PNNL Scientists Leverage AI to Optimize Glass Formulas for Liquid Radioactive Waste

Why It Matters

Higher waste loading directly cuts disposal volume and operational costs, accelerating the decades‑long Hanford cleanup and demonstrating AI’s tangible value for nuclear waste management.

Key Takeaways

• AI model lifts waste loading ~1% per 20% increase, cutting logs
• Higher waste loading trims disposal containers, lowering overall project footprint
• Active learning approach experimentally validated for first time in glass design
• PNNL algorithm, used since 2012, now cuts glass logs by 5%
• DOE Genesis Mission leverages AI to speed nuclear waste cleanup nationwide

Pulse Analysis

Vitrification has been the cornerstone of the United States’ strategy to immobilize high‑level radioactive waste, but the chemistry of Hanford’s legacy tanks—containing every element on the periodic table—makes formula optimization a daunting, data‑intensive task. Traditional trial‑and‑error methods rely on static equations that cannot efficiently explore the combinatorial space of glass formers and waste constituents. By integrating active learning, PNNL’s AI platform rapidly iterates through thousands of candidate recipes, learning from each experiment to home in on formulations that maximize waste loading while preserving glass durability.

The breakthrough lies in the model’s ability to predict a modest yet cumulative gain: roughly a 1% increase in waste incorporation for every 20% baseline loading, translating into an estimated 5% reduction in the total number of glass logs required over the life of the Hanford vitrification project. Fewer logs mean smaller storage footprints, lower transportation and handling expenses, and a shortened mission timeline. Crucially, the AI‑derived recipes have been validated in laboratory melters, confirming that the higher waste fractions do not compromise melt viscosity, corrosion resistance, or long‑term stability—key performance metrics for regulatory compliance.

Beyond Hanford, the success of this active‑learning framework signals a broader shift in nuclear waste management. The DOE’s Genesis Mission, which earmarks AI as a catalyst for accelerating cleanup across the nation’s complex sites, now has a proven use case to scale. As more laboratories adopt similar data‑driven approaches, the industry can expect faster decommissioning schedules, reduced fiscal burdens, and a clearer pathway toward meeting stringent environmental standards. The PNNL study thus not only advances glass science but also showcases how AI can turn decades‑old challenges into actionable, cost‑effective solutions.