0:00 All right. Thank you, Duann, and thank you. It’s really it’s awesome to see everybody. And so, I kind of want to start the presentation just on a little bit more like personal note in some ways, which is like, you know, a lot of times I I often get asked like, you know, what’s our what’s our ultimate goal with nTop? And to kind of touch on that, kind of have to go back to what I was really interested in since I was a kid.

0:24 And basically since I was a kid, I had like two obsessions since like eight years old. Airplanes and CAD. And you know, maybe it’s a strange thing to to like, but so for with nTop, you know, my goal since the early days of NTOP was really to see a computer be able to model an aircraft from requirements all the way to design in one parametric aircraft model.

0:49 U and that’s because you know start to finish every surface every interface every constraint and you know that might sound like a niche goal with nTop but I think it actually you take a step back you know in aerospace that’s where we model it’s like one of the best examples of like physics driven performance-driven design you know we operate in a you know these are these are physics driven surfaces they have to operate at a high performance other safety critical right it’s about integration certification everything it all matters and if we could solve the problems in aircraft design we can solve those problems in other industries as well and so aircraft design has really from the early days been been a serious goal of mine and it’s also where some of the CAD and simulation tools that I’ve looked up to since my early days it’s where they began also from KIA to NX George Allen with us used to work at McDonald Douglas when they were developing when they owned NX actually etc.

1:55 And you know, with with today when I look at aircraft design, those current tools that I’ve looked up to since I was a kid, you know, they’re making it harder and harder to deliver on these designs faster, right? Because engineering is where kind of aircraft engineering is where ambition gets pushed to the limit. And what’s happening in the industry is really kind of two things. One, the design cycles are tightening up, right?

2:27 And you know what’s happening is programs from the DoD, programs in aircraft design that used to have five years to get to field a new aircraft, we’re being asked to deliver in 18 months. And this is without losing performance, without losing certainty. And this is a really tough shift. And especially because a lot of the engineers that I talk to and speak with and meet with every day, they entered this industry to really push the limits to push the boundaries to deliver on the ambition and the the the reason they entered this industry was to deliver that.

3:00 And it wasn’t just about like delivering fast and shoddy performance because aircraft design you can’t deliver shoddy performance, right? It’s about safety. It’s about delivering things right. And this becomes hard because of something that I’ve heard referred to as the lock-in trap, right? And that’s in the early stage of development, decisions that you make have massive, massive budget, safety, and time implications over the course of the program.

3:25 And the compressed timelines today now have kind of pushed the current tools that we use over the edge. And that’s a big problem. And it’s it’s not just a design problem. It’s a it’s a timing problem. And you know it’s this lock-in trap that’s really c pushing teams to fall into local optima early on which basically create all these like massive budget implications downstream. And one example of that in the old world is the F-22.

3:57 So if you look at this is an image I found of the design changes from the YF22 to the F-22. And if you look at it right like the the inlets had to move backwards on the aircraft from the YF22 which was the prototype version that was fielded it was six years 14 changes and today with those compressed timelines we don’t have the time to to make these changes and you know this is a problem this lockin trap is a problem in aviation but I would expect this to be a problem in automotive in industrial and consumer.

4:36 We see this across all of our customers. And you know, it’s wherever that we’re seeing the kind of engineering ambition run up against the complexity, this lock in trap. It’s like sitting down there kind of waiting to rear its head, right? And it seems see the core problem seems really really simple, right? Which is it takes too damn long to go from requirements to field of design.

4:59 And it you know, if I if I just like take a step back, it should be really easy, right? Have an initial design, make some updates to my geometry, run some simulation. Has it converged? Maybe not. Let me go back and change the geometry. But but there’s issues, right? The whole process is serial. It’s run. And then when I do make those updates, those updates are manual.

5:22 And uhoh, my model turned red. Now I need to go in there and find where the missing con edge was that disappeared that I was referencing downstream. And so that takes some time. Then once the CAD model is updated, oh wait a second, you know, now it actually doesn’t update at all because of the model’s too complex or it takes too long. But once I once I finally converge on some shape, some geometry, then I’m going to run my analysis.

