Scholarlink

<p>This research introduces a novel approach that integrates vibration and acoustic data to improve the reliability of conventional vibration-based systems for diagnosing wind turbine (WT) blade faults. Three blade conditions—fault-free (good), bend, and erosion—are investigated, with 120 samples collected for each, yielding a dataset of 360 samples. Fault-related features are independently extracted from both vibration and acoustic signals and classified using a probabilistic multiclass support vector machine (SVM). To enhance diagnostic performance, two decision-level fusion (DLF) strategies are implemented: a non-trainable (NTR) method based on Dempster–Shafer Theory (DST) and a trainable (TR) method using the softmax regression. Experimental validation on test blades demonstrates that the DLF-based approach using NTR achieves up to 100% classification accuracy and greater robustness under noise compared to traditional single-signal methods. These results highlight the practical advantage of combining heterogeneous signals, offering a more dependable solution for WT blade fault diagnosis.</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

Wind Turbine Blade Fault Detection through Combined Analysis of Vibration and Acoustic Data

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • Manas Ranjan Sethi,
  • Banala Hemasudheer,
  • Sudarsan Sahoo

摘要

This research introduces a novel approach that integrates vibration and acoustic data to improve the reliability of conventional vibration-based systems for diagnosing wind turbine (WT) blade faults. Three blade conditions—fault-free (good), bend, and erosion—are investigated, with 120 samples collected for each, yielding a dataset of 360 samples. Fault-related features are independently extracted from both vibration and acoustic signals and classified using a probabilistic multiclass support vector machine (SVM). To enhance diagnostic performance, two decision-level fusion (DLF) strategies are implemented: a non-trainable (NTR) method based on Dempster–Shafer Theory (DST) and a trainable (TR) method using the softmax regression. Experimental validation on test blades demonstrates that the DLF-based approach using NTR achieves up to 100% classification accuracy and greater robustness under noise compared to traditional single-signal methods. These results highlight the practical advantage of combining heterogeneous signals, offering a more dependable solution for WT blade fault diagnosis.