Research On The Graph Based Health Status Assessment And Early Fault Diagnosis Of Rotating Machinery

As a common machinery,rotating machinery is widely used in various industrial applications.With the development of industrial techniques,rotating machinery is be-coming increasingly large,automated and integrated.The failure of key element will result in heavy losses of life and property.Thus,the health management of rotating machinery has been attracting more and more attentions nowadays.This study focuses on two core problems in the health management of rotating machinery,i.e.,health sta-tus assessment and early fault diagnosis,and explores an unified framework for health status assessment and fault diagnosis of rotating machinery based on graph model.By regarding rolling bearing,one of the most widely used parts in rotating machinery,as the object,the performance of proposed method is explored and evaluated from both theoretical and experimental perspectives.First of all,the short-time periodogram is extracted from the vibration signal which is collected during the operation of rotating machine.Then,via graph modeling strat-egy,this short-time periodogram is modeled as an undirected weighted graph where two types of formulas for weight value calculation are introduced.The first one is used for machine health status assessment,while the anomaly is detected,the second one will be used for fault diagnosis.For the issue of health status assessment,this study use the modeled undirected weighted graph to map the short-time periodogram,resulting in a series of graph spec-tral frequencies.From both theoretical and experimental view,this study demonstrates that the health status of monitored bearing can be reflected by the principal graph spectral frequency.Subsequently,this study proposes using the Gaussian change point detection algorithm to detect the anomaly status during the operation of monitored rotating machine.For the issue of fault diagnosis,this study proposes using K-nearest neighbor(KNN)classifier for roller bearing diagnosis with graph model.When anomaly is captured,the second weight value will be calculated and fed to KNN classifier for diagnosis.The training samples in KNN are the graphs belonging to different fault type.It is worth mentioning that distance metric also plays a critical important role in KNN,influencing the finally diagnosis result dramatically.Thus,this study also investigates the diagnosis performance of KNN under four commonly used graph distance metrics(i.e.,spectral distance,entropy distance,weighted edge distance,modality distance)respectively.Final result shows that weighted edge distance based KNN achieves the best diagnosis performance which also outperforms its competitors in this area.In summary,this paper integrates above two modules into an unified framework which enables the health assessment and fault diagnosis to be conducted continuously.The performance of presented framework was validated using two publicly-available data sets.One is provided by Case Western Reserve University,and the other is pro-vided by University of Cincinnati.The result shows that the presented framework can conduct the rotating machine health status assessment as well as fault diagnosis effectively and the performance outperforms the others existing in the literature.At last,the main contents of this study are summarized and the future work is dis-cussed.