Permutation entropy-based improved uniform phase empirical mode decomposition for mechanical fault diagnosis

Abstract

Uniform phase empirical mode decomposition (UPEMD) is an effective signal separation method, which is proposed to solve the mode mixing phenomenon of empirical mode decomposition (EMD) by adding uniform phase sinusoidal signal as masking signal. However, the decomposition effect of UPEMD depends on the selection of amplitude and phase number of masking signal. Besides, there are also some issues such as noise residue and incomplete decomposition that need to be solved. In this paper, a novel permutation entropy-based improved uniform phase empirical mode decomposition (PEUPEMD) method is proposed to address these problems. In PEUPEMD method, first of all, the sinusoidal signals with uniform phases are added to the raw signal as masking signals. Second, the superimposed signal is decomposed using EMD and the final intrinsic mode functions are obtained via an ensemble way. Third, the obtained IMFs with high-frequency are detected by permutation entropy algorithm. Finally, the residual signal containing low-frequency components is decomposed through EMD completely. Last, the simulated signal and tested data are applied to verify the feasibility of PEUPEMD via comparing it with EMD, UPEMD, CEEMDAN and ICEEMDAN methods. The results analysis indicated that PEUPEMD was superior to the comparative methods in decomposing accuracy and mode mixing suppression.