Bearing fault diagnosis method based on multi-level information fusion

Bearing fault diagnosis is crucial for saving time and reducing costs, as it targets one of the most critical components in rotary machines. The decision fusion method has proven to be an effective approach to improving the performance of fault diagnosis. However, the conflicting results will affect the accuracy of the final classification. This paper proposes a multi-level information fusion network (MLIFN) to improve the classification performance of multiple classifiers in decision fusion through feature fusion method. A two level decision fusion of deep features and empirical features are applied to improve the classification accuracy. Furthermore, an innovative hybrid attention mechanism, combining self-attention and mutual attention, is introduced for deep feature extraction. The Paderborn bearing dataset is employed to validate the effectiveness of the MLIFN. The results demonstrate that the diagnostic accuracy of the MLIFN can reach up to 99.82%, significantly outperforming other methods.