CDFAM Berlin 2024 · Berlin · 7–8 May 2024

Industrializing Physics-Driven Generative Design

Listen as a podcast

Abstract

Physics-driven design tools such as topology optimization, while transformative at the conceptual level in engineering, have yet to significantly impact production components due to limited adoption in industry workflows. Through a series of case studies and technical analyses, this presentation examines the barriers to the integration of these tools into industry workflows, and in particular those preventing the mass production of topology optimized parts (such as cost and manufacturing constraints). We will demonstrate technical solutions which allow engineers to harness these tools across various sectors, enhancing value creation and efficiency for generalist engineers.

Transcript

From YouTube’s automatic captions, lightly cleaned; expect some errors. Each timestamp opens the video at that moment.

Read the full transcript · 2,707 words

Okay, hello, my name is Thomas Rees, I’m from a company called ToffeeX, much the same way as Morit I decided the title of this presentation about four months ago, and I wrote the presentation about yesterday, so things have changed a bit, but I I think I’ll still probably be around this this topic here. So I’ll first talk a little bit about our, as in Tox’s view of of engineering and engineering workflows and where computational design tools can sit, and then I I think the second half of my presentation is going to be a little bit different from a lot of the ones I’ve seen so far. I’m not going to talk so much about our workflow, our features, our user interface, all this sort of stuff, if you want to know about that come speak to me, I’m just going to show pictures of parts that have been manufactured that were designed with Toffee and talk about them a little bit, because that’s kind of like what I love to do.

1:02 So first, who is, who am I, what is Toffee? We’re a spin out from Imperial College London, we’re based in central London, we’ve existed for about 3 and a half years now, and we’re growing, so 25 plus is not true, I think it’s basically 30 now. We make a CAE tool, it’s cloud-based generative design software, it basically uses high fidelity physics simulations to generate optimal designs for thermofluids components, and it’s a cloud native software as a service. I’ll maybe I’ll talk a little bit about why that is a little bit later, but yes, and these are all the people who have given us lots of money.

8:53 So this is kind of how I see engine ering workflow from my background, which is aerospace, right, so you first update the design, because you never really start with a blank sheet design, especially in super high value manufacturing, you start with an existing design and you modify it to improve it. You normally check that your improvements are working by running computer simulations, because building and testing things are, really is really quite expensive, but eventually you do have to build it and you do have to test it. Just to give you an example, even with additive, you know, it still can take a very long time to build something, we had a customer who built a heat exchanger, it cost them 80,000 and it took a month of build time, so it’s not like you can iterate quickly with these super high value components.

Eventually you’re going to run out of time or money, and then you have to go to market, and then you do it all over again, and this is quite a long slow process. Fortunately with with the advent of like actually good additive manufacturing, and people have started to come up with sort of DfAM tools, I guess that’s why we’re all here, which are actually super cool, and they basically, I think, changed the game a little bit for for additive, because people were no longer just printing existing designs that were designed to be manufactured using a traditional manufacturing approach, they were actually making designs that you couldn’t make otherwise.

But these tools essentially have a similar problem as conventional design approaches, in that it is essentially still driven by the engineer drawing stuff, and they need to go through iterative this iterative loop that I talked about before, and you still run into the same problems. So we were working with Rico, one of our customers, who took a conventional heat SNC design, redesigned it using a DfAM process, and found when they tested it that the performance was actually worse. And we’re sort of wondering, well, what the hell are we doing here then.

So that there’s possibly a solution to this in the form of topology optimization, right, so topology optimizations, you know, you have physics simulations, you have optimization algorithms, and you can use all this information and this data and these gradients to generate super optimal designs, but often these designs are quite complicated, very difficult to manufacture. So so your sort of design cycle for topology optimization was, you know, start from an existing design, do the topology optimization, have to redesign the topolog, topology optimiz, optimized part, because you can’t manufacture it, because there might be overhangs or something like this in for additive, or if you don’t want to manufacture using additive you need to like completely redesign it, and then you get your to inspired design.

And this is fine, but it still kind of sucks, because you still haven’t solved this problem, right, and this is kind of where I hope Toffee can can step in and maybe accelerate this whole cycle and really improve time to market by another step. How do we, how, how do we see the, how can we do this, right, so what are the problems with this, like, what do we need to address? This is my personal view. First of all, it’s, we need to address the complexity of like how people use these softwares, often with engineering, not often, always with engineering, and it’s necessary, you have these really interdisciplinary problems, right, so it’s not just, you can have a CFD expert, you can have a structural expert, and you can have a manufacturing expert, and you need to get them all together in the same room, you get them talking to each other.

