are less than the NRC reporting guideline of 104 for the current ARC database (3.8x104) and after the database is updated (5.7x104) to include the recent pulled tube results for Farley Units 1 and 2. Similarly, the SLB teak rates at EOC-14 are less than the Farley SG allowable limit of 13.7 gpm for both the current database (7.9 gpm) and the updated database (9.9 gpm). These burst probabilities and

-wmy.,% , a em 2-1

__-- j

o l

' 1 i

leak rates incorporate the pulled tube R2C85 results of no burst at SLB conditions and a SLB i

' leak rate of 0.72 gpm. The Monte Carlo calculations are performed excluding the R2C85 l indication and the 0.72 gpm measured leak rate is added to the leak rate from the Monte Carlo analyses. Consequently, condition monitoring structural and leakage F.m '.y requirements are satisfied at EOC 14.

EOC 15 Overational Assessment For the EOC 15 operational assessment, the projected SLB burst probability of 9.9x10 4 fo the current database with a POD of 0.6 is below the NRC reporting guideline of 10 4 . For the 4

updated database, the burst probability for a POD of 0.6 increases to 1.2x10 which slightly 4

exceeds the NRC guideline. However, the burst probability reduces to 8.0x10 for a very conservative POD that only assumes that allindications above 10 volts are detected (POD = 1.0 for > 10 volts and 0.6 for < 10 volts). When the updated database is combined with voltage dependent growth rates based on a methodology developed for another pla.it, a voltage dependent POD, such as the EPRI recommended POPCD currently being reviewed by the NRC, results in a burst probability below the NRC reponing guideline. Overall, the operational assessment for structural integrity is acceptable for full cycle operation based on the small difference from the 4

NRC guideline (1.2-1.4x10 versus 1.0x104) and that realistic POD assumptions for detection of high voltage indicatiotis reduces the burst probabilities to below the NRC guideline.

The current leak rate analysis method with biased leak rate parameters and POD = 0.6 results in projected EOC-15 SLB leak rates of 15.7 aid 20.4 gpm for ARC correlations based upon the current and updated databases, respectively. Although these results exceed the allowable limit of 13.7 gpm, applying the expected EPRI POPCD rather than a POD of 0.6 reduces the corresponding leak rates to 9.9 and 12.9 gpm which are within allowabic .!mits. An acceptable leak rate of 7.8 gpm is obtained for a POD of 0.6 with a methodology applying empirical leak rate parameters (measured data as a cumulative probability distribution) with an exponential tail to account for the potential of a larger than measured leak rate. Application of the empirical leak rate parameters with voltage dependent growth rates and using the EPRI POD results in a leak rate of 5.3 gpm for the updated database. An acceptable leak rate is also obtained without a correlation when the unbiased leak rate parameters are applied in conjunction witii the EPRI POPCD.

Acceptable (< 13.7 gpm allowable leak limit) SLB leak rates are also obtained whenever leak rates are calculated with a voltage dependent leak rate correlation. The current and updated databases result in a voltage dependent correlation when the leak rate correlation is based on the Farley pulled tube and model boiler data (or the French data are excluded) and when the revised EPRI Data Exclusion Criterion 3 is applied. The best estimate SLB leak rate of 2.3 gpm is obtained using the voltage dependent leak rate correlatica which results from using only the Farley pulled tube and model boiler leak rate data. The resulting leak rates are in the range of 2.3 to 4.5 gpm dependent on the POD assumed for the analysis and whether or not voltage dependent growth rate distributions are applied. Exclusion of the French data is currently being reviewed by the NRC and the revised Criterion 3 is expected to be submitted to the NRC for approval in the near term.

It is concluded that acceptable structural and leakage integrity are found for the EOC-14 ww , .,%., - 2-2

' operational assessment. SLB burst probabilities and leak rates are acceptable upon inclusion of

' realistic PODS in the analyses. Use of only the Farley pulled tube and model boiler data for the ARC leak rate database results in a voltage dependent correlation for which SLB leak rates analyses show large margins against allowable limits.

The following summarizes the database and/or methodology changes that result in acceptable SLB tube burst probabilities and leak rates when applying the updated ARC database:

Burst Probability -

= Application of conservative POD = 0.6 for indications less than 10 volts and 1.0 for indications exceeding 10 volts.

Application of EPRI voltage dependent POD based upon POPCD (submitted for NRC review).

When voltage dependent growth rates are included in the analysis, a voltage dependent POD based upon POPCD is required to obtain a burst probability < 10.

SLB Leal Rate Application of EPRI voltage dependent POD based upon POPCD when applying the current analysis method with biased or unbiased (submitted for NRC review) leak rate parameters and leak rate independent of voltage (no correlation).

Application of empirical leak rate parameters and an exponential tail with either POD = 0.6 or the EPRI POPCD.

Application of the EPRI recommended database (submitted for NRC review) that utilizes th:

. Farley and model boiler leak rate data and results in a voltage dependent leak rate correlation (excludes the French data from the correlation).

Application of the revised EPRI Date Exclusion Criterion 3 (to be submitted for NRC review) which excludes one model boiler data specimen and results in a voltage dependent leak rate correlation.

e

- -vi, ,,,,m.,mim 2-3

• 3.0 ANALYSES DATABASES AND CORRELATIONS USED FOR SLB LEAK AND BURST it has been noted in Reference 7.1 that addition of the recent Farley-1 pulled tube leak rate data to the ARC database results in a significant change to the leak rates and an update of the correlations used for ARC analyses of 7/8" diameter tubing is required. The recent Farley 1 and Farley 2 pulled tube data are described in Section 3.1. The current reference correlation and updated correlation are described in Sections 3.2 and 3.3. Since five pulled tube indications from Farley SGs are the only domestic pulled tube data in the EPRI ARC leak rate database for 7/8" tubing, a SLB leak rate correlation using only the Farley and model boiler data provides the best estimates of Parley SG leak rates for this operating cycle evaluation. The correlation resulting from the use of only Farley data is described in Section 3.4. The NRC is currently reviewing the EPRI recommendation of Reference 7.3 to exclude the French data from the ARC correlat Sensitivity analys:s are included in this report using the correlations resulting from er.clusion of the French data as described in Section 3.5. EPRI Data Exclusion Criterion 3 has been revise in Reference 7.6 to incorporate NRC comments on the prior correlation and will be submitted to the NRC for review in the near future. Application of this criterion results in a voltage dependent leak rate correlation as described in Section 3.6.

3.1 Summary of Farley Units 1 and 2 Pulled Tube Examination Results A description of the destmetivc examination results for Farley 1 pulled tubes is given in the 1997 90 day report (Reference 7.1). This section provides a summary of the results that are used to update the SLB leak and burst correlations. Pulled tube examination results from 1996 Farley-2 pulled tubes are given in Reference 7.2 and summarized herein as used to update the cor:clations.

The Farley 1 results for tube R2C85 significantly impact the correlations while the Farley.2 results have a minor impact on the correlations. It is important to note that removal of large voltage indications from the Farley SGs has been emphasized with the result that the only 7/8" diameter tubing, domestic indications that leak have been removed from the Parley SGs. A total of 5 pulled tube indications with measured leak rates have been obtained from the Farley SGs.

The only other 7/8" tubing, pulled tube indications with measured leak rates are two indications '

removed from French SGs. It has been recommended (Reference 7.3) that the French data be excluded from the ARC correlations due to distinct differences in leakage and burst trends with voltage compared to domestic data. This difference may be due to uncertainty in the voltage adjustment from French to ARC vcitage normalization or to differences in crack morphology.

Given the dominance of Parley pulled tubes in the leak rate database, it is appropriate for this Farley-1 operating cycle evaluation to use only the Farley pulled tube results in combination with the model boiler results to develop a leak rate correlation. This use of the Farley pulled tubes results in a voltage dependent leak rate correlation as developed in Section 3.3 and applied for the SLB leak rate analyses of this report.

The 1997 Farley-1 pulled tube indication at R2C85-TSP 1 contributes to the ARC database. This indication had a 13.55 bobbin volt pre pull indication which increased to 28.5 volts after the tube pull. The measured burst pressure for this indication was 3990 psi at room temperature and the n easured leak rate adjusted to reference SLB conditions was 0.72 gpm. This indication was turther evaluated to determine its appropriate use in the ARC database, particularly with regard to the cause for the large voltage indication and increase in voltage from the tube pulling o-a -, .,m , -

3-1

, operation. Figure 3-1 shows the + Point response for the field pre pull and laboratory post pull conditions. Figures 3 2 and 3 3 show the destructive exam results and a comparison of the

' destructive exam results with the " blind", pre exam + Point depth profile (Reference 7.4) for the large indication. For comparisons with NDE data, the destructive exam profile is averaged over the + Point coil field size (about 0.16" for axial indications) and corrected for uncorroded ligaments as shown in Figure 3 2. The predicted + Point depth profile shown in Figure 3-3 is in good agreement with the destructive exam profile. Figure 3-4 shows a sketch of the R2C85 OD degradation including the location of the axial burst crack opening and the axial tensile rupture location. The axial tensile rupture test was performed to support development of an axial rupture pressure versus voltage correlation for ARCS with " locked" TSPs.

Figure 31 shows a substantial increase in the secondary crack response as a result of the tube pulling operations. This is a result of opening of the cellular corrosion patches seen in the sketches of Figure 3-4. The + Point maximum voltage response of the primary axial crack is essentially unchanged as a result of the tube pulling operations (both are approximately 11.6 volts). The dip in the primary crack, + Point response of Figure 31 below the center of the TSP remains after the tube pull. It is believed that this dip in vertical amplitude is associated with +

Point analyses in the presence of a cellular patch. At a cellular patch the axial response coil goes up (positive voltage) while the circumferential goes down (negative voltage). It is believed that the circumferential coil response at the cellular patch causes the dip by reducing the net vertical amplitude. This dip does not appear to be associated with an uncorroded ligament within the macrocrack and is not found in the RPC 80 mil coil response. The sharp drops in voltage above and below the center of the TSP are likely associated with the ligaments at about 0.25 and -0.23 inches. These ligaments are at each end of the throughwall crack length for this indication.

Figu e 3 5 shows a comparison of the + Point pre and post pull depth and voltage profiles.

Diff;rences between the pre and post pull profiles are modest and do not permit identification 6f the cause of the large voltage increase as a result of the tube pulling operations. The pre and post-pull voltages in the vicinity of the uncorroded ligaments are similar and do not indicate that the ligaments tore during the removal operations. These results tend to indicate that the primary crack was not strongly affected by the tube pulling operations. It is clear from Figure 3-1 that the secondary cracks with cellular corrosion patches were opened by the tube removal operations and these effects would contribute to an increase in the bobbin voltage. The maximum corrosion depth found away from the primary crack was 74%. While it would not be generally expected that opening of the secondary cracks would contribute an increase of 14 volts from pre to post-pull, this effect is the caly identifiable effect from + Point data that helps to explain the voltage increase.

