Research On Fault Diagnosis Technology Of Large Steam Turbine Based On Information Fusion

Neural network possessed well in non-linear mapping, and D-S evidence theory has advantage over uncertainty, which are widely used in the filed of diagnosis.A new fault diagnosis method was put forward based on BP neural network, Elman neural network and D-S evidence theory in this paper. The outputs of each network were regarded as the inputs for evidence theory, these outputs of network could be fused in order to reduce the uncertainty and enhance the nicety. Furthermore, this method could be used to make full use of the fault phenomenon information for steam turbine.According to particulars of stream turbine fault diagnosis, the objectives and important significance for stream turbine fault diagnosis were introduced, and the development of stream turbine fault diagnosis at home and abroad was expatiated also. And then, the systematic analysis about the vibratory mechanism, main faults, fault symptoms and fault feature extraction was made.Aiming at the shortcoming of satisfactory result for the steam turbine couldn’t be obtained by a single network, this complex problem could be decomposed into many simple issues then single network to diagnosis could be utilized. After reviewed the fundament principles of BP net and Elman net, the characteristic frequency spectrum of vibration for steam turbine was got as neural network input vectors after unitary. Then the BP Network was trained by these input vectors. The outputs of two neural networks were turned into degree of belief for the evidence of D-S evidence theory. Finally, the outputs of the two networks were fused by D-S evidence theory and the final results of diagnosis were got also.Lastly, the new method of diagnosis for steam turbine based on neural network and D-S evidence theory has been developed by MATLAB R2009b. Compared with the results of simulation which BP net or Elman net was only used in the diagnosis method, the uncertainty of fault diagnosis was reduced by new method in this paper. The main work in this paper was summarized, and the shortage and direction of further research in the future was also discussed.