0:00 Thank you so much, Duann, and it’s so good to see everybody, and such a wonderful set of presentations for the morning as well. Today I’m going to start by giving kind of our view by the way, my name is Brad Rothenberg, CEO founder of ntopology, now called nTop. We shortened our name, or lightweighted the name. I’m going to give our view on how we how we think about computational design, computational modeling, and then I’m going to be joined by one of our local customers from Seaman Energy, Marcus Key, as well, to kind of talk through some applications that he’s been working on, and then I’m going to finish up with some peaks into End Toop five that’s coming soon as well.

0:39 So, you know, first off, the way that we think about computational design you know, we think computational design is best applied to the really hard, physics driven problems, where iteration is really key through the process. And, you know, this is a AM conference, but I thought it would be interesting to start with a nonAM example from one of our customers at Tesla. You know, I think the Giga casting kind of exemplifies these problems, and they’ve been working on this for much longer than, you know, we’ve existed as a company, even. But, you know, taking 70 individual metal stampings from the rear underbody of a car and redesigning that into two gigantic metal gigas components requires like consistent iteration. It’s a very physics driven problem, GE, the geometries. You know, back in my days in architecture school, we said form follows function. You know, it’s very much: the form is driven by the simulation, by the physics.

1:36 But moving into AM, you know, AM introduces even additional complexity into the process. You know, you’re forming the shape and the mater you’re forming the material while you’re generating the shape. You can print shapes that you couldn’t design, or you couldn’t manufacture, with other processes.

1:54 And, you know, when you look at kind of the current status quo engineering tools, and, you know, the the the way that they were built, they were not built for computational design. Back in the, you know, long before I was born, and probably almost many people in here were born, they weren’t built for computational design. And, and, you know, while great, I mean, I started ENT toop because I was obsessed with CAD, really into 3D modeling software. You know, the manual process, CES, the manual modeling of low-l design features up front, it’s, know, it’s very limiting to the process. You need to manually draw: this is a screen grab from some of our customers at locky to Arrow, modeling the bulkhead ribs, and they literally manually Click on each rib and design fillets for each of those ribs.

2:45 Second, it’s very tricky to kind of set up relationships between geometry and a parametric CAD system, like, for example, this is a this, this is a manifold for a rocket intake, and what, you know, you’d want to do is have that the manifold diameter changing over the course of the curve while it joins together, based on kind of some target pressure in and target pressure out, where those two pipes joined together, that parameterization might break because of the way it needs to be defined, U, with a traditional model. And then, lastly, capturing the physics that’s driving these parts and products is very difficult. It’s difficult to capture. And then, even if you can capture it, using those physics to drive the relationships between geometry is also, it’s, it’s very difficult as well.

3:34 And so, when we think about computational design, you know, I think of it as a as a new process, right? And so, looking at the kind of oversimplified engineering process, right, like, you, you start with the design, you mesh it, you simulate, you start with the design, redesign it, mesh it, simulate, etc, etc, etc, and, you know, more iterations should ultimately lead to more performance. Each iteration, you’re learning something new, you’re learning something new, you’re learning something new. And so, within a computational design process, the way that, the way that we think about it, is really three, three steps: one, you build a computational model, so you kind of encode the logic of the design with inputs and outputs. Once you have that model built, you can then run it to come up with what the optimal design is, based on some target, you know, fitness criteria. And, lastly, you need to integrate that design into a larger system, right? So you need to go back into your CAD system for the assembly, you need to go into CAE for further downstream validation, maybe you need to do fatigue analysis, creep analysis, etc, you also need to get into Cam tools for manufacturing as well.

4:44 And so, you know, the current processes ultimately leads to this business problem, where, you know, you’re only getting a certain number of iteration in per time that you have, right? It’s manual. With the computational design process, you know, maybe it takes a little bit longer up front to define that model, if you could also draw it in a CAD system, but the power of this, you know, process is that once that model is defined, you can rapid fire iterate through the design, kind of automatically, or in collaboration with the computer, to kind of get to those better results. And so, you can get an order of magnitude more design iterations in the same amount of time. And so, you know, if the limiting factor in a traditional process is how quickly you could model something, the limiting factor in computational design is compute: how quickly can we compute these results, how quickly can we, can we solve them.