You need to address real world requirements, and so we’re talking about manufacturability here, which is super important, but there there’s other really important real world requirements, like certification, lifetime assessment, cost, both operational cost of your component, but also the capital cost to build it, and so on. And the other thing is, and this is more specific to maybe topology optimization approaches, is you need to be able to solve many different types of problems, right, so you can’t just solve, you can’t just make another bracket, right, you can’t just make another C lever, you need to be able to solve the wide range of problems that that engineers are facing today.

So just to illustrate where we end up here, so if you remember this, when we run this same problem through through Toffee, we get this design, and when Rico tested it, we did quite well, and just to illustrate that it’s real, the one on the right is is a photograph of the printed part. So the rest of this this presentation now is just pictures of really cool components, and some which I have, we haven’t shared publicly before. So first is, I’m going to talk a little bit about additive manufacturing, obviously that’s why we’re all here, and first we, we can, can constrain our designs using basically geometric constraints, which which describe the requirements for a particular additive manufacturing process.

So obviously overhand control, very important, we can do that. So you can see here on on the left we have a, essentially a heat sink, we have cold fluid coming in one pipe and then cooling the bottom surface which is hot and exiting the top pipe, and we can get, if we run this through our generation design software, we get some really complicated unconstrained design, but we can switch on an overhand control and we can get a constrained design for the bottom. Again, we can control for things like warpage or shrink lines, right, so on the left is an unconstrained design, but let’s say we’re particularly concerned about warpage, well, we can switch on a constraint and we can still do the same optimization, but it’s optimized under the this constraint.

So what can we do with this? We can, again, this’s another example we did with Rico, where they were actually, this looks like the thermal resistance has gone down, which is bad, which is true, but Rico are actually looking to really reduce the pressure losses, and they were willing to pay with some some thermal heat transfer there. So we can basically designed this, Rico built it, this on the right is a CT scan of the component, and on the left is basically the file we provided to Rico, and you can see it’s, it looks pretty good, so we’re quite confident that our parts can actually be manufactured.

Another example is this one is one we did with EOS, so the this is a heat exchanger, it’s not a DfAM talk without some heat exchanger. So this is the first of three here, the first, this is a plate heat exchanger, so you can imagine you have a hot fluid going in and out and a cold fluid going in and out, and in between them you just have alternating layers of hot and cold channels, and the question is, when you have just one pipe coming in with a fluid and you want to distribute it as evenly as possible among, I don’t know how many there are, there maybe five different layers, while also sort of minimizing any losses and ensuring a uniformity of flow across that exit, how do you do that, because it’s really hard.

9:29 And the first attempt, essentially, which is basically an old style approach, where you draw the picture in a CAD software, you run your CFD simulation, you look at the results, you get a clever engineer to start changing the CAD software, iterating and repeating over and over again, this can take a long time, and you have to do a lot of these iterations. So after three iterations this is sort of like the design we got to, so you can see the interweaving red and blue, blue, there, there on on that image showing the interweaving are hot and cold fluids, we can optimize for two fluids at the same time in the same design space.

10:11 And when we put this into Toffee, we get got these pictures, so you can see the red is the hot fluid design, sort of like the fluid parts, right, and the bottom is the cold fluids, and then on the left here you can see the printed component there, made by us. So these are kind of like some con proof that we can sort of constrain for additive manufacturability, but another thing we think is quite important is to allow engineers to be engineers, right, so we don’t want to say, just put your boundary conditions on the box and then go have a donut, right, like we want people to be creative, to to play with these tools, to to basically really, maybe, maybe we get it wrong, maybe it needs a kick in the right direction, right.

10:58 And and we want to be able to give engineers these tools to do this while also still optimizing for the best performance from a physics perspective, and just one of the features we, we have to do that is we allow users in their design space to specify if they want to have the same repeated part anywhere, right, so if you imagine on the left here there’s a lattice there of nine repeated parts, that’s just because the engineer said anything you build here I want you to build it here, here, here, here, here, and here, right, and you can see that you can do that in rotational periodicity in the middle, or even on the right, every outlet it looks different, but if you look at it above every outlet you have exactly the same geometry.

