(621a) Applying Machine Learning and Quantum Chemical Simulations to Actinide Process Modeling

With the kind cooperation of other US Department of Energy (DOE) Laboratories, Lawrence Livermore National Laboratory (LLNL) has been tasked to create predictive chemical equilibrium-based process models that simulate plutonium processing from recovery to desired products. To overcome the “gaps” in literature properties and solutions data, our project combines Quantum Chemical (QC) simulation predictions (e.g., from Gaussian16, VASP) with Machine Learning (ML) techniques to estimate “missing” compounds’ thermochemical properties (due to plutonium’s complex chemical behavior), and will leverage ML tools again to develop unavailable non-ideal mixtures properties to improve our models’ accuracy. This presentation will discuss our multi-faceted model-building approach, ML training data collection strategy, and our current model development efforts.