Optics InfoBase is the Optical Society's online library for flagship journals, partnered and copublished journals, and recent proceedings from OSA conferences.

For data-driven remaining useful life (RUL) prediction of rolling bearing (RB), deep learning methods usually do not consider the difference in distribution due to different operating conditions, which adversely affects the prediction results. Recently, transfer learning has been a research hotspot, and it can mitigate the above problem effectively. However, for different bearings under the same working condition, the data distribution changes due to the location and time of failure. To solve these problems, a transfer prediction model based on the dynamic benchmark (DB) has been proposed to predict the RUL of bearings in this study. The method is divided into three processes. First, a fully convolutional variational autoencoder is used to perform unsupervised learning; next, a domain adaptation method based on the DB is employed for reconstruction; finally, extracted hidden layer features are input to the fully connected layer of the proposed model to obtain the final prediction result. The prediction and health management (PHM) Challenging 2012 and XJTU-SY RB accelerated life test datasets were used for verification. The results showed that the proposed method could significantly improve prediction accuracy and had good stability and generalizability.

The existing bearing temperature fault detection and early warning system has a high false alarm rate and insufficient early warning ability. For this reason, in this study, a method for detecting the abnormal bearing temperature of high-speed trains based on spatiotemporal fusion decision-making was proposed. First, the temperature characteristics of similar bearings were compared and analyzed with different spatial distributions. Then, a bearing abnormal temperature rise detection model based on the analytic hierarchy process (AHP) entropy method was proposed. Second, the temperature characteristics of the same bearings were compared and analyzed with different time distributions. A real-time prediction model of high-speed train bearing temperature anomalies based on Bi-directional Long Short-Term Memory (BILSTM) was proposed. Finally, the D-S evidence theory was used to combine the anomaly detection model based on the AHP entropy method and the anomaly detection model based on BILSTM real-time prediction. Through the comprehensive diagnosis and decision-making of high-speed train bearings from two dimensions of space and time, a more comprehensive and accurate anomaly detection model was realized. The experimental results showed that the spatiotemporal comparison fusion decision model successfully eliminated the misjudgment phenomenon of single-dimension model diagnosis and that it has good early warning ability.

The transfer learning (TL) method represented by domain adaptation (DA) can effectively improve the prediction accuracy of rolling bearings’ remaining useful life (RUL) under different working conditions. However, the difference in the bearing degradation process under the same working conditions limits the reliability and generalization of the transfer RUL prediction model. Owing to the aforementioned problems, this study proposed an RUL transfer predicting method for rolling bearings based on working conditions’ common benchmark. An attention mechanism autoencoder (AMAE) is proposed to extract the common benchmark under each working condition and improve the commonness between deep features. The dynamic benchmark constraint (DBC) under the same working conditions was proposed to ensure the reliability of a common benchmark and improve the prediction accuracy in the process of benchmark transfer under different working conditions. Verified by Xi’an Jiaotong University (XJTU) and the Changxing Sumyoung Technology Co., Ltd. (SY) bearing datasets, the proposed method can effectively obtain a common benchmark that can be used for DA under various working conditions. In the experimental design of six sets of RUL prediction tests under different working conditions, more than 50% of the experimental tasks have better prediction results using the proposed method. The proposed approach increases the overall prediction accuracy to 11.74% compared with the method without DA. Experimental results show that the proposed method can better meet the needs of intelligent operation and maintenance in practical engineering

Intelligent mechanical fault diagnosis has developed very fast in recent years due to the advancement and application of deep learning technologies. Thus, there are many deep learning network models that have been explored in fault classification and diagnosis. However, there are still limitations in research on the relationship between fault location, fault type, and fault severity. In this paper, a novel method for diagnosis of bearing fault using hierarchical multitask convolution neural networks (HMCNNs) is proposed, taking into account the mentioned relationships. The HMCNN model includes a main task and multiple subtasks. In the HMCNN model, a weighted probability is used to reduce the classification error propagation among multitasks to improve the fault diagnosis accuracy. The validity of the proposed method is verified on bearing datasets. Experimental results show that the proposed method is very effective and superior to the existing methods.

Transformer model is being gradually studied and applied in bearing fault diagnosis tasks, which can overcome the feature extraction defects caused by long-term dependencies in convolution neural network (CNN) and recurrent neural network (RNN). To optimize the structure of existing transformer-like methods and improve the diagnostic accuracy, we proposed a novel method based on the multiscale time-frequency sparse transformer (MTFST) in this paper. First, a novel tokenizer based on shot-time Fourier transform (STFT) is designed, which processes the 1D format raw signals into 2D format discrete time-frequency sequences in the embedding space. Second, a sparse self-attention mechanism is designed to eliminate the feature mapping defect in naive self-attention mechanism. Then, the novel encoder-decoder structure is presented, the multiple encoders are employed to extract the hidden feature of different time-frequency sequences obtained by STFT with different window widths, and the decoder is used to remap the deep information and connect to the classifier for discriminating fault types. The proposed method is tested in the XJTU-SY bearing dataset and self-made experiment rig dataset, and the following work is conducted. The influences of hyperparameters on diagnosis accuracy and number of parameters are analysed in detail. The weights of the attention mechanism (AM) are visualized and analysed to study the interpretability, which explains the partly working pattern of the network. In the comparison test with other existing CNN, RNN, and transformer models, the diagnosis accuracy of different methods is statistically analysed, feature vectors are presented via the t-distributed stochastic neighbor embedding (t-SNE) method, and the proposed MTFST obtains the best accuracy and feature distribution form. The results demonstrate the effectiveness and superiority of the proposed method in bearing fault diagnosis.