Bayesian nonlinear finite element model updating of a full-scale reinforced-concrete bridge column under low-to-moderate seismic excitation

This study investigates the capabilities and limitations of Bayesian nonlinear finite element (FE) model updating for extracting physically meaningful information under low-to-moderate seismic excitation. A distributed-plasticity FE model of a full-scale reinforced-concrete (RC) bridge column tested on the UC San Diego Large High-Performance Outdoor Shake Table is calibrated using the Transitional Markov Chain Monte Carlo (TMCMC) method and the first three earthquake ground motions (EQ1−EQ3) of the test sequence in a cumulative data assimilation framework. The analysis systematically evaluates parameter identifiability, the influence of prior bounds, the physical consistency of Rayleigh damping formulations (mass plus initial elastic stiffness versus mass plus tangent stiffness, and global versus event-specific damping coefficients), and the dependence of inherent damping on excitation intensity across these events, transitioning from quasi-linear to moderately nonlinear structural response regimes. Results show that low-intensity excitation (EQ1) provides insufficient information to uniquely constrain material and damping parameters, often leading to bound-driven or compensatory parameter interactions associated with low identifiability and yielding deceptively good fits with non-physical parameter estimates. In contrast, stronger excitation (EQ3) activates nonlinear mechanisms, producing stable, physically interpretable parameter estimates and enabling robust generalization to unseen seismic excitations. The study also shows that the choice of stiffness matrix in Rayleigh damping, initial elastic versus tangent, governs the response amplitude dependence of the instantaneous modal damping ratios which provide a physically meaningful characterization of inherent damping. This behavior reflects compensatory interactions between parameters controlling inherent damping and those governing hysteretic energy dissipation. Finally, this study underscores the need for large-scale experimental and field data, beyond numerically simulated data and small-scale experiments, to support the development and validation of physically meaningful Bayesian nonlinear FE model updating methodologies for digital twin and real-world applications.