基于多尺度正交PCA-LPP流形学习算法的故障特征增强方法

Fault feature enhancement method based on multiscale orthogonal PCA-LPP manifold learning algorithm

针对齿轮箱故障声发射信号特征增强问题,提出一种多尺度正交PCA-LPP非线性流形学习特征增强方法,兼顾PCA的全局方差增大变换特性以及LPP的局部非线性特征保持特性,并通过正交化消除投影分量间的冗余信息,使处理之后的齿轮箱故障信号内含的故障特征得到增强,一方面增强后信号包络谱中的故障谱线清晰明显,另一方面增强后信号以小波包能量熵为特征量,故障类型的辨识率显著提高,可以达到93.75%。

Aiming at the feature enhancement problem of gearbox fault acoustic emission signals,a novel method based on multiscale orthogonal PCA-LPP manifold learning algorithm was proposed here considering the global distribution variance enhancement of PCA and the local nonlinear characteristics preservation of LPP.The redundant information between projection components was eliminated with the orthogonal treatment.In processing gearbox fault acoustic emission signals with the proposed method,on one hand,the fault lines in the enhanced signal’s envelope spectrum were more clear,on the other hand,the eigenvectors of fault signals were constructed with the wavelet packet energy entropy,the fault identification rate rose obviously,and it could reach 93.75%.