Development and Application of a Data-Driven Methodology for Validation of Risk-Informed Safety Margin Characterization Models

The proposed project aims to develop and demonstrate a data-driven methodology for validation of advanced computer models used in nuclear power plant safety analysis. Specifically, the advanced computer models are those in the toolkit developed to support risk-informed safety margin characterization (RISMC), an integrated deterministic/probabilistic safety analysis methodology developed in the Department of Energy’s Light Water Reactor Sustainability (LWR-S) program. The project envisions that the new validation methodology for safety analysis codes will build upon the U.S. Nuclear Regulatory Commission’s Code Scaling, Applicability and Uncertainty (CSAU) methodology, and its subsequent regulatory guide NRC 1.203 on “Transient and Accident Analysis Methods”, also known as Evaluation Model Development and Assessment Process (EMDAP). The resulting methodology, called Risk-informed EMDAP, should meet requirements of the RISMC methodology. The proposed project will bring to bear advanced methods and tools in verification and validation (V&V), sensitivity, and uncertainty analysis to facilitate the implementation of already demanding EMDAP in a risk-informed application. One major challenge in validation is a lack of relevant data, including lack of confidence in the applicability of models and their supporting data in prototypical reactor conditions. In addition to this, the computational and methodological limitations of previous eras led to a reliance on human judgment that can now be reduced. It is not that we no longer need data: rather, nowadays, it is possible to improve the use that we make of data we have (or can obtain). In particular, a new physics-guided validation strategy based on first principles will rigorously map and bound simulation errors in the domain of intended model use. To our knowledge, this represents a first-of-a-kind approach in the determination of the validation domain in the nuclear engineering community, and one that presents a significant shift from the current one based on expert-determined scale distortion uncertainties. This is what we mean by “data-driven.