Phase-field simulations of dendrite growth with melt flow remain challenging from a computational cost perspective. However, as the dendrite growth morphology is significantly affected by melt flow, continuous development of high-performance computational techniques is essential to better understand the dendritic solidification. Large-scale parallel computations with multiple graphics processing units (GPU) were enabled to accelerate phase-field lattice Boltzmann (PF-LB) simulations of dendrite growth with melt flow [S. Sakane et al., J. Cryst. Growth, 474 (2017) 154-159]. Subsequently, a two-relaxation LB model was introduced into the PF-LB model to use the kinematic viscosity coefficient of real materials efficiently [S. Sakane, T. Takaki, Comput. Mater. Sci., 186 (2021) 110070]. Furthermore, an adaptive mesh refinement scheme in a parallel-GPU environment was developed (parallel-GPU AMR) [S. Sakane, T. Aoki, T. Takaki, Comput. Mater. Sci., 211 (2022) 111542]. Performing parallel-GPU AMR simulations of PF-LB model, we investigated three-dimensional (3D) columnar dendrite growth of Al-Cu alloys with natural convection [T. Takaki, S. Sakane, T. Aoki, ISIJ Int., 63 (2023)]. This study investigates the effect of natural convection on 3D dendrite morphology in more detail through high-performance parallel-GPU AMR PF-LB simulations. In particular, the possibility of dendrite fragmentation owing to natural convection is studied in detail.

Moderated by: Andreas Ludwig / Alain Jacot