Research on Real-time Wheel-rail Force Identification Network of Rail Transit under Compound Line Conditions

Abstract

Objective Wheel-rail force plays an important role in the study of train running state, wheel off roundness and rail corrugation in rail transit. Current wheel-rail force identification methods have some problems, such as difficulty in data collection, high cost and inability to identify wheel-rail force under rail transit line conditions, so it is necessary to study the real-time wheel-rail force identification network of rail transit lines. Method A real-time wheel-rail force identification network of rail transit lines is constructed based on convolutional neural network. According to the measured data of a type of urban rail transit trailing car, the simulation model is established using SIMPACK software. Based on the simulation data, the accuracy and speed of the real-time wheel-rail force identification network for the lateral and vertical wheel-rail forces under working conditions, such as different curve radii lines and different train running speeds are studied. Result & Conclusion The real-time wheel-rail force identification network has an excellent ability to identify the wheel-rail vertical force, with the correlation coefficients all up to 0.99, and the average absolute error about 500 N and only 1% of the true value. The identification ability of the network for wheel-rail lateral force decreases with the increase of train operating speed, but it is still within the acceptable range. The correlation coefficient decreases from 0.93 to 0.65, and the average absolute error increases from 1 480 N to 3 000 N, which is approximately 20% of the actual value. The proposed network has a fast calculation speed and can meet the needs of wheel-rail force real-time identification.

Keywords

• rail transit
• composite line
• real-time wheel-rail force identification network