基于声纹的高泛化性风机叶片异常检测方法研究

High Generalization in Anomaly Detection of Wind Turbine Generator Based on Voiceprint

  • 摘要: 对风力发电机组叶片异常检测进行研究,发现依靠单台风力发电机3个叶片声纹之间的参考和对比即可诊断该台风力发电机叶片是否故障。为此,该文提出基于聚类和中值收敛的周期性音频切割方法,对声纹进行有效的切割,减少了计算量,为后续异常检测提供了基础。采用风力发电机组3个叶片之间的稳态差异法对异常进行检测,绕开了待检物,信道等变化所带来的算法迁移失效问题,具有良好的泛化性。为风力发电机组叶片检测提供了一种有效的技术手段。

     

    Abstract: The major research work of the paper is the anomaly diagnosis of wind turbine generator based on voiceprint. The research finds that the reference and comparison between the three blades of a single wind turbine can diagnose whether the wind turbine is faulty. On this base, the paper proposes a periodic audio cutting method based on clustering and median convergence, which effectively cuts the voiceprint, reduces the amount of calculation, and provides a basis for subsequent anomaly detection. The steady-state difference method between three blades of wind turbine is used to detect anomalies, which avoids the migration failure of algorithm caused by the changes of channels and objects to be checked. The paper provides an effective technical means for fan blade inspection.

     

/

返回文章
返回