Fault Detection of Motorized Spindle Based on Clustering by Density Peaks

Motorized spindle is the critical component of CNC machine tool. Accurately detecting fault in motorized spindle can effectively reduce maintenance time and corresponding costs. Feature extraction and fault recognition are the two main aspects in fault detection based on data-driven method. In this study, frequency interval energy and empirical mode decomposition energy entropy are selected as the features of signal. A clustering approach by finding density peaks is then applied to adaptively classify these features. The decision graph is modified based on local density ratio to overcome the disturbance of noise or marginal points in the process of searching density peaks with low local density and to automatically cluster the features. Bearing fault test is used to indicate the necessity for improvement. The effectiveness in fault recognition of the proposed method is verified by fault data of motorized spindle. Finally, recognition results are compared with traditional fuzzy c-means clustering. This comparison reveals that the suggested approach can recognize various fault conditions of the motorized spindle with high accuracy.