Fault diagnosis model of rolling bearings based on the M-YOLO network
ObjectiveThe algorithms developed for the combination of deep learning and bearing fault diagnosis have achieved initial results, but most of them are processed by processing one-dimensional vibration data and input into the network structure for diagnosis, while the research on fault diagnosis tech...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Office of Journal of Mechanical Transmission
2025-04-01
|
| Series: | Jixie chuandong |
| Subjects: | |
| Online Access: | http://www.jxcd.net.cn/thesisDetails#10.16578/j.issn.1004.2539.2025.04.020 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | ObjectiveThe algorithms developed for the combination of deep learning and bearing fault diagnosis have achieved initial results, but most of them are processed by processing one-dimensional vibration data and input into the network structure for diagnosis, while the research on fault diagnosis technology using two-dimensional signals as input is still on the surface, and the analysis of such methods is rarely reported. The rolling bearing is taken as the research object, and the fault diagnosis algorithm with two-dimensional signal as the input is studied, and the fault diagnosis model of rolling bearing based on M-YOLO network is constructed for the problems of multi-condition fault diagnosis, small data sample, and long model training time.MethodsFirstly, the mosaic data augmentation method was used to enrich the samples to improve the interference of unbalanced data on the diagnostic results. Then, through the Markov frequency domain image conversion method, the discrete signal was transformed into a probabilistic model, different strategies were used to classify the time series, and the fault diagnosis with the two-dimensional frequency domain image as the model input was completed. Finally, the traditional Dropout structure was replaced by Dropblock, and more refined optimization was carried out from the spatial and temporal levels to improve the robustness and diagnostic accuracy of the model.ResultsThe test results show that the diagnosis results of the M-YOLO diagnostic model are significantly higher than those of the traditional fault diagnosis methods, and the frequency-domain conversion feature image has better robustness than the time-domain image, which is more suitable for the training and classification of the object detection model. |
|---|---|
| ISSN: | 1004-2539 |