Modularized physics-driven dynamic response prediction of long-span bridge and traffic system subjected to stochastic loads based on neural operators

The rapid and reliable prediction of dynamic responses in long-span bridges (LSBs) and traffic systems considering dynamic interactions under stochastic loads is crucial for understanding the operational risk for LSBs (e.g., structural safety and vehicle driving safety). However, the high computational cost per simulation with existing refined models greatly hinders the ability to conduct studies requiring multiple simulations and/or with long durations covering realistic time-varying load scenarios. To address this challenge, a new modularized physics-driven approach for predicting dynamic responses of long-span bridges and traffic systems is proposed. This approach integrates neural operators and the equivalent dynamic wheel load (EDWL) method to formulate a highly efficient and robust computational platform for both research and engineering applications. By leveraging physics-driven implementations of neural operators, the dynamic responses of the LSB and each vehicle in the stochastic traffic flow can be efficiently derived. This is achieved by solving the dynamic equations with neural operators based on the consideration of LSB/wind/traffic interactions using the EDWL concept. The results derived from the proposed method on a prototype bridge have shown excellent agreement with those from the traditional EDWL numerical method, with an efficiency improvement of over 90%. The novel contribution of this study is the development of the first physics-driven, neural-operator-based method for the dynamic analysis of LSBs as they interact with traffic and other stochastic loads like winds, bearing great potential for near-real-time applications because of its superior efficiency. Furthermore, the modular design significantly enhances the portability and transparency of this method, making it easily adaptable to different engineering applications involving various bridges and/or different on-site load conditions。