5:52 But of course, meshing takes time, right? I need to make sure that the Jacobian value of my meshes is correct. So I’m not I don’t have too skewed of elements. The simulation is expensive for some of our aircraft customers. They run a full flight envelope and it takes them like six days for one design point, right? If we’re going to iterate through thousands of different variations, that’s a problem.

6:16 If it takes that long. And so at nTop to address this problem headon we basically put together a workshop and this is what I want to talk about just for a little bit. And basically, you know, we we took it we took the time to step back and talk to some of our aerospace customers to say, okay, what is the the core problem? And you know, how can we solve the geometric bottlenecks and the simulation bottlenecks to allow them to deliver faster?

6:50 And so, you know, with at NTOP in the early days, designing NTOP to push the 3D printing, to push the lattises, to push the heat exchangers, it it really pushed us to develop the most advanced, most powerful modeling technology on the planet. And you know, it was our belief that we could go in and actually apply the same modeling technology to aircraft design problems. And that was what this workshop was for.

7:14 We wanted to see in one week, could we actually deliver a full requirements to design system? And so we brought together some of the world’s experts on this. We brought together George Irving who actually drew up the YIF22 that I was showing before. He was from Loheed. He was a senior configurator there. And Yan Von debrand program manager at DARPA as well as leading the modeling group at Boeing.

7:36 We brought them together, brought our engineers together and you know this was the intersection of aircraft experts, our modeling experts, our simulation experts for one week teaming up for this challenge. And you know I was excited for this not just because of the you know the the compressed timeline but it was also a chance to like put our tools up against some of the most challenging hard problems.

8:07 Across design, right? In aircraft design, you have to represent, you know, conic lofts and all the different surface types. You have to run 1D, 2D, high fidelity analysis, etc. And George and Yan showed up with deep aircraft experience. Our team brought, like I was saying, the most powerful modeling technology out there. And together, we pushed each other over the course of this week. And it started with an idea or a drawing, right?

8:32 This was a sketch that George Irving made of a of a two seat fighter aircraft. And we started from this sketch and we set out to build blocks that could represent a parametric lofted fuselage, a parametric two panel wing, parametric weapons bay, sensor bay, u rubber engines, right? A rubber engine is kind of exactly what you would think it is. It’s an engine that you can stretch out in size based on performance parameters.

9:02 And most of the week we spent developing those blocks for this initial configuration, but we threw in a sec second configuration just to push us to the limits. This is like a CCA like unmanned drone. And the idea was okay, could those same sets of blocks that we built in Ntop be applied to the same type of problem but a different species of problem. And so, you know, most of the week was spent building these initial blocks.

9:31 At the end of the week, in about an hour, we were able to reconfigure those blocks to hit hit on this second configuration. And so, it wasn’t just about faster modeling throughout the week. It was really about better thinking, more exploration, more insights through the process. And, you know, it was really this is kind of what happens when we take, you know, ambition on from our team and experts in the industry, engineering and technology and align around a core problem, right?

10:06 Gathering the requirements, parameterizing up the model, exploring the design space. And it really starts with, you know, how you actually parameterize that model. And so here we have a fully parametric fighter-like vehicle and nTop. You can see the wing go through a few different changes of the configuration. Here we have a lofted fuselage. And so what I’m really proud of from this week is really figuring out between all of us, how do we actually use implicit modeling to generate a correct lofted wing with dihedral with twist with correct air foils at different sections?

10:39 How do we loft an inlet duct? How do we size an inlet duct to match some cross-sectional area target that we have? So, this is actually I don’t think you I don’t know if you can see the red lines under here, but basically that you can update and edit the green curve to have a target cross-sectional area throughout your shape. You could do this on a fuselage, you could do this on an inlet duct as well.

11:01 And then the offset function, the variable offset function is there’s a there’s a design parameter that’s optimized to try and match that cross-sectional area along the duct. You can connect that model to different analysis methods. Maybe it’s one more simplified panel methods for CFD. One of our team members connected X foil into NTOP or you can connect to Vorlax solvers and stuff like that. But I think what was most impressive was those those same blocks were then just reconfigured into the CCA vehicle in like less than an hour put together.

