Global sensitivity analysis based on distance correlation for structural systems with multivariate output. (15th July 2018)

• Record Type: Journal Article
• Authors: Xiao, Sinan; Lu, Zhenzhou; Wang, Pan
• Is Part Of: Engineering structures. Volume 167(2018)
• Page Start: 74
• Page End: 83
• Publication Date: 2018-07-15
• Subjects: Multivariate global sensitivity analysis -- Multivariate output -- Dependence measure -- Distance correlation; Structural engineering -- Periodicals; Structural analysis (Engineering) -- Periodicals; Construction, Technique de la -- Périodiques; Génie parasismique -- Périodiques; Pression du vent -- Périodiques; Earthquake engineering; Structural engineering; Wind-pressure; Periodicals; 624.105
• DOI: 10.1016/j.engstruct.2018.04.027
• Languages: English
• ISSNs: 0141-0296

Abstract

Highlights: The distance correlation is used for multivariate global sensitivity analysis. The proposed method can be regarded as an extension of the PCC based method. The whole probability distribution of multivariate output is considered. The proposed sensitivity index can be easily estimated with higher robustness. Abstract: Models with multivariate output are often used in practical structural systems. Multivariate global sensitivity analysis (GSA) plays an important role in quantifying the contribution of uncertainty in the model input to the output, which is quite useful for simplifying the models and improving the model performance. Many traditional global sensitivity indices can be considered as dependence measures of model input and output. However, these dependence measures can only measure the dependence between two scalar random variables. In this paper, the distance correlation, which can measure the dependence between random vectors, is utilized for multivariate GSA for structural systems with multivariate output. The distance correlation based sensitivity index not only considers the whole probability distribution of multivariate output but also can be easily estimated with only a single set of input–output samples. A numerical example is adopted at first. Then the distance correlation based sensitivity index is applied to a vibration problem in structural dynamics, a truss structure model and a wing box structure model. The results show that the distance correlation based sensitivity index has a higher robustness compared to the covariance-decomposition based sensitivity index.