The R2C85 indication had a growth of 12.4 volts over Cycle 14. At the end of Cycle 14 Pressure Pulse Cleaning (PPC) was performed and followed by the eddy current inspection. Tube R2C85 is located near the pulser nozzle where PPC can cause vibration of the tube. The destructive exam identified striations due to fatigue within the oblique components of a cellular patch which is likely a result of PPC. No evidence of fatigue was found on the face of the primary axial crack. Consequently, the influence of PPC on the primary axial crack cannot be clearly determined. Since the destructive exam shows some influence of PPC on propagating the secondary cracks, an increase in bobbin voltage occurred as a result of PPC although the magnitude cannot be estimated. Overall, it is concluded that PPC affected the secondary cracks, which could have had some influence on the primary crack and resulted in some indeterminable

-~m,.,,- 3-2

' increase in bobbin voltage. The R2C85 indication is conservatively included in the ARC

' database since the effects of PPC cannot be quantified for application of the EPRI .

s ondata esclu i

' criteria. Tube R2C77 in SG A had a 6.35 volt bobbin indication and the second largest growth rate of 3.22 volts. This indication is also located near the PPC pulser nozzle. The + Point response for this indication also tends to show cellular corrosion patches near the major axial crack similar to that found for R2C85. It is reasonable to expect that the voltage increase for R2C77 may also have been influenced by the PPC operations. Although both R2C85 and R2C77 may have had bobbin voltages and growth rates influenced by PPC, they are conservatively included in the ARC analyses.

Tube R34C53 with a 6.8 volt bobbin indication was removed from Farley 2 in the 1996 inspection. The laboratory measured leak rate for this indication was found to be 2.02 liter /hr.

This leak rate is very close to the regression fit the 7/8" leak rate database and the destructive t

exam results for this indication did not significantly influence the ARC correlations. This indication was pulled since the current ARC reference database was defined and, together with the R2C85 indication, provide a basis for updating the ARC correlations as described in the Farley 190 day report (Reference 7.1).

Three additional Parley SG pulled tubes had measured leakage at SLB conditions. Farley-1 tube R28C35 with a 4.03 volt indication bemoved in 1995) had a SLB leak rate of 2.19 liter /hr and Farley-2 pulled tubes R4C73 with a 2.81 volt indication and R21C22 with a 7.56 volt indication had SLB leak rates of 0.08 and 0.07 liter /hr, respectively. These indications are the only domestic pulled tubes with SLB leakage in the ARC database. The only other pulled tubes with SLB leakage were two indications removed from French SGs. Since the Farley pulled tube indications dominate the ARC pulled tube leak rate database, it is appropriate to utilize only the

! Farley data and the Model Boiler data to assess Farley SG leak rates for an operational assessment of full cycle operation. The resulting leak rate database without the two French indications results in a SLB leak rate versus voltage correlation satisfying the NRC statistical guidelines for a correlation as described in Section 3.4 below.

I 3.2 Current NRC Approved Database The current NRC approved ARC database is described in Reference 7.5 and was used for SLB leak rate and burst probability analyses in the Farley-190 day report. These results are compared in this report with updates to the database, as described in the following sections, resulting from the addition of the new pulled tube data from Farley-1 and Farley 2.

3.3 Current NRC Approved Database with Addition of Latest Farley Pulled Tubes The addition of the above described two data points from Farley SGs provide a basis for updating the ARC correlations. In this case, the two data points are added to the current NRC approved database. The resulting burst pressure, probability of leakage and leak rate correlations were included in the Farley-190 day report (Reference 7.1). The addition of the 0.72 gpm leak rate for the R2C85 indication results in a significant change to the correlations. Table 3-1 shows the effects of addition of the new pulled data from Farley-1 and Farley-2 on the leak rate database.

The extent of impact of the revised database on SLB Icak rates relative to the reference database is described in Section 5. -

w w m,.g.,- - 3-3

, ,3.4 SLB Leakage Using Only Farley Pulled Tubes and Model Boller Data

' As noted above, Farley SG pulled tubes contribute five data points to the SLB leak rate database and represent the only domestic pulled tubes with measured leak rates. Consequently, the most accurate estimates of Parley SG leak rates for an operating cycle evaluation can be made using only the Farley pulled tubes plus the Model Boiler data for the leak rate correlation. Using only Farley pulled tube data eliminates the two French data points from the leakage correlation and results in a voltage dependent leak rate correlation satisfying the NRC statistical requirement for a < 5% p-value to obtain a correlation. The resulting regression St is illustrated in Figure 3-6 along with the data points in the reference database and new data. Parameters defining the leak rate correlation are shown in Table 3 2. This leak rate correlation is the primary basis for estimating Farley.1 SLB leak rates in this report Only the leak rate correlation is modined for this option.

s 3.5 Current NRC Database with Exclusion of French Data 2

Exclusion of the French data from the EPRI ARC database was recommended in Reference 7.3 and is currently under review by the NRC. This results in exclusion of two data points from the reference leak rate correlation discussed in Section 3.2, one point from the burst correlation and thirteen data points from the probability ofleakage correlation. The resulting leak rate correlation is the same y that obtained by using only Farley pulled tube data for the correlation (Table 3-2).

However, this database option also raodifies the POL and burst pressure correlations.

3.6 Revised EPRI Data Exclusion Criterion 3 Applied to Current and Latest Farley Database The original EPRI Data Exclusion Criterion 3 was not approved by the NRC in GL 95 05.

Exclusion criterion 3 has been modified in Reference 7.6 to reflect NRC concems with the original criterion. Criterion 3 was developed to identify erroneous leak rate measurements although the specific cause for the measurement error cannot be identified. Typically, this criterion applies to very low leak rate measurements for the size of flaw being tested such as can

' occur if deposits lodge in the crack and restrict leakage. Criterion 3 requires two conditions to be satisfied to exclude the data point. The measured leak rate must be below the 95% prediction interval for the voltage correlation and also below the 95% prediction interval for the leak rate versus throughwall crack length data, if destructive data is not available for the indication, the second requirement is that the SLB leak rate must not exceed the normal operating leak rate measurement by more than 10% whereas factors of > 2 are expected. For the EPRI 7/8" database, application of the revised Criterion 3 results in excluswn of Model Boiler specimen 542-4 from the database. With exclusion of this very low leak rate for a 50 volt indication, a voltage dependent leak rate correlation satisfying the NRC statistical requirement is obtained. A summary of regression analysis results obtained by exclading the Model Boiler specimen 542-4 is shown in Table 3-3 Figure 3-7 illustrates the distribution of the mean and median predicted i

with the correlation. It is anticipated that Reference 7.6 will be submitted to the NRC in the near future requesting approval of the revised Criterion 3.

s Table 3-1: Effect of Inclusion of New Farley 1 & 2 Data on the Reference Leak Rate Database for 7/8" Tube APC Applications (All French Data Included)

Leak Rate (lph) Log ( Leak Rate )'-

Parameter Reference w/ A2 & Al Reference w/ A2 & Al Database Database Database Database Sample Size 27 29 27

~

29 SamNe p 13.31 18.I2 0.5696

~

0.6172 Sample a 20.84 34.57 0.8188 0.8482 Population Estimates Based on Log (Leak Rate) Sample Estimates Biased Unbiased Population p 21.961ph 27.89 Iph 18.76 Iph 23.661ph Std. Dev., o 128.00 lph 185.75 Iph 63.80 sph 88.98 Iph p Value 6.4% 7.6%

Notes: 1. The distribution of sample leak rates has been previously shown to not contradict a null hypothesis that the parent population is a log-normal distribu-tion.

Table 3-2: Effect of New Farley 1 & 2 Data on the Leak Rate vs. Bobbin Amplitude Correlation (Model Boiler and Farley 1& 2 Data Only) log (G, ,) = 3

+ , log ( Volts) p Reference Database with New / Old 7

Database Value A2 & Al Data Ratio 3 -0.8392 -0.7557606 0.9016

p. 1.2636 1.2461 0.986 2

r 22.9 % 19.9 % 0.868 og,,,,, ( ,) 0.7132 0.7546 1.058 N (data pairs) 25 27 p Value for 1.6% 2.0% 1.27

,w~u.*, . cam, nu.n 3-5

9 Table 3-3: Effect of Excluding Model Boiler 542-4 Data on the Leak Rate Database for 7/8" Tube APC Applications log (Q,,,) = 3 + p, log (Volts) aster Reference Database with New / Old Database Value A2 & Al Data Ratio 3 -0.9207 -0.8343 0.9061

$4 1.3211 1.3005 0.9844 r 2 20.4 % 18.0 % 0.882 o r,,,,, ( $3 ) - 0.7510 0.7877 1.049 d N (data pairs) 26 28 l p Value for $2 2.0% 2.5% 1.25 I

I

< % mm,.,,a-, aim 3-6 l

. Figure 3-1 Forley-1, R02C85

). - -

Pre and Post Pulled + Point Response

. Pre Pull

,2

..,. ., m ,,

_ _ ,,,,.u. .....s .......

j , , . . . . .. ..... . ...... .. - . - - " - "

,a. . . . . . ......,... ,,.........

, , . , ,,.w... ...... ..m . . m T.,,a, n .,,.,,

. . . . . . .....-. .o...

,, i .

u.,.,.,,., ..,, .,.

I I

l

\

t l

m> w~

I l  ! h i r' s g p f

- .. k ,

g C ' .

% n .. -.

4

%x3x .

o:

' ' " " = '

N,.

1

• wm a

Post-Pull

! EEE* E4 i

j s

.wi, vg.

i Y

• ME l

1 ggs 1

.i n..f E,n ,

- [

' 35 gg i' -

EM

4. .

4 sm l

i m..

i l

e se E

J a ,

4 i

3 ~1 i - - . _ . . - .-. . - _ -_ .. _. -

. o s

Figure 3-2  !

Farley-1,'97 Pulled Tube, S/G C R02C85-1H, Crack 1 seio.

Destructive Exam Results 7

including Coil Field Average (0.16") and Ligament Correction 100 % 0 O -

7 -O-Local Depths 90 % --

-+-Cod Fekt Average 5

m _

g -G-Cod Average wth Le c ,,e_

-e-Lig Cod Avg 70% -

, Cod Cod Avg I f

• m I Local Field w/ Lag m u ca.

6

}  ?  !

g

= ATD-- 76 99 % 76 99% 76 48%

50 % -- - 1 f* Max.

