0:02 Pleasure to be here, sorry that I didn’t get the memo to put a heat exchanger on the first slide, there will be heat exchanges, I promise. And I’m going to talk about how the cloud helps us to get to a real-time design insights with cloud native AI and physics simulation. It’s a lot of words in there, we’re going to unpack this over the next couple of minutes, and it’s going to make more sense, I promise. But I did want to start with a very provocative picture that I’m going to show you now, and do I already thinks on my God, what is he, what is he about to do? But it’s not that bad. I want to use this little picture as a small analogy.

0:32 We’ve, we’re all in computational design, we’re all either driving simulation tools, we sort of consumer of simulation insights, we all have, you know, seen simulation images. And I sell simulation software for a living, so I feel like I can make this joke that is about to happen over the next minute. And each time I see simulation results, our own, from other vendors, from simulation engineers, it reminds me of that embroidery, I think it’s called, right? I thought it’s knitwear, but of this embroidery, because on the front the simulation result looks great, it’s colorful, you can make design decisions. You know, you know where this is going, right? You can make design decisions based off of it. But what, most of the time, in those situations, is not talked about, is the simulation process. How did you get there? Yeah, how did you get there?

1:20 And I really don’t mean this, in, you know, for a specific tool, for a specific solver, for a specific vendor, for a specific engineer, it’s really the nature of the beast, right? It’s a nature of the beast that we’re trying to predict physics on spatially complicated domains, and we have those spatially complicated domains being shaped in all forms. We’ve seen many different, you know, modeling engines today, many, many different modeling paradigms, physics, multiple nonlinearities, it’s just nasty, right? And that is a single simulation.

1:45 Typically, most of the customers we work with are larger teams, taking automotive tier 1, automotive tier 2, they have many different programs that are support, they, they’re supporting, take a white goods menu factur, many different product platforms. So you have all those engineers that you want to make design decisions, and at the same time they should have able, they should be able to just validate their design decisions as they go, but they can’t, right? Because it’s, you, you layer on top all of this physics complexity, you layer on top user management, license management, data management, process management, and all the good things down here. I need to check, yeah, all of these things, right?

2:23 And it’s not to say that, that a single simulation solution or a single tool isn’t good at anything, it’s just like the nature of the beast. As an engineering team grows, as a real world engineering workflow is happening, as a real world larger engineering organization is, is designing products, it gets tough, right? It gets tough to roll out simulation. And typically what the response to this is, is a team starts to simply compromise and say, okay, look, this is so complicated, is so nasty, we’re not going to roll it abroad, we actually keep its central, we’re going to manage this complexity central with a dedicated team of people.

2:54 As a result, though, we’re compromising simulation lead time, meaning, because it’s central, because only few people have access to it, as a result, most of the design engineers, program engineers, support engineers that are actually making design decisions on a day-to-day basis don’t have access to real-time results. They’re left there, we heard it earlier, and are waiting, often times for weeks, to get their design generated until the design ass simulated. And again, design generation, we’re going to, I’m going to show later a few example workflows with SimScale, design generated could be done with any type of methods, any type of engine, it doesn’t matter, there’s then until it’s simulated there’s a long simulation lead time.

3:32 And I want to double click here once, and I’m not going to shake the table anymore, I want to double click once, because often times this is a bit mixed up with simulation cycle time. Simulation cycle time is really from where a simulation engineer, or for that matter any engineer, sits down and sort of, all right, look, let’s get this simulated, right? And then all the nasty stuff around pre-processing, meshing, so that is like, yes, there’s time loss there for sure, it needs technology, it, it needs good tools. But all the stuff before this is outlook, this is the simulation backlog, simulation teams backlog. This is, you know, this is a process issue, it’s not like necessarily a core simulation technology issue, it’s an issue that arises from the fact that simulation has been deployed only always centrally and larger engineering teams.

And so, why is that, right? Let’s double click, why does it look like this, why do most engineering teams deploy simulation this way? And typically, when we start working with a customer, typically the simplified way of how we find a simulation stack to look like when we start working with them, it kind of looks like this. It, it historically grew, and so it started with a, a first simulation tool they rolled out a decade ago, maybe two decades ago, maybe just a couple of years ago, for an particular physics, typically on some workstation, for some user. They start adding some additional tools, maybe investing in a cluster, they start sort of, you know, putting some homegrown automation around it, not the good stuff from Synera, right, we friends, I’ll mention them later again, homegrown automation, some scripts, they dabble into cat integrated CAE, because they actually would like the design engineer to make design decisions. And then eventually, for some peak workflows, some peak loads, they, they, they put stuff into the cloud, and that has created a lot of value, right? There’s, there’s, this is a, this is a process, this is a stack that has supported many design engineering decisions, many great products, you know, aircrafts fly, cars drive, so that’s good, right?

