Abstract: To study the relationship between wheel wear and equivalent conicity of high-speed trains, the UIC519 integration method is applied in combination with partial least square, backpropagation neural network, and Ridge Regression models. The wear of LMA-type wheel profile and its relationship with equivalent conicity are analyzed both qualitatively and quantitatively, resulting in the establishment and validation of mathematical models based on these three methods. The study results indicate that the Ridge Regression model is more suitable for analyzing the correlation between wheel wear and equivalent conicity in high-speed trains, effectively capturing the impact of wear on equivalent conicity. Further analysis shows that when the wheel wear reaches 0.42～0.84 mm, corresponding to a mileage of approximately 57 000 km to 122 000 km for high-speed trains, concave wear is observed near the nominal rolling circle of the wheel profile. This phenomenon significantly influences the variation in equivalent conicity, leading to larger modeling errors in this range. It is further demonstrated that the primary factor affecting the relationship between wheel wear and equivalent conicity is the concave wear of the wheel profile.

Key words: wheel wear, equivalent conicity, partial least squares regression, back-propagation neural network, ridge regression model