Description
In this PhD dissertation defense, LaGrande Gunnell presents research on improving Hanford nuclear waste vitrification algorithms through advanced computational methods. The work specifically focuses on integrating machine learning (ML), uncertainty quantification (UQ), and gradient-based optimization techniques to optimize nuclear waste glass formulations. The goal is to maximize waste loading while ensuring the glass meets necessary constraints, such as viscosity, electrical conductivity, chemical durability, and corrosion resistance.
Key contributions include:
- Integrating various ML models, including Gaussian Process Regression (GPR), neural networks, and support vector regression (SVR), into optimization frameworks.
- Developing efficient uncertainty propagation methods, which significantly reduce computational costs compared to traditional Monte Carlo simulations.
- Conducting sensitivity analyses to identify the most influential factors affecting glass property uncertainties.
This work leverages open-source scientific computing tools, setting the stage for future advancements in nuclear waste immobilization processes.