A bearing fault detection method based on an improved YOLOv10 model of transfer learning
By improving the transfer learning method of the YOLOv10 model, the problems of low efficiency and high false negative rate in bearing fault detection were solved, and the cross-domain adaptability and detection accuracy of the model were improved, especially the ability to detect tiny defects under small sample conditions.
CN122156892APending Publication Date: 2026-06-05WUXI UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUXI UNIV
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-05
Smart Images
Figure CN122156892A_ABST
Abstract
The application discloses a bearing fault detection method based on an improved YOLOv10 model of transfer learning, collects a plurality of image data of key parts of a bearing, labels a fault area of a fault sample image, forms an image data set, constructs a C2f-ScConv module in a Backbone backbone network of the YOLOv10 model and embeds an EMA attention module, introduces a BiFPN bidirectional feature pyramid network structure in a Neck neck network, constructs an improved YOLOv10 model, inputs an ImageNet data set and a NEU-DET data set into the improved YOLOv10 model for full-parameter training, sets the first three layers of the Backbone backbone network as untrainable states, obtains initial model weight parameters, inputs the image data set into the improved YOLOv10 model for training, and migrates the initial model weight parameters to the improved YOLOv10 model, collects image data of key parts of a bearing to be detected, inputs the image data of the bearing to be detected into the trained improved YOLOv10 model, and the trained improved YOLOv10 model outputs a bearing fault detection result of the bearing to be detected.
Need to check novelty before this filing date? Find Prior Art