Research On Probabilistic Safety Assessment Methodology For System Dynamic And Passiveness

Nuclear power is currently the only sustainable new-clean energy to replace the fossil energy, and nuclear safety is the lifeline of the development of nuclear energy. Probabilistic safety assessment (PSA) is an important approach to carry on the safety analysis of nuclear power plants (NPPs), which has been widely used in the stages of design, operation and maintenance in nuclear industry.The traditional PSA methods (based on Fault Tree/event tree) have some limitations to solve the problems of "dynamic" and "passiveness":the "dynamic" problems are transferred into the static ones in the traditional PSA models, which causes their inherent deficiencies in dealing with "dynamic" problems; the traditional PSA method cannot be used to assess the functional failure of the passive system.Some dynamic PSA methods have been developed to solve the "dynamic" problems during the last several decades with that a problem of computational cost arises. The problem of computational cost also arises in the most frequently used approach, Monte Carlo (MC) simulation, for the reliability analysis of the functional failure of the passive system. In order to solve the problems in the PSA analysis, the following researches have been carried out in this dissertation:(1) Study on the features of dynamic PSA methods. The object methods include MC simulation, Dynamic Event Tree (DET) and Monte Carlo Dynamic Event Tree (MCDET). The three object methods are benchmarked with two systems. The "dynamics", such as aging, state dependence between component and system, reparability, ordering and timing of the failures, have been considered in the models. The results show that the computational cost of DET model increases exponentially as the mission time increases, while the computational costs of MC and MCDET models increase linearly as the mission time increases. Thus, DET is only applicable for the simulation with a short mission time, and MC simulation and MCDET are more powerful for long mission time.(2) Propose a biasing transition rate method based on the direct MC simulation. A visual component is added to the biased component in series, which increases the transition rate of the integration. And the probability of the occurrence of the rare unexpected event increases. The efficiency to capture the evidence of the rare event and the performance of the method are improved by the biasing transition rate method. In the application of the benchmark system, the efficiency to capture the evidence of the rare event has been improved considerably.(3) Propose a coupling method of the MC simulation and the MCDET method. In this paper, the above mentioned biasing transition rate method is coupled with the MCDET method. In the application, some selected components who rank high in the structural analysis and have small probabilities are biased. In MCDET model, the direct MC samplings are replaced by the biasing transition rate samplings for the biased components. The coupled method is applied to the reliability analysis of ESPS system. In this study, the efficiency to capture the evidence of the rare event of the biased MCDET model is much better than that of the MCDET model.(4) Propose two PSA methods for constructing response surface for the functional failure of passive systems, namely, the advanced response surface fitting method and a coupled method based on artificial neural network. They can fit proper mathematical models to replace to deterministic model. The two proposed methods are used to analyze the functional failure of the passive decay heat removal system of CLEAR-I. And the results show the system has high reliability.The study on the features of the dynamic PSA methods can be the basis for the selection of the methods in the engineering application; the efficiency of the dynamic PSA methods based on MC simulation has been improved considerably by the biasing transition rate method, which has great significance for the application of the dynamic PSA methods; the proposed response-face-constructed methods can effectively fit accurate mathematic models with small samples, which means the proposed methods can carry on the PSA analysis of the functional failure of the passive system with less computational cost.