A Residual-Variational Mode Decomposition to Detect Impact Signal

Variational Mode Decomposition (VMD) has been widely used in signal process. However, number of modal and balance parameter limit the application of VMD seriously. To solve this problem, a novel method, residual-VMD (RVMD), is proposed in this paper. RVMD can be separated into two stages. In the first stage, VMD is run to decompose raw signal into a series of modals. Then, an improved ensemble kurtosis, named smooth ensemble kurtosis (SEK), is applied to evaluate the ratio of impulse component in each modal. Then, the modal which has the smallest SEK will be subtracted from the raw signal and replace the raw signal by the residual signal. This process will be repeated until the modal number is below 3. Then, the modal which has a larger SEK will be set as the input of the second stage. In the second stage, the balance parameter is obtained by a fix step length search method. Finally, the modal will be analyzed by envelope technology to observe the fault characteristic frequency clearly. The simulation and experiment results indicate the rationality of the proposed method. Meanwhile, the comparison results highlight its superiority than the others.