Scholarlink

<p>As the core transmission component of rotating machinery, the accuracy and generalization of bearing fault diagnosis are critical for the safe and stable operation of industrial systems. Under complex operating conditions such as load fluctuations and speed changes, the diagnostic performance is prone to sudden drops due to differences in domain distribution. To overcome these limitations, we propose a Dual Branch Domain Adaptive fault diagnosis network (DBDA-Net) based on drive-end data and fan-end data. This model adopts the WiDe Convolutional Neural Network (WDCNN) as the backbone network and constructs a dual branch architecture. The fault information extraction branch mines domain specific fault features of dual-end signals, while the domain similarity extraction branch learns domain invariant features. The channel attention fusion module in the model integrates sensor data from the driver and fan ends to achieve information complementarity. In the loss function section, a multidimensional loss function system is designed, which includes domain related loss, contrastive learning loss, multi task classification loss and domain classification loss. In the result verification section, feature visualization and the SHAP interpretability framework are used to reveal the basis for fault discrimination of the models and improve the transparency of the diagnostic process. The experimental part was validated on the CWRU public dataset and the laboratory unlabeled bearing dataset. The results showed that the proposed model achieved the highest target domain diagnostic accuracy of 99.75% in domain adaptation tasks, with F1 scores exceeding 0.9900. The proposed method is comprehensively compared with state-of-the-art models in terms of diagnostic accuracy and model scale, verifying its superiority and efficiency. Meanwhile, the ablation experiment on dual-end data verified that the model performance was optimal when integrating dual-end data and the attention module, with an accuracy of 99.75%. DBDA-Net possesses both high diagnostic accuracy and strong interpretability, providing a new technical solution for the intelligent fault diagnosis of rotating machinery in industrial scenarios.</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

DBDA-net: An interpretable multi-source domain adaptive model for bearing fault diagnosis using dual-end vibration signals

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月16日

查看全文
  • Ziang Du,
  • Yanfei Liu,
  • Zhaoqiang Wang,
  • Congbao Tian,
  • Qian Zhang

摘要

As the core transmission component of rotating machinery, the accuracy and generalization of bearing fault diagnosis are critical for the safe and stable operation of industrial systems. Under complex operating conditions such as load fluctuations and speed changes, the diagnostic performance is prone to sudden drops due to differences in domain distribution. To overcome these limitations, we propose a Dual Branch Domain Adaptive fault diagnosis network (DBDA-Net) based on drive-end data and fan-end data. This model adopts the WiDe Convolutional Neural Network (WDCNN) as the backbone network and constructs a dual branch architecture. The fault information extraction branch mines domain specific fault features of dual-end signals, while the domain similarity extraction branch learns domain invariant features. The channel attention fusion module in the model integrates sensor data from the driver and fan ends to achieve information complementarity. In the loss function section, a multidimensional loss function system is designed, which includes domain related loss, contrastive learning loss, multi task classification loss and domain classification loss. In the result verification section, feature visualization and the SHAP interpretability framework are used to reveal the basis for fault discrimination of the models and improve the transparency of the diagnostic process. The experimental part was validated on the CWRU public dataset and the laboratory unlabeled bearing dataset. The results showed that the proposed model achieved the highest target domain diagnostic accuracy of 99.75% in domain adaptation tasks, with F1 scores exceeding 0.9900. The proposed method is comprehensively compared with state-of-the-art models in terms of diagnostic accuracy and model scale, verifying its superiority and efficiency. Meanwhile, the ablation experiment on dual-end data verified that the model performance was optimal when integrating dual-end data and the attention module, with an accuracy of 99.75%. DBDA-Net possesses both high diagnostic accuracy and strong interpretability, providing a new technical solution for the intelligent fault diagnosis of rotating machinery in industrial scenarios.