A NEURAL NETWORK APPROACH FOR WHEEL-RAIL FORCES PREDICTION

In railway system, track inspection vehicles regularly measure the vertical and lateral irregularities of each rail in order to ensure the safe running of railway. However, more and more researches indicated that the irregularities of rail are not sufficient for safety evaluation, and that wheel-rail forces should be considered. The instrumented wheelsets that have been used for measuring wheel-rail forces are too expensive and costly. In this paper, a neural networks based method to predict wheel-rail forces by the measured track irregularities is proposed. The track irregularities including vertical and lateral irregularities, cross-level, curvature and gauge are used as inputs for neural networks, and the wheel-rail forces are the predicted outputs. The neural network employed in this paper id the widely used multi-layer perceptron neural network. The data of track irregularities are collected by a certain track inspection vehicle, and the data of wheel-rail forces are generated by the ADAMS/Rail software. In order to speed the convergence the neural networks, the Levenberg-Marquardt learning algorithm is employed to train the neural networks. The numerical simulations are carried out and the simulation results show that the proposed technique can predict wheel-rail forces precisely.