Research On Gas Turbine Fault Diagnosis Based On Complex Network

As a typical complex system,there is a nonlinear relationship between faults and symptoms of gas turbine.As a kind of research method of complex system,complex network can analyze the internal connection between each node in gas turbine according to the network characteristics,and realize the analysis and judgment of the fault from the point of view of the whole theory.It is helpful to excavate the correlation and characteristic pattern hidden in the running state of gas turbine.Therefore,based on the complex network theory,the method for modeling gas turbine network is studied.The topology characteristics of gas turbine network under the fault and the severity of different faults are analyzed.The diagnosis method of typical gas turbine fault and fault severity based on complex network is put forward,which provides a new method for the fault diagnosis of gas turbine.The specific research contents are as follows:(1)The method of establishing complex network for gas turbine is studied.The characteristics of complex network are analyzed from three aspects which are microcosmic,mesoscopic and macro.In the light of the gas turbine thermal fault diagnosis,the low dimensional gas path parameters are used as the node and the inverse proportion similarity function is used as the edge to establish the gas turbine network.Aiming at the fault of gas turbine bearing,the symbolic time series method is adopted to reduce the dimension of the high-dimensional bearing vibration frequency to the symbol coding probability node.The relationship between the average path length,the aggregation coefficient and the similarity coefficient and the threshold value of the gas turbine complex network is obtained.The gas road fault and bearing fault gas turbine complex network is realized.(2)A gas turbine fault diagnosis method based on traditional modular degree community detection algorithm is studied.Based on the prior knowledge of gas path fault modeling and bearing physical failure test,a gas turbine fault diagnosis method based on modular community detection is proposed,and the gas turbine state recognition problem is converted to the network community detection.The gas turbine fault training set network and the bearing training set network are established.The accuracy and generalization ability test of the fault diagnosis method are carried out.It shows that the proposed fault diagnosis method based on the module degree community detection can realize the diagnosis of the typical fault of gas path and bearing,and has generalization detection ability for the unknown fault.(3)Fault diagnosis of gas turbine based on automatic classification community detection is studied.For realizing automatic classification of the gas turbine network community and solving the problem that the module degree societies community detection rely on the prior knowledge,a method is proposed which the average clustering coefficient and the average path length are used to find the threshold of the optimal community characteristics of the gas turbine operating state network.A gas path fault training set network of gas turbine and a bearing training set network are set up.The change trend of average clustering coefficient and average path length of gas turbine network under different thresholds is analyzed.The accuracy and generalization ability test of fault diagnosis method have been carried out.It is proved that the gas turbine fault diagnosis method based on automatic classification community detection can realize the diagnosis of typical fault of gas path and bearing without prior knowledge,and has the generalization and self-learning ability of diagnosis knowledge.(4)Research on fault severity diagnosis of gas turbine.Based on the prior knowledge of gas turbine components fault,a diagnostic method for the severity of component faults based on modular community detection is proposed.The fault diagnosis of performance decline of gas turbine gas path and the fault degree of bearing with different sizes is carried out respectively.It is proved that this method can diagnose the fault severity of gas turbine components.In view of the characteristics that fault severity is gradual changed for the whole life cycle of the gas turbine,the life cycle time sequence is taking as the node.The network average path length is used to find the optimal threshold,and the network model of the whole machine fault severity is established.The relationship between the fault severity of the whole life cycle and node degree of the sparse network node is obtained,and the diagnosis method of the fault severity of the whole machine fault based on the node degree is proposed.The simulation test proves that it provides a new method for the diagnosis of gas turbine fault severity.