Research on Prediction of Metro Wheel Wear Based on Integrated Data-Model-Driven Approach

Existing research on wheel wear prediction uses either data-driven or model-based methods. However, due to the high reliability and limited sample characteristics of metro wheel wear, data-driven methods are not accurate enough and require relatively high data costs, and model-based methods mainly lack verification with measured data and generalization ability. To address the shortcomings of the two types of methods, a new approach combining model-based and data-driven methods is used to predict wheel wear in this paper. First, the least-squares algorithm is used to analyze and calculate the difference between the wear measurement for a specific running mileage and the corresponding simulated wear, with the minimum difference taken as an objective function. By means of optimization algorithms including Genetic Algorithm, Particle Swarm Optimization, Tabu Search and Simulated Annealing, the wear coefficient k in Jendel wear model is optimized, thereby obtaining an optimized Jendel wear model. Later, metro wheel wear for additional running mileage is simulated and predicted through combined application of the vehicle system dynamics, wheel-rail contact, and optimized Jendel wear models. Finally, the paper analyzes the wear prediction results obtained by the integrated data-model-driven approach and compares them with the results of traditional methods and measured data. The results suggest that the integrated data-model-driven approach effectively reduces the uncertainty in selecting the wear coefficient by experience, lowers the experimental data costs, and improves the wear prediction accuracy. Therefore, it is a promising approach to wheel wear prediction.