Fault Diagnosis Technology For Nuclear Power Plants Based On Deep Expert Knowledge

In order to ensure the safe operation of nuclear power plants,research of fault diagnosis technology for nuclear power plants carried out is necessary and urgent.The focus of fault diagnosis technology is on computer modeling,which aims to strive to accurately determine the current system operation status in the shortest possible time and playing a role in assisting operators in decision-making.Compared to analytical models and data driven fault diagnosis methods,expert system-based fault diagnosis methods have obvious advantages in simple modeling and fault path display.However,inherent bottlenecks such as difficulty in obtaining deep expert knowledge,inability to diagnose unknown faults,and inability to quantitatively diagnose faults also affect the application of expert systems in nuclear power plants.Due to the complexity of nuclear power plants,there are many alarm parameters after a fault occurs which greatly tests the operators’ ability to respond on the spot.Based on the above issues,bond graph theory is applied to the modeling of expert system inference engines and a decision support system interface is developed in this study.The purpose is to increase the engineering applicability of expert systems in nuclear power plants to a certain extent and to help operators extract important steps from complex procedures,reduce the burden and avoid human error.The main contribution of the thesis is listed as follows:Bond graphs,time causality graphs and variable relationship graphs are established for main equipment of the reactor coolant system.The effort-flow relationships of the pressurizer,the steam generator,the main pump and the reactor are analyzed in order to establish bond graph models through in-depth understanding of bond graphs.Based on the bond graph models,time causality graphs and variable relationship graphs are established to obtain deep expert knowledge of the equipment.Finally,fault signature matrixes are established to obtain the effort-flow relationships which lay a foundation for diagnosis.Combined with the causal relationships of variables and the fault signature matrixes,a nuclear power plant fault diagnosis expert system based on bond graph theory and the expert system G2 platform is developed.By verifying the output fault data from the simulator PCTRAN,the specific equipment level faults can be diagnosed well.It is shown that the propagation paths of parameters and effectively interpret deep knowledge within the equipment such as effort-flow relationships could be reflected by the system.Therefore,a new idea for the design of expert system inference engines could be provided.A nuclear power plant fault diagnosis system based on signed directed graph(SDG)models is constructed.In order to carry out system level faults diagnosis and provide more levels of knowledge to decision support interface,a nuclear power plant fault diagnosis system using the SDG methods is established by this study.Twenty-five SDG models are established and the accuracy of the results is verified by simulation.A decision support system based on expert knowledge is constructed.In order to avoid mis operation of procedures by operators under a large number of alarms conditions,a decision support system is designed by this study.