Research On Fault Diagnosis Method Of Rolling Bearing Based On Machine Learning

With the increasing demand for high-performance modern equipment,the requirements for equipment have also grown toward large-scale,automated,and complex.The more complex and precise the equipment is,the more difficult it is to monitor and troubleshoot the equipment.Rolling bearings are the most commonly used components in rotating machinery.The normal operation of rolling bearings often directly determines the performance of the equipment.At present,Researchers are introducing unsupervised learning into fault diagnosis to realize automatic feature extraction and recognition.So this paper focuses on fault diagnosis of rolling bearing based on unsupervised machine learning and mainly research the fault diagnosis process fault detection,fault feature extraction and fault pattern recognition of these three stages,And a complete fault diagnosis method of rolling bearing is proposed.According to the rotating machinery failure vibration signal non-stationarity caused by the fault detection difficulties,in this paper,the fault detection method of rolling bearing based on high-order spectral analysis is studied,and a rolling bearing fault detection method based on fuzzy information granulation and high-order spectral analysis is proposed.The more sensitive features after granulation are selected to constitute effective parameter information,and the higher-order spectrum analysis of third-order statistics is used.The simulation results show that the rolling bearing vibration signal granulation greatly reduces the data dimension of the bispectrum,and meanwhile,it can also express the fault information of the rolling bearing vibration signal,which can be applied to the fault detection of the rolling bearing.In order to realize the automatic feature extraction of rotating machinery failure and avoid the transition dependent man-made feature calculation,the fault feature extraction method of rolling bearing based on extreme learning machine and sparse self-encoder are respectively adopted.This paper introduces the basic principle of extreme learning machine,constructs a deep learning model built by multi-layer extreme learning machine and uses it to extract the feature of the generated bispectrum.Then for the unsupervised learning method of complex network structure,the parameters of the unsupervised learning method are long and the training time is long.However,the accuracy of network diagnosis with simple structure is not ideal.Therefore,a new feature extraction method based fuzzy information granulation and sparse self–encoder constructed parallel structure is proposed.At the same time,performing feature extraction on multiple effective parameter information.The simulation results show that the eigenvectors extracted by the two methods can effectively express the fault characteristics of the rolling bearing.In order to realize the intelligent diagnosis of the rolling bearing fault,a rolling bearing fault identification method based on stochastic forest is proposed.The fusion classification experiment is carried out by using the fault eigenvector extracted from its own characteristics,classification experiment based on fault feature of rolling bearing is extracted by parallel structure based on sparse self-encoder.The simulation results show that the random forest has a high accuracy as a diagnostic classifier.Finally,this paper presents a complete intelligent fault diagnosis method of rolling bearing based on fuzzy information granulation,sparse self-coding learning and random forest,and presents a complete intelligent fault diagnosis method of rolling bearing based on fuzzy learning granulation,higher order spectral analysis and deep extreme learning machine auto encode.Experiments were carried out using the bearing data from Case Western Reserve University in the United States.The results show that both of them can effectively realize the fault diagnosis of rolling bearing.