Research On Hybrid Intelligence Fault Diagnosis Method For Nuclear Power Plants

As the nuclear power plants have a potential radioactive risk,the safety has always been an important issue for nuclear energy development.The research about condition monitoring,fault diagnosis and fault prognosis have been conducted to ensure the nuclear power plants operating safely and reduce the operators' misjudgment of thesystem status;once the plantsare deviated from the normal operation,the appropriate operating recommendations could been given,which may release the mental pressure for the operators,diminish the damage to the plants,thereby the service life of the equipment is prolonged and the safety performance of the plants is improved.Based on a review of the relevant research on fault diagnosis methods at home and abroad,the key technologies of anomaly detection,fault type identification,fault severity evaluation are studied in this paper,and a hybrid intelligence fault diagnosis systemfor nuclear power plants is designed and developed to assist the operators to ascertain the operating status of the plants and provide a reference for the further measures.The main works of this paper include:(1)The modeling process of principal component analysis(PCA)and the fault detection technology based on PCA model are studied,which realized timely detection of anomaly in nuclear power plants system.(2)The fault diagnosis method based on signed directed graph(SDG)model is studied,and a SDG model of the reactor coolant system is established.Since the fault is detected with principal component model,the state of related parameter nodes in SDG model is definite,the cause of system anomaly is obtained through reverse inference,and namely,the fault type is confirmed.(3)The fault severity evaluation method of Elman neural network is studied,and the mapping relationship between the fault failure and the variation of related parameters is established by Elman neural network to approximately evaluate the fault failure of some typical ruptured-type fault in nuclear power plants.(4)Taking the nuclear reactor coolant system as a specific object,a fault diagnosis system for nuclear power plants is designed and developed,which integrates PCA model based fault detection,SDG model based fault diagnosis and Elman neural network based severity evaluation.The relevant fault simulated by the PCTRAN software is used to test the response of the developed diagnosis system,and the various designed functions of the system are testified.The examples show that the proposed method is able todetect the anomaly of plants system in time and make a correct judgment on the fault type,and approximately evaluate the fault failure of some ruptured-type fault;in addition,the effectiveness of the various functions in the diagnostic system is certified.The research of this paper may lay a theory foundation for the further engineering application of the diagnosis system.