An MD-continuum multi-scale model to simulate the ramp compression of an elastic-plastic material

The simulation of ramp compression is a multi-scale problem; wave propagation occurs on significantly greater spatial and temporal scales than is tractable for atomistic simulation, but material behaviour - which is most effectively simulated atomistically - can play a significant role in wave propagation.

This thesis describes the development and use of a novel multi-scale methodology for simulating the ramp compression of an elastic-plastic crystalline solid. The first piece of work investigated the following assumption: for sufficiently low strain rates, the local ramp compression of a single Lagrangian element can be approximated in molecular dynamics (MD) by uniform, uniaxial compression (UUC). This assumption was analysed by comparing a piston-driven non-equilibrium MD simulation with a UUC MD simulation, both seeded with initial voids. For the tested ramp rise time of 100 ps, the assumption was found to be sound: the thermodynamic evolution of a Lagrangian element was effectively reproduced.

The second piece of work, inspired by the first, developed a multi-scale model to enable the simulation of ramp propagation for an arbitrary piston velocity profile. Three major components and their development are presented: 1) a one-dimensional hydrocode, 2) an MD data set of UUC simulations, and 3) a neural network, trained on the MD data set, to act as the equation of state/constitutive model for the hydrocode. The developed simulation tool was used to simulate a sample driven by a piston, ramped from rest to 1100 m/s in 100ps. Using voids as a plasticity parameter, the tool was found to agree well with an NEMD simulation of the same scenario but at a significantly reduced computational cost.