Long-distance settling simulation of equiaxed dendrite by a moving-frame algorithm: phase-field lattice Boltzmann study with parallel-GPU AMR

Abstract

In large-ingot castings, the settling of equiaxed dendrites often results in distinct cone-shaped negative segregation in the lower region of the ingot. To accurately predict and control such macrosegregation, it is important to understand the kinetic behavior of equiaxed dendrites in the melt. The phase-field lattice Boltzmann (PF-LB) model is powerful for simulating dendrite growth with melt convection and solid motion. However, it is computationally expensive and represents only the short-distance motion of dendrites in three-dimensional (3D) simulations. For an efficient 3D evaluation of the effect of dendrite motion and rotation on growth behavior, we introduce the moving frame algorithm to PF-LB simulations. Here, the computational domain tracks the settling dendrite to express long-distance settling without restricting the domain size. The PF-LB simulations were accelerated by parallel computing using a combination of multiple GPUs and adaptive mesh refinement (AMR), also referred to as parallel GPU-AMR. The moving-frame algorithm was modified to adapt to AMR. From the simulation results, we demonstrate that the proposed method helps evaluate the effect of dendrite rotation on the settling and growth velocities of equiaxed dendrites in 3D.