Multi-GPU Based Immersed Boundary-lattice Boltzmann Method For Simulation Of Submerged Vegetation Water Flow

Aquatic vegetation is widely found in rivers,lakes,and wetlands,and has significant effects on the characteristics of open channel flow,such as reducing flow velocity,enhancing water turbulence,changing sediment transport and streambed morphology,and improving water ecology,etc.These changes further affect the physical,chemical,and biological processes of the water ecosystem.Therefore,the study of vegetated water flow is of great importance for ecological protection and river sediment management.In the field of submerged vegetated water flow research,numerical simulation methods are in the development stage,and the difficulty lies in the fact that numerical methods need to have the ability to simulate large-scale flow fields and deal with large deformation fluidstructure interaction(FSI)at the same time,especially for the study of flexible submerged vegetation,where the characteristics of water flow are affected by the deformation of vegetation while changing the morphology of vegetation.The immersed boundary-lattice Boltzmann(IB-LB)method is well suited to handle such complex FSI problems.The method has the advantages of simple algorithm,high parallelism,and can use the graphics processing unit(GPU)to improve computational efficiency.The IB-LB method has been widely used after more than ten years of development.However,this method still has disadvantages in flexible submerged vegetated water flow applications: the single GPU computation is not computationally efficient under large-scale grids,and the numerical stability is poor when simulating high Reynolds number turbulent flow problems.To address these problems,this paper develops the following aspects of research.(1)A multi-GPU optimization method without the message passing interface(MPI)is implemented.An optimization method is implemented based on the existing MPI-based multiGPU approach.The optimization is that one(CPU)controls two to four GPUs,and the data of GPUs are exchanged directly within the CPU without exchanging through MPI.The accuracy of the optimization method is verified by calculating the lid-driven flow at different Reynolds numbers,and the computational performance of two-GPU and fourGPU parallelism is tested.Multi-GPU parallelism is more suitable for large-scale meshes in three dimensions,and can perform higher computational performance after the number of meshes reaches the level of 10 million or more,and the computational speed is up to2.7 times of single-GPU parallelism when four-GPU parallelism is applied.(2)Introducing the wall-adapting local eddy-viscosity model into the IB-LB method.The WALE-based large eddy simulation(LES)model is combined with the IB-LB method to replace the original standard Smagorinsky subgrid model.The twodimensional flow around a cylinder is used to validate the method,and the distribution of the time-averaged pressure coefficient on the cylindrical surface agrees well with the experimental values.After that,the method is used to calculate the three-dimensional FSI problem of elastic filaments,and the results of flow velocity and vortex distribution in the flow field and the motion characteristics of elastic filaments are obtained.When the Reynolds number based on filament length is 3000,the elastic filament can keep the shape stable and the oscillation state is consistent with the flexible vegetation,which proves the feasibility of the IB-LB method to simulate the high Reynolds number flexible submerged vegetation water flow.(3)Simulation of flexible submerged vegetation water flows with a grid of 40 million.The IB-LB method is combined with the multi-GPU optimization method and the WALE-based LES model to first simulate the two-layer rigid submerged vegetation flow to further verify the accuracy of the method in this paper.Then,the multi-GPU numerical simulation of the flow field of flexible submerged vegetation on a grid of 40 million is implemented on four GPUs.The arrangement of flexible vegetation is divided into two ways: 100 plants aligned and 95 plants staggered,and the Reynolds number of the flow field reaches 3000.The results were analyzed in terms of time average flow velocity distribution,Reynolds stress distribution,and the deceleration effect of different vegetation arrangements on the water flow.The flow velocity was reduced by 46.2% when the vegetation is staggered,which is conducive to increasing sediment deposition and reducing streambed erosion.In this paper,a multi-GPU method without MPI is implemented,and the stability of the IB-LB method for calculating the FSI of flexible structural bodies is enhanced by introducing the WALE model.Also,the flow field of 100 elastic filament arrays under a grid of 40 million is simulated.The computational results show the value of trying to further apply the IB-LB method to the study of submerged vegetation flow.