Visibility Graph based Feature Extraction for Fault Diagnosis of Rolling Bearings

Abstract

In the vibration signal of rolling bearing, there are correlations between different frequency components, which is an important information for rolling bearing fault diagnosis. In this paper, in order to extract the correlation information effectively, visibility graph based feature extraction is introduced. Firstly, power spectrum is extracted by the welch method, therefore the signal is converted into frequency domain. Then, through the visibility graph (VG) modeling of power spectrum, correlation information between frequency components is obtained. After that, eight representative features will be extracted from VG, i.e., edge numbers, link density, average closeness centrality, average degree, graph entropy, average distribution weight, weight entropy, and clustering coefficient. Finally, the fault types are diagnosed by one-versus-one support vector machines (OVOSVMs). Compared with the existing fault diagnosis methods, it is proved that the proposed method can diagnose the faults of rolling bearing effectively.