CDFAM NYC 2023 · New York · 14–15 June 2023
Physics-Driven Generative Design: The Future of Engineering
Abstract
This recording from the CDFAM Computational Design (+DfAM) Symposium features Nina Korshunova of Hyperganic.
Presentation Title: Physics-Driven Generative Design: The Future of Engineering
Utilizing her expertise and drawing upon her experiences at Hyperganic, Korshunova provides in-depth insights into the tangible applications and potential of this novel approach, enhancing our understanding of this emerging field.
This presentation is tailored for professionals, researchers, and academics in the fields of computational design, additive manufacturing, generative design, and physics. Those intrigued by the integration of physics and generative design in the context of modern engineering will find this presentation particularly valuable.
Please note: The CDFAM Computational Design (+DfAM) Symposium is a respected platform for sharing knowledge and innovations in computational design and additive manufacturing. This recording is shared for educational and professional development purposes.
Transcript
From YouTube’s automatic captions, lightly cleaned; expect some errors. Each timestamp opens the video at that moment.
Read the full transcript · 3,208 words
0:08 All right. Welcome everyone. My name is Nina and I am from Hyperganic to present you today something really really exciting. Before we go on, let’s watch a short clip that just went live on LinkedIn. Somehow the audio doesn’t come, but never mind. All right. So, we finished up with a very interesting question, what’s next? And today I would like to talk about physics-driven generative design. And maybe we can get a quick peek on what the future of engineering could look like and especially what we imagine it in Hyperganic to be.
0:51 But before that, I’d like to introduce myself because I think I see my PhD in Technical University of Munich. That’s where everything started with investigation of additive manufactured structures. In particular, I’m a computational engineer by training, so I’m also a software engineer. And what we had to do is develop new numerical methods that could digest the complexity of additive manufacturing. And the beginning, it was just a PhD.
1:18 I never thought that we are going to have a startup, but my colleagues and I together founded DirectFEM. That was a company that brought these numerical methods actually to commercial world. And last year, Hyperganic acquired this company and we embarked on this journey all together to build something really important. So, it’s been already a year and that’s how we started. And you know, before this, I’d like to say how I, together with my colleagues, saw hyperganic.
1:58 Before this, hyperganic was about algorithmic engineering. We always talked about a paradigm shift. And why so? Well, basically, you know, we all learned how to design in CAD, how to design with these surface-based designs. I tell you my experience. In this case, I spent hours on designing something like what you see on the screen. Algorithmic engineering was something completely different. We took a motto and say, “Engineers will never have to solve the problem twice.
2:36 Why? Let’s encode this knowledge. Let’s put it in terms of the code in something that can be easily automated, easily repeated without an engineer touching anything on your computer.” And this allowed hyperganic actually to focus on something else. Engineers are trained to solve problems. I ended up in front of the computer trying to fix and learn the new tool instead of actually solving an engineering problem. That’s what was for me algorithmic engineering and what it was all about when we joined.
3:16 Moreover, hyperganic was not just algorithmic engineering. I already talked about the concept, but you know, this was extremely fast. I really was surprised. I really have to admit it because when as a software engineer you think, “Oh, how did they do this?” And when I speak about speed, I speak about two things. I speak about the speed of the kernel, about the computation, and visualization. I never thought that a voxel-based models with signed distance fields can be visualized on my laptop without any any lag.
3:54 That’s when I talk about speed and it brought so much flexibility because you could design something that you never thought before and it could have been complex, it could have been simple, it doesn’t really matter. You just need to put your imagination and start designing efficiently. But you know what? I have to say that the acquisition was actually for a reason. This was not enough. This was a design.
4:25 And I think I will speak to everyone in this audience that every chair that you’re sitting on has purpose. You’re sitting on it. Every cup for coffee has also its purpose. You put something hot in it and you don’t want to burn. Physics was missing. Physics was something that gives the design a purpose for later. And that’s how we came to be. But I actually never spoke about why such an acquisition.
