0:00 Awesome. Hello everyone. Yeah, my name is Thomas Rees. I’m from ToffeeX. And I’m going to talk about this in vaguely. I actually had a different presentation planned where a bunch of cool examples from aerospace that we’ve been working on and then basically the embargo wasn’t lifted in time so I had to quickly rewrite stuff. But so first half is going to talk about metal amerospace and then the original idea was bunch of examples from aerospace but instead it’s going to be a bunch of examples from sort of aerospace.

0:34 And you can sort of like maybe make the connections yourself. Cool. This is Toffee X. We’re a company based in London. We make topology optimization software fundamentally for fluid dynamics heat transfer that sort of stuff. Except we try to do it in a smart way. We try to build a user experience which doesn’t suck. We try to allow people to iterate as quickly as they can. We give people the tools essentially to use topology optimization in a good way rather than a painful way.

1:09 This is us. We’re about 25 people. These are the people we work with. So you can see there’s probably a lot of aerospace companies in there. So the next slides is kind of like intro to topology optimization, but there’s been like five or six decent talks about to up until now. But it’s got a cool animation. So just bear with me. Yeah, look, you start with a design domain.

1:30 It’s empty. You set up a physics problem. So in our case, it’s CFD, right? So you set up your boundary conditions. You set your parameters. You also set up an optimization problem. So what do you want to achieve? In fluid topology optimization there’s probably some measure of pressure losses across your system and then probably some in our case thermal objective as well. So maximize this temperature minimize this heat transfer whatever you want.

1:56 There’s the picture. You discretize your domain. It’s it’s a computer simulation after all. And then you follow this loop right? So you solve your CFD problem. You solve your optimization problem. You calculate your gradient. You update your design. You loop around loads of times. And then once you’ve converged, come on, come on, come on, come on, come on, come on, come on, come on. Yeah. Okay, here we go.

2:17 This is the animation. And All right, cool. So, you see eventually it converges to a design with loads loads and loads and loads of little fins. In this case, we’ve got cold food coming in from the left, hot walls, and you’re trying to cool the walls as much as possible. So, it creates these fins, creates loads of recirculation regions. There’s some streamlines. And so you get lots of convection, but the fins are designed to not cause any separation or turbulence turbulent losses.

So, you got a very low pressure loss across the system. So that’s the principles of what’s going on. I’m not going to talk about our tool that much. I think you can do that look look it up online and stuff. So, for the rest of the talk, it’s just going to be about engineering problems that need to be solved and so on. Oh yeah, there you go.

3:05 Don’t ask me anymore. So, okay. So, the talk LPBF and aerospace. So our history, Toffee X, we spun out from the department of aeronautics at Imperial College London. So, aerospace is in our blood. My background is aerospace. I worked at Rolls-Royce before I worked at Toffee X. I like planes. I like making planes. And basically applying additive in an aerospace environment is important to me. So I I have this this sentence I came up with and I felt very smart about myself when I came up with this.

3:41 But then I realized probably there’s there’s two sort of like fields or industries which really sort of are the bell weather for how successful the additive manufacturing industry is. It’s aerospace and its medical devices. And I sort of came up with this, right? So, airspace relatively low volume compared to literally almost anything else. Right. So, your big Airbus will maybe produce a couple hundred planes a year, right?

4:08 So, it’s not huge volume. Additive also very good for making low volume, high value stuff. Aerospace, they only care about performance. They’re not so interested about cost. Yeah. All right. Cool. There you go. Not cost constraint. Aerospace very expensive. Airspace is a huge supply chain, right? It’s enormous. It’s a problem. You can see that if you follow kind of like what’s going on in the in the market right now.

4:32 Loads of delays and deliver deliveries and stuff because of supply chain challenges. Well, guess what? Additive enables, right? Okay, cool. Aerospace also innovation, R&D driven culture. You know, you see it in marketing everywhere. This is aerospace grade aluminium or something like this or carbon fiber sort of reinforced plastics was a you know another big innovation that came out of aerospace is kind of like a a crucible for for a lot of innovation.

4:56 Wow, this is an innovation in design manufacturing process. So this seems like a perfect fit. But if you actually look at what is flying parts manufactured with additive you actually probably only need one hand to count the number of certified parts which are flying on aircraft. So if we can’t get AM to work in aerospace, what are we even doing here, right? And you can probably follow this exact same sort of logic for for medical devices as well.

