Unstructured mesh based elastic wave modelling on GPU: a double-mesh grid method
Unstructured mesh based elastic wave modelling on GPU: a double-mesh grid method
复制标题
GPU 上基于非结构化网格的弹性波建模：双网格方法
DOI：
10.1093/gji/ggx339
复制
发表时间：
2017-11
影响因子：
2.8
通讯作者：
Gao Hongwei
中科院分区：
文献类型：
--
作者：
Yang Kai;Zhang Jianfeng;Gao Hongwei
我们提出了一种基于非结构网格的数值技术来模拟弹性波在复杂几何设置的非均质介质中的传播。该方案是通过采用所谓的网格方法和双网格实现来开发的。双层网格是通过将一级网格的每个三角形网格细分为一组相等的较小网格，并等分三角形的每个边来生成的。该方法结合了结构化网格和非结构化网格的优点。基于Delaunay三角剖分的质心Voronoi镶嵌生成的不规则非结构化一级网格具有速度依赖的密度函数，可以准确地描述表面形貌和界面，并且网格单元的大小可以根据局部速度变化。与不规则、非结构化的整体网格相比，一级网格的每个网格单元内一致的小网格大大减少了几何系数的内存需求。采用双网格方法还可以缓解由于采用完全非结构化网格而导致的内存访问不连续问题。结果表明，采用该方案的GPU实现可以获得较高的加速速率。数值算例验证了双网格弹性网格法的良好性能。
We present an unstructured mesh based numerical technique for modelling elastic wave propagation in heterogeneous media with complex geometrical settings. The scheme is developed by adapting the so-called grid method with a double-mesh implementation. The double-mesh is generated by subdividing each triangular grid of the first-level mesh into a group of congruent smaller grids with equally dividing each edge of the triangle. The resulting double-mesh grid method incorporates the advantages of structured- and unstructured-mesh schemes. The irregular, unstructured first-level mesh, which is generated by centroidal Voronoi tessellation based on Delaunay triangulation with a velocity-dependent density function, can accurately describe the surface topography and interfaces, and the size of the grid cells can vary according to local velocities. Congruent smaller grids within each grid cell of the first-level mesh greatly reduce the memory requirement of geometrical coefficients compared to a whole irregular, unstructured mesh. Applying the double-mesh approach can also alleviate the discontinuity of memory accessing mainly caused by adoption of fully unstructured mesh. As a result, the GPU implementation of the proposed scheme can obtain a high speedup rate. Numerical examples demonstrate the good behaviour of the double-mesh elastic grid method.
登录
查看更多内容
影响因子：
3
作者：
P. Micikevicius
通讯作者：
P. Micikevicius
影响因子：
3
作者：
D. Komatitsch;J. Vilotte
通讯作者：
D. Komatitsch;J. Vilotte
影响因子：
0.9
作者：
J. Virieux
通讯作者：
J. Virieux
影响因子：
2.8
作者：
Z. Jianfeng;Liu Tielin
通讯作者：
Z. Jianfeng;Liu Tielin
影响因子：
3.3
作者：
R. Weiss;J. Shragge
通讯作者：
R. Weiss;J. Shragge