Gradient-Free Importance Sampling Scheme for Efficient Reliability Estimation

This work presents a novel gradient-free importance sampling-based framework for precisely and efficiently estimating rare event probabilities, often encountered in reliability analyses of engineering systems. The approach is formulated around our foundational Approximate Sampling Target with Post-processing Adjustment (ASTPA) methodology. ASTPA uniquely constructs and directly samples an unnormalized target distribution, relaxing the optimal importance sampling density (ISD). The target's normalizing constant is then estimated using our inverse importance sampling scheme, employing an ISD fitted based on the obtained samples. In this work, a gradient-free sampling method within ASTPA is developed through a guided dimension-robust preconditioned Crank-Nicolson (pCN) algorithm, particularly suitable for black-box computational models where analytical gradient information is not available. To boost the sampling efficiency of pCN in our context, a computationally effective, general discovery stage for the rare event domain is devised, providing (multimodal) rare event samples used in initializing the pCN chains. A series of diverse test functions and engineering problems involving high dimensionality and strong nonlinearity is presented, demonstrating the advantages of the proposed framework compared to several state-of-the-art sampling methods.