Scholarlink

<p>This research proposes an advanced method for monitoring flood discharge vibration and diagnosing abnormal vibration of submerged radial steel gates. By employing a data-level dynamic fusion method based on the covariance variance dedication rate (CVDR), vibration signals that reflect the comprehensive vibration characteristics of the supporting arm are obtained. Then recurrence plots and recurrence quantification analysis are used to determine the alarm thresholds for abnormal vibration. Furthermore, a joint diagnostic method combining CVDR, non-threshold recurrence plot, convolutional neural networks (CNN), and multi-head attention (MA) mechanism is proposed for fault diagnosis. The vibration signal is first fused using the CVDR method to minimize the risk of misdiagnosis. This fused signal is then processed with kernel density estimation to calculate the probability density distribution of the laminarity index, from which the vibration alarm threshold is determined. This approach enables real-time monitoring and timely warning of flood discharge vibrations. Using the CNN-MA method, 50 repeated experiments were conducted, with the least accurate case achieving a test accuracy of 93.06 %, and the average accuracy across all 50 experiments reaching 95.00 %. These results demonstrate that the proposed method enhances feature extraction capabilities, significantly improving the diagnostic accuracy of abnormal gate vibrations.</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

Vibration monitoring and fault diagnosis of a radial gate subjected to discharge excitation using an improved recurrence plot-based method

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Yangliang Lu,
  • Yakun Liu,
  • Ze Cao,
  • Di Zhang,
  • Hengxin Ren

摘要

This research proposes an advanced method for monitoring flood discharge vibration and diagnosing abnormal vibration of submerged radial steel gates. By employing a data-level dynamic fusion method based on the covariance variance dedication rate (CVDR), vibration signals that reflect the comprehensive vibration characteristics of the supporting arm are obtained. Then recurrence plots and recurrence quantification analysis are used to determine the alarm thresholds for abnormal vibration. Furthermore, a joint diagnostic method combining CVDR, non-threshold recurrence plot, convolutional neural networks (CNN), and multi-head attention (MA) mechanism is proposed for fault diagnosis. The vibration signal is first fused using the CVDR method to minimize the risk of misdiagnosis. This fused signal is then processed with kernel density estimation to calculate the probability density distribution of the laminarity index, from which the vibration alarm threshold is determined. This approach enables real-time monitoring and timely warning of flood discharge vibrations. Using the CNN-MA method, 50 repeated experiments were conducted, with the least accurate case achieving a test accuracy of 93.06 %, and the average accuracy across all 50 experiments reaching 95.00 %. These results demonstrate that the proposed method enhances feature extraction capabilities, significantly improving the diagnostic accuracy of abnormal gate vibrations.