PLoS ONE, Год журнала: 2025, Номер 20(4), С. e0319202 - e0319202

Опубликована: Апрель 11, 2025

In industrial production, obtaining sufficient bearing fault signals is often extremely difficult, leading to a significant degradation in the performance of traditional deep learning-based diagnosis models. Many recent studies have shown that data augmentation using generative adversarial networks (GAN) can effectively alleviate this problem. However, quality generated samples closely related For reason, paper proposes new GAN-based small-sample method. Specifically, study continuous wavelet convolution strategy (CWCL) instead operation GAN, which additionally capture signal’s frequency domain features. Meanwhile, designed multi-size kernel attention mechanism (MSKAM), extract features vibration from different scales and adaptively select are more important for generation task improve accuracy authenticity signals. addition, structural similarity index (SSIM) adopted quantitatively evaluate signal by calculating between real both time domains. Finally, we conducted extensive experiments on CWRU MFPT datasets made comprehensive comparison with existing methods, verified effectiveness proposed approach.

Язык: Английский