5:23 Now this we we’re based in the UK and the UK government fortunately seems to have realized this and is pouring loads of money and investment into additive in aerospace. I we speak with a lot of companies big OEMs as well as suppliers and these are the kind of like three main pillars of problems we’ve encountered or that they see right one is productivity so can we actually make the stuff we need to make right so at the moment even the biggest metal additive machines in LPBF and I guess other manufacturing processes as well are pretty space constrained you you know maybe the biggest ones you can sort of like half meter by half meter by half meter, right?

6:09 So, anything bigger than that, you can’t make it. That’s a problem. Laser power, parameter choice, beam shaping, choosing, you know, size of the the laser, so on. Anyway, loads of stuff here, lots of words. This is a boring slide. The main thing is also design for additive, right? It people still are using legacy techniques to design for additive. And I know you all know this because if there’s any room of people in the world who understands how frustrating that is, it’s this room here.

6:40 Okay, what did I say there? Yeah, also design cycles. Oh yeah, this is very important. This is specifically for aerospace, right? So design cycles are 10 to 15 years long. It’s probably the worst kept secret in aerospace right now that Airbus have started work on their next airplane project. It’s 10, 15, maybe 20 year design cycle. So, if additive doesn’t get on this plane, then it’s not getting on any plane until 2050, right?

So, I don’t know. I’d like to be living on an island by then. So, Dexam, this is kind of like one of the solutions. So, this is a huge UK, government funded project led by Airbus, all these really cool people here and us as well. So, that’s really cool. Yeah, basically everything that I said, but we’re going to spend money to try and solve it. This is one of the things I’d like to speak about and then I couldn’t.

7:31 Okay, so let’s talk about good CDFAM. So this is the rest of the talk now and these these next couple slides kind of presents our view at toffee of like what is good design? What is good engineering design? And I want to start with like this is a throwback slide. This is like back when we were toffee am back in 2020. I kind of dug this one out of the attic.

7:59 So the the the most important thing we think when we’re trying to design for additive is we want to challenge engineering biases, right? So not not like oh engineers are terrible but we have our biases based on experience and data that we’ve already sort of lived through. And so when we start to design a new product it’s normal that we don’t start from a blank sheet of paper.

8:17 It’s normal that we start from some existing knowledge, right? And so here’s my here’s my analogy. The top there, those are fuel injectors in gas turbines. I don’t know which company they’re from, but they’re a bunch of fuel injectors for gas turbines. And on the bottom is basically a wooden bridge across the temps. This is like a drawing from Victorian times. Okay, you get the idea. It’s a strut bridge.

8:39 This is a fuel injector built with additive by General Electric for one of their gas turbine sort of research projects. I don’t know if you noticed it looks exactly the same before. It performs exactly the same as the ones came before. It’s just like much more expensive. So like what’s the point? And this is the first metal bridge built in the UK. It’s made from iron. I think it’s got this awful name like the iron bridge.

9:06 But you’ll notice it’s exactly the same bridge just built out of metal. And that’s because essentially the engineers at the time were basically knew what they worked with before, wood, right? And they said metal or that’s kind of like wood. And then they they they kind of like built the same bridge. But if you look at sort of like once we’ve learned how to engineer for metal, we can do much crazier things.

9:27 Now, I don’t think this is like a better bridge, right? Hopefully it works just as well as the other bridge, but it sort of allows you to sort of see what else is possible. In this case, it’s more like architectural expression and so on, but in terms of performance, I don’t think it’s that much of a stretch to say, okay, yeah, maybe maybe we can do a better fuel injector.

9:46 So that’s the first thing. Can we challenge engineering biases? Second is we’ve been talking about this for a while and it seems everyone else is caught on as well. So that’s great. Engineer in the loop and explainability. You can’t just make designs and say, “Hey, it’s good. Go run with it.” because engineers are human and they’re going to say why. And then also I’m not getting on a plane.

10:13 You can’t explain how it works. So this is important. For us specifically with topology optimization and especially topology topology optimization for fluids, we can only find local minima in the optimization problem. Which means that for a single setup there might be loads of good solutions and we need some way of choosing what is the good the good solution we want. And because we want things to be explainable and understandable maybe the engineer should have some input on finding the minima that they want.

