Scholarlink

<p>As core transmission components in high-end intelligent manufacturing systems, face gears pose challenges in fault diagnosis, such as weak feature discriminability, few-shot fault samples, and cross-domain generalization failure under various working conditions. To address these issues, a digital twin-enhanced domain generalization (DTEDG) framework is proposed for diagnosing few-shot faults in face gears. Analytical dynamics and finite element analysis are combined to construct a digital twin model, enabling the acquisition of virtual vibration measurements under different conditions. With latent features extracted from both virtual and real vibration data, a mutual information alignment is proposed to fuse digital twins for feature enhancement. A one-dimensional U-net diffusion model is then employed to augment those few-shot real vibration features. On this basis, a multiscale information unification strategy is put forward to improve the cross-domain diagnosis capability under different working conditions. The proposed DTEDG was evaluated on a face gear experimental setup, achieving an average accuracy of 92.95% and an F1-score of 92.30% for the few-shot fault diagnosis under multiple working conditions. Digital twin construction fidelity was quantified by the meshing frequency deviation between simulated and measured signals, yielding a mean absolute percentage deviation of 2.63%. The results demonstrate that the proposed method exhibits superior generalization ability and diagnostic performance for face gears in intelligent manufacturing systems compared with existing methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Digital twin-enhanced domain generalization for diagnosing few-shot faults in face gears

  • 期刊论文

  • 发表日期：

    2026年3月27日

查看全文
  • Chuan Li,
  • Qiyi Liu,
  • Qibing Yu,
  • Shuai Yang,
  • Ziqiang Pu

摘要

As core transmission components in high-end intelligent manufacturing systems, face gears pose challenges in fault diagnosis, such as weak feature discriminability, few-shot fault samples, and cross-domain generalization failure under various working conditions. To address these issues, a digital twin-enhanced domain generalization (DTEDG) framework is proposed for diagnosing few-shot faults in face gears. Analytical dynamics and finite element analysis are combined to construct a digital twin model, enabling the acquisition of virtual vibration measurements under different conditions. With latent features extracted from both virtual and real vibration data, a mutual information alignment is proposed to fuse digital twins for feature enhancement. A one-dimensional U-net diffusion model is then employed to augment those few-shot real vibration features. On this basis, a multiscale information unification strategy is put forward to improve the cross-domain diagnosis capability under different working conditions. The proposed DTEDG was evaluated on a face gear experimental setup, achieving an average accuracy of 92.95% and an F1-score of 92.30% for the few-shot fault diagnosis under multiple working conditions. Digital twin construction fidelity was quantified by the meshing frequency deviation between simulated and measured signals, yielding a mean absolute percentage deviation of 2.63%. The results demonstrate that the proposed method exhibits superior generalization ability and diagnostic performance for face gears in intelligent manufacturing systems compared with existing methods.