Real-time X-ray data analysis with DONUT accelerates materials science

Real-time analysis of X-ray data is revolutionizing materials science research at the U.S. Department of Energy's Argonne National Laboratory. The DONUT tool, which stands for Diffraction with Optics for Nanobeam by Unsupervised Training, is a physics-aware neural network that delivers results almost instantly, allowing scientists to adapt experiments on the fly and uncover deeper insights into advanced materials.

Traditionally, analyzing X-ray data from techniques like scanning X-ray nanodiffraction microscopy (SXDM) has been a slow and laborious process, often taking weeks. DONUT changes this by using artificial intelligence to learn directly from experimental data, eliminating the need for labeled training examples. This physics-aware approach enables the system to quickly interpret complex X-ray images, revealing the internal structure of materials at the nanoscale.

Aileen Luo, an assistant computational scientist at Argonne, explains that DONUT allows researchers to see what's happening inside materials as the experiment unfolds, providing results hundreds of times faster than traditional methods. This real-time feedback enables more productive experiments and opens up new opportunities for discovery.

DONUT's flexibility and customizability make it adaptable to various scientific questions. Researchers can train the system on their data and adjust what they want it to predict during the experiment, like having a fresh recipe for every new research question. This capability is particularly valuable as the upgraded APS delivers brighter X-ray beams and collects data at higher speeds.

The tool also lowers the barrier for new users, including graduate students and visiting scientists, by eliminating the need for expert-labeled datasets. Traditionally, preparing labeled data required significant time and specialized knowledge, slowing down research and limiting participation to those with advanced training. DONUT removes this hurdle, allowing more researchers to take advantage of high-speed data collection.

Looking ahead, the team is working on expanding DONUT's impact to experimental automation, exploring new flavors of DONUT for autonomous microscopy and other advanced imaging techniques. This physics-aware training framework could support major initiatives like the DOE's Genesis Mission, which aims to double scientific productivity and accelerate innovation through AI. DONUT is poised to play a crucial role in tackling new scientific questions and making the most of next-generation research facilities.