Improved uniform phase empirical mode decomposition and its application in machinery fault diagnosis

As an adaptive non-stationary signal decomposition method, empirical mode decomposition (EMD) has a serious mode mixing problem. The uniform phase empirical mode decomposition (UPEMD) was proposed by adding a sinusoidal wave of uniform phase as a masking signal to overcome the shortcomings of EMD. However, the amplitude of added sinusoidal wave lacks adaptability, the mean curve cannot be completely separated from the signal in the iterative sifting process and thus the residual noise will affect the decomposition accuracy. To enhance the performance of UPEMD, in this paper, the improved uniform phase empirical mode decomposition (IUPEMD) method is developed to adaptively select the amplitude of added sinusoidal wave and then choose the optimal result from the iterative sifting of mean curves with different weights according to the index of orthogonality. The simulation signal analysis results show that IUPEMD has better decomposition ability and accuracy than the original UPEMD and ensemble empirical mode decomposition (EEMD). Finally, IUPEMD is applied to the rolling bearing and rotor rubbing fault diagnosis by comparing it with UPEMD, empirical wavelet transform (EWT) and variational mode decomposition (VMD) methods. The results show that IUPEMD can effectively identify the rolling bearing and rotor faults, and have better recognition effect than the comparison methods.