5:38 And so, at End Toop, we kind of think about building tools purpose built for this computational design process. First off, we have a software, End Toop, it’s the desktop software, based on this really powerful modeling technology, which I’ll show some previews of, of at the end here, using sign distance fields to capture the geometry. So it’s fast, built for modern computers on GPUs, and it’s very robust. Second, we have a product, End Toop Automate. You could think of End Toop Automate as like End Toop robots that are cloud deployable, to process and run these models for you to create the results. And then, lastly, we have a product called nTop Core. ENT Toop Core is for our partners, to kind of integrate the implicit model into to their design ecosystem, and we’ve had some exciting announcements last year, and some more exciting announcements to come as well. And that, it’s a relatively easy SDK: you read in the End Topop generated implicit, you can query the implicit for its distance, or for its gradient, or for you can extract a mesh, or you could extract contours from that.

6:45 And so, definitely, this is open to all of the kind of new tools in the room as well, excited to for you guys to integrate and build on top of some of the work that we’ve done, because we put a lot of time, a lot of effort, into producing these tools, really built, purpose built, for computational design. And so, you know, as of today, we’re in about a little over 400 customers, kind of across the space, across aerospace, medical, automotive, industrial, consumer and manufacturing. And so, I’m really proud and excited to be joined, being as we’re in Berlin, my first time in Berlin was actually to meet with Marcus and the Siemens Energy team several years ago. It’s just been exciting and impressive.

7:27 I mean, little quick story, which is, when I first met Marcus and the team, they were like, okay, we’re deploying 3D printing to help with the conversion of our large gas turbine engines to hydrogen power. I learned last night that the hydrogen power is basically for better energy storage, so you can collect the energy when it’s cheap, store it in hydrogen, which is obviously renewable, and then you can burn it when the energy is expensive. And it’s was really exciting to hear that, now have large gas turbines running fully on hydrogen, with a lot of 3D printed components, enabled by 3D printing, really. And, you know, these large gas turbine engines, I think, are some of the most advanced products I’ve ever seen in my life, like, I think they define simulation driven design. Imagine, think, like the jet engine, like a 7 S trip 7 engine, is probably, like, I don’t know, the size of this screen, we’ll think about that, but the size of like a tram, that’s how big these engines are. It’s unbelievable. And so, with that, I’ll have Marcus kind of explain some of the work that they’re doing.

8:32 Thank you. Yeah, thanks a lot for the introduction. Actually, before I dive into the applications, I’m have a public service announcement to make, because we often realize that that people don’t understand yet fully that Siemens and Siemens Energy are now two completely separate companies. We both deal with AM, however, Siemens is really mostly focusing on the on the software side of things, whereas our goal, coming from the gas turban business like we just heard, is really around serial production of LPBF parts, in particular, and not only do we do that for ourself, but through our service provider and our locations, we also service external customers in the aviation and space, automotive, and the tooling industry, which gives us quite a unique situation where we really see, U, an array of of applications, but we can leverage also our internal knowledge and offer it to external customers.

9:51 And just a quick view: we have a global setup, and just to stress the the serial production part, this is by now not the most accurate slide anymore, but you can see we really have large printer fleets. We have 28 LPB, more than 28 printers, in Wer, and 19 in Finspang. Those are our serial production facilities, but we also have smaller locations, and in particular the one in Berlin, where we do a lot of our R&D work.

10:35 Okay, so much about AM at Siemens Energy. And here we are again, another feed exchanger. For me, personally, also, that was the gateway drug into implicit modeling, because the, this definition that we talked about was a signed distance field, that’s really ideally suited for the the triple periodic minimal surface structure. What you see on the left hand side here is one of our latest designs, and I mean, you can always argue how large does the STL file really have to be, but for me to be content with the quality, and that I’m sure that when I put it on the printer I will not see facets when it comes out, I came up with, okay, roughly need to be around 5, 5 gab in the print file. Even if I run CFD on it, I still need to export the the fluid domains, and both of them come also out at at least 1.5 GB. For me, that’s just too much to handle, I have my, my hard drive is full, and sending that to the printer is a real nightmare. When you think about Ser production, whereas with the implicit format of End Topop, where you basically then have it represented as the sign distance field, the print file is 6 megab. I can send it to an EOS printer, get sliced, and and and vectorized, and what you see here on the right hand side, that’s actually a larger prototype model that we built, it filled up nearly a complete m290, and there was just no way in hell that we could do that with an ST, at least we didn’t want to. I know there’s people out there that do 30 – 40 GB STLs, but I don’t think that’s a, that’s a good approach.

