Identification of circular cracks, extending to the surface of a pipe using the finite element method and artificial neural network

Abstract

The current paper describes the crack defect identification on the internal and external surfaces of the pipe. Defects represent circular crosscut cracks extending to the surface. It is assumed that the boundaries of the crack do not interact with each other. The problem of determining the depth of the crack leads to geometric inverse problem in elasticity theory. The solution to geometric inverse problem is based on the combination of finite element analysis and artificial neural networks. In addition, amplitude-time characteristics of components of the displacement vector are required for inverse problem. In solving direct problem, since a finite piece of a pipe is considered, the radial and axial displacements are measured during a short period of time. For that time, the waves reflected from the ends of the pipe cannot reach the receiver. The process of displacement measurement in this paper is simulated using calculation of finite elements software ANSYS. In the shown numerical example of this report, the depth of a crack is identified. Moreover, we investigate the dependence of this identification's accuracy upon input data, the architecture of neural network, the time-consuming of the training process as well as the accuracy of data inputs.