Crack and shaft misalignment are two common types of fault in a rotor system, both of which have very similar dynamic response characteristics, and the vibration signals are vulnerable to noise contamination because of the interaction among different components of rotating machinery in the actual industrial environment, resulting in great difficulties in fault identification of a rotor system based on vibration signals. A method for identification of faults in the form of crack and shaft misalignments is proposed in this paper, which combines variational mode decomposition (VMD) and probabilistic principal component analysis (PPCA) to denoise the collected vibration signals from a test rig and then achieve signal feature extraction and fault classification with convolutional artificial neural network (CNN). The key parameters of the CNN are optimized and determined by genetic algorithm (GA) firstly, and the domain adaptability of the trained network is verified by the signals with different signal-to-noise ratio (SNR) values; then, the noisy vibration signals are decomposed into multiple band-limited intrinsic modal functions by VMD, and further data dimension reduction is performed by PPCA to realize the separation of the useful signals from noise; finally, the crack and shaft misalignment of the rotor system are identified by the optimized CNN. The results show that the proposed method can effectively remove the interference noise and extract the intrinsic features of the vibration signals, and the recognition rates of crack and shaft misalignment faults for the rotor system with different SNR values are more than 99%, which is considered to be very effective and useful.
Keywords: convolutional neural network; crack; fault identification; noisy environment; shaft misalignment.