Ultrafast analysis of individual grain behavior during grain growth by parallel computing

Abstract

The possibility to characterize in an automatized way the spatial-temporal evolution of individual grains and their properties is essential to the understanding of annealing phenomena. The development of advanced experimental techniques, computational models and tools helps the acquisition of real time and real space-resolved datasets. Whereas the reconstruction of 3D grain representatives from serial-sectioning or tomography datasets becomes more common and microstructure simulations on parallel computers become ever larger and longer lasting, few efforts have materialized in the development of tools that allow the continuous tracking of properties at the grain scale. In fact, such analyses are often left neglected in practice due to the large size of the datasets that exceed the available physical memory of a computer or the shared-memory cluster. We identified the key tasks that have to be solved in order to define suitable and lean data structures and computational methods to evaluate spatio-temporal grain property datasets by working with parallel computer architectures. This is exemplified with data from grain growth simulations.