| Abstract Scope |
Iterative phase retrieval in Bragg Coherent Diffraction Imaging (BCDI) is computationally expensive. This work aims to generate physically realistic training data for BCDI using two complementary approaches to create diverse defect structures typical of plastic deformation in metallic systems. First, molecular dynamics (MD) simulations under varied loading conditions (e.g., tension, compression, shock) naturally induce a wide range of defects. Second, specific defects such as edge and screw dislocations, stacking faults, and twin boundaries are systematically introduced into single crystals, followed by MD relaxation to obtain energetically favorable configurations. Simulated diffraction patterns from these relaxed crystals are paired with their structures to train machine learning (ML) models. This scalable approach enables rapid, one-shot reconstruction of defected crystals from experimental diffraction data, bypassing slow, iterative phase retrieval. The methodology also supports future extensions to reconstruct polycrystalline materials and complex alloys, with relevance to mechanical properties, defect-mediated phase transformations, and structure-property relationships. |