12:29 What is also very special about the the implicit modeling to me is a, to me as a designer, it gives a lot of of freedom to adjust the geometry. I me, we’ve also seen just in the demo, right, how with the point cloud, you can basically affect the structure, and the same is is true here, just we’re not using points, I’m using Fe, and basically it’s again the bias that that we just heard about in OV, it’s called the the wall offset, right, like how far do I put the the center wall to the left and the right. And by just having this field that gives me different values, I can then really blend geometry into each other, right, I can have, I don’t need even a belel anymore, I can just go ahead and have the the two walls collapse and the channel vanishes just by itself. That’s basically what you see in those 45 and 35 degree regions here. At the same time, you also see this shade going from green to orange, that’s also just changing the bias, and that is important in a heat exchanger, because with the temperature the density changes, and therefore the volume flow changes, and all this has an impact on your heat transfer, but also especially on the pressure loss. And if you want to have a maximum heat transfer for a minimal pressure loss, this is one of the lever that you can work with.

14:09 Compared to a lot of the other examples that we saw, this is not fancy at all, that’s really a very simple heat exchanger we’re looking at right now, usually they are conventionally manufactured. However, we do have customers, internal customers, that very often have a very unique situation, like they need a very unique shape, or it really needs to be fine tuned to certain operating conditions that they need, and so additive still can be a very viable option there for them. So we were approached, and in order to find the the optimal design, I looked at, okay, how can I set up a very lightweight parameter optimization, right? I first said, okay, I want to build it upright, so I’m going to take care that the walls are at 45, 135. I wanted to have some stagger, and some basic dimensions that I can change, I ended up with four parameters, and what you can see in the GIF here, that with just these four parameters, you can actually create quite a lot of different shapes already. And in the beginning I also tried to do something like this in a traditional B software, and I can tell you that didn’t work out so well. It’s implicit, the implicit modeling approach here is really much more robust, which I think is an absolute must when we’re talking about optimization.

15:45 I then went the very traditional route: you can go from the science distance field to an, to an, I imported that in just standard CFD solver to get the the temperature and pressure loss information, and I then used an optimizer to close the loop and vary the parameters to get a design that I wanted. In this case I went for a Paro study, just to figure out where where I really want to go, I have two conflicting targets, pressure loss and te transfer, as you can see, and the bottom left, the horizontal and the vertical lines, these denote my limits that I have to stay under, and there’s only a very small portion of the par front that actually even is in the area where I need to go. But what I really want to highlight here is that I came up with quite a robust tool chain, however optimization chain, however, given that it’s a very simple example where I took a symmetric sub model, it’s still already takes quite long. Yes, I’m computing that on my local workstation, but still, if you look at the time that is spent for just converting the implicit model into a surface mesh, and then volume meshing it in the CFD solver, and then having the CFD solve, you know, we end up with an average time for the loop that this 1 hour, I calculated 200 designs, you can do the math how long that took. However, it is quite robust, I mean, for for me in doing optimizations, only having 20, 23 out of 200 designs fail, that’s already a good day.

17:43 I also have to to say, this is a real example, so I didn’t take out anything, eight of those designs, that was just because the license server was down, so that’s, that’s not the fault of any software, that just happens, right? But all the rest of the failures, they actually revolved around surface meshing and volume meshing, which gives me the conclusion, okay, this meshing, that’s something that is something that takes a lot of time and is error prone. So is there maybe something we can do about it? And therefore I’m really happy that we could also partner with with Cloud Fluid, that are currently partnering also with Anop, to look at a different approach. I put the meshless in in quotes at the specific request of Max, because of it’s not, it’s not really meshless, right, it’s it’s basically, you can call it a voxal grid, but that’s a completely different thing than a finite volume mesh in the end, right? So on the top you have again the traditional approach, and at the bottom you have this tighter integration between Anop and Cloud Fluid that eliminates certain steps completely, and also the solving of the CFD is actually very performant, working on GPUs. So very important is that ENT to U that Cloud Fluid, that’s also an L bman solver, you basically end up with Lees compared to the rans, so you get instationary results, and this can read in directly the implicit file, right, which already means that I can just cut out the the surface meshing step, and then it actually uses End Toop Core and the functionality there to do the voxelization, and boom, that takes less than a minute, compared to the 15 minutes I would spend on this specific case for the volume measuring. The the solving is also working really fast. I have to admit there’s still a lot of things to to work out, this is in an exploratory state, but these examples have me quite happy already, and I wanted really explore that more. I really think, and we heard a lot about meshless processes, I really think that can also be a huge time saer, which would be very important for everything with the optimization.

