Solution of Inverse Problems in Electromagnetic NDT Using Neural Networks

This paper presents a technique for solving inverse problems in electromagnetic nondestructive testing (NDT), using neural networks (NN). They are trained to approximate the mapping from the signal to the defect space. A crucial problem is signal inversion, wherein the defects profiles must be recovered from calculated signals by using finite element method (FEM), this method give good results by using the refinement mesh but in very long time. The idea of this paper is the exploitation of the FEM but with a middle mesh where the results are approached in short time. This signal was exploited in the inversion problem, where the maps represent the defects in the plate. The inversion results obtained with the NN are presented. The presented approach has permitted to realize good maps in a very reasonable training time with respect to others approaches.