5

, Depth 100% 100 % 100%

40% - - --- -

N - - - -

Lgth.: 0 81- 0 81- 0 81-f 30 % -- - -- - - - -

20 % ---- -- - --

10% - - - -- -- --- -- -- --- -

0% ' ' * * "

-=-.:.:. **'** "*'*'*'* '

- - - - - ~;. - - - - - - - - :2.*:::: .*:::.'s  :.':. :::.

4.450 -0.375 -0.300 -0.225 -0.150 -0.075 0.000 0.075 0.150 0.225 0.300 0.375 0.450 Antal Distance (in.)

L Chart 1 Far(bc5) xis 9/4/97 3-8 e

1 ,

__ ___.-_-_J

Figure 3-3 Farley-1, '97 Pulled Tube, SIG C R02C85-1H, Crat.k 1 Comparison of +Pdint and Destructive Exam TSP TSP 100 % , ----------------------- ,

90 %

I fa A O

W g- -, n  : + Pcht 80 %

4 -A a

1. - - * - - Destructive Exam *sth Cod 70% -- ---

Field and Ligament -

I .' 'l Correction A b g 60% -

g - -- - -

a .

g N 50 %

}

,- l 4

A-A _C 4 Max. Volts 11.55 3 a a g 40% - -

-g -- - - - - - - - -- -

g ATO. 71 82 % 76 99%

z a A

> f A 30% -

g --- -

A- Max Depth 97 0% 100 %

4 6

20% - - - --- -- . - -- - --- - - ~ - -

tength o es- oss 10% - --- - - - - - - -

t 0% -

i

-0.450 -0.375 -0.300 -0.225 -0.150 -0.075 0.000 0.075 0.150 0.225 0.300 0.375 0.450 AXIAL LENGTH (IN.)

Far(bc5).xis ChR20C85 9/4/97 1

1 I I

g l' .,

33 _ h El ,, .s - s , . - TOP 73 P/

-6 a

,,, 9 - - - -i f[

2 9

( ,

v

, .3 t- v -l 8 i s 0.00 -

a O '

~Q y,

> ,- ,i.

1. 8 El _ o.n .

.w \

_ wee ,sr i 1 I I

o* 90* 180* 270* 360*

Circumferential Position (degrees)

Figure 3-4 Sketch of the OD crack distribution found at the first tube support (TSPI) segion of Tube R2C85. Also shown is she h> cation of the burst fracture opening. The burst opening extended beyond the TSP crevice region enore than th intergranular corrosion. It is estirnated that the corrosion cracking on the burst fracture extended 0.03 inch above and below the crevice region. Finally, the post-burst test, tensile fracture face location is shown.

i 3-10

t Figure 3-5 Farley-1 SG C R2C85 Pre and Post-Pull + Point Voltages and Depth Profiles 100 --

r--- rrrr----- r r---- s- 14.0 C Destructive Exam Coil Avg.weth Lig. l 90 i i *-

l * .-

E i 12.0 0 +R Propust Depth l

80 l- j- .

!:1 & - j {

l "".,*****. *,

i.

--NF---+Pt. Post. Pull

I -

Depth  !

70 I +

l , '0*O l

,3i , - ior - -+Pt. Pre-Pull Volts, g [. ,' l

c. 60 t f

g.-. . l: E..5

+ , .

.E?0 . 8.0  :. e . ..Pt. Post #ull Volts

= 1 A ". . ',

, , - t 3 50 r

1

.l .--a *. l } i

. i 4

a j - a ' . -- i l j

o 40 2

i

~~

. a.

6.0 F I 4

'~ N.E' + ..

_DL. ffLPrtPuN fhPgst i

!  ; .  :: . . i '

pun l A' .h. j g  ! A j'E .

1 30 /. .' Length o et o as u as a ." A".E' 15 i

' i .

i - 4.0

.[

O

.- l *i 5 Max i

! ^ *- ,

20 gl  :

i ::  ;

's. ! vo*) F55 '2 38 E .: ' t s. -

5' 2,g ATD(%) 76 94 71 82 69 93

^ E 10 E 3- --

gA'j 2- -

g,, .

.E' 'a , -g Depth % 100 0 97 0 99 0 ,

^ I

O  ! 0.0

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 Axial Distance (in.)

Chart 1 (6) Far(bc5) xis 9/29/97 3-11  ;

1

2 Figure 3-6: Leak Rate vs Bobbin Amplitude 7/8" Tube Model Boiler and Farley Data Only, Effect of New Farley Data 1000.0 _

. m_.___._._. -

o Model Bo.ler i Database .__ - _ _ _

__ m -

4 _. . . u .

__._-___- _ _ , _ _ .a _ ~______.___.__.a..a.

u_p_. . . .. 4. _

_. _ _ , _ . _ . ._1__ ._

4.,. 4._ i_.

o US (Farley) Pulled Tubes l4 i p._ + _ 7 . ,

_ _.___ 1 E

.X ! -

100.0 --- a Farley-1,2 Data _n x _

r tn r

._a.o J _.x 9 _ .-. 4_u x Retained Per NRC (MB) -~- _ __ '

~*~  ? -+ +-

o . _~_ ~_~ ~ ~- _ x x s-+ r-r.

n x Reference Mean(Q) --

,,j

- - X -- -f i " 4-4-~ Q 9l4 H-

.c x o +

a 10 0 ~

j *x

4  !

c J + Reference Median (Q)

_% _-x gx4 _0;4_gg___ _;=_ g rg y -

u . . . .

. y _. O q . . . o _ . g__- _ ,, -~ .- .y 7;- .

+g - e_ e

. _ 1 u

4 - p.

d .

x r-- -o r----- -o------- -----:---- ---

_ f ix3  ; D' ,g 0 ~+ ~ ~ ~ ~ ~ ~' ~4 ~ ~. L'

,x

.M d

. 1.0 x x

l m.-,

' .f+!+I!t

, l 0 o 2

i i

d __x .

z[

~

y _I" 4

i  : i j_

d Cg-_x <:- i o ~~T i i M.+

~

_. ,. I i r  !

/e /X w

_x___ .d +

i i

_L_

j

,*.'+ l

/a x*

l _a __ L_._

l 1

L x

+

d o l i I

,r-x -- -

lw l

^

+ Fd-n[x _,. . :['

! l ,

y._1 m4

,.s

+ .._

1 3_ ,

i

.- + ,

!l i i

.- +  ;

.l * . ,

t 0.01 0.1 1. 10. 100.

Bobbin Amplitude,V(Volts) >

d.\apc\78_ anal \780Vwcor [OvsV78wa>EDF_28.5v] 3-12 ROC W2W97.1214 PM 1

Figure 3-7: Leak Rate vs Bobbin Amplitude '

7/8" Tube Data without MB 542-4, Effect of Plants A2 & Al Data 1000.0 _ .1 - - . _ _ . . .. ._

r -

_ w.c 2_._ _: .___ :tr r t

o Model Boiler Database .___ _ .- _

n: ~~ tr 1. r 2:ci 12 ::= Ittt 7 ; .._ c

_._ _ w _q _ _ _ _ _ . _ _ _ _ _ . _ . _ . . _ _ _____.i__ u.._4 J a US (Farley) Pulled Tubes _ - _ .-

+_ ; ,_J Q_

j -

.x: f

[

100 0 -E = Farley-1,2 Data 3 l ..

y_ -

4. 4 _%

. 6

- . _ _ . - _ - _ g ..pVM_1_m A_-4 4

m . _ m__ . _. . - _. a x__. _

A EdF Data

r. p'y_.g".

-- - ~~ - -

x

,xo _, i

_t.

~

p- , ._.1_

_-_x 2_. ._._ O __ ___1 ,_+._.1_ b}

24 x Reference Mean(Q)  !

h x o

- -f ' # e. !

10*0 - = ^

=

G- + Reference Median (Q)

---x - - &Q w. 'L , - + .

o; ,_o_._.____..-.

4.. g+. . o--+. . . _ .. _

_x

._ -.-..-_3 ._ .

t #

g

-x . -._

_x-x D _ __._o-s *t_,___ , +s'_

9

s. o - --- . - - . _ -. - .- . - a _._

~

s

_x ___, ._ _._. _ . _ _ _. _ _.l._ c 1.

N x . . .

x O

sh[t+

M 1.0 ~

= l O l e _ ____ , - . 1 a - .

_/_.x_ _x_x

. > u_ __ m

.l_ o ___ _ _ _ _ _._ __j _ t

._ m_4._

,{ 4 ,+

,  ; y _.p

_n_._ ~ -

/ ,x

_ _ /+

,y

,I 6

2.___a__> i a

l

_i,._

0.1

/,x *

- +

, l l ! l, ! a f l;  ;

f ,

/ x e+ '>

gi  ; 10-+ _t _ _ r;;.

x

. . . i t

_x , -i ,

H -_ r - - -- ,

Y e l

______i__1_ L ._

4_ - _ -

a n, ,

4  ;

q _L. .

/._ _

4 _. .

, j i } . -

+ -

, i  ! i !

0.01 2

0.1 1. 10. 100.

Bobbin Amplitude, V (Volts) 780Vwcor OvsVw-oS42-4 9/2919712.07 PM 3-13 RFK: 9729/97.12 07 PM 1

l

, 4.0 ANALYSIS METHODS EVALUATED The current NRC approved analysis methods and database result in a SLB leak rate correlation

, independent of voltage and a correlation that t.iases the leak rate parameters to the high side. A 4

recommendation to utilize unbiased leak rate parameters was submitted for NRC approval in Reference 7.3 and is currently under review. The analysis methods for biased and unbiased parameters are briefly described in Sections 4.1 and 4.2. NRC review of the unbiased

- methodology has identified an alternate method for consideration tnat utilizes a " boot-strap" approach based on using the direct leak rate measurements with an adjustment for the high leak rate tail. This methodology employs the empirical leak rate parameters with an exponential tail fit that adjusts for the possibility that the largest leak rates may not yet be included in the measurements. This empirical method is described in Section 4.3. As noted in Section 3, a voltage dependent SLB leak rate correlation can be obtained by any one of three paths: 1) Using only the five Farley pulled tube leak rate measurements together with the model boiler data: 2)

Excluding the French data from the ARC database; and 3) Applying the revised EPRI Data Exclusion Crite-ion 3 to the database. These alternatives lead to SLB leak rate analyses applying a voltage dependent correlation. The ahernate analysis methods, three variations without a voltage dependent correlation and three variations that result in a voltage correlation, are described in this section and analysis results are given in Section 5.

