Uncertainty Quantification Framework for Design Optimization

Difficulty to obtain neutron sources of interest have driven the need for optimization
techniques to tailor a neutron generator as a replacement. A proposed solution uses
off-the-shelf neutron sources coupled with an energy-tuning assembly to mimic the
source of interest (i.e. AmLi, AmBe, thermonuclear fission spectra, etc.). These
energy-tuning assemblies have been designed with complex optimization algorithms
coupled with Monte Carlo simulations. These new system surrogate designs often
do not have an experimental counterpart for validation and comparison, and lack
non-statistical uncertainties. This work aims to improve confidence in the predictions
by providing a tool for fast uncertainty quantification to use with transport tools,
necessary for future validations. The tool, TOFFEE, has been developed to use the
sensitivity coefficients and covariance data along with the sandwich rule to assign
variance in cross-section data and subsequentially to reaction rates, neutron flux, and
k-eff.
TOFFEE is a Python framework that uses MCNP6.3 to calculate sensitivity
coefficients (generated by KSEN, PERT, etc.) and NJOY to generate a covariance
library. The tool generates new input files and calculates the cross-section uncertainty
with the sandwich rule. To test the tool, TOFFEE is used to generate cross-section
uncertainty on three models: two benchmarks, Jezebel, BeRP ball, and a newly
generated energy-tuning assembly designed with an advanced optimization algorithm.
The results presented in the example application are the uncertainty on kef f for
Jezebel, the uncertainty on kef f and the volumetric flux for the BeRP ball. The examples are then verified by comparing them to the stochastic sampling method for calculating uncertainty, used in the SAMPLER routine by SCALE. Lastly, the
uncertainty in the energy-dependent surface flux leaving the energy-tuning assembly is
calculated. These three examples give confidence that the tool can be used standalone
and in the optimization process of energy-tuning assemblies for source replacement.