PRAISE analyses are performed on a piping location-by-location basis to provide the cumulative failure probability as a function of time, whereas the desired end result is the system failure frequency within various time frames. The time increments of interest are now (0-25 years), the near future (25-40 years) and the more distant future (40-60 years). If the reactor transients analyzed are per reactor-year (rather than calendar year), then the times considered are also per reactor-year. As an average, one calendar year corresponds to 0.8 reactor years, but no adjustments for this are made in this appendix. The system failure frequency is obtained from the failure frequency for individual locations by analyzing the most highly stressed location and multiplying by the number of such highly stressed locations in the system. The failure frequency for a flow rate exceeding q for a given time increment from tj to t2 is obtained from the following relation

The values of P& (t) are the output from PRAISE for the dominant location(s) in the system. The failure

frequency for the system is then obtained by multiplying by the number of locations in the system that have the high stresses of the dominant location.

F. 3.1 Material Properties

Numerous material properties enter into a PRAISE analysis, many of which are described in the references cited above. A Ramberg-Osgood stress-strain curve is used in the computation of the applied value of the /-integral, as represented by the following relation

Crack instability is governed by exceedance of a critical net section stress and/or a /-integral based tearing instability using a bilinear tearing resistance curve.

A set of default material tensile and fracture properties are provided in WinPRAISE [F.4], which are summarized in Table F.3. Unless otherwise stated, these properties are used in the base case analyses.

The properties of Table F.3 are generally somewhat conservative and are representative of undegraded materials. In some instances, degraded material properties are considered, as discussed at the particular component involved.

In the case of fatigue of initial cracks, the distributions of the initial crack depth and aspect ratio are probably the most important random variables. Unless otherwise stated, the depth distribution is taken from Reference F.10, which is the default in WinPRAISE and is also included in Reference F.9.

The approach taken by Karen Gott to the elicitation was to first consider how her experience from the Swedish nuclear fleet was applicable to the US fleet of nuclear power stations. In this respect she took into account the known histories of the various degradation mechanisms which have troubled the two fleets as well as the mitigation methods which have been developed. This led her to amongst other things to the conclusion that the likelihood of an unexpected mechanism leading to failure is probably larger than the likelihood of a known mechanism resulting in failure in a region which is inspected on a regular basis. In general a new area of concern with regard to component degradation has arisen on a seven to ten year cycle over the lifetime of commercial nuclear plants.

To produce the numbers she used her database of failures and degradation in mechanical components in Swedish plants. The degradation mechanisms are the same, but the numerical figures are different because of differences in design and construction. She based her elicitation figures on the number of leaks in proportion to the number of reported cases (many were detected early) and took these to be the current figures for a good safety culture situation. The database includes other mechanical components than pipes, but does not cover steam generator tubes, so she was able to generate figures for pump and valve housings, for example. She then considered the differences in the philosophies concerning qualification and application of inspection programmes between the two countries. This she incorporated into her thinking about the safety culture aspects, both for the current time and for the extrapolation to 40 and 60 years.

The surge line elbow weld was analysed at a 60-year life assuming the same cyclic conditions as for the elbow itself, but with the stresses factored down by 20 percent as suggested at the Elicitation Base Case Review Meeting on June 4 and 5, 2003 in Bethesda, Maryland. The two hydro cases were, however, maintained at their original values.

In this analysis, RR-PRODIGAL first simulates the weld construction, including any build inspections, to establish the start of life defect density and distribution for both buried and surface breaking defects. As stated earlier, conditional failure probabilities are assessed for both situations and combined to give the final failure probability.

The failure probability evaluated for this case was:

Table G.3 Results for PWR Surge Line Weld Analysis

It can be seen that this failure probability is over an order of magnitude higher than the base case.

This is due to the difference between having to initiate a defect and then grow this defect to failure, and having the probability of pre-existing defects in the weld. The base case values from the base material failure as reflected in Table G.2, i. e. crack initiation leading to failure, have been used in Table D. 1 in the main body of this report. Note, however, that the values reported in Table G.2 are cumulative probabilities of failure in 25, 40, and 60 years whereas the values reported in Table D.1 of the main body are frequencies. Consequently, the Table G.2 values need to be divided by 25, 40, and 60 years, respectively to facilitate any comparisons. Furthermore, the values in Table D.1 are for leak rates greater than the threshold leak rates, i. e., 380 lpm (100 gpm) while the values in Table G.2 reflect the totals.

Generally speaking making estimates of LOCA frequencies for non-piping components is more challenging than making estimates for piping systems. There are multiple components to consider, each having different operating requirements, design margins, materials, and inspectability. There are also widely varying failure modes and scales to consider. For PWRs for the smaller category LOCAs, one must consider SGTRs and small penetration failures. For the larger category LOCAs, common cause bolting failures and component shell failures need to be considered. For the larger components, the bigger design margins (compared to those for piping) are somewhat offset by the decreased inspection quantity and quality. Compounding all of this is the fact that there is generally not as much precursor information available for the non-piping components as there is for piping.

