The large volume of waste, along with its complex composition and variable chemistry, make finding the optimal recipes for creating glass that meets performance and regulatory standards a challenge.
Two-part project: The work is part of two-part project funded by the DOE Office of Environmental Management’s Hanford Field Office in partnership with glass scientist Albert Kruger. Results from the first part, which were published in May 2024 in the Journal of Non-Crystalline Solids, focused on the development of the glass formulation models.
The second part, published in the April 15 edition of the Journal of Non-Crystalline Solids, provides experimental validation of ML-based glass property models developed during the project’s first phase.
According to the paper authors, “The updated models and formulations showed increased waste loading while reducing the failure rate, demonstrating improved predictive accuracy, reduced uncertainties, and the effectiveness of active learning in guiding high-dimensional, nonlinear LAW glass design.”
The practical benefits of the models, the authors wrote, include “higher waste loading, shorter mission duration, and lower operational risk.”
Quote: Glass scientist Albert Kruger told Nuclear Newswire, “This effort will allow meaningful economies to expanding the validated glass formulation domain. When the WTP mission kicked off, the contractor adopted an active design to generate line rules to meet the minimum contractual targets for waste loading. That concept was inconsistent with the operational mission but was adequate for starting the plant.
“With the variability of wastes to be treated and lack of maturity for detailed chemical analyses of the inventory, it was evident that a better and more robust process control model would be beneficial. That epiphany opened the door for developing glass formulation activities that accounted for the development of forward process control—understanding the contribution to critical properties from each of the constituents of the waste and the dry materials used for making good glass within the design constraints of the facilities being built.
“The current success in the application of AI should allow for the further expansion of flexibility to address the waste variability, while supplanting the older method of formulating a sufficient quantity of glasses, melting, and testing to add the data to our knowledge base.”
