0:01 Thanks, it’s good to be here today. I’m Matt Cedarberg from Coreform and I’ll be talking about bridging the CAD simulation gap. Before we start though, I’ve been really inspired by a lot of the presentations the last two days and and especially the one from Ryan McClelland from NASA. So, I wanted to share my favorite NASA anecdote. In 1995, Robert Williams was given permission to take 11 days of the Hubble Space Telescope to stare at a tiny sliver of the sky where there appeared to be nothing.

0:53 And then when that was developed, it turned out that there were thousands and thousands of galaxies in that place where there appeared to be nothing. I I love that application in so many areas of my life, perhaps none other than the fact that my entire career I seem to have been drawn to people who stare at tiny slivers of CAD models, not just for 11 days but for years upon years.

1:20 The first person in my life who did that was my dad, Dr. Tom Cedarberg, who was a professor of computer science. He’d come to my soccer games and people would come to see what statistics he was taking on his yellow legal pad and they were disappointed to realize that he was only doing NURBS math. So, here’s a picture of him doing NURBS math on the chalkboard. The the problem that increasingly drew his attention later in his career was the foundation of NURBS.

1:48 It turns out that these these surfaces that are used to define all complex CAD shapes, if you trim a NURBS, which is done all the time, and then combine two NURBS together, mathematically they are not watertight. And so, this is illustrated in this example of the famous Utah Teapot, which actually leaks. And so, for two decades he worked on this problem and came up with a solution that he called T-splines.

2:15 T-splines are fully compatible with NURBS, and you’re able to create mathematically watertight models of any complexity. And so, he made that invention when I was just exiting college, so we decided to start a little company together called T-splines. And we made plug-ins to Rhino and SolidWorks, and then eventually Autodesk bought the technology, and it’s it’s used to make generative design for products such as those in Fusion 360, like something like this this bracket here on the stage.

And again, the key to success there though for us was to in to introduce a new innovation in a way that was compatible with the existing representations. And so, for the last 10 years of my career, I I found some other stairs of CAD geometry, this time more than one, and in fact, an entire field of researchers who are been in this field called isogeometric analysis. Now, this is led by Dr.

3:15 Tom Hughes, who’s on the left from UT Austin, and with him is Dr. Michael Scott, one of my business partners, who was Tom’s student. But they what they were staring at since night since 2005 was this realization that everyone’s had, including my own wife, every time I tried to describe to her what I do for a living, it makes zero sense that the foundational CAD tools and the foundational FEA tools used in our industry are based on completely different math.

3:43 So, you you can take a long time to create painstakingly the exact CAD design that you’ll manufacture, and then the very next thing that you do when you’re wanting to predict how that manufactured object will behave is you change it. So, you reduce all of the details, some of which actually matter, to create something that is actually meshable. And that’s very painful. Sandia National Labs, who we work closely with at Coreform, years ago published this this study that that Jan referenced earlier from Entop that meshing and model prep consume the vast majority of simulation time.

4:20 And as we’ve been talking with customers over the years, this this percentage is not going down because the complexity of the models continues to go up. And I I think there’s been some variation of this slide in most presentations this conference that this chasm, this separation between simulation and analysis, which is largely driven by the different geo- geometrical representations, has really limited the adoption of analysis over the years.

4:47 And now that is becoming even more problematic as agents are coming online and it’s just it’s just an issue to to to have agents that can actually simulate on on the actual CAD design with this meshing process. And and there’s really only kind of two options that have emerged over the years and they they each have their own tradeoffs. And so, hexahedra are more accurate, those are definitely preferred, but they can be very time-consuming to create and there is no way to arbitrarily create hexahedra on on any arbitrary geometry.

5:22 So, so many simulations are on tetrahedra, which can be automatically generated, but they come up with their own issues. If you keep the geometry exact, they can be very expensive to compute and then with poor condition, they can also lead to other robustness issues. At Coreform, we actually bought the commercial rights to commercialize the Sandia Cubit technology so we could really understand this problem deeply before we tried to solve it.

5:50 So we’ve we’ve been digging into that world as well. So what many of our customers find themselves in is this this repeating loop of defeature, meshing, repeating that almost like using shampoo, but this isn’t takes minutes. This takes this can take weeks each time you run a different categorization which can prevent simulations from being run on multiple iterations and limit the the benefits of analysis. So with this field of IC geometric analysis, Tom Hughes, what they discovered from from their years of staring at these these these gaps in CAD was a way to actually create a new methodology that is still finite element analysis without the manual meshing process where you can analyze directly on the geometry that you designed and not in approximation.

6:39 And so it’s illustrate this on this fully detailed CAD, let’s just zoom in a little bit on the on the caliper of this brake assembly to to highlight what this what this allows. So the traditional finite element method, this I I mentioned we we bought the Cubit meshing technology and so one of our the best Cubit modelers in the world took him 20 hours to defeature and hex mesh this brake caliper and he actually didn’t even finish.

7:02 He just ran out of time. And then running IC geometric analysis on the same part it took under a minute to actually run the simulation. And you can see this is critical because where the stress concentration lies is on the same fillet that needed to be removed in order to create a hex mesh. And so when you’re creating these hex meshes, then there isn’t a chance in the likelihood that there the simulation may be impacted by the reduction in fidelity.

