selection of effective singular values using difference spectrum and its application to fault diagnosis of headstock

the noise reduction effect of singular value decomposition (svd) relies on the selection of effective singular values. the characteristic of singular values of normal signal and noise being studied, it is pointed out that there is a sudden change in the singular values of normal signal, but not in the ones of noise. the concept of difference spectrum of singular value is put forward, which consists of the forward differences of singular value sequence and can describe the sudden change status of singular values of a complicated signal. the automatic selection of effective singular values can be realized by the peak of the difference spectrum. if the maximum peak of difference spectrum is located in the first coordinates, it means that a strong direct current (dc) component is contained in original signal and the number of effective singular values will be determined by the second maximum peak coordinates, while what the first singular value corresponds to is the dc component, or else the number of effective singular values is determined by the maximum peak coordinates. the relationship between column number of matrix and noise removing quantity of svd is also studied using difference spectrum and the result shows that this relationship is like a symmetrical parabola. by dint of the difference spectrum, the hidden modulation feature caused by gear vibration in headstock is isolated from a turning force signal and the fault gear is accurately located by this modulation feature.