Scholarlink

Conventional machine learning (ML)Machine Learning (ML) and deep learningDeep Learning (DL) strategies for acoustic emissionAcoustic Emission Data (AE) data-driven condition monitoring exhibit a range of reliability concerns. These include variations due to fluid pressure, flange vibrations, varying leak dimensions, and AE signal noise, all of which shift with changing pipeline conditions. Furthermore, the complexity of interpreting sensor data is heightened by noise and fluctuating pressure conditions, particularly in multi-variate systems where discerning spatial relationships among sensors proves challenging. In response, we have pioneered the application of Graph Convolutional Networks (GCNs) to AE-based pipeline monitoring. This novel approach leverages a publicly accessible dataset, GPLA-12, which includes AE signals to both train and assess our GCN model. Our innovative graph construction method is crafted to decode and analyze the complexities of AE signals recorded under diverse pressure scenarios in a multi-sensor environment. This technique is poised to redefine standards in pipeline monitoring research and applications.

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

Enhancing Acoustic Emission  Driven Smart Gas-Pipeline Monitoring with Graph Neural Network

  • 章节

  • 发表日期：

    2025年

查看全文
  • Murshedul Arifeen,
  • Md. Junayed Hasan,
  • Ali Rohan,
  • Somasundar Kannan,
  • Anil Prathuru

摘要

Conventional machine learning (ML)Machine Learning (ML) and deep learningDeep Learning (DL) strategies for acoustic emissionAcoustic Emission Data (AE) data-driven condition monitoring exhibit a range of reliability concerns. These include variations due to fluid pressure, flange vibrations, varying leak dimensions, and AE signal noise, all of which shift with changing pipeline conditions. Furthermore, the complexity of interpreting sensor data is heightened by noise and fluctuating pressure conditions, particularly in multi-variate systems where discerning spatial relationships among sensors proves challenging. In response, we have pioneered the application of Graph Convolutional Networks (GCNs) to AE-based pipeline monitoring. This novel approach leverages a publicly accessible dataset, GPLA-12, which includes AE signals to both train and assess our GCN model. Our innovative graph construction method is crafted to decode and analyze the complexities of AE signals recorded under diverse pressure scenarios in a multi-sensor environment. This technique is poised to redefine standards in pipeline monitoring research and applications.