Application And Research On Machinery Fault Diagnosis Based On Complex Network Community Clustering

With the progress of industrial technology, large complex machinery aredeveloping toward the large-scale, complexity, integration. When major faults ofequipment take place, it will seriously affect the industrial production and lead toeconomic loss. Therefore, making accurate diagnosis for large complex machinery,and ensuring the safe and accurate operation of mechanical equipment is one of thecurrent researching hot-spots in the field of mechanical fault diagnosis. Complexnetwork has been a new research method in recent years, and is an important tool andmodel used to describe complex systems. It will treat elements in the system asnetwork nodes, and the edges of nodes represent the connection between elements.Through the analysis of the nodes and edges, we can find out the features such asself-organization, self-similarity, small-world, and so on. The small world showssmall connection of interconnected nodes and these collections are closely combinedwith less external connection. This feature is also known as community feature. It justcorresponds to the characteristics that the relationship between similar fault samplesis tight in the field of fault diagnosis, and the connection between different faultsamples is sparse. The paper will treat fault samples as the nodes in complex networks,build up the complex network model of fault samples, and carry out the analysis ofcommunity clustering diagnosis. The main work is just as follows:（1）It carried out analysis of fault samples complex network communityfeatures, to determine the factors of similarity function, network side power, edgethresholds, and so on. And fault samples network model was set up; it researched thecalculation method based on mutual information to evaluate important nodes in thenetwork model and verified the superiority of this method compared with othermethods.（2）It studied the fault diagnosis method based on complex network communityclustering, applied complex network community structure characteristics to divide thenetwork into a number of communities, used the changes of community module valueto carry out community cluster, and finally merged community corresponding todifferent fault types, it realized diagnosis; It verified the effectiveness of the methodby the rolling bearing fault diagnosis examples.（3）It studied improve K-means clustering diagnosis method based on complex network community clustering. For the drawbacks of K-means clustering algorithmwhich depends on the initial cluster number K value and the initial clustering center. itused complex network community clustering for K-means clustering algorithm todetermine the K value, through calculating correlation of network nodes to selectimportant node as the initial clustering center, carried out the clustering diagnosis. Iteffectively overcomes the difficult problem that is to choose K-means clusteringalgorithm of K value and the initial clustering center. It verified the effectiveness ofthe method by the rolling bearing fault diagnosis examples.（4）It studied compound fault feature separation method based on complexnetwork community clustering. It applied empirical mode decomposition todecompose compound fault signal into a number of the IMF component of differentfrequency bands, each IMF component as the network community, merged similarcommunity, and finally got the community corresponding to different single fault,achieving the effective separation of compound fault. It used unbalanced rotor andbearing inner ring compound fault and bearing inner ring and ball compound fault asexamples to verify the effectiveness of the method.