Sequential machine learning based fragility analysis: Sequential ML-FA for reactor containment vessel subjected to internal pressure

With the introduction of probabilistic safety assessment in nuclear power plants, a fragility analysis is critical to evaluating the probability of a failure of structures. However, such fragility analysis requires a large amount of finite element analyses due to explicit consideration and quantification of all sources of uncertainty. This study aims to present a sequential machining learning-based framework that can sequentially and efficiently estimate the fragility of containment vessels in nuclear power plants while minimizing finite element analyses, and the proposed framework is applied for performing a fragility analysis of a prestressed concrete containment vessel subjected to internal pressure. Within the framework, machine learning models are used to predict the behavior of the containment vessel based on collected analytical data from finite element analyses. The predicted data are used to estimate fragility curves through maximum likelihood estimation within the proposed framework, and the number of analytical data for training machine learning models is sequentially increased until the required convergence index of the estimated fragility curve is reached. In addition, the final fragility curves obtained from the proposed framework are compared with the fragility curves (benchmark) obtained from 1000 analytical data. This proposed framework can significantly reduce computational costs by estimating the fragility curve with the minimum number of finite element analyses.