Speaker
Description
\noindent Los Alamos National Laboratory\
P.O. Box 1663, MS A143\
Los Alamos, New Mexico 87545\
Email: \url{arclark@lanl.gov}\
LA-UR-21-29584\ \
Nuclear data are widely used by the international community for a variety of applications, including criticality safety, reactor performance, and material safeguards. Despite the breadth of use-cases, critical benchmarks are primarily used for nuclear data validation, which are sensitive to specific energy regions and nuclides and are unable to uniquely constrain, for instance, neutron fission parameters (\ie the average fission multiplicity, fission cross section, and fission spectrum). As a consequence, general-purpose nuclear data libraries, such as \eviii~\cite{Brown2018}, may have deficiencies that, while not apparent in criticality applications, negatively impact other applications, such as non-destructive analysis of special nuclear material \cite{Miller2014, Evans2014}.
Recent work by the \underline{E}xperiments \underline{U}nderpinned by \underline{C}omputational \underline{L}earning for \underline{I}mprovements in Nuclear \underline{D}ata (EUCLID) project demonstrated \cite{Neudecker2020} that machine learning (ML) can effectively augment expert judgment in identifying discrepant nuclear data. The ML algorithm used random forests and the SHAP metric to determine which nuclear data contribute most to bias between measured and simulated \keffns -eigenvalues. Specifically, the authors were able to identify issues, both known (\eg ${}^{239}$Pu thermal and resonance parameters in \evii) and unknown (\eg ${}^{19}$F neutron inelastic scatter cross section from 0.4-0.9 MeV). EUCLID further applied \cite{Neudecker2021} the ML algorithm to studying the impact of combining critical benchmarks with the LLNL pulsed sphere measurements \cite{Wong1972}. Because the pulsed sphere time-of-flight neutron leakage spectra were sensitive to nuclear data in a different energy regime than compared to critical benchmarks, nuclear data for several nuclides could be identified as reliable (${}^{1,2}$H, ${}^{7}$Li, ${}^{9}$Be, ${}^{14}$N, ${}^{235, 238}$U, and ${}^{239}$Pu) or having potential shortcomings (${}^{6}$Li, ${}^{12}$C, ${}^{16}$O, ${}^{24-26}$Mg, ${}^{27}$Al, ${}^{48}$Ti, ${}^{56}$Fe, and ${}^{208}$Pb).
This work explores how the diverse set of integral and semi-integral responses,
\begin{enumerate}
\item Critical benchmarks,
\item pulsed spheres measurements,
\item Beta-effective measurements,
\item Reactivity coefficient measurements,
\item Reaction rate ratio measurements, and
\item Neutron multiplicity counting benchmarks,
\end{enumerate}
\noindent are differently (or similarly) sensitive to the same nuclear data and discusses how that adds to understanding shortcomings in nuclear data libraries. The focus will be on how changes between the response sensitivities can potentially inform the ML-based identification of discrepancies in plutonium fission parameters. Future work will extend this analysis to a larger nuclear data set and discuss the impact on nuclear data validation.
\section*{Acknowledgments}
Research reported in this publication was supported by the U.S. Department of Energy LDRD program at Los Alamos National Laboratory. This work was supported by the US Department of Energy through the Los Alamos National Laboratory. Los Alamos National Laboratory is operated by Triad National Security, LLC, for the National Nuclear Security Administration of the US Department of Energy under Contract No. 89233218CNA000001.