Acoustic monitoring of rail faults in the German railway network

Paper i proceeding, 2019

The early detection of rail surface defects such as squats, poor welds, or wheel burns is important to prevent further rail deterioration. In this paper, a methodology for acoustic monitoring of squats in the German railway network is proposed based on the measurement of axle box acceleration (ABA) on the DB noise measurement car (SMW) and the previously developed numerical model WERAN for wheel/rail interaction. Specific characteristics of squats in the ABA signals are determined with the model and verified by pass-by measurements combined with direct geometry measurements of the squats. Based on these re- sults, a logistic regression classifier is devised for the detection of squats in the measured ABA signals of the SMW. Trained with simulated and measured data, the classifier identifies all of the known severe squats and 87% of the known light squats in the measured test data.