Study Of Information Fusion Methods For Fault Diagnosis Of Nuclear Power Plant

Nuclear Power Plant(NPP)is a thermal system with complex structure.It consists of major equipment such as nuclear reactor,steam generator,main coolant pumps,voltage regulator,turbines,and other related functional loops.Diagnosis of NPP' s faults is related to comprehensive processing of information from various sources;therefore,information fusion attracts a lot attention in the field of NPP fault diagnosis.However,some information fusion algorithms cannot achieve effective application if used directly.Therefore,studying nuclear power plant fault diagnosis related methods and technique based on information fusion is with important theoretical significance and practical value.This dissertation includes two aspects:the fault feature extraction and the diagnosis strategy of NPP.Taking the typical faults of the primary circuit main coolant system equipment as examples,and focusing on existing problems for NPPs fault diagnosis,information fusion methods of NPP' s fault diagnosis have been studied and explored from the multiple levels and perspectives.This dissertation conducts researches to solve following issues:? how to recognize signs of main coolant pump faults accurately;? how to recognize signs of concurrent faults of main coolant pump accurately;? in different conditions,how to correctly recognize three fault modes:barrel breaking,barrel fastener cracking off and heat treatment deformity of lower core support plate and core barrel;?how to recognize faults modes that control rod driving mechanism getting stuck,drop and slip.For these problems,this dissertation conducts researches below and obtains creative results:For fault feature extraction of NPP fault diagnosis,this dissertation analyzes the relationship between fault sources and fault features,and states the properties that ideal fault features should have.Additionally,the mathematical expression of fault features and fault modes is provided,and the framework of NPP fault diagnosis is defined,the hierarchical structure based fusion diagnosis model of NPP fault information is built,and the implementation strategy of NPP fault diagnosis is proposed.For the typical fault mode recognition problems of main coolant pump,a method of the wavelet entropy of neural network is propsed.The features of main coolant pump fault signals are extracted by using spectrum analysis and wavelet entropy analysis,respectively.On this basis,wavelet entropy neural network is constructed.For the concurrent fault problems of main coolant pump,a method of DSmT on Decision Level is provided.Adopting free DSm module and mix DSm module,dynamic fusion calculation of multiple independent proof sources contain fault information is conducted,the decision mode of main coolant pump concurrent fault diagnosis is built,then verified by simulation.For the typical fault problems of core hanging basket,a method of the wavelet packet energy analysis and DSmT is propsed.By using wavelet decomposition and reconstruction method,the fault signal of core barrel is decomposed to basis function family formed from wavelet expanding and contracting,then sub-band energy distributed in different bands are obtained,and used as BPA evaluation signal of core barrel DSmT.By simulation and analysis,the effectiveness of this method is verified.For the typical fault problems of CRDM of control rods,a method of rough set neural network is put.The CRDM internet of things(IoT)framework and IoT CRDM fault diagnosis recognition system are developed.From the perceptual layer,network layer and support layer of IoT,MEMS sensor,ZigBee module and Multi-Agent rough set neutral network module are introduced,respectively.From CRDM fault recognition viewpoint,the feasibility and effectiveness of applying IoT on NPP fault diagnosis are proved,and the accuracy of the rough set neural network fusion method is verified.With lots of simulations and comparison experiments,this dissertation solves issues of fault signal detection in information fusion data layer by using spectrum analysis and wavelet entropy analysis,fault feature extraction in feature layer by using wavelet energy analysis and rough set theory method,and fault feature recognition and fault information location in decision layer by using BP neutral network and DSmT fusion method.Applying multiple information fusion methods above,this dissertation studies various fault modes of main loop reactor main coolant pump,core barrel and CRDM,including the situation of concurrent fault mode,and verifies the feasibility,effectiveness and reliability of those methods.The work of this dissertation provides theoretical foundation of applying information infusion method on NPP fault diagnosis.