EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

Engineering-oriented multi-phase-field models aim at precisely matching the sharp-interface asymptotic of grain boundary and triple junction motion. However, numerical solution of the governing partial differential equations introduces an inherent discretization error. Quantification of this error is of critical importance: a) for developers to guarantee accuracy, reliability and robustness of the code over a wide parameter range, and b) for users to optimize numerical input parameters for a specific application range and error tolerance. Manually starting and evaluating the required series of benchmark simulations is tedious, inefficient, and prone to subjective bias. We here present an integrative numerical framework to benchmark MICRESS simulations of anisotropic grain growth. The benchmark example addresses the evolution of a central grain interacting with neighbouring grains by boundary curvature and junction dynamics. The limits for the physical parameters (numbers of neighbours, domain size, boundary energies and mobilities) as well as for the numerical parameters (grid spacing and diffuse interface width) can be specified. A Python program then configures parameter variations within the selected ranges and automatically generates associated simulation input, to be processed into MICRESS input files using the template engine Jinja. The software solution is batch-capable and designed to be integrated into an ICME infrastructure, holding all software dependencies and delegating the workload to a HPC clusters. Alternatively, benchmarks can be orchestrated interactively via IPython. At the end of each simulation, the rate of fraction of the central grain is evaluated and compared to the analytic benchmark solution, derived from the general Neumann-Mullins equation. The dependency of the overall error on the input parameters is evaluated by regression analysis. To allow a deeper insight, the framework additionally offers interactive plotting of the error and the associated phase-fields over time.