Time-Variant Bayesian Updating Using Surrogate Models for Damage Identification

Abstract

The main objective of structural health monitoring (SHM) is the timely damage diagnosis in infrastructures to support decision-making. Bayesian finite element (FE) model updating is a robust tool for damage identification in the presence of uncertainty, offering a probabilistic approach that accounts for various sources of error and variability. Nevertheless, model calibration based on Bayesian inference often requires high computational resources. Recently, surrogate models and metamodeling techniques have become increasingly prevalent in approximating system responses, reducing the time and resources required. Therefore, leveraging surrogate models for rapid decision support is crucial. This study presents a novel approach for damage identification of civil structures subjected to seismic excitation through time-variant Bayesian updating using surrogate models. The proposed approach begins with the training and validation of a Polynomial Chaos Expansion (PCE)-based surrogate model that replaces a linear FE model, which calculates modal properties such as modal frequencies and mode shapes. Then, the validated surrogate model is used in the Bayesian updating process within the likelihood function, where the time-variant modal properties serve as a target for calibration using the Sequential Monte Carlo (SMC) method. This process enables efficient and precise estimation of the FE model parameters, ensuring accurate detection, localization, and quantification of damage in civil structures during earthquakes. These findings significantly enhance the feasibility and effectiveness of real-time SHM.