Resolution of Generic Safety Issues: Issue 194: Implications of Updated Probabilistic Seismic Hazard Estimates ( NUREG-0933, Main Report with Supplements 1–34 )
Beginning in the early-1980s, the NRC sponsored the development of a Probabilistic Seismic Hazard Analysis (PSHA) methodology by LLNL. For the purpose of conducting a systematic evaluation of the licensing criteria for older plants, a limited study of the seismic hazard at the sites where these plants are located was conducted in 1982 and documented in NUREG/CR-1582.1834 In a 1982 letter, the USGS suggested that deterministic and probabilistic evaluations of seismic hazard should be made for the Eastern United States (EUS) to assess the likelihood of large earthquakes along the eastern seaboard. This led to the 1989 publication of the PSHA study of all 69 sites in the Central and Eastern United States (CEUS) by LLNL in NUREG/CR-5250.1835 In conjunction with funding the LLNL study, NRC also recommended that the nuclear power industry conduct an independent study to present a coordinated utility position on PSHA estimates. The industry study of 56 CEUS sites was conducted by EPRI and the results were published in EPRI-NP-4726 in 1986.
A draft report on the trial implementation of the Senior Seismic Hazard Analysis Committee (SSHAC) guidance1838 for the probabilistic seismic hazard assessment of the Watts Bar and Vogtle1839 nuclear plants showed a higher probabilistic seismic hazard estimate for the Watts Bar site than the value obtained from NUREG-1488.1836 The increase in the seismic hazard estimate was investigated in a follow-on study which identified the root causes to be a combination of characteristics of the Watts Bar site, such as the site-specific source zones characterization, and more generic ones, such as the modified ground motion model. Depending on whether new information becomes available, other sites could have similar conclusions, such as in the case of Vogtle, for which the mean estimates of the seismic hazard slightly decreased between the 1993 EUS and the 1998 Trial Implementation Plan (TIP) studies. This represented a new interpretation of new seismicity data and resulted in the identification of this issue.1837
The safety concerns were: (1) Did the new data warrant concerns regarding the seismic design bases for nuclear power plants in the region around the Eastern Tennessee Seismic Zone (ETSZ)? and (2) Were other nuclear power plants in the region adversely affected?
Large differences in the seismic hazard results between those from the LLNL study and the EPRI study led to the examination of the conflicting results. The staff decided to supplement the LLNL study by improving the elicitation of data and its associated uncertainty from the experts to better capture the uncertainty in our knowledge. The results of this study were published in NUREG-1488.1836
Although the PSHA results in NUREG-14881836 show that there is reasonable agreement on plant-specific SSEs, the LLNL seismic hazard estimates in the 10-4 to 10-6 range are systematically higher than the EPRI hazard results for this range. This is the range of seismic hazard that typically has the most influence on the contribution to seismic risk for nuclear power plants. In an attempt to better understand the reasons for the differences in the two methods, the SSHAC was established under the sponsorship of NRC, EPRI, and DOE in early-1993. The SSHAC report1838 was published in April 1997 and stated: "Originally, some of the sponsors and participants proposed that one key objective should be to 'resolve' the differences between the LLNL and EPRI studies. However, the Committee quickly realized that the new project would be most useful if it were forward-looking rather than backward-looking - specifically, if it could pull together what is known about PSHA in order to recommend an improved methodology, rather than specifically attempting to figure out which of the two studies was 'correct,' or which specific problems with either study were most important in affecting the study's specific results."
In order to apply the SSHAC methodology, LLNL was contracted to perform a study1839 (the TIP) of two trial sites (Watts Bar and Vogtle) in the Southeastern United States, a draft of which was completed in 1998. The TIP results for the Watts Bar site indicated that, at the mean annual frequency of 10-4, the peak ground acceleration (PGA) value is about 0.45g, compared to a PGA of about 0.28g at the same mean annual frequency of 10-4 from NUREG-1488.1836 In order to investigate the reasons for the difference in the results from the TIP and the earlier LLNL study, another study was conducted and documented in the draft report UCRL-ID 142039, "Comparison of the PSHA Results of the 1993-EUS-Update and the 1998-TIP Studies for Watts Bar," in March 2002. The introduction of the ETSZ, and to a lesser extent the change in the ground motion attenuation model, increased the potential for higher seismic hazard at sites in the proximity of the ETSZ. A comparison of the TIP and NUREG-14881836 hazard curves for the PGA values is shown in Figure 3.194-1 below.
