Neural Network usage in structural crack detection

Abstract

Artificial Neural Network is becoming an efficient tool in online structural health monitoring. ANN enables, due to its promising inherent capabilities, to predict existence of undesirable effects such as cracks within the structure. Natural frequencies of the structure particularly the first three vibration modes are the most pronounced features of the structure to be evaluated for the health monitoring tasks. Crack in the structure make it weaker and under certain loads it may extend to complete fracture and sometimes to catastrophic failure. In this paper, the ANSYS software which employs finite element (FE) techniques is used to generate data for solid cantilever beams and simply supported beams. Natural frequencies are obtained for the first three vibration modes taking into account that the structure is linear for the healthy and the cracked structures. For different crack locations and crack depths, the ANSYS data on natural frequencies and vibration modes show lower values compared with healthy structure. These are good indicators to be used for training the Artificial Neural Network (ANN) tools. Results of ANSYS software is first verified with some available theoretical solutions and then results of the trained artificial neural network (ANN) for defected structure are compared with ANSYS solutions. The findings of this study suggest high accuracy of ANN on structural health monitoring with robust prediction of size and location of cracks.