Cubic DUCG(Dynamical Uncertain Causality Graph) Recursive Algorithm And Its Implementation In Nuclear Power Plant Fault Diagnoses

Three Mile Island Accident in the United States indicates that,once a nuclear power plant(NPP)fault takes place in a complex situation,it may lead to serious and incalculable consequences if we can't locate the fault root cause,forecast the fault development,and take effective fault recovery and protection measures timely and accurately.At present,the theoretical algorithms and computer system developments for intelligent fault diagnosis of NPPs are still lacking.In-depth researches on advanced artificial intelligence theories and developments of fault diagnosis software systems for NPPs are of great significance,which can be used to assist NPP operators in monitoring parameters,in diagnosing fault root cause,and in predicting fault development.This paper carries out the above-mentioned work based on Dynamic Uncertain Causality Graph(DUCG)theory proposed by Professor Zhang Qin.DUCG is a probabilistic graphical model which can represent any uncertain causality compactly and graphically,DUCG can simplify graphics and expand events based on observed evidences,thus calculating the updated probabilities of the queries that are in concern.However,DUCG needs to do some special treatments when dealing with dynamic systems with a negative feedback,which greatly reduces the interpretability of DUCG.In order to solve the above limitation of DUCG,Professor Zhang Qin proposed Cubic-DUCG,a further development of DUCG aiming to solve more complex situations such as dynamic negative feedback.Cubic-DUCG allows connections of causal relationships crossing any number of time slices,whose main idea is to generate the cubic causal relationships graph dynamically based on the dynamically observed evidences,which can model the evolution of fault in time slices intuitively and make fault diagnoses and predictions accurrately.However,the computation amount of Cubic-DUCG will be huge if we directly expand the dynamic evidences and calculateBased on Cubic-DUCG,this paper presents a time slice recursive algorithm,which can be applied to expression expansions,simplifications and probability calculations.The algorithm can effectively solve the problem of the combination explosion in expression expansions,simplifications and probability calculations of Cubic-DUCG.On the basis of this efficient algorithm,the JAVA software for reasoning machine in NPP fault diagnosis system based on Cubic-DUCG is designed and developed.The software can carry out parameter monitoring accurately and efficiently,and assist operators of NPP to diagnose fault root cause and predict fault development.An experimental research and an efficiency comparison are carried out based on the developed software and knowledge base of NingDe NPP(24 faults in the secondary circuit),which verifies the accuracy and efficiency of the proposed time slice recursive algorithm.The proposed time slice recursive algorithm and the realized fault diagnosis reasoning machine based on Cubic-DUCG can be used for further theoretical studies of DUCG and for fault diagnosis in chemical plants,aerospace and other complex industrial systems.