5:26 But unfortunately, it’s not very standardized, it’s a heterogeneous software and hardware stack that is hard to scale. And so, as a result, as we’ve heard before, think about the embroidery, it typically means that a central simulation engineer ends up running the simulations, driving the stack, right? And you can’t deploy it, because as engineers we know, if you want to scale something, you need, you need to standardize. Tough to do with this, because actually also, like what we’ve heard in, in, in, in presentations before, hardware limitations, data, da da da, it gets tricky. And so the, the hours and seconds that we actually want all of our design engineers to be able to make design decision instantaneously, gets tricky. And so you see where this is going, this was the red slide, right, now comes the blue slide, right? So that was, that afforded us the chance, and that’s sort of the, the story of SimScale.

6:16 We thought about how does, how should a simulation stack look like, in order to enable engineering teams to use simulation aggressively, early, and broadly in engineering processes, they make, thereby make better design decisions earlier, and thereby really enable many of the novel workflows that we’ve seen today, many new modeling paradigms. And it starts with the fact from our point of view, that, which might be a bit biased because we’re in the cloud, but from our point of view, it starts with the fact that user data process management can’t be an afterthought, it needs to be baked in. We see many organizations today that actually have deployed really good tools, really good solvers, right, but they want to scale it, and they are struggling with the fact that they try bolt on user data process management on top of it, which is tough.

7:00 SimScale starts the other way around, so SimScale runs cloud native, it comes native, natively integrated, all HPC commissioning is done automatically, we’re going to see it in a second, user data process management for broad deploys. And then the central simulation team always has access to everything that is happening, right? So they have full visibility into how simulation methods are being used, and thereby it’s not just, use it broadly, we heard earlier, trust, right, good, good point, so the simulation team needs to be a able to monitor what’s happening to ensure quality. On top of that, we integrate simulation technologies. And I also didn’t, again, for the presentation before, I didn’t clean this presentation from all the three-letter acronyms, so I want to pre-apologize for that, but we can see a bunch of different methods integrated here, we’re going to see some in action in a second, that then end to end bundled, to end to, in physics workflows.

We come from, from the cvd world, I have a cvd background. The second biggest part of our revenue is structural, and then thermal, and then emac, and all of that is then shipped really in a way that the customer wants to consume it. And that starts with, of course, most of our usage sits on our web-based GUI, we’re going to see it in a second, I planned on a life demo, while I’m not shaking the table, we’re going to see how that works out. But also the, and so that makes it easy for, we’re going to see later a case study where, when you want to deploy simulation to 100 engineers a specific method, that’s a huge pain, right? But if everything is, is vertically integrated, then sort of deploying becomes easy, so that is where the, the web GUI comes into play, and then the web API, meaning you have programmatically accessible physics, you can put it into third party workflows, which then leads to the port integrations.

8:43 And yes, you guessed it, this sort of gets us to hours and seconds in simulation turnaround time, and as a result, program engineers can sort of innovate faster without the simulation team being the bottleneck. So let’s see how this looks. I know there’s the mission slide, I’m the CEO, I cannot make a presentation without stating SimScales mission. So, in a, in a nutshell, we see our SimScale is not about one specific solver method or about one specific physics, SimScale is about how can we help engineering teams innovate faster by making simulation truly accessible at any scale, that means early, broad, while maintaining central quality control of an expert team that, that controls the methods. This is how it looks like, and everybody thinks, great, another simulation software, colorful pictures, right, we’ve all been there. H.

9:33 So let’s unpack this a bit, so what’s special about this one, by the way, again, an e-motor, apologies. First of all, cloud native, meaning SimScale.com, and you just log in and you’re good to go. Again, of course, sounds fancy, sounds nice, but if you want to deploy a simulation to 100 engineers, it just becomes like, we’ve, we know customers where it takes 2 months to commission award station there, right? So it’s just like, it helps to drive adoption. Many products in the industries we work in are not governed by single physics, but are constrained by multiple different physics. And if you want a program, sometimes called platform engineers, application engineers, to own functional performance of the part of the assembly that they’re responsible for, they’re not going to learn five different tools.

