Reliability sensitivity analysis via the likelihood ratio method

This paper investigates the likelihood ratio (LR) derivative/gradient estimation approach to sensitivity analysis in reliability applications. The implementation of the LR method in the two commonly adopted approaches to Monte Carlo simulation in reliability analysis, namely the component-based approach and the system-based approach, is first worked out. To speed up the simulation, we then apply the LR method within the framework of Z-VISA, an importance sampling scheme we have developed recently. A simple example with analytical solution available is studied to test the effectiveness and behaviors of the LR method when applied to reliability sensitivity analysis and also the effectiveness of the Z-VISA biasing technique to reduce the variance of LR derivative estimators.