Uncertainty Quantification (UQ)


Uncertainty quantification (UQ) process devises a metric to quantitatively characterize uncertainties in simulation or experiment for an engineering/physics system. A comprehensive UQ process propagates and prioritizes all major sources of uncertainties, to develop insight for the follow-up analysis of reducing uncertainties. It is an essential component of the best-estimate reactor analysis as it provides reliable metric by which the quality of the predictions can be assessed.

We have designed novel algorithms such as Physics-Guided Coverage Mapping (PCM) to extrapolate experiment-based uncertainties to real-world applications via scaling/mapping. Our research is future-focused on addressing a key challenge in the existing regulatory/licensing framework, i.e., on how to transition from heavy reliance on experiments to a science-based risk-informed validation strategy. Such a strategy is capable of leveraging advances in predictive science, information theory, and data analytics to develop transformational approaches for the evaluation of biases and uncertainties when limited experimental data exist. Addressing this challenge is critically needed to support the expected adoption of advanced reactor technologies, which require scientifically-defendable analysis capabilities for their model validation.

Relevant work

Addressing ambiguities in constrained sensitivity analysis for reactor physics problems

This work by Jeongwon Seo presents an algorithm to perform sensitivity analysis under specific constraints imposed by the model. While uncertainty quantification attempts to address the uncertainty in model parameters, sensitivity analysis studies the impact of these parameters on the model output. Together, they test the robustness of model predictions.

Theoretical development of cross-section uncertainty library for core simulations

This article by Dr. Huang condenses the vast uncertainty space of various reactor parameters using ROM techniques while also preserving the accuracies of these parameters by propagating the uncertainties in an accurate manner. This allows for less conservative estimates and more realistic assumptions to be made for regulatory purposes efficiently.