Ames Lab scientists have developed an AI workflow to help discover magnets without rare earths (an area in which AI has already demonstrated it can excel). The pipeline uses AI models trained on real-world physics and electron behavior rather than existing data to make educated guesses about the specialized materials needed to build permanent magnets, which must be able to maintain magnetization even under extreme conditions like high temperatures.
Ames National Laboratory is a national laboratory of the U.S. Department of Energy (DOE) and the project is part of DOE’s Genesis Mission. The program’s official website describes it as an initiative that leverages government resources alongside academia with the goal of creating AI resources aimed at “breakthroughs in energy dominance, scientific discovery, and national security.”
Permanent magnets are ideal for this mission because they are frequently used in defense applications, ranging from radar systems and fighter jets like the F-35 Lightning II to submarines and drones. Despite a huge deal with Apple to sell the company’s rare earth magnets, the United States currently has only one rare earth mine (in Mountain Pass, California) and exports more than 95% of mined minerals to Asia for refining, meaning a rare earth-free solution could simultaneously address both safety and cost concerns.
Defining AI breakthrough
Ames Lab’s breakthrough is based on an existing AI model, called DuctGPT. DuctGPT was originally designed to help find materials that could survive inside fusion power plants (like the rare earth superconducting magnets developed by MIT). This means materials that can withstand considerable heat, radiation, and mechanical stress, but are nevertheless ductile enough (able to be stretched and formed without losing toughness) to be made into workable parts. The main advancement of DuctGPT is that it incorporates physics-based modeling, instead of just being trained on old data.
This essentially means that instead of just looking for patterns in previously collected data, AI understands the underlying science and can use it to invent new materials. Rather than guessing from a limited sample, the AI has the rules of the game and can search for new materials rather than modifying known ones.
Models can also take into account logistical considerations, such as the cost of producing these materials or the difficulty of sourcing basic components. The idea is to ensure that AI doesn’t end up suggesting replacement materials that are as hard to find as the rare earths they are intended to supplant.
New technology to solve old problems
The search for a rare earth-free replacement for permanent magnet materials is not a new endeavor for Ames. In April last year, the lab issued a press release about a rare earth-free magnet it had developed by combining bismuth and manganese. It was developed specifically for use in permanent magnet motors, which require magnets capable of retaining their magnetization despite extreme temperatures or other magnetic interference. The Ames scientists were able to develop a process by which they coated the crystals of the magnetic material with a polymer that prevented them from coming into contact with each other, which could cause a cascading loss of magnetization.
The AI initiative aims to accelerate the discovery of materials such as manganese/bismuth composite while ensuring their commercial viability. Reducing reliance on rare earth elements could have implications far beyond defense, such as supply chain flexibility for sectors such as renewable energy, transportation and consumer electronics. Materials developed using DuctGPT’s physics-based models could also expand the range of materials available to engineers designing next-generation technologies.
