Adaptive quaternion multivariate local characteristic-scale decomposition and its application to gear fault diagnosis

Abstract

Local characteristic-scale decomposition (LCD) has been widely applied in the field of gear fault diagnosis. However, LCD is only suitable for processing single channel signals and has poor noise robustness. Gear fault information collected by single channel of the sensor is not sufficient when the background noise is severe, based on which, a novel adaptive multivariate signal noise reduction method, quaternion singular value decomposition fault information spectrum (QSVDFIS), is defined in this paper. Based on QSVDFIS, a new multivariate signal decomposition method, adaptive quaternion multivariate local characteristic-scale decomposition (AQMLCD), is proposed in this paper. AQMLCD can adaptively reduce the noise of multivariate signal components, effectively fuse the fault information of each channel. Compared with multivariate empirical mode decomposition (MEMD), the noise robustness of AQMLCD is significantly improved, while the computation efficiency of AQMLCD is similar to that of MEMD. AQMLCD is applied to gear fault simulation and experimental signals, and the results illustrate the superiority of AQMLCD method.