11:38 And it’s a full again fully parametric edit. We can edit the dihedral of the wings if you wanted to move the ducts, the inlet duct. So you instead of having an upper inlet duct, if you wanted to change and see those inlet ducts going on the side, that’s just a block in nTop that you can drag in and move that. And so the blocks for those that are familiar, how many are how many of you guys are familiar or use nTop?

12:03 Okay. So for those of you the same block system that you’re used to, you can use to capture things like a two panel wing, a lofted two panel wing with input in inputs for the air foils of the different sections, the location of that wing. And so this is kind of representing the, you know, earlier stages of conceptual design into early preliminary design cycles and speeding that up.

12:26 But for something more detailed like a group one UAS, so this is a much smaller vehicle sub 55 pound drone, this model we actually spent another roughly day on with those same blocks and just adding the detail needed to actually manufacture this component. And so here we have what I believe is the first requirements to manufacture design for a group one UAS built in an implicit model.

12:56 And you know this is made from about 23 different unique parts that come together. This is the airframe. There’s obviously motors and electronics that go into this as well. You can see all the different parts. If you are interested in the assembly instructions for this, my friend in New York has a company that you guys should look into called DRock. That’s Phil Arnstein. It’s really interesting. Each of these components is optimized to be printed with no support structure and you push a button from this and if you have a a couple cheap FDM machines, you can print all those different components.

13:32 And here we have George Irving holding his first version of this aircraft, which is awesome. And so, you know, I think this story it illustrates something even bigger, which is it’s not really about being afraid to fail anymore, right? It’s about leveraging kind of a more effective tool set and you know not burning up precious time and being able to to go through and make mistakes and and you know I’d actually be more afraid of not failing fast enough and not tightening up these iteration cycles not learning fast enough not getting the insights because at the end of the day as a software provider you know our tool should work with the engineer to help provide insights to move a program to move a product through that development cycle.

14:23 Ultimately to get the insights needed to close out a design because when you’re when you’re winning, when you’re showing up to win, you’re not just showing up with like a design that meets the requirements, but it’s the design that you have the most confidence in. It’s the design that you understand the most. It’s a design that you really understand what those parameters are going into the design, how those parameters affect the performance of that design.

14:46 And so back to the workflow for a second because I talked mainly about the geometry. But in terms of the design space exploration, some of the measurements and the analysis, I want to hand it off to Max who recently joined our team. Max created Cloud Fluid. So for those of you that know our new CFD fluid solver, just joined the team earlier this year and it’s been awesome to to work together on this.

15:11 So I want to hand it off to Max to talk about this section for a second. Thank you. Yeah, welcome. All right, exploring that design space. It’s not only about geometry and parametric designs, but also about physics and performance. So when we want to look at this design space, we really want to know what is the performance of that part and those set of parameters. And traditional design and simulations are not necessarily looped together.

15:44 They’re separate parts and because of that they’re slow process. So to truly optimize and automatically find very efficient and good performing parts early in the loop, we need to connect them and we need to close that loop tighten that loop. Why is that separate? It’s not by design. And I mean we’re not the only one following that vision to close that loop is traditional CFD methods do not support rapid iteration because they’re complex.

16:27 They’re hard to do. One of the first steps you encounter when setting up CFD simulations is meshing and it’s errorprone. It’s complicated and just a single cell that has a too large deformation can basically ruin your whole simulation workflow and you waited for days to to return something and it’s not there. It doesn’t converge. Solve times are not necessarily quick and not at least what we consider quick.

17:01 In academia it might be okay to wait for 15 days for a large ed simulation on a high performance cluster. In rapid iteration workflows this is not fast enough. So we want something that goes with the speed of of iteration of geometry and thirdly computational cost is quite high. You can scale to clouds. You can scale to GPU clusters. But as we know from AI mechanisms, time is money.

17:33 And the longer these algorithms run, the more costly they are. Let’s look into a faster, more scalable approach and that is introducing Ntop Fluid. What was previously cloud fluid. This is a GPU native CFD solver based on the Lettus Bolzman method. We can show that we can speed up traditional CFD run times by magnitudes, sometimes 100 times, sometimes 1,000 times faster. So, this is scaling down from days of simulation time to just hours, sometimes even seconds.

