Machine Learning Enabled Models to Predict Sulfur Solubility in Nuclear Waste Glasses.

Area of Science:

• Materials Science
• Nuclear Engineering
• Artificial Intelligence

Background:

• The Hanford site is considering a direct feed approach for vitrifying low-activity waste (LAW) and high-level waste (HLW).
• This approach involves processing nuclear waste with higher sulfate concentrations than previously anticipated.
• Existing empirical models struggle to predict sulfate solubility and design glass formulations for these conditions, especially for high-level waste compositions outside their calibration range.

Purpose of the Study:

• To address the limitations of existing models in predicting sulfate solubility in nuclear waste glasses.
• To leverage artificial intelligence (machine learning, ML) to develop accurate predictive models for sulfate (SO3) solubility.
• To identify key compositional and processing variables influencing SO3 solubility and develop improved analytical models for glass formulation.

Main Methods:

• Trained three machine learning (ML) models using a comprehensive database of over 1000 low-activity waste (LAW) and high-level waste (HLW) glasses.
• Assessed and ranked the influence of glass compositional and processing variables on SO3 solubility using the best-performing ML model.
• Developed two closed-form analytical models based on the understanding of influential variables, with varying complexity.

Main Results:

• The developed analytical models accurately predict SO3 solubility in LAW and HLW glasses, achieving accuracy comparable to ML models.
• These new analytical models significantly outperform existing state-of-the-art models in predicting sulfate solubility.
• The study identified critical variables affecting SO3 solubility, providing a data-driven foundation for future glass design.

Conclusions:

• This research presents a data-informed, AI-guided approach for designing nuclear waste glasses with enhanced sulfur loadings.
• The developed models offer a reliable method for predicting sulfate solubility, crucial for optimizing vitrification processes at the Hanford site.
• The findings pave the way for creating advanced nuclear waste glass formulations with unprecedented sulfur-holding capacities.