自适应代理粒子的非线性有限元与不可压缩流体实时耦合  

Adaptive Agent Particles for Real-Time Coupling of Nonlinear FEM with Incompressible Fluid

作　　者：袁志勇[1] 廖祥云[1] 郭甲翔[1] 童倩倩[1] 郑棉仑 

机构地区：[1]武汉大学计算机学院,武汉430072

出　　处：《计算机学报》2016年第7期1319-1333,共15页Chinese Journal of Computers

基　　金：国家自然科学基金(61373107;61190125);国家"九七三"重点基础研究发展规划项目基金(2011CB707904);北航虚拟现实技术与系统国家重点实验室开放课题基金(BUAA-VR-13KF-15)资助

摘　　要：该文提出了利用生成自适应代理粒子来实时耦合TLED非线性有限元与隐式不可压缩流体的新方法.首先,该文根据固体表面的局部高斯曲率对固体表面采样生成不同分辨率的代理粒子.其次,为了避免代理粒子和流体粒子间过早的耦合,文中设计代理粒子光滑核函数的支持域为椭球型.其次,为了防止流体粒子在固体发生形变时产生穿透现象,作者在每个时间步长内对固体表面进行重采样生成代理粒子,并自适应地改变其支持域的大小,最后,在处理边界时运用正则化方法计算边界处粒子的物理量,并利用基于统一计算设备架构的GPU并行计算方法对耦合过程进行加速.实验结果表明,该文提出的方法能在减少采样粒子、提高计算效率的同时,实现实时逼真的流固耦合效果.This paper presented a novel method that adaptively sampled agent particles for real-time coupling TLED nonlinear FEM solid with implicitly incompressible fluid.The agent particles were sampled with different resolution according to the local Gauss curvatures of solid surface model.The agent particles’support domain was designed as ellipsoid to avoid premature coupling between agent particles and fluid particles.To avoid penetration of fluid particles on boundaries in case of deformation,the agent particles were adaptively resampled after deformation of nonlinear FEM solid and its support domain was adaptively changed.To accurately handle the coupling,regularization method was adopted to calculate the particles’physical quantities on the boundary to obtain better coupling effects.The CUDA-based GPU parallel computing was utilized to accelerate the computation of coupling process between agent particles and fluid particles.The experimental results demonstrate that the proposed method can achieve significant improvement in time and memory,reducing the sampling particles and meanwhile obtaining real-time realistic fluid-solid coupling effect.

关 键 词：自适应代理粒子 流固耦合 非线性有限元 不可压缩流体 GPU加速 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]