The Image Segmentation And Computational Simulation Based On GPU-accelerated LBM

The pore-scale porous media flows(PSPMFs)is common in the nature and engineering application.Analyzing the law of flow in porous media can help us better understand the flow phenomenon in the complex structures,which can serve the optimization of industrial process and the control of disease.Because the porous media usually has arbitrary structure,the solution of Navier-Stokes equation based on unstructured mesh is difficult.It is important to develop a new method combining easy modeling and effective resolving.So GPU-accelerated Volumetric lattice Boltzmann method(VLBM)is proposed in this paper to simulate fluid flow in porous media.Structured mesh is usually used in lattice Boltzmann method(LBM),meshing the complex structure easily.To realize the accurate simulation with less grids,VLBM is developed herein.The concept of VLBM is using continuous value,from 0 to 1,to represent boundary cells by calculating the percentage of skeleton.Then the volumetric grid is involved in the control equation of LBM.Based on the continuous data,microscale pore-structure upscaling is designed in this paper.The density of grid can be decreased by calculating the average value of adjacent cells,which can guarantee the invariance property of porosity.To ensure the small fluctuations in the transport properties of porous media,the upscaling times should be tested using representative samples for different models Theoretically VLBM upscaling method is applicable before the pore space has obvious fusion At the same time,VLBM,like the traditional LBM,has the natural parallelism of explicit dynamic resolving.So GPU technology is adopted in this paper to accelerate the solution of VLBM.This operation can not only improve the speed of calculation,but also increase the size of model which can be simulated.We can use traditional segmentation results to do the simulation and upscaling of VLBM.But to realize the accurate image segmentation and reserve more image information,the level set equation of geodesic active contour model is solved based on LBM concept in this paper and the boundary is represented using distance field.Combining the local refinement method and symbolic function,the continuous data for VLBM can be obtained.We developed relative codes for above methods and simulated the crude oil flow in the digital sand stone.808.7MLUPS,1400 times of CPU calculation speed,was realized on 256×256×512 model by using GPU accelerated VLBM.Meanwhile,2~3~32~3 upscaling was tested for the sandstone model.As the other application,we simulated the blood flow in the human choroid capillary firstly,analyzed the effects of choroid structures and observed the blood flow status at different AMD stages.