Thanks. Okay. I’m excited to be here with all of you today. My name is Brandon DeMille representing General Atomics Aeronautical or GAASI for short. For those who don’t know, GAASI is a world leader in unmanned aerial systems primarily for the defense industry. And today I’m here to talk about topology optimization in our airframe design process. All right. So, quick outline of the talk. We’re going to start with some introduction and context on our workflows.

0:55 And then we’ll get into an example problem. And the output of that will be fuselage mass estimates. We’ll use that data and we’ll zoom out and look at it in the context of a wider trade study scoring effort. Then we’ll talk about some conclusions from that and some future work to continue to improve on this. Okay. So, airframe design process. This is a very high-level overview. Many of you probably are very familiar with this, but I think it’s good to level set.

1:29 Initially, we will start with some kind of a proposal. If things go well, we’ll get a contract award and that will come with mission requirements for the aircraft. GAASI will then convert those into aircraft requirements. And then at that point, we can begin pre-conceptual design and MBAO efforts followed shortly thereafter by conceptual design. Now, admittedly, at this stage in the design process, there’s a bit of churn sometimes.

1:56 We may go back and revisit requirements. We may recycle some ideas and that’s okay early in the process. But what we don’t want to do is let that linger. We don’t want to have that kind of issue later in the design process. So for completeness, let’s just get through the rest of the the process here. After conceptual design, we’ll go to preliminary, critical design, and then eventually on to full-scale production, right?

2:22 And what we’ve found is that it’s really worthwhile to take a little bit of extra time and effort at this stage and do a bit of a trade study to look at all of our material and manufacturing options to make sure that we’re confident and move forward for the rest of the design process. And that’s what we’re going to talk about today. So, the idea here is we’ve got a we think of it as a funnel, right?

2:49 So, we’re we’re starting at what we call qualitative trade study level. We pull together all our subject matter experts. We read them the requirements. We throw everything into this top of this this funnel. It’s broad. We’ll take anything here, right? And at this stage we have too many options to dive into detail on every one. So, we’re going to use engineering judgment, experience qualitative methods to whittle that list down to something more manageable, one or two dozen options.

3:22 And that’s where we enter what we we call the quantitative portion of the trade study. At this point, we shift gears. We’re able to spend more time per concept and so we rely on data at this stage, right? One example might be our supply chain team goes and gets quotes or lead time estimates or whatever, right? The the analysis team might do some preliminary analysis. The designers will start doing a little bit of design work.

3:46 We we’ll use data to move on from here. And we’ll whittle that list down even further to one or two, maybe three options, and then that’s where we begin preliminary design, right? And what we’re going to talk about today is kind of giving a boost to that quantitative portion of the study. And we’re going to use design optimization methods including topology optimization to add some important data to that decision-making process.

4:14 And as we already kind of mentioned, as you move down this slide or through this funnel, we’re reducing the concept count, and we’re increasing the amount of design detail we’re allowing ourselves to do for each concept. Okay, so with that in mind, here’s a example. This is our example airframe. It’s kind of a mashup of features from past programs and some things that we’re interested in the future.

4:38 It’s a class 3 UAV. It’s meant to be air-launched. There’s no landing gear. It’s got a payload bay. It’s got locations for a payload to be mounted, ducting, space claim for an engine, but it’s got all the basics that we need. One other feature that it does have is it’s got a constant cross-section in the center there, where the where the wing roots attach. And that was something that our conceptual design team was interested in.

5:06 So, how do we even how do we start this? Of course, we need the output from the qualitative portion of the study. We need a list of things we’re going to evaluate, right? We already talked about that. We want to include packaging and system integration considerations. We we just briefly touched on that. That’s all the the things like the payload bay size and shape, the payloads we want to consider.

5:30 If you want to include fuel volumes, you could do that. We don’t want to we want to be careful here not to over-constrain ourselves. We want to include reasonable things, but we don’t want to go too far. And then we don’t always have all the details yet. We need manufacturing constraints for each of the manufacturing methods. So, this is where a computational designer needs to really interface with the manufacturing team so they can understand what the limitations are and roll those into the FEM that they’re going to create.

