Modeling microstructural evolution in single-phase, composite and two-phase polycrystals.

The grain boundary curvature driving force for grain growth arises from the angular conditions imposed by the balance of interfacial energies at three- and four-grain junctions. Interference with such angular conditions substantially alters grain evolution. Monte Carlo Potts model computer simulations of grain growth have been employed as an experimental tool to develop, clarify, and test physical theories of curvature-driven grain growth in pure and multiphase systems. The simulation lattice itself may interfere with grain junction angular conditions, causing deviations from normal grain growth behavior; decreasing the lattice anisotropy or increasing the simulation temperature minimizes these effects. In polycrystals which contain a static, inert second phase, certain boundary/particle intersections remove grain boundary curvature by compensating for grain junction angular conditions; grain growth slows or stops in such systems. A unique, pinned grain structure may evolve in a d-dimensional polycrystal only when pinning particles are infinitely long in fewer than d $-$ 1 dimensions. Because requirements for grain boundary connectivity differ with dimensionality, particle shape and volume fraction both contribute to grain boundary pinning in two dimensions; the effects of particle shape are minimal in three dimensions. When two phases evolve simultaneously, the interfacial energy differences between phases change grain junction angular conditions. Two-dimensional polycrystals continually evolve when these angles are thermodynamically fixed, while grain growth ultimately ceases when grain junction angles may vary. Since grain shapes, phase volume, and phase arrangements are dictated by interfacial energies, clustered-, alternating-, isolated-, and single-phase microstructures occur in different interfacial energy regimes. Efficient serial and massively parallel grain growth simulation algorithms have been developed, analyzed, and tested. While the computing time per simulation timestep of conventional serial algorithms is a constant proportional to the system size, a faster n-fold way algorithm increases in efficiency as grain growth progresses. The computing time per simulation timestep of an efficient massively parallel algorithm is a small constant.Ph.D.Applied SciencesEngineering, Materials scienceMaterials sciencePhysicsPure SciencesUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/128992/2/9308334.pd