A spatial bearing fault classification method based on improved APSMOTE-WKMFA
ObjectiveAiming at the problem of low accuracy of classification of time domain features of spatial bearings, time domain indicators and wavelet packet decomposition algorithms are combined to obtain the time-frequency distribution features of spatial bearings. The maximal overlap discrete wavelet p...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Office of Journal of Mechanical Transmission
2025-01-01
|
| Series: | Jixie chuandong |
| Subjects: | |
| Online Access: | http://www.jxcd.net.cn/thesisDetails?columnId=100610469&Fpath=home&index=0 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | ObjectiveAiming at the problem of low accuracy of classification of time domain features of spatial bearings, time domain indicators and wavelet packet decomposition algorithms are combined to obtain the time-frequency distribution features of spatial bearings. The maximal overlap discrete wavelet packet transform (MODWPT) is used to obtain the energy distribution of time series in different frequency bands as fault features, which solves the problem that traditional time domain features are difficult to effectively differentiate the operating state of spatial bearings. Aiming at the problem of poor classification performance of a model for a few classes of fault samples in the case of category imbalance, the spatial bearing fault classification method based on an improved affinity propagation synthetic minority oversampling technique-wavelet kernel marginal Fisher analysis （APSMOTE-WKMFA） was proposed.MethodsFirstly, the geodesic distance was used as the similarity metric for the affinity propagation algorithm, and the synthetic minority oversampling technique （SMOTE） was used to generate samples in the filtered subclusters up to the class balance. Secondly, the projection mapping was performed using the kernel marginal Fisher analysis based on the wavelet function. Finally, the <italic>k</italic>-nearest neighbor classifier algorithm was used to train the classification model on the transformed low-dimensional features. Test validation was carried out using the spatial bearing dataset of Harbin Institute of Technology’s aero-engine.ResultsCompared with the Euclidean distance, the geodesic distance can more accurately reflect the similarity between the spatial bearing data, and the intra-class aggregation and inter-class separation of the data are enhanced after the projection mapping, which improves the separability of faults. The test results show that under the same conditions, the classification accuracy of the improved APSMOTE-WKMFA is improved by 10.2 percentage points on average, compared with that of the class imbalance data, the k-means SMOTE, the APSMOTE, the improved APSMOTE and the improved APSMOTE-LPP, realizing the effective diagnosis of spatial bearing faults under class unbalance and variable speed conditions. |
|---|---|
| ISSN: | 1004-2539 |