Non-destructive testing of cracks using eddy-currents and a generalized regression neural network (GRNN)

In this paper, we propose a new method for the robust estimation of crack dimensions. The method is based on the eddy current evaluation and a generalized regression neural network (GRNN) scheme. The network is trained by several known crack shapes based on the input impedance of a magnetic probe using a finite element solution for the eddy currents. The target value to be trained was the shape of the crack using a window based on the probe impedance. Noisy data, added to the probe measurements, is used to enhance the robustness of the method. We present a comparison of the results obtained using the proposed method with those obtained from a feed-forward neural network. It is shown that the GRNN is faster both in training as well as in identification of the cracks.