On information fusion for reliability estimation with multifidelity models

Multifidelity models attempt to reduce the computational effort by combining simulation models of different approximation quality and from different sources. Information fusion combines outputs from a model hierarchy in order to obtain efficient estimators for a quantity of interest. In this paper, information fusion is applied to reliability estimation. To this end, efficient multifidelity estimators for the probability of failure are developed by combining additive and multiplicative information fusion with importance sampling and importance splitting (notably the moving particles method). Importance sampling and importance splitting based multifidelity reliability estimators are compared focusing on relative error and coefficient of variation.