Signal segmentation for isolating the influence of PQ variation and machine manufacturing imperfections on bearing fault detection

The purpose of this research is to introduce a tree-based signal segmentation technique for improving the reliability and sensitivity of bearing fault detection by Electrical Signature Analysis (ESA). To illustrate the motivation of the proposed approach, the nature of bearing anomaly especially the generalized roughness is employed to show the difficulty to detect the bearing defect by electrical signal. Furthermore the influence of Power Quality (PQ) variation and machine manufacturing imperfections on the bearing defect by ESA was detailed, which also verified by the experiment observation. The proposed signal segmentation technique can be used to adaptively partition the non-stationary voltage signal into locally stationary process sets, which can be concatenated into a set of approximately stationary signal set for processing. The corresponding segmented current signal can also be grouped together with voltage signal according to the distance between different groups of approximately stationary signal. Finally model-based detection scheme based on electrical signals is applied to track the trend of bearing health condition change. To verify the effectiveness of the proposed procedure, experimental results are provided for different load levels.