Research On Fault Diagnosis And Prediction Of Gas Turbine Lube Oil And Fuel System

Lube oil system and fuel system are very important to ensure the normal andstable operation of gas turbine. By providing lube oil of certain temperature andpressure, lubricating oil system helps the gas turbine prevent the occurrence of asurface friction and reduce mechanical wear. On the other hand, fuel system, which isresponsible for supplying the operating machine with fuel, shows great influence onthe economy and stability of gas turbine. In this paper, the lubricating oil system andfuel system of gas turbine that is used to generate electricity are investigated,especially focusing on the following three aspects, that is modeling, fault diagnosisand fault prediction.At the outset of the paper, it analyses the working principle of lubricating oilsystem and fuel system, and common faults are obtained. The possible causes andconsequences of the faults are further analysed combining the frequency of the faultsoccurred in the process of running system. The main components of lube system andfuel system are modeled, including pumps, coolers, filters, piping, valves, oil tank, etc.Moreover, by simulating the main six faults of both systems, the altering trends ofrelevant parameters are obtained when the faults occur. These trends are used as thereferences for the consequent fault diagnosis and prediction.In the second part, according to the features of lube oil and fuel systemsmonitoring parameters, the rough set theory is selected as the fault diagnosis method.The basic theories and processes of modeling are introduced to decide the methods ofdata discretization and attribute reduction. The decision table is set up according to theextraction of fault data and the reduce attribute is performed. The feasibility ofreduction results is discussed and the optimal sets of properties are obtained. Thisfinally provides the diagnostic rules. The results show that the established model issuitable for dealing with uncertainty and small samples, with a high diagnosticaccuracy.In the third part, the self-organizing feature maps (SOM) neural network isselected as the fault prediction method. Taking the characteristics of input and outputparameters into account, proper network structures and learning algorithm are determined. The prediction results are displayed by three visualization modes (Umatrices, principal components chart, and data tracking chart). By extracting the faultsdata as the training sample, the faults prediction net model is developed. It is shownthat the SOM network model not only can intuitively reflect the health condition of thesystem and estimate the failure trends, but also shows bright application prospects andpractical value in the field of failure prediction.