Machine learning surrogates for surface complexation model of uranium sorption to oxides.

Area of Science:

• Geochemistry and Environmental Science
• Nuclear Waste Management
• Computational Modeling

Background:

• Geological disposal of spent nuclear fuel necessitates understanding radionuclide mobility, particularly uranium, which is highly mobile in its oxidized state (U(VI)).
• Uranium sorption to surrounding mineral surfaces is a key process limiting its migration, but traditional surface complexation models (SCMs) face numerical convergence challenges.
• Accurate modeling of uranium mobility is crucial for the safety assessment of nuclear waste repositories.

Purpose of the Study:

• To develop and evaluate machine learning (ML) surrogates for complex surface complexation models (SCMs) of uranium sorption.
• To address the computational cost and convergence issues associated with traditional SCM solvers.
• To provide ultrafast AI/ML tools for enhancing nuclear waste migration models.

Main Methods:

• Explored two ML surrogates: Random Forest Regressor and Deep Neural Networks (DNNs).
• Trained and validated ML models using predictions from the 2-pK Triple Layer Model for uranium retention by oxide surfaces.
• Integrated the developed ML surrogate model into a larger-scale contaminant migration model.

Main Results:

• Both ML surrogates accurately reproduced SCM predictions for uranium sorption.
• Deep Neural Networks (DNNs) demonstrated particular efficacy, achieving high accuracy at a significantly reduced computational cost.
• The ML surrogates successfully avoided the convergence problems inherent in numerical SCM solvers.

Conclusions:

• AI/ML surrogates offer a computationally efficient and reliable alternative to traditional SCMs for modeling uranium sorption.
• These ultrafast ML surrogates can be readily integrated into larger contaminant migration models for improved repository safety assessments.
• The study presents a novel ML-based approach to enhance the understanding and prediction of radionuclide migration in geological repositories.