A three-stage CE-IS Monte Carlo algorithm for highly reliable composite system reliability evaluation based on screening method

Abstract

This paper proposes an interesting three-stage algorithm targeting at highly reliable high dimension composite system reliability evaluation using Importance Sampling (IS). The central idea is at the first stage (the Screening stage) picking out those bottle-neck components which have the most main impact on composite system reliability indexes calculation. The Screening process is specially customized for composite system to adaptively achieve the recognition process once the bottleneck percentage parameter μ is set reasonably. The relative perturbation value of each component is calculated firstly as the basis of recognition progress. In one time of iterations in recognition progress, a given percentage of the exciting bottle-neck components will be removed. After some iteration, those bottle-neck components will be screened out. The remaining Cross Entropy Optimization and Importance Sampling Evaluation stages are performed only considering the distortion of those bottle-neck components' sampling parameters. Numerical tests show that the method has good estimation accuracy performance and substantial variance reduction on highly reliable high dimension system. This also verifies the existence of degeneracy phenomenon of likelihood with the increase of dimension.