Power spectral density-guided variational mode decomposition for the compound fault diagnosis of rolling bearings

Compound fault diagnosis with stochastic interfering impulses remains to be a tricky task due to the difficulty of determining multiple resonant frequency bands. Variational mode decomposition (VMD) can extract instinct mode functions (IMFs) adaptively and effectively reduce the modal aliasing issue. However, such adaptivity requires a pre-known number of IMFs, which cannot be automatically determined. This paper develops a novel VMD method for compound fault diagnosis guided by power spectral density (PSD), which enables the estimation of the IMF amount and optimization of VMD processes by applying PSD's representation for energy distribution in the frequency domain. Several model amount determination methods and empirical wavelet transform (EWT) method are used to compare with the proposed method. Through comparative experiments of simulation and challenging experimental data sets, the proposed method has remarkable advantages in terms of fault feature detection capability and result display, and shows superior robustness.