Scholarlink

<p>The reliability and efficiency of Permanent Magnet Synchronous Generators (PMSG) are crucial for the long-term performance and fault-free operation of wind power systems, where condition monitoring and early fault detection are essential. This study aims to improve the accuracy of the fault detection model and reduce processing time. To achieve these goals, a robust fault diagnosis framework is developed that combines the Finite Element Method (FEM), feature engineering, and experimental validation to detect various levels of demagnetization in PMSG. Four fault conditions are examined: healthy, 50%, 75%, and 100% unipolar demagnetization using stray flux and current signals. Features are extracted using the Discrete Wavelet Transform (DWT), ranked with the Kruskal–Wallis filter, and optimized through Feature Dimension Coordination (FDC). Three fundamental machine learning (ML) algorithms (KNN, SVM, and Ensemble), including seventeen sub-classifiers, are evaluated for fault classification. Results show that the Cosine KNN and six features (Range, L2 Norm, L1 Norm, Standard Deviation, Median, and Mean Absolute Deviation) are the most effective for classifying faults using flux signals. The feature engineering approach improves processing time (from 1.9 to 0.6&#xa0;s) and accuracy (from 95 to 100%). The method was validated with experimental data, confirming its robustness for wind energy systems.</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

An optimized machine learning framework for diagnosing demagnetization faults in PMSG using FEM, feature engineering, and experimental validation

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Nadeem Shahbaz,
  • Yu Chen,
  • Feng Liang,
  • Du Siyu,
  • Sichao Zhang,
  • Shouwang Zhao,
  • Shuang Wang,
  • Zhang Min,
  • Yong Ma,
  • Chong Li,
  • Wei Deng,
  • Yong Zhao

摘要

The reliability and efficiency of Permanent Magnet Synchronous Generators (PMSG) are crucial for the long-term performance and fault-free operation of wind power systems, where condition monitoring and early fault detection are essential. This study aims to improve the accuracy of the fault detection model and reduce processing time. To achieve these goals, a robust fault diagnosis framework is developed that combines the Finite Element Method (FEM), feature engineering, and experimental validation to detect various levels of demagnetization in PMSG. Four fault conditions are examined: healthy, 50%, 75%, and 100% unipolar demagnetization using stray flux and current signals. Features are extracted using the Discrete Wavelet Transform (DWT), ranked with the Kruskal–Wallis filter, and optimized through Feature Dimension Coordination (FDC). Three fundamental machine learning (ML) algorithms (KNN, SVM, and Ensemble), including seventeen sub-classifiers, are evaluated for fault classification. Results show that the Cosine KNN and six features (Range, L2 Norm, L1 Norm, Standard Deviation, Median, and Mean Absolute Deviation) are the most effective for classifying faults using flux signals. The feature engineering approach improves processing time (from 1.9 to 0.6 s) and accuracy (from 95 to 100%). The method was validated with experimental data, confirming its robustness for wind energy systems.