Multivariate empirical mode decomposition and its application to fault diagnosis of rolling bearing

Rolling bearings are widely used in rotary machinery systems. The measured vibration signal of any part linked to rolling bearings contains fault information when failure occurs, differing only by energy levels. Bearing failure will cause the vibration of other components, and therefore the collected bearing vibration signals are mixed with vibration signal of other parts and noise. Using multiple sensors to collect signals at different locations on the machine to obtain multivariate signal can avoid the loss of local information. Subsequently using the multivariate empirical mode decomposition (multivariate EMD) to simultaneously analyze the multivariate signal is beneficial to extract fault information, especially for weak fault characteristics during the period of early failure. This paper proposes a novel method for fault feature extraction of rolling bearing based on multivariate EMD. The nonlocal means (NL-means) denoising method is used to preprocess the multivariate signal and the correlation analysis is employed to calculate fault correlation factors to select effective intrinsic mode functions (IMFs). Finally characteristic frequencies are extracted from the selected IMFs by spectrum analysis. The numerical simulations and applications to bearing monitoring verify the effectiveness of the proposed method and indicate that this novel method is promising in the field of signal decomposition and fault diagnosis.