Research On Rolling Bearing Fault Recognition Method Based On Two-Stage Multi-Criteria Feature Selection Model

In view of the low accuracy of fault identification of rolling bearings,which are the core components of mechanical equipment,the theory of feature evaluation and selection is applied to fault diagnosis of rolling bearings from the perspective of signal processing.The multi-domain feature extraction,key feature selection and bearing fault status recognition under variable working conditions are completed.The basic theory of feature evaluation and selection is introduced,and the selection basis of feature evaluation criteria and the necessity of stability evaluation of feature selection are emphatically analyzed.On this basis,a two-stage multi-criteria feature selection model is established,including key feature selection based on multi-criteria fusion and stable feature selection based on stability analysis.In view of the non-stationarity and non-linearity of bearing vibration signals,it is difficult for single feature or single domain feature to fully characterize the change of bearing state.A multi-domain feature set which can accurately describe the bearing state is established,including time-frequency domain,entropy,energy,complexity and low-dimensional reconstruction features.Based on this,the variation characteristics of bearing signal in different states are analyzed.According to the discreteness of the distribution of feature samples corresponding to each state,the classification separability and sensitivity of each feature in multi-domain feature set are preliminarily evaluated,which lays a foundation for subsequent feature evaluation and selection.The input is to establish a multi-domain characteristic matrix.Aiming at the high dimensionality of feature matrix,firstly,a feature selection method based on lasso regularization and spectral clustering is used to remove a large number of irrelevant features in feature set.Secondly,considering the limitations of independent evaluation criteria,a multi-criteria feature selection method based on maximum correlation distance and correlation evaluation criteria is proposed to complete key feature selection.Aiming at the problem whether the key feature set obtained under a single workingcondition can be used to accurately identify the bearing fault state under complex working conditions.By means of feature selection stability analysis,three stability evaluation methods are selected to complete the selection of stable features.The fault pattern recognition of rolling bearings is accomplished by three classifiers,such as pnn,which are input as state eigenvectors,and a high recognition accuracy is obtained.The results show that the feature subset screened by the established feature selection model has good class separability,stability and robustness,and bearings fault state can be effectively distinguished under variable operating conditions.