A neural network approach to ECT data inversion for materials quality evaluation

The aim of this paper is to present a novel NDT technique for detecting and estimating the location of a defect inside a conductive object by neural network based processing of eddy-current data. The electromagnetic interaction between the conductive specimen and the eddy-current probe is simulated by a 3D numerical technique, which reproduces the differential impedance profile seen on the specimen's accessible surface, depending on the defect location; the obtained data are used to train a multilayer neural network which provides an analytical approximation of electromagnetic interaction phenomena; a maximum likelihood inversion technique is then proved to be effective in estimating the flaw location.