Predictive models for evaluating seismic demands for bridge portfolios

• Developed closed-form predictive models for seismic demands in highway bridges. • Incorporated both geometric variability and record-to-record uncertainty in predictions. • Predictive accuracy on held-out test set: R² up to 0.93 (means); R² = 0.79–0.88 (dispersion). • Models enable rapid, bridge-specific vulnerability assessments for large inventories. In seismically active regions, vulnerability assessments of highway bridges are essential to ensure infrastructure resilience and support risk-informed management decisions. Existing fragility models that are often utilized for bridge portfolio assessments, lack the level of specificity required to enable an informative, structure-specific seismic evaluation. To address these limitations, this study proposes a predictive framework for generating transparent, closed-form equations for probabilistically estimating key seismic Engineering Demand Parameters (EDPs) of highway bridges in a fraction of time that would have been required for a bridge-specific assessment and of greater accuracy compared to the assessments that adopt class-level treatment. The proposed framework offers estimates for the conditional mean (μ EDP ​) and predicts the conditional dispersion (σ EDP ​) via a log-linear model, that accounts for the bridge geometrical properties to capture structural variability within the considered bridge inventory as well as the record-to-record (RTR) uncertainty. A dataset of 1,800 nonlinear time-history analyses was generated by subjecting 30 bridge prototypes to 60 hazard-consistent ground motions for a 475-year return period. From this, second-order polynomial regression models were calibrated to predict column curvature ductility and elastomeric bearing deformation. On a held-out test set, the models reached R² up to 0.93 for mean predictions (0.93 for mean column ductility; 0.90 for mean bearing deformation) and R² = 0.79-0.88 for dispersion predictions. This provides a practical tool: the closed-form equations enable rapid, geometry-aware demand estimation for portfolio screening.