0:00 It’s not every day is that I wake up the morning and I have to walk 5 minutes to be in such an awesome crowd, thank you Duann for bringing all these people here, thank you very much, this is really exciting, and again welcome to Berlin. But second, I got a little bit nervous this morning because I, looking at the crowd, and I’m a manufacturing guy, I’m a process guy, and I saw all this expert in computational engineering and design, I got a little bit nervous because I think what I’m going to talk about is not what you’re dealing with, with. Therefore I decided the first picture I will share with you is this one, okay, because I want to build a bridge with you and share with you why it is important to connect your discipline to the manufacturing process, or process engineering discipline, and I couldn’t find a better example than this one. The raw material is the same, and yet the geometry and the properties are solely defined by the process, this is a pure geometry designed, engineered by the process, so-called recipe, in in the cooking art, but I would assume in manufacturing it’s pretty much the same thing, and that’s what I know one or two thing about, and that’s where my presentation will be focused on.

1:12 But I think it’s sometimes really nice to go back to the fundamentals, and my friend danan told me, Omar, no marketing, just fundamentals, so I hope you’re served. So manufacturing, obviously, is a transformation using energy of raw material to a final geometry, but your geometries, I hope, have a final function, and that function is mainly defined by the process that the process engineer or the manufacturing engineer will Design, so-called process route. Coming back to additive manufacturing, there are so many things actually that are blocked by this process, you guys are able to design and engineer extremely complex component, yet when we try to manufacture them there are few challenges.

2:00 The first one, I think, is the expertise, as you know, if you look at the picture on the extreme right, a lot of successful companies are able to produce component because they have expert Engineers able to know where to put the support, how to orient it, but please don’t underestimate that, that’s a rare knowledge and a lot of people are not able to gain it, and this is one of the reasons why am is not going as as we want. So lack of knowledge, and even when we have that knowledge, that knowledge is mainly built on years and years of expertise, which mean trial and error etc. The second one is we’re trying to bring more materials, and once we understand one material then there’s already a second and the third, and when we want to understand these materials well, we need to know how to process them so that we can get the right properties.

2:50 Someone said this morning, rip 3D printing, Long Live additive manufacturing, well, that’s the challenge we need to overcome, how do we produce this beautiful copper component without what you’re seeing here. And lastly, there is another topic that I see more and more, is the consistency, you guys are bringing quite challenging geometries to this production system, you’re not being easy on additive, and obviously when you look at heat exchangers, and I really love heat exchangers, the first thing the user of the heat exchanger will do is to put it under differential pressure, and if it’s not consistent it will explode, so no wonder why no one is using heat exchangers yet. And what you want, obviously, is your very thin walls to be consistent, so every layer, every section, if you decide that it’s going to be 2 mm then it has to be 2 mm, but if your process recipe or signature or whatever you want to call it is not really engineered, then this is what happened. Well, you could see that the Melt pools became one melt pool and it’s not good.

4:03 And what we obviously want to have, and I’m already going to start my marketing session, is something like this, where all the walls and the wall, all the thicknesses are consistent, and how do we do that, by creating process recipes. So I want to go one step back to tell you what we do in Thousand Kelvin and what is the main hypothesis of the work we do. Obviously all of you hear about thermal management, it’s a challenge in the EV and space, but also in additive it’s a thermal management problem, and we as Engineers need to solve this problem, and our approach is to solve it computationally. However, it’s not an easy one, it’s one of the most challenging in my view, as you could see it’s multi scale, it’s multi-physics, and obviously if you really want to master the process, you want to go to the ultimate level, which is the Melt poool level.

5:01 Simul, simulation Technologies looking at the part level are already available, I would even claim they’re boring, because they’re not really solving the problem, but the more we go to the layer base, scan strategy base, it becomes a bit more complicated, but the challenge here is not the physics, it’s just the computation, we don’t have illimited computations in this world. There was a paper from a friend of mine in France that really explained this nicely, and it’s always nicer to use the words of other professionals than ours, and he explained, using data, and I recommend you to read this paper, that ultimately if you want to go to the ultimate level of understanding we need to go down to the scan strategies etc, but it’s obviously very expensive.

5:38 So what have we done over the last three years, a thousand keving, where we had an idea, not so genius, very simple, where we’re going to use probabilistic compression, that’s what machine learning is doing. If we have great physicists capable of modeling this complex interaction, if we have cheap compute thanks to our friends from AWS, if we have the capability to generate a lot of data, so why not using probabilistic compression and building models that are able to predict all this in a nanc, and I recommend you very very much to listen to the presentation from my friend Matias, because he’s going to talk about the same thing for automotive. So the future of engineering is going to be partly built on probabilistic compression of physics, and this is exciting time for all of us.

6:30 So we built a maze that we’re actually selling to companies, and people are using it, and it’s a very easy to ous Tool, and I don’t know what category is a maze, to be honest, I don’t know if it’s a simulation, if it’s a cam, but it’s okay, it’s a new category and we will see more and more of these products. Extremely accurate, and today I will demonstrate to you the ultimate level of accuracy, and how machine learning is going to become the name of the game. Extremely computationally efficient, I threw a number here, a lot of people think this is a sales number, but it’s a true number, and I will show it to you in the next slide, 10 to the8 faser than finet element, good luck simulating melt poool based tool path on a finet element. Very easy to use, cloud-based, fully integrated with oems etc.

