Scholarlink

<p>To overcome the shortcomings of conventional bearing fault diagnosis methods relying on artificial feature extraction, a new bearing fault diagnostic method based on dual-scale convolutional neural network and bidirectional gated recurrent unit (DSCNN-BiGRU) is presented in this paper. The method first inputs the original vibration signal into the model, extracts feature from the data through parallel convolution blocks with two different kernel sizes, and then uses the channel attention mechanism to weigh and fuse these features. Finally, the weighted features are input into the classifier module for fault classification, and the end-to-end diagnosis from the input signal to the fault classification is realized. To verify the validity and robustness of the DSCNN-BiGRU model, the bearing datasets of the DDS test bench and the MFS-MG test bench of our group are adopted in this paper. The experimental outcomes describe that the average diagnostic accuracy of DSCNN-BiGRU is 99.60%. Even when the signal-to-noise ratio is reduced to − 6 dB, the model can still accurately recognize the bearing fault type, and the accuracy rate is 80.43%.</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

DSCNN-BiGRU: A Novel Deep Learning Model for Bearing Fault Diagnosis Under Noisy Conditions

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Ming Zhao,
  • Haimeng Sun,
  • Xianwang Li

摘要

To overcome the shortcomings of conventional bearing fault diagnosis methods relying on artificial feature extraction, a new bearing fault diagnostic method based on dual-scale convolutional neural network and bidirectional gated recurrent unit (DSCNN-BiGRU) is presented in this paper. The method first inputs the original vibration signal into the model, extracts feature from the data through parallel convolution blocks with two different kernel sizes, and then uses the channel attention mechanism to weigh and fuse these features. Finally, the weighted features are input into the classifier module for fault classification, and the end-to-end diagnosis from the input signal to the fault classification is realized. To verify the validity and robustness of the DSCNN-BiGRU model, the bearing datasets of the DDS test bench and the MFS-MG test bench of our group are adopted in this paper. The experimental outcomes describe that the average diagnostic accuracy of DSCNN-BiGRU is 99.60%. Even when the signal-to-noise ratio is reduced to − 6 dB, the model can still accurately recognize the bearing fault type, and the accuracy rate is 80.43%.