基于小波—奇异值分解差分谱的弱故障特征提取方法

Extraction Method of Faint Fault Feature Based on Wavelet-SVD Difference Spectrum

对于一些复杂信号中的弱故障特征信息,以往的两种小波—奇异值分解(Singular value decompositiom,SVD)组合模式的特征提取效果不佳,从小波的频率窗特性出发分析了出现这种问题的原因,进而对复杂信号的奇异值分布规律进行研究,据此提出一种新的小波-SVD差分谱组合模式。对原始信号做小波分解得到一系列细节信号后,不再将这些信号简单地排列成矩阵,而是利用每个细节信号构造特定结构的Hankel矩阵,再通过SVD对每个矩阵做正交化分解,并利用奇异值差分谱来选择特征奇异值进行SVD重构,由此实现对弱故障特征信息的提取。对一个轴承振动信号的处理结果证实该方法对复杂信号中的弱故障特征信息具有优良的提取效果,其获得的故障特征波形非常清晰,克服了以往小波-SVD组合模式对弱故障特征提取效果不佳的缺陷。

For faint fault feature information hidden in the some complicated signal,the usual two kinds of wavelet-singular value decomposition（SVD） method can＇t achieve the good feature extraction effect,and the reason of this defect is analyzed from the aspect of frequency window property of wavelet.Based on the study about the distribution law of singular values of complicated signal,a new method of wavelet-SVD difference spectrum is proposed.In this method,an original signal is decomposed by wavelet transform and a group of detail signals are obtained,however,these detail signals are not simply arrayed a matrix any more,whereas each detail signal is used to create a Hankel matrix of specific configuration,and then SVD operation of each matrix is made to obtain its orthogonal decomposition results,furthermore,the feature singular values are selected by dint of difference spectrum of singular value for the SVD reconstruction,through these procedures the faint feature information can be extracted.The processing results of the vibration signal of a bearing demonstrate the excellent effect of this method on the extraction of faint fault feature information from the complicated signal,and the fault feature waveform extracted is very clear.The defect of usual two kinds of wavelet-SVD methods that they can＇t achieve the good effect on extraction of faint fault feature is overcome.