Abstract
Railway infrastructure is the backbone of global transportation systems, necessitating constant attention to safety and reliability. Recent years have witnessed an increasing focus on monitoring rail defects, particularly surface defects known as squats. Traditionally squats are identified through labor-intensive visual inspection, ultrasonic inspection, and eddy currents; the need for more efficient and accurate detection methods has led to Axle Box Acceleration (ABA) use and the development of new methods based on machine vision systems. However, today still challenges persist in terms of quantity of defects generated and consequent severity level assignment. This paper introduces a novel approach to automatically assess rail squat defects through the fusion of machine vision data with vertical acceleration measurements from a train’s axle box.
Our methodology involves data collection from both the machine vision and the ABA systems during regular operations. The integration of these data sources aims to correlate visual anomalies with the wheel dynamic behavior, focusing on squat defect location and severity. This integration aims to identify squats in their very early stage with a rate close to 100% and to automatically classify their severity level.