11:44 And you can imagine this is quite useful if you want to make one design that you use in many different places, but it’s also useful for making heat exchanges. So this is what we did with Boeing, for a, we made a periodic lattice, basically it’s an air nitrogen heat exchanger for an aerospace application, it’s about yay big, it’s made with this patterning tool, they built it at the AMR, AMRC in Sheffield in the United Kingdom, and it was tested in Germany, and we hit all our performance targets, it weighed less than the target weight, and we still managed to make a design which could be manufactured using a 300, actually I think it was 280 mic on all thickness.

12:31 And just so no one gets left out, we did the same thing for Airbus, but in this case they were interested not just in the thermal performance, the thermofluidic performance of the heat exchanger, they also wanted it to be loadbearing, so instead of just having a hot fluid and a cold fluid and do the best you can, they wanted to make a lattice which could be used as the ham and cheese in the sandwich structure, right, so on the top and the bottom we had these plates, and then we also had to op ize for the stiffness of the repeating lattice, and this was subjected to a three-point load test as well as the thermal fluid testing.

13:09 So our heat transfer is much lower than like a conventional heat exchanger, which is really not surprising, because the wall thicknesses here are much much thicker than the conventional heat exchanger, however a conventional heat exchanger under the 3O loading would explode immediately, and we’re stiffer than Nox honeycomb, in fact our our LCE didn’t fail, it was the bonding between the carbon fiber and the lattice that failed. So that’s kind of just another example of how we can combine different types of physics, different types of manufacturability, and generate these super cool designs, the pictures, that’s a stress distribution on the top right, and then that’s the periodic unit cell that we made.

13:54 Okay, so that’s kind of like some pictures of additive approaches, I I just want to mention here, here also, despite the fact this is a D CDFAM symposium, we also think it’s very important that we can manufacture using other methods, and so we have tools to control for bit size, for example, if we’re doing a machined approach, or basically we can give a a particular machining direction, and so you can see the results here of that of what happens when we change the machine tool. We also have tools to control for stamping, and we’ve even done some stuff with chemical etching as well.

14:38 So I think that’s the end of my talk, maybe it’s pretty short, but that’s pretty much all I wanted to say. Just to summarize, we’re a physics driven design tool, I haven’t talked about your user interface, I haven’t talked about anything like that, it it it looks like a CAE tool, right, it looks like someone took a CAD tool and and a CFD tool and like smash them together, and I’m happy to demo it if people want to see more about more about it, but I think what’s really cool is seeing what we can achieve, w, with our technology, right, so we we can look at new innovative solutions that maybe we would never have thought of before.

15:14 We can accelerate design time, by the way, this, from a blank sheet of paper design, sheet of design specs, to a file ready for printing, took two and a half weeks, with performance data, with experiment, with, sorry, simulation data to back it up, so we can really really accelerate design cycle, and I think there’s there’s other examples where we’ve cut basically design time in half. We optimize for performance, right, so we know a priori that we’re going to hit all the functional requirements, because we have these simulations going on behind the scenes, and we also think maybe engineers should be creative, and they should be allowed to play with things and not just have essentially a black box where some topology optimization happens, and we want them to be able to use their creativity to create basically the best new next best thing.

16:16 Okay, so that’s, that’s it, thank you so much.

More from CDFAM Berlin 2024

Computational Processes for Adaptive Biomechanics

Computational Processes for Adaptive Biomechanics

Jesus Marini Parissi · MoonrabbitX

Cognitive Design Systems

Cognitive Design Systems

Rhushik Matroja · Cognitive Design Systems

Generative DfAM in Footwear Industry

Generative DfAM in Footwear Industry

René Medel · Framas

Spherenes: A New Class of Minimal Surfaces

Spherenes: A New Class of Minimal Surfaces

Christian Waldvogel · Spherene

Process Automation for Engineers

Process Automation for Engineers

Daniel Siegel · Synera

Design Parts not Shapes

Design Parts not Shapes

Chelsea Cummings · The Barnes Global Advisors

Additive Flow

Additive Flow

Alexander Pluke · Additive Flow

Navasto – AI Accelerated Engineering

Navasto – AI Accelerated Engineering

Matthias Bauer · Navasto

Shape of Generative AI

Shape of Generative AI

Onur Yüce Gün · New Balance

Register for Updates and Discounts on CDFAM events.