Abstract This manuscript presents a novel, comprehensive study focused on predicting the complex interactions between rail wheels and tracks under wear conditions. The study introduces a unique combination of advanced regression techniques — Gaussian Process Regression (GPR), Support Vector Machines (SVM), Decision Tree (DT), Linear Regression (LR), Non-linear Regression (NLR), and Artificial Neural Networks (ANN) — integrated with a Finite Element (FE) model. The FE model, meticulously developed to simulate the intricate dynamics of wheel-rail interactions, allows for a detailed analysis of wear-related parameters, such as contact pressure, stress distribution, and contact area. A key innovation of this work is the integration of multiple machine learning techniques with FE analysis to predict wear evolution in rail-wheel systems. By validating the FE model against established references and generating an extensive dataset, this study provides a rigorous comparison of each regression method’s predictive accuracy. The findings offer crucial insights into the strengths and limitations of each technique, enhancing the applicability of predictive modeling for real-world railway maintenance. The combination of FE analysis with machine learning techniques constitutes a significant advancement in railway engineering, offering an efficient and reliable predictive framework for managing wear in rail-wheel interactions, with direct implications for optimizing maintenance strategies.
Bendine et al. (Tue,) studied this question.