A machine learning based voting regression method for adhesion estimation in wheel-rail contact

The majority of control methodologies for railway vehicles depend on adhesion data to attain optimal traction. Therefore, researchers have been actively investigating practical and feasible approaches to ascertain precise adhesion data with the objective of optimising the efficiency of railway vehicle operations. While several methods are accessible for indirectly measuring or estimating adhesion, there remains a need for more accurate and expedient estimation techniques. Therefore, in this study, we propose a novel machine learning-based voting regression (VR) model that can efficiently estimate the adhesion between wheel and rail. The proposed VR model is constructed using weighted averages of individual models such as Histogram-based gradient boosted trees (HistGBT), random forest (RF), and linear regression (LR). It is demonstrated by various real-world measurements that the proposed method achieves a high score of 0.922 R2 for the estimation of the adhesion coefficient and outperforms the benchmark model-based Polach's method.