Autoregressive model‐enhanced variational mode decomposition for mechanical fault detection

Abstract

Mechanical fault detection is an important research direction attracted a great attention. Variational mode decomposition (VMD) has been widely used due to the ability to generate in highly localised time-frequency (TF) representation. However, the method usually fails to analyse the non-stationary signal with the unknown low-frequency interferences. Combined with autoregressive (AR) model technique, an enhancement version of VMD method is proposed. First, VMD is employed to decompose the raw non-stationary vibration signal into a set of intrinsic mode functions (IMFs), which contain the faulty information, and the interference of high-frequency noises can be almost eliminated. Second, AR model is introduced to further purify each IMF through decreasing the interferences of unknown low-frequency components. Finally, Hilbert envelope analysis is applied to demodulate the purified signal and the fault can be determined by comparing the demodulation frequency with the theoretical faulty feature frequency. Numerical simulations and experimental investigations are conducted to validate the performance of the proposed method. The comparisons with two combinations, AR model and empirical mode decomposition (EMD), AR model and local mean decomposition (LMD) are given, and the comparative results verify the proposed method is superior to other methods for mechanical fault detection.