Behind the Scenes of Autonomous Meshing: An Interview With Kelly Senecal

If you’ve ever talked to someone who works at Convergent Science, you will undoubtedly have heard us extolling the virtues of CONVERGE’s autonomous meshing. Got a complicated geometry? No problem! Moving boundaries? Easy! No time to waste on meshing? We’ve got you covered!

This enthusiasm, we would argue, is not unwarranted—CONVERGE’s autonomous meshing strategy was truly a novel innovation. So much so that when CONVERGE was first released, the Convergent Science founders were met with more than a little skepticism. As Convergent Science Co-Founder Kelly Senecal puts it, “Nobody believed us.” The founders had to prove the worth of this new feature, asking companies to provide their hardest geometry so they could see for themselves that, in a matter of minutes, the geometry could be up and running in CONVERGE.

Fast forward 16 years, and CONVERGE’s autonomous meshing has become the industry gold standard. As other CFD solvers are releasing their own versions of automated meshing, I wanted to find out what it is that makes CONVERGE’s autonomous meshing different. To do so, I sat down and talked with Kelly, one of the original developers of CONVERGE and, arguably, the number one fan of autonomous meshing. 

To start off, what exactly is autonomous meshing?

Autonomous meshing is truly automated meshing, in the sense that the user just has to supply a few parameters in the user interface, and all the actual meshing is done at runtime by CONVERGE. So it takes the meshing completely out of the hands of the user. You still have control over the mesh, though. As a user, you can define fixed embedding regions if you know ahead of time that you want to have fine resolution near a boundary, for example. CONVERGE also uses Adaptive Mesh Refinement—at every time-step the code is intelligently figuring out where mesh is needed and where mesh can be removed based on the flow physics to be very efficient with the cell count.

What prompted you, Keith, and Eric to develop an automated meshing approach?

We spent a lot of time during our graduate school days, and during the early days of Convergent Science back when it was a consulting company, making meshes for people in a code called KIVA. And even though we could do it relatively quickly—we had created tools to help us—it could still take days or even weeks to make a mesh for a complicated geometry. And when you’re an engineer, you want to spend your time running your CFD simulations, analyzing results, and using them to make design decisions; you don’t want to spend all your time making meshes. So that’s what motivated us. We thought there had to be a better way. 

What was the process like writing the code for autonomous meshing?

That’s a good question. Scary? Because we didn’t know if it was going to work or not. And we had some missteps. We originally based the code around an immersed boundary method, as opposed to the modified cut-cell Cartesian approach we use now. We thought the immersed boundary method could work, and we got something running—not quickly exactly, it probably took a year and a half to get the code up and running for a 3D engine simulation. But we realized it wasn’t going to work because it was hard to get that approach to conserve robustly. So we had to scrap essentially all the code we had written and go to this new approach. So that was a bit scary. And we still weren’t sure the new method was going to work. Originally, the automated meshing took minutes or even hours to create the mesh in the solver. We get that question a lot: “Doesn’t this take forever?” And originally, yes it did. One of the real breakthroughs we came up with was making that process almost instant. It adds very little time to the overall CFD calculation, even though we’re remaking the mesh entirely at each time-step. Once we had that eureka moment, we knew we were onto something big. And so it went from scary to very exciting. 

What makes CONVERGE’s autonomous meshing capabilities unique?

A lot of CFD codes these days throw around the automated meshing terminology fairly loosely, I would say. There are different levels of automated meshing out there, and how truly automated it is depends on a lot of factors, like how complicated your geometry is and whether or not you have moving boundaries. There are some codes that can do automated meshing for certain cases, but it’s really hard to be able to have automated meshing work in general for all cases. But that’s what we have. We have yet to find a case where we throw a geometry at CONVERGE and it isn’t able to mesh it. And so that’s what makes us unique—we have truly autonomous meshing for every case, no matter how complicated the geometry or the motion profiles are. 

What kinds of practical benefits do companies see as a result of CONVERGE’s autonomous meshing?

Really it’s about efficiency. Maybe you have a design out in the field that is having problems, and you want to use CFD to figure out what happened. Or maybe you want to design a brand new flow device from scratch using CFD. In the past, you’d spend a lot of time just making your mesh. And once you’ve made your mesh, the next question is, “How do you know if it’s fine enough? How do you know that you’re grid-converged?” It’s very hard to answer that question with traditional meshing techniques, because it’s so difficult to make the mesh in the first place, you’re probably not going to want to make another one. With autonomous meshing in CONVERGE, it’s very easy to make multiple meshes and show grid convergence. So again, it makes the process much more efficient. It also gives you confidence in your solutions because you can very easily double or triple the resolution and see how that affects your answer. Of course, you still have to run those simulations, so that takes some computer time. But the actual engineer time is minimal. So it’s much more efficient, you’re more confident in your solutions, and you get more accurate results. And in the end, that leads to a better design.

What applications benefit the most from autonomous meshing?

The ones that benefit the most are cases with complicated geometries and moving boundaries. That’s what’s hardest to do traditionally in CFD, and most real fluid devices are complicated. There are approaches that can handle moving geometries, but a lot of them add numerical error because you’re deforming the mesh near the boundary, for example. Whereas with our autonomous meshing technique, we recreate the mesh at every time-step while the motion is occurring, so we avoid those numerical artifacts. Autonomous meshing can also handle large differences in scales—maybe you have really tiny channels in your geometry as well as very large areas. CONVERGE can handle those very different scales efficiently, automatically putting fine resolution in the small channels and very coarse resolution in the large areas. So varying scales, complicated geometries, and moving boundaries benefit the most. But again, even simple geometries benefit because you’re not spending any time making the mesh. 

Are there any new meshing features currently in the works for CONVERGE?

In version 3.0, we released something called inlaid meshing. It’s not required for any simulation, but if you want to add a boundary layer mesh or a non-Cartesian mesh in a portion of your domain, you can do that through inlaid meshing. We already have all the tools implemented in the code to read those meshes and have them interface with the traditional cut-cell Cartesian mesh. What we’re working on now is automating the inlaid mesh generation similar to how we automate our traditional CONVERGE meshing. When this feature is implemented, the inlaid meshing will also be fully autonomous. 

If someone is interested in trying autonomous meshing out for themselves, what should they do?
Reach out to us! We have a variety of licensing options available, whether you want to use CONVERGE for commercial purposes, to conduct academic research, or to learn a new skill for your resume. We also offer on-demand licensing and access to computing hardware through our cloud computing platform, CONVERGE Horizon. We would love to work with you to find the right license for your needs so you can experience the power of autonomous meshing for yourself!