And now finally, back from the heat exchanges, more way to our core business. I, I brought this VE for you that we’ve been, well, this one, we, we’re mainly using, I would say, as a technology example that we can freely show everywhere and share with everyone. Obviously we have much much bigger ones as well. So this is a supposed to be a way one that you would see after the combustion system in a gas gas turbine, and obviously, due to its shape, we have to use quite a lot of support structures, to support it, to make sure that it doesn’t war in a way that we don’t want, and just for the heat to get conducted to the build plate. On the left hand side you basically see where we’re coming from, that would be our regular approach, with just solid supports, and this can be have a, this has a very huge cost impact, right? We print them, well, we have to design, design them, then we need to print them, and then we need to remove them, so three times it’s just pure ways, and whatever we can take away of those, it’s going to be really good for us. So the idea was to use a thermal driven lus lettuce optimization, that we use End software for, to come up with a way that we can actually have a TPMS structure, rather than the the solid supports, but still be able to hold the vein in place, fully supported, and eliminate hotpots that would otherwise occur in the in the build process, right?

20:53 So this is the example I’m showing: in the first step, we only actually actually started with a with a gradient in one direction, varying the thickness, but already this gave us a really really strong benefit, and just showed that what we have had done in the simulation, we could put it on the printer and get a real real part out, and the benefits were already huge, with 50% less support material and 20% less sprint time, and for us in cial production that is that means a lot of money that we’re saving. We we’re now taking it one step further, that’s the the overall flow chart, and we’re not only taking the solid supports and mating them luses, now we’re basically, together with, well, Anis, developing it with our input, on our example for us, we are looking at a combined topology optimization with the letters, so basically, there’s now you can either have void Letts or solid material in the optimization, and all this happens in a lettuce meta model. So in the En software, the process simulation, that’s completely done, the optimization is completely done, without any defined geometry, but it’s it’s done with a model, and then in the end we end up with the the the topology shape, the new shape, and the density feel of how dense the lettuce needs to be. And this is then again where ENT Toop comes into play, with this very elegant description of the TPMS structure, and the ability to do field driven design, we just take in the topology body and a field, and like this we can then create the final geom, R, and this again would probably create, if you mesh it well, we know that it creates a large STL again, but with the implicit interop we again now have the ability to go directly to the printer from a very small file.

24:44 And as my final slide, just really the latest and greatest that we just produced on Saturday, this is what it then looks like if you, if you do the the optimization together with the with the topology optimization also, and because that’s work in progress, we don’t have numbers yet, but as I told you, that was 50% already material reduction, and we’re looking at 30 to 40% more now with this kind of support structure, where we also go to very very thin walls. Yeah, so I really just can say, all those people that talked about, well, we need a a paradigm shift in design software, I’m a designer and I agree with you 100%, there’s good things out there, we started to leverage implicit modeling and field driven design in nTop, and that’s become a really huge and important part of our computational design to box, so we would also be very happy about what Brad suggested, you know, have integrations, collaborate wherever you can, because I think that will create then even more value. A

26:05 Thank you, Marcus. I me, I think what’s most impressive is the control over the fields and the field driven design, to to kind of get that precision and accuracy in the design, the precision and control over the design, which I’m going to come back to and touch on at the end of the this, but wanted to start just to review. So in 2023, you know, we shipped over 130 new features, we released software roughly every two weeks, not quite 26, I think there was a couple misses on releases potentially, maybe holidays potenti, something like that. But three new features from 2023 that are getting some traction in the market: one, our field optimization framework, so this is a multi-objective optimization framework, built kind of as an extension of our topology optimization, that allows you to to use kind of the meta material properties of lates in the optimization framework, and there’s actually two PhD students here, Peter and Rebecca, who just published their PHD, they did an internship with nTop, contributing to this framework, the term is called dehomogenization in the industry, so I urge you to find them, o students, they’re in the crowd here, and it’s really really impressive work. Second, we released End Topop Automate, which I had mentioned earlier, and nTop Core, to enable the implicit interop.

27:26 And so I’ll start with a couple of enhancements to our to our software. One, obviously when we’re dealing with precise geometry and control over the geometry, you really need to understand what it is that you’re building, and so we’re introducing a number of new fields to the software, a better viewability of the fields, things like thickness, draft angle, etc. On the analysis side, we’ve been working with some of these, you know, we put in quotes mesh free, it’s not mesh free, extended finite element method, still uses a mesh, it just happened to use a mesh that’s maybe more suitable for GPUs and parallel compute, which is a structured grid. This is developed kind of in collaboration with Kurt mate from CU Boulder as well. This is just kind of showing, by cutting the cells to the mesh, the analysis could handle kind of much more complex structures like these thin shell, so GRS and stuff like that.

