Alternative Sampling Approaches for Integrated Safety Analysis: Latin Hypercube Versus Deterministic Sampling

Abstract

Integrated safety analysis combines both deterministic and probabilistic safety analysis, which brings several advantages including improved modeling of dynamic interactions and treatment of uncertainties. Dynamic event trees provide a framework to realize integrated safety analysis for nuclear power plants and process plants. Simple random sampling based Monte Carlo simulation is used to propagate epistemic uncertainties in dynamic event trees. This setup requires simulation of accident scenarios for each set of input epistemic parameters. The computational time to perform such calculations can even challenge today’s computational infrastructure, especially for complex accident scenarios. Alternative approaches have been under investigation to overcome the computational issues. This work explores two alternative sampling approaches for dynamic event trees, namely Deterministic Sampling (DS) and Latin Hypercube (LH) sampling approaches. A chemical batch reactor problem solved with simple random sampling approach is used for comparison of the current results. The analysis of results reveals that alternative sampling methods are computationally economical as well as their results are on par with the reference. Further, strengths and weaknesses of the considered alternative sampling approaches are discussed.