Intelligent Crack Recognition Based on XFEM and GA-BP Neural Networks

Abstract

Based on the extended finite element method (XFEM) and the error back propagation multilayer feedforward (BP) neural network (GA-BP) algorithm optimized by genetic algorithm (GA), an inverse analysis model for identifying cracks in structures is established. The GA-BP neural network is trained by the displacement data of measuring points obtained by XFEM forward analysis, and the network is used for crack inverse identification. The feasibility and accuracy of the model are verified by two typical examples, and the effects of mesh density, measuring point layout and input data noise on the accuracy of network recognition are discussed. The results show that the proposed method can retrieve the geometric information of straight cracks, which is the major focus of linear elastic fracture mechanics, and has good noise tolerance. Besides, it has better accuracy than the traditional BP neural network in general.