Combining artificial neural network and finite element methods for detecting cracks

This study presents a new procedure based on Artificial Neural Network (ANN) and Back-Error Propagation (BEP) algorithm for crack identification in Functionally Graded Beam (FGB). The main idea that was used for crack identification in this study was the variation of vibration characteristics with the change of crack parameters. A cracked cantilever FGB was modeled using Finite Element Method (FEM) for different location and depth of crack. For applying FEM, ANSYS commercial Software was used. The modal analysis was done on the modeled beam and the first three natural frequencies of structure were obtained in different situations of crack. The modal analysis results were validated by data of one of available references. It was concluded that the FEM analysis has been done with good accuracy. Then two multi-layer neural networks were created for predicting the location and depth of cracks. The obtained data were used to train ANNs. For training ANNs, BEP algorithm was used. For applying BEP algorithm MATLAB commercial software was used. Inputs of neural networks were natural frequencies and outputs were locations and depths of cracks. The all of data were applied to neural networks in normalized form. Then the natural frequencies of FGB with different crack conditions as inputs applied to trained neural networks and corresponding locations and depths of cracks were calculated. The trained neural networks predicted crack location better than crack depth. Results showed that cracks characteristics were computed with good approximations. It was concluded that the presented procedure can be used for identification of crack in FGBs.