一种基于视觉词与扩散模型的工业轴承数据增强方法
By combining visual words with a diffusion model, the problems of non-convergence and data imbalance in GAN models during training are solved, generating richer bearing fault diagnosis data and improving the effect of data augmentation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING JIAOTONG UNIV
- Filing Date
- 2025-05-15
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, generative adversarial network (GAN) models are prone to non-convergence during training, the generative model and the discriminative model are highly coupled, and the problem of data imbalance is not effectively solved, resulting in poor data augmentation effect.
A method based on visual words and diffusion models is adopted. Grayscale images are generated through continuous wavelet transform, interest points are extracted and clustered, mapping relationships are increased, and rich samples are generated using diffusion models, which are independent of the discriminant model and improve the diversity of generated data.
The training process is more stable, the generated samples are richer, overfitting is avoided, and the data augmentation effect is improved, making it suitable for bearing fault diagnosis.
Smart Images
Figure CN120705627B_ABST