5:58 One of my favorite parts about this whole process is it helps to pull people together from across our whole organization early in the design process. This can’t happen without inputs from a whole bunch of different people. And along that line, right? Load cases. We need all the critical load cases. Now, a lot of airframes at G A S I eventually get evaluated against hundreds or thousands of load cases.

6:25 At this point, we can’t do that. It’s too computationally expensive and often we don’t have all those details ready yet anyway. So, we need to rely on our loads team, our analysis team to really help us focus those load cases. And I think even still, it’s important to remember that we are going to operate using a subset of all the load cases that will matter eventually, right?

6:48 And then in response to those load cases, we also need our structural requirements. Now, typically, not always, but typically it’s not enough to just say we need our structure to survive these loads. Sometimes that’s enough, but usually we need some kind of structural performance in response to these loads, right? What is the maximum deflection allowed? What’s a modal frequency that we need to avoid? What is a torsional rigidity that’s needed to to stay out of a flutter issue?

7:18 All those kinds of things. And if we’re not used to designing this way, this can be sometimes a challenge to collect all those those requirements, but it’s important. And and sometimes we have to go back to previous models or previous designs that have been successful and extract those from there as a reference. And then the last item I’ve included here is is new for us where we’re we’re including some aero performance based on the OML shape that can be rolled into this at the same time.

7:48 And so one way we’re doing this is we took that that constant cross-section portion of the fuselage, we parameterized it, we ran CFD, we did a DOE, and so now we can have a drag as a function of those shape variables. And that allows us to put an upper limit on drag and still fine-tune the shape of the the fuselage while these topology optimizations are running. Here’s a super quick example schematic showing how the loads are applied to our fuselage.

8:21 And here’s an example of rigidity constraints. We have torsional rigidity and then flexural rigidity of the fuselage at different stations along the length. Whereas it’s just an example. On the left-hand side you’ve got values at the nose and at the right-hand side you’ve got the those at the rigidity at the tail. Okay. I told Joanne I have to at least include one of these. I love these.

8:45 But I’ll only make you watch one slide worth of these. And I think this is helpful to show all the different methods that are being used in parallel. On the left you’re going to see 3D topology optimization of the substructure. I’m sure everybody here is very familiar with this. You’ve seen this kind of thing before. Ooh. Little pixelation going on. That’s okay. On the right, on the upper right, you’ll see skin thickness optimization.

9:10 The the color contour plot represents skin thickness. And then on the bottom right, even though it’s kind of weird looking right now, you can see both of those things happening at the cross-section. And then if you you can also see the shape of the fuselage changing as well. Right? So this is just a way to illustrate all the different methods that are being used simultaneously in response to those load cases and those requirements.

9:36 Okay. So once that’s done, we end up with results like this. What the one we were just looking at is a laser powder bed fusion substructure with a variable thickness composite skin. Now, once we have that up and running, we can quickly change materials and manufacturing constraints and generate more options. And of course, all of these come with mass values, right? We’ve got machined aluminum substructures in a in more of a traditional bulkhead.

10:05 We’ve got wire DED with a little bit of a simplified structure there. We’ve got cast aluminum. And you can see even though none of the settings were changed for the skins, those are changing as well in response to the the substructure. The skin and substructure have to work together. The next three options are cases that we wanted to evaluate where the skin and the substructure are the same material.

10:29 And this is interesting because then we can potentially simplify assembly and reduce mass by removing the joints that connect the skin and the substructure. Of course, there are downsides to this as well. There’s only so thin you can go on say something like wire DED. And so those those skin thicknesses, the lower bound is not as thin as others and you you’re going to have a mass trade-off there, right?

10:53 And then on the far right, we have a few options where we purposely simplified the skins and we said, “Hey, we’re going to break this up into simple panels and the panels can be whatever thickness they want, but each panel is a is a constant thickness panel.” We have a composite version and a and a stamped version. Stamped aluminum. Now, those are probably going to be less expensive to manufacture, but as you can see soon, they’re not as efficient.

