The Study On Fault Diagnosis Technology Of Steam Turbine Vibration Based On Decision Tree Algorithm

With the increasing development of power technology, the capacity and parameters of generating unit are improved and the thermodynamic equipments in power plants are becoming complicated and intelligent. This all leads to the complication of turbine fault. It will cause tremendous economic losses to the companies and our country if the turbine goes wrong. Data mining has many advantages in dealing mass data. As a result, it will be an efficient way to attain knowledge if data mining and fault diagnosis system of turbine are combined together. We can attain diagnostic rules from historical data by the ways of data mining to diagnose faults.In the paper, the decision tree C4.5 algorithm is applied to the fault diagnosis of steam turbine unit. The simulation experiments are conducted on the experimental platform of steam turbine rotor. The vibration tests include:normal running of rotor, imbalance of rotor, friction of rotor, oil film oscillation at the rotation rate of 4000r/min, and each running condition was tested for 20 times. Statistical characteristic parameters are calculated by matlab from primary vibration data.These parameters includes skewness, kurtosis, average value and maximum value.40 sets of training data sets are randomly selected for each operation. The rest sets of vibration data form the testing set. The training set will be processed by the C4.5 algorithm according to the procedure of fault diagnosis based on C4.5 algorithm. Diagnostic model of the decision will be tested by the testing set. The diagnosis accuracy of the model is 90% after be tested by the testing set. The C4.5 algorithm can be applied to the system of fault diagnosis on steam turbine after it has been optimized. The framework of the fault diagnosis system based on C4.5 algorithm has been constructed and the functional modules are separated based on the system framework. The software architecture and the diagnosis interface of the system have been designed preliminarily. The basic functions of the diagnosis module based on C4.5 algorithm on also have been realized.