Global sensitivity analysis for multiple importance sampling centres using a novel adaptive line sampling method

Abstract

To achieve more efficient resolution of reliability analysis problems in engineering, it is essential to enhance the line sampling (LS) method, which is typically effective for rare events. This article proposes an enhanced method for rare events by combining LS with kriging. In this method, the updating of kriging is conducted from a one-dimensional perspective along the important direction. The samples used to assess the failure probability are obtained through importance sampling (IS) from multiple IS centres rather than Monte Carlo simulation (MCS). During each iteration, both upper and lower bounds of the failure probability are simultaneously evaluated to determine whether to halt the iteration process. Finally, global sensitivity analysis is conducted by filtering the samples generated at multiple IS centres, rather than generating MCS samples. By comparison with existing methods, this article demonstrates the efficiency and accuracy of the proposed method through three numerical examples and an engineering problem.

KEYWORDS:

• Line sampling
• rare events
• active learning kriging
• global sensitivity analysis

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The simulation data within this submission are available upon reasonable request to the corresponding author.