Fault Diagnosis of Rolling Bearings Based on Undirected Weighted Graph

One of the main functions of rolling bearing condition monitoring is to diagnosis the type of fault that is occurred during its continuous operations. This paper presents a new method for rolling bearing fault diagnosis based on the graph model. Concretely, through Fourier transform, the periodogram is computed from the condition monitoring (CM) signal and then modeled into an undirected weighted graph. This graph is subsequently fed to K-Nearest Neighbor (KNN) Classifier for fault type diagnosis. In particular, to perform KNN upon graph model, a robust graph distance metric so-called sum of the difference in edge-weight values (SDEWV) is adopted via investigating four candidate metrics existed in the literature. Based on experimental results in the publicly-available database, we demonstrate exciting results of the proposed method in bearing fault diagnosis, indicating its great potentials in real engineering applications.