|
| Bearings Fault Diagnosis under Variable Operating Conditions Based on Invariant Risk Minimization |
| Received:June 21, 2024 Revised:August 13, 2024 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7643/issn.1672-9242.2024.10.012 |
| KeyWord:invariant risk minimization bearings fault diagnosis vibration signal deep learning transfer learning |
| Author | Institution |
| MA Jiteng |
Naval Aviation University, Shandong Yantai, , China |
| LYU Weimin |
Naval Aviation University, Shandong Yantai, , China |
| LI Gen |
Naval Aviation University, Shandong Yantai, , China |
| DAI Ziqin |
Naval Aviation University, Shandong Yantai, , China |
|
| Hits: |
| Download times: |
| Abstract: |
| The work aims to achieve intelligent fault diagnosis that can stably identify bearing fault types under unknown working conditions. To address the problems of weak model generalization ability in traditional intelligent fault diagnosis methods and the strong data dependency of transfer learning methods such as domain adaptation, an unknown working condition bearing fault diagnosis method based on invariant risk minimization was proposed. The vibration signal data from multiple available working conditions were preprocessed through data augmentation techniques. The invariant risk minimization strategy was adopted to constrain the feature extractor and domain classifier, enabling them to learn features and knowledge that remain invariant across unknown working conditions. This method could effectively improve the generalization ability of the model under unknown working conditions and reduce the reliance on specific working condition data. Experiments demonstrated that this method achieved satisfactory diagnostic accuracy and robustness in diagnostic tasks under multiple unknown working conditions. In conclusion, when facing situations where the working condition information is unknown but fault characteristics are related, this method allows deep models to generalize better and achieve high-precision bearing fault diagnosis without requiring corresponding data from unknown working conditions. |
| Close |
|
|
|