A fault diagnosis method using bond graphs in an expert system is proposed for a reactor coolant system. Firstly, the time causality graph and the variable relationship graph are derived from the bond graph. Secondly, the fault signature matrix is obtained by combining the change relationship of fault parameters. Finally, the fault signature matrix is used as the rule of the inference engine design in the expert system for fault diagnosis. In this paper, the key equipment of the reactor coolant system is used to verify the fault diagnosis method of the bond graph expert system, and the path reasoning relationship between alarms is obtained, which can accurately obtain the deep knowledge required by the operators. A new idea for fault diagnosis in a nuclear power plant’s expert system is provided by this method.

The bond graph can be used to simulate and model conservation processes in different energy domains. In this paper, based on bond graph, the pressurizer was modeled and diagnosed at full power. Two parameters of pressure and mass flow were selected. Firstly, the bond graph of pressurizer was established. The pressurizer was divided into three zones: the fluctuating zone, the saturated liquid zone and the gaseous zone. The conservation equations of mass and energy were given respectively for these three regions. Secondly, the conservation equations of mass and energy were combined with the conservation of volume, and the model was constructed according to the combination rules. Then, according to the established bond graph model, the time causality diagram was established to determine the influence relationship between the coefficients and listed the fault characteristic matrix. Meanwhile, the reasoning structure between the coefficient was realized by the expert system G2. Finally, the fault data simulated by simulator was imported into the bond graph model. The failure of the safety valve, the spray valve and the immersion type heater were introduced to verify the modeling and diagnosis effect. It concludes that the diagnosis effect is consistent with the imported fault. Compared with the data-driven method, the bond graph method is more reliable, and compared with the expert knowledge method, the bond graph method can diagnose unknown faults and has a wide range of applications. In addition, a new method for nuclear power plant simulation modeling was provided. This was a new attempt to model bond graphs outside the research of mechanic, electronic and hydraulic.