Eddy currents testing defect characterization based on non-linear regressions and artificial neural networks

Abstract

Feature extraction and defect parameters estimation from eddy current testing data has received special attention in the last years. Principal component analysis, wavelet decomposition and Fourier descriptors are some of the tools used for feature extraction. Particular interest is devoted to using artificial neural networks to perform parameters estimation and profile reconstruction of defects. This work reports the use of non-linear regressions for feature extraction based on the modeling of the measured response by a set of additive Gaussians and artificial neural networks to estimate the width and depth of defects.