For the past 18 months we have been working with MIT on an U.S. Air Force funded project to automatically generate an unstructured or hybrid mesh on and around virtually any geometry model. This work was presented at the Pointwise User Group Meeting 2018 and again at AIAA SciTech 2019. As this research comes to a close, we’re beginning to move the technology into the production software.
Truth be told, all of us have been able to automatically generate meshes for quite some time. More truth be told, this automation comes with a price.
- You had to start with a clean, watertight geometry model.
- The resulting mesh, while interesting, was unusable.
Archimedes said something along the lines of “Give me a fulcrum and I shall move the world.” Perhaps if he had done meshing he would’ve said “Give me a watertight geometry model and I shall mesh the world.” I think we all understand why a decent geometry model is a predicate for automatic meshing.
The easiest way to automatically generate a mesh is by using all tetrahedra because the methodologies (like the Delaunay criterion) are so robust. And while these meshes may be interesting for our solid mechanics friends, they are virtually useless for viscous CFD simulations.
Where does one get a clean, watertight geometry model? This research was focused on automatic mesh generation so we needed to start with Archimedes’ fulcrum and we did that by working with geometry models from the Engineering Sketch Pad (ESP) conceptual design software (what our USAF funding agency uses).
And because ESP is what one of its developers calls “analysis CAD” or aCAD, the geometry models are extremely analysis friendly. That friendliness has several aspects, but the critical one for meshing is the ability to apply attributes directly to various components of the geometry model and pass them to the mesher.
These attributes are imported by the mesher and are used to define how certain regions of the model should be meshed. This is what allows a usable, properly clustered, boundary layer resolving mesh to be generated.
From Geometry Model to Mesh
If we stuck to the Archimedes theme we would’ve called the implementation of the research results “Eureka” but it’s only a research artifact so it’s not all that important. The value of having a scripting language like Glyph that covers all of Pointwise’s functionality is that we were able to easily implement this research as a script that goes from import of the attributed geometry model (input #1) all the way through to export of a volume mesh to a CFD solver’s format. The meshing itself is all governed by best practices that have been codified in the script and supplemented by a parameter file (optional input #2). The script uses only existing methods and techniques in Pointwise; all it manages is workflow and the application of best practices. The image below shows some of the details of one particular mesh generated by this technique and shows how the viscous clustering was applied for both the surface and volume elements with absolutely no user interaction.
What’s next for you is to get on over to pointwise.com so you can watch the video replay of Dr. Steve Karman Jr.’s presentation from our user meeting on this work. He explains and demonstrates the technique in much more detail.
What’s next for us is to implement this script’s functionality into the production version of Pointwise. You’ll see the building blocks of this automation being released over the next several versions.