Machine Learning Prediction of Nitric Acid Extraction Behavior in PUREX Process

Area of Science:

• Nuclear Chemistry
• Chemical Engineering
• Data Science

Background:

• The Plutonium Uranium Reduction Extraction (PUREX) process is vital for recovering plutonium and uranium from spent nuclear fuel.
• Predicting extraction parameters in PUREX is challenging due to complex aqueous feed compositions.
• Understanding nitric acid behavior is crucial as it influences actinide extraction in the PUREX process.

Purpose of the Study:

• To develop predictive models for nitric acid distribution in the PUREX system.
• To utilize machine learning to overcome limitations of traditional thermodynamic models.
• To provide an in silico tool for optimizing PUREX process design and experiments.

Main Methods:

• Compiled extensive solvent extraction literature data on nitric acid behavior.
• Developed machine learning models to predict organic phase nitric acid concentration.
• Investigated the influence of initial acid and tributyl phosphate concentrations and diluents.

Main Results:

• Machine learning models demonstrated high accuracy in predicting nitric acid equilibrium concentration.
• The models successfully correlated acid and TBP concentrations with nitric acid extraction across various diluents.
• Generated ML-aided response surfaces provide a novel predictive capability.

Conclusions:

• Machine learning offers a powerful approach for modeling complex chemical processes like PUREX.
• The developed ML models serve as a valuable in silico tool for PUREX process optimization.
• This work represents significant progress in predicting and optimizing actinide separation in nuclear fuel reprocessing.