Global sensitivity indices for multivariate outputs using RBF-PCE metamodel

Global sensitivity analysis aims to identify the most influential input variables of structural systems, and Vector Projection-based Sensitivity Indices (VPSIs) are efficient to character the comprehensive effects of input variables in a model with multivariate outputs. However, these indices were estimated by Monte Carlo simulation, leading to expensive computational costs. Therefore, an efficient surrogate-based method for vector projection-based sensitivity analysis is proposed in this paper. The hybrid metamodel constructed by sparse Polynomial Chaos Expansion (sPCE) and Radial Basis Function (RBF) is generalized to a multiple-output model, and VPSIs are estimated by the known information of the RBF-PCE model. The experiment results show that the proposed method has high accuracy and efficiency with a few samples.