Automatic eddy current classification of signals delivered by flaws – application to nuclear fuel cladding

The paper proposes an automatic system for the classification of flaws using eddy current testing. The classification system contains two main blocks: a data pre-processing block based on a sliding window filtering and a Neyman-Pearson detector for evaluating whether or not the zone of a signal contains information about a flaw, with a fixed probability of detection. The data post-processing block contains a system for the normalisation of signals in both amplitude and phase, a feature extraction system based on the calculation of the 6th to 10th harmonics in the Fourier space of signals corresponding to flaws, a system for the calculation of the Mahalanobis distance between the clusters, previously constructed for the types of flaws that can appear on nuclear fuel cladding, and a classification into a class of defects depending on the minimum Mahalanobis distance. The system has been tested, with a 95% probability of detection and a 95% reliability coefficient, on nuclear fuel cladding made of Zircaloy 4 from a PHWR nuclear power plant.