PNNL

Abstract

Borosilicate glass is the widely accepted waste form for immobilization of high and medium level nuclear wastes [1,2]. Advances in nuclear energies and new reactor designs require the development of new waste forms. For example, wastes from molten salt reactors and reprocessing of nuclear fuels lead to salt-based wastes that are difficult to be immobilized by conventional borosilicate glasses due to limited solubility and waste loading. In designing new waste forms, machine learning (ML) and artificial intelligence (AI) based approaches [3,4] are much needed and can be beneficial in enabling a more efficient design in large parameter spaces as compared to traditional Edisonian trial-and-error approaches. We report in this paper the rationale and latest progress of our ML/AI-based design of phosphate-based waste forms.