基于模糊C均值聚类和转子轴心轨迹特征的转子状态诊断

Aiming at the problem of using existing axial center orbit features to identify rotor fault level having lower recognition accuracy and poor effect, a new feature extraction approach for rotor axial center orbit was proposed based on rotor axial center orbit quadrant information entropy. With this method, rotor axial center orbit was divided into four ranges according to quadrants, the information entropy of each range was calculated, respectively and taken as fault features. Then the fuzzy clustering was applied to do fault pattern recognition and fault level one. Effects of mesh size on clustering effect were analyzed to judge the method for determining key parameters in process to acquire quadrant information entropy. The stability of the fuzzy C-mean value clustering was improved by initializing clustering center. Fault simulation tests with different patterns and different levels were performed on a test platform. New indexes proposed here were compared with existing rotor axial center orbit features. The results showed that the proposed approach has a remarkable performance in recognition effect and data visualization, and can provide a new idea for further real time state monitoring and fault accurate diagnosis.