7:16 And because I claim 10 to the 8, I just want to do a little example, because we have this part that we show a lot, so we have all the data on it, and this is typically a part that has some 200ish kilom of tool path, and if you want to simulated use INF finet element, it will take you 24,000 years. With compression, probabilistic compression, this goes back to 2 hours on the same computer, this is massive, this is GameChanger. And if it continues like this, that means that this Technologies are going to go to the shop floor, the analysts have been the consumer of these Technologies, the more they become computationally efficient, easy to use, validated, trusted, I am imagining that the shop floors are going to be different, and hopefully we’re going to see more and more of your parts in our cars and planes.

8:06 So this will not be possible without integration, obviously, and we are really really happy to be one of the first companies in the world that work closely with the leaders, and we have a deep integration with EOS, but also generalization, so we are also very very to bring this to the market through Autodesk and their integration, so we really want this to be easy to use, Easy to access etc. Okay, so concretely, I will take you through an example, very simple example, and then I would finish my presentation on the validation. So this is typically what happens, someone in the world sends to a service bureau, probably one of you guys or ladies, and ask them, I want this part to be printed, then they come back to you and say youo, you know you, we need to change the design and we need to do a few iteration and things like that, and the cost keeps growing, and the interaction gets long, and the lead time, everything becomes more complicated, and this is really one of the main really simple tangible blockers of this industry.

9:09 If we can make this easier, a lot of people will print more, and I hear all of you, and some of you telling me, you know, am, so these are the issues behind. So how does that look, well it is a complex physics problem, the part is being built, and suddenly you will see it here, an intricate corner of this geometry will start becoming liquefied, I let you watch it carefully, so there is something related to the thermal recipe here that is deviating, and it’s extremely hard, if not impossible, for an Engineers to see this, unless you tell me, I could see that, please be my guess, but this is really intricate and complicated, and only with technologies that are capable of predicting this upfront we can solve these issues.

10:00 So how does these predictions look like, well we have the tool pad, we have access to all the parameters, we have a a very accurate prediction, we can find specifically the regions and vectors that will lead to that with high Precision, but the story does not stop there. If we have physics to predict, why don’t we use compute to optimize, so we just frame a simple optimization, reduce X, minimize X by changing y, given the parameters, and we throw this to the computer. Back as a result, the designers, before even printing, they can see where the issue will happen, and you could see the prediction here, could see generally the the the bottom neck in the top, but even that is not enough, you need to see really on which layer the problem will start.

10:43 But you can also verify digitally how this will be fixed, and obviously the proof is in the pudding, as like my customer like to tell me, so we are one of the software companies who really have to do physical things, so we have hard time with our customers, but we like those, those examples. Another example, similar geometry, very intricate, topology optimized, a lot of you like this, you could see clearly on the picture that only a few branches actually had the issue, and the others are good, very hard to predict, very hard to see, but the prediction is able to see it and correct it, as you could see. So this is one of the first blockers that we try to address, how do we make this technology easy to use, little bit less failures, a little bit faster, a little bit cheaper, so that it can go go from highend, high touch, to mid-end, Mid touch, let’s put it that way.

11:39 But obviously we’re not yet there, and we work with a lot of companies in the Aerospace and medical, and they keep telling me, Omar, we need validation, we need qualification, accelerating all this topic, and I have to admit it took me one year to understand a little bit how we can plug ourself into this, because it’s not really obvious, and this is one of the main topic for us at the moment, how do we help our customers accelerate their qualification, certification process, if I may use these terms.

12:08 So there is a well-known knowledge in the industry that actually the thermal signature, well the recipe, is driving mainly the issues that needs to be seen, cold spot, Hot Spot, microstructure defect. So the assumption is, if we know that the thermal signature will will drive the defect as well as the micr structure, and I think there is enough empirical data proven this, then understanding the thermal signature and controlling it is the potential way to solve all this. So the hypothesis here is that a homogenized, and I will explain the homogenized part, a homogenized micr structure will lead to a more homogeneous quality of micros, of of properties.

12:55 So why, why are these companies looking for homogenate micr structure, well there is an economical reason behind, so this is a blade, I see some of my friends here printing blades, and they know they have massive amounts of data. When you print this blades, you end up actually with three data sets, the top, the bottom and the middle, and apparently the FAA, which is the organization that certifies these component, request for each material, micr structure, different data sets, and the cost of of these data sets, I see some of my friends here, is multi-million dollars, so the barrier really to bring am is really expensive, shockly expensive, even that I’m in this industry for many years. So if we homogenize it, we’re going to just make it one data set, and therefore it’s bit faster, bit cheaper.

13:46 So the hypothesis from thousand Kelvin is, if we give you, as an expert, the tool to homogenize this, then you can save money, but then we need to prove to you that we see what needs to be seen, that we predict accurately this thermal process, that we can give you the level of confidence that you can use our tool to do that. And with our friends at EOS, not only we are integrating, but we do a lot of fun research and we look at Optical tomography, so what you see in the right, my right, is an optical tomography image of a titanium component, and what you see on the left, my left, is the AI prediction.

14:24 Now besides the accuracy of the prediction and the efficiency in predicting this in few seconds, there is another layer of amazing things happening here, what you see on the right is the ultimate level of sensing, it’s a gray values, what you see on the left is temperature data, very precise temperature data, which mean now you can correlate, start connecting to micr structure and things like that. Zooming a little bit on the details, you could see the AI is capable of predicting the cold spot, the hot spot, with a high level of precision, and this is, I would say, in my opinion, a breakthrough. I can continue with multiple P