Scholarlink

<p>Machine learning has enabled predictive maintenance in industrial fault diagnosis through vibration analysis, which still faces significant application resistance due to the high demand for data and computing power. TinyML enables low-power, on-device machine learning inference on resource-constrained microcontrollers with minimal data transmission through extreme model optimization, demonstrating potential for broad industrial applications. This paper presents an edge-deployable TinyML approach for efficient bearing fault diagnosis in a vibration monitoring system. By employing transfer learning, the methodology effectively adapts the model with limited training data and deploys it on an ESP32-S3 microcontroller. Experimental results on a proprietary bearing dataset covering four bearing fault types demonstrate a fault state classification accuracy of 88.28% while completing inference within 45 ms and consuming only 17.7 mJ of energy. The proposed solution also integrates remote monitoring and data management capabilities, providing real-time insights into equipment health status. This paper provides an effective and economical approach for condition monitoring, fault prediction, and preventive maintenance of industrial equipment.</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

An edge-deployable TinyML approach enhanced by transfer learning for efficient bearing fault diagnosis

  • 期刊论文

  • 发表日期：

    2025年11月6日

查看全文
  • Zheng Gao,
  • Zhichao Jiang,
  • Zefang Dong,
  • Xianpeng Fu,
  • Yuanfen Chen,
  • Chi Zhang

摘要

Machine learning has enabled predictive maintenance in industrial fault diagnosis through vibration analysis, which still faces significant application resistance due to the high demand for data and computing power. TinyML enables low-power, on-device machine learning inference on resource-constrained microcontrollers with minimal data transmission through extreme model optimization, demonstrating potential for broad industrial applications. This paper presents an edge-deployable TinyML approach for efficient bearing fault diagnosis in a vibration monitoring system. By employing transfer learning, the methodology effectively adapts the model with limited training data and deploys it on an ESP32-S3 microcontroller. Experimental results on a proprietary bearing dataset covering four bearing fault types demonstrate a fault state classification accuracy of 88.28% while completing inference within 45 ms and consuming only 17.7 mJ of energy. The proposed solution also integrates remote monitoring and data management capabilities, providing real-time insights into equipment health status. This paper provides an effective and economical approach for condition monitoring, fault prediction, and preventive maintenance of industrial equipment.