ABSTRACT
This study integrates measured data with a high-fidelity finite element (FE) model to develop a physics-based digital twin for PC girder bridges using a Bayesian model updating framework. The Fourier Neural Operator (FNO) serves as a surrogate model for Bayesian model selection and parameter estimation, while Transitional Markov Chain Monte Carlo (TMCMC) estimates posterior probability density functions (PDFs). By pre-training the FNO on a small dataset, it approximates PDE solutions and accelerates Bayesian updates. The Gaussian random field and random variables enable Bayesian model selection, informed by estimated model evidence. The updated model’s nonlinear load-deflection curves align with experimental data, and TMCMC-derived parameters yield FE simulations that match FNO estimates. The FNO-aided approach is over 15 times faster than conventional FE model updating, demonstrating both accuracy and efficiency. This method enables near real-time structural assessment, advancing model-based structural health monitoring (SHM) for PC girder bridges.
