Scholarlink

<p>Diagnosing faults in electrical submersible pumps using intelligent methods is a challenging task, especially when deep learning techniques are used to extract features directly from vibration signals instead of relying on predefined human features. A key limitation of this approach is the lack of foundational models for machine fault diagnosis using vibration data, unlike the abundance of pre-trained networks available for image classification. To address this, we propose a method that applies various 2D transformations to time domain signals, combines them into RGB images, and leverages these images to fine-tune existing image classification networks. Our results demonstrate that this approach outperforms the state-of-the-art previous deep learning method based on metric learning applied to this task and is comparable to the solution using human-defined features.</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

Electrical submersible pump fault diagnosis based on 2D transformation of vibration signals and transfer learning of image classification networks

  • 期刊论文

  • 发表日期：

    2025年1月31日

查看全文
  • Luciano Henrique Peixoto da Silva,
  • Alexandre Rodrigues,
  • Flavio Varejão,
  • Marcos Pellegrini Ribeiro,
  • Thiago Oliveira-Santos

摘要

Diagnosing faults in electrical submersible pumps using intelligent methods is a challenging task, especially when deep learning techniques are used to extract features directly from vibration signals instead of relying on predefined human features. A key limitation of this approach is the lack of foundational models for machine fault diagnosis using vibration data, unlike the abundance of pre-trained networks available for image classification. To address this, we propose a method that applies various 2D transformations to time domain signals, combines them into RGB images, and leverages these images to fine-tune existing image classification networks. Our results demonstrate that this approach outperforms the state-of-the-art previous deep learning method based on metric learning applied to this task and is comparable to the solution using human-defined features.