[2E01-04] Fuel Research with Data Science and Machine Learning, Fast Reactor Recycle

[2E01] New Developments in Nuclear Fuel Research through Integration with Data Science

(8) Harnessing Machine Learning for the Discovery of High Thermal Conductivity Fuel Candidates

*Yifan Sun¹, Masaya Kumagai¹, Mingyu Jin¹, Eriko Sato¹, Masako Aoki¹, Yuji Ohishi², Ken Kurosaki^1,3 (1. Kyoto Univ., 2. Osaka Univ., 3. Univ. of Fukui)

Keywords:Advanced nuclear fuel, Machine learning, Thermal conductivity

In the wake of the Fukushima Daiichi Nuclear Power Plant incident, the development of advanced nuclear fuels has become a priority. While promising high-density fuels like UN and U₃Si₂ have been proposed, they are still far from commercially applicable, underscoring the need to explore a broader range of uranium compounds. Our study aims to expedite this exploration process by introducing a machine learning model capable of identifying uranium compounds with high thermal conductivity. This classification model was trained on 168,916 data points to predict thermal conductivity based on a compound’s composition and temperature. The model successfully identified 119 stable uranium compounds with thermal conductivities exceeding 15 W/mK, offering a significant leap forward in the search for advanced nuclear fuels.