Domain‐Knowledge‑Guided Rule Extraction for Governing Nepheline Amount in High‑Level Waste Glasses

ABSTRACT

Nepheline crystallization during canister centerline cooling is one of the most constraining factors limiting waste loading in high‐level waste (HLW) borosilicate glasses. Existing predictive models treat nepheline formation as a binary outcome, yet durability degradation depends continuously on nepheline fraction. This study moves beyond binary classification by combining domain knowledge with association rule mining to extract compositional and thermal‐history rules associated with specific nepheline fraction thresholds. A physics‐based feature selection strategy reduced 70 compositional, structural, and kinetic variables to 15 features spanning the thermodynamic, charge‐compensation, modifier‐partitioning, and kinetic domains. Rules were mined from 727 quantified HLW glass compositions using three nepheline thresholds. Single‐feature and pairwise analyses reveal that the onset of nepheline formation is governed by the sodium–aluminum–boron balance and reinforced by kinetic factors, whereas the amount of nepheline, once formation has occurred, depends on the partitioning of modifier cations and the structural connectivity of the glass network. These results suggest that nepheline crystallization is governed by two mechanistically distinct processes, namely a compositionally sharp onset and a diffuse severity regime, a distinction applicable to any aluminosilicate system where nepheline amount is the target of compositional optimization.