Pattern recognition approaches for the detection and characterization of discontinuities by Eddy current testing

Abstract

Eddy current signals (ECS) generated under varied experimental conditions from different types of discontinuities like partial/through thickness holes and notches of various dimensions, fatigue cracks, stress corrosion cracks, etc. in AISI type 316 stainless steel sheets/plates have been analyzed using pattern recognition (PR) approaches to understand their quality of performance for detection and characterization of several aspects of the discontinuities. The PR analyses have been carried out using linear discriminant (LD), minimum distance (MD), empirical Bayesian (EB) and K-nearest neighbor (KNN) statistical classifiers, and multilayered perceptron (MLP) and Kohonen's artificial neural network (KANN). The MLP approach has been extended to eddy current images also to achieve deblurring. The practical feasibility and application potential of ANNs is demonstrated through a case study on nuclear fuel cladding tubes where both the online and the offline approaches have been implemented.