Development Of Reliability Analysis Framework For Passive Safety Systems In Nuclear Power Plants

The safety of nuclear power plants has long been a widespread concern for the public,while recently many applications of passive systems have improved the safety performance of reactors.The core feature of passive safety systems is that under accidental conditions it is able to maintain reactors safe by natural forces even though no power input or human interfere.For this reason passive systems are credited with higher reliability than traditional active systems.However,the reliability analysis frameworks are quite different for passive systems and tradition active systems because of their differences in working principles.This is why an accurate reliability analysis method for passive systems should be developed.This paper summarized three common frameworks for passive system reliability analysis,and developed an evaluation method based on one of the three methods.Surrogate model method is one of many useful ways to calculate the reliability of passive systems.It replaces the original long-running thermal hydraulics program by a fast-running mathematic function.In this paper a Gaussian regression model is developed,and Artificial Neural Network is employed as a comparison to it.Based on results by test functions,the accuracy and robustness of Gaussian regression model are verified.The sampling method which is combined with surrogate model,or called design of experiment,is also crucial for improving the efficiency and accuracy of reliability analysis.This work combined importance sampling,directional sampling and adaptive sampling into a comprehensive sampling method called directional adaptive importance sampling.This method initializes with the directional sampling framework and moves sampling areas to more important areas gradually.Based on some test functions,we compared the results of importance sampling,importance sampling with surrogate model,uniform sampling with surrogate model,this mixed sampling method with surrogate model,with the accurate result calculated by direct Mont Carlo simulation.Results verified the accuracy of the directional adaptive importance sampling method.In the last part of the paper we developed a RELAP5-3D model for ACP1000 Passive Containment Cooling System as a case study.We selected several uncertain parameters and then calculated its functional failure probability.Bootstrap method is employed to calculate the confidence intervals for results gained by different methods,and the final results showed that the combination of directional adaptive importance sampling method and Gaussian regression model has the narrowest confidence interval.