Early Fault Detection Of Rolling Element Bearings Based On Graph Modeling Of Component Signals

Early fault detection of rolling element bearings (REB) is of vital importance for avoiding unforeseeable machine failures, and it makes helps to operators to shut down the machine for maintenance in the early stage, decrease the operation and maintenance costs. And most importantly, it helps reduce the occurrence of greater safety accidents. Component signal is a powerful tool to describe the non-stationary and non-linear signal at different scale. Motivated by recent advances in graph modeling, we propose a novel method based on graph theory and local mean decomposition (LMD) method in this paper. We first utilized the component signals decomposed by LMD as inputs for graph modeling, thus a better description of signal state with multi-scale information can be obtained. Based on the generated graph, the anomaly score is calculated to support a null hypothesis testing for fault detection. In order to validate the effectiveness of our method, experiments are carried out on the publicly available XJTU-SY database. We also make comparisons with some benchmarking methods, it also shows the robustness and priority of our method.