A highly efficient 3D level-set grain growth algorithm tailored for ccNUMA architecture

Abstract

A highly efficient simulation model for 2D and 3D grain growth and recrystallization was developed based on the level-set method. The model introduces modern computational concepts to achieve excellent performance on parallel computer architectures. Strong scalability was measured on ccNUMA architectures. To achieve this, the proposed approach considers the application of local level-set functions at the grain level. Ideal and non-ideal grain growth was simulated in 3D with the objective to study the evolution of statistical representative volume elements in polycrystals. In addition, microstructure evolution in an anisotropic magnetic material affected by an external magnetic field was simulated.

A novel highly efficient Lagrangian model for massively multidomain simulation applied to microstructural evolutions

“…Performance and speed-up of the model will be given and compared to other highly efficient parallel methods in the litterature in the context of GG [28,30]. Also, in the following, the term domain will reference an individual grain in a microstructure, however, the parallel TRM approach can be extended to any massively multi-domain problem immersed in a 2D triangular mesh.…”

A vertex model of recrystallization with stored energy implemented in GPU

“…The methodology of [20] allowed for simulations of systems of 100 to 1000 grains. Mießen et al [21] proposed a two-and three-dimensional level-set algorithm adapted for ccNUMA architecture that allows highly efficient numerical simulations using OpenMP. Diffusion generated level set methods were proposed in [22], where about 4 million grains are nucleated and a comparative analysis is provided relative to Monte Carlo simulations.…”

Computationally Efficient Phase-field Simulation Studies Using RVE Sampling and Statistical Analysis

“…In fact, it has been pointed out by many investigations [7,8,9,10] that the more heterogeneous a microstructure is the finer it has to be resolved. An example is the problem of grain growth where despite the possibility for studying large-scale simulations [11,12,13] it was demonstrated that the size of the RVEs utilized in several grain growth simulations may not be enough to observe self-similar behaviour depending on the ini- 15 tial grain size distribution [14]. Evidently, the computational cost of representing a microstructure as a continuous and contiguous space is immense.…”