Please Excuse the Mesh: CFD and Pointwise

You probably know that Cadence acquired fluid dynamics meshing company Pointwise recently. I covered it briefly in my post Update: Pointwise, PCIe, RISC-V. I wanted to find out more, so I had a call with Pointwise CEO (or CEO Emeritus) John Chawner in Fort Worth (Dallas-ish) TX.

Meshing

First, Pointwise does the meshing part of computation fluid dynamics (CFD) but doesn't actually do the solving. Instead, It supports a broad range of industries (planes, submarines, blood pumps, curved architectural walls, fish dam bypass, and more). It supports a broad range of solvers, each with its own quirks, formats, and restrictions. “That keeps us honest since can’t assume what the user is going to do.”

However, meshing is really important:

Mesh generation for CFD is something everyone hates and thinks of as a necessary evil. It can take 75% of an analyst's time. They can get relly complex with meshes of 100M or half-a-billion elements around an external aircraft simulation. In about 2015, NASA produced a CFD Vision 2030 Report with a blue riband panel, which said that "meshing needs to be made invisible to the user". This report might as well have been our product strategy. Since it is 75% of the effort, reductions in meshing time are a really big deal. The analysts want to just get through it to get to the real meat of computing the flow fields.

Their product is called the same as the company, by the way: Pointwise. The original product, superseded in 2007, was called Gridgen (no, not a character in Lord of the Rings, short for Grid Generation).

I told him to explain how the meshing technology works, but to be gentle. He started with an analogy:

Say we wanted to model the HVAC airflow in your living room. Perhaps there's vent blowing air into the room. Perhaps there's a ceiling fan. At some level, we could model every molecule of air, but that's obviously computationally impossible. Instead, we can approximate using brick-shaped elements. The smaller we make the bricks, the closer we will approximate reality. If we just break the room up into nine bricks, like a Rubik's Cube, the solution will be completely useless. We could go the other way and break the room up into tiny cells. It would be very accurate but it would run forever. The trick with good meshing is to put the small cells where they are needed the most, and use larger cells where nothing much is going on. For example, modeling a plane, most of the interesting part is close to the surface of the aircraft, and further away we can be much coarser.

John told me that was an oversimplification in many cases, since sometimes there are advantages to using shapes that are not just "bricks". The shape can be skewed, which can affect the accuracy depending on the solver. Or the mesh can be broken into tetrahedra, which can be done really fast compared to the bricks, but which not all solvers support well (as in giving accurate results). For some solvers, Pointwise uses a hybrid approach, with really nicely structured bricks near the surface of the plane where accuracy is required, and then transition to tetrahedra further away where accuracy is not required so much. As with this car.

When Pointwise started, meshing was something where you got one go. You would be given the object (plane, car, pipe, whatever) and produce a mesh for the fluid surrounding it or flowing through it. You had to produce a perfect mesh first time. This would then be passed to the solver to work on the partial differential equations of the Navier-Stokes equations. However, nowadays there is mesh adaptation, where the solver can essentially report back "I need more mesh here" and so the initial mesh is no longer so critical, it just needs to be good enough to get started, and then the mesh can be refined where extra accuracy is required (for example, a shock wave, or a wake coming off a wing).

Another area of flexibility is that there is a scripting language. The toolkit and the scripting language together allow the users, who are the domain experts, to tailor the tool to what they require. It has allowed people to program things that the tool was never planned to do out of the box. A&D, turbomachinery, biomedical..."We are not experts in all these areas but customers can take the scripting language and build what they need".

Pointwise History

I asked John to give me the story of the company:

Pointwise was started in 1994 by myself and John Steinbrenner. We built on prior work we had done and decided to see if we could commercialize it. We met Rick Matus, who had been a product manager at Fluent. There three of us took it forward and never looked back. It was 26 people when Cadence acquired us.

About half the installed base is aerospace and defense, which is where most of us at Pointwise came from. Virtually every major aerospace program in the US in last 40 years was meshed with Pointwise. Beyond that, blood pumps, and 3D printing at architectural scale. For architecture can print freeform shapes like curved walls. That’s a mesh. One of our customers took a tour and said “Did you know that’s just a mesh like they use in CFD”. They 3D print large architectureal features using our meshing. [See the video at the end of this post.]

Submarines are especially tough since they require resolution of very fine details in the flow, so very small mesh cells. But a submsrine is very large, so something like 10⁷ scale difference in things you are trying to resolve.

One of our customers in Washington State was looking for ways for fish to bypass dams and power stations. They had to design fish bypass ducts, some passive, some active. They needed to model that flow to move the fish around.

One of our meshes was even on a US postage stamp with a picture from NASA. It's a $14.50 express mail stamp so you've probably never seen it.

I asked John whether the meshing in CFD was all that different from the meshing in finite element analysis. He told me that their software could mesh for solid mechanics if we wanted to and that they actually have some customers that do that. However, they don't have all the materials libraries needed to a real analysis and measure deflection load or anything like that. Their main focus is fluid dynamics: planes, trains, and automobiles, as the saying goes.

But no, there's nothing that we couldn't mesh. In fact, there is a meshing competition at the International Meshing Roundtable. That's the annual conference for meshing geeks.

As an example, Pointwise's entry was the guitar at the top of this post and the abdomen above. Read all the details in Meshing the 24th IMR Fender Jazzmaster Guitar and The Abdominal Atlas. They won the Meshing Contest Award with these two entries that year.

Videos

Here's the video of how Branch Technologies uses Pointwise meshing to construct large architectural-scale 3D printed structures (30 mins).

Watch the video of a webinar in which Amine Ben Haj Ali, Senior Engineering Specialist in Advanced Aerodynamics at Bombardier, details the Meshing & Adaptive Re-meshing Server (MARS) he has developed, which has been used by engineers at Bombardier to generate over 250,000 high-quality meshes since 2017. MARS automates the meshing process, reduces meshing time, and ensures consistent meshes across geometry variations no matter who is using it.

CFD Blogging

Pointwise has a blog called Another Fine Mesh and almost every Friday John writes a post called This Week in CFD. We have created a Computational Fluid Dynamics blog here on the Cadence website. For now this contains stubs from both the Pointwise blog and the NUMECA blog, plus a little original content around the Celsius Thermal Solver, which also uses CFD under the hood to measure airflow in electronic systems.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.