For the BWR plants, the three major non-piping components that were considered were the RPV, the pumps, and the valves. In general, many of the same degradation mechanisms that are of concern for BWR piping are also a concern for BWR non-piping components. Stress corrosion cracking (specifically PWSCC) is a concern for many of the smaller Alloy 600 components, such as the CRDMs and other penetrations. As with piping, multiple cracks and fast propagation rates could lead to LOCAs. While the mechanism (PWSCC) is more severe at higher temperatures associated with PWRs, this mechanism could become a more significant issue later in the life of the BWRs. Thermal fatigue is another degradation mechanism associated with BWR non-piping components that is common with BWR piping. Thermal fatigue is especially of concern at inlet nozzles and other locations that experience thermal stratification, especially at the feedwater nozzles. For the same reasons as highlighted above for BWR piping, thermal fatigue can possible lead to larger leaks or LOCAs.

Other mechanisms for non-piping components that were not of concern for BWR piping are radiation embrittlement, common cause bolting failures, and thermal aging of cast stainless steel components, such as pump and valve casings. Radiation embrittlement reduces the base metal toughness of the RPV. Fortunately, for BWRs, it is not as of much concern as it is for PWRs due to the increased shielding available with the BWRs. Common cause bolting failures are important for manways and bolted valves. The common cause mechanisms may possibly include: improper installation or maintenance of the bolts,

i. e., improper torque, external corrosion of multiple bolts, and steam cutting of bolts. One participant thought that these common cause failures will cause the greatest risk. Thermal aging of cast stainless steels can cause a significant reduction in the fracture toughness of these materials, however, fortunately to date no cracking mechanisms have emerged for these materials.

Figure L.22 shows the Category 1 LOCA frequencies for the RPV, pumps, and valves at 25 years. The RPV shows the biggest expected Category 1 LOCA frequencies. The Category 1 RPV LOCA frequencies are driven by the CRDM penetration failures. However, the severity of the CRDM failures for BWRs was reduced by about one order of magnitude with respect to PWR CRDM failures due to the BWR heads operating at a lower temperature. Other than these head penetrations, nozzle and body cracking were mentioned as possible sources of failures. A number of precursor cracking incidents have been seen in service. The valves and pumps contribute to a lesser extent. Most of the panelists generally treated these components the same. At least one panelist (F) had some experience with manufacturing
defects in valve bodies which led to some increased concern with valves. Other issues with valves, and pumps, included the potential for bolt failures for the reasons outlined above, and the fact that the material they are made of (cast stainless steel) is notoriously difficult to inspect and is subject to toughness degradation due to thermal aging.

Valves

Pumps

RPV

LOCA Frequency (CY-1)

Figure L.22 BWR Category 1 Non-Piping LOCA Frequencies by Major Component at 25 Years of

Plant Operations

Figure L.23 shows the Category 3 non-piping LOCA frequencies at 25 years of plant operations. The most noticeable difference between the Category 3 and Category 1 LOCA frequencies is the three orders of magnitude reduction in the median value of the estimated LOCA frequencies for the RPV as the CRDM concerns disappear. A single CRDM failure cannot support a Category 3 LOCA. Only about half of the participants were concerned about RPV nozzle failures, but those that were assigned comparatively high frequencies to them. For the pumps and valves, the corresponding decrease in LOCA frequency is only about one order of magnitude. Consequently, the pumps, valves, and vessel now contribute about equally to the overall LOCA frequency.

Valves

Pumps

RPV

LOCA Frequency (CY-1)

Figure L.23 BWR Category 3 Non-Piping LOCA Frequencies by Major Component at 25 Years of

Plant Operations

Figure L.24 shows the Category 5 non-piping LOCA frequencies at 25 years for the BWR non-piping components. As was the case for the Category 3 LOCAs, the pumps, valves, and vessel are all now of about equal importance. For these large LOCAs, the panelists felt that the valve, pump, and vessel bodies were the most likely subcomponents to fail. For the vessel body, the concern was with LTOP while for the valve and pump bodies, the concern was with fatigue and SCC.

Valves

Pumps

RPV

LOCA Frequency (yr-1)

Figure L.24 BWR Category 5 Non-Piping LOCA Frequencies by Major Component at 25 Years of

Plant Operations

Figure L.25 shows the cumulative LOCA frequencies for the BWR non-piping components at 25 years of plant operations. On average there is about a one order of magnitude shift in the cumulative LOCA frequency between each successive LOCA category. The median value for the estimate of the Category 1 LOCA frequency is approximately 10-4 while the median value for the Category 6 LOCA frequency is about 10-9.

Cat 6 Cat 5 Cat 4 Cat 3

Cat 2 Cat 1

10-14 10-13 10-12 10-11 10-10 10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2

LOCA Frequency (yr-1)

Figure L.25 Cumulative BWR Non-Piping LOCA Frequencies at 25 Years of Plant Operations

Figures L.26 and L.27 show the effect of time on the Category 1 and 3 cumulative LOCA frequencies, respectively, for the BWR non-piping components. For all intents and purposes there is almost no effect of time on the predicted LOCA frequencies. The median values do not change nor does the variability, i. e., the interquartile ranges remain the same. Non-piping components are affected by similar partially compensating factors as the piping components. In addition, a number of participants expressed the belief that the maintenance and mitigation issues raised for piping also apply for non-piping components. The only thing that changes is the minimum value predicted by Participant H for LOCA Category 3. Participant H foresees the non-piping LOCA frequencies increasing at both the 40 and 60 year time interval. Figure L.28 shows the effect of time on the Category 5 frequencies. In this case the median values do not vary with time, nor does the maximum values, however, a number of participants started to see the LOCA frequencies increasing near the end-of-plant license renewal (60 years) such that the lower end of the IQR (i. e., the 25th percentile) increased an order of magnitude at 60 years over what it was at 40 and 25 years. This increase in the Category 5 predictions was driven by aging concerns of a few of the panelists at 60 years.

