Modeling of shock wave propagation in porous magnesium based on artificial neural network

Abstract

Data transfer from the level of atomistic simulations to the level of continuum modeling is one of the urgent problems in the mechanics of materials. In this work, we perform multiple molecular dynamics (MD) simulations of the uniaxial (along different crystallographic directions) and volumetric (isotropic) compression of porous HCP magnesium single crystals in order to study the atomistic processes of the plastic deformation of anisotropic porous material and to create a reference dataset (strain dependencies of stresses, porosity and dislocation density at different temperatures and initial porosities). The MD data are used to train two artificial neural networks (ANNs): the first ANN is a surrogate of the constitutive equation of porous magnesium, while the second ANN describes the dislocation nucleation/emission as the onset of plastic flow. In contrast to the previously considered case of porous FCC aluminum, the present approach takes into account the anisotropy of the matrix material. The ANN-based constitutive equation is successfully used within the continuum mechanics model for numerical study on the shock wave structure in porous magnesium in comparison with direct MD simulation of the shock wave problem.