Axle-box bearings are one of the most critical mechanical components of the high-speed train. Vibration signals collected from axle-box bearings are usually nonlinear and nonstationary, caused by the complicated operating conditions. Due to the high reliability and real-time requirement of axle-box bearing fault diagnosis for high-speed trains, the accuracy and efficiency of the bearing fault diagnosis method based on deep learning needs to be enhanced. To identify the axle-box bearing fault accurately and quickly, a novel approach is proposed in this paper using a simplified shallow information fusion-convolutional neural network (SSIF-CNN). Firstly, the time domain and frequency domain features were extracted from the training samples and testing samples before been inputted into the SSIF-CNN model. Secondly, the feature maps obtained from each hidden layer were transformed into a corresponding feature sequence by the global convolution operation. Finally, those feature sequences obtained from different layers were concatenated into one-dimensional as the fully connected layer to achieve the fault identification task. The experimental results showed that the SSIF-CNN effectively compressed the training time and improved the fault diagnosis accuracy compared with a general CNN.

Despite the numerous studies on bearing fault diagnosis based on frequency domain or time-frequency domain analyses, there is a lack of a fair assessment on which method or methods are practically effective in identifying the fault frequencies of damaged bearings in noisy environments. Most methods were developed based on experiments with simple lab test rigs equipped with bearings having manufactured artificial defects, and the signal-to-noise ratio under lab conditions is too ideal to be useful for verifying the effectiveness of a signal processing method. The purpose of this study is to evaluate the effectiveness of advanced signal processing methods applied in a high-speed train operating environment with multi-source interference. In this work, the most advanced signal processing methods (including spectral kurtosis, deconvolution, and mode decomposition) are studied, and the shortcomings of each method are analyzed. Based on the characteristics of high-speed train wheel set bearings (HSTWSBs), the concept of fault characteristic signal-to-noise ratio (FCSNR) is put forward to quantitatively evaluate the fault periodicity intensity, and corresponding improved methods are proposed by combining the FCSNR with existing signal processing methods; all these methods consider the periodic characteristics and impact characteristics of the bearing fault. The simulation signal and actual signals of HSTWSB with natural defects help verify the effectiveness of the proposed methods. Finally, the advantages and disadvantages of the different signal processing methods are objectively evaluated, and the application scope of each method is analyzed and prospected. This study provides a reference and new ideas for the fault diagnosis of HSTWSB and other industrial bearings.

Accurate remaining useful life (RUL) prediction of bearings is the key to effective decision‐making for predictive maintenance (PdM) of rotating machinery. However, the individual heterogeneity and different working conditions of bearings make the degradation trajectories of bearings different, resulting in the mismatch between the RUL prediction model established by the full‐life training bearing and the testing bearings. To address this challenge, this paper proposes a novel RUL prediction method for roller bearings that considers the difference and similarity of degradation trajectories. In this method, a feature extraction method based on continuous wavelet transform (CWT) and convolutional autoencoder (CAE) is proposed to extract the deep features associated with bearing performance degradation before the degradation indicator (DI) is obtained by applying the self‐organizing maps (SOM) method. Next, a dynamic time warping (DTW) based method is applied to perform the similarity matching of degradation trajectories of the training and testing bearings. Driven by the historical DIs of the given bearing, the grey forecasting model with full‐order time power terms (FOTP‐GM) is applied to model the degradation trajectory using a parameter optimization method. Then, the failure threshold of the given testing bearing can be determined using a data‐driven method without manual intervention. Finally, the RUL of the given testing bearing can be estimated using the preset failure threshold and the optimized degradation trajectory model of the given testing bearing. The experimental results show that the proposed method retains the individual differences of bearing degradation trend, realizes the independent and reasonable bearing failure threshold setting, and improves the prediction accuracy of RUL.

Fault diagnosis under specific working conditions has achieved remarkable success. However, due to variations in working conditions, the assumption that training and test samples are independent and identically distributed is often violated, which makes the diagnostic model brittle under unseen working conditions. To this end, a generic generalization strategy, namely, regularized domain adaptive weight optimization strategy (RDAWOs), is devised for fault diagnosis of rotating machinery. We first design the architecture of a 1-D convolutional neural network. Then, the hyperparameter regularization term and an adaptive pooling layer are designed to control the complexity and improve the adaptability of the overparameterized deep model, respectively. Finally, domain adaptive weight optimization is established to identify the working condition abundant in spurious label-related information and to mine the robust fault knowledge under various working conditions. Obtained results indicate the strong generalization ability for out-of-distribution samples, as well as relatively high diagnostic accuracy of the RDAWOs-based deep model under unseen working conditions.

More layers in a convolution neural network (CNN) means more computational burden and longer training time, resulting in poor performance of pattern recognition. In this work, a simplified global information fusion convolution neural network (SGIF-CNN) is proposed to improve computational efficiency and diagnostic accuracy. In the improved CNN architecture, the feature maps of all the convolutional and pooling layers are globally convoluted into a corresponding one-dimensional feature sequence, and then all the feature sequences are concatenated into the fully connected layer. On this basis, this paper further proposes a novel fault diagnosis method for a rotor–journal bearing system based on SGIF-CNN. Firstly, the time-frequency distributions of samples are obtained using the Adaptive Optimal-Kernel Time–Frequency Representation algorithm (AOK-TFR). Secondly, the time–frequency diagrams of the training samples are utilized to train the SGIF-CNN model using a shallow information fusion method, and the trained SGIF-CNN model can be tested using the time–frequency diagrams of the testing samples. Finally, the trained SGIF-CNN model is transplanted to the equipment’s online monitoring system to monitor the equipment’s operating conditions in real time. The proposed method is verified using the data from a rotor test rig and an ultra-scale air separator, and the analysis results show that the proposed SGIF-CNN improves the computing efficiency compared to the traditional CNN while ensuring the accuracy of the fault diagnosis.