A COMPARATIVE-STUDY OF CONVENTIONAL AND ARTIFICIAL NEURAL-NETWORK CLASSIFIERS FOR EDDY-CURRENT SIGNAL CLASSIFICATION

A series of eddy current signal trajectories have been obtained on art ificial round and rectangular defects in thin stainless steel plates. These signals have been processed to achieve a single waveform charact erising the eddy current trajectories from which a large number of fea tures have been derived both in the time and frequency domains. The op timised number of features to characterise a defect has been ascertain ed with the help of four reported conventional classifiers and an Arti ficial Neural Network (ANN) classifier. A comparative assessment of th e potential of these classifiers has been carried out within the domai n of the given signal trainings in this investigation. The results ind icate that only five signal features are sufficient to gain an underst anding about the nature of defect.