4.1 WCAP 14277: Blased Leak Rate Parameters, No Leak Rate Correlation The currently approved leak rate simulation methodology, Reference 7.8, is based on the statisti-cal test result that a null jypothesis that the logarithm of the leak rate test results is distributed normally is not contradicted. Thus, ahhough the actual distribution is still unknown, it may be considered that the leak nte test results are distributed log normally. The original reason for investigating the distribution of the log leak rates was based on the acceptance of a valid liriear "

regression of the log-leak rates on the log bobbin amplitudes for 3/4" diameter steam generator (SG) tubes For the linear regression the residual errors associated with the predicted values of the log leak rates were verified to be approximately normally distributed. In addition, current guidelines indicate that parameter uncertainties should be accounted for. If a theoretical '

distribution is postulated, the uncertainties in the parameters become relatively easy to simulate.

Because the distribution of the log-leak rates is considered to be normal, the simulation of the leak rates proceeds as follows:

1) For one simulation of all of the indications in a SG the population parameters of the log leak rate distribution are simulated. The population variance, cr 2, is estimated as a random function of the distribution of number of data points minus one over x 2 corresponding to the degrees of freedom for the variance (number of data points minus 1). This estimate is possible because of the assumption that the log-leak rates are normally distributed. The same estimate is not valid if applied to the leak rate data because the leak rate data are not considered to be normally distributed.

2) Given the population variance, random values of the log-leak rate population mean, p, are estimated as normally distributed with a mean equal to the data mean, m, and a variance equal to the population variance divided by the number of data points, e.g.,29.

,wn ai, w, .,,.o. ut i ,

4-1

3) The leak rate for each indication that leaks' in the SG is then simulated as randomly following a normal distribution with a mean of p and a standard deviation of c. The sum of the leak ratesof all of the indications that leak is calculated and retained.

4) The simulation steps 1) through 3) are repeated many times, e.g.,250,000. The leak rate totals are arranged in ascending order and the total appropriate to the level of confidence desired is identified.

The results obtained from this process are significantly biased to yield leak rate values exceeding those that should be estimated from the SG tube indications. If the mean and standard deviation of the population of log leak rates are known, values of the mean and standard deviation of the population of leak rates are given as, pg =c Y , and o* = a g pg (,,i , g ) , (4,11)

It is known that if the mean and standard deviation of a sample of log-leak rates, m and s respectively, are substituted for p and o, the resulting estimates of the parameters of the distrib ,fleak rates is positively biased (always larger than expected). Since c2 s simulated 2

..s being distributed as ,2 (n /)/x , and p is simulated being normally distributed with a mean of m, the Monte Carlo simulation of the individual leak rates is distritsuted with a mean and standard deviation ofleak rates greater than those from Equation (4.1-1). This is because the mean of the 2

distribution of(n 1)/x is greater than unity. The key point is that a positively biased distribution of leak rates is being simulated, not an unbiased distribution of leak rates, and that bias is significant.

4.2 Unbiased Leak Rate Parameters, No Leak Correlation ~

Two approaches may be considered to obtain unbiased estimates of the leak rates from a lognormal distribution. The first is to consider m and e s as o unbiased estimates of the leak rate, and then use, 2

' '2 m =In #

,and s =ln I +1 .

(4.2 1)

}sf + ml , .r *0s to obtain normal distribution parameters of the log-leak rate which when simulated will retum a distribution of leak rates about the parameters of the test data.

The second approach, which has been previously recommended for use, is to obtain unbiased estimates of the parameters of the distribution of the population ofleak rates, p and a from the

' Whether or not an indication leaks is determined by sampling the probability of leak correlation.

- , .,<a , m - 4-2

• i log leak rate sample parameters m and s. These are then used in Equation (4.2-1) instead of m p

and sg to. obtain normal distribution parameters of the log leak rate which will return a distribution of leak' rates about the unbiased estimates of the leak rate parameters. Note that

' neither of the approaches discussed in this section are aimed at leading to best and/or unbiased estimates of the parameters of the log leak rate population. The intent of the transformations is only to obtain parameters which when simulated will return a distribution of leak rates about the observed leak rate parameters. The rationale for recommending the use of the second approach is that the efficiency of the estimates of the leak rate parameters is improved. In Section 5, calculated 95% confidence values for the leak rates based on the second approach are compared with those based on the biased leak rate parameters.

4.3 Empirical / Exponential Leak Rate Simulation, No Leak Correlation A third approach is to simulate the empirical data directly, without consideration of the underlying distribution of the data. Tais approach does not result in a simulation of the parameters of the distribution of leak rates, because the parameters of the distribution-are not directly used. Conservatism in the analysis is achieved because the upper tail of the simulated d;stribution is longer than the tail of the empirical distribution.

Introduction The simulation is based on using a mixed empirical and exponential distribution model of the leak rates. The methodology is described in Reference 7.9. Reference 7.10 also' provides worthwhile information for simulation methodologies. The method is similar to the bootstrap method of estimating the parameters of a distribution in that the empirical distribution is repeatedly sampled with replacement to obtain leak rate estimates for individual indications.

However, in order to account for variation in the upper tail of the leak rate distribution that rnay * '

not be reflected in the data, the leak rate distribution is simulated as the empirical distribution for the first n k data points and an exponential distribution fitted to the last k data points where n is the total number of data points obtained from the testing. The parameters of the mixed distribution, i.e., the mean and k, are calculated to have the same mean and variance of the stepwise cumulative distribution 2

function (CDF) of the data. For the simulation, a uniformly distributed random number (URN)is generated and the inverse of the cumulative distribution of leak rates is calculated to obtain a leak rate.

The Empirical Model The base model for the stepwise, or discrete, cumulative distribution function (CDF) of the leak 2

The word random is taken here to mean pseudo-random, the term applied to computer code generated estimates of a sequence of random numbers.

-m.e, .,m , nu-, 4-3

rates when ranked in ascending order is given by, F(X,) = 1, (4.3 1) n where iis the index number of the value of the leak rate for which the CDF of the leak rates is wanted. It is assumed that F(0) is zero so i ranges from 1 to n. For the leak rate data, n is 29, hence each increment of the CDF corresponds to a value of 1/n or 0.0345. For example, the twentieth value of the leak rate, Xw, corresponds to a cumulative probability range of 0.655 to 0.690. The leak rate for any indication would be calculated by generating a random integer between 1 and n and returning the corresponding value of the leak rate. For example, a random number value of 0.73 would result in an integer value of 22, thus, the random leak rate would be Xn. The value of the integer is found as the truncation of 0,73 29 plus I so that the leak rates range from the 1" to the n* values of the leak rates.

In practice, the CDF may be piecewise linearly interpolated to obtain values of F corresponding to leak values, t, between the test data points. This is accomplished as, (t-X,)

F(t) = -i +

(4.32) n n ( X, , , - X, ) ,

where i ranges from zero to n l. The second term in the equation is simply the CDF incremental probability times the ratio of the leak rate increment from i to the first test leak rate less than t, i.e., X, , to the leak rate increment from X, to X,., .

in order to obtain a random leak rate from the empirical CDF, a random uniform number, U, is calculated and the index number, i, for the corresponding leak rate value less than or equal to a cumulative probability of occurrence of U is found as U n. The leak rate, t, is then interpolated between the P and P plus 1 value, t = X, U1n, (X, , , - X,) . (4.3-3)

The drawback to using only the empirical distribution of the leak rates is that a leak rate larger than the largest test value will rever be simulated. Each leak rate in the database will be simulated 1/n.100 percent of the time. Computer generated random numbers, U, usually lie in the range of 0<U<!, hence, the probability of occurrence of the exact largest leak rate will be 1/n minus the smallest significant value repre:ented in the computer. However, in operation, leak rates larger than the largest sample leak rate could occur. To simulate this eventuality, one option is to simulate the first n k data points as an empirical distribution, and fit an exponential distribution to the last k data points to simulate the tail of the distribution. The data are assembled in ascending order and a piecewise-linear CDF is fitted to the first n k observations.

A sluftea exponential distribution is fitted to the right of observation X,,., with a mean chosen so that the means of the overall fitted distribution and the sample are the same.

. um,,mvieww, .,oo. . io. - 4-4 '

~

, Fitting of the Exponenti:1 Distribution The exponential distribution is a one parameter function with a CDF, F(t), given by, F( r) = 1 - e -"' , (434) where B is the mean of the distribution, estimated from the mean of the data, and I is any number greater than or equal to zero. The maximum value of F is unity and occurs for a value of : of positive infinity. The minimum value of F is zero corresponding to a value of f of zero. The value of the variate corresponding to a random value of the CDF of U is given by, t = - 0 in(l - U) . (4.3 5) r Thus, if the mean is known, or estimated, random values of the distribution can be generated from URNS However, it is desired to only fit the tail of the empirical distribution, i.e., the last k data points, to an exponential distribution. This is done so that the mean of the total simulated distribution is equal to the mean of the empirical data. Reference 7.9 provides the expression to be used for the parameter of the exponential distribution as, 0=I "[ * ..

+ f. . i (X, - X, .,)

(4.3-6) k .

Using the parameter of the exponential distribution as given by Equation (4.3 6), the tail of the leak rate distribution is then simulated exponentially as, '

t = X, ., - 0 in ". ( 1 - U) , if U > 1 . . . (4 3-7) k n and empirically as, t = (n U - i)(X,,, - X,) + X,, otherwise, (4.3-8) where i is found as the truncation of nU. The distribution tail, i.e., the exponential part, is sensitive to the number of data values used to find the exponential distribution parameters. This is illustrated by the tail curves on Figure 4-4, discussed later. When k is equal to 4 the expo-nential distribution is used for probability values equal to or above the probability associated with the 25* data point because n-k is 25. Thas, the last k+1 data points are actually used to fm' d the exponential tail through the determination of the value of 0. The lar;,er the value of k, the greater the probability range over which the exponential distribution is used. However, the rate of decay of the distribution is also greater, hence the length of the tail of leak rates is decreased relative

• , using lowe.r values of k. This is a result of the requirement to maintain the same mean for the c mb;oen -  : -1 and exponential distribution as for the original data. The decision of which

(* . - w 1 e w ta g., +.re ~g. w m u 4-5 b d)

t

?

_ , value of k to use is based on seeking the best match of the variance of the mixed distribution to

, , the variance of the data, Reference _7.9.

4

, Shifting of the Exponential Distribution In order to determine the appropriate expression for the exponential tail to be substituted for the last k data points of the empirical distribution, we note that if Equation (4.3-4) is written as,

,.x F(t) = 1 -Ie T, -

(439) n 1

that the maximum value is still unity as ie, however, the minimum value is one minus k/n when t is equal to X instead of being zero when t is zero. In essence, X is a shift of the distribution in the i direction with the value of k/n determining the corresponding value of the

. cumulative probability when t =:X Thus, by shifting the distribution the minimum cumulative probability of the exponential tail will match the cumulative distribution of the empirical values at the k* data point.

