Scholarlink

<p>To address the challenges of insufficient precision and limited adaptability in conventional rotor fault diagnosis methods, we propose a new approach using convolutional neural networks. This aims to effectively identify complex and diverse fault patterns in a steam turbine rotor through a thorough examination and in-depth investigation. An HZXT-009 sliding ball bearing simulation rig was used to conduct fault tests such as rotor misalignment, unbalance, and touching faults. The data collected experimentally was used to perform a comprehensive analysis of the temporal and spectral signal characteristics. Use the obtained data to train and enhance the convolutional neural network fault diagnosis model. The results of the model testing accuracy can reach 99%. The generalization test is introduced to verify that the model trained by the simulation test data can detect multi-condition faults in the operation of the power plant. The network detection results show that the accuracy rate can reach 97.5%, which is expected to be widely used in actual production and improve the efficiency and accuracy of fault diagnosis of rotor system.</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

Diagnosis of steam turbine rotor based on improved convolutional neural network algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月28日

查看全文
  • Zhongtao Zhou,
  • Miao Zhou,
  • Hui Huang,
  • Yanghai Li,
  • Wanbing Xu

摘要

To address the challenges of insufficient precision and limited adaptability in conventional rotor fault diagnosis methods, we propose a new approach using convolutional neural networks. This aims to effectively identify complex and diverse fault patterns in a steam turbine rotor through a thorough examination and in-depth investigation. An HZXT-009 sliding ball bearing simulation rig was used to conduct fault tests such as rotor misalignment, unbalance, and touching faults. The data collected experimentally was used to perform a comprehensive analysis of the temporal and spectral signal characteristics. Use the obtained data to train and enhance the convolutional neural network fault diagnosis model. The results of the model testing accuracy can reach 99%. The generalization test is introduced to verify that the model trained by the simulation test data can detect multi-condition faults in the operation of the power plant. The network detection results show that the accuracy rate can reach 97.5%, which is expected to be widely used in actual production and improve the efficiency and accuracy of fault diagnosis of rotor system.