Design of an Optimal Neural Network for Evaluating the Thickness and Conductivity of the Metal Sheet

This paper presents the application of non-destructive evaluation by eddy currents for the determination of the geometrical and physical parameters of metal sheet, obedient to a sensor of a double coil (method of Adding-Opposing (AO)). The forward problem is solved by using an analytical model. The electrical impedance for coil is measured for two frequencies ranging from 1 kHz and 1 MHz. The inversion method is implemented using neural networks; it consists to introduce the real and imaginary parts of the impedance for the evaluated thickness and conductivity. The neural network (NN) implementation of this problem is determined by the split-sample method and the adjustment of the internal parameters of the neural networks so as to minimize the mean square error (MSE). The inversion results obtained with both NN (MLP and RBF) are presented and compared. The presented approach has permitted to achieve good parameters estimation in a very reasonable training time with respect to others approaches.