Parallel computing of direct simulation Monte Carlo (DSMC) based on compute unified device architecture (CUDA) is developed and improved data transmission in multi-GPU parallel computing is devoted to promote parallel efficiency.A two-dimensional Couette flow and lid-driven cavity flow by CPU,single GPU and double GPU parallel computing are simulated,respectively.Precision of results by GPU is consistent with that by CPU and speedup ratio can reach to 10~30 by single GPU acceleration and 40~60 by double GPU acceleration.Speedup efficiency by multi-GPU is approximated to 100%.

We propose a fast random simulation strategy based on differentially weighted MC. The strategy improves computation efficiency significantly, and guarantees enough calculation accuracy, thus coordinates contradiction between computation cost and computation accuracy. The main idea is based on majorant kernel. It is possible to transfer a traditional coagulation kernel to a majorant kernel through splitting and amplifying slightly. The maximum of majornant kernel is obtained by single looping over all simulation particles. The maximum majornant kernel is used to approximate the maximum coagulation kernel in particle population, and is further used to search coagulation particle pairs randomly with acceptance-rejection method. The waiting time (time-step) for a coagulation event is calculated by summing coagulation kerenls of particle pairs involved in acceptance/rejection processes.. Double looping in normal Monte Carlo simulation is avoided and computation efficiency is improved greatly.

An LB-based gas-solid two-phase model with two-way coupling is developed considering feedback forcing of particles in evolution equation of fluid particles. Smagorinsky subgrid model is also introduced in simulation of flow field with high Reynolds numbers. Classic particle-laden flow over a backward facing step is simulated and velocity profiles of gas phase and particles (considering one-way coupling and two-way coupling respectively) are compared with experimental results. The results of two-way coupling LB model are obviously better than these of one-way coupling LB model. Furthermore, preferential concentration of particles with different Stokes numbers (St) is investigated. It is found that small particles (St~0(0.1)) show better following behaviors with gas phase and are uniformly distributed in the flow field. Particles with moderate Stokes numbers (St~0(1)) are hard to be entrained into the vortex and show strong preferential concentration. On the other hand, large particles (St~0(10)) can enter into the vortex because of great inertial and are distributed more uniformly in flow field.