Fault diagnosis method of rotating bearing based on improved ensemble empirical mode decomposition and deep belief network

Abstract

Abstract Based on the analysis of the bearing rotating speed feature and the vibration analysis technique, a novel fault diagnosis method of rotating bearing by adopting improved ensemble empirical mode decomposition (EEMD) and deep belief network (DBN) was proposed. Firstly, the EEMD method is adopted to decompose the collected vibration data into the combination of the several intrinsic mode functions (IMFs). Then the spectrum of IMF components and the spectrum of original data are compared to eliminate the false components and interference signals. Because the redundant extreme points or pseudo-extreme points seriously affect the construction of the mean curve of EEMD algorithm, three methods are put forward to eliminate the pseudo extreme points and the redundancy extreme points so as to improve the algorithm performance. Finally, the remaining IMF components are entered into the DBN to extract the data features and realize the fault diagnosis. Simulation results on the rolling bearing data of the Bearing Data Center in Case Western Reserve University show the effective of the proposed method.