Adaptive sampling with neural networks for system reliability analysis

There are two types of uncertainties in engineering, and performance functions are usually implicit functions involving simulations. In this paper, a neural network-based reliability analysis method for structural systems with mixed variables is proposed. The proper intervals for p-box variables are selected, then a new learning function is proposed. The stopping iteration is used to terminate the proposed algorithm. The lower and upper bounds of probability of failure are calculated based on the final constructed surrogate models. The proposed reliability analysis method can be used for systems with mixed variables. The proposed method is applicable to any existing surrogate models. A numerical example is investigate to show the applicability of the proposed method.