Quantitative eddy current testing using radial basis function neural networks

Abstract

For quantitative eddy current testing in the presence of disturbing variables, radial basis function neural networks have been investigated. The digitized in phase and quadrature data from a dual frequency eddy current instrument have been given as input to an optimized radial basis function network to test quantitative information online. The generalization, interpolation and extrapolation abilities of the radial basis function network have been studied on stainless steel plates having a variety of machined notches of varying length and depth in regions of permeability variations. The radial basis function network could detect and quantify notches with a maximum deviation in depth quantification of 35 μm (1.4 x 10^-3 in.). The radial basis function network technique has also been applied to another industrial application involving quantification of thickness of stellite coating on carbon steel and that of NiAl coating on UNS N07718.