A novel Bayesian-inference-based method for global sensitivity analysis of system reliability with multiple failure modes

Abstract

Global reliability sensitivity analysis aims at quantifying the effects of each random source on failure probability or reliability over their whole distribution range and is highly concerned in reliability design and uncertainty control. And in practice, a structure or product usually has more than one component impacting their performance safety, which is essentially a system reliability problem. Therefore, this paper proposes a novel Bayesian-inference-based method for moment-based global sensitivity analysis of system reliability with multiple failure modes. First, the limit-state function of each component involved in the system is linearly approximated based on the reliability index. Then, the global reliability sensitivity is transformed into a problem of multivariable Gaussian probability within a given safe region where the dimension number is double of the failure modes. In this case, the Bayesian-inference-driven expectation propagation technique is introduced to solve this intractable problem in an analytical manner, based on which the closed-form solution to the global reliability sensitivity for system with multiple components is accordingly derived. Finally, a numerical case, a vehicle subjected to impact, a cantilever beam and a practical engineering application to a four-panel spaceborne deployable plane antenna are studied to demonstrate the effectiveness of the proposed method by comparison with Monte Carlo simulation.