Improved metamodel-based importance sampling for the performance assessment of radioactive waste repositories

In the context of a probabilistic performance assessment of a radioactive waste repository, the estimation of the probability of exceeding the dose threshold set by a regulatory body is a fundamental task. This may become difficult when the probabilities involved are very small, since the classically used sampling-based Monte Carlo methods may become computationally impractical. This issue is further complicated by the fact that the computer codes typically adopted in this context requires large computational efforts, both in terms of time and memory. This work proposes an original use of a Monte Carlo-based algorithm for (small) failure probability estimation in the context of the performance assessment of a near surface radioactive waste repository. The algorithm, developed within the context of structural reliability, makes use of an estimated optimal importance density and a surrogate, kriging-based metamodel approximating the system response. On the basis of an accurate analytic analysis of the algorithm, a modification is proposed which allows further reducing the computational efforts by a more effective training of the metamodel.