Feature Fusion Research On Variable Operating Condition Assessment Of Rolling Bearing

Rolling bearing, as the critical component of aero-engine, directly affects the air safety. In order to decrease the air accident rate effectively, it is important to conduct state inspection for rolling bearing and to find the failure symptom. The current research of rolling bearing condition assessment mainly focus on feature extraction and data description, the parameters are incomplete and the parameters are not optimized. The effects of loading and rotate speed changing on condition assessment are ignored and the assessments are not accuracy. In this paper, variable operating condition assessment research based on feature fusion is studied. The main research work is embodied in:(1) Three methods for multi-feature fusion are proposed, namely distance discriminance, one-class classification method and posterior probability method. The distance discriminance is based on Euclidean distance by using the vibration characteristics of the normal condition data, and compare the distance between the unknown condition and the normal condition. One-class classification method describes the distribution of normal condition data and test whether the unknown condition data obeys the distribution. Posterior probability method is the SVM arithmetic based on posterior probability. The training samples are formed by using the data of normal condition and major failure condition and then conduct the SVM study. By using the method, classification problems are solved and the probability estimation is obtained by combining the Bayesian Decision Theory.(2) The rolling bearing single point fault simulation experiments are conducted and 4 groups vibration acceleration signals are obtained. 13 dimensionless characteristics are extracted from the time domain, frequency domain and time-frequency domain. By using these characteristics, the sensitivity of characteristics are analyzed and the influence of rotating speed on fault feature are compared. By the means of the multi features fusion, the research results fully prove the correctness of the proposed methods.(3) Rolling bearing performance degradation experiments are conducted, and experiment data of different bearing working conditions, from normal to severe abnormal, are obtained. Condition assessment is conducted by using the above methods and the results are compared. The results show that the condition assessment using the multi feature fusion can distinguish the normal and abnormal conditions obviously, and the validity of the proposed method is fully verified.