Rolling bearing is an essential part of various rotating machines, and its signal is the typical nonlinear signal. Traditional fault diagnosis usually relies on manual experience to extract the features of signals first. Deep convolutional neural networks (DCNN) can make fuller use of time series than traditional convolutional neural networks (CNN). Because of the low accuracy rate, fault diagnosis using gated recurrent unit (GRU) alone is not unsatisfactory. In order to improve the temporality of one-dimensional convolutional neural networks (1D-CNN) and enhance the accuracy of GRU, a novel fault diagnosis method called deep convolutional neural networksgated recurrent unit (DCNN-GRU) is first put forward, which combines DCNN with GRU. The original signals without preprocessing are input into the DCNN, and the outputs of DCNN are input into the GRU consequently. Then the faults of rolling bearing can be diagnosed effectively. As the post-processing method, the t-distributed stochastic neighbor embedding (t-SNE) method is applied to visualize the fault diagnosis results. Six different network models, including the DCNN-GRU, are used to train the same fault dataset for comparison. The simulation results show that the proposed method can reach more than 99.9% accuracy stably for the given dataset, which can verify the feasibility and effectiveness of proposed method. And the DCNN-GRU can also be verified with good generalization ability using different dataset.

Deep learning is gradually being widely used in fault diagnosis now, because deep learning networks are more advantageous in processing data, especially image data. However, research using frequency spectra image of fault signals as inputs to deep learning networks are extremely rare in the field of fault diagnosis. Therefore, a brand-new intelligent fault diagnosis method is proposed in this paper which combines discrete random separate (DRS) frequency spectrum images with deep learning networks (DRSFSI-DL). To investigate the fault diagnosis effects of the method mentioned above, several deep learning networks are utilized for comparisons, such as GoogLeNet, residual network, and Inception_ResNet_v2. The vibration fault frequency spectrum images processed by the DRS method are input to train several deep learning networks. Under the same circumstance of deep learning networks, the fault diagnosis using the DRS frequency spectrum image (DRSFSI), is also compared to the fault diagnosis using traditional frequency spectrum, including the power spectrum density (PSD) and cepstrum. The fault diagnosis results show that the proposed method has a better classification accuracy than the PSD image and cepstrum image, with the same deep learning networks. The fault diagnosis accuracy can reach up to 100.00% for some deep learning networks with better generalization performance than the PSD image and cepstrum image. Lastly, the proposed method is further verified using the brand-new bearing fault dataset, and excellent accuracy and generalization ability are achieved.

No abstract

Gear vibration fault signals are non-stationary and nonlinear, so it is very difficult to accurately extract the fault characteristics for diagnosis. As symplectic geometry mode decomposition (SGMD) has shown excellent decomposition performance and noise robustness in signal processing. A novel gear fault diagnosis method, that is, SGMD-CNN, is proposed combined SGMD with a convolutional neural network (CNN). The noise of the gear vibration fault signal is reduced through the use of SGMD method, and several symplectic geometry components (SGC) are screened out according to the kurtosis value respectively. The reconstructed signal based on the selected SGCs is used as the input of the convolutional neural network. The ability of the convolutional neural network to learn features and classification is used to realize intelligent fault diagnosis of gear. The proposed SGMD-CNN is verified by the gear fault data. Based on the same convolution neural network model, the accuracy of the established SGMD-CNN model is 100%. It is about 32% higher than the accuracy of the simple CNN model and about 16% higher than the accuracy of the EMD-CNN model. The results show that the method is effective in classifying and identifying different types of gear failures.