Bayesian Model Updating for a Cable-Stayed Pedestrian Bridge Using DREAM and Kriging Model

Abstract

Modeling error and measurement noise are inevitable and lead to a significant discrepancy between Finite element model (FEM) and a real structure. Finite element model updating (FEMU) is, therefore, necessary to match the measured data with a predicted response from FEM for advancing structural health monitoring (SHM). Bayesian approach has been proposed to identify the most probable values (MPVs) of physical parameters and provide parameters uncertainties. However, the current Bayesian approach has challenges in high-dimensional problems and requires high computational costs in the complex structure. In this study, a new Bayesian updating framework is proposed using Differential Evolution Adaptive Metropolis (DREAM) sampling method with a variance-based global sensitivity analysis (GSA) and Kriging model to enhance the Bayesian approachs performance and computational efficiency. Firstly, variance-based GSA is used to eliminate insignificant parameters to measured responses and reduce model dimensionality. Secondly, a Kriging model is employed as a surrogate of the time-consuming FE model for reducing the computational burden. DREAM is essentially a multi-chain sampling method, which parallelly runs different paths for all possible solutions and accurately approximates the posterior distribution density function (PDF) for the Bayesian approach. The demonstration of the proposed updating framework of a real-world cable-stayed pedestrian bridge is presented.