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

Steam turbine not only continuously operates under high temperature, high pressure and high speed work environment, but also work with complex gas, water, oil, steam system together. It will occur some failures and accidents inevitably. It plays an important role for ensuring steam turbines to operate under a safe and economic conditions.Potential uncertainty factors which induces steam turbine faults increase accordingly.There are many reasons which result in uncertainty factors such as complex system structure,variable speed, dynamic stress and strain, excessive temperature gradient distribution and working medium corrosion, etc. With increasing of the abruptness of mechanical equipment damage, it is a great challenge to identify coupling among faults and uncertainty factors in the steam turbine generator units. The uncertainties of expert system of fault diagnosis are analyzed. A probability uncertainty method of fault classification for steam turbine generator sets based on probability causal relation and the 2D-holospectrum is proposed in this paper.The main contents of this paper are as following:(1) The reasons of uncertainty during the process of fault diagnosis are investigated. The reasons lead to expert system fault diagnosis uncertainty are analyzed in detail. The sources of uncertainty are studied and summarized in expert system fault diagnosis based on case and rules respectively. The solutions are also put forward in this paper.(2) Based on probability causal relation and the 2D-holospectrum, the fault diagnosis models is described in this paper. Firstly, the theories of diagnosis models are introduced.Then, the possibility of fault set covering is determined through selecting fault signs and comparing fault signals with the standard signals in frequency domain graph and2D-holospectrum. Then, probability causal relation theory is adopted to select fault types by prior probability of fault and probability among the intensions of cause and effect. Moreover,the ratio of overlap region corresponding to area in 2D-holospectrum can give the probability of the fault. Finally, the model is verified by the model established by the dynamic equation based on single span single disk rotor system. The results show that the method proposed can achieve the purposes of attribute reduction, pattern recognition and judgment of fault constitution.(3) A kind of method is put forward to judge the dominant factors among multiple faults in this paper. The method is combined the 2D-holospectrum with set theory and normalized frequency information. Firstly, the possibility of fault set covering is determined, fault types is chosen and verified by 2D-holospectrum. Then frequency information is normalized byregarding the ratio of overlap areas between fault signals and the standard signals as the probability of single fault. Finally, combining with set theory, the probability of multiple fault is obtained. It can provide a possible judgment to deal with coupling in troubleshooting.