Multi-modal characterization techniques will dramatically accelerate materials research and discovery; however, this development will result in the extraction of significant quantities of data.

We apply machine learning methods trained on simulated data to tackle the interpretation of experimental data, with a goal of achieving real-time data interpretation and experiment steering capabilities. Our team has demonstrated the use of advanced X-ray emission (XES) spectrometry for simultaneous non-resonant X-ray emission at multiple edges/elements and developed the advanced concept of simultaneous multimodal X-ray microprobe measurements (X-ray fluorescence, emission, and diffraction) for APS-U enhancement beamline 25-ID, which could achieve full structural, chemical, and compositional information in an X-ray image.

We have also developed user-friendly software that applies unsupervised machine learning to process XES data from detector signals. Using simulated data from charge transfer multiplet theory, we have also used supervised machine learning to determine charge, spin, and electronic structure properties from operando XES data of transition metal oxide battery materials. Initially funded as an LDRD, the project is currently funded by DOE as a collaboration between Argonne National Laboratory, Brookhaven National Laboratory, and Lawrence Berkeley National Laboratory. Achievements include the following:

• Developed software package to process X-ray emission spectroscopy (XES) with unsupervised machine learning
• Developed method to analyze XES data with supervised machine learning
• Developed advanced XES spectrometer for simultaneous non-resonant X-ray emission at multiple edges/elements