[SY-F13] Advanced analysis tools for atomistic microstructure modeling
Invited
Direct molecular dynamics (MD) simulations of materials microstructures become feasible thanks to fast growing computing capacities and can deliver a wealth of important observations on fundamental mechanisms. Such simulations are cross-scale rather than multi-scale, i.e. simultaneously large enough to represent meso- and macroscopic structures and processes and yet fully resolved, tracing every "jiggle" of atomic motion. Our recent MD simulations of dislocation-based crystal plasticity will be presented as an example [Nature 550, 492-495 (2017)].
To deal with the overwhelming complexity of processes on these scales, we develop advanced computational methods and tools to recast the massive transient trajectory data generated in large-scale MD simulations into a human comprehensible and analyzable form using methods of "in-situ computational microscopy". These algorithms automatically identify grain boundaries, surfaces and interfaces, dislocations, other defects, structural phases and deformation fields and can track their evolution. The data reduction and transformation of the underlying MD model into high-level microstructure representations make insightful visualization, quantitative analyses and linking with higher-scale materials models possible.
To deal with the overwhelming complexity of processes on these scales, we develop advanced computational methods and tools to recast the massive transient trajectory data generated in large-scale MD simulations into a human comprehensible and analyzable form using methods of "in-situ computational microscopy". These algorithms automatically identify grain boundaries, surfaces and interfaces, dislocations, other defects, structural phases and deformation fields and can track their evolution. The data reduction and transformation of the underlying MD model into high-level microstructure representations make insightful visualization, quantitative analyses and linking with higher-scale materials models possible.