Scientists use machine learning to boost research into new materials

The material on a powder X-ray diffraction memory plate is analyzed at the University of Rochester Laser Energy Laboratory’s Omega Laser facility. Scientists are developing deep learning models to analyze the enormous amounts of data produced by these experiments. Credit: University of Rochester Laser Energy Laboratory Photo / Danae Polsin and Gregory Ameele

Deep learning models developed by scientists are now capable of analyzing the extensive data produced by X-ray diffraction techniques.

Researchers at the University of Rochester say deep learning could turbocharge a technique that is already the gold standard for characterizing new materials. His work, published in Npj computer materialsdetails how they created models to more efficiently use the extensive data generated by X-ray diffraction experiments.

During X-ray diffraction experiments, bright lasers illuminate a sample, producing diffracted images containing important information about the material’s structure and properties. Project leader Niaz Abdolrahim, associate professor in the Department of Mechanical and Scientific Engineering at the Laser Energy Laboratory (LLE), says conventional methods of analyzing these images can be controversial, time-consuming and often ineffective.

“There is a lot of materials science and physics hidden in each of these images, and terabytes of data are produced every day in facilities and laboratories around the world,” says Abdolrahim. “Developing a good model to analyze this data can really help accelerate materials innovation, understand materials in extreme conditions, and develop materials for different technological applications.” »

Innovations in high energy density experiments

The study, led by Jelardo Salgado ’23 MS (materials science), is particularly promising for high-energy density experiments like those conducted at LLE by researchers in the Center for Matter at Atomic Pressures. By examining the precise moment when materials change phase under extreme conditions, scientists can discover ways to create new materials and learn more about the formation of stars and planets.

Abdolrahim says the project, funded by the U.S. Department of Energy’s National Nuclear Security Administration and the National Science Foundation, improves on previous development attempts. machine learning models for X-ray diffraction analysis that have been trained and evaluated primarily with synthetic data. Abdolrahim, Associate Professor Chenliang Xu in the Department of Computer Science, and their students incorporated real data from experiments with inorganic materials to train their deep learning models.

According to Abdolrahim, more experimental data from X-ray diffraction analysis needs to be publicly available to help refine the models. She says the team is working to create platforms for others to share data that can help train and evaluate the system, making it even more effective.

Leave a Comment