Wayside monitoring for railway wheel flat identification: A multiclass supervised learning approach

Wheel defects are a significant concern in rail transport, impacting safety and increasing infrastructure wear. This study presents an automated framework for early detection and severity classification of wheel flats using data from a single wayside-mounted accelerometer. Acceleration signals are processed to extract time-domain features, which are then classified using machine learning (ML) algorithms, including K-Nearest Neighbors (kNN), Multi-Class Support Vector Machine (MSVM), Decision Tree (DT), Ensemble Tree (ET), Gaussian Mixture Model (GMM), Naive Bayes (NB), Random Forest (RF), and Discriminant Analysis (DA). The results indicate that the proposed approach consistently achieves over 90% accuracy in detecting and classifying wheel flats, irrespective of the sensor’s placement—whether on the rail or between sleepers—and with the use of only a single sensor. This cost-effective and scalable approach minimizes sensor requirements, making it practical for widespread implementation across large-scale railway networks while ensuring high accuracy in defect detection.