A Novel Bayesian Framework for Earthquake Damaged Steel Model Updating

This study proposes a novel method for updating steel models based on mode shape to update complex earthquake damaged steel models. By simulating the original structure and inputting a large number of seismic records collected from the surrounding area, the damage indices of numerous structural components were calculated and various post-earthquake mechanical damage states were simulated. These damage states, in conjunction with the underlying mechanism, serve as the foundation for updating the model of the complex structure. The relationship between damage indices of different components is established through neural network. Finally, the Metropolis–Hastings sampling method is employed, combined with previously established parameter relationships by neural network, to generate samples and fit the distribution of posterior probability density function. A numerical truss steel bridge in Japan is built, and the numerical model in different damage states and cases are employed. Both damage cases of the numerical model are updated using the proposed method. The results of updated model demonstrating the reliability and feasibility of the method.