Adaptive Prediction of Rotor Crack Depth and Fault Diagnosis

Aiming at the problems existing in traditional identification methods of rotor crack fault such as difficulty in feature extraction，inability to quantitatively identify the crack depth and susceptibility to noise pollution，a prediction model of crack depth based on axis orbit is proposed. The proposed model is based on singular value decomposition and denoising convolutional autoencoder（SVD-DCAE），which can effectively extract the fault characteristics of the cracked rotor and accurately predict the crack growth stage. The center orbits of the cracked rotor are considered as the input of the proposed model. Simulated data and experimental data are used to train and verify the proposed model respectively，and random noise is added to simulate different noise environments. The results suggest that the SVD-DCAE model can realize the accurate prediction of the crack stage. In the weak noise environment（signal-to-noise ratio（SNR）of 10 dB），the prediction accuracy of the crack stage is higher than 98%. Meanwhile，SVD-DCAE possesses strong anti-noise ability and robustness. In the strong noise environment（SNR of -10 dB），the prediction accuracy of the crack stage still reaches 80%，which is much higher than other classic convolutional neural network prediction models. © 2024 Nanjing University of Aeronautics an Astronautics. All rights reserved.