Crack detection in a rotating shaft using artificial neural networks and PSD characterisation

The diagnosis of cracks in rotating shafts using non-destructive techniques provides a route for avoiding catastrophic failure of these common components. This study measured the dynamic response of a full-scale rotating shaft with three different crack depths. A novel non-destructive system is developed and described. The system uses vertical vibration of the system measured over time and characterises its behaviour using elements of the power spectral density (PSD) gained from a fast Fourier transform of the time-history. The PSDs were used as an input into an artificial neural network (ANN) to detect the presence of cracks using changes in the spectral content of the vibration of the system. A novel method for reducing the amount of data input into the ANN is described. The Peak Position Component Method (PPCM) reduces data transfer by using statistical characterisation of the position of the peaks in the PSD. The peak positions represent a small fraction of the information contained in the total frequency range. The number of the PSD peaks used as input to the neural net is a small fraction of the total frequency range. The ANN was a supervised feed-forward network with Levenberg-Marquardt back-propagation algorithm acting on the PPCM results. The frequency spectrum for the three different crack lengths examined show clear shifts in the peak positions of the PSD and the results clearly demonstrate the feasibility of using the new system to detect cracks in-service.