4:52 You know, you can tell me FEA and element analysis exists already. I agree. I studied myself. The thing is when you talk about such complexity, I think meshing is quite a problem. I think I can challenge each and every one of you to just mesh the object from before. So at Indirect FEM, we developed a technology that has no meshing at all. It’s a quasi meshless method called finite cell method that allowed to eliminate this meshing process.
5:25 And guess what? It allowed to come from a design to simulation with one algorithm. There was no any more setup like CT scans, point clouds, floated STLs, just even start CAD models. It doesn’t really matter. That’s what this technology was really capable of. And we went on and the first challenge that we had to achieve when joining Hyperganic is to embed and make it work together like one.
5:53 So, one plus one design of Hyperganic and simulation of Dark Femme. We achieved it and we also had a little bit of a discussion because we also heard oh, you need to introduce assemblies. So, as of now you can run linear elastic analysis on homogeneous and non-homogeneous material laws. And when I speak about non-homogeneous material laws, it’s something that comes from CT scan where every point is really different material.
6:21 You can also simulate different assemblies and it’s absolutely geometric agnostic. And the first question that I get asked very, very often yeah, cool. New methods. How fast is this? Because of course, this is not really a black magic. I’ll answer this question right now. So, this was actually the publication was accepted today. So, you can all go and read it on. This was done together with ETH and we were working on developing of auxetic materials with Hyperganic core.
6:56 This ran within 30 seconds on this laptop. That’s the speed we are talking about. It’s rather simple. I will never claim that large objects will never take time, but that just gives you an example or an anchoring point where we stand. The next question is how accurate it is. Physics actually requires accuracy. I brought an example with the colleagues and this is comparing to Nastran. On the left you see actually our code.
7:30 At the right you see Nastran of very simple geometry. It was just because we actually had a lot of conversations about it. We can confidently claim that our method converges in the same rate as finite element method and to the same values. We also have quality control. We also have exactly the same troublesome convergence process. But we converge to the correct values. And because we are coming from the PhD world, we can say that please go ahead and read our publications and you’ll see that we actually spent quite some times on validating and seeing whether the math is correct.
8:13 But this is still also not enough because when we were faced with hypergolic geometries, nothing is linear elastic. Nothing is actually structural. There are quite a lot of other applications. So we went ahead and we had to face and implement thermal simulation, heat conduction, new physics. So we went instead of going depth, we went to the breadth of the physics possible. So you again can simulate multiple domains and see which which law you would like to choose.
8:44 Still not enough. Next step was to implement electromagnetostatics and this we released in the past November of last year in our platform. Of course, we are not doing merging. It’s one-way coupling. Electromagnetic coil analysis is possible, but it’s electromagnetostatics. And again and again, I need to repeat that there was a mind shift of company. For us, it was important to deliver accurate results that are robust and have no step in between design and simulation.
9:21 And I remember in the past year when I joined them in in my education, I studied structures. I often got asked, “Okay, but you know, we want to do this fluid simulations.” I was like, “All right, yeah, sure. I’ve never heard of it and I don’t really know how to do it, but we actually had to sit down and do it. And as of now, I can proudly say that we had to challenge ourselves to find a new discretization method that would allow to work on Hyperganic Core and enable CFD simulations.
9:57 We have already implemented the flow models with boundary conditions and straight and curved boundaries. And as of now, as we speak, we are introducing advection convection support such that we can comprehensively simulate thermal fluid dynamics way of delivering this to the public. So, all in all, simulation took a very, very big step. Inside Hyperganic, you can save this manual effort. And instead of focusing on meshing and trying to see where the surface is reconstructed well or not, whether it’s water or it’s fine, focus on actually what matters, solving your workflow in different physics, advanced, accurate, and fast.
10:48 Well, you would say, “Yeah, good. Still not enough.” For me, this is not enough. Because we spoke quite a lot during this morning that there are very siloed steps. There is design, there’s simulation, there’s production. Usually, I very much agree these teams never talk to each other. The designers are just there with their cubicles. Simulation guys really don’t talk to designers. And guess what? This really slows down the process.
11:19 This slows down the production of products that actually could matter. It can take years. We asked actually one of our customers how long it takes, and they told me 5 years. I was like, “Cool, that’s kind of a lifetime for me.” But, imagine actually the world that these silo steps would not exist. Imagine that they would merge together in a seamless feedback loop without anyone touching the computer.
11:52 That was the next challenge. And that’s what we are currently embarking on, and you already see what is possible. That’s actually what for me means physics-driven generative design. This is the first step towards it. And in order to really say again on accuracy and what it will achieve, we can go on and see, for example, this is just an aluminum ring, and we talked about the heat sink application, so we wanted to really position our boundary conditions right and see where it is going to lead us.
12:25 Well, there were six total iterations with 30 evaluations, and we achieved 59% increase in total heat dissipation. There was no one sitting in front of the computer while this was ending. It was 30 minutes. So, as I said, we don’t do black magic. We still need some time, but this is fully automatic. And this is exactly what opens up a whole world of generative design, because this is optimization.
12:53 There is nothing in there more than what we talked about before. But, for me, generative design is exploring the whole design space without any interaction. Where you can actually see how the objects can look like, and maybe have different trade-off of costs, of manufacturability, of performance, and so on and so forth. This is where we have to arrive. And for this, we still need something else. It is still not enough.
13:27 And I I must admit that it will never be enough for you the PhD people inside Hyperganic. But you know, last year we had the platform launch. And everyone knows that this Hyperganic was a platform that was fully based on coding. We got a lot of feedback. And I must say that it was quite challenging. But we learned something very important. No matter the software tool, we are all humans.
14:00 And we are all absolutely different. I as I said, for me software engineering is just to my heart. I can code. I can program. And this is what I would choose when I have to solve an engineering problem. However, I’m so sure that there are people in this audience who would choose a UI. And it’s totally fine. So, I would like to ask a question and it’s more a rhetoric question inside for you.
14:33 How long did it take you in your past jobs or in your current job to enter and understand the software to become productive? Just answer it for yourself. I’ll just tell you my experience which doesn’t need to be necessarily true. It takes me ages. Although I’m an engineer I’m a programming tools I’m a buttons somewhere else. It slows us down. So, we learned that the tool has to be human-centric.
15:03 We learned that we all are different and we all need different tools to achieve the same goal. And that’s okay. And in the past year I told you we were a coding platform. That’s of course gives quite a lot of flexibility especially if you know how to code, you can just go crazy and generate these kind of pictures in terms of minutes. But we have introduced the UI such that you can easily get into our platform.
15:33 Moreover, since then we brought it to the web to make it accessible for you such that you can just go on and try it for yourself without any further installations. You see here there’s a missing piece. There is a very big piece missing because there is something that we learned from being a kid. We learn language. So therefore you should be able to talk. To talk to your tool without any further interaction and be able to solve your problem.
16:10 And that’s what Hyperganic has been working on for quite some time. We all have heard today too that there artificial intelligence is on the rise. For me this is a very vague thing, you know, I know deep learning, machine learning and AI for me something vague so but the thing is that ChatGPT and other tools showed us what is possible. But I think it’s also possible in engineering where the tool can be made accessible.
16:44 Where the tool can become creative. And I already see quite confused faces. I’ll tell you one more motto of Hyperganic. Instead of talk show. And show what is possible and say where you are, be accurate, be humble and show to the world what else we can do and what we cannot do. So I’ll invite you on a journey to finally unveil Hyperganic. Because we had a lot of comments about this, too.
17:21 So, I had to record this demo and I will start it. And I’ll tell you one more thing. This is not sped up. This is real time. So, we go to our website and instead of just clicking on the UI that you can see up there, this is Please, really just a prototype. You can go inside your text to 3D and try to generate object. So, here I wrote that I would like to create a cylinder with an outer radius and an inner radius.
17:52 Also, forgive me. So, you already hear me speaking. I forget articles. I’m from different language and I also can mistype things. But guess what? We are confident that this can be produced. And one more thing. With Hyperganic, we are sure that the objects that are produced are watertight and engineering ready. So, there is no more this kind of, you know, point clouds where you would see that there are holes inside the object.
18:24 There is a real cylinder that I produced, but you would tell me, “Oh, cylinders. We also cylinders.” Well, as simulation engineers, we actually like cubes because that’s a perfect benchmark for all of the tests. But instead of this, let’s really infill it with some pattern. Simple. Gyroid pattern, I put some cell sizes and some wall thickness. Maybe it will help me to do something else. Well, the tool is not magic.
18:51 I also promised you this. This is an MVP. And this is really happening real time. There is no GPU based. It’s really all CPU and you get your cylinder out. With all gyroid patterns. And I promise you yet another time, you can export it to Actually, where is the one 3MF? As of recently weeks. And you can just go on and try to manufacture that. Still not enough.
19:21 Let’s try to instead of just in-filling it with a pattern and transition it radially to another pattern, something like a fissure cone in this case. Precise I just said transition to fissure cone. So, it will produce something and you need to be very careful with what you’re doing. If you really want to be precise, you can put three additional widths and see how it is. And you already see it on the screen, right?
19:44 So, you see that the pattern is transitioning already to the right side without me yet another time clicking on something else. But still, let’s try to You would say, “Yeah, it sounds quite small, right?” Let’s change the radius. Let’s go to 25 mm. And while we are waiting, I know this is happening on this laptop yet another time. Instead of this, there is no connection to the web and there is just a CPU node running with eight CPUs.
20:15 No magic. There is no GPU. And the visualization is live. There is no acceleration. There is no caching. And of course, you already see that it takes a bit longer and we can produce just a simple cylinder. So, this is what new Hyperganic is. This is what we’ve been developing for quite some time and we really believe that this will be the future of engineering. Because each one of us is different.
20:46 People will still be able to code here. You can go on the browser, open the code workspace and just go crazy with that. You can also go for UI. You can go for talking and the objects will be produced. So, that’s what we believe in Hyperganic is a small peak in the future. And now, I need to show you this video again. I urge you to go to LinkedIn because there is unfortunately a very cool sound inside here where we show what actually is possible with the new Hyperganic with the new platform.
21:24 And we’ve been building it for quite some time. So, actually today is a very important day for us because you can see what is currently possible. And instead of just ending on this note, I’d like to invite you to scan this QR code and just apply for our exclusive private beta release where we will talk openly about our products where you will be able to test it yourself because this is a time where we need your help.
21:57 We need your help to improve our tools further. We have heard you before in our software platform. Now, we built something else, but still you are the people who are doing the future of engineering. And we would like to thank you to being with us on this journey and continuing with us further. Thank you very much. And if you want to see the software live, come back to me. It’s on my laptop. Thank you.
Listen, search and cite
Listen to this talk as a podcast
More from CDFAM NYC 2023

Generative Design & Digital Manufacturing at NASA Goddard
Ryan McClelland · NASA Goddard Space Flight Center

Generate-to-Validate: Optimal Designs and the Need to Deliver Under Uncertainty
André Wilmes · Rafinex

Defining Performance not Just Geometry with Software Process + Material Science
Hardik Kabaria · Carbon

Reinventing the Wheel – Representing & Manufacturing Geometry with a Program
Mark Burhop · Siemens Technology

Interspecies Forms: Robotic Feedback Systems Mycelia Growth & Computational Form
Natalie Alima · Adidas / Biolab Studio

Designing Material Properties with Advanced Toolpath Generation
Alex Roschli · Oak Ridge National Laboratory

Design Automation Using Machine Learning & Computational Design
Rhushik Matroja · Cognitive Design Systems






















