Feature frequency extraction based on singular value decomposition and its application on rotor faults diagnosis

Abstract

The selection of effective singular values using the singular value decomposition (SVD) method has always been a hot topic. In this paper, we found that there was a special relationship between effective singular values and feature frequency components. Theoretical derivations illustrated that each frequency component produced two adjacent nonzero singular values with one ranking another closely. Size of singular values was directly proportional to amplitude of feature frequency. The number of singular values was only related to the number of feature frequency components. For these discoveries, a novel feature frequency separation method based on SVD was proposed, through which axis orbits of large rotating machines were readily purified. The results show that the algorithm was very accurate in feature frequency extraction.