Bayesian Nonlinear Structural Model Parameter Updating Using Visual Damage Indicators

Abstract

Postearthquake decisions regarding the repair, retrofit, and safety of damaged infrastructure often rely on accurate estimates of structural response demands. These estimates are highly sensitive to uncertain material parameters whose true values often deviate from those used in the numerical model. This study proposes a Bayesian framework for updating nonlinear structural model parameters using postearthquake visual damage indicators (VDI) as evidence. Observed damage is probabilistically linked to engineering demand parameters (EDPs) through limit-state relationships defined in established literature and standards of practice. Prior parameter distributions are propagated through Monte Carlo nonlinear response history analyses with and without record-to-record ground-motion uncertainty to obtain probabilistic EDPs. These are then used to compute likelihoods and update both marginal and joint parameter distributions. The framework is demonstrated using an experimentally tested reinforced-concrete structural wall. Updated maximum-a-posteriori model parameter estimates are shown to substantially improve predictions of global and local seismic response demands relative to priors and produce fragility curves that differ appreciably. Results also show that incorporating parametric correlations can further enhance these estimates. The proposed methodology provides a practical pathway for leveraging observed damage to improve postearthquake performance assessment and decision support.