Wear Prediction Of Rolling Contact Of Rail Steel Based On Experiment Data

In recent years, rail wear are getting more serious with the rapid development of highspeed rail and the improvement of axle load. The alteration in wheel and rail profiles due to wear involves considerable vehicle and track-maintenance costs, and influences the loading capacity of the rails, as well as the operation safety and riding comfort of the vehicles. The complexity of the wheel/rail wear limits the feasibility of wheel/rail wear evaluation, forecast and reliable quantitative calculation in various working conditions. Mining the relationship between influence factors and wear with the analysis of the experiment data is a effective way of prediction of wear volume. Prediction of rail wear can guides the choice of rail material, predicts rail life, helps to make effective overhaul plan of rail, limits the wear of wheel/rail profile, and has tremendous social benefits and economic benefits.This paper achieves the data of rail wear by JD-1rail-wheel tribology simulating machine.Prediction function based on the BP neural network mines the law of rail wear, mainly useing the GA-BP neural network to forecast rail wear volume in the experiment conditions, and hybrid algorithm based on PSO is applied to neural network to solve the right values. Finally numerical calculation is used to local situation analysis of the wheel/rail contact.The main results are as follows:(1) The result of GA-BP neural network is more stable and stronger convergence,avoiding the limits of complex theory and unknown factors.nonlinear mapping is realized between the axle load, speed and attack angle and wear.(2) Using GA-BP neural network simulates the relationship between axle load, speed, attack angle and wear. It indicates that the load (axle load), speed, attack angle (curve radius) have great effect on rail wear.(3) The relationship between the axle load, speed, attack angle and wear abtained by the network simulation shows that Railway construction should try to avoid small curve radius and improve matching speed with it, and train should try to improve the speed to wear smooth stage to extend wheel/rail service life.(4) The PSO hybrid algorithm based on PSO algorithm is applied to solve the weight and threshold values of the BP neural network, avoiding the conventional algorithm(gradient descent algorithm and so on) relying on initial values, greatly enhancing the possibility of optimal weight and threshold values solved, improving the approximation precision of BP neural network, and having excellent fitting features for the samples of wear data.the key to the availability of prediction results based on the PSO hybrid algorithm is the degree of sample data noise and the effectiveness of sample distribution.Because this algorithm has outstanding Approximation performance, but fault-tolerant ability is weaker and more sensitive to the noise of sample data.(5) With the increase of single axle load, the maximum stress of contact patch is increasing. Change rate is maller with heavier axle load; with the increase of single axle load, the change rate of the maximum contact stress become gradually smaller due to the increasing rate of contact area greater than the increasing rate of the single axle load.