Enabling Rapid Chemical Analysis of Plutonium Alloys via Machine Learning-enhanced Atomic Spectroscopy Techniques

Analytical atomic spectroscopy methods have the potential to provide solutions for rapid, high fidelity chemical analysis of plutonium alloys. Implementing these methods with advanced analytical techniques can help reduce the chemical analysis time needed for plutonium pit production, directly enabling the 80 pit-per-year by 2030 manufacturing goal outlined in the 2018 Nuclear Posture Review. Two commercial, handheld elemental analyzers were validated for potential in situ analysis of Pu. A handheld XRF device was able to detect gallium in a Pu surrogate matrix with a detection limit of 0.002 wt% and a mean error of 8%. A handheld LIBS device was able to yield univariate detection limits as low as 0.1 wt% Ga with mean error of 3%. Implementing machine learning methods for spectral analysis with the handheld LIBS device reduced error to 0.27%, but the limited device resolution impedes improvements in sensitivity. A compact Echelle spectrometer was implemented with a laboratory LIBS setup to reach a detection limit of 0.006 wt% Ga when coupled with an optimized extra trees regression. A Gaussian kernel regression trained on this high resolution data set yielded the most accurate predictive model with 0.33% error. Lastly, the phenomenon of self-absorption was quantified and corrected for in Ce-Ga LIBS spectra. By implementing a Stark broadening based correction, the univariate detection limit for Ga from LIBS spectra was reduced to 0.008%. Overall, this research indicates that implementing a compact, high resolving power spectrograph for recording Pu alloy spectra and developing optimized machine learning models for spectral analysis can yield high fidelity solutions for Pu alloy chemical analysis and quality control