Automated identification of dislocation structures from experimental Laue microdiffraction patterns

Benjamin Udofia, Ruhr University Bochum, Germany

In this study, a cross-modal machine learning framework is presented for associating X-ray diffraction patterns with their corresponding dislocation microstructures. Using discrete dislocation dynamics simulations and virtual diffraction modeling, paired structural and diffraction data are embedded into a shared latent representation through contrastive learning. The learned representation enables efficient retrieval of corresponding dislocation microstructures directly from diffraction data, while representative subset selection significantly reduces the amount of training data required. The proposed framework demonstrates the potential of machine learning to bridge diffraction measurements and underlying microstructural information, providing a foundation for future diffraction-based characterization of crystalline materials.