09-18-2018 | Juliette Pardue: Three-Dimensional Prism-Dominant Mesh Generation for Viscous Flows Around Surface Slope Discontinuities

105th NIA CFD Seminar: Three-Dimensional Prism-Dominant Mesh Generation for Viscous Flows Around Surface Slope Discontinuities

Date: Tuesday, September 18, 2018

Time: 11am-noon (EDT)

Room: NIA, Rm137

Speaker: Juliette Pardue

Weblink: http://hiroakinishikawa.com/niacfds

Abstract: The NASA CFD Vision 2030 Study has identified mesh generation to be a significant bottleneck in the CFD pipeline. Human intervention is often required due to a lack of robustness and automation while generating the mesh. An automated approach to generating unstructured three-dimensional, prism-dominant meshes for viscous flows is presented. Meshes comprised of prisms are advantageous because they yield a more accurate solution and have fewer elements than their purely-tetrahedral mesh counterpart. An extrusion-based approach using multiple normals is used where anisotropic prisms are formed from the surface mesh facets and blend prisms are used to fill the cavities between multiple normals. Multiple normals are needed at a node to satisfy the visibility requirement for all incident surface facets. These multiple normals arise at regions of the surface mesh where there are convex discontinuities in the slope of the surface across edges. Nodes where blend region meshes must be formed are classified using an exhaustive enumeration scheme based on the number of convex edges and concave edges that are incident upon the node. Templates are presented to robustly mesh all the enumerated types of blend regions that occur at ridges, cusps, and corner nodes, including non-Lipschitz nodes. Intersections are detected in the boundary layer using an efficient spatial partitioning tree and are treated using Laplacian smoothing of the normal vectors or by reducing the total height of the boundary layer in the identified regions. The remainder of the volume is then tetrahedralized with an isotropic Delaunay mesh generator.

Speaker Bio: Juliette Pardue is a Ph.D. candidate in Computer Science at Old Dominion University under Dr. Andrey Chernikov and Dr. Nikos Chrisochoides. She is part of the Center for Real Time Computing, National Institute of Aerospace, and NASA Langley. Her research interests include mesh generation, parallel algorithms, computational fluid dynamics, and computational geometry. She has been published in the International Meshing Roundtable, the International Conference on Parallel Processing where her paper won the best paper award, the Virginia Modeling, Analysis, and Simulation Center Capstone Conference, and AIAA’s Aviation Fluid Dynamics Conference. The first revision of her 2D distributed-memory parallel mesh generation work and code are currently under review by ACM’s Transactions on Mathematical Software.