Response Prediction of Stochastic Dynamics by Neural Networks: Theory and Application on Railway Vehicles

Stochastic dynamics is a research topic for railway vehicles involving a wide range of randomness or uncertainty. However, the modeling and calculation of stochastic dynamic systems are often high-cost and low-efficiency. Neural network is an effective machine learning tool driven by data; this paper devotes to bridge the gap between neural networks and stochastic dynamics and to attain proper uses of this technique in railway vehicles. The mapping capability of neural networks for various stochastic suspension dynamics is validated by the proposed random repetition scheme. And this powerful computational tool is applied to predict the dynamic performance of high-speed trains in service instead of dynamics calculations; a typical case is analyzed to emphasize the advantage of the dynamic performance evaluation considering the coupling of various factors that it can enhance the security and reliability by attaining prognostic and health management and condition-based maintenance.