Research On Early Fault Diagnosis Of Rolling Bearing Based On WPD And Graph Theory

With the rapid development of science technology and the rapid progress of intel-ligent industrial technology,the mechanical equipment in current manufacturing in-dustry is becoming more and more automatic,large,precise and intelligent,which not only makes the structural design of the key parts of the mechanical equipment more complex,but also requires higher manufacturing accuracy,and puts forward higher requirement for the installation and smooth operation of key parts of mechanical e-quipment.Rolling bearing is an essential part of mechanical equipment,especially in rotating mechanical equipment,meanwhile it is also the most vulnerable parts in ro-tating mechanical system,which has been attracting more and more attentions nowa-days.Thus,this study focuses on rolling bearing and explores an unified framework for health status monitoring and early fault diagnosis based on wavelet packet decomposi-tion and graph theory.The performance of proposed method is explored and evaluated from the both simulation and experiment perspectives.First of all,the wavelet packet coefficients are extracted from the vibration signal which is collected during the operation of rolling bearings.Considering the correla-tion between the wavelet packet coefficients of each sub-band,an undirected weighted graph model is constructed for the wavelet packet coefficients of each sub-band to real-ize the dynamic characterization of vibration signals.After the early fault of the rolling bearing was detected,the fault features were further extracted at the early fault time for fault diagnosis.For the issue of early fault monitoring,this study use the undirected weighted graph to model the wavelet packet coefficients.In order to meet different working condition-s,an adaptive weighted method is proposed to fuse the multi-dimensional similarity scores to obtain one dimensional index reflecting the health state of the rolling bear-ing.It is proved theoretically that the health state of the rolling bearing can be reflected by the fused index i.e.,anomaly scores.Furthermore,this study used the anomaly de-cision making algorithm based on the gaussian distribution hypothesis to monitor the fused index,and then assess if any early fault occurred in the process of operation,so as to realize the early fault monitoring.For the issue of early fault diagnosis,this study proposes using K-nearest neighbor(KNN)classifier for early fault diagnosis of rolling bearing.When early fault is cap-tured based on the early fault monitoring,the singular value decomposition is carried out on the graph set at the time of the early fault to construct the fault features.And then,the constructed fault feature is fed to K-nearest neighbor classifier for fault type identification,among which the training samples in the K-nearest neighbor classifier are the singular value sequence features of various health states of rolling bearings.In summary,this paper integrates the early fault monitoring and diagnosis of rolling bearings into a unified framework which enables the early fault monitoring and di-agnosis to be conducted continuously.The performance of presented framework was validated using two publicly-available data set.One is provided by Case Western Re-serve University,and the other is provided by Xi'an Jiaotong University.The result shows that the presented framework can conduct the rolling bearing early fault mon-itoring as well as early fault diagnosis effectively,the performance outperforms the others existing in the literature.At last,the summary and prospect of the main contents and innovation points of this study are summarized,and the future direction of study is discussed.