| Abstract Scope |
The complex and variable composition of spent nuclear fuel (SNF) poses a challenge in designing reliable wasteforms due to (de)stabilizing cation-cation interactions (between fission products, actinides, transuranics, etc.). The mixing elements in solid-solutions can distribute randomly or present short-range features, resulting in both enthalpic and entropic effects which can impact the long-term stability of the materials under storage and repository conditions. While calorimetric methods are essential techniques used to interrogate the thermodynamic impacts of mixing, building predictive models using traditional techniques becomes exponentially more challenging as wasteforms increase in chemical complexity. In this work, we use brannerite materials (AB2O6, prototypically UTi2O6) as a model system to study how machine learning (ML) can be used in tandem with calorimetric techniques to accelerate the study of element mixing in realistic wasteform materials. These materials are chemically homogeneous and phase pure, making them ideal for calorimetric measurements, while Ce(III)-U(V) charge balancing introduces added complexity. A suite of binary and ternary (U,Ce,Th)Ti2O6 mixtures have been synthesized and rigorously characterized, and preliminary calorimetry has been performed to study select binary mixing of Ce(III)-Ce(IV) and Ce-U materials. We are currently pursuing ML techniques to predict thermodynamic properties of the more complex ternary mixtures, which we can then benchmark with selective laboratory measurements rather than mapping the entire energetic landscape. This workflow will facilitate quicker screening of wasteforms by streamlining time consuming calorimetry and enabling the rapid extraction of critical stability constants, and the targeted, rational design of wasteforms for realistic SNF. |