AI powers modeling of safer sustainable nuclear reactors

AI powers modeling of safer sustainable nuclear reactors
by Robert Schreiber
Berlin, Germany (SPX) Jan 15, 2025

Researchers from Skoltech and the Institute of High Temperature Electrochemistry of UB RAS have developed a machine learning-based model that predicts the properties of molten salts. These substances are crucial in metallurgy and present promising solutions for the growing challenge of nuclear waste management. Published in the *Journal of Molecular Liquids*, this work provides tools to make nuclear power safer, more sustainable, and reduce costs in metal production.

Molten salts form a diverse class of compounds with numerous physical properties critical for industrial applications. Materials scientists aim to refine the composition of these mixtures to optimize processes like producing pure titanium, calcium, aluminum, and more. They also play a vital role in advancing next-generation nuclear reactor technology.

As the world seeks carbon-free energy solutions, nuclear power remains a pivotal player alongside solar and wind generation. While fusion technology holds potential but remains distant, molten-salt reactor (MSR) technology is much closer to realization. These reactors depend on molten salts with carefully tailored physical and chemical properties.

MSRs offer several advantages over traditional reactors. They operate at nearly atmospheric pressure, significantly reducing the risk of hydrogen explosions like those seen in the Fukushima disaster. This design also lowers safety and operational costs. Unlike conventional reactors, MSRs can be refueled during operation without requiring a shutdown. Operating at double the temperature of current reactors, they achieve higher power generation efficiency and provide opportunities for waste heat utilization.

Furthermore, MSRs address the challenge of accumulating nuclear waste. By using highly radioactive minor actinides such as neptunium-237 and americium-241 as fuel, they turn hazardous materials from traditional reactors into a resource for energy generation.

Understanding the properties of molten salts is critical for advancing these technologies. However, the vast array of chemical combinations and technologically relevant properties makes experimental investigation prohibitively expensive and complex. The corrosive nature of molten salts and the extreme temperatures required further complicate this task.

"Computationally guided search for melts with particular physico-chemical properties might substantially simplify and accelerate the development of next-generation nuclear reactors, since the number of real experiments will be minimized," said Nikita Rybin, lead author and research scientist at Skoltech AI's Laboratory of Artificial Intelligence for Materials Design.

"In this study, we presented and tested a methodology that allows one to calculate thermophysical properties of molten salts at finite temperatures. Our findings for the salt known as FLiNaK (comprising LiF, NaF, KF) coincide with available experimental data, prompting us to expand our work to other salt compositions and properties. This will eventually make computationally guided developments in next-generation nuclear reactors feasible."

The team employed machine-learned interatomic potentials to calculate molten salt properties. These potentials, trained on data from smaller-scale quantum mechanical models, enable the large-scale computations necessary to derive physical properties. Without machine learning, such calculations would be computationally prohibitive.

Research Report:Thermophysical properties of Molten FLiNaK: A moment tensor potential approach