PGV–PRGD-Based Probabilistic Seismic Demand and Conditional Fragility Analysis of a Fault-Crossing Bridge

This study develops a PGV–PRGD-based probabilistic seismic demand and conditional fragility framework for a fault-crossing bridge. Spatially variable ground motions are generated using a broadband hybrid method and applied to a nonlinear OpenSees model. Elastic net regression, SHAP, and permutation importance identify PGV and PRGD as a physically interpretable vector-valued intensity measure. Gaussian process regression is used to establish probabilistic seismic demand models and derive marginal fragility curves and conditional fragility surfaces. Results show that PRGD provides complementary information to PGV and reveals component- and direction-dependent risk differences under fault-crossing seismic input.