Abstract：To solve the problem of modal aliasing in the nearfrequency part of the existing decomposition method and improve the accuracy of the timevarying parameter model for direct diagnosis of the fault signal, the typical fault signal analysis method of the multiaxis rotor system was investigated. A method for analyzing fault characteristic signals was proposed based on the combination of variational mode decomposition (VMD) and AR spectrum. The number of decomposition modes k in VMD was selected by the instantaneous frequency mean method. After the VMD decomposition, the inherent modal functions (IMFs) were generated. The AR spectrum was used to extract the features of IMFs, and the characteristic frequency bands corresponding to the typical failure modes were analyzed. The results show that the problem of selecting the decomposition mode number k can be solved by the fault feature extraction method based on VMDAR spectrum, and the k value can be avoided in the empirical selection. The modal aliasing of the nearfrequency part of the signal decomposition can be suppressed by VMD. The windowing effect of Hilbert separation is overcome by the AR model, which has strong resolution in frequency band division. The proposed method can effectively extract fault features and provide feature information for the hybrid kernel support vector machine algorithm of the improved particle swarm optimization algorithm.