10:47 So there’s a bunch of tools which can be considered design constraints but are not sort of super important in the context of of like manufacturing or so on. And you can sort of like in this case we can make a pattern of of whatever but we got loads of other examples like this. I’m running out of time. Okay. And then speed of iteration. Everyone knows about this.

11:08 It’s very good. We want to go really quick and learn as much as we can. These are six designs made with toffee. Here’s the purto front. This took a day. Okay. And education as well. Okay. We all know this. Everything I just said, we need to educate the market that this is how it is and this is how they need to work to get the most out of it.

11:25 Okay. Halfway through. Let’s go. Right. So, the rest is all examples. So, the first one’s airspace. So, good. Although I don’t have the experimental data, so not so good. But and then the rest you can use your imagination. So this is a project done by Jack Tuft at University of Glasgow. He’s currently at Orbex which is a launch provider based in Scotland. And he’s was interested in applying topology optimization to regeneratively cool rocket engines.

11:56 So those of you who don’t know rocket engines, they’re very hot, right? So you get really really hot gas coming through and it’s hotter than the melting temperature of the metal. And so you need to cool the the outside of the nozzle. And normally the way people do that is they run cooling channels using the fuel. They run the fuel through the cooling channels around the walls and back up and they inject the hot fuel into the combustion engine.

12:18 And so this cools the walls. So that’s good. And it improves the efficiency of combustion. So that’s better. So he said, “All right, I want to I want to have a go at doing this.” so he did okay. He did a bunch of stuff. And he sort of basically started from this empty domain here on the left. Set up the known heat fluxes from the expanding exhaust gases generated a 2D design domain from this and then set up the optimization to minimize the wall temperatures and minimize the losses of pumping the the fuel through the channels and then using the patterning feature I talked about earlier he could pattern it that go all the way around the engine.

13:00 So this is the kind of design he got. He did a bunch of designs. So, I’ve only got three here. Actually, two here. That’s kind of like the baseline there. I don’t know if you can see the colors, but like there’s blue and red where red is like super hot and red blue is not super hot, but still hot. And he generated a bunch of designs, iterated through them, estimated their performance, made a couple selections as to which ones he wanted to test, manufactured them, put them on the test rig, and then tested them.

13:30 And I’ve seen the video. It didn’t explode and it was cold. So, that’s great. But unfortunately I can’t share it with you. But that was a relatively successful project. And in terms of like challenging biases, one of the things we we were expecting essentially was for the So, you can’t see it. All right. So, basically the highest heat fluxes are at the neck the throat of the nozzle.

13:55 And so we would expect to be there to be sort of a higher heat transfer surface area, more fins built in that region. What we found was this wasn’t true. It was building very few fins in that region and loads of fins at the end. So many small features there. And we were a little bit confused by this cuz that’s not what we were expecting. And then we did some simulations and we studied it and then we realized, oh, actually what’s going on here is by the time the the coolant gets to the exit of the nozzle, it’s already very hot.

14:23 So it’s not offering much cooling and so you need a higher heat transfer surface area to cool the thing down. Whereas if at the net at the throat it’s pretty cold anyway. So you don’t need a lot to cool the thing down. And so this kind of was it’s obvious when you look back at it but when you start from a blank sheet of paper it’s not.

14:41 And this is what we could learn. Okay. This one is a project done by Okay. I’m sorry. This is done by our customers Rico. So they are interested in building parts for automotive applications and a lot of electric vehicles around right now obviously and they need to be cooled down. Their thermal challenges are very different from internal combustion engine challenges. So one of the things that needs to be cooled down is an AC/DC inverter generates a bunch of heat.

15:16 We don’t like this. And they kind of set up toffee x design problem for an AC/DC inverter and they wanted to manufacture it using their binder jetting technology but as we’ll talk about a little bit later it’s a little bit more interesting than that so what they did is they started by splitting their domain into three this was an engineering choice they didn’t need to do this but they chose to do this to enable a higher rate of production I believe did a bunch of designs really rapid iteration a lot of quick learning and then they chose the three designs they like the most.

