Achieving advanced CFD capabilities with high-performance computers

By JAMES MASTERS AND CAROLYN WOEBER|December 2019

The Meshing, Visualization and Computational Environments Technical Committee explores the application of computer science to pre-processing, post-processing and infrastructure in support of computational simulation in the aerospace community.

As problem size and complexity increases, the meshing community continues to grapple with effective utilization of high-performance computing platforms, which was one of the impediments laid out in NASA’s Computational Fluid Dynamics Vision 2030 study. In January, representatives from Pointwise, Symmetric, Cascade Technologies and Cambridge Flow Solutions made strides toward this goal by demonstrating the ability to generate and process meshes up to 13 billion elements. Exascale computing on extremely large data sets was further addressed in June when compression algorithms developed at Oak Ridge National Laboratories in Tennessee demonstrated the ability to compress time-accurate, unsteady, CFD data from 4.7 terabytes to 568 gigabytes. And Tecplot’s subzone load-on-demand data structure allowed transient solutions of 10 billion elements to be saved and stored on an engineering workstation.

New Jersey-based Intelligent Light, funded by a U.S. Department of Energy Phase II Small Business Innovation Research grant, developed an uncertainty quantification proof-of-concept for large datasets called Spectre-UQ, which demonstrated in January a technique to arrive at the total uncertainty for a numerical study.

Computational environments enabling rapid multidisciplinary design are becoming increasingly important. In response to this trend, the Computational Aircraft Prototype Syntheses research program, of which the Massachusetts Institute of Technology and Syracuse University are primary developers, provided infrastructure to enable multifidelity, multidisciplinary, physics-based aircraft design. In June, the Aerospace Analysis and
Design in Virtual and Augmented Reality toolKit, or AArDVARK, an engineering-oriented virtual and augmented-reality framework leveraging the CAPS infrastructure, demonstrated several relevant applications, ranging from packaging to modal analysis. These demonstrations highlighted how the appropriate utilization of virtual reality can augment traditional engineering analysis techniques.

Also on the computational environments front, the Computational Research and Engineering Acquisition Tools and Environments program released version 10.0 of Kestrel in May and version 10.1 in August. Support for axisymmetric 2D flow solutions was added to the primary finite volume flow solver in version 10.0 and to the high-order finite element solver in version 10.1. This allowed for 3D bodies of revolution to be simulated in 2D while still reporting loads consistent with the full 3D configuration. With the newest versions, it is also possible to restart a simulation with a different mesh system and visualization error from the Mach Hessian.

High-order meshing algorithms continued to make significant progress. In June, researchers at Mississippi State University demonstrated an advancing-layer method for generating curved boundary layer meshes. Pointwise entered its second year of a NASA Phase II SBIR to develop a high-order mesh capability that involves mixed-order degree elevation and adaptation. High-order mesh generation up to degree three is also now available in Capstone, as well as the corresponding high-order visualization. MIT, Pointwise and Intelligent Light also continued to explore high-order visualization. Intelligent Light won a Phase II SBIR in June to pursue visualization for high-order simulations.

In January, Capstone released version 10, which includes hexahedral elements in the boundary layer and tangential adaptivity for boundary layers, allowing users to generate automatic smooth transitions with the unstructured mesh. Version 10 also enhanced geometry capability with new repair techniques.

Pointwise completed an SBIR Phase I contract for the U.S. Air Force in March that focused on access to computational geometry from within the flow solver. MeshLink, an open source library released in May as part of the contract, solves the problem of
geometry-mesh associativity while providing simplified access to the supporting geometry during mesh manipulation and adaptation.

Finally, researchers from the French National Research Institute made a significant advancement in turbomachinery applications when they demonstrated the viability and efficiency of unstructured anisotropic mesh adaptation techniques to turbomachinery applications including internal regions with periodic boundaries.