On the Importance of Uncertain Parameters in Seismic Probabilistic Fragility Assessment

Seismic fragility analysis is essential for establishing the reliability of structures and components of nuclear reactors and is an integral part of the seismic probabilistic risk assessment process. However, importance measures capable of communicating risk analysts the factors on which to focus modelling activity and data collection are missing. We investigate the decision-support criteria of interest, namely, the mean fragility curve and the "High Confidence of Low Probability of Failure" value. We define three sensitivity measures consistent with these criteria. Because fragility curves are cumulative distribution functions (CDF), we study the properties of CDF-based importance measures. The definitions are then specialized to the Electric Power Research Institute safety factor method, which is nowadays used worldwide for seismic risk assessment of nuclear plants. We show that, for this method, the importance measures of safety factors can be obtained analytically. We illustrate the methodology by application to an industrial case study, namely the fragility analysis of a nuclear reactor service water pump. Numerical findings reveal that the joint use of the three sensitivity measures provides risk analysts with a rigorous way to detect the most important factors and screen-out the non relevant ones.