A novel multi-scale convolutional neural network incorporating multiple attention mechanisms for bearing fault diagnosis

This paper proposes a multi-scale convolutional neural network (CNN) fault diagnosis model incorporating multiple attention mechanisms (MMCNN) to address the limitations of conventional CNNs in learning critical fault features, which impacts the accuracy of rolling bearing fault diagnosis. The proposed approach first introduces a convolutional structure characterized by multi-channel and multi-scale attributes, designed to expand the network's receptive field and effectively capture salient features across various dimensions. Subsequently, we enhance and integrate three distinct attention mechanisms-position attention mechanism (PAM), channel attention mechanism (CAM), and squeeze-and-excitation attention mechanism (SEAM)-into the multi-scale feature extraction model. These mechanisms collectively optimize the network's learning process by mitigating the influence of irrelevant signal components and adaptively amplifying the response to fault features. Finally, by visualizing the feature maps from the intermediate layers of the model, we demonstrate that the proposed method significantly improves both the feature extraction capability and fault type recognition accuracy of the model.