A neural-network classifier of flaws for multifrequency Eddy-current tests of heat-exchange pipes

A neural-network classifier has been developed that evaluates the geometric characteristics of a detected flaw on the basis of the parameters of the corresponding multifrequency signals obtained via scheduled eddy-current tests conducted with a through-type probe. The classifier is intended for testing heat-exchange pipes in steam generators of a nuclear power plant with a water-moderated water-cooled power reactor. The representative library of signals required for the design was formed on the basis of theoretical and experimental data. The theoretical data were obtained in a numerical physical and mathematical model of the electromagnetic testing procedure implemented with the MagNum3D program for finite-element analysis. The experimental data were obtained through measurement of multifrequency eddy-current signals from test specimens with artificial flaws.