And so enabling them with a templatized way, in a single application, to drive many different simulations, is the, of the, the reason why we have such a broad physics footprint. Any scale, sometimes cloud vendors talk about unlimited scale, but we all know it’s somebody’s else computer, but it gives you more scale. So HPC commissioning is done automatically, that means you have High number of cores available, can run, can run large model sizes, and at the same time the number of simultaneous runs is practically unlimited. Let’s stay with that. Real-time collaboration, again, it’s 2024, right, every other software has embraced that, engineering software has not yet, we think, of course it’s nice to have it.

11:04 But it really comes, if you want to drive simulation adoption to simulation novice users, it’s, from our point of view, not a nice to have, it’s, it’s important, because only that way they have direct access to a simulation engineer. Simulation engineer can go in real time into the project, can see what they are doing, can tweak workflows and the likes. And it really surprised me that AI wasn’t such a big topic today, maybe tomorrow, D, tomorrow, okay, sorry to screw with the agenda. So it is AI integrated, we’re going to see more in a second what that means, but also here the point is, because suddenly all simulation data, all design data, everything that is being simulated, suddenly lives online in databases, in structured databases, right next to unlimited G, practically unlimited GPU computing power, just opens the door to a gradual path to AI adoption. Exact examples of what we’re going to see in a second. So much to theory, now let’s see it in action, and then afterwards I’ll show a few customer examples.

12:08 Okay, nice. So important, the important thing about SimScale demos is always, this is a Google Chrome browser, right, so we’re in a browser. We’re starting with SimScale.com, right? So imagine a scenario, T1, T2 automotive supplier, or for that matter, you know, from an electronics, om, a program, an application engineer wants to simulate a first version, responds maybe to an RFQ, to an RFP, in this case a piece of rotating equipment, a centri fugal pump. We’ve seen that the user would start with a setup template. So we can see here on the left, this is a simulation method to predict the hydraulic performance of a centrifugal pump, my simulation team has provided me with that method. I go in, I bring in a cat model, SimScale is cat agnostic, in this case we’re bringing it in from, as an XT file.

13:05 We’re assigning boundary conditions, again, also here, I’m not touching any numerical setup, I’m assigning the flow region, no, that’s the rotating region, and then I’m, I, I didn’t need to touch mesh generation, ET, because my methods team has pro provided it with. And then I press and start it, and my first run is going, and you can see in the lower left, this is already a performance curve. So we’re simulating multiple operating points of that centrifugal pump in parallel, and of course there’s colorful pictures. And so, as an application engineer, I can now, within the browser application, do a first performance check on this piece of rotating equipment. You get the idea, right? And again, if I’m not sure if I’ve used this method correctly, have I used the wrong template, do I want somebody else to take a look, because I’m, you know, not sure if I’ve done this the right way, real-time collaboration is built in, so I can share this with my simulation expert, I can share it with SimScale support team.

14:03 Here’s the application engineer going through a first design version, we can see eight blades, 12 blades on the left, same story, I make a design change, again, in whatever, SimScale is authoring agnostic, and whatever editing method you’re preferring, and then the application engineer can already now sort of make a first design decision, right? Again, crucial, in meeting, when, when a white GS manufacturer tries to meet all the deadlines in all the different platforms they are supporting, or when a, when an automotive supplier tries to support multiple different vehicle programs at the same time and tries to sort of respond to all of those rfqs, the other speakers have done the logistics somehow better than me, right, good.

15:04 Then, moving on. So that, that story right now showed us how the application engineer gets down, gets away from, I’m waiting for the simulation team to respond to me with the simulation, down to hours of me iterating through the first designs. That’s still hours, because that, that video was cut, it was a standard final volume solve, so that takes about 30 minutes. Yes, it’s in parallel, so all operating points in parallel, but at the same time it’s still, you know, under the order of magnitude of hours. So suddenly the engineering team starts producing simulation data in the cloud, and we’ve partnered up with navasto, Matias is somewhere in the crowd over there, all deep AI questions go to Matias, just to preempt those, a, oh, no, no questions, anyways, so we’re good.