Figure L.26 Effect of Operating Time on the Cumulative Category 1 LOCA Frequencies for BWR

Non-Piping Components

Figure L.27 Effect of Operating Time on the Cumulative Category 3 LOCA Frequencies for BWR

Non-Piping Components

———— 1———— 1———— 1———— 1———— 1———— 1———— 1————

10-14 10-13 10-12 10-11 10-10 10-9 10-8 10-7 10-6

LOCA Frequency (yr-1)

Figure L.28 Effect of Operating Time on the Cumulative Category 5 LOCA Frequencies for BWR

Non-Piping Components

Figures L.29 and L.30 show the cumulative MV estimates, along with the 5% and 95% bound values for the various participants for the BWR Category 1 and 3 non-piping LOCA frequency estimates, respectively, at 25 years of plant operating time. Of note from these figures is the fact that a number of the participants (A, E, F, and H) predicted greater uncertainty for the Category 3 LOCAs than they did for the Category 1 LOCAs. This is not unusual in that one would expect the uncertainty to increase for lower frequency events, such as larger LOCAs. It is probably somewhat more surprising that the other four participants predicted comparable uncertainty for the Category 1 and 3 LOCAs. Overall the predictions for BWR non-piping were more consistent than for PWR non-piping discussed next. For the BWRs, there are less components and failure modes to consider and the approaches used to estimate the frequencies were more closely related.

LOCA Frequency (yr-1)

Figure L.29 BWR Non-Piping Category 1 LOCA Frequencies Showing MVs, 5% LB, and 95% UB
Values for All Participants Who Responded to the BWR Non-Piping Questions

LOCA Frequency (yr-1)

Figure L.30 BWR Non-Piping Category 3 LOCA Frequencies Showing MVs, 5% LB, and 95% UB
Values for All Participants Who Responded to the BWR Non-Piping Questions

The PIPExp database has evolved over a period of about ten years. A first database version was developed with financial support from the Swedish Nuclear Power Inspectorate (SKI). Since the conclusion of the initial R&D effort in 1998 an active maintenance program has supported the database.

D. A.1 Database Structure

Designed in Access, the database consists of searchable free-format text fields and a large number of data fields that are used as data filters in support of a range of data processing needs. Table D. A. 1 is a summary of text and data fields.

D. A.2 Completeness and Quality Management

The completeness of the pipe failure data is addressed through a continuous database management program. Extracted from Monthly Summary Reports, Table D. A.2 provides snapshots of the database evolution from 1998 to the present. In PIPExp, each record is assigned a ‘Quality Index’ (Table D. A.1, item #4, and Table D. A.3) as one means of monitoring the completeness and technical accuracy of source information as well as the process of classifying and coding of the source information.

Table D. A.1 Description of Data Fields in PIPExp

Database as of 12-31-1998 Database as of 12-31-1999 No. Damage Reports by QA-Index22_____________________________ No. Damage Reports by QA-Index Plant Type Totals 1 2 3 4 5 6 Totals 1 2 3 4 5 6 BWR 673 210 66 3 74 7 277 1595 1000 168 2 146 53 226 PHWR 100 30 3 — 56 1 10 100 30 3 — 56 1 10 PWR 1376 386 123 6 152 84 746 1656 645 176 5 211 208 411 RBMK 57 3 6 — 19 28 1 66 7 6 — 22 31 — 2291 629 198 9 301 120 1034 3417 1682 253 7 435 243 647 Plant Type Database as of 12-31-2000 No. Damage Reports by QA-Index Database as of 12-31-2001 No. Damage Reports by QA-Index Totals 1 2 3 4 5 6 Totals 1 2 3 4 5 6 BWR 1711 1111 164 2 175 58 201 1784 1172 166 2 197 63 184 PHWR 95 43 1 — 41 10 — 96 44 1 — 41 10 — PWR 1748 696 181 5 260 209 397 1952 811 194 5 329 221 392 RBMK 125 12 5 1 77 30 — 125 12 5 1 77 30 — 3679 1862 351 8 553 307 598 3957 2039 366 8 644 324 576 Plant Type Database as of 12-31-2002 No. Damage Reports by QA-Index Database as of 06-01-2004 No. Damage Reports by QA-Index Totals 1 2 3 4 5 6 Totals 1 2 3 4 5 6 BWR 1872 1216 174 12 219 75 176 2033 1370 189 77 233 93 71 GCR, HWLWR — — — — — — — 12 12 — — — — — PHWR 106 51 2 — 42 11 — 101 47 1 — 41 12 — PWR 2077 1011 198 6 351 233 278 2280 1213 218 103 344 270 132 RBMK 160 48 — — 18 81 — 160 12 5 4 109 30 — 4215 2290 379 22 721 349 454 4586 2654 413 184 727 405 203

