ML Framework Detects Railway Wheel Defects Using Passive Ultrasonics

Researchers have developed a machine learning-based system to diagnose various defects in railway wheels using passive ultrasonic technology. The system relies on acoustic emission signals captured without direct contact, offering a non-invasive inspection method crucial for railway safety and maintenance. The framework processes data collected from eleven full-scale railway wheelsets, representing nine distinct health conditions. It uses statistical tests like Kruskal-Wallis and mutual-information analysis to pinpoint the most effective features from both time- and frequency-domain signals. Key discriminative indicators identified include decay rate, kurtosis, skewness, and envelope low-frequency power. A Random Forest classifier, trained with these selected features, achieved a balanced accuracy of approximately 0.66 and a Macro-F1 score of 0.65 across the nine defect classes. This demonstrates the viability of combining passive ultrasonic sensing with machine learning for automated, non-contact defect classification, laying the groundwork for future field-deployable inspection systems.