AbstractFast estimation is a critical feature of the proposed modeling approach as a “forward” fast solver that has a key role in solving the “inverse” problem involved in crack sizing, especially for real‐time applications. To this end, two different inversion methods have been proposed to estimate the crack depth. The first one uses an interpolation model of the 3D direct model simulation data. The second approach is based on artificial neural networks (ANN). In this article, the rotating uniform eddy current (RUEC) probe is used to detect the normal magnetic component Bz which is defined as the characteristic signal for reconstructing the crack length and depth concurrently. Using the first approach, The Bz characteristic 3D surface is presented, and this can be modeled by a fitted polynomial interpolation equation. Thus, the crack depth can be inverted using this equation referring to the experimental or simulated Bz signal and the crack length. The ANN is outlined to determine the crack depth based on the simulated Bz signature. The move from the first inversion method to the second was flexible and useful, in which the interpolation model is used as a defect signal generator for fast building of a large and efficient database for ANN training. Both of the proposed methods proved the objectivity and accuracy of the inversion results and offered more robust engineering support for automated NDT, reducing production time and increasing productivity.
Read full abstract