Abstract The demonstration of large operative safety margins is the fundamental requirement for the transition of new generation nuclear reactors from the early concept phase to the initial industrial design one. In order to do so, it is necessary to perform a thorough characterization of the uncertainties affecting the models used for describing the physical behavior of the reactors and to quantify their impact on the safety performances of the plant, typically in terms of the probability of occurrence of deviations from the nominal, or design, operative conditions, often referred to as failure probabilities (reliability analysis). Conceptually, this problem can be solved by framing the problem within a stochastic setting and propagating the uncertainties, represented by suitable probability distributions, by resorting to sampling-based, Monte Carlo (MC) schemes. However, the high reliability of these plants and the complexity of the computer codes used to model their behavior are such that standard MC approaches would require prohibitive computational times. Further computational issues then arise due to the fact that the stochastic models adopted, i.e., the probability distributions, are, in turn, affected by uncertainties, typically related to the inference process adopted to estimate their relevant parameters, so that the estimates of the failure probabilities also become uncertain, with potentially significant impacts on any safety-based decision making process. In this context, the objective of this work is that of providing an innovative computational tool for efficiently performing Sobol-based global reliability sensitivity analysis that allows to satisfactorily quantify the impact of these additional uncertainties on the safety performances of innovative nuclear reactors concepts. The method exploits an original adaptive surrogate modeling with MC variance based techniques, in order to reduce the computational efforts while maintaining sufficient degrees of accuracy. The resulting algorithm is demonstrated with reference to a real innovative nuclear reactor concept, i.e., the Lead Bismuth Eutectic eXperimental Accelerator Driven System (LBE-XADS) reactor. The results of the analysis show the potentiality of the method, offering important insights on the robustness of the concept reactor safety design and suggesting possible paths of improvement, at acceptable computational times.
Read full abstract