Research on Fault Feature Extraction and Pattern Recognition of Rolling Bearing

Abstract

Rolling bearings play a very important role. If the state of the rolling bearing is wrong, the whole equipment will fail, so we need to check the state of the running bearing. Fault Feature Extraction of Rolling Bearings. CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) analyzes the collected oscillation signals to obtain various natural state functions. Kurtosis and Correlation Factor Evaluation Index are used to screen IMF components with a large number of error data, and independent component analysis is carried out on the selected components. The corresponding processing and reconstruction are carried out to extract the feature frequency. The built-in signal processing method can deal with the problem of bearing fault signal and identify the specific fault frequency. Fault Pattern Recognition of Rolling Bearings. The decomposition acceleration sensor is used to analyze the data of bearing in different states, and then, several components are obtained. The components related to the error signal are found, and the corresponding feature vectors are calculated. Then, the dimension vector is reduced, and a particle flow optimization algorithm is used to identify specific error conditions after dimension reduction. Experiments show that this method has higher detection rate in identifying specific fault states.