Why is meshing so hard?
Today’s thought-provoking reading is a two-year old blog post from CFMS that asks “Why is meshing so hard?” I have often asked myself the same question and today decided – as is standard practice nowadays – to ask Google. I didn’t even have to complete the search to see what everyone else is asking.
Here are the thoughts the article provoked in me.
- The article quotes a meshing researcher as saying “anyone can write a CFD solver before lunchtime, but to write a good mesh generator takes a lifetime.”
- I am proving the latter. Still striving for good.
- While this person is obviously engaging in a bit of hyperbole to make a point, I have actually heard CFD people seriously claim to be able to write a CFD solver over a weekend.
- A prospective customer once told us that their company’s policy was that any software that cost more than $10,000 per year was cheaper for them to develop themselves.
- The challenge in meshing, according to the author, is the volume mesh. I disagree a bit because my personal experience has shown that most volume meshing problems are the result of a poor surface mesh.
- The author wonders why has meshing remained challenging for so long.
- The Navier-Stokes equations are the Navier-Stokes equations are the Navier-Stokes equations. Meshing has no such equivalent, common, mathematical basis and virtually anything goes to generate points. (Structured grid techniques using elliptic PDEs and the Delaunay technique for unstructured meshes come close.)
- All CFD solvers (do not take the word “all” literally) solve the N-S equations. But they each solve them in different ways that a mesher must support.
- There was a structured grid flow solver that required meshes to be left-handed. Really?
- Cell type support varies from hexes (structured and unstructured) to tets, prisms, pyramids (but really, don’t use pyramids they say), polyhedra. This doesn’t account for codes that allow (or don’t allow) hanging nodes and edges.
- Of course, the mesh quality levels desired by each solver differ, not only in value but in method of computation.
- CFD problem size and scope continue to evolve and expand. What took me a year to do in 1986 wouldn’t even qualify as an undergraduate homework assignment now. Today the geometry models to be meshed have more entities in them than there were grid points in the meshes I used to generate decades ago.
- In the beginning, everyone doing CFD had to be an expert. Now, with more and more people using CFD, there are a lot more non-experts. (This is a good thing.) But this makes designing software user experience (what we used to call graphical user interface back in the day) more challenging when you have to serve a broader range of users. Keep in mind that UX is important for meshing because we lack N-S equations that can simply be converged.
- Bottom line: all the unglamorous, dirty work flows downhill from the solver to the mesher.
- The author was quite prescient when wondering whether there should be mesh generation workshop. In fact, the 1st AIAA Geometry and Mesh Generation Workshop was held this past June and plans are underway for a second.
- Finally, the author (who has many more degrees than I) opines that the future of meshing may be massively parallel and involve machine learning. Smarter people than I will have to figure that out.
And y’all would be disappointed if I didn’t end with a comment I’ve repeated more than a couple of times, “Mesh generation exists in order to make turbulence modeling seem respectable.”
Upcoming Pointwise Events
- Webcast on 25 July: Multi-block Structured Meshing of a Diffusing S-Duct
- Workshop on 22 August in Maryland: Efficient Meshing with Pointwise
- Workshop on 16 October in Daejeon, South Korea: Efficient Meshing with Pointwise
- MSC released Adams 2017.2 for multibody dynamics simulation including a new aerodynamics module.
- Beta CAE released v17.1.2 of their software suite.
- Mentor released a new version of FloTherm for electronics cooling with a new feature called the command center with which users can define variations of base models.
- ANSYS Fluent has demonstrated scalability to 200,000 cores. [Wow.]
Applications, Events, and Jobs
- CFD helps with the design of vertical turbine pumps that lead to less cavitation and wear in boilers.
- Inside HPC delves into how the Exaflow project is working toward exascale simulations.
- DCS Computing’s Open Source CFD Workshop will be held 14 September 2017 in Linz, Austria. Keynote speakers will include Drs. Darrin Stephens and Chris Sideroff from Applied CCM and Applied CCM Canada, respectively. (Full disclosure, Applied CCM is a Pointwise partner.)
- Areva in Washington state seeks a thermal-hydraulics CFD engineer.
- Registration for TFAWS (Thermal & Fluids Analysis Workshop) is now open.
More Than an Array of Quads
Sometimes it’s just nice to see something that has a surface simplicity that hides an underlying complexity. The painting below is the work of Cheyney Thompson.