10. Demonstrative examples of the component-level importance measure

A demonstrative example application was carried out using a simplified PRA model to calculate the component-level risk importance measures from the PRA-calculated risk importance measures of the basic events relating to the components. The simplified PRA model used for the demonstrative examples is presented in Appendix B.1. It is based on a pressurized water reactor and is built on RiskSpectrum® PRA computer code. It is pointed out that this model was developed for demonstration of risk importance measures only and should not be related to actual NPP risk.

In a typical PRA computer code only the risk importance measures associated with the basic events are usually provided. As an example, all importance measures which are calculated by the RiskSpectrum® computer code are described in Appendix A.

In the example PRA model from Appendix B.1, a selected number of components are focused upon to show the component-level importance measures and their interpretations. Table 10-1 presents a summary of those components. For each component, it also shows the quantitative input which is needed from the RiskSpectrum® PRA model for calculation of component-level importance measures by the formulas described above. This input consists of the probabilities of representative basic events and FC measures for those basic events, as calculated by the RiskSpectrum®.

Table-1 Summary of components used as examples.

The component-level risk importance measures for those components are presented in Table 10-2. They were obtained in two ways. The first one is by appropriate modification of the PRA model (e.g. to reflect the failed or “perfect” status of the considered component) and its rerunning. This is, actually, the most “exact” way achievable in practice to calculate the importance measures. It is here referred to as “full requantification of PRA model”. The results are shown in the third column. The second way to obtain the component-level importance measures is to use the formulas described above. Corresponding results are shown in the fourth column. The details on how the importance measures were, in all cases considered, obtained in both ways can be found in Appendix B.3. As can be seen from the comparison of the values in two columns, the results obtained by the formulas are the same or very close to those obtained by the full PRA model requantification. Slight differences are attributable to rounding errors and to approximations and truncation process incorporated in the top event probability (frequency) quantification by the PRA model.

Table10-2 Summary of calculated component-level importance measures.

These two tables demonstrate the advantage of obtaining the component-level importance measures in deciding the risk significance of components or prioritizing the components in terms of their risk significance. As discussed earlier, in many applications, judgments are needed for components and the user would not have to sort through risk importance of multiple basic events associated with a component; instead, a direct measure for the component could be conveniently used.

Additionally, Table 10-3 focuses on the results for the component-level RAW importance measure for those components which are involved in a CCF group. This is of special interest in many PRA applications. The Table 10-3 shows, for each component, obtained RAW values for the three types of unavailability discussed in Sections 8 and 9. The representative component-level RAW is shown in the middle position of the three values. This is “unavailability or failure with CCF potential” (type 1 in Sections 8 and 9). It reflects an importance of unavailability (or failure) of a component which may (depending on its cause) have implications on the status of other components in the CCF group. On the left side of the representative RAW shown is the RAW for the case when the component is taken out of service for preventive maintenance, without any implications on the status of other components in the CCF group (type 2 in Sections 8 and 9). On the right side of the representative RAW shown is the RAW for the unavailability due to a CCF of all components in the group (particular case of type 3 in Sections 8 and 9). The details on the calculations can be found in the corresponding examples presented in Appendix B.3.

Table10-3 Summary of calculated component-level RAW for three different types of unavailability of components involved in a CCF Group.

As shown in the Table 10-3, for all cases:

RAW (Prev. maintenance) < RAW (CCF potential) < RAW (CCF of all components)

The representative component-level RAW (i.e. with CCF potential) is higher than the RAW associated with preventive maintenance. This is expected and is because the representative component-level RAW is adjusting for the common-cause contribution. This increase in the component-level RAW depends on the CCF contribution. On the other hand, it is shown that the representative component-level RAW is significantly lower than the RAW for the CCF event of all components involved in the CCF group. This also comes as expected because the RAW for the total CCF event does not represent the importance of the considered component alone. Rather, it represents the collective importance (due to this particular cause) of unavailability of all components in the group.

Generally, changes in the different risk measures from a basic event for the component to the component-level are different. For example, considering the information from Tables B-7 and B-11 in Appendix B, one can see that: RAW for EDG 1 “failure to start” is 6.03 and component level RAW for EDG 1 is 17.5; FC for EDG 1 “failure to start” is 3.35E-02 and the component-level FC is 3.16E-01; RRW for EDG 1 “failure to start” (considering ) is 1.03 and the component-level RRW for EDG 1 is 1.47. This is expected and consistent with the understanding of the measures.

The use of the component-level risk importance measures can be considered more appropriate as they would take into consideration different failure modes of the component, including the common-cause failures, in an integrated manner removing the judgment process of the user. This would simplify the SSC categorization process and make PRA results more understandable and directly usable.