Rolling bearing fault diagnosis method based on a multi-scale and improved gated recurrent neural network with dual attention

Regarding the problem that the diagnosis aeeuraey of rolling bearing fault diagnosis models decreases under the variable working conditions and environmental noise interference, a rolling bearing fault diagnosis method (DAMSCN-BiGRU) composed of a multi-scale convolutional network based on dual attention mechanism (DAMSCN) and an improved bidirectional gated recurrent unit (BiGRU) was proposed. Firstly, using the multi-scale feature fusion module with different kernel sizes to obtain a variety of receptive fields and extract the multi-scale feature information of the original vibration signal of the bearing, which were fused adaptively according to their importance. And the multi-scale features were weighted and fused using a dual attention module composed of channel attention and spatial attentionto weaken the redundant features in the fused features. Then, the attention layer was added and the segmented activation was used to improve the BiGRU to mine the time-domain features of the signal to improve the performance of the bearing fault diagnosis. Finally, the classification of different faults was completed by the Softmax layer. The experimental results show that compared with other intelligent diagnosis models, DAMSCN-BiGRU can achieve an average diagnostic accuracy of 98.2% under variable working condition and still has an accuracy of 85. 3% in the strong noise background, and the effect is better than other commonly used models under different levels of noise intensity, which is beneficial to promote the research and practical application in the intelligent fault diagnosis of rolling bearings. © 2024 Chinese Vibration Engineering Society. All rights reserved.