The research work in this paper elaborates on the theoretical effectiveness of the proposed method based on the multivariate EMD. It also clearly indicates through numerical simulations and applications to bearing monitoring that the expansion from standard EMD to multivariate EMD is a successful exploration. Using multiple sensors to collect signal from different locations of the machine and using the multivariate EMD to analyze multivariate signal can contribute to comprehensively collect all the frequency components related to any bearing fault, and is beneficial to extract fault information, especially for early weak fault characteristics. Both the characteristic frequencies of simulated signal and the fault frequencies of practical rolling bearing signal can be extracted from the same order of IMF groups, thus showing that multivariate EMD is an effective signal decomposition algorithm and can be competently applied to fault diagnosis of rolling bearings when combined with a multiscale reduction method and fault correlation factor analysis. In signal acquisition and processing, given the circumstance that there is a trend toward the use of multiple sensors, multivariate EMD appears to be very useful and meaningful as a kind of multivariate data processing algorithm. By analyzing the simulated signal and two different practical multivariate signals, the results demonstrate the significance of the proposed method in the field of fault diagnosis of rolling bearing.
