Improved MPS Method And Its 3D Parallel Computation

As a meshless particle method, MPS(Moving Particle Semi-Implicit) method has a great advantage in dealing with large deformed free surface flows. However, it suffers from strong non-physical pressure oscillation, which to a large extent decreases its reliability of pressure computation. In addition, computational cost of MPS method is high, severely limiting its application in 3D flows involving a large number of particles. Therefore, it is of significant importance to develop a 3D parallel MPS method which is able to produce smooth pressure field.The present study investigates the pressure oscillation behavior of MPS method, focusing on three aspects, i.e. the pressure gradient model, the source term of the pressure Poisson equation(PPE) and the free surface detection. Numerical tests are carried out against hydrostatic problem and violent sloshing flows. Results of hydrostatic problem show that using the momentum conservation gradient model or the mixed source term method produces limited improvement on pressure field, while the combination of the two schemes successfully suppress the pressure oscillation. However, the performance of the combined method is not satisfactory in violent sloshing in which free surface is largely deformed. Detail analysis is conducted, showing that the reason of pressure oscillation under such circumstance is due to misjudgement of surface particles. To overcome this, a new surface detection method is proposed in the present study, which judges the surface particles based on asymmetry distribution of neighboring particles. Numerical tests show this new detection method greatly improves the accuracy of surface judgment. Therefore, an improved MPS method is constructed by combining the momentum conservation gradient model, the mixed source term and the new surface detection method together. Simulation of violent sloshing flow shows that the improved MPS method successfully suppresses the pressure oscillation behavior, thus computed pressure by improved MPS is in good agreement with experimental data.To achieve an efficient 3D simulation with a large number particles, a parallel strategy for MPS computation is developed, which is based on a background grid with dynamic load balance strategy. By means of MPI(Message Passing Interface) library, the parallel MPS is able to perform parallel computation on distributed memory cluster. Speed-up is tested by computing a 3D dam break flow with various number of particles. Thanks to the dynamic load balance, the measured speed-up is satisfactory. Further analysis compares the computational time and speed-up of each procedures in a single time step computation and indicates that PPE is the key factor for parallel efficiency of MPS. In view of this, GPU is adopted to accelerate the solution of PPE. An open source library, CULA, is used to run the parallel computation on a GPU card, i.e. Tesla C1060. Test results show that GPU can achieve quite high speed-up when the number of particles is large, implying a great potential in MPS parallel computation.Based on the improved MPS and parallel strategy, a 3D parallel MPS solver, MLParticle-SJTU(Meshless Particle Solver developed at Shanghai Jiao Tong University), is developed in the present study, and applied into several typical flows in field of ship and ocean engineering, including dam break, liquid sloshing, green water and ship-wave interaction flows.3D dam break flows are simulated to analyze the particle convergence and validate the developed MLParticle-SJTU solver. Results show that the computed free surface and impact pressure are in good agreement with experimental data and other nemerical results. In addition, increasing particle resolution can improve the capturing of details of free surface deformation.3D violent sloshing are investigated in this paper. 2D tank sloshing is first computed to analyze the reliability of MLParticle-SJTU in context of sloshing flows. the tank is subjected to pitch, surge and coupled motion. Good agreement is obtained between computed free surface and impact pressure and experimental ones. In addition, the effects of excitation frequency and excitation amplitude on sloshing flows are analyzed. Results show that excitation frequency significantly affect the evolutions of wave and impact pressure, while excitation amplitude mainly affect the impact pressure and this effect becomes more evident in the vicinity of resonant frequency. To analyze the difference of 2D and 3D simulation, sloshings in rectangular tank forced to horizontal move are computed in 2D and 3D. The general motions of liquid in 2D and 3D computation are quite similar, however 3D simulation can produce smoother pressure field, and shows more natural behaviors of breaking wave and liquid splashing. In addition, 3D membrane tank sloshing in coupled motion of six degrees of freedom is computed, showing a much more complicated wave pattern compared with waves in 2D tank. 3D sloshing in baffled tank are also investigated. The effect of baffle length on sloshing is analyzed. Results show that fluid motion can be decreased with an increase of baffle length, this effect becomes less evident when the baffle length is close to the water depth.3D green water flows are investigated in the present study based on MLParticleSJTU solver. To achieve this, a numerical wave tank is first constructed in which wave can be generated through piston-type wave maker and damping region is incorporated to absorbe the reflected wave. To validate MLParticle-SJTU solver, 2D green water is simulated and the shipping wave and impact pressure on the super structure mounted on the deck obtained numerically are compared with experimental ones, showing a good agreement. Furthermore, 2D and 3D wave-body interaction flows are studied. 2D floating breakwater interacting with regular wave is simulated. The computed motion of floating body is in good agreement with experimental data. Besides, motions of 3D Wigley hull in regular waves with large amplitudes are simulated. Results show that the Wigley suffers from evident pitch motion and green water phenomena are observed. MLParticle-SJTU shows good stability in the simulations of such complex flow phenomena.