Bearing Fault Diagnosis Method based on Multiple-level Feature Tensor Fusion

Fault diagnosis of bearing in mechanical equipment is critical for ensuring safety and saving costs. The feature fusion technology is a effective way to improve the performance of fault diagnosis. However, how to extract and fuse the complementary fault features from multi-sensor data is still an important challenge. This study proposes a multiple-level feature tensor fusion network (MLFTFN) for bearing fault diagnosis to address this problem. A modified mutual attention is applied to feature extraction in MLFTFN to perform interactions between different modes of signal (signals in time domain and frequency domain). Afterwards, a feature tensor construction strategy is proposed, in which the feature tensors contain complementary fault information to the original features. The features extracted from feature tensors and the original features are concatenated for fault classification. Paderborn bearing data set is used to verify the effectiveness of the MLFTFN. Results show that the diagnostic accuracy is greatly improved, which can be up to 99.5%.