New mesoscale approaches to microstructural evolution and analysis

As the link between the atomic and continuum scales, microstructural models and simulations are a critical element of computational materials science. However, while standardized computational tools have become widely accepted at the electronic, atomic, and continuum scales, mesoscale simulation methods remain diverse and are still evolving. In part, this stems from the breadth of phenomena included under the microstructural umbrella, but it is also due to intrinsic limitations in the prevailing methods. Discrete microstructural evolution simulations (i.e. the Monte Carlo Potts model, probabilistic cellular automata) are computationally efficient and easy to implement; they suffer, however, from various artifacts of the underlying computational lattice. This talk will introduce a lattice-­‐free, discrete kinetic Monte Carlo method. The Material Point Monte Carlo (MPMC) method uses randomly placed material points to overcome the unphysical effects of lattice anisotropy on interfacial and volumetric energies and to enable the correct evolution of systems that undergo shape distortion. MPMC simulations retain most of the computational benefits of other discrete methods and provably reproduce the physics of interface motion by surface and bulk driving forces. One goal of mesoscale simulations is to understand rare events, such as failure initiation, hot spot formation, and abnormal grain nucleation. Because these phenomena typically arise from the localization of long-­‐range interactions, identifying an incipient rare event is challenging. This talk will present a network theory approach to analyzing long-­‐range grain neighborhoods using the random walk graph kernel. Utilizing data from microstructural evolution simulations, a machine-­‐learning system can be trained to classify potential abnormal growth events within the grain network. By operating beyond a nearest-­‐neighbor or mean-­‐field interaction distance, this method has promise for characterizing a number of long-­‐range microstructural phenomena.

Elizabeth A. Holm, Philip Goins, and Brian DeCost