Scholarlink

Introduction <p>Most wind turbines are located in harsh environments and are prone to failure.Early fault signals are typically characterized by low-amplitude, short-duration periodic pulses, often masked by strong shock noise, which makes fault features easily submerged. Most existing decomposition methods rely on envelope calculation, but shock noise easily leads to envelope distortion, affecting fault feature extraction and making it difficult to identify early faults.</p> Methods <p>Consequently, a feature embedded deep learning-based mechanical fault identification approach is proposed in this study. Firstly, the original vibration signal is subjected to the feature mode decomposition (FMD), which fully considers shock signal characteristics, decomposes the original signal based on its periodicity, shock, autocorrelation and other features.The filter is then updated using correlation kurtosis (CK) to enhance the anti-shock signal interference capability. To increase the signal decomposition accuracy and prevent local optimality, the FMD's two parameters are optimized using the sparrow search algorithm. The optimal mode and characteristic frequency of early faults are obtained. Secondly, the traditional one-dimensional convolutional network model is improved by introducing the gated recurrent unit (GRU) and the attention mechanism, and the deep dynamic focusing network fault recognition model is proposed to realize the wind turbine fault identification.</p> Results <p>Finally, experiments and cases are used to confirm the effectiveness of the suggested approach.</p> Conclusion <p>The results show that the proposed method effectively inhibits the adverse effect of shock signals on early fault feature extraction, and is superior to alternative methods in fault identification accuracy.</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

Anti-shock Noise Intelligent Fault Identification for Wind Turbines Based on Feature-Embedded Deep Learning

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • Jian Dang,
  • Xiaoqin Cheng,
  • Haolin Yin,
  • Shaopeng Zhang,
  • Guangyi Liu,
  • Rong Jia

摘要

Introduction

Most wind turbines are located in harsh environments and are prone to failure.Early fault signals are typically characterized by low-amplitude, short-duration periodic pulses, often masked by strong shock noise, which makes fault features easily submerged. Most existing decomposition methods rely on envelope calculation, but shock noise easily leads to envelope distortion, affecting fault feature extraction and making it difficult to identify early faults.

Methods

Consequently, a feature embedded deep learning-based mechanical fault identification approach is proposed in this study. Firstly, the original vibration signal is subjected to the feature mode decomposition (FMD), which fully considers shock signal characteristics, decomposes the original signal based on its periodicity, shock, autocorrelation and other features.The filter is then updated using correlation kurtosis (CK) to enhance the anti-shock signal interference capability. To increase the signal decomposition accuracy and prevent local optimality, the FMD's two parameters are optimized using the sparrow search algorithm. The optimal mode and characteristic frequency of early faults are obtained. Secondly, the traditional one-dimensional convolutional network model is improved by introducing the gated recurrent unit (GRU) and the attention mechanism, and the deep dynamic focusing network fault recognition model is proposed to realize the wind turbine fault identification.

Results

Finally, experiments and cases are used to confirm the effectiveness of the suggested approach.

Conclusion

The results show that the proposed method effectively inhibits the adverse effect of shock signals on early fault feature extraction, and is superior to alternative methods in fault identification accuracy.