Research On Fault Diagnosis Method Based On Expert System For Nuclear Power Plants

Nuclear power plant is a complex system with large significantly interacting parameters.To ensure operational safety,the condition monitoring for key equipments and systems is needed.Once a fault occurs,in order to take timely measures to mitigate the effect of faults and ensure safety operation,the operators need to analysis the alarms and identify the root causes.However,nuclear power plant is a plurality of systems to complete the assigned functions,event if a single equipment failure,the effect of fault will be passed between systems and internal systems,so the alarm flow rising up.Although the operators have been trained,they cannot sort out the the relationship between the alarms without the assistance.Meanwhile,they cannot identify the originating alarms and secondary alarms,finding the fundamental root cause is vary hard.The operators are often subjected to psychological pressure and workload,easily lead to human error and endanger the safety operation.Therefore,design and develop the fault diagnosis system including condition monitoring,alarm analysis,fault diagnosis to assist the operators has become the hotspot issue in nuclear safety field.In this dissertation,the domestic and overseas research status for fault diagnosis is analyzed.Considering the technological requirement and the operators' cognitive characteristics,the fault diagnosis system is designed using expert system method,including the data collection,condition monitoring,alarm analysis and fault diagnosis to assist the operators to make strategy.The main work is as follows:(1)Utilizing the combination of multilevel flow model and domain map to design the knowledge base of expert system.Firstly,to describe the complex system from the aspect of cognitive science,the multilevel flow model is utilized.Secondly,the multilevel flow model further integrated with domain map,and the reasoning rules are embodied in the instances.In this way,it performs concisely knowledge representation and revised knowledge base easily.(2)Event propagation,event logic and reasoning rules have influence on the reasoning machine of expert system.In this dissertation,the reasoning module is designed.The domain map serves as an input of the reasoning machine.The simulation shows that the diagnostic result is in accord with practical situation.(3)Assuming that the data from the sensor is real and effective,the uncertainty of the data due to sensor failure is ignored.Within this context,the observation points' protection technology is proposed to resolve the loop modeling and mutually exclusive events in the generic fault model.The simulation shows that the statuses of observation points are protected during reasoning.(4)The fault diagnosis expert system is developed including data collection,condition monitoring,alarm analysis and fault diagnosis.Independent and interrelated two parts:process surveillance unit and fault diagnosis unit,they are connected by triggering rules.In this way,it is easier for developers to revise the individual module.The simulation shows that the diagnostic result is in accord with practical situation.In addition,the minimum global solution is designed.The simulation shows that the calculation result is in accord with reasoning machine,this approach provides a convenient way to find the root cause.Finally,the fault diagnosis expert system is tested.The simulation shows that even if a part of events is triggered,the simulation shows that the fault diagnosis expert system can also identify the root cause.In summary,the main innovation points of this research can be summarized as follows:(1)A knowledge base design method for expert system based on multilevel flow model is proposed.To facilitate knowledge acquisition,knowledge representation,knowledge base modification and expansion.(2)Real-time alarm analysis and fault diagnosis are realized by using expert system.A fault diagnosis method based on the correct condition of the sensor is proposed,and the fault diagnosis range is reduced.(3)The minimum cut set algorithm is used to identify the minimum global solution of the system abnormal state.