An optimized strategy for using asymptotic sampling for reliability analysis

Asymptotic sampling is a recently developed method for calculating small failure probabilities of high-dimensional problems. It is based on Monte Carlo simulation, however, extremely small probabilities can be estimated with reasonable computational effort. The present paper presents an optimal strategy for using asymptotic sampling. In particular, it is explained in which cases the accuracy of the results can be increased by applying low-discrepancy sampling. A new surrogate technique is presented for obtaining accurate results also in cases in which low-discrepancy sampling is not applicable effectually. Furthermore, an extension of the method is shown for estimating threshold exceedance probabilities.