On multinormal integrals by Importance Sampling for parallel system reliability

The ability to compute multinormal integrals to any required accuracy is a key issue for an efficient computation of failure probabilities, particular important in context with system reliability analysis. Hence in this paper, an accurate Importance Sampling procedure to compute multinormal integrals in high dimensions is presented. The novel method allows to sample exclusively in the failure domain which substantially increases the efficiency of the Importance Sampling procedure. The proposed approach is extended for slightly non-linear limit state functions which typically result from non-Gaussian distributed input variables and linear limit state functions of the response in linear structural analysis. ggested procedure is easy to implement, accurate and convenient for practical applications.