And yeah, one core driver in this is the lettuce boltsman method. It gets away from meshing. We use very simple voxal meshes that can be derived directly from Anttop implicit files. Yeah, high performance GPU power and seamlessly integrated. And also we use large ed simulations as a turbulence modeling scheme which is very robust and very generic. So we don’t have to adapt to certain flow phenomena. Let’s look at one very simple example.

18:56 We see this aircraft here, this preliminary design aircraft and we have this inlet duct. And in this situation we have a duct with two inlets and one engine at the end. And yeah, we know want to to know its performance. And one very important thing early on is whether this duct induces recirculation areas which would be bad if the engine was would suck in certain vortices, stand in vortices.

19:29 So this dock is driven by some spline parameters. And to show you just how simple it is to set up a flow simulation in Ntop, we have this block structure here on the left. It’s pretty much just linking Ntop existing objects to solver settings. In this case, we have the flow domain that we saw on the green slide earlier. We have inlet velocities and the pressure outlet just given as implicit bodies.

19:56 We define the fluid to be air. And the only numerical parameter left to state is the voxil size. And we have good practice for that how to choose that. And from one of these simulations you can extract recirculation areas or separated boundary layers. Because this is robust and it runs fast, you can very easily set up a DOE or an optimization workflow for that. Just to give you an idea, running this with 400,000 box takes about 10 seconds on a laptop GPU.

20:34 Running it with 4 million B box voxels takes about 5 minutes. So, this whole series here is set up and executed in about an hour. And this allows pretty much to iterate through CFD performance early and pretty much before validation in preliminary designs. Awesome. Thank you, Max. And so, you know, the idea with nTop is ultimately about keeping engineers in the loop, right? Because today with computational design and machine learning that design loop that I was talking about before that forward design loop right that’s changed right today it’s it’s okay to fail fast it’s okay to learn quickly you can refine that design with confidence over the course of design and so you know instead of getting one shot on goal or two shots on goal you can get dozens or thousands or hundreds of thousands and you In some workflows, our customers are pushing a button and running thousands of simulations overnight, training up these surrogate models and then bringing those surrogate models back into nTop.

21:50 So instead of iterating through to find that design, you can punch in, okay, these are the requirements that I want, give me that design that meets those requirements. And so the term that we’ve been using at NTOP is really co-intelligent engineering because at the end of the day, it’s ultimately about keeping the engineer in the loop. You’re defining precise highquality geometry, precise air foils, precise conic lofts, precise splines, driving lofts, precise duct, inlet ducts.

22:20 And it’s not about AI replacing the design intuition, but rather it’s about AI amplifying engineering intelligence with fast, accurate physics informed generation and faster iteration through the design. And you know I think this is really the shift right in in the past ultimately it was about how do we digitize drawing and drafting geometry and today it’s really about capturing intent and shaping behavior. And so it’s it’s no long co-intelligent engineering is not kind of working around the tools and trying to get hack the tools to do what you want but it’s ultimately about the tools working with you.

23:00 The geometry that you define just generates. The analysis is fast. It’s really quick helping you find that. And so it’s not ultimately just about faster modeling, but it’s a change in the way that you actually engage with your own ideas, with your own tools. And so, you know, whether you’re designing, you know, a next generation UAS or a turbine blade or electronics out there, you know, I think this shift applies to you and how you engage with those tools because at the end of the day, this ambition that I’ve been talking about, it’s not industry specific.

23:36 I mean, I think we all entered this industry because we want to make the next high performing product out there. And so, you know, whether it’s working on the next billion dollar advanced aircraft program or just rethinking a critical component in your system, it’s this ambition, I think, that still matters. And so, you know, it’s not just designing parts, but you’re defining what’s possible. And at the end of the day, we built nTop to help you execute on that and to help you execute on the full extent of your ambition.

24:05 So, thank you, Duann, and thank you, guys. And happy to chat for us today. To learn more about the CDFAM computational design symposium series, to see the archives of previous presentations, and to learn about future events, visit CDFAM.com.