22 See Table A-3 for a definition of QA-Index.

QA-Index Definition 1 Validated — all source data has been accessed & reviewed — no further action required 2 Validated — source data may be missing some, non­critical information — no further action anticipated 3 Validated — incomplete source data — assumptions made about material grade and/or exact flaw location — no further action anticipated 4 Validation based on incomplete information — depending on application requirements, further action may be necessary 5 Validation based on available, incomplete information — further action expected (e. g., retrieval of additional source data) 6 Not validated — validation is pending, or record is subject to deletion from database

In the VIPER software, cracks are assumed to exist in BWR vessel welds due to two causes — original manufacturing defects and service-induced cracks which initiate in the stainless steel cladding. These cracks are assumed to grow as a function of operating time due to fatigue crack growth and SCC of the low alloy steel vessel material. Simultaneously, the vessel beltline region is assumed to embrittle due to irradiation. Monte Carlo simulations of these processes are employed in VIPER, which include fracture mechanics crack growth calculations due to fatigue and SCC, and a comparison of predicted crack sizes to the critical crack size due to normal operation as well as possible transient conditions. The governing transient condition was determined to be a LTOP event, since BWRs are not subject to PTS.

The effects of ISI are imposed at appropriate inspection intervals, assuming a POD curve for the inspections. Flaws that are detected during ISI are assumed to be repaired, and thus eliminated from the population, such that they can no longer grow to a leak or vessel failure.

The axial vessel beltline welds are divided into a series of segments, and each segment is analyzed separately to account for axial gradients of irradiation fluence in the welds, which peaks at the core centerline, and decays at elevations above and below that location. The failure frequencies from each segment are weighted by their respective weld volume, and summed to determine failure frequency for the entire vessel.

Modes of failure considered are:

1. Vessel fracture during normal operation (KI > KIc)

2. Vessel fracture during an assumed LTOP event. The LTOP event considered is pressurization to 1,150 psi (7.93 MPa) at 88°F (31°C), which is assumed to occur at a frequency of 1E-3.

3. Predicted crack growth to 80% of wall thickness before failure modes 1 or 2 occurs (LBB)

The results for a typical BWR are given in the following table:

Table I.2 Summary of BWR RPV Beltline PFM Results

These provide an estimate of the probability (per vessel year) of breaks of various sizes due to vessel beltline failures. To complete this table, it was assumed that a leak (LBB mode failure) corresponds to a crack of length = 60 inches (1525 mm) that breaks through and begins leaking as a through-wall crack of this length (since the wall thickness is approximately 6 inches (150 mm), and cracks in VIPER are assumed to have a ten to one aspect ratio). Dave Harris ran this case using the PRAISE code leakage rate prediction capability (See Appendix F for description), and computed a leak rate of 193 gpm (733 lpm) for an axial crack of this size in a BWR vessel. Thus, predicted LBB mode failures from the RPV beltline were treated as Category 1 breaks. Predicted vessel fractures, either during normal operation or due to LTOP events were treated as complete RPV ruptures, which were assumed to result in very large, Category 6 breaks. Intermediate break sizes were then determined by log-log interpolation between these two extremes. The sharp increase in large break probability between years 40 and 60 is attributable to the combined effect of two aging mechanisms — crack growth and RPV embrittlement.

None

Comments Related to Section 7 of NUREG-1829

Comment Number: 7-1

Submitted by Zouhair Elawar — Palo Verde Nuclear Generating Station Comment: Draft NUREG-1829 used plant experiences to estimate the steam generator tube rupture (SGTR) frequency which amounts to greater than 50% of the total small LOCA frequency. Estimate of the remaining 50% of Category 1 LOCA was entirely based on expert elicitation. The resulting Category 1 frequency estimates from the panel showed a significant divergence of opinions. I strongly recommend that Category 1 LOCA frequency estimates should continue to be related to the large number of years of plant experiences similar to the method used in NUREG 5750. The current lengths of those experiences amount to thousands of reactor-years. They are statistically significant to be used to estimate the annual frequency of events at the 1E-2, 1E-3, and 1E-4 levels. Similar estimates are used in PRA models for numerous other important PRA parameters (such as SGTR).

Response: Operating experience becomes more relevant as the LOCA size decreases, especially for events such as SGTR, because the database is populated with actual events. However, there is still scant data on SB LOCAs due to limited passive piping or non-piping component failures. The panelists were made aware of this operating experience data, and SGTR was specifically used as a base case estimate, along with frequencies for other piping and non-piping precursor events (e. g., cracking). However, one of the cautions with using operating experience solely to calculate Category 1 LOCA frequency estimates is that past experience is not necessarily indicative of current or future performance. Common cause material degradation can result in systematic increases in generic frequencies as a function of time compared to frequencies based on operating experience. Conversely, wholesale mitigation programs (as with IGSCC) can result in systematic decreases in LOCA frequencies over time.

These common cause considerations are one reason why elicitation is preferable to simple operating experience estimates for even small break LOCAs. In fact, the elicitation results are most justified for this LOCA category because of operating experience data. The operating experience data was fully considered by the panelists to estimate past LOCA frequencies, and they then used their best judgment to modify these estimates, as appropriate, based on their current knowledge. It is preferable to use current knowledge in this way rather than wait for several years for operating experience to adequately reflect the new conditions. It is worth noting, however, that the elicitation results and operating experience data are generally consistent for LOCA Category 1. The only differences, which are not statistically different, are for the frequencies associated with non-SGTR, Category 1 PWR failures. The panelists justified the higher NUREG-1829 estimates as a result of the effects of PWSCC on small diameter piping failures.