I The next step is to find a value for 0 that results in the average leak rate from the piecewise, continuous mixed distribution matching the average leak rate from the stepwise empirical distribu-tions If the data are plotted in ascending order with the abscissa representing probability or sample number and the ordinate representing the leak rate, Figure 4 2, the stepwise cumulative distribution function is visualized. The mean, m, of the stepwise distribution is found as I

t a y es y .

y 4

m={J={J+{,.i n J,

... n . .. .. . i n (4.3-10) <-

I 4

-where the X, are the individual measured leak rates. However, the mixed model considers the empirical distribution of the first n k data points to be piecewise continuous based on a linear

' interpolation between the data points. Hence, the mean from the portion of the mixed distribution

' -to the left of the n k* will be less than that of the stepwise distribution to the same data point.

' - Since it is the mean of the overall stepwise distribution that is matched, the exponential tail of the mixed distribution will be longer than if the entire distribution was simulated as piecewise

).

continuous. This is evident from the depiction of both distributions on Figure 4-3. In this case i the empirical model is used up to the twentieth data point or 69% of the distribution is modelled empirically and 31% is modelled exponentially. The curve illustrated on Figure 4-3 is based on I

k being.9. The effect of matching the mean of the mixed distribution to the stepwise mean of the data is apparent from Figure 4-3 because the leak rate for cumulative probabilities above 70%

is larger than would be obtained from,the data. This is especially true for cumulative probabilities above about 93%. Figure 4-4 illustrates several solutions corresponding to different values of k. The variance of the distribution of all of the stepwise leak rates is 1195.1Iph3 .

Setting k to 9 for the mixed distribution results in a variarce of 1194.81ph2, in effect identical to the variance of the data. . The variance changes by me,; 'han 120lph2 when k is set to either 8 or 10.

- =.,.w m ,. o - 4-6 4

i d

-,- .--,+- , , . - . . - ~ , - -, ,-, ,_ , _

, n . . , _ - - - - , -

, Conclusions The simulation of the leak rate when there is no evident correlation between the leak rate and bobbin amplitude may be performed by simulating the distribution of the test leak rates with the tail of_ the distribution modified to simulate an exponential distribution. The form of the exponential distribution is selected to match the mean of the sample distribution and, as much as practicable, the variance of the sample distribution. The use of the current method of simulating a normal distribution of the log leak rates results in grossly biased estimates of the leak rate distribution and the 95% confidence level of the total leak rate.

4,4 SLB Voltage Dependent Leak Rate Correlation As given in Section 3, a correlation of _SLB leak rate with voltage that satisfies the NRC statistical requirement for a p-value < 5% can be obtained based on industry recommended

' changes to the current database with or without inclusion of the latest Farley-1 and Farley 2 pulled tube data in the database. The recommended changes are: -

Use of only Farley pulled tube and Model Boiler data in SLB leak rate correlation. This results in exclusion of French data points R8C74 TSP 1 and R22C26 TSP 1 from the

, SLB leak rate correlation Exclusion of French data from the current database. This results in exclusion of two data 4

points from the SLB leak rate, thirteen data points from the POL correlation and exclusion of one point from the burst pressure correlation.

Application of revised EPRI Data Exclusion Criterion 3. This results in exclusion of Model Boiler specimen 542-4 from the SLB leak rate correlation. The correlations applied

! in this case retain the French data points.

The bobbin voltage for Farley-1 pulled tube R2C85 increased frca a pre pull voltage of 13.55 to a post pull voltage of 28.5 volts. The 15 volt change is the largest increase found between pre and post-pull measurements for any inoication in the EPRI database. Therefore, it is appropriate to consider including this indication in the SLB leak rate correlation at the post-pull value of 28.5 volts. The resulting correlation is included in the analyses of this report.

4.5 Voltage Dependent Growth Rate for Farhy 1 The need for including BOC voltage dependent growth rates in the ARC analyses was first identified at Braidwood-l (Reference 7.7) upon application of a 3 volt ARC repair limit. For BOC voltages abave about 1.5 to 2 volts, the frequency oflarge voltage growth rates was found to be significantly higher than found at lower voltages. The growth rate magnitude did not significantly increase with BOC voltage but large growth rates occurred at a higher frequency.

The method applied in Reference 7.7 to include voltage dependent growth rates in the ARC analyses was to develop multiple cumulative probability distributions over varying ranges of BOC voltages. With this method, the limiting SG growth distribution is enhanced for large growths by assuring that the largest three growth rates in all SG3 are incl:ided in the distribution. For q WWNspenWIMky wpf Ckteer lik ten 4*7

-~

application with a voltage dependent POD, a growth distribution is developed for a minimum of

. the largest 25 BOC voltages but with a voltage range that includes at least one of the three largest growth rates found in any SG. The ext voltage bins for growth rates must include at least 200 indications. When applied with a PG) of 0.6, all voltage growth bins must include at least 200 indications or the methodology becomes excessively conservative.

Figure 4-5 shows a scatter plot of voltage growth versus BOC voltage as reported in the Parley-1 90 day report (Reference 7.1). This figure shows the one large growth rate for R2C85 in SG C but the next two largest growth rates include one each from SGs A and B. As discussed in Section 3, there is a possibility that the large growth of R2C85 could have been affec'ed by pressure pulse cleaning (PPC). The next largest growth rate for R2C77 in SG A of about 3.2 volts is at a tube location in the PPC pulser zone and could also have been affected by PPC.

Except for these two indications, the growths rates are small and < 2 volts. The data of Figure 4-5 show only a modest increase in frequency of large growth rates above about 1.3 volts.

However, to conservatively assess the potential effects of growth dependence on BOC voltage, the guidelines of Reference 7.1 were applied to the SG C data of Figure 4 5. The resulting growth distributions are shovin in Figures 4-6 and 4-7 for application with a voltage dependent POD and with a POD of 0.6, respectively.

~w."vuu, .,<m., m ,,, 4.g

,- . , i' t ' i' i! ,! t' . j

- 0

.:,' . !il 0

1 M

. ,l i l.+I , A 3

. 1, l ! ti ,4I  !

.!! . *,ji .t t 2

!I jj, ft  ;

1 1

{,l l ii  ! .! ' . ' 4 .i ,.!iii  !

7 ji .f+ > 9

/

9 2

/

l ,  !

.!} ' ;j- .jI , iil l ,.f,  ;} .

9

^ K F

R l  !  ; - .!! . :i o l r!. i o

!,l I- l

• . t ' l I' . '

d e o d

u 'A

.- I il. ! 1 ' -

- l e c o d In n u

t a o o i t I;.!t l -

l ' .h.

p a

. t .' . +j l l i! . ' ; .

o )

s mD A l t

o AFd o o o V

(

i nE o o V b ,

b a e ot o d BaD

)

0 t

u s .:.l 'i ; l ji. * , . ,

1 i l

vet

. i  ! I 4 . .!' . l p

.: : ' o e a < l ; ' ' .  ! : . i .l ;t

.i .  !~ m t

aR  ! ,l f1 l l ; il I .!  ! .

o . : J.

o- A n

ij!

t + f;< }

'

Rk o D. - i k a . jI il l ' j , -

. . .> b e

4! j,s' ,  !

b aL o e ..,;+ j: ' . ;:{e iI B LC r :-i I+ . ;I' 1

C )

S ,.

e B  !

.: I . .; . +}!  ;.f,t; s

4D e a M

(

rO us t b

t a C ge a s R e I . ' , ,; :1 i l  ! rt . ' . e- l' ib F u DbeN u r t

a =

o-r eT e P R T

l k -

d o ed t a ae

_ i 8" .

Bl u enDaL

. ;; . i {l l ,ii i'

/ l 7 l n

deP i

t aF a -

]

a oS ed e t a

MUREM D Q

[

is

. m o o o a i v

v 1

. q

- 8 1 7 0 0 0 0 1 1 t

la 0 0 -

0 n 0 1 0 0 0 1 0 a_

0 1 8 7

1 t c

p a

T 2e.N'@ u3Z4dj' t

c 4 d

P Figure 4-2: Distribution of ODSCC Test Specimen Leak Rates 180.0 IP7Am 7/8" ODSCC Test Data I60.0 L-

---

Mean Leak Rate r l

140.0 _

1 l

1, 120.0  !! _

Ea C 100.0 $l

$e I M

.X 80.0 e -

60.0  ; p-

/ l ll .

l  !

ll' l

40.0 ' '

l 20.0 _ . . _ . . _ . . _ . _ . _ . _ . . _ . . _ . . _ . . _ . . _ . . _ _ . _ . _ . . _ . _ . . _ . _ . _ _ _ _ . ..

r r hr -

0.0  :  :  :  :-:-:  :":":":":" ":E:E: #: E: 5: : : : : ': ': : : : :

0 1 2 3 4 5 6 7 8 9 10 11 52 13 14 15 16 17 18 19 20 31 22 23 24 25 26 27 28 29 Sample Cumulative Probability d unarst\ alloy 600\78qvboot xts [O Hest) 4-10 RFK. 9/29/97.1123 AM 1 ,

Figure 4-3: Bootstrap Simulation CDF of Leak Rates for 7/8" Tubes 300.0 Selected Mixed Model to be Simulated 4

-*- Raw Data (Piecewise Continuous)  ;

i i i 250.0 Exponential Tail  ! 4 i

j i

}

4 , 4 .

s*%

.c '

i , -

9 200.0 ,

' b s- _

I 3es  ;

i M

• u '

a

$ 1E0.0 '

-a i

, 1 -

# i g  !

I

.S a i i p -- i l I

$ j j f i f f 5 100.0 ,' I I '

C l l '

+* { l  !

• ,/ I i l I l'

~ -

l ,f x

50.0  !

-- ' l' #

' 1 l t i i / ,

l l , ,A '

..a-.. .l l ,8l .

^ ^ ^

r, _ l 0.0 v a a a i ^ ^ ^

^^^- ^

~ #- ~

~!

0% 10% 20% 30% 40'i 50% 60% 70%. 80% 90% 100%

Cumulative Probability of Occurrence (Sample) d WC8_anan78qvboot zis [O Cum]

• 4-11 RFK: 9/29f97 1124 AM 1

Figure 4-4: Bootstrap Simulation CDF of Laak Rates for 7/8" Tubes Effect of the Number of Data in the Exponential Tail i  :

i i  !

-*- Raw Data ' '

1 350 C

Exponential Tail, k = 9  !

4

+ Exponential Tail, k = 4 -

r i i l 300 o Exponential Tail, k = 2 There is a balance that takes place 1

---

Exponential, All Data between the extreme values in the 2 -

tail and the starting point for the 3  ; { .

5 250 tail. Lower values of"k" lead to '

'q g

longer tails over a shorter range of o

. probability. The best fit curve in  ;

j i this case is for k=9, matching the ,

j 200  ;

variance of the original data. ,

o

.5 150 I  !

