Scholarlink

Abstract <p>High-pressure hydrogen valves are subjected to the instantaneous impact of hydrogen flow and repeated start-stop action during service, and there is a potential risk of leakage. This paper investigates monitoring and identification of hydrogen valves leakage to ensure their operational reliability. Firstly, an acoustic signal monitoring system was built based on a high-pressure hydrogen gas-tightness test platform, and the time-domain feature of valves under different leakage conditions was analyzed. Secondly, the frequency-domain feature is extracted using a combination of &#xa0;variational modal decomposition and wavelet packet decomposition. Ultimately, the backward propagation network (BP) and convolutional neural network (CNN) are used to recognize patterns of acoustic signals, with the time-domain and frequency-domain parameters as feature inputs independently. The results show that the accuracy of BP and CNN networks based on frequency domain features has significantly improved, 93.33 and 91.67%, respectively. This paper obtained the feature extraction and pattern recognition method for hydrogen valves, which provides a reference for accurate and efficient recognition of the leakage condition of high-pressure hydrogen valves in the service process.</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

Research on Leakage Monitoring and Recognition Method of High-Pressure Hydrogen Valves

  • 期刊论文

  • 发表日期：

    2025年2月1日

查看全文
  • Yi Qin,
  • Zhe Yang,
  • Zetian Kang,
  • Qian Wu,
  • Yuchen Wang,
  • Anfeng Yu,
  • Huan Liu,
  • Yun Luo

摘要

Abstract

High-pressure hydrogen valves are subjected to the instantaneous impact of hydrogen flow and repeated start-stop action during service, and there is a potential risk of leakage. This paper investigates monitoring and identification of hydrogen valves leakage to ensure their operational reliability. Firstly, an acoustic signal monitoring system was built based on a high-pressure hydrogen gas-tightness test platform, and the time-domain feature of valves under different leakage conditions was analyzed. Secondly, the frequency-domain feature is extracted using a combination of  variational modal decomposition and wavelet packet decomposition. Ultimately, the backward propagation network (BP) and convolutional neural network (CNN) are used to recognize patterns of acoustic signals, with the time-domain and frequency-domain parameters as feature inputs independently. The results show that the accuracy of BP and CNN networks based on frequency domain features has significantly improved, 93.33 and 91.67%, respectively. This paper obtained the feature extraction and pattern recognition method for hydrogen valves, which provides a reference for accurate and efficient recognition of the leakage condition of high-pressure hydrogen valves in the service process.