| ZHONG Xi,LI Pengpeng.A Fault Diagnosis Method for Imbalanced Aeroengine Bearing Samples Based on SAGAN-DRSKN[J].航空发动机,2026,52(1):20-28 |
| A Fault Diagnosis Method for Imbalanced Aeroengine Bearing Samples Based on SAGAN-DRSKN |
| DOI：10.12482/ISSN.1672-3147.20241115001 |
| Key Words:fault diagnosis bearing generative adversarial networks residual shrinkage networks attention mechanism small
sample aeroengine |
|
| Hits: 1857 |
| Download times: 665 |
| Abstract:To address the issue of low diagnostic efficiency caused by data imbalance in aeroengine bearing fault diagnosis, a fault
diagnosis method combining self-attention generative adversarial network(SAGAN) and adaptive deep residual shrinkage kernel network
(DRSKN) is proposed. The SAGAN method is utilized for minority class sample augmentation, leveraging the self-attention mechanism to
preserve the temporal correlation and fault feature consistency of the time series signals. For feature extraction, a DRSKN model is
constructed, which employs a multi-branch structure and multi-scale convolution kernels to capture both local and global features, and
incorporates a shrinkage activation function to suppress weak noise interference. Experimental validation was conducted based on the case
western reserve university (CWRU) Bearing Dataset. The results demonstrate that under the imbalance ratios of 10:1 and 5:1, the diagnostic
accuracy of the proposed method reaches 99.5% and 99.3%, respectively, which is significantly superior to that of the traditional ResNet
and DenseNet methods. The excellent feature classification capability of the model was verified through t-SNE visualization analysis. The
proposed SAGAN-DRSKN method can effectively address the data imbalance problem in bearing fault diagnosis, providing a reliable
technical solution for intelligent fault diagnosis of aeroengine bearings. |
| View Full Text View/Add Comment Download reader |
|
|
|