1

\ m

+

9 1 i

c t ,

+

t o I/  :

100 '

i. .

^

i j i,

( /

/

! l l ,_-

'~~~,

50 I M . - -

.......a......... ..-...- - --- w

_ _f. - a __-l ,

~

70% 75% 80% 85% 90% 95% 100 %

Cumulative Probability of Occurrence dDpc\78 anali78qvboot mis [O cum Ks 2] 4-12 RFK 9/29/97.1126 AM 1

m

t w,  :  : I  : ' '  ! i ti! l  ! il; t i!'L f t II i,ll i

bI l :t

- 5

. _ 4 A B G G -

. S. S - -

U i n -

- h -

_ a c t B G i

o w -

1

& _ i n

r G. - - 3 A i s_

mt s

e _ - _

g _

G _ i a

r -

S __

a.-

i n _

L s t _

a_ . -

ht - - - - _

w o b r i 5

G _. .

l 2

t _

s .

e -

g _ _ _.

r _

L a _

s e u gl t

aP uC- -

r a

OG _

- 2 e

7 9 S g

9 ro _. a 5- 1 f

_ - l t

_ - o 4 I e

g V e

r 4 3

1

. a. 1 -

u g n 4 i

u o C-

. F 4 O i

t 1

_. i1 ' 5 B n

l!

C- - 1 U

yO o I

l eH A #

r s a i F4 1 l

e

- c y

. C I

f. I g

i n

r u

D ht

. w o

r _

G e 5

e g 0 t

a l

o V

u r

u s

n O v 0 9 8 7 a

1 6 5  :. 3 2 g U g n

r. e1U aIsc $5EC *y o>

m w

d Fige.e 44 Farley Unit -1 Cycle 14 Growth SG-C CPDF Distribedienes Used for Projecting EOC-15 Resents with POPCD

o . _ _ _ _

. _-  ; . _ .p,. ._... - - *

-=- - - g - - - -

, _m.---o .+ ---

0.9 -

y-s jn . .... *. .

/

s

, -- .d ~ _ - -

g .

/ -

/ Y ~

0.8 -

K *

/

I f -

a 1 -

.."4 f 0.7 -  !  !

• f..

g I

~

/

a i  :

sa.  ; I

• a 0.6 - , ,

Ea r -

l 1 -

.D / .

• 3 0.5 - -

f ,.  : Up to 13 voit (922 Inds.)

/

~

5 0.4 - ,#-

.- - e-- 13 to 1.8 volts (l89 inds) - --

2=

6 .

E U

=03- ,

.+ --+ -Over 1.8 voit; (31 Inds.) - - - - -

+~

0.2 - . . - *- SG-C Only (All Volts) - _ . - -

p  :

./

• 0.1 - .' . - ---- - -

l l -

~

0.0 " . -~ '~~~.4 " " " " " " " " "

,. - o - .". , e . - ." . e. , -, -

9 9 , o o o o o o o o o - - e. e Veitsge Growth

~ 4-14 1

- - _ -. . ~

Figure 4-7 Farley Unit -1 Cycle 14 Growth SG-C CPDF Distributions Used for Projecting EOC-15 Results with a POD =0.6 1.0 -

-=- - - = -- - = , , , , , g. !- - . . x. - - - - x - - - - - - e - -

,- _,y.--- -x-Y ,x*..-

1-o_9 - s _ - . . .

/ *

/ ,=

0.8 -

f t .

a i .

• 0.7 - I -

~

- - - - 2-

.% I .

c .

a 1 .

k ca 0.t- t i 1 .

n i .

a .

a i .

I 0.5 -

I --+--Up to 13 volt (922 Inds.)

c .

e  :

.t- 0.4 - .- - - -

j ,

---x --Over 13 volts (220 Inds.)

E .

= 03- -

U .

~

~

0.2 -

- e- SG-C Only (Ali Volts)

~

0.1 - .- -- - - - - - - - - - -

..K 0.0 - ' " - " ~ " " - - ^ -

a. - o -- a. - , - e n e ., n -

9 9 o o o o e o o o o -- - ,-

!, Voltage Growth

.4-15 1

a 5.0 SLH LEAK RATE AND TURE HURST PROHAHILITY ANALYSES 5.1 Comparison r,f EOC 14 and HOC.15 SLH Leak Rates for POD = ti.%

When eva. ding cycle lengths to reduce SLB leakage, it is important to note that a very conservative POD such as a constant value of 0.6 can have as large an influence on leak rates as the cycle length. For Farley 1, this is shown in Figure 51. The effect of tube plugging at EOC 14 is to reduce the SLB leak rate (calculated with database including recent Farley pulled tubes) from about 9.9 gpm to about 7.1 gpm. After a POD = 0.6 is applied to define the BOC 15 voltage distribution, the BOC leak rate is about 13.5 gpm (3.6 gpm higher than before plugging) or an increase of 6.4 spm from only the POD application. After the POD application, the increase in leak rate over the operating cycle from BOC 15 to EOC 15 is only 6.9 gpm compared to the 6.4 gpm POD effect. Consequently, shortening the cycle length can have only a modest reduction in leak rate for a constant POD. It is therefore the goal of this report to demonstrate that high leak rates predicted for Farley 1 are the result of unnecessarily conservative databases and analysis methods.

The use d more realistic databases and methods leads to acceptable leak rates such that there is a negligible risk of exceeding allowable leak rates after a full cycle of operation. For example, application of the POPCD voltage dependent POD results in a 1.1 gpm (1.7 gpm less than before plugging) increase from the POD effect compared to the 6.4 gpm found for POD = 0.6 and an ,

operating cycle increase of about 4.7 gpm compared to 6.9 for POD = 0.6.

5.2 Analysis Results for EOC 14 As Measured Voltage Distribution SLB burst probability and leak rate analysis results for the EOC 14 measured voltage distribution are given in Table 51. Analysis results are given for the current ARC database and the database "

updated to include the recent Farley SO pulled tubes. In general, the updated database significantly increases leak rates and moderately increases burst probabilities.

5.2.1 SLH Hurst Probability at EOC 14 For the measured EOC 14 voltage distribution,it is seen from Table 51 that the updated database only increases the burst probability from 3.8x10d to 5.67x10d using current analysis methods and excluding the indication at R2C85 from the burst analyses on the basis that this indication was pulled and burst tested with a resulting burst pressure well in excess of SLB conditions. The strong influence of pulled tube R2C85 demonstrating a burst pressure in excess of SLB conditions is seen by comparing Cases 1 and 2. Exclusion of R2C85 from the burst probability analyses 4 4 results in more than a decade reduction (5.2x10 to 3.8x10 )in the burst probability. Thus, the 4

EOC 14 burst probabilities are less than the 10 burst probability, NRC reporting guideline for both the current and updated databases.

From case 6, it is seen that exclusion of all French data from the burst correlation, as recommended in Reference 7,3 and currently being reviewed by the NRC, results in a burst 4

probability of 4.51x10 with the updated database which is also acceptable and slightly smaller than obtained with the updated database,

,-w , ,,,m, m - 51

The 60C-14 burst probability is thus acceptable for the cunent and updated databases.

5.2.2 SLB Leak Rates at EOC.14 From Table 51, the SLB leak rate is increased from 7.9 gpm to 9.9 gpm for the updated database when the cunent ARC analysis methods (biased leakage parameters with no voltage dependent) and the measured leak rate of 0.72 spm is applied for pulled tube R2C85. This leak rate is less than the site allowable SLB leak rate of 13.7 gpm (room temperature gpm used for both calculated and allowable gpm). Additional analyses described below were performed to demonstra!c the reduction in predicted leakage for improved database assumptions and analysis methods.

The NRC is cunently reviewing a recommended (Reference 7.3) change in the analysis method from biased to unblased leakage parameters when there is no voltage dependent leak rate correlation as obtained with the current and updated databases. Case 3 shows that the use of unbiased parameters reduces the leak rate to 7.6 gpm. An additional method being evaluated (See Section 4.3 for a methodology description)is the direct use of the measured leak rate data in an empirical method with an exponential adjustment to the high leak rate tail to recognize that the largest leak rate may not yet have been measured. The EOC 14 leak rate using this methodology (Case 4 of Table 51) is reduced to 4.5 gpm.

As previously discussed in Sections 3 and 4, it is appropriate for a Farley SO assessment to use I

only the five Parley pulled tube leak rates with the model boiler data in the SLB leak rate correlation since the Parley data are the only domestic pulled tube indications with measured leak rates. With use of the Parley only data, a voltage dependent leak rate correlation satisfying NRC statistical guidelines is obtained. From Case 5, it is seen that this results in a predicted EOC 14 SLB leak rate of 2.1 gpm which is judged to be the best estimate of the leak rate. Case 8, based * '

on applying the revised EPRI Data Exclusion Criterion 3 which also leads to a voltage dependent leak rate correlation, results in a further reduction of the leak rate to 1.6 gpm.

Overall, it is concluded that acceptable EOC-14 leak rates are obtained using the cunent or updated database and condition monitoring requirements are satisfied. Revisions to the database and methodology currently under review by the NRC would further reduce leak rates and increase magms against the allowable limits.

5.3 Projected Leak and Burst Results at EOC.15 SLB burst probability and leak rate analysis results for the projected EOC-15 voltage distribution and operational assessment are given in Table 5 2. Analysis results are given for the current ARC database and the database updated to include the recent Farley SG pulled tubes. As found for the EOC 14 analyses, the updated database significantly increases ieak rates and moderately increases burst probabilities.

5.3.1 Projected SLB Burst Probability at EOC 15 As given for Case 1 in Table 51, the updated database combined with the reference analysis methods increases the burst probability from 9.92x104 to 1.21x10'2 From Case 2, it is seen that

-we , .,m , m -

5-2

O the burst probObility exceeding the NRC reporting guideline of3 10 is due to the assumption of POD = 0.6 independent of voltage which results in 0.67 of a 13.5 volt indication left in service at BOC 15. With the still very conservative assumption of a POD of 1.0 only above 10 volts so that very large voltage indications are not left in service as undetected indications, the burst 4

pmability is reduced to an acceptable 8.02x10 . The application of an expected POD such as the EPRI POPCD of Reference 7.3 further reduces the burst probability to 5.51x10 4 . Exclusion of the French 4 data from the burst correlation slightly increases the burst probability from 5.51x10 to 5.87x10' (Case 11).

It is concluded that the EOC 15 operational assessment, burst probability is acceptably less than 102 NRC acceptance guideline for both the current and updated databases when any realistic assumption on detection of large voltage indications is applied.