Key elements of this response have been incorporated into Section 7.10 of the revised NUREG. This comment is also very similar to GC5. Please see the response to this comment for additional information.

Comment Number: 7-2

Submitted by Zouhair Elawar — Palo Verde Nuclear Generating Station

Comment: The draft NUREG combined a variety of LOCA sources into each LOCA category. Piping LOCAs and several non-piping LOCAs were pooled together to form each of the LOCA categories. It would be useful for each of the 6 LOCA categories to add a table of LOCA sources and frequency contributions. This breakdown is particularly important for the small and medium LOCA categories.

Some contributors to the small and medium LOCAs are modeled separately in most PRA models (SGTR, RCP seals, inter-system LOCAs and others). If the end user does not subtract the separately-modeled LOCA contributors, then the contribution to CDF (core damage frequency) from those contributors would be conservatively and redundantly modeled.

Response: First, it is important to recall that this study was only concerned with estimating passive system failure frequencies of structural system components (SSCs) within the primary system (Section 2). Therefore, LOCAs associated with RCP seals, interfacing system LOCAs, and active system LOCAs are not included in the NUREG-1829 estimates. There are previous estimates for these types of events (e. g., NUREG/CR-5750). The revised NUREG (Section 7.8) includes a table of SGTR frequencies as requested by this comment because all the panelists provided sufficient information to determine individual LOCA frequency estimates for this system. Section 7.8 of the revised NUREG also contains estimates for all non-SGTR passive-system SSC failures.

In general, however, it was not possible to develop information for individual system failures from the elicitation responses. For each LOCA category there was a potential contribution from either 12 or 13 different piping systems (12 for the PWRs and 13 for the BWRs) plus a contribution from either 3 or 5 non-piping components (3 for the BWRs and 5 for the PWRs). In addition, for each of the non-piping components, there were up to 8 (for the RPVs) subcomponents (e. g., vessel body, CDRMs, nozzle, head bolts, etc.) that were potential contributors. In order to make the responses tractable, the panelists were requested to only provide information for systems that they expected to make up at least 80 percent of the failure frequency contribution. Thus, not every panelist provided responses for every piping system or non-piping component or subcomponent. Furthermore, it was not possible to impute this data accurately because the relative frequency contributions to each panelist’s estimates are small. Consequently, it was not possible to calculate group estimates of individual system failure frequencies because this would have required each panelist to provide estimates for every piping system and non-piping component..

Key elements of this response have been incorporated into Section 2 of the revised NUREG. Additionally, Section 7.8 has been added to the revised NUREG to provide the SGTR and non-SGTR estimates for PWR plants.

Comment Number: 7-3

Submitted by Zouhair Elawar — Palo Verde Nuclear Generating Station

Comment: The steam generator tube rupture frequency (merged with LOCA category 1) was reported as mean value = 3.5E-03 based on number of industry events averaged over years of reactor operations. This mean value frequency should be separated from the main small LOCA category and estimated as a “range” consisting of upper and lower bounds. Plants with aging steam generators are encouraged to use the upper bound of the range. And, plants with new steam generators may use the lower bound of the range. Of course, numerous plants would use the overall mean value of 3.5E-03.

Response: Separate SGTR frequencies have been calculated from the elicitation responses and are provided in the revised NUREG (Section 7.8). The current-day median STGR frequency is 2.6E-3 and the mean frequency is 3.7E-3 (Table 7.18). These values compare favorably with the operating-experience Category 1 frequency of 3.5E-3 for SGTRs as reported in Section 4.4.1. The 5th and 95th percentiles of the group estimates are 5.0E-4 and 1.0E-2, respectively. Additional discussion concerning the SGTR estimates is also contained in Section 7.10 of the revised NUREG.

Comment Number: 7-4

Submitted by Zouhair Elawar — Palo Verde Nuclear Generating Station

Comment: The various LOCA frequencies are reported in the several tables as cumulative values. In order to isolate the frequency of each LOCA category, one has to subtract the frequency of the next higher ranking category. This reporting format may lead to human errors. Some users may not become aware of the cumulative table format since that description is briefly stated at the later sections of a very large report. Please add a footnote under each LOCA frequency table explain how to obtain the frequency of each LOCA category.

Response: The six LOCA categories were defined in terms of cumulative thresholds because the panelists felt more comfortable with providing their responses as cumulative thresholds rather than to intervals defined by consecutive thresholds. (Appendix B) Because the differences between the results for consecutive LOCA categories are typically much smaller than their uncertainties (see Table 1, Figure 1, Figure 7.36 and Figure 7.37), there is no statistical difference between using the cumulative frequencies or interval frequencies determined by subtraction. If interval-defined LOCA frequencies are required, simply use the NUREG cumulative threshold estimates in Section 7.7. However, Section 7.9 does use interval-defined frequencies for consistency in comparing the NUREG-1829 estimates with other prior study estimates. This point has been clarified in the revised NUREG in the Executive Summary, Section 3.4.1, and Section 7.9.

Comment Number: 7-5

Submitted by Zouhair Elawar — Palo Verde Nuclear Generating Station

Comment: The equivalent break diameters used in various PRA models that form boundaries between various LOCA categories are not necessarily matching those used in the draft NUREG. For example, a small LOCA range may extend up to 0.03 square feet equivalent break area (derived through existing capability of high pressure safety injection). The draft NUREG should give a clear guideline on interpolations between the various LOCA frequency values, including advice on arithmetic or geometric preference for interpolation.

