Interpreting 2D X-ray Diffraction Patterns of Dynamic Compression Experiments using Machine Learning, Simulation and Constraint Algorithms

Recent developments in dynamic compression experiments are allowing collection of in-situ 2D x-ray diffraction (2D-XRD) patterns which yield important structure and structural orientation information. This data is crucial to determining and confirming deformation and transition mechanisms. However, interpretation of dynamic 2D-XRD patterns is extremely challenging. The experiments generally yield small datasets with low signal-to-noise ratio and often feature non-ideal x-ray sources and setup geometries. We describe two computational methods for simulating 2D-XRD patterns (using the LAMMPS MD code, and DENNIS application, respectively). Machine learning methods based on these simulations are used to (1) preprocess data from experiments and simulations, and (2) to automate identification/separation of component crystal structures and orientations from mixed state compression experiments. We find that implementing physics-based geometric constraints significantly reduces computational expense for both powder and single-crystal diffraction geometries via iterative implementation of the Bragg/Laue equations. This enables focused refinement on a drastically smaller solution space. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.