Stratified sampling algorithm based reliability sensitivity

Reliability sensitivity analysis is implemented by the stratified sampling technique in this paper. The basic idea of the stratified sampling method can be stated as follows. The random space is divided into many exclusive sub-spaces firstly, and then the number of the samples lying in each sub-space is determined by its' contribution to the estimator of the reliability sensitivity. More failure samples can be obtained by this strategy than by the crude Monte Carlo simulation. Hence, the corresponding variance of estimator of the reliability sensitivity can be reduced evidently and the convergence rate can be improved remarkably. The expression for the reliability sensitivity, the variance and variation coefficient of the estimators are derived in details. Three examples are employed to demonstrate the efficiency and accuracy of the presented approach. The results' comparison between the proposed technique and the Monte Carlo method is given. The results show that the presented method can obtain satisfying results with high efficiency and is versatile for single failure model, serial system, parallel system, etc.