15:50 And these were essentially optimized again for questions about like differences in temperature at the inlet and the outlet and so on. Now the interesting thing about this as well was they could in principle have manufactured this all in one part from the way they’re buying a jetting technology at a very competitive cost competitive price. So this is for automotive they care about you know saving 5 cents here and there.

16:16 And what they chose to do instead was essentially just manufacture the blue pits, these three blocks. And then using their binder technology and they get a base plate and then diffusion bond the two pieces together. So this is quite it was quite an involved project involving many different manufacturing engineers, many different design engineers. But the end result was that this is okay this is experimental data. So, ignore the two pictures at the bottom.

They’re, the, oh, sorry, the words. They’re on the wrong order. On the left, it’s the aluminium, forged baseline design. So, this pin fin structure, which might be difficult to see. In the middle is kind of like the super version. It’s made out of copper, highly conductive, a little bit more a little bit heavier. But, you can obviously see the gray bar is much lower. Lower is better here in both cases.

17:11 And then the the design with that they generated with toffee. It’s as good or better performance than copper despite being made of aluminium. And it’s also lighter than both other designs. And I challenge you to come up with a design as complex as that just with a piece of paper and a pencil. Okay. Last one. This is my this my baby. All the other ones I didn’t work on them.

17:38 This one this one’s me. So we we were approached by IKM. So they are a LNG field services and hardware provider in based in Norway. And they came to us and they said we want to make a vaporizer for sampling of liquid natural gas. I knew nothing about this before but it turns out if you want to transfer shipments of LG you don’t just measure how much there is.

18:04 You need to measure the energetic cont content of the of the gas because apparently that’s what people care about rather than how much of it there is. And so what they do is they just sample it and they take samples and they take sort of a minutes of of like samples of the energy quantity and then with this they can sort of price the gas both for the sellers and the buyers and also for regulatory reasons you need to be able to calculate how much energy is in the gas.

18:33 And so this is very commercially important to be able to achieve accurate samples of LNG at a low cost and basically they came to us and said we want to have a go and we have all these partners which is great. It was great to work with them. So this was a problem for us because it involved sort of expansion and evaporation and combustion and all the heat transfer involved in that and the pressure losses and the cryogenic nature of the liquid natural gas coming in and it was all very difficult.

But we we we we said we’ll try we don’t have any of those models, right? So so all we have is a nice incompressible fluid that’s well behaved. We can do turbulence but that’s about it but we thought we’d have a crack anyway and just that we should be smart about how we approached this problem. So we started by just iterating. So here are some examples of iterations we went through.

19:32 These are all just basically outputs from our software. So top left we just said heated it up. So the big bar in the middle that’s a heater. Gas comes in from the left it comes out sort of like this hole here. And we got a spiral wrapping around the heater. So that’s great. Looks really cool. Oh god, my text is too big. Okay, that’s fine. Whatever. And then we thought, okay, this is we can probably do a little bit better than that.

19:55 What if we tried to evaporate the gas as much as possible in like the first third and then we used the latter 2/3 of the of the vaporizer to mix the gas. So what we did is we changed our objective functions. We said maximize the heat transfer at the beginning. And you can see essentially the spiral right at the beginning is much tighter around it. So you get many more spirals around there.

20:14 And then the rest is essentially once the gas is vaporized hopefully for mixing everything and just make sure you finish the the vaporization process. And then basically we start applying manufacturing constraints. We also wanted to insulate the walls and so on. And so you can see this sort of like evolved over and over and over and over again. Come on you right. And so this was the final design we got to kind of looks like a pretzel or a pelvis.

20:39 But essentially what you get is we get two basically on each side there’s two recirculations, two spirals. And this allows you to basically mix it quite well. And then get it out the other side. Then Vinenzo, a bunch of other really smart engineers did some more work on it. I wish I could talk more about this. Come speak to me afterwards. Generated this design and it was all good.

21:05 They tested it on the left baseline existing vaporizer. You kind of you can see where they installed ours, right? So the standard deviation of the measurements is half across a huge range of flow conditions. And so what that means is to get essentially the same sort of like accuracy of energy measurement you need to burn half the gas. Okay, I’ve run out of time. We also do to cooling for tooling.

21:33 So this is metal die casting. This says we’re really good. I ran out of time. There we go. Thank you. To learn more about the CDFM computational design symposium series, to see the archives of previous presentations, and to learn about future events, visit CDFAM.com.