And so graph Neal net method is directly, sits directly next to the simulation solvers, that means I can now take all of those simulations that my team runs, and that’s par free, right, I don’t need a parametric model. In this case, admittedly, it’s parametric, but I don’t need my team to simulate, in some shape or form, a parameterized model, they can just like keep on running centrifugal pumps, and I can start using that data. Go to SimScale, sort of select the runs, and contrain a graph neural net, physics agnostic, that now gives me a deep learning surrogate model that allows me now to do real-time predictions. And in the lower right, we can see a little link that says link from for action, right?

16:28 So that’s the second demo, we just chatted earlier about that. It’s tricky to do live demos in presentations, but we’ll try, right? So this is now live, if something doesn’t work, it’s totally the conference Wi-Fi, it’s not the SimScale application, right, can we agree on that? Again, but here the story is like, that’s the little rotating equipment, central fle pump, we’ve seen a second ago. And what we want to do is now, my, my engineering manager, my team lead, my simulation team has pro, has trained the first AI model for centrifugal pumps. And so I can now go in, and instead, and let’s say I have a new version of my model, so this one, this nine blade version, in the top left we can see all of the different versions. I went through the geometries on the left, we can see the, the simulations.

17:12 And so what I’m going to do is, this is still an old geometry, and so what I’m going to do is instead of simulating, is, as we’ve seen it before, are going to select an AI model, turn the AI model on. We would need to talk about how AI models are verion and managed and used, but in this case I’m choosing the appropriate AI model. And what happens now in the back end is, instead of doing a full solve, it, it uses a deep learning surrogate, and there’s an inference, right? So within a second or less, we have a first result here, and I can see on the right an AI model confidence, sort of, that tells me, is the geometry that I’ve just thrown against this AI inference, is that within the training space, right, for a more technical explanation, Matias.

17:52 And now the good thing is, like, this is, is now parameter free, meaning I can go in and change things in the setup. So let’s say I want to change the speed, so I’m changing the flow rate, and again, same story, simulation setup has changed, instead of a full solve I get to invoke an AI prediction, and within a few seconds I get, I get an answer. That’s, conference Wi-Fi really sucks, okay, there we go, right? So this one, and then the last one, before, I think you get the idea, right, and, and the last one is, okay, this, this simulation setup now evaluated still an old version of my model. Now let’s look at an, at this one, this was the nine blade one, so you can see I’m changing this, sort of assigning a new geometry to the, to the simulation. The AI model is still turned on, you can see the valute looks a bit different, right, we have a different outlet. And I guess you can now guess what happens next.

18:57 Again, bringing this back to the idea of, we talked in the earlier slides about, it’s tough to, to make a broad fast decision-making, and engineering teams work, right? And frankly, yes, there’s a lot of sort of work that goes into actually build, you know, the, the embroidery thingy, training that model was not so nice, it took eight hours on a GPU to train it, you need to, you outlier, da da da, so it’s not like as smooth that it looks here. But because everything is in the cloud, because data is manage, because I have a digital threat between the data I use for training, I have that under control, right? And I can just gradually start rolling this out, gradually building such model, models in a practical way. I had a backup video, full disclosure.

Moving on. Good, and let’s just wrap this up, I think I’m already a bit over time. All of what we’ve seen, we didn’t talk a lot about methods, we didn’t talk a lot about physics to SimScale is a general purpose simulation platform with, you know, structural capabilities, it’s an imp finite element solver, fluids, four or five different solvers in production, IBM finite volume, thermal, conduction, convection, radiation, all of it coupled, and then first, low frequency emac release happened earlier, no, late last year. And all of that sits orthogonal to the compu infrastructure, user data, process management, as well as the AI capabilities.

20:22 Good, and now a few, sort of, just a few examples of how customers put this to use, because again, none of what I showed was an end to end workflow, as we’ve seen earlier, right, those really nice workflows where you, where you have authoring of the geometry, and then sort of the simulation in loop, ET, said, this is really just, just physics, right? So whatever, sort of, the idea is, whatever you throw against it, you get, at, at marginal unit cost, a very fast prediction of how that product design performs. And, and it allows you to scale that within an engineering team.

