Research On Fault Feature Extraction And Classification Recognition Of Rolling Bearings

In the modern machine manufacturing and application,the various components are increasingly inseparable.When one of the components fails,it may affect the normal operation of the entire device.Bearing fault diagnosis technology can extract the effective fault feature information from the rolling bearing signal to find faulty parts as early as possible and avoid some accidents in time.This paper mainly studies the fault diagnosis technology of the rolling bearing system,and focuses on the fault feature extraction,feature selection and fault classification and identification of the rolling bearing.In response to the above problems,this paper has done the following work:First,the feature extraction method of rolling bearing faults is studied.Aiming at bearing vibration signal,an improved fractal box dimension algorithm is proposed,and the information entropy feature extraction algorithm and Hilbert-Huang transform marginal spectrum power feature extraction algorithm are studied.Since single features cannot distinguish fault types well,this paper uses the features generated by the three extraction methods as the combined features.Second,the feature selection method of rolling bearing faults is studied.Three kinds of feature selection algorithms,including Sequential Forward Selection(SFS)algorithm,Sequential Floating Forward Selection(SFFS)algorithm and ReliefF algorithm,are introduced.Using these three feature selection algorithms,all the features from feature extraction are selected,an optimal subset of features is selected,and simulation results are analyzed.Simulation results show that the feature subset generated by ReliefF algorithm can well represent the original feature set.Third,the classification and identification methods of rolling bearing faults are studied.The principles of three classifiers are introduced,including Adaboost,Gradient Boosting Decision Tree(GBDT)and Xgboost classifier.These three classifiers and two traditional classifiers including k-nearest neighbor algorithm(KNN)and support vector machine(SVM)are used to classify features,and the results are analyzed and compared according to the simulation results.The innovation points of this paper are as follows:1.In this paper,an improved fractal box dimension algorithm is proposed for frequency spectrum and window frequency shift of vibration signals,which greatly reduce the amount of computation,save the hardware resources needed in the sampling process.Also the high accuracy rate can be ensured.2.A feature selection method combining ReliefF algorithm with GBDT classifier is proposed,which effectively improves the recognition rate of rolling bearing fault diagnosis,and is superior to other classifiers and feature sets.The accuracy rate reaches 99%.Compared with Xgboost and other algorithms,it can effectively save hardware resources while ensuring high recognition rate.Experiments show that the bearing fault detection method proposed in this paper has the advantages of high accuracy,low computational complexity,low hardware requirements and so on.It has great application value in practical engineering.