Uncertainty and sensitivity analysis in severe accidents

Core melt sequences are studied at the design stage of third generation reactors due to the high threat posed to population and environment. Phenomenological models employed in the deterministic codes for transient simulation are often based on experimental campaigns and past accidents, thus subject to uncertainty that is propagated to the code outputs. As the impact of uncertain model inputs on the evolution of system variables and key timings of a severe accident is determined, critical components can be identified and operator procedures adjusted. This task requires a large number of model runs which is often practicable only on an HPC cluster. A Python wrapper for the severe accident code MAAP, integrating the open source OpenTURNS library, is developed and presented in this work. It allows the design of an uncertainty quantification study, parallel submission of the individual simulations to an HPC cluster, result post-processing and plotting for a variety of different studies. The wrapper is then used to perform a preliminary study of a Total Loss of AC Power transient in an EPR plant. A global uncertainty and sensitivity analysis with 48 Latin Hypercube sampled input parameters is chosen. Statistical measures (mean value, variance), PDFs and CDFs provide an estimate of the output uncertainty, while the most relevant parameters are identified based on correlation and linear regression coefficients.