The main coolant piping system is one of the base case systems. The failure probability of the large piping of this system is dominated by the hot leg to pressure vessel weld, because this location is at the highest temperature and sees the highest stress.

F. 3.2.1 Dimensions and Welds — From the piping isometrics made available to the panel members, the hot leg has a 29 inch inner diameter, and a thickness of 63.5 mm (2.5 inch) (OD=34 inches), fabricated from SA-376 (which is an austenitic stainless steel). The example plant has two coolant loops. There are several welds in the hot leg, including shop and field welds and safe ends. There is a safe end and field weld at the pressure vessel.

F.3.2.2 Stresses and Cycles — Table 1-2, page 9 of Reference F.1 summarizes the deadweight, pressure and restraint of thermal expansion stresses for the 14 field welds in one loop of the large primary piping in the plant considered in that report. Joint 1 is the hot leg to pressure vessel joint, and it has the highest stresses. The seismic stresses are included in Table 1-3, page 10 [F.1]. They are generally quite low. The postulated list of transients is provided in Table 4-1, page 152 [F.1], and is the transients occurring over the 40 year design life, which corresponds to reactor years. The list contains 11 types of transients. There is sufficient information in Reference F. 1 to consider all of these transients, but only the heat-up cool-down transient will be considered in this case, because it is the dominant transient contributing to fatigue crack growth [F.1]. The heat-up cool-down transient was postulated to occur 200 times in 40 years (5/year). This is excessive, and 3/year is used herein.

The seismic stresses are given in terms of the maximum load controlled stress (deadweight + pressure + max seismic), and the summary of the stress history needed for fatigue crack growth analysis is also provided. This summary, denoted as S, is the sum of the cyclic stresses as follows

S4 = Z^x Ao2 [F.4]

se ism icstressh isto ry

This is what controls the amount of fatigue crack growth during a seismic event for a fourth power crack growth law that includes Я-ratio effects (See Reference F.1). Table F.4 summarizes the suggested stress history for the hot leg to pressure vessel joint.

Table F.4 Summary of Stress History for Hot Leg to Pressure Vessel Joint
deadweight stress = 2.08 ksi
pressure stress = 6.49 ksi (axial)
restraint of thermal expansion stress = 6.50 ksi
3 times per year

The right-hand column in the above table is the cyclic stress if the seismic event contains 200 stress cycles all of the same amplitude, with a low minimum load. This column is derived from the value of S4 and Equation F.4, with omax=Ao, and is included just to provide an idea of the size of the seismic stresses. They are not large.

Residual stresses, when considered, are taken to be the default values for large lines, as reported in Reference F.2.

F.3.2.3 Results — WinPRAISE runs were made for the hot leg to pressure vessel weld using the above stresses and default material properties. Table F.5 summarizes the results. “Good” inspections at 0, 20 and 40 years were considered. These results are the cumulative leak probability. The left hand column gives the leak rate in gallons per minute. The next column in gives the time (25, 40 and 60 years), and the probabilities are directly from the PRAISE output for these times. For the Monte Carlo simulation, the crack size plane (a/h — a/b) was divided into 20 by 20 strata, with a maximum of 2000 trials drawn from each stratum. Sampling from a given stratum was stopped when 20 failures occurred in that stratum. Sampling began at the corner of the a/h-a/b plane corresponding to long deep cracks (1,0), and continued to shorter, then shallower cracks until no failures occurred in a stratum within 2000 trials. Sampling was then stopped. This procedure is referred to as automated stratification, and is a feature unique to WinPRAISE [F.4]. Earlier versions of the PRAISE software require the user to define each stratum and the sampling from each.

Table F.5 Cumulative Probability PRAISE Results for Hot Leg-Pressure Vessel Weld for Fatigue

Crack Growth from Pre-Existing Defects

noticeable effect of hydro

noticeable effect of seismic when hydro test is performed, less effect when no hydro aging has about x2 effect >1500 gpm same as DEPB

Runs were made with and without a hydro test, and it is seen that hydro testing has a noticeable effect. Moderate material degradation is considered, with the values of JIc and (dJ/da)matl identified in the table. The failure probabilities are all very small, even the leak probabilities. The influence or seismic events is seen to be quite small.

Table F.5 provides the base case results for the hot leg. Additional runs were made to study the following variables: [15]

• The effects of the fatigue crack growth relation employed were studied. The fatigue crack growth relation in PRAISE for austenitic stainless steel is based on information available during the original software development. More recent crack growth relations have been suggested [F.11]. For the simple stress history in this case, it is possible to run PRAISE with a crack growth relation that is equivalent to the more recent relation.

• PWSCC crack initiation and growth has been identified in the control drive mechanisms (CRDM) in PWRs. This occurs in the Alloy 600 weldment. This alloy is also used in the safe end of the pressure vessel to main coolant piping welds, so is present in the hot leg to pressure vessel weld under consideration. In order to model the initiation and growth of PWSCC cracks, the initiation kinetics were assumed to be the same as for Type 316NG stainless steel as currently in PRAISE [F.2], but the crack growth kinetics were changed to be representative of Alloy 600. Based on information in Reference F.12, the crack growth kinetics is represented by the relation

da__ CKr’

dt

where m equals 1.16, and C is lognormally distributed with a median value that depends on the temperature and material (weld, base metal, etc.). The median value of C for a weld at 315 C (600°F) is 7.86×10-7, when crack growth rates are in inches/hour and K is in ksi-in1/2. Combining the within-heat and heat-to-heat variation in C, the second parameter of the lognormal distribution is 1.193 (standard deviation of lnC = 1.193).

