A Probability Method For Steam Turbine Generator Set On Uncertainty Of Fault Classification Based On Bayes And Holospectrum

As an important equipment of power plant, the unplanned shutdown of the steam turbine generator units would not only result in huge economic loss but also cause great panic to people’s life. Rapid development of the machinery stimulates the increasing complexity of the steam turbine generator set, which gives rise to the danger and the failure ratio to the steam turbine generator units.The fault diagnosis system of steam turbine has the characteristics of the wide source, a variety of information, etc. In addition, the input information may be disturbed, incomplete and captured under many uncertainty factors. The incompleteness and impreciseness of information collected in fault diagnosis system and may cause the certainties in fault diagnosis system. A probability method for steam turbine generator on the uncertainty reasoning of fault classification is proposed in this paper, which is based on the2D-holospectrum and Bayesian decision theory.The main content of the thesis includes:(1) The sources and reasons that the uncertainty factors occur in fault diagnosis expert system are studied about a steam turbine generator. The representation of uncertainty and the uncertainty reasoning process are analyzed compared in some widely used uncertainty reasoning models. The sources and causes of uncertainty are analyzed and summarized in both the fault diagnosis expert system based on case and based on rule.(2) A probability classification model of the uncertainty reasoning based on2D-holospectrum and Bayesian decision theory is described in this paper. The theoretical foundation and the detail steps of the model establishment are provided. The completeness of the sample spaces is discussed. The frequency bands where faults located can be selected through the comparison between the fault spectrum and the standard spectrum. Next, types of fault can be identified according to the Shore’s chart. Bayesian decision theory is adopted for the preliminary fault estimation. Finally, Overlap of2D-holospectrum between the fault signal and the standard signal can indicate the degree that the fault occurs. The ratio of overlap region to the corresponding area in2D-holospectrum and the evidence theory can give the probability of fault. The model is verified by the experimental data performed on the rotor vibration test bed. The results show that the method proposed is feasible for reasoning with imperfect information. (3) Further improvement of the faults classification on multiple faults is also studies.The coupling relationship between the multiple faults is also considered into the new model. The theory about the improved model shows the classification model is feasible dealing with multiple faults.