Abstract
Nucleation of dislocations in a homogeneous crystal lattice is relevant for small-scale plasticity or ultra-fast loading. Previously, we improved the dislocation nucleation theory and proposed to use artificial neural networks (ANNs) trained by molecular dynamics (MD) data to obtain a self-contained description [1]. The ANNs were used to approximate material properties, such as stress-strain relationship, shear modulus and generalized stacking fault at the elastic stage prior to the nucleation of dislocations. In the present work, we consider the case of copper single crystal in a wide range of pressures from-10 GPa to +50 GPa. At preparation of training data, we apply a polynomial extrapolation of MD data beyond the nucleation limit, which allows us to improve the precision of the trained ANNs and make the theory predictions more accurate. Also we develop an approximate approach, which requires smaller and simpler MD data for training, but gives the strain rate dependence of the nucleation threshold close to the rigorous theory of dislocation nucleation.
NSTL
Elsevier