• The effects of more severe material degradation were studied, with the values of the degraded toughness given along with the results. Since PRAISE can not consider time-dependent material properties, the degraded material properties are present even in new pipe. The values of the degraded properties are from Reference F.13.

The results of these additional runs are summarized in Tables F.6 and F.7.

Table F.6 Cumulative PRAISE Results Additional Runs for Hot Leg Pressure Vessel Weld

The design limiting stress was 40.0 MPa (4.49 ksi). In the case of PWSCC growth, initial fabrication defects were considered with the default depth distribution discussed above. Both initiation and growth were considered for the column identified as PWSCC initiation. The higher large leak rates for the initiation relative to the PWSCC growth are due to the possibility of multiple initiation sites, whereas the growth considers only one initial crack.

Table F.7 Additional Hot Leg Pressure Vessel Runs Considering Material Aging

Good Inspection at 0, 20 ,40 Updated da/dN

3 HU-CD per year No Hydro Unless Specified

Type 304 Stainless

Degraded Properties Used for All Times

The biggest effect in Table F.7 is not having a hydro test. This assumption is necessary, because when degraded material properties are used, everything that fails does so during the hydro test.

“Extremely sensitive aged” material properties are needed before degradation has a large effect.

For each plant type and for piping and non-piping Dr. Harris selected a reference system and attempted to scale other systems relative to that reference system. He tried to use estimates based on operating experience to the maximum extent, and then scaled the relative frequencies for the LOCA categories using results from the PFM analyses. In many instances, operating experience was not applicable, so he then relied more on the PFM results. If he felt that a given system was not a significant contributor to the leak frequency for a given LOCA category, then he was less concerned about the accuracy of the frequency estimate for that system.

PWR Non-seismic LOCA: For the PWR case, he used the hot leg as the reference system for large leak flow rates. Operating experience is not readily applicable. The PFM analyses for the hot leg showed a very wide range of results depending on the assumptions and input to the analyses. Therefore, he scaled the hot leg results by use of the RPV reference case results. Results presented at the wrap-up meeting in February 2004 provided an estimate of the RPV > 500,000 gpm (1,900,000 lpm) as 10-10 (per plant-year) for the first 25 years. He doubled this value to account for 2 hot legs. He assumed that the leak frequency for 100 gpm (380 lpm) LOCA is 3 У orders of magnitude higher, and then interpolated on a log-log scale. This fixes the frequency-leak rate for the hot leg at 25 years. He assumed that the frequency for > 500,000 gpm (1,900,000 lpm) in the time increment 25-40 years is twice that for 0-25 years, and four times as large in the increment 40-60 years. The leak frequency for 100 gpm (380 lpm) is assumed to be independent of time.

The cold leg is then assumed to have frequencies 1/3 those of the hot leg, because the cold leg operates at a somewhat lower temperature. The surge line is assumed to have leak frequencies 100 times as large as the hot leg, because the surge line sees a lot more cycles than the hot leg, and is just as hot. These estimates then define the very large leak frequencies for the entire plant.

At the low end of the leak rate scale, he assumed the plant results to be bounded by the past operating experience for steam generator tubes. An estimate from the wrap-up meeting for the steam generator LOCA frequencies is 3.5×10-3 per plant-year.

He used the HPI make-up nozzle as a surrogate for all 2 to 6 inch diameter lines. He used the reference case results from the PFM results that was presented at the wrap-up meeting, but reduced the leak frequencies by an order of magnitude at 5,000 gpm (19,000 lpm). He assumed that the SIS accumulator and RHR systems have about an order of magnitude less contribution than the surge line, so they have a small contribution to the overall plant.

This procedure provides his best estimate. The 5% and 95% estimates are scaled up and down from the best estimate. He estimated the 5% to be 1 У orders of magnitude below the best estimate (multiply by

0. 03), independent of time and leak rate. He varied the multiplier for the 95% estimate, making it larger for the larger leak categories. The multiplier varied from 30 to 1000. He believes that we have a better handle on the smaller leak rates, because they are bounded by steam generator tubing experience, which is plentiful.

BWR Non-seismic LOCA: He selected the recirculation system for the reference for BWRs. For intermediate leak rates 100 to 25,000 gpm (380 to 95,000 lpm), the 12 inch diameter portion of the recirculation system dominates. He used the base case results from the wrap-up meeting, but reduced them by an order of magnitude, because his PFM analysis underestimated the benefit of the post-remedial action residual stress.

The feedwater system is also important, because it has lots of welds, and is prone to FAC (which is not related to welds). Since this is a dominant system, he assumed it to be comparable to the surge line in a PWR (which is a dominant system for that type of plant). The steam line is about the same size and same material as the feedwater, but is not prone to FAC, therefore he assumed the steam line to have leak frequencies that are two orders of magnitude below the feedwater system. He assumed the RHR line to be the same as the PWR surge line, which is about the same size. NUREG/CR-6674 shows very low probabilities of through-wall cracks in the HPCS/LPCS system, so the contribution of this system was assumed to be negligible. The recirculation, feedwater, steam line and RHR are assumed to be the dominant systems, and no estimates were made for other systems.

The estimated uncertainty bands are generally tighter than the PWR estimates, because they are based more on experience for the dominant system (recirculation). They are independent of time, but do vary somewhat with leak category.

Non-piping: Dr. Harris felt less confident making estimates for non-piping components, because most of his experience is related to piping. He relied heavily on results provided by others in the wrap-up meeting, and used CRDM nozzle PWSCC, rector pressure vessel and steam generator tubing data for reference purposes. He also scaled relative to piping in many instances. He did not estimate time dependencies. For instance, for pumps and valves, he figured that they are less failure prone than the piping system in which they are located (passive failures). He estimated probabilities, and then calculated relative contributions of failure scenarios.

Having established a basic failure probability, the COD can be evaluated independently of this probability. Once this is established, the leakage area of the defect follows, and given this leak area, the flow rate can be evaluated using the information from Figure G.1. A mean power law was then used to calculate the mean flow rate given a leakage area. The table below gives the elastic COD values evaluated for this case and the resultant flow rate.

Interpolating between the results in Table G.4, it can be seen that a defect approximately 160 mm (6.2 inches) long, which is approximately 15 percent of the pipe circumference, results in the first leakage category of 380 lpm (100 gpm).

Clearly it is the behaviour of the defect beyond the elastic range that is of interest for the larger leak categories. If it were to be assumed that at the critical defect size the pipe would simply tear, in an unstable manner, to result in a Double Ended Guillotine Break (DEGB) failure, then the leak rate would simply jump from a Category 1 failure to the gpm associated with the DEGB. In this case that would be 250,000 lpm (65,000 gpm) or a Category 4 leak. The probability of a Category 2 leak rate would then be the same as a Category 3, which would be the same as the Category 4!

Such an assumption could be considered valid. However, in this work, it was assumed that the defect would continue opening in a stable, but plastic manner. Whilst models do exist to evaluate the plastic deformation of defective pipes, no such model was used in this analysis. Instead expert judgement was used to assess how the COD would develop beyond this elastic point, and at what defect size the pipe would finally tear into a DEGB failure. The results of this judgement are shown in Figure G.2. The area of leakage can then be calculated, and the leak rate, given a defect length also follows. The resulting gallon per minute flow rate, for this example, is shown in Figure G.3.

The failure probability gives the basic probability of a breach of the pressure boundary. Figure G.3 shows the leak rate in gallons per minute, given a defect of a given length. In order to obtain the probability of a leak rate greater than ‘X’ gallons per minute, it only remains to provide a distribution of the defect size at the moment of failure.

G. 4.4 Defect Distribution and Leak Rate at Failure — No Leak Detection

First consider the case with no leak detection. For this case the instantaneous size of the defect, and its associated COD, at the moment of snap through to a breach of containment is required. As an example, if the aspect ratio were of the order of 8/1 at snap through, then given a pipe wall thickness of about 36 mm (1.4 inches), the defect length would be approximately ten or eleven inches long. If it were then pessimistically assumed that this was the full through wall defect length, then the instantaneous leak rate would be just above (actually about twice) our ‘Category 1’ failure criteria of 380 lpm (100 gpm). Thus, the probability of a leak rate greater than Category 1 becomes the basic probability of failure times the probability that the defect at snap through was greater than 250 mm (10 inches), i. e., the defect had an aspect ratio at snap through of about 8/1 or greater. It then follows, from Figure G.3, that in order to exceed the Category 2 leak rate, the instantaneous defect size at snap through would have to be greater than 380 or 405 mm (15 or 16 inches), i. e., the defect had an aspect ration of about 11/1. Furthermore, the defect snapped straight open to the fully plastic COD.

As stated earlier, RR-PRODIGAL has the capability of simulating the crack growth both around and through the pipe wall. However, this is not generally used as the solutions require a detailed knowledge of the stress distribution around the pipe, including any weld residual stress, and generally such knowledge is not well enough defined. Thus, expert judgement was again used. The expert
judgement required is to generate a defect distribution at the moment the defect snaps to the COD of Figure G.3, assuming no leak detection.

This base case is for the surge line elbow and it has been assumed that most of the deformation and high stress will result from large bending moments at the elbow. It was felt that this would initiate a defect preferential on the hogging side of the elbow, and promote a crack to grow through the wall thickness on this side of the elbow. This would then imply that the crack growth around the pipe diameter would be restricted. Figure G.4 represents the distribution decided upon for this analysis. This distribution shows the most likely defect length to be up to about 250 mm (10 inches), which is about a quarter of the way around the pipe circumference. The probability of the defect being over halfway around the pipe is seen as a rare event, being about 0.025 or a 1 in 40 chance. If the loading were not dominated by bending, then this distribution would probably be judged to be flatter, with perhaps a 1 in 10 chance of being greater than halfway round the pipe circumference.

Combining Figures G.3 and G.4 gives the conditional probability of a leak greater than a given leak rate. This final plot is given in Figure G.5 and is combined with the basic failure probability to derive the values given in Table D. 1 in Section D of the main body of this report.