Scholarlink

<p>This article proposes a transformer vibration diagnosis method based on convolutional neural network transfer learning to address the problems of large differences in the distribution of fault vibration signal data and small sample size in practical engineering. The method combines differential mutual inductance sensor technology to achieve high-precision monitoring of transformer vibration. For the diagnosis of transformer vibration, one-dimensional time-series vibration signals are transformed into two-dimensional images through recursive graphs, and source domain and target domain data are constructed. Adopting the ResNet network with ECA attention mechanism for pre training, extracting universal features and transferring them to the target model, fine-tuning is completed with a small number of samples, significantly improving the accuracy of fault recognition and model robustness. At the system level, a differential graphene mutual inductance vibration sensor was designed, which efficiently converts mechanical vibration into electrical signals through electromagnetic induction principle, and optimizes system parameters by combining the equivalent circuit model. Simulation and experiments have shown that the graphene sensor has high sensitivity, strong anti-interference ability, and stability, and can effectively capture weak vibration signals of transformers under complex operating conditions. The joint application of two technologies not only solves the problem of intelligent diagnosis in small sample scenarios, but also provides a reliable hardware perception method, providing theoretical support and technical path for transformer status monitoring and fault warning.</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

Transformer Vibration Fault Diagnosis Based on Recursive Graph Transfer Learning and Differential Graphene Sensor

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Jiaqi Peng,
  • Yulin Ma,
  • Yuan Li,
  • Shou Li,
  • Suofu You,
  • Guoping Liu

摘要

This article proposes a transformer vibration diagnosis method based on convolutional neural network transfer learning to address the problems of large differences in the distribution of fault vibration signal data and small sample size in practical engineering. The method combines differential mutual inductance sensor technology to achieve high-precision monitoring of transformer vibration. For the diagnosis of transformer vibration, one-dimensional time-series vibration signals are transformed into two-dimensional images through recursive graphs, and source domain and target domain data are constructed. Adopting the ResNet network with ECA attention mechanism for pre training, extracting universal features and transferring them to the target model, fine-tuning is completed with a small number of samples, significantly improving the accuracy of fault recognition and model robustness. At the system level, a differential graphene mutual inductance vibration sensor was designed, which efficiently converts mechanical vibration into electrical signals through electromagnetic induction principle, and optimizes system parameters by combining the equivalent circuit model. Simulation and experiments have shown that the graphene sensor has high sensitivity, strong anti-interference ability, and stability, and can effectively capture weak vibration signals of transformers under complex operating conditions. The joint application of two technologies not only solves the problem of intelligent diagnosis in small sample scenarios, but also provides a reliable hardware perception method, providing theoretical support and technical path for transformer status monitoring and fault warning.