A high-precision surrogate model for seismic vehicle-track-bridge system based on hybrid deep learning for nonlinear structural restoring forces

Abstract

This paper presents a novel high-precision surrogate model for seismic vehicle-bridge interaction, leveraging hybrid deep learning techniques to predict nonlinear structural restoring forces. By integrating deep learning predictions within a coupled high-speed vehicle-track-bridge (VTB) system model, this approach offers a significant advancement in simulating the complex nonlinear hysteretic behaviour of critical track-bridge components during seismic events. The innovative surrogate model effectively replaces traditional finite element-based nonlinear components with a machine learning-driven solution, thereby enhancing both computational efficiency and accuracy. Extensive evaluations under varying seismic intensities confirm the model’s precision in capturing structural and vehicular responses, as well as performance metrics related to vehicle derailment during earthquakes. The results demonstrate the robustness of the hybrid deep learning approach in accurately predicting dynamic responses and mitigating the risks of high-speed train derailments on seismically impacted bridges, making it a valuable tool for safety assessments in high-speed rail infrastructure. The methodology and code implementation are publicly available at https://github.com/kanepro1998/Surrogate-Model.