Data-driven surrogate model to predict isotopic composition using dynamic mode decomposition

Reduced-Order Modeling (ROM) has become an indispensable tool for reducing the cost of repetitive executions common to Sensitivity Analysis (S A), Uncertainty Characterization (UC). Presented here is the application of Dynamic Mode Decomposition (DMD) to build a data-driven, reduced-complexity surrogate that predicts the fuel concentration of a single TRIGA fuel element over time. The ultimate goal is to produce a surrogate for rapid determination of the composition of KSU TRIGA MARK II research reactor at any time within its forty years of operation. Such a surrogate would enable forward or inverse propagation of uncertainties from the initial fuel loadings and recent measurements, and enhance the current core model and reduce these uncertainties. The methodology was applied first to single fuel elements to assess the reliability of the predictions when both testing and training data come from the same configuration.The resulting surrogate was then tested with different initial conditions from a different fuel element. The tests verify that even with some perturbations of the initial conditions the DMD surrogate is able to predict the concentrations of the isotopes of interest. These single elements tests represent a first step towards modeling the whole core and propagating uncertainties in the fuel composition over the 40 year period.