Speaker
Description
The high brilliance of modern X-ray and neutron sources enables temporally and spatially resolved measurements with high acquisition rates, resulting in vast amounts of data. Given the achieved experimental performance, the data analysis becomes a substantial bottleneck on the way to new discoveries. This applies, inter alia, to surface-sensitive scattering techniques, such as grazing-incidence diffraction (GID) / grazing-incidence wide-angle scattering (GIWAXS). Moreover, the analysis of 2D GID/GIWAXS data is inherently challenging due to the high data dimensionality and various parasitic contributions. To address these challenges, we develop an mlgid data analysis pipeline [1], which divides the task into different stages. After the initial data reduction, an object detection ML model detects Bragg peaks in GID patterns. After additional peak refinement, the extracted peak parameters can be used for further analysis, e.g., probabilistic ML-based matching to known crystalline structures or determination of unit cell parameters. The pipeline achieves performance comparable with the typical data acquisition rates and was also shown capable of on-the-fly analysis during in situ GID experiments. These findings pave the way for more efficient experiments, allowing the experimental plan to be adjusted based on analysis results obtained directly during the experiment. They ultimately have the potential for automated surface-scattering experiments.
We acknowledge funding via DAPHNE4NFDI (DFG grant no. 460248799) and OSCARS (Horizon Europe grant No. 101129751).
- D. Lapkin et al., (2026, in preparation)