28:31 Second, twoo, is more implicit interoperability, and so in 2023 we announced integrations with EOS and Autodesk. Autodesk will go live in about a month or so, which will come out in Fusion, to be able to read in and top implicits into fusion and do downstream operations from those implicit models. 2024, we’re working with about 15, a few more than that now, and top core part ERS, and the the goal is really to build out the implicit ecosystem, so that more tools could read in implicits, slice them, do analysis on them, go back into a CAD system, make drawings, do CNC tool paths, etc. Additionally, CFD is a major topic to our customers as well, you could see in the heat exchanger examples or fluid flow examples, so really excited about integration with scale, that will come out, David is in the crowd, and then GPU accelerated fluid solvers, Max, from cloud fluids in the back back there, definitely say what’s up to him, again, not not meshfree but using a voxel type grid approximation of the geometry, which again is more suitable for GPUs, right, with with the regular grid, so this is very fast. We’ve seen some example problems that would have run in in, you know, weeks in a normal traditional finite volume based CFD system, running Cloud Fluid in about a day, or a little bit less than that, and in some cases we can bring it down to an hour.

30:03 And then, lastly, coming to nTop Core is also parametric interoperability, and so this was a demo we made in the early days of nTop Core, where some of the geometric parameters that drive the model can also be exposed for further gradient based optimization process CES, and you know that exists in nTop today with the field optimization, it’s a term that I think Blake quarter may have coin, differentiable engineering, or maybe maybe that’s a ter, I’m not entirely sure if he the term, but it’s a, I think it’s a really good term, where the idea is to kind of trace back the to to optimize over key design parameters, right, like if you have some design parameters, you have some fitness results, and you have a shape that’s output, you can basically, using the derivatives and the chain rule, kind of tracing back the derivatives all the way from, you know, the actual, you know, from the fitness to the shape to the parameters. nTop implicit via nTop Core gives you the derivatives of the shape, it gives you the parameters, through our nTop Core partners, they can then evaluate the fitness of that shape itself, so that you can do further optimization.

31:28 And so this is kind of an example where field optimization is doing this, there’s happens to be one parameter here that’s spatially varying across, this is an overly simplified canal beam example, but it’s one parameter varying the shell thickness of the beam that’s optimized. And so, by doing this, you’re, we’re basically decoupling the geometry from the physics, so that the idea is any implicit model that’s parametric could be tied to any physics optimization, to kind of create these more optimal workflows. And then lastly, we’re releasing nTop 5, so nTop five is coming out in a few months, and the features around nTop five, more to come soon on this, but really the three areas we’re highlighting, we’re doubling down on the precision and the speed, and introducing more of this interoperability.

32:16 And so just a little demo here, this was a model that I had made on an airplane on the way coming here, looks like it’s a little bit pixelated from the from the video, but this is a a simple parametric implicit, and you know, as you can see, you can change the parameters, they update in real time in the model. Okay, now it’s actually a better view, so it’s it cleaned up, so you know you can edit these parameters, this is evaluating the model on the GPU in real time, and any of these design parameters as you change them, right, they they theoretically could be tied to these kind of blackbox optimization that I was mentioning. But one of the issues with nTop is, you know, as you zoom in, you kind of see the voxelization or the rendering of the model, so you would need to press contrl h in nTop to get the high resolution view. We built a new, we put a lot of time into a new modeling engine and a new rendering technology, that basically if you switch it on, it renders it in high precision in real time as you’re navigating through the model, and so that enables you to kind of control features that are, you know, in millimeters, or you know, in very very small, over the whole course of an airplane, like if this was some kind of cooling system or antenna in the nose of the plane.

33:40 And so, you know, when you open that as a model, you know, we have these packaged custom blocks, you can kind of see inside of the model, you know, you can start to see all those features resolved at the precision of the actual geometry, because again, that’s the power of the nTop, the implicit is that you’re resolving actually smooth surfaces, sharp features, sharp corners. And so again, if you zoom in here, there’s a part of the plane is cut out in the nose, just to kind of highlight how how precise and small the features are, oops, was the wrong direction. So you can, you know, if you zoom in and turn on that antenna view, so this happens to be a conformally mapped structure to the nose, and again, you can see, you know, if you zoom in it re resolves, re-update the rendering, again, with like perfect sharp features, precise. And so again, with nTop five, it’s really about, for us, doubling down on the precision, doubling down on the control of that geometry, and giving the tool set so you can view and see these while while you’re working, while you’re building that computational model.

34:58 And so with that said, super excited about showing this to people, getting more customers using the software, and it’s it’s awesome to see everybody here. So thank you.