Research On Bearing Fault Identification Model Based On Support Vector Machine

Rolling bearing is one of the most commonly used components in rotating machinery.Its normal working state directly affects the performance of the whole equipment,even the safety of the whole production line.Therefore,the research of rolling bearing condition monitoring and fault diagnosis technology has important theoretical significance for avoiding accidents and reforming the maintenance system.In this paper,the rolling bearing is taken as the research object,aiming at the key problems of the rolling bearing fault diagnosis,such as feature extraction,pattern recognition model,etc.,a series of research work is carried out from the bearing vibration signal processing^[1].First of all,the background and significance of this topic are described,and the research status of rolling bearing diagnosis information acquisition,fault feature extraction and fault pattern recognition at home and abroad are summarized.On this basis,the research ideas and contents of this paper are established.This paper analyzes the basic structure and vibration mechanism of rolling bearing,lists the main forms and causes of rolling bearing failure^[1],and makes a solid foundation for the research of bearing fault diagnosis and pattern recognition.Secondly,the adaptive wavelet threshold denoising method is used,which is implemented according to the data of the electrical laboratory of Case Western Reserve University.Compare the noise reduction signals before and after normal bearing condition,bearing inner ring fault,rolling element fault and bearing outer ring fault.The feasibility of wavelet adaptive threshold denoising is proved.Then,the entropy analysis method is selected to study the problem of bearing fault feature extraction.According to the data of rolling bearing,two kinds of entropy with the highest discrimination are selected for research,namely arrangement entropy and approximate entropy.Then,the time-domain signal,frequency-domain signal and envelope signal of the data are analyzed,the multi-domain eigenvalues of the bearing signal are extracted,the extracted time-domain entropy,frequency-domain entropy and envelope entropy eigenvalues are combined into the eigenvector as the input of the classification model,and are input into the training model and tested^[2].Finally,multi domain eigenvectors are input into the Support Vector Machine model（SVM model）and the particle swarm optimization（PSO）algorithm is used to find the optimal solution of the parameters in the RBF kernel.Then the function comparison between SVM model and PSO-SVM model is carried out and the conclusion is drawn.Through the comparison between SVM and PSO-SVM model and multi domain entropy eigenvector as input,it can be concluded that PSO-SVM model has a strong adaptive ability,does not need to set parameters manually,and the recognition rate is higher.When multi domain eigenvectors are used as input,the effect of time domain eigenvectors is better,the recognition degree is higher,the envelope eigenvectors are the second,and the frequency domain eigenvectors are the worst.By using this multi domain eigenvector as input and comparing SVM with PSO-SVM,the fault identification model of rolling bearing is studied systematically.