Running Safety Assessment of High-Speed Train–Bridge System During Earthquakes Based on a Wheel–Rail Contact Surrogate Model

Efficient and accurate simulation of the wheel–rail (WR) contact state is vital for assessing the running safety of high-speed trains on bridges during earthquakes. This study proposes a hybrid contact searching method to enhance the accuracy and robustness of contact point detection under extreme conditions. Based on this method, a comprehensive data pool of WR contact states is established. A WR contact surrogate model is then developed using a backpropagation (BP) neural network, enabling rapid prediction of the creep coefficient. Finally, the surrogate model is embedded into a coupled dynamic system of the high-speed train–bridge, and seismic running safety assessments are conducted under multiple operating conditions. The results show that the proposed hybrid contact searching method is both accurate and reliable. The surrogate model achieves a prediction error within 8%, significantly improving computational efficiency. These findings highlight the promising engineering applicability of the surrogate-model-based running safety assessment approach for high-speed trains running on bridges during earthquakes.