When growth rates dependent upon the BOC voltage are included in the analyses with the updated database, the burst probability for POD = 1.0 above 10 volts is increased from 8.02x10 4 to 1.37x10 (Case 8). Inclusion of the EPRI POPCD in the analysis with the updated database (Case 9) reduces the burst probability to 7.93x104. The structural requirements for the operational assessment with the combined updated database and voltage dependent growth rates are then satisfied with application of a voltage dependent POD.

5.3.2 Projected SLB Leak Rate at EOC 15 When the current SLB leak rate methodology with biased leak rate parameters and no leak rate correlation is applied, the updated database increases the leak rate from 15.7 to 20.4 gpm as shown by Case 1 in Table 5 2. Both leak rates exceed the Farley allowable leak limit of 13.7 gpm. However, this large leak rate results from an excessively conservative analysis procedure.

' Without a leak rate conelation, the leak rate is very sensitive to the an:Jysis methodology' " as shown by Cases 3 to 5 in Table 5 2. Use of the EPRI POPCD (Case 3) reduces the leak rates to 9.9 (current database) and 12.9 gpm (updated database). Use of POD = 0.6 with the unbiased leakage parameters (Case 4) and the empirical leak rate parameters (Case 5) reduce the leak rate for the updated database to 14.7 and 7.8 gpm, respectively. The 7.8 gpm value using the empirical parameters is less than the allowable limit. Based on the effects of POPCD between Cases 1 and 3,it can be expected that the use of POPCD with the unbiased leak rate parameters would also reduce the 14.7 gpm leak rate to below the allowable limit.

The best estimate of the EOC 15 SLB leak rate is obtained using only the Parley pulled tube and mod:1 boiler data for the leak rate correlation together with the EPRI POPCD. This database results in a voltage dependent leak rate correlation as also obtained by excluding the French data from the correlation. The resulting leak rate is only 2.3 gpm (Case 7). Even with a POD of 0.6, the Farley data result in a leak rate of 3.8 gpm (Case 6), When voltage dependent growth rate distributions (See Section 4.5) are applied, the 3.8 gpm leak rate for Case 6 increases to 4.5 gpm (Case 9) which remains well below the allawable limit of 13.7 gpm. Acceptably low leak rates are obtained as long as a voltage dependent leak rate correlation is applied. A voltage dependent leak rate is also obtained when the revised EPRI Data Exclusion Criterion 3 is applied to exclude model boiler specimen 54?-4 from the leak rate database. As shown by Case 12, this results in a leak rate of 3.1 gpm even with a POD of 0.6.

-.%m , .n.- m -

53

4

,' 5A Conclusions EOC H Condition Monitorine Assessment 1

d The burst probability of 5.67x10 calculated applying the updated database and the as measured voltage distribution for the EOC 14 condition monitoring assessment is acceptable and less than the NRC reponing guideline of 10 2.

l Acceptable EOC 14 leak rates are obtained using the updated database and condition monitoring i requirements are sa'isfied. 'Ite current methodology of applying biased leak rate parameters with '

the current and updated databases result in SLB leak rates of 7.6 and 9.5 gpm, respectively, which are below the allowable leak limit of 13.7 gpm. Revisions to the database and methodology currently under review by the NRC would further reduce leak rates and increase margins against the allowable limits. The best estimate of the EOC 14 SLB leak rate is 2.1 gpm using only the Parley pulled tube and model boiler data to evaluate the leak rate. These data result in a voltage dependent leak rate correlation and the resulting leak rates are small compared to the SLB allowable leak limit of 13.7 gpm.

EOC 15 Onerational Assessment The EOC 15 operational assessment burst probability for the updated database is acceptably less than 10~2 NRC acceptance guideline when any realistic assumptien on detection oflarge voltage indications is applied. Even for the very conservative assumptioni of POD = 0.6 for indications less than 10 volts and 4 POD = 1.0 for indications above 10 volts, the EOC-15 burst probability is reduced to 8.02x10 . The NRC is currenity reviewing the EPRI proposed voltage dependent POD based on Probability of Prior Cycle Detection (POPCD). If this POD is accepted, the EOC-d 15 burst probability is further reduced to 5.5x10 . When censervatively applying the constant '

POD of 0.6 with the4 updated database, the resulting burst probability of 1.2x10 is only slightly higher than the 10 NRC reporting guideline. When growth rates dependent upon the DOC voltage are included in the analyses with the updated database, the burst probability for POD =

4 1.0 above 10 volts is increased from 8.02x10 to 1.37x102 (Case 8). Inclusion of the EPRI POPCD4 in the analysis with the updated database (Case 9) reduces the burst probability to 7.93x10 . The structural requirements for the operational assessment with the combined updated database arf voltage dependent growth rates are then satisfied with application of a voltage dependent POD.

Applying the current analysis method with biased leak rate parameters and POD = 0.6 together with the current and updated databases results in projected EOC 15 SLB leak rates of 15.7 and 20.4 gpm, respectively, which exceed the allowable limit of 13.7 gpm. However, applying the expected EPRI POPCD rather than a POD of 0.6 reduces the corresponding leak rates to 9.9 and 12.9 gpm which are within allowable limits. Without applying a voltage dependent leak rate correlation, an acceptable leak rate of 7.8 gpm is obtained for a POD of 0.6 with a methodology applying empirical leak rate parameters with an exponential tail. An acceptable leak rate is also obtained without a correlation when the unbiased leak rate parameters are applied in conjunction with the EPRI POPCD.

Acceptable (< 13.7 gpm allowable leak limit) SLB leak rates are obtained whenever leak rates are calculated with a voltage dependent leak rate correlation. The updated database results in a w ma .,= m - 5-4

4" voltege dependent correlation when the leak rate correlation is based on the Farley pulled tube data (or the French data are excluded) and when the revised EPRI Data Exclusion Criterion applied. The best estimate SLB leak rate of 2.3 gpm is obtained using the voltage dependent leak rate correlation which results from using only the Parley pulled tube and model boiler leak rate data. The resulting leak rates are in the range of 2.3 to 4.5 gpm dependent on the POD assumed for the analysis and whether or not BOC voltage dependent growth rate distributions are applied.

Exclusion of the French data is currently being reviewed by the NRC and the revised Criterion 3 is expected to be submitted to the NRC for approval in the near term.

The following summarizes the database and/or methodology changes that result in acceptable SLB tube burst probabilities and leak rates when applying the current and updated ARC databases:

Burst Probability Application of conservative POD = 0.6 for indications less than 10 volts and 1.0 for indications exceeding 10 volts.

+

Application of EPRI voltage dependent POD based upon POPCD, When voltage dependent growth rates are included in the analysis, a voltage dependent POD based upon POPCD is required to obtain a burst probability < 10'2 SLB liak Rate Application of EPRI voltage dependent POD based upon POPCD when applying the current analysis method with biased or unbiased leak rate parameters and leak rate independent of voltage (no correlation).

+

Atplication of empirical leak rate parameters and an exponential tail with either POD =

0" or the EPRI POPCD.

Application of the EPR) recommended database that utilizes the Farley and model boiler leak rate data and results in a voltage dependent leak rate correlation (excludes the French "

data from the correlation).

Application of the revised EPRI Date Exclusion Criterion 3 which excludes one model boiler data specimen and results in a voltage dependent leak rate correlation.

g WWs9NepenWifulky opf ander HA 1991 55

~

~~-

Tcble 51,

, , SLH Leak Rate and Hurst Probability Analyses for Limiting SG C. EOC 14 Hased on Measured EOC 14 Voltage Distribution, POD = 1.0 Hurst Probability'" SLH Leak Rate gpm Case Method of Analysis Current Updated Current Updated Database'2' Databasei Database Database

1 Current WCAP 14277, Diased leakage 5.2x 10') 6. l i x 10'8 7.6 9.5 parameters, No leak rate corr.

2 Same as Case 1 except pulled tube

' 3.8x 10" 5.67x 10" 7.9 9.9 l data (no burst, leak rate = 0.72 gpm) used for R2C85.

Additional SLH Leak Rate Sensitivity to Methods and Database 3 Case 2 with unbiased leakage - - -

7.6 parameters, No leak rate correlation 4 Case 2 with Empirical leakage - - -

4.5 parameters and exponential tail, No leak rate correlaticn 5 Case 2 with use of only Farley pulled - - -

2.1 tube and model boibt leakage data.

Leak rate correlation ebtained.

6 Case 2 with exclusion of French data. -

4.51x 10" -

2.1 Leak rate con. '

~

7 Case 2 with exclusion of French data. - - -

1.4 Farley l tube at 28.5 volt, Leak rate corr.

8 Revised Exclusion Criterion 3 applied. - - -

1.6 MB 542-4 excluded, Two French data retained, Leak rate corr.

Notes:

1. Burst probability for one or more tubes.

2. Current NRC approted database without latest pulled tubes from Farley 1 and Farley 2

3. Updated database including Farley 1 and Farley 2 pulled tubes. Method of analysis column rnay modify this database.

==wa, -,, m in ie 5-6

' Table 5 2.

, SLH Leak Rate and Hurst Probability Analyses for Limiting SG C, EOC.15 (485.8 EFPD)

C i Hurst Probability" SLB Leak Rate . gpm l .a Method of Analysis POD i s e

Current Updated Current Updated Databasei8 ' Databasei8' Database'8' Database

1 Cunent WCAP 14277, Biased. No 0.6 9.92x10 3 1.21x108 15.7 20,4 leak rate conciation.

la Case I with all French data in corr. 0.6

• 8.06x 10 ' -

24.4 2 Same as Case 1. 0.6 <10v 6.64x10 ) 8.02x10$ 15.4 20.4 1.0 >10v 3 Same as Case 1. POPCD 4.66x 10'8 5.51 x10 ' 9.9 12.9 Additional SLB Leak Rate Sensitivity to Methods and Database 4 Case I with unbiased leak rate 0.6 - .

11.4 parameters No leak rate con. 14.7 5 Case I with empirical leak rate 0.6 . . -

7.8 parameters and exponential tail, No leak rate correlation 5a Case 5 with voltage dependent growth 0.6 . . -

8.3 Sb Casc 5 with voltage dependent growth POPCD - - -

5.3 6 Use of only Farley pulled tube and 0.6 - .

2.7 3 model boiler data for leak rate. Leak ate correlation.

7 Same as Case 6 POPCD - -

1.6 2,3 8 Case 6 for leakage and Case 2 for 0.6 <10v -

1.37 x10 -

4.5 burst with voltage dependent growth. 1.0 >10v 9 Case 6 for leakuge and Case 2 for POPCD .

7.93 x10'3 -

2.7 burst with voltage depender.t growth.

