CEEMD-assisted bearing degradation assessment using tight clustering

Rolling element bearing is a critical component of various rotating machineries. As the demand of reliability of machinery gradually increases, the accurate diagnosis of bearing degradation becomes increasingly important to ensure safe production and reduce operation cost. With more knowledge and data of the bearing degradation accumulated, vibration data of bearings with different fault patterns and indicators are obtained. A diagnosis model with self-learning capability helps the model to understand various features in different degradation stages of bearings. Hence, the model provides more accurate diagnosis information of the current conditions of bearings. In this paper, a tight Gaussian mixture clustering unsupervised learning algorithm is implemented with the assistance of an optimized complementary ensemble empirical mode decomposition (CEEMD) to diagnose the damage severity of rolling element bearings. The obtained information is used for characterizing the severity of damage existed within the machine and facilitating the decision-making of machinery maintenance. The experimental vibrational signals of rolling element bearings are decomposed using the improved CEEMD. After obtaining the critical intrinsic mode function from the CEEMD, the features are calculated, and a tight clustering algorithm is implemented to categorize the bearing degradation stage. The tight clustering algorithm overcomes the incapability of traditional clustering algorithm in distinguish scattered features. A more stable categorization is generated by using the proposed algorithm. Less quantity and more accurate training data are used to improve training efficiency. The proposed model can be implemented in expert systems to distinguish different degradation stages with a self-learning capability.