Reliability Analysis Using Adaptive Kriging Surrogates with Multimodel Inference

Abstract

This work addresses the issue of model selection in adaptive kriging-based Monte Carlo reliability analysis. It is shown that arbitrary model selection (kriging trend and correlation) can lead to poor probability of failure estimates for complex systems. We propose a method for kriging model development that employs information-theoretic multimodel inference and introduces an averaged kriging model derived from the associated model probabilities. The proposed multimodel kriging model is then integrated into an adaptive sample selection method that merges the surrogate enhanced stochastic search method with a learning function modified from the adaptive kriging—Monte Carlo simulation (AK-MCS) method. The result is an efficient method for a surrogate model–based reliability analysis that converges as fast as, or faster than, the AK-MCS method but with significantly improved robustness providing greater assurance in model accuracy.