11:17 So, that’s a trade-off that the the program has to make. Now, looking at all of these, you might say, “Well, yes, this is interesting, but I don’t know if we’ve really captured all the DFM constraints that we need to capture, right?” And to that I would say I think you That’s true, and that’s why we have refinement steps. So, I’m going to go through a couple examples of the ways we refine these models.

11:43 On the left, you’ll see the same result that we just looked at for the machine bulkhead option. We take that initial result, and we use it to reduce the volume of the design space, and then that allows us to afford to reduce the the the element size, which gets us a higher resolution answer on the right. Okay? Without it getting all too all without a avoiding excessive computational expense.

12:13 And with any accident I’d like to point out you can see kind of a traditional wing carry-through bulkheads, and also tail carry-through bulkheads there. Here’s the investment cast aluminum option. Again, on the on the left is the the solution we already saw. We take that, make it a make it into the design space for a subsequent step, and and then we end up with a higher resolution result.

12:39 That may be a difficult-looking structure to cast, but I got to say in my in my career I’ve seen pretty complicated wax injection molding tools. And of course, if we had to you could I could always be printed wax patterns as well. That all affects the cost, and that can be evaluated separately. Wire DED option. So, this was another refinement method where we take a simpler-looking result on the left, and we shell that, we core it out, and we do a shell thickness optimization, and that’s what you can see on the right.

13:15 There’s a little bit of 3D structure that’s built in at the wing root there. And it can go on and on, right? So, these are tricks that we do, and there’s different ways to do this for other manufacturing methods as well. Once that’s done, you got a mass roll off. So, we’ve included here some some details of the smaller bars at the top, but the main things are you’ve got substructure in orange, and you’ve got skin skin mass in light blue and dark blue, right?

13:43 Couple things to note, design freedom in the substructure is a big deal. Powder bed fusion’s doing great here. Cast aluminum is also doing well. And then another trend that we can pull out of this is just that like I said earlier, the design freedom in the skins, and a sub- significantly thin mean gauge for your skins is it turned out to be important here. So, we’ll take all these mass values, and we can convert them into a score for airframe mass, right?

14:20 And then I think it’s important here to zoom back out. The The program has to decide how important is airframe mass. Airframe mass isn’t everything, right? In my job, it’s it’s important, and it’s what I focus on generally, but there’s other things that they may need to worry about. They need They care how much it’s going to cost. They may care about capital investment, supply chain, how much technical risk is there, right?

14:45 So, the program team has to decide along with the customer how much each of these things will weigh. All right. So, let’s get into conclusions here. I’m getting a 2-minute warning, so we’ll wrap it up. So, this is really just a combination of really well-established methods that I’m sure everybody here has has seen before, but I’m hoping it’s it’s at least somewhat helpful to see them in this combination and used in this way.

15:15 It’s it’s been very helpful for us in terms of quantifying these trade-offs between materials and manufacturing options. We’re able to cover a variety of methods relatively quickly. The multi-step optimization is a key feature that helps us define manufacturing constraints. And we’re really trying to throw out the bad ideas early. We’re not designing the whole thing at in one shot. And there’s still plenty of room for improvement here.

15:43 I feel like I’m in the right place, right? We need to accelerate this. So, AI-augmented structural optimization is one area that I’d love to see us move into. Direct inclusion of trade study metrics like cost would be great to add. Again, we’re in the right place. We have already We just talked about this this morning. Deeper integration with our MDA team. We’re adjacent to them. They helped contribute to this talk.

16:09 And And I’m not exactly sure how this is going to turn out, but we’re we have to figure that out. Automated workflow, right? This is This is more efficient than designing everything for every combination, but there’s still lots of automation that could happen either via scripting or AI agents. Multi-physics and multi-functional structures could be included in this kind of thing. And then, of course, partnerships with leading providers, people like you, it would be immensely helpful. And I look forward to talking more with all of you at the conference. Thank you very much.