At the reference annual frequency of 10-4, the TIP results are about 1.6 times higher than the 1993 EUS-Update estimate. Sites with operating plants in the proximity of the ETSZ are Browns Ferry, Sequoyah, and Watts Bar. Based on the results for the Watts Bar site, there is a potential that the ETSZ could influence the seismic hazard at these other sites as well. The effect of changes in ground motion model, although secondary in nature, can increase the response spectrum shape in the high frequency range from 9 Hz to 50 Hz. A recent study1840 also showed the increase of spectral ordinates in the high frequency end. Seismic input in the high frequency end of the response spectrum can cause relay chatter and other effects to vibration-sensitive components. The USGS seismic hazard maps for the Eastern Tennessee area also indicated a higher seismic hazard.
The assessment of seismic risk using seismic PRA models starts with a seismic hazard curve (e.g., frequency of exceedence versus PGA), as described above. Then, fragility curves (conditional frequency of failure versus PGA) for each structure, system, and component of interest must be derived. Finally, the fragility curves are convolved with the seismic hazard curve using event tree and/or fault tree logic models to calculate the frequency of various end states (e.g., CDF) - a fairly involved numerical integration. This calculation can be rather formidable - much more so than the usual internal events PRA, since a seismic event can both initiate an accident and also serve as a common mode failure mechanism for many components, structures, and systems in the plant.
If the change in the seismic hazard curve were a constant multiplicative factor, constant over the domain of the curve, the resulting change in seismic CDF would also be a simple multiplicative factor, since the proportional change would carry through the entire calculation. However, the TIP
Comparison of the Mean Seismic Hazard Estimates for the Watts Bar Site Figure 3.194-1
curve does not differ from the original curve by a constant factor. This does not change the Boolean logic of a PRA, but does change the numerical integrations. Another complication is that many plants do not have a seismic PRA, but rather as part of their IPEEE, many licensees performed a seismic margins analysis (SMA). This results in no quantification of the seismic risk at these plants, though it does provide a determination that there are safe shutdown paths that meet a required review level earthquake (RLE) and also identifies any potential vulnerabilities associated with those paths. For these plants, the IPEEE typically does identify an overall plant high confidence of a low probability of failure (HCLPF) value, though this value may take credit for plant modifications to resolve the identified vulnerabilities, anomalies, outliers, etc.
Fortunately, an August 1999 paper by Robert P. Kennedy ("Overview of Methods for Seismic PRA and Margin Analysis Including Recent Innovations," Proceedings of the OECD-NEA Workshop on Seismic Risk, Tokyo, Japan,) presented an approximate method of estimating seismic risk using the plant HCLPF value. This method assumed that the seismic hazard curve can be approximated by an exponential curve and that the fragility curves can be approximated as being log-normally distributed. Both assumptions are reasonable approximations for the purposes of the screening of this issue. Using these assumptions, this method develops a closed form solution for the seismic risk which was developed for use in sensitivity studies such as this. This method was used to develop a sense of the change in the risk estimates, based on the different seismic hazard curves (i.e., LLNL 1993 vs. TIP 1998) for the Watts Bar site. As a caution, these are simplistic calculations that give a rough estimate of the seismic CDF. However, a reasonable estimate of the expected change in CDF resulting from the change to the latest seismic hazard estimate can be obtained by applying the same approach to both sets of seismic hazard information.
The TIP results indicated that the mean seismic hazard estimate for Watts Bar was about two times greater than that estimated in NUREG-1488.1836 To compare the impact of this new seismic hazard information on CDF for Watts Bar, a simple calculation was carried out using the approximate method described above. The specific steps of the approach are identified in Section 6.2.1 of the Kennedy paper.
This calculation addressed only the seismic contribution. It did not address random equipment failures/unavailabilities or operator errors. However, it was noted from the NRC contractor's TER on the Watts Bar IPEEE submittal that "... non-seismic failures are not expected to be significant for WBN [Watts Bar Nuclear] because there seems to be sufficient diversity and redundancy in the equipment selected in the SSEL [safe shutdown equipment list] for the success paths ..." and that "... significant human action problems are not expected for WBN." Therefore, neglecting any contribution to the CDF from simultaneous random equipment failure or adverse human action in this simple calculation should not lead to erroneous results.
