Seismic derailment-induced train–wall impact on high-speed railway bridges: A two-stage numerical and data-driven assessment framework

Earthquake-induced derailment can lead to train impacts against bridge protective walls, but the collision state is rarely derived from the preceding derailment motion. This study links a train–track–bridge seismic model to a train–reinforced concrete wall impact model to assess the resulting spatial intrusion. The framework transfers pre-contact attitude, velocity, and position and measures the maximum leading-car upper-body volume fraction beyond the wall plane within 1.0 s. For the investigated CRH2 train, China Railway Track System type III (CRTS III) slab track, and six-span bridge, 182 of 225 seismic cases enter the right-wall terminal class. A balanced database of 400 impact simulations supports comparison of six surrogate models and evaluation of prediction uncertainty. Gaussian process regression achieves a root mean square error of 0.0384 and a coefficient of determination of 0.927 on 80 independent holdout cases. Its 95% conformal prediction interval provides 97.5% empirical marginal coverage. Motion variables account for 93.6% of normalized positive permutation importance, while increasing stirrup diameter reduces the predicted exceedance ratio by about 0.069 at a fixed high-demand state. These results show that collision demand depends strongly on the inherited motion state, with wall reinforcement providing additional restraint. The framework supports preliminary comparison of protective-wall options and prioritization of detailed analyses; applications to other railway systems require independent validation and calibration.