Study On An Uncertain Reasoning Method Based On Multilevel Flow Models

When fault or accident occurs in nuclear power plant, the operators in the main controlroom have to perform fault diagnosis in time to analyze and locate the fault according to thesymptoms of system. However, the causalities between faults and symptoms are usually verycomplex because nuclear power plant is huge and complex and the operation may be affectedby the environment. The correctness of diagnosis result can be also affected due to manyrandom, fuzzy, or uncertain factors, such as the uncertainty of the operating informationcollected from sensors and the uncertainty of reasoning rules for diagnosing.On the basis of Multilevel Flow Models (MFM), this thesis presents an uncertainreasoning method for dealing with the effect to the diagnosis results caused by the uncertaintyof fault diagnosing rules. The proposed method consists of the following four procedures.(1) Establish MFMs of the target system and describe the uncertain causalities andimpact intensities between faults and function states.(2) Extend function states into mini-causal trees according to the relationships of theMFMs.(3) Connect mini-casual trees into fault tree based evidence model according to theoccurred system symptoms.(4) On the basis of evidence model and applying Bayesian principle, calculate thehappening probability of each fault under given faulty symptoms.In order to validate the proposed method, this thesis first establishes MFMs for the stablepowered operation of Pressurized Water Reactor (PWR) nuclear power plant, then performsqualitative diagnosis and uncertain diagnosis for some typical faults including Loss ofCoolant Accident (LOCA) and Steam Generator Tube Rupture (SGTR) respectively, andfinally compares the diagnosis results. In all inducted cases, the diagnosis results arereasonable which validates the correctness of the proposed method. The proposed method isexpected to assist the operator’s decision-making when they perform in-depth fault diagnosis.