Modified MPS Method For Free Surface Flows And Its Heterogeneous Parallel Acceleration

Free surface flow widely exists in natural and engineering problems.Free surface flow is highly non-linear and often occurs with complex phenomena such as breakage,rollover and fusion of free surface.It is difficult to study it theoretically,while the experimental method is expensive and usually needs a long period.The moving particle semi-implicit(IPS)method is a Lagrangian description particle method.It avoids the numerical dissipation on the interface capture or track-ing and can simulate the large deformation free surface problem well.However,the MPS method suffers from the fundamental instability problem and high computational cost in its practical appli-cation.Therefore,it is of great practical significance to develop a three-dimensional parallel MPS method which can calculate pressure steadily and accurately.Firstly,this paper introduces the theoretical basis of the MPS method,verifies its feasibility in free surface simulation by dam-break and jet fracture problem,and applies the MPS method with surface tension model to liquid rocket engine atomization simulation for the first time.The qualitative comparison between simulation results and experimental observations shows that the MPS method successfully simulates three modes of atomization,which verifies the applicability of the MPS method in atomization simulation.In order to improve the stability of pressure calculation of the MPS method,this paper synthet-ically analyses and considers the factors of equation property of pressure Poisson equation(PPE),tensile instability and accuracy of solving the equation of Navier-Stokes equation,and puts for-ward a whole scheme to improve the stability and accuracy of pressure calculation of the MPS method.To improve the accuracy of the boundary conditions of the PPE equation,comparision of the accuracy of various free surface detection algorithms is made,and the most effective light source method is selected as the free surface detection method for subsequent calculations.Dy-namic stabilitzation(DS)algorithm and particle shifting(PS)algorithm are introduced to deal with tensile instability problem and improve the accuracy of pressure solution in the MPS method.The numerical simulation results of a hydrostatic pressure problem show that the PS algorithm and the DS algorithm can effectively improve the unstable pressure calculation of the MPS method,and the DS algorithm achieves nearly perfect simulation results.The simulation results of a square patch rotating problem show that the DS algorithm can effectively improve the stability of the cal-culation,while the PS algorithm can improve the particle distribution in the calculation process.Two-dimensional dam-break problem shows that the pressure calculation results of the MPS-PS-DS method,which combined with various improved schemes,are more stable and accurate than those of the original MPS method.In order to improve the computational efficiency of the MPS method and realize large-scale three-dimensional computation,a MPS method based on GPU heterogeneous acceleration is devel-oped and the acceleration ratio is analyzed.The computational results show that the explicit part of the program has a high acceleration ratio,while the implicit part limits the overall acceleration ratio,and the overall acceleration ratio reaches a maximum of 16 times.The developed GPU MPS method is applied to the simulation of high energy fuel-coolant interaction(FCI)problem.The penetration process of the jet is successfully simulated and the simulated penetration depth is in good agreement with the experimental results.Moreover the developed GPU MPS method qual-itatively simulates two stages of the FCI problem.The atomization of direct impingement nozzle is simulated,and the effects of different jet velocities and impinging angles on the atomization characteristics are analyzed.In order to solve the problem of single-GPU memory capacity limitation,a multi-GPU accel-erated MPS program is developed.A region decomposition method based on background grid and particle number is adopted to achieve better load balancing.Communication in distributed memory computer cluster is realized using Message Passing Interface(MPI)library.The results of a three-dimensional dam-break problem show that the program has strong scalability and weak scalability,and the parallel efficiency is high.As a demonstration example,the dam break problem with three obstacles on the scale of tens of millions of particles is successfully simulated using the developed multi-GPU MPS program,which shows the practicability of the program.The problem of swirling liquid film atomization is simulated.The process of liquid film breaking into liquid filament and liquid filament breaking into droplets is successfully captured.The typical image of swirling liquid film breaking process is in good agreement with the experiment.