A GPU-accelerated 3D PF-LBM modelling of multi-dendritic growth in an undercooled melt of Fe–C binary alloy

Abstract

In the present study, a GPU-accelerated 3D PF-LBM model, which is a combination of the phase field method (PF) and the lattice Boltzmann method (LBM), is established to predict the multi-dendritic growth of Fe–C binary alloy. In order to solve the problems of insufficient memory and low data transfer efficiency faced by the GPU-based parallel computation for 3D dendritic growth, the memory segment loading method is implemented with the cudaMemcpy function in parallel computing architecture CUDA, and the optimum thread block size of 128 is determined. The computation performance analysis of the present GPU-accelerated 3D PF-LBM model shows that the present model gets rid of the limitation of GPU memory and the speedup ratio of present model reaches 1700 times for the cell number of 5123. Then, the present model is used to investigate the stationary single/multiple dendritic growth in an undercooled melt of Fe-2mol pct C binary alloy with or without a forced fluid flow of 0.05 m/s, and the results show that the rejected solute at the solid/liquid interface is washed away by the forced flow and the solute is enriched at the downstream. But the solute enrichment at the interdendritic space is hardly washed away by the forced flow, when the dendrites are fully developed. Thus, the solute enrichment becomes more significant in the downstream region, especially at the interdendritic space. Therefore, the equiaxed dendrite morphology shows a significant asymmetrical morphology for the multiple-dendritic growth with forced fluid flow.