A fast approximate method for parametric probabilistic sensitivity estimation

The analysis of parametric probabilistic sensitivity analysis is important for reliability-based design, which shows changes of system reliability caused by the change of basic variances. In this paper, a fast approximate method of reliability analysis based on the commercial FE simulation-artificial neural network-Monte Carlo simulation is proposed, which can save the calculation cost with efficient precision. With this quick-response model, a new scaling parameter is presented here considering the global dispersity of stochastic parameters, and the parametric probabilistic sensitivity is analyzed too. The sensitivity indices can be computed by the simple and approximate formula in engineering. A numerical example is presented to validate the accuracy and efficiency in reliability and parametric probabilistic sensitivity by comparing with the analysis of ANSYS.