NSF backs Texas A&M tool that speeds materials discovery
A new National Science Foundation award will fund free, open software that uses machine learning to answer a question every materials scientist asks.
Researchers are using open-source machine learning tools to predict material properties and guide the discovery of new materials with unprecedented speed.
A Texas A&M University materials science and engineering professor has received a National Science Foundation (NSF)award to build free, open software that can compress materials calculations once measured in years into a single day.
The project — PhaseForge — is led by Dr. Raymundo Arróyave, professor in the Department of Materials Science and Engineering. It applies machine learning to a decades-old method called CALPHAD — short for calculation of phase diagrams — that tells researchers whether a material they want to build can exist at all.
“In materials science, the first thing you need to know is whether a material you want to make indeed can be made,” Arróyave said. “The answer to that question is given by thermodynamics.”
The catch is that building those thermodynamic models takes significant time and deep expertise. Most material systems simply don’t have one. Without that foundation, hardly anything about a material can be simulated, leaving researchers to work by experiment alone.
Arróyave compares it to finding your way through a house in total darkness with a tiny flashlight.
The payoff arrived faster than he expected. When his students and a postdoctoral researcher first showed him that the method worked, he said he had trouble sleeping that night.
“When I was a graduate student, such an analysis would have taken an entire Ph.D.,” Arróyave said. “Now it could be done in one day.”
The tool targets inorganic materials with immediate practical stakes: thermal protection systems for spacecraft, shape memory alloys and ultra-hard coatings that shield other materials from wear.
PhaseForge is being distributed free and openly online, and anyone can contribute to it. The code is available on GitHub. Tutorials and training materials are planned.
Arróyave credits the project’s origins to a hunch and a group culture that rewards experiments.
“We always like to try stuff and to work together,” he said.
Everyone in the group, himself included, contributes to projects on the side.
“PhaseForge shows how machine learning can help metallurgists answer fundamental questions much faster,” said Dr. Ibrahim Karaman, head of the Department of Materials Science and Engineering. “By making the tool free and openly available, Dr. Arróyave will help accelerate materials discovery across academia, national laboratories and industry.”
The award will support one doctoral student and one postdoctoral researcher.
Arróyave is collaborating with Karaman, associate department head and professor Dr. Miladin Radovic, Brandon Bocklund of Lawrence Livermore National Laboratory, and industry partners.
