Adaptive sampling-based RBDO method for vehicle crashworthiness design using Bayesian metric and stochastic sensitivity analysis with independent random variables

For many engineering design problems, traditional most probable point (MPP)-based reliability analysis using sensitivity information to find the MPP is difficult for practical use. In addition, the sensitivities of performance function are often unavailable for problems such as crashworthiness. Using Monte Carlo simulation method to calculate the sensitivities of probabilistic responses, which are often obtained by using finite difference method, is very time consuming and inaccurate. This paper presents a stochastic sensitivity-analysis method for computing the sensitivities of probabilistic response by using Monte Carlo simulation incorporated with a metamodel, which is selected by using Bayesian metric considering data uncertainty. An adaptive sampling-based RBDO methodology based on Bayesian metric and stochastic sensitivity analysis is developed for design optimisation of large-scale complex problems. This method not only produces an accurate metamodel, but also yields an accurate optimal design efficiently. This methodology is demonstrated by a crashworthiness optimisation example.