ABSTRACT Deep learning has achieved significant progress in the field of bearing fault diagnosis due to its efficient feature extraction capabilities. However, most data‐driven intelligent diagnosis methods neglect to incorporate domain knowledge, leading to a lack of interpretability. To address this limitation, a dual‐channel fault diagnosis model is proposed in this paper, which includes a time‐domain channel and a frequency‐domain channel network. The two channels can extract features independently, and the diagnostic results are fused into probability values for decision‐making so that the model has better diagnostic performance. In the frequency‐domain channel, a network architecture combining attention mechanisms is proposed based on the characteristics of bearing fault frequencies. A multi‐scale convolution module (MSCM) is employed to extract local features of the spectrum at multiple scales, while a global self‐attention module (GSAM) is utilised to capture global features. Since time‐domain signals only reflect the variation of vibration amplitude over time and cannot directly reveal fault features, a multi‐frequency band feature attention module (MBFAM) is introduced in the time‐domain channel to adaptively focus on different frequency band information. The proposed method integrates frequency information into the network model, providing higher interpretability, which is also confirmed by experimental validation.

Bearing fault diagnosis is of great importance to ensure the safe and stable operation of mechanical equipment. The actual collected bearing fault signals are susceptible to strong noise interference and bearing samples for each fault state may be insufficient, which increases the difficulty of capturing effective features. Most of the existing diagnostic methods extract features from a single sensor signal for pattern recognition and fault diagnosis. The fault information provided by a single sensor is limited and incomplete, which is usually very difficult to meet the demand for accurate and reliable fault diagnosis in complex scenarios. To solve these problems, this paper proposes a multi-sensor residual convolutional fusion network (MRCFN) for intelligent fault diagnosis of bearings. Firstly, a convolutional pooling module (CPM) is coupled with the designed double ring residual module (DRRM) to rough feature extraction and deep feature mining, which not only captures the discriminative fault features from multi-sensor signal, but also avoids the performance degradation of network. Secondly, a spatial channel reconstruction module (SCRM) is further introduced to eliminate redundant information in the features and improve the network training efficiency. Finally, the presented global interactive perception fusion module (GIPFM) is connected with a classification block (CB) to globally fuse the features extracted from the acoustic and vibration signals and conduct automatic fault identification, which both can realize the complementarity and calibration of multi-sensor feature information and high precision diagnosis. The experiments and a series of comparisons are implemented on two datasets to efficaciously verify the superiority of the proposed method over five existing representative multi-sensor fusion diagnosis methods (i.e., FAC-CNN, MB-CNN, MsACNN, MRSDF , and IMSFDFL) under strong noise and small samples.

Significant advancements have been achieved in the realm of wind turbine fault diagnosis through the application of deep learning. However, most methods generally assume that the distribution between training samples and testing samples remains consistent, which is impractical owing to the varying working conditions. Furthermore, the interpretability of the deep model has always been a focus of attention. Therefore, a novel interpretable empowered deep transfer framework (IEDTF) based on deep contrastive convolution, Wasserstein distance (WD), and improved Gramian angular summation field (IGASF) is proposed. By blending multichannel parallel and spatial polling mechanisms, the proposed method can adequately exploit multiscale information at a deep level from the source domain. Then, contrastive learning combined with the WD is proposed to establish relationships between data from different domains, which generalizes labeled data to unlabeled data. The IGASF is proposed for interpreting distinguishable transfer features intuitively. The effectiveness of the proposed method is demonstrated on the wind turbine bearing dataset with different bearing fault positions. Experimental results indicate that IEDTF can locate fault positions exactly under varying working conditions and shows high stability, which outperforms many existing transfer models applied in wind turbine bearing fault diagnosis.