On-line rail defect diagnosis with differential eddy current probes and specific detection processing

This paper presents a non destructive evaluation system for rail inspection with a device embedded in a subway train. The difficult operating conditions lead to the design of a specific double-coils and double-frequencies sensor that delivers 8 complementary signals. After the presentation of the technological solution, the paper deals with the detection of minor defects and presents three approaches: the first one is a time heuristic approach, the second one is based on wavelet basis projections, and the last one implements an inverse filtering which allows the transformation of the differential probe outputs into estimated absolute ones. The methods are validated statistically on a labeled defect data set with the help of ROC curves.