时频表示特征约简的旋转机械故障特征提取方法

Rotating machine fault feature extraction based on reduced time frequency representation

针对二维时频表示特征提取困难这一问题,在分析基追踪与二维非负矩阵分解方法(Two Dimensional Nonnegative Matrix Factorization,2DNMF)的基础上,提出一种基于时频表示特征约简的时频特征提取方法。利用基追踪方法将信号分解成基于信号多特征冗余原子库的一组原子的线性组合,组合各分解原子的Wigner-Ville分布获取信号基追踪时频表示,采用2DNMF对基追踪时频表示的幅值矩阵进行特征约简以获取蕴含在其内部的低维特征。将提出的方法应用于8种不同状态轴承信号的特征提取中,实验结果证明了方法的有效性。

Aiming at extracting fault features from the two-dimensional time frequency representation,a novel time frequency feature extraction method based on reduced time frequency representation is proposed after investigating the principles of basis pursuit and Two Dimensional Non-negative Matrix Factorization（2DNMF）.Combined with Wigner-Ville distribution,the basis pursuit method which represents the original signal as a set of atoms is introduced to compute the basis pursuit time frequency representation,and then 2DNMF is employed to reduce the dimension of the amplitude matrix of basis pursuit time frequency representation and extract its corresponding low dimensional features.The proposed method is applied to extract the fault features from eight different state rolling bearings,and the results verify its effectiveness.