Scholarlink

<p>This paper proposes a semi-supervised ensemble learning-based anomaly detection method to detect small leaks in pipelines using acoustic emission (AE) signals. The proposed method combines bagging and random feature generation, employing 1-Class Support Vector Machines (SVMs) as base estimators. Random feature generation extracts features for each base estimator from AE signals in the frequency domain using randomly selected frequency ranges and bin widths without requiring prior knowledge of frequency bands affected by leaks. The extracted features used to train the base estimators. Leak detection is achieved by aggregating predictions from the base estimators. By training solely on data from normal operating conditions, the proposed method addresses the class imbalance issue commonly encountered in anomaly detection. Experimental results demonstrate that the proposed method outperforms other models in detecting small leaks.</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

Detecting Small Leaks in Pipeline with Semi-Supervised Ensemble Learning Using Acoustic Emission Sensor

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Byungjae Park,
  • Simaek Lee,
  • Hyeongmin Yoo

摘要

This paper proposes a semi-supervised ensemble learning-based anomaly detection method to detect small leaks in pipelines using acoustic emission (AE) signals. The proposed method combines bagging and random feature generation, employing 1-Class Support Vector Machines (SVMs) as base estimators. Random feature generation extracts features for each base estimator from AE signals in the frequency domain using randomly selected frequency ranges and bin widths without requiring prior knowledge of frequency bands affected by leaks. The extracted features used to train the base estimators. Leak detection is achieved by aggregating predictions from the base estimators. By training solely on data from normal operating conditions, the proposed method addresses the class imbalance issue commonly encountered in anomaly detection. Experimental results demonstrate that the proposed method outperforms other models in detecting small leaks.