Research On Measurement Technique Of Wheel/Rail Force Based On Improved BP Neural Network

The wheel/rail contact relationship is one of the important research topics in rail transit. As the train run faster, more intense interaction between wheel and rail, wheel/rail loading is more complicated. The accurate measurement of wheel rail force is of great significance for the stability and safety of vehicle operation system. At present, the most widely method of the measurement of wheel/rail forces is use of the wheel-sets as the force sensor. Organize a certain bridge on the wheel plate to get the relationship between the bridge output and the wheel/rail fore through the calibration test.By using finite element method to simulate the wheel-sets loading conditions to find the best patch positions. According to the simulation results process the instrumented wheel-sets and conduct static calibration test of instrumented wheel-sets for the calibration coefficient and mutual influence coefficient.Artificial neural network has good robustness and fault tolerance, and has strong ability of self-study and parallel processes. The simulation speed of the neural network is much higher than that of the traditional modeling methods. Therefore, in the field of engineering, artificial neural network is widely used in pattern recognition, modeling and control of nonlinear systems. In this paper, for the purpose of accurately predict wheel-rail force, we discuss the wheel-rail research method based on genetic algorithm optimized BP neural network.In this paper, the CRH2wheel is the object of study. The theory of continuous measurement for instrumented wheel-sets is introduced. This paper combines with the finite element method, genetic algorithms and artificial neural networks, analyzes the relationship between the wheel/rail force and the output of the static calibration test of instrumented wheel-sets. Establish the decoupled of the horizontal and vertical of neural network model, different conditions of the calibration test data for the study sample. By repeatedly training and learning, comparing the errors value between General BP neural network and genetic algorithm optimization of BP neural network, GA-BP neural network to obtain smaller error value and better prediction; comparing GA-BP neural network prediction of wheel-rail force and wheel rail force calculated by the calibration coefficient. GA-BP neural network can be applied to the wheel/rail force measurement, the neural network has the application value.