The work presents a multiscale investigation of deformation mechanisms in nanocrystalline (NC) aluminum, combining large-scale molecular dynamics (MD) simulations with machine learning techniques to transfer MD data into continuum mechanics simulation at macroscale. We examine symmetric tilt grain boundaries (GBs) with misorientation angles ranging from 8.79° (Σ85a) to 36.87° (Σ5), establishing the dependence of GB mechanical response on GB structure within fixed misorientation angle. Three distinct plastic relaxation mechanisms are identified: 1) GB migration in low-angle systems; 2) coupled GB sliding and grain rotation in intermediate- and high-angle systems; and 3) dislocation plasticity. The first two mechanisms (GB migration and coupled sliding/rotation) predominantly serve as the primary plastic relaxation pathways during initial deformation stages, while dislocation plasticity activates at the late deformation stages. The dislocation plasticity initiation depends on the primary deformation mechanism: 1) following GB migration, dislocation activity originates from segment emission at annihilation sites of the migrating dislocation walls; 2) in systems with GB sliding/rotation, the transition to dislocation-mediated plasticity occurs when localized GB stresses exceed critical thresholds (4–5 GPa in 1 nm regions). The timing of this transition varies significantly (50–600 ps) depending on GB character, reflecting fundamental differences in defect nucleation barriers between low-angle and high-angle GB systems. A new hardening phenomenon is discovered during GB sliding with grain rotation, resulting from arising normal stress components that oppose sliding. The MD results demonstrate that statistical treatment of atomistic-scale data is essential for reliable transfer to continuum-level modeling. Our analysis reveals critical aspects demanding careful statistical consideration: 1) GB migration stresses vary by 35–40%; 2) sliding activation stress shows 80% variation. The MD results are transferred to continuum modeling through an artificial neural network framework approximating plastic potential reconstructed from the MD. Implemented within smoothed particle hydrodynamics simulations, this data-driven approach provides an efficient procedure to transfer atomistic data to continuum level for prediction of NC aluminum behavior.
扫码关注我们
求助内容:
应助结果提醒方式:
