Simulation And Research Of Phase-field Model Based On GPU

Phase-field method does not need to track boundary in real time as well as judge the explicit boundary conditions repeatedly, and it has become an effective computational tool to simulate complex interface in the crystal growth process. As phase-field method is applied more and more widely, the established phase-field model is more and more complicated in solidification microstructure simulation.Large-scale simulation is required to realize more realistic phase-field simulation,which makes the calculation amount is a prominent bottleneck by using finite difference method, and the problems of large computational amount, low computational efficiency and small-scale simulation exist on a single CPU computation. Therefore, it is very important to adopt reasonable and efficient numerical computation method to solve phase-field model, which has become one of the research focuses in this field. In this thesis, a parallel computing method based on CUDA + GPU architecture is explored, and the parallel solution to phase-field model is implemented. This thesis mainly carries out the following research work:(1) The Sola algorithm is combined with the phase-field model of pure diffusion dendrite growth, and a phase-field model of dendritic growth under forced flow is established, called Sola-phase field model. Taking pure SCN for example, the dendritic growth evolution process in the presence of flow is studied. However, the diverse solidification microstructure styles observed in the researchers’ experiments clearly show the understanding limitations of academia to the dendritic solidification process at present. Therefore, the present two-dimensional solidification microstructure researches cannot predict the real three-dimensional solidification microstructure by using the two-dimensional information. So, it is necessary to study the solidification phenomenon by means of three-dimensional phase-field simulation.In this thesis, based on the phase-field model proposed by Karma, a three-dimensional phase-field model coupled with temperature field is established, and the simulation research on the three-dimensional phase-field model is also carried out by using pure SCN as the simulation object.(2) The parallel solutions to two-dimensional and three-dimensional phase-field model are implemented on GPU which has powerful computing capability by the concurrent execution of multiple threads, which improves the computational efficiency when using the finite difference method, simulates dendritic growthevolution process in lager-scale to further understand the microstructure formation mechanism during solidification process and reappears the evolution trends of interface morphology and growth kinetics.(3) The computation time based on GPU and CPU are compared under the same condition. The results show that the acceleration effects in different degrees have obtained with the GPU parallel computing, which improves the computational efficiency greatly.(4) The simulation results of the dendritic tip velocities and tip radius based on GPU are compared with the simulation results based on CPU, the values of microscopic solvability theory and the related references’ simulation results. The results show that they are identical with each other, which indicates the reliability and validity of the GPU parallel algorithm.