| GUO Xiaopeng,DU Shaohui,AN Zhongyan,LI Xiuwen,JIN Yitao,LU Yang.A Bearing Fault Diagnosis Method for Aeroengines Based on Self-Attention and Improved ACGAN[J].航空发动机,2026,52(1):59-65 |
| A Bearing Fault Diagnosis Method for Aeroengines Based on Self-Attention and Improved ACGAN |
| DOI：10.12482/ISSN.1672-3147.20231013001 |
| Key Words:bearings self-attention improved auxiliary classifier generative adversarial network fault diagnosis aeroengine |
| Author Name | Affiliation | | GUO Xiaopeng | AECC Shenyang Engine Research Institute, Shenyang 110015, China | | DU Shaohui | AECC Shenyang Engine Research Institute, Shenyang 110015, China | | AN Zhongyan | AECC Shenyang Engine Research Institute, Shenyang 110015, China | | LI Xiuwen | Tangzhi Technology Hunan Development Co., Ltd., Changsha 410083, China；
BeiJing TangZhi Technology Development Co., Ltd., Beijing 100038, China | | JIN Yitao | Tangzhi Technology Hunan Development Co., Ltd., Changsha 410083, China | | LU Yang | AECC Shenyang Engine Research Institute, Shenyang 110015, China |
|
| Hits: 1813 |
| Download times: 513 |
| Abstract:In practical aeroengine engineering applications, engines are not allowed to operate with faults, resulting in a scarcity of
fault samples. This sample imbalance significantly degrades the accuracy and stability of data-driven bearing fault diagnosis models, neces?
sitating a reduction in reliance on bearing fault data. To address these issues, this paper proposes a bearing fault diagnosis method for
aeroengines based on self-attention and an improved auxiliary classifier generative adversarial network (ACGAN). Firstly, the original
input feature dataset is constructed using vibration data and operating condition data, which effectively preserves the characteristics and
operating conditions of the original data. Secondly, a data augmentation model based on self-attention and improved ACGAN is developed
to generate synthetic data. The original and synthetic data are then fused and used as input features to train the Softmax classifier, thereby
establishing the aeroengine bearing fault diagnosis model. Finally, the fused dataset is employed to enhance four types of bearing fault
datasets, and the proposed model is verified. The results show that the model based on self-attention and improved ACGAN achieves a
diagnostic accuracy of 98.9%, precision of 90.59%, recall of 96.25%, and a balanced F-score of 93.33%, demonstrating its ability to
significantly improve the performance of the aeroengine bearing fault diagnosis. |
| View Full Text View/Add Comment Download reader |
|
|
|
