Ahlbrink, Ralf; EURAILSCOUT Inspection & Analysis; Germany
Gao, S.; Eurailscout Inspection & Analysis; Germany
Szugs, T.; Eurailscout Inspection & Analysis; Germany
Ahlbrink, R.; Eurailscout Inspection & Analysis; Germany
Session: Transportation, railway & automotive 2
Time: 15:40 - 16:00
Rolling contact fatigue is caused by interaction of train wheels with railheads. To minimize operational risks railways infrastructure owners can ask for regular inspections to determine critical sections of assets in a predictive maintenance scheme.
Some defect types, for instance (sub-)surface anomalies like squats, were herein assessed with our train UST02 by specific ultrasonic (UT) and eddy current (ET) testing probes, but each with inherent limitations. Additionally surface imaging video (VT) was recorded.
Machine learning techniques were applied to combine data from these three inspection systems from a recorded track, which had a squat defect population. The target was to provide optimized classification as basis for prediction of rolling contact fatigue growth.
We describe results of this ‘combined systems method’, CSM, with classification features derived from densely spaced clusters with input from UT B-scans, and parameters from ET and VT raw data after signal and image processing, respectively. In comparison to classification per single system, CSM gave significantly improved hit rate without degradation of specificity.