Quantitative Reconstruction of Stress Corrosion Crack by Using Neural Network

As stress corrosion crack(SCC) mainly propagates along grain boundary, it has a very complicated microstructure similar to the branches of a tree, which makes its nondestructive quantitative evaluation more difficult. A method for modeling SCC and a sizing scheme based on neural networks were introduced and evaluated. Impedance signals from a pancake probe scanned just over the crack were taken as the source signal for crack reconstruction. A lot of sample datasets were calculated by using a finite element method and boundary element method(FEM-BEM) hybrid code for network training. A parameterization method of crack for the output of neural network was introduced. The numerical results demonstrated that the neural networks approach is a suitable way for reconstruction of SCC.