Researchers Create Database of AI Models to Find New Alloys for Nuclear Fusion Plants

A study led by the Department of Energy’s Oak Ridge National Laboratory details how artificial intelligence researchers created an AI model to help identify new alloys used as shielding for fusion application components in a nuclear fusion reactor. The findings mark a major step toward improving nuclear fusion facilities.

The project began several years ago under the leadership of David Womble, former director of the AI ​​Initiative. Massimiliano Lupo Pasini, an AI data scientist at ORNL, advocated for its continuation as part of the initiative’s Artificial Intelligence for Scientific Discovery (AISD) work area. The results of the study are published in the journal Scientific data.

“These alloys are necessary to achieve exceptional performance at very high temperatures, both in terms of high temperature resistance and structural mechanical properties necessary for their use in complex nuclear power plants,” said Lupo Pasini.

Traditionally, these materials were made from tungsten as the primary element, with additional elements injected in. This alloy composition proved resistant to high temperatures, but was not consistent in maintaining adequate shielding.

“Recently, the materials science community has been exploring the possibility of replacing these standard technology materials with something completely new and disruptive,” said Lupo Pasini.

Identifying potential metal combinations, however, is a major challenge given the sheer number of possibilities. Guided by AI, researchers can bypass the seemingly endless period of trial and error to more efficiently find viable candidate alloys.

Lupo Pasini teamed up with German Samolyuk, Jong Youl Choi, Markus Eisenbach, Junqi Yin and Ying Yang and generated the data to create an AI model that identified three elements to test as potential candidates for creating new alloys. Choi, Eisenbach and Yin work in ORNL’s Computer Science and Computational Sciences Directorate, while Samolyuk and Yang work in the Physical Sciences Directorate.

However, this AI-generated database represents only the first half of the project. The generated data will be used by the authors for further research dedicated to the development, training, and deployment of ML models for materials discovery and design.

“To support the design of new high-entropy refractory alloys, we need to cover six elements,” said Lupo Pasini. “In addition, since quantum mechanical calculations are very expensive to run on existing supercomputers, data alone will not be enough.”

Expensive quantum computations weren’t the only challenge the team faced when creating the foundations of its AI model, Lupo Pasini said.

“It took a fair number of hours of work on the Perlmutter and Summit supercomputers to generate the data we just published,” he said. “The data generation took over a year.”

The Perlmutter supercomputer is located at Lawrence Berkeley National Laboratory, while Summit, part of the Oak Ridge Leadership Computing Facility, is hosted at ORNL. Both computing systems are facilities used by the DOE Office of Science.

The team’s next step is to take this generated data and use it to train the AI ​​model that will accelerate the vast array of compounds that come from mixing the six elements at different concentrations in the form of alloys.

“We are trying to help materials scientists with their trial-and-error approaches to identify the relative percentage of different elements that need to be mixed in order to create alloys that can lead to disruptive technological advances in fusion,” added Lupo Pasini.