Bayesian methodology for generic seismic fragility evaluation of components in nuclear power plants

Bayesian methodology for updating the seismic fragility of components in nuclear power plants is presented. The generic fragility data which have been evaluated based on the past SPSAs are combined with the seismic experience data. Although the seismic experience is limited to the acceleration range below the median capacity of the components, it has been found that the evidence is effective to update the fragility tail. In other words, the uncertainty of the fragility is reduced although the median capacity itself is not modified to a great extent. The annual frequency of failure is also reduced as a result of the updating of the fragility tail. The PDF of the seismic capacity is handled in discrete form, which enables the use of arbitrary type of prior distribution. Accordingly, the Log-N prior can be used which is consistent with the widely used fragility model. For evaluating posterior fragility parameters (Am and BU), two methods have been proposed. Furthermore, it has been found that the importance of evidence used in the Bayesian methodology can be quantified by the entropy of the evidence. Only the events with high entropy need to be considered in the Bayesian updating of the fragility. The currently available seismic experience database for typical components can be utilized to develop the fragility tail which is contributive to the seismically-induced failure frequency. The combined use of generic fragility and seismic experience data, with the aid of Bayesian methodology, provides refined generic fragility curves which are useful for SPSA studies. (author)