A team of scientists from the National Laboratory of Ames has developed a machine learning model for searching for constant magnetic materials without critical elements. This model predicts the temperature of Curie for new material combinations, which is a crucial step in using artificial intelligence to predict new materials with constant magnets. The findings of this work have been published in the journal “Chemistry of Materials”.
High-performance magnets are essential for various technologies such as wind energy, data storage, electric cars, and magnetic cooling. However, these magnets usually contain critical materials like cobalt and rare earth elements, which are scarce. This scarcity has prompted researchers to find ways to create new magnetic materials with reduced critical material content.
To train their machine learning model, the team utilized experimental data on the temperatures of Curie. The Curie temperature is the maximum temperature at which a material retains its magnetic properties. Yaroslav Mudryk, a scientist from the Ames Laboratory, emphasized the significance of finding compounds with a high Curie temperature as the first step towards discovering materials that can maintain magnetic properties at elevated temperatures.
Mudryk further explained that traditional methods of searching for new materials rely on expensive and time-consuming experiments. However, leveraging machine learning techniques can save both time and resources.
The team trained their model using well-known experimental magnetic materials and then tested it with compounds based on cerium, zirconium, and iron. Andriy Palasyuk, another scientist from the Ames Laboratory, proposed this idea of focusing on unknown magnetic materials composed of widely available elements. Palasyuk emphasized the importance of the next supermagnet being not only high-performing but also based on accessible domestic components.
The researchers discovered that their machine learning model successfully predicted the Curie temperature of the material candidates. This success represents a significant step towards developing a rapid method for designing new permanent magnets for future technological applications. “We are creating physics based on machine learning for a sustainable future,” said Plazant Singh.