Fault diagnosis of rolling bearings based on attention module and 1D-CNN

Aiming at the problem of poor feature recognition ability of traditional convolutional neural network(CNN), this paper proposes a rolling bearings fault diagnosis model combining the attention module and the one-dimensional convolutional neural network (1D-CNN). Firstly, the noise-added vibration signals are used as the input of the proposed model, and their multi-dimensional features are extracted by using the "convolution + pooling" unit. Then, the attention module assigns different weights to the extracted features. The double pooling layer is adopted to replace the fully connected layer in the traditional CNN for feature extraction and information integration. Finally, a Softmax layer is used to achieve bearing status classification. Experimental results show that the diagnostic accuracy of the model proposed in this paper reaches 99%. Compared with the traditional models, the proposed model has higher diagnostic accuracy, faster convergence speed, a more stable training process, and better generalization performance under variable load. © 2022, Solar Energy Periodical Office Co., Ltd. All right reserved.