Variational Identification of Nonlinear Dynamics for Seismic Response Estimation

Key Points

• The aim is to improve seismic response estimation and uncertainty evaluation in structures exhibiting complex elastic behavior.
• Utilized a data-driven nonlinear system identification technique based on variational inference.
• Applied Bayesian estimation to model linear regression for state equations during strong seismic events.
• Extended basis functions to incorporate hysteretic restoring forces and elastic-plastic responses.
• Tested the method on one-degree-of-freedom systems and finite element models of steel bridge piers.
• Demonstrated high-accuracy estimation of displacement responses, including residual displacements.
• Enabled quantification of uncertainties associated with the estimated seismic responses.
• Showed potential to better account for modeling errors and observational uncertainties.