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

Real-time X-ray data analysis with DONUT accelerates materials science Scientists at the U.S. Department of Energy’s Argonne National Laboratory have introduced a groundbreaking tool called DONUT, which enables real-time analysis of X-ray data and significantly speeds up materials science research. Developed through collaboration between the Advanced Photon Source (APS) and the Center for Nanoscale Materials (CNM), both DOE Office of Science user facilities, the tool leverages deep learning to interpret complex X-ray diffraction data instantly. This innovation marks a shift toward more dynamic and responsive experimentation, allowing researchers to adapt their strategies in real time. DONUT, short for Diffraction with Optics for Nanobeam by Unsupervised Training, is a physics-aware neural network designed to understand how X-ray beams interact with materials. By integrating fundamental principles of physics into its architecture, the tool enhances the accuracy of predictions regarding material structures. The system was developed and tested using data from the Hard X-ray Nanoprobe beamline, which combines resources from the APS and CNM. Published in the journal npj Computational Materials, the study highlights the potential of DONUT to revolutionize how scientists approach materials characterization. Traditional methods of analyzing X-ray diffraction microscopy data, particularly scanning X-ray nanodiffraction microscopy (SXDM), were known for being slow and labor-intensive. These techniques often required weeks or even months to produce results, limiting the ability of researchers to iterate rapidly or explore new hypotheses. DONUT addresses these challenges by processing data in real time, offering results that can be hundreds of times faster than conventional approaches. This capability empowers scientists to refine their experiments dynamically, enhancing efficiency and depth of insight. SXDM is a sophisticated technique that maps the crystal structure of materials by directing a focused X-ray beam over a sample. This method provides critical information about how materials function in applications ranging from battery technology to catalysis and electronics. However, the complexity of SXDM-generated data, characterized by multiple layers and dimensions, has historically made interpretation difficult. Scientists previously relied on manual comparison of experimental data against simulated patterns, a process that was not only time-consuming but also susceptible to human error. DONUT introduces a novel approach by merging artificial intelligence with embedded physics models. Unlike traditional AI systems that require labeled training data, DONUT learns directly from experimental inputs, eliminating the need for pre-labeled datasets. This self-training mechanism makes the tool highly adaptable, enabling users to tailor predictions according to specific experimental goals. According to Mathew Cherukara, a computational scientist at Argonne, the flexibility of DONUT allows for customized analyses, effectively providing a unique “recipe” for each scientific inquiry. The real-time capabilities of DONUT open new possibilities for autonomous experimentation. Researchers at the APS can now design and execute self-driving experiments, where decision-making is guided by immediate feedback from ongoing data collection. This feature is particularly valuable for studies involving real-world material behavior, where rapid adaptation is essential. Aileen Luo, an assistant computational scientist at Argonne and Cornell University, emphasized that the immediacy of results transforms the experimental process, enabling quicker responses and more informed choices. As DONUT continues to be integrated into routine operations at the APS and CNM, its impact on materials science research is expected to grow. The tool not only streamlines data analysis but also lowers entry barriers for new users, fostering broader participation in cutting-edge scientific exploration. With further refinement and application, DONUT stands poised to redefine the pace and scope of materials discovery in the years ahead.