The results of the Watts Bar IPEEE seismic analysis, performed in accordance with the EPRI SMA methodology as described in EPRI-NP-6041-SL, "Nuclear Power Plant Seismic Margin," Revision 1, August 1991, indicated that the plant HCLPF value exceeded the review level earthquake value of 0.3g PGA. There were no significant issues identified in the staff's SER or contractor's TER of this analysis, and there were no identified seismic vulnerabilities, anomalies, or outliers.
The simple calculation included some assumptions regarding the plant's seismic capability and the logarithmic standard deviation of 0.4 that was recommended in the Kennedy paper was used. A lower logarithmic standard deviation would result in higher calculated CDF and change in CDF values. In addition, Watts Bar had identified two success paths that both exceed a HCLPF value of 0.3g PGA. Using the HCLPF Max/Min method rules, the plant HCLPF is equal to the greater of the HCLPF values for these two success paths. However, it was not clear from the SER or TER what precise HCLPF values were achieved for each success path; only that they both exceeded 0.3g PGA. Therefore, in this analysis both success paths were assumed to only just meet the 0.3g PGA and, thus, this capacity was also used to represent the plant HCLPF in the analysis. If a higher HCLPF value were used, lower CDF and change in CDF values would be calculated. With the plant HCLPF of 0.3g PGA and assuming the logarithmic standard deviation of 0.4, the simplistic approach was used to estimate the risk associated with seismic events for the different seismic hazard information.
Using this method and the LLNL seismic hazard information documented in NUREG-1488,1836 the Watts Bar seismic CDF was estimated to be about 10-5/RY. Using this approach and the new seismic hazard information from TIP, the Watts Bar seismic CDF estimate increases to about 4 x 10-5/RY. This approach implicitly assumed no change in the spectrum shape from the IPEEE study. But the TIP uniform hazard spectrum, which is based on a 10-4 mean PGA value, has higher spectral acceleration values than the design SSE spectral acceleration values above about 7 Hz and the increase peaks at about 25 Hz. However, in the 1 to 7 Hz range, the spectral acceleration values are significantly below those from the SSE spectrum. In order to account for the effect of this difference in spectrum shape on the CDF, the Watts Bar plant HCLPF value (0.3g) was scaled to the spectral acceleration values at 5 and 10 Hz, and the scaling relationships for 5 and 10 Hz spectral ordinate from the TIP uniform hazard spectrum were used to determine the CDF values at 5 and 10 Hz. The resulting average CDF was 1.8 x 10-5/year. Therefore, accounting for the TIP uniform hazard spectrum shape, there was an increase in CDF of about 0.8 x 10-5/year.
In order to determine the sensitivity of the estimated CDF for the Watts Bar site using the TIP seismic hazard curve, several CDF estimates were made using the mean, 15th, and 85th percentile hazards, with varying uncertainties (beta values). From Figure 3.194-2, it is apparent that the CDF values are not very sensitive to the percentile level of the hazard curve. This is because the HCLPF value is high and at the low end of the annual frequency of occurrence.
This issue specifically addressed plants in the ETSZ. However, at the time of this analysis in 2003, the USGS had undertaken a nationwide effort of seismic hazard mapping under the National Earthquake Hazard Reduction Act. In early-2003, the USGS issued revised hazard maps using a methodology quite similar to the SHAAC approach and the NRC was conducting a study of the USGS methodology as a part of the 10-year seismic data base updating activity. This project was expected to lead to an assessment of seismic hazard at existing plant sites. At the end of the NRC study, a comprehensive perspective of the increase or decrease of plant seismic hazard and its effects on the SSE ground motion at all the EUS plants was expected to be available.
Based on the risk estimates associated with the spectrum shape for the Watts Bar site and Figure C5 of Management Directive 6.4, the issue regarding the adequacy of deterministic seismic design criteria for the licensing basis of plants in the ETSZ was excluded from further consideration. A generic study may be required to assess the significance for other plants, if the revised USGS results confirm the TIP results and show increases in the seismic hazard for more sites.1841