广义变分模式分解降噪在齿轮早期故障诊断中的应用

针对齿轮早期故障特征的微弱性和耦合性，本文提出广义变分模式分解(generalized variational mode decomposition, GVMD)-峭度-包络谱法诊断齿轮故障。首先利用GVMD的频域多尺度定频分解属性，根据齿轮故障频谱信息和信号特点设置GVMD主要参数，按需分解信号，准确获取微弱特征分量，避免VMD对微弱特征提取存在的不足和小波包变换能量泄漏引起的微弱特征混淆问题。然后结合峭度准则和齿轮故障频率信息选择故障冲击分量，融合更多故障信息重构降噪信号。最后对降噪信号进行包络解调分析，实现齿轮故障诊断。实际信号分析表明，由于GVMD能够按需获取微弱特征分量，本文所提方法能够获得更丰富的微弱故障信息准确识别齿轮早期故障位置。

Application of Generalized Variational Mode Decomposition De-noising in Early Gear Fault Diagnosis

Aiming at the weakness and coupling of gear fault characteristics, a novel method based on generalized variational mode decomposition (GVMD)-kurtosis-envelope spectrum is proposed in this paper to diagnose early faults of gear. First, the main parameters of GVMD are set based on the signal characteristics and gear fault spectrum information owing to its property of the multi-scale and fixed-frequency decomposition in the frequency domain. The signal can be decomposed as required by GVMD to accurately obtain the weak feature components. And the deficiency of VMD to extract weak feature and the information confusion of weak feature caused by the energy leakage of wavelet packet transform (WPT) are avoided by GVMD. Then, the de-noised signal is reconstructed by the feature components that are selected via kurtosis criterion and prior knowledge of fault features, which is expected to include more fault impact features. Finally, the envelope demodulation analysis for the de-noised signal is carried out to diagnose gear faults. Actual signal analysis shows that because GVMD can extract weak feature components as desired, the proposed method can integrate richer weak information to accurately identify the early fault location of gear.