10 Exclusion of French data, Leak rate 0.6 9.63x10 3 1.28x108 2.7 3.8 correlation.

I1 Same as Case 8 POPCD 4.27x 10-2 5.87 x 10 l.6 2.3 12 Revised Exclusion Criterion 3, MB 0.6 - - -

3.1 542-4 excluded. Two French retained, Leak rate correlation.

Notes:

1. Burst probability for one or more tubes.

2. Cunent NRC approved database without latest pulled tubes from Farley l and Parley 2

3. Updated database including Farley.1 and 2 pulled tubes May be modified by Method of Analysis column.

o wwu , .e, m i ,

57

~ _ -

~

s Figure 5-1 Farley Unit -1 Sacani Generator C Projected EOC-15 Leak Rate - Updated Database that includes Farley I& 2 Data Applied 25 29 - -

'E - - - - - - - - - - - - - - - - - -- -

CYCLE LENGTH EFFECT .

_ 6.9 gpm E

@ l5 -

p A Anowable SLB leakage lmt ------------

S cc u POD EFFECT

}m 10 - (Li.wr==g Basis) 6.4 gpm I -

a u __

v3 dk a m EALISTIC R POD EFFECT (1.1 p) 7 g-- -_ _ ._

Cytte IS - N)i>=0 f.

5- -- - - --

X IiOC-14 Actual a BOC-15 (No N)D Adjuument)

A BOC-IS (N)It'D) 0 "-

l l l 0 100 200 300 400 500 Assumed Cycle Duration (EFPD) 54 s

.' 6,0

SUMMARY

OF SUPPORTING HASES FOR VOLTAGE DEPENDENT Stil LEAK RATE CORRELATION This section summarizes the supporting bases for applying a voltage dependent SLB leak rate correlation for 7/8" tubing. General engineering arguments for a correlation as well as database considerations are included in this discussion.

1.

Irrational Leakage Trends When Leak Rate is Independent of Voltage When the leak rate is applied independent of voltage, the leak rate of very small voltage indications is the same as that for large voltage indications, in this case, the distribution of

' leakage as a function of voltage is a function of the product of the number ofindications and the l

probability of leakage. Although the probability of leakage is small below a few volts, nearly all of the indications are in this voltage range and this range dominates the number ofindications predicted to leak. Figure 71 shows this product as a percentage of total leakage versus bobbin voltage for the Parley 1 actual EOC 14 and projected EOC 15 voltage distributions. It is seen that the dominant contribution to leakage occurs in the lower voltage range with about 60% of the leakage occurring for indications below 2 volts and about 85% of the leakage below 3 volts.

The lowest voltage indication found to leak for 7/8" tubing in more than 50 pulled tube indications was .t.8 volts which had a measured leak rate of about 0.08 liter /hr. Four pulled tube indications that had < 10 volt indications leaked at less that 3 liters /hr. However, the voltage independent leak rate correlation results in assigning these indications an average leak rate of 28 liter /hr with a standard deviation of about 186 liter /hr (biased correlation of Table 31). Th effect is that the large voltage indications which have and are expected to have larger leak rates dominate the voltage independent leak rates and these large leak rates are assigned to the 8$% "

of the predicted leaking indications below 3 volts, in this case, an occurrence of a large volt indication with a significant leak rate, such as the recent Farley 1 pulled tube indication, results in a large increase in the leak rate for the 85% of leakers predicted at < 3 volts.

It is totally against engineering logic that the expected large leak rates of high voltage indications would draniatically impact the total leak rate by assigning the large leak rates to the 85% of predicted leaking indications below about 3 volts. A voltage dependent leak rate correlation is essential to obtaining a rational leah rate calculation.

2.

General Considerations on Data Exclusions That Result in a Leak Rate Correlation The indications that prevent a statistically acceptable leak rate correlation for 7/8" diameter tubing are a 31 volt French indication that leaked at about 0.13 liter /hr and/or a 50 volt model boiler specimen that leaked at about 0.85 liter /br. Specific recornmendations to exclude these data points have been previously identified and are further discussed in item 6 below. The emphasis in this discussion is noting the irrational consequence that including these very high voltage indications with very small leak rates in the leak rate database results in assigning very large leak rates to low voltage indications. This inclusion of both indications in the database prevents a statistically acceptable correlation with the net effect of assigning large leak rates to the dominant population of indications below 3 volts as discussed in Item 1 above.

- w m ,., % ,m -

6-1

_ _ _ _ _ _ _ _ - _ --____-___ - - - - - - ~

In eddition, if these two indic:tions did not leak at all, they would not be included in the leakage

' database and a statistically acceptable conelation would result. Overall, the NRC requirement to maintain these indications in the database while applying statistical requirements on a p value for a correlation results in leak rate behavior inconsistent with physical reality.

3.

Comparison of 7/8" and 3/4" Tubing Trends of Leak Rate With Voltage For 3/4" tubing with a larger number of indications in the leak rate database, a statistically acceptable correlation ofleak rate with voltage has been obtained. These data demonstrate that leakage is conelated with voltage when sufficient data is obtained. For 7/8" tubing with a smaller number of leaking indications, the database is not large enough to overcome the impact of including the two high voltage, low leak rate indications in the database and satisfy statistical correlation requirements. The similarity of the leak rate data trends with voltage for both 3/4" and 7/8" tubing is shown in Figure 7 2 which overlays the data for both tube sizes. The 3/4" voltages have been increased by a factor of 1.36 to more closely approximate the 7/8" voltages.

The 1.36 factor is comprised of a product of two tube diameter ratio factors of 1.17. One factor results from using the same hole diameter for voltage normalization (rather than scaled hole sizes) between the two tube sizes and the second factor results from scaling crack length for equivalent structural response.

Figure 7 2 shows the data points recommended for exclusion from both data bases. The 7/8" exclusion recommendations are based on excluding the French data and application of the revised EPRI exclusion Criterion 3. The 3/4" points recommended for exclusion are based on application of the revised Criterion 3. The data points recommended for exclusion have leak rates well below the remainder of the database.

The data of Figure 7 2 show comparable trends of leak rate versus voltage for the two tubing '

sizes with the 7/8" data tending toward a more shallow slope than the 3/4" data. The fact that leak rates have been demonstrated to correlate with voltage for the 3/4" data and that the 3/4" and 7/8" leak rate data show similar trends with voltage support the application of a voltage dependent correlation for 7/8" tubing.

4. Significance of Different p. values Between 5% and 7.6%.

The NRC GL 95 05 requirement to apply a voltage dependent correlation is that the resulting correlation resuhs in a p-value of less than or equal to 5% for the slope parameter. The p value represents the probability of obtaining the observed results even if there is no correlation between leak rate and voltage. The 5% value is a traditional application of 95% confidence levels that there is a correlation for safety related analyses. As noted in Table 31, the inclusion of the latest Farley pulled tube data in the database, while retaining the two high voltage, low leak rate indications, results in a p value of 7.6% for a leak rate correlation. Thus, a leak rate correlation cannot be applied to the 7/8" leakage data. However, it is questionable whether a difference in p values between 5% and 7.6% is significant for application of a leak rate correlation. That is, the difference between 92.4% and 95% confidence that there is a correlation is not considere to be significant for leak rate applications. Statistics are controlling the leak rate issue rather than engineering logic.10 particular, this modest difference in p values should not be permitted to drive leak rate analyse to irrational results as shown in item I above and a voltage dependent

- w ,. % ,nu,a 6-2

correl: tion should be cpplied even if resulting in a p value of 7.6%.

5. Utillration of Only Farley Pulled Tube and Model Holler Data in Leak Rate Correlation As noted in Section 3 and Table 3.2, the use of only the Parley pulled tube and model boiler data in the leak rate database results in a leak rate correlation with a p-value of 2% which satisfies GL95 05 requirements for a correlation. The five Parley pulled tubes with measured leakage are the only domestic pulled tubes found to leak so that the resulting leak rate correlation reflects all domestic data. This selection of data is equivalent to excluding the French data from the correlation which has been recommended for other technical reasons as discussed in Referen 7.3. The fact that all domestic data supports application of a leak rate correlation for 7/8" tubing should be adequate to permit the use of a correlation.

6. Industry Recommendations That Result in a 7/8" Tubing Leak Rate Correlation As noted in Tables 3 3 and 3 3, exclusion of the French data from the leak rate correlation results in a p value of 2% or exclusion of model boiler specimen 542 4 while retaining the French data results in a p value of 2.5%. Both of these options result in an accel table correlation. Exclusion of the French data has been recommended in Reference 7.3 although the NRC review of this recommendation has not yet been completed. EPRI data exclusion Criterion 3, which has not been approved by the NRC, results in exclusion of model boiler specimen 542-4 from the database. Data exclusion Criterion 3 has been modified (Reference 7.6) to reflect NRC comm i

on the prior version of this Criterion. Reference 7.6 is undergoing an EPRI QA review and will be transmitted to the NRC in the near future. NRC approval of either or both of these recommendations will result in the necessary correlation of leak rate eith voltage.

i w%vu, . ram, m - 6-3 i

Figure 6-1 Farley Unit Leak Rate for Steam Generator C Percentage Contribution from Voltage Bias

!1 ~- ,.

EOC-15 Projection 12.0% --  ?

- c- EOC-14 Actual 8 4

4

. 10.0% - fg

.3

s 3 j fI 8.0% - '1 5 9 II w

9

1. g fl II II 11 6 i El

.e 6.0*/o - i s n I~I

{' i qpg ii d V\ 1 1 88 g I hI I I b 1 1 , I I 4.0% - , g- g- -

t7 88 y

d i'i 1 q I I I D 2""

r u; e, 9

7

,, -'i e

9 1 g s '1 s d 4 i t i i

-20 1

tT i ^^^^^

OM; W - ^ -^ ^ - ""

E ; $ E E 2 I I 5 % % % % % % 2 % % % % 0 ; O yy3yyE s.ai. v.its 1

m

1 Westinghouse Proprietary Class 2C , '. l

. 1 Figure 6-2: Combined 7/8" and 3/4" Tube Data SLB Leak Rate vs. Bobbin Amplitude, AP = 2560 psi @ 650*F 3/4" Bobbin volts Multiplied by 1.36 to Obtain Equivalent 7/8" volts a

1000.

i 3 o 5 & a l

100. u s a

x q o o to a

N O x

@ io~ n -

oo" o",

3 oU e C3 o ao x x *x

• E CC au x o

q x o xxa o  ;

e 1 x

~

a 1. o 7 , , , , , ,

3 4

x o 3/4-Inch Data o! '

!'  ! x 7/8 Inch Data  ;

1 xj i

!I!

x; i I

l l I 1

I  ! I l 4 i .

o

1 i

l.  ! l i  ! i  !

0.01 1 10 100 Bobbin Amplitude (Volts) iw_nrionv a 6-5 "*"' '2 23 m L_-