Scholarlink

<p>Health monitoring and fault diagnosis of rolling bearings are crucial for the continuous and effective operation of mechanical equipment. However, traditional methods often face challenges in accurately extracting fault features in complex and noisy environments, as conventional feature extraction techniques are highly susceptible to noise interference, which can result in the loss or distortion of fault information. Accordingly, a time–frequency dual-channel model based on dynamic feature fusion is proposed for fault diagnosis of bearing. The two channels employ conditional convolution (CondConv), residual connections, and Bi-LSTM to extract features from both time-domain and frequency-domain signals. CondConv dynamically generates convolutional kernels and biases, enabling the model to adaptively capture fault-specific impact features. This dynamic adjustment mechanism not only enhances the model's robustness to noise but also improves the adaptability and flexibility of feature extraction. The multi-head self-attention mechanism is adopted to capture the global correlation. Gating mechanism is introduced to dynamically regulate feature fusion, reducing redundancy and enhancing feature representation quality. Experimental results on the CWRU bearing fault dataset show that the proposed model significantly outperforms others under various noise conditions. Notably, under −&#xa0;6&#xa0;dB noise, the model achieved a maximum accuracy of 98.5%. Furthermore, ablation studies demonstrate that the gating mechanism plays a key role in enhancing noise resilience and classification accuracy. By dynamically optimizing the fusion of time-domain and frequency-domain features, the gating mechanism improved classification performance in noisy environments. The effectiveness of the proposed time–frequency fusion method in improving model robustness and fault diagnosis performance has been demonstrated and provides an effective reference for the accurate extraction of fault features in complex environments.</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

A novel dynamic feature fusion method with conditional convolution and gating mechanism for rolling bearing fault diagnosis

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Wentao Dong,
  • Zanhua He,
  • Daojin Yao,
  • Xiaoming Wang,
  • Jiahao Yan

摘要

Health monitoring and fault diagnosis of rolling bearings are crucial for the continuous and effective operation of mechanical equipment. However, traditional methods often face challenges in accurately extracting fault features in complex and noisy environments, as conventional feature extraction techniques are highly susceptible to noise interference, which can result in the loss or distortion of fault information. Accordingly, a time–frequency dual-channel model based on dynamic feature fusion is proposed for fault diagnosis of bearing. The two channels employ conditional convolution (CondConv), residual connections, and Bi-LSTM to extract features from both time-domain and frequency-domain signals. CondConv dynamically generates convolutional kernels and biases, enabling the model to adaptively capture fault-specific impact features. This dynamic adjustment mechanism not only enhances the model's robustness to noise but also improves the adaptability and flexibility of feature extraction. The multi-head self-attention mechanism is adopted to capture the global correlation. Gating mechanism is introduced to dynamically regulate feature fusion, reducing redundancy and enhancing feature representation quality. Experimental results on the CWRU bearing fault dataset show that the proposed model significantly outperforms others under various noise conditions. Notably, under − 6 dB noise, the model achieved a maximum accuracy of 98.5%. Furthermore, ablation studies demonstrate that the gating mechanism plays a key role in enhancing noise resilience and classification accuracy. By dynamically optimizing the fusion of time-domain and frequency-domain features, the gating mechanism improved classification performance in noisy environments. The effectiveness of the proposed time–frequency fusion method in improving model robustness and fault diagnosis performance has been demonstrated and provides an effective reference for the accurate extraction of fault features in complex environments.