December 1 - 6, 2024
Boston, Massachusetts
Symposium Supporters
2024 MRS Fall Meeting & Exhibit
EN12.06.04

Machine Learning the Chemistry, Structure and Properties of Molten Salt Across the Fuel Cycle

When and Where

Dec 3, 2024
4:00pm - 4:30pm
Hynes, Level 3, Room 305

Presenter(s)

Co-Author(s)

Stephen Lam1

University of Massachusetts Lowell1

Abstract

Stephen Lam1

University of Massachusetts Lowell1
A central challenge to deploying molten salt nuclear technologies lies in our ability to accurately characterize, predict, and monitor the chemistry of molten salts throughout the fuel cycle. In synthesis, dissolved species can exist in multiple oxidation states, whose ratio must be adjusted to control purification processes, volatilization and corrosivity. Similarly, during operation, the fuel salt composition evolves continuously with generation of numerous fission products, which produces significant changes in the thermophysical and thermochemical properties of the melt. Finally, in back-end fuel recycling processes, lanthanides and impurities must be separated from the salt while retaining fissionable actinides. Each of these steps requires the study of an enormous array of evolving chemical and thermodynamic conditions in the solution phase. Here, current experimental and computational approaches are not sufficiently accurate and expeditious for assessing the large design spaces demanded across the fuel cycle. As such, it is unlikely that we will achieve the robust chemical understanding required for commercial deployment and regulatory approval under conventional research paradigms employed in the study of molten salts. This talk will discuss our latest advances in applying artificial intelligence to overcome these challenges for studying the chemistry-structure-property relationships in molten salt, which include 1) machine learning-assisted atomistic simulation for speed and accuracy, 2) chemistry-informed machine learning for interpolating thermophysical and thermochemical properties across the periodic table, and 3) machine learning-enhanced characterization and online monitoring with spectroscopic methods. We will show how state-of-the-art methods have been applied for uncovering structure-property of molten salts with unprecedented speed and resolution and discuss future opportunities for improvement in each of these areas.

Symposium Organizers

Dan Gregg, ANSTO Synroc
Philip Kegler, Forschungszentrum Juelich
Josef Matyas, Pacific Northwest National Laboratory
Tomofumi Sakuragi, RWMC

Session Chairs

Pavel Ferkl
Stephen Lam

In this Session