Dynamic response prediction of vehicle-bridge interaction system using feedforward neural network and deep long short-term memory network

Vehicular loads represent one of the most critical external dynamic actions on the bridge structures, especially the short- and medium-span bridges. The dynamic interactions between the vehicle and bridge become more and more significant due to the rapid development of transportation nowadays. This paper proposes a methodological framework for efficient and accurate prediction of dynamic responses of the vehicle-bridge interaction system using artificial neural networks. Based on the developed vehicle-bridge system model composed of 3D train vehicle model, bridge finite element model, and wheel-rail contact model, numerical experiments are carried out to produce data required for training neural networks. Feedforward neural network and deep long short-term memory network are developed and designed to forecast various dynamic responses of both the vehicle and bridge in time- and frequency-domain. The proposed framework is illustrated on a high-speed train vehicle and bridge system. To further examine its robustness, effects of track irregularity and noise level on the prediction performance are investigated. The present framework can provide an efficient alternative to the vehicle-bridge interaction analysis, and has significant ability of inclusion of field measurement data and promising potential for the realization of online response prediction.