A novel feature extraction method of compound faults of bearing based on ITD and information fusion

Abstract

A method combining signal decomposition and information fusion is proposed to solve the difficulty in extracting the compound fault features of rolling bearing. Firstly, the method decomposes the homologous signals into the sum of a series of proper rotational components, which are obtained by the sensors in horizontal and vertical directions with intrinsic time scale decomposition (ITD). Secondly, autocorrelation noise reduction and normalization processing are performed for the component signals obtained; weight coefficients of signals are adaptively determined according to normalized autocorrelation function (AF) of components. Thirdly, autocorrelation functions of all component signals are weighted and blended according to weight coefficient obtained. Finally, fault feature frequency of rolling bearing is extracted with power spectrum of signals blended. Equally, the failure types of bearings are judged. To verify the effectiveness of proposed method, the data which corresponds to different failure types is analyzed and verified; meanwhile, other methods is compared with presented method. The result indicates that the feature information of compound failure can be extracted precisely, and the compound failure type of bearings can be judged accurately with presented method.