Reliability sensitivity analysis method based on weight index of density

In order to improve the efficiency of digital simulation in approximating reliability sensitivity, a method is proposed which works by generating deterministic and low-discrepancy samples uniformly in the design space and applying the value of joint probability density function as a weight index at any sample. The weight indexes ensure the estimated values of the reliability sensitivity are converged to the true values. This way of getting points by low-discrepancy sampling instead of depending on a variable's probability density can ensure smaller error bounds and a higher possibility for the samples to fall into failure domain, so that the convergence speed becomes much higher for small failure probability events. Additionally, the steps to calculate the reliability sensitivity with related variables are the same as those with independent variables, which is another advantage that makes the method simpler and easily applicable. Several examples in this paper demonstrate the advantages of the proposed method sufficiently.