7:32 Just a reminder of what it means to to trim. So surface trimming is what enabled all of CAD and that’s because these nerves patches are all rectangular, so to create non-rectangular shapes you just cut them out. And then to create more complex NURBS representations, you you fit them all together. And these are called boundary representations because they’re hollow in the middle. So, at Coreform, what we’ve done to bring isogeometric analysis to industry is just extended this very simple idea of surface trimming to volumes.

8:01 And so now we actually have a full fully volumetric parameterized model that we run finite element analysis on. So, what this process looks like is you take a fully detailed CAD model, you immerse that in a a Cartesian or higher order smooth spline grid. And then we again we just do a Boolean operation between the the grid and and the CAD objects. So, now we have a volumetric CAD that we run simulation on.

8:32 This is a very robust procedure. Again, we compare this now with robustness against the tet meshing in Cubit. This is a more robust procedure than even tet meshing. And the reason being is that like we were noting earlier, there are gaps in all CAD models, but if we don’t need those to be completely watertight to within any given tolerance, we just need to be able to determine the inside outside to be able to run these Boolean operations to create the CAD, and then go ahead and run the full fidelity simulations.

9:02 So, the the the critical lesson that I learned with with my with my prior company was to really drive innovation in industry and drive adoption. The the way to make life easiest on everyone involved is to keep compatibility to the greatest degree possible with the existing tools. There’s plenty of of interesting innovations that that I’ve seen come out that that are very exciting, but if that solves one part of a 15-part workflow, then it’s just difficult to get commercial adoption.

9:30 So, at Coreform, we’ve we’ve developed isogeometric analysis solvers on our own to prove the concept, and now we’ve begun integrating these also with with existing solver codes. We’ve integrated with a number of government codes, and then what we’re excited to announce at this conference is that we’ve now integrated this into Abaqus, which is one of the the the leading non-linear finite element analysis codes. And so, you can see the same procedure that we have that we’ve developed in the penalty is now available using the Abaqus solver.

10:01 The workflow is is so simple. It’s actually simpler than the traditional Abaqus workflows. You begin by importing the fully detailed CAD, and then you have all sorts of options available to you. You can simulate every every part of the CAD that you brought in, or you can apply traditional tried and true finite element techniques. So, you can assemble the assembly, perhaps. If you want to remove the the bolts and replace those with connectors, you can do things like that.

10:27 And then you apply the same loads and boundary conditions that you would with traditional finite element analysis, except when it comes to the meshing, you keep the fully featured CAD, and you just immerse every object in a bounding box. This is fully automatic. It can happen with Python, or it can happen in the GUI. You can control the density of each of each of each object. This is what the IGA mesh looks like.

10:51 Again, it’s it’s this immersed generated mesh with cut cells on the boundaries. And then we go ahead and run the simulation. We can again post-process that using standard Abaqus workflows. This is made possible because of the the the isogeometric analysis is a finite element method. And so, we we do the setup in Abaqus CAE. We create the core for my G mesh, and then when we create the mesh, we use the Abaqus user element, which is a very robust part of their solver.

11:25 We call to Abaqus throughout the solver. We use the Abaqus materials, and we contribute the element stiffness through through the IGA element that we’ve added. And so the capabilities that this opens up the door to is basically all of the Abaqus. And so for example, we have linear stress linear static stress. We have thermal expansion. Metal plasticity. Large deformation. Nearly incompressible plasticity. Mechanical contact. Frequency extraction. And one of the one of the powerful things about this is that when you’re adopting a new technology, if you’ve been using finite element analysis for decades, there’s things that you trust.

12:14 And so as as customers are beginning to use this, you don’t they don’t need to completely throw out traditional finite elements. We can do mixed FEA IGA simulations. And so part of the model can be remain as FEA, part of it can be IGA. We find that the value of IGA is greatest the more complex the physics and the more complex the geometry of the part. And so you that’s an option to have the most complex complex parts be IGA and then leave the rest FEA if that increases your comfort level.

12:44 Another unique thing is that everything that we’re doing in IGA is based on solid models. And so often in industry when there’s thin sheet metal parts, then those are used as shell elements and the mid-surface of the CAD is taken. That’s also a process that can be problematic and again you’re not simulating on the exact CAD. So what we do is we just take everything as a solid.

13:07 And because we have these higher order smooth spline elements, we can get away with fewer than a single element through the thickness and it still resolves beautifully. This is also very nice when there’s more complex cast parts such as giga cast parts where it’s a little bit unclear is that should that be a solid or should that be a shell because it’s one complex car part that that morphs together.

13:28 And again, that’s just a single solid part in in with Coreform’s IGA technology. And of course, we can run assemblies as well. And so at this point, from that from Tom Hughes’s group, there’s been thousands of published papers that have shown that IGA can do everything that FEA can do, but better. And and now we have this available in traditional commercial finite element analysis software. So, the future of FEA is here, and surprisingly, your CAD is already ready.

14:03 There’s no need to change the CAD, and your solver is ready. So, the question I’m leaving with with is are you? I I hope you are, and I hope you’ll come talk with me, and we can figure out how to get you set up to run isogeometric analysis on your on your CAD data that you have today. Thank you. I need to do a simulation with our less than one unit.

14:24 Is it what’s less than one cell? Yeah, so with whatever the with whatever the bounding boxes you use, fill it with however many elements that you want. And because these are high-order elements, we have additional quadrature points inside the elements, and so you don’t need you can have less than a single element there. So.