Inverse Analyses by Means of Hierarchical Neural Network and Computational Mechanics with Application to 3D Crack Identification.

This paper describes the application of a hierarchical neural network to the identification of a crack hidden in a solid with the electric potential method. The present method of crack identification consists of three subprocesses. First, sample data of identification parameters vs electric potential values are calculated by the finite element method. Second, the error-back-propagation neural network is trained using the sample data. Finally, the trained network is utilized for crack identification. The present method is applied to the identification of a two-dimensional crack in a plate, and then applied to the identification of a semi-elliptical surface crack in a plate with electric potential values measured on the back surface. The accuracy and efficiency of the method are discussed in detail.