Global sensitivity measure for uncertainty distribution parameters and effective solution for obtaining it

For the engineering structure involving uncertain distribution parameters, a global sensitivity measure based on failure probability is established. To get the effect of uncertain distribution parameters on the failure probability, traditional Monte Carlo method generally needs a "triple-loop" crude and time consuming sampling procedure to compute the established global sensitivity. To overcome the disadvantage of MC method, we propose an improved sampling method for global sensitivity measure of failure probability, in which the triple-loop is simplified into a "double-loop" and the computing efficiency is greatly improved. The main idea of the proposed method, which is explained in section 1 and 2 of the full paper, consists of: (1) generating samples, (2) searching suitable estimators and establishing the relationship between failure probability based global sensitivity measure and the estimators, (3) obtaining the global sensitivity measure. Compared with the traditional MC method, the proposed method is more efficient for the same acceptable precision, due to the fast convergence of the estimators. Calculated results of 1 numerical and 2 engineering examples, presented in section 3, and their analysis demonstrate preliminarily the reasonability of the proposed sensitivity measure and the efficiency of the proposed method.