Smart Rail Infrastructure: Onboard Monitoring with Machine Learning for Track Defect Detection

Detecting damages in railways is essential for ensuring the safety and reliability of train operations.This study introduces a methodology for detecting damage in railway tracks by employing an onboard monitoring system installed on a freight wagon.The methodology proposed here adopts a comprehensive approach involving data acquisition, feature extraction, data fusion, and outlier analysis.Initially, data is collected using the onboard monitoring system, capturing diverse responses from both the axle box and carbody during wagon operation.Subsequently, feature extraction is conducted on these acquired responses utilizing continuous wavelet transform techniques.Additionally, feature normalization via principal component analysis is applied to mitigate environmental and operational variations, enhancing sensitivity to damage detection.The Mahalanobis distance is then employed to merge features, yielding a damage index for each scenario.Finally, the fused features undergo classification using appropriate machine learning algorithms to distinguish between undamaged and damaged tracks.This methodology promises to enhance railway maintenance practices by offering an automated and dependable approach for detecting damages in railway tracks.