20:52 So this is bua, one of our customers from Switzerland, 11,000 employees, with the goal of cutting down 50%, in, of their entire fleet of machines, 50% water consumption, 50% energy consumption, and 50% carbon footprint. And they wanted to deploy simulation very broadly, and within record time deployed SimScale to 120 users. I need to validate record time, if I can say the word, I haven’t validated that yet, next time maybe have a more precise than record time. And also here, like everybody that manages larger engineering teams, or, or thinks about engineering infrastructure, 120 engineers with workstations to do, you know, thermal, CT, fluid analysis, Etc, it’s a pain. So that is sort of really this cutting down lead times to hours.

21:42 Then a good example of R, putting the AI capabilities to use, R, an automotive engineering service provider that we heard about it earlier already in the Sara presentation, they’ve built a foundational error prediction model on SimScale. So on the left we can see a generic core shape being modulated, and then in the middle we can see how all of this has been programmatically assimilated with a finite volume solve, or lce bman solve, one of which, and then an AI model has been trained. And now they have a foundational AI model where, with retraining, that can now be used for, you know, responding to customer specific models.

22:13 Then, traditional shape optimization, here would G you, a flow control, flow control and valve manufacturer, again, shape optimization, we’ve heard earlier that optimization is still, you know, not being broadly adopted. We can see here, again, we didn’t do the, the generation of the geometry, that’s with our friends at cases, at Friendship Systems, there the geometry is being moduled, and you can see in the middle, this is the important piece here, 330 simulations, each of which an hour, so that gets tricky, right? And so that’s why typically it becomes this large engineering project, and somebody needs to sit down and da da da, but because HPC management is included, it just becomes more accessible.

ENT toop, the, we’ve heard Brad and Marcus talk earlier, also here, I’ve, I’ve kept on begging Brad to not produce those very complicated geometries, but he kept on pushing, that was sort of not nice, was a joke, right? We, we’re excited about many new ways of how geometry is being authored, right, and specifically from a physics driven perspective, and I think it’s, it’s a great example where, where simulation plays like an important role in geometry generation. And so we’re excited about the collaboration with enop, equally with Cera, we, we’ve seen earlier in the presentation how they, in another way, provide new ways for engineers and engineering teams to explore design spaces.

23:43 Here we can see a conformal cooling example, something that would be very hard to do in a traditional history based parametric cat model, in scar can just, like, you know, push these models out. And then we can see that now, suddenly, we have this plethora of, of different models, all of them need to be validated and, and simulated, and this is where the integration with SimScale comes to play, and where we can, you know, very quickly produce, like, sweep the parameter space, sweep the design space, see what’s a good, what’s a good outcome. And then also shout out to the team at Octon, because they can’t be here today, similar, another approach of how sort of physics driven design is being done, and also here simulation challenges, large models, complicated models, excited about that one.

24:27 And then the last one before I wrap this up. I was about to use another buzz word, trying to get around this, but, so, but generative engineering is already at the top, right, so past this point now, but I do think it’s a very interesting application. Ksb, a pump manufacturer, basically faced the challenge of, for reasons that are too long to explain, face the challenge of needing to simulate a lot of their product portfolio, and instead of simulating all of it, because it was a regulatory change, it was a change in thinking, and saying, okay, what if we can build generative model where we can actually put in the requirements for the centrifugal pump and produce a geometry out of it, right?

25:06 And without now, again, going into the detailed method, it’s a, we can see at the top here, it’s about a thousand simulation of a, of a centrifugal pump run. Again, we’re talking about serious compute, we’re talking about serious effort if you want to pull this off with a standard stack. But because HBC commissioning is done easily, data management is done easily, you’re not dealing with all of this gigabytes on your hard drive, Etc, it just becomes possible, and they’ve built this generative engineering workflow.

25:30 So, just wrap this up. SimScale is unfortunately not the solution to everything, but it is a novel way of using the cloud, and deploying in a cloud native and a cloud first approach, simulation methodologies, and thereby making them aggressively more accessible. And that is to, you know, more loads to more users, while managing all data, while managing HPC, and thereby really enabling, yeah, the next generation of computational design workflows that we’re excited about. And we kind of feel like the folks at the basement getting the physics done, that’s kind of the story about SimScale.

26:08 And I realized everybody had this nice slide at the end, introducing by name and the LinkedIn profile, and I really just have that, so I realized that might have been not the smartest thing. But we are headquarted in Munich, and our office is 500 meters south of the teresian Visa, where the October Fest is happening, so it’s kind of still on brand, right? But my name is David, you can find me at LinkedIn, or just on the internet, next time. I’ll put it here. All right.