Research On Bearing Fault Early Warning And Diagnosis Method Based On Feature Integration Interactive Association Analysis

As one of the important parts in mechanical equipment,rolling bearing plays an important supporting role in rotating machinery.However,in actual working conditions,the working environment of the bearing is often relatively harsh,and long-term overload operation will further accelerate the fatigue wear and crack deterioration of the bearing.Therefore,the irreparable accidents caused by the failure of rolling bearings are numerous.How to issue early warning at the early stage of rolling bearing failure and diagnose the bearing failure position and failure degree with a small sample size is of great significance to ensure the smooth operation of the equipment and the life safety of the operating workers.In addition,due to the time-consuming and laborious disassembly and assembly of mechanical equipment such as fans,false alarms during normal operation of the equipment are also a problem that cannot be ignored.In view of the above problems,this paper analyzes the significance of the correlation between variables to the fault diagnosis of mechanical equipment based on the correlation of fault features.Based on the correlation between features,neural network and extreme value theory,a feature-based Correlation and adaptive threshold of bearing fault early warning method,and a rolling bearing fault recognition method suitable for small samples and multi-classification in complex working conditions.Based on the filtering feature selection method,a hybrid feature selection method with improved maximum weight and minimum redundancy is proposed.The main research contents of this paper are as follows:(1)Aiming at the problem that the traditional feature selection method cannot remove the redundancy between features while selecting good features,a hybrid feature selection method with improved maximum weight and minimum redundancy is used to select features: this method uses common filtering of various characteristics Based on the feature selection method,based on the specific evaluation function and the accuracy of the specific classifier,the optimal feature subset with the largest weight value and the smallest redundancy between features is selected.Experiments show that this method can achieve good results in the subsequent bearing classification when the Variable Predictive Model Based Class Discriminate is used as the classifier.(2)Aiming at the problem that the fixed threshold will be falsely alarmed due to extreme conditions during normal operation,the extreme value theory and the variable prediction model are combined to perform abnormal detection on the rolling bearing.This method uses the Variable Predictive Model Based Class Discriminate to fit the normal bearing data and test The reconstruction error of the sample is used as the indicator of the running state of the bearing,and the adaptive threshold is set by using the generalized extreme value theory.Experiments show that this method can accurately alarm when the bearing fails early,and effectively avoids false alarms caused by extreme conditions.(3)Aiming at the problem that the common classifiers have low accuracy on the small sample multi-classification problem,the Variable Predictive Model Based Class Discriminate of weight fusion is used to identify bearing faults.This method adds the Gaussian model and the Gaussian model to the original four models of the Variable Predictive Model Based Class Discriminate.RBF neural network and generalized regression neural network are used to fit the relationship between features,and the training error of the model is used to calculate the weight of each model,and finally a fusion model of weights is obtained.Experiments show that this method still shows high classification accuracy in the case of small samples and multiple classifications;and it is superior to the original Variable Predictive Model Based Class Discriminate and commonly used pattern recognition algorithms in recognition accuracy.