Simulation Based Workflows for Reliable Safety Assessment for Radioactive Waste Disposals

Summary The performance of nuclear waste disposal facilities is increasingly being assessed based on failure probabilities. Traditionally, Monte Carlo Samplings (MCS) are used to quantify failure probabilities with the known drawback that very large sampling sizes become necessary to achieve reliable rare event probability estimates, which lead to prohibitive CPU requirements especially when the probabilistic performance assessment needs to be repeated frequently and the underlaying high-fidelity numerical simulations are time consuming. An approach to significantly reduce the sampling size of high-fidelity simulation, originally developed for certification process of autonomous driving systems, was adopted to Nagra’s current probabilistic performance assessment approach for the Swiss radioactive waste repository. By using a combination of inflated distribution functions, metamodeling and advanced reliability analysis on metamodels the probability of failure of the current Nagra design can be reliable estimated to about 4e-8. The overall number of high-fidelity simulations used in the workflow is 1000 samplings only. Having efficient automated workflows in place to reliably evaluate rare event probabilities opens the space to install automated processes to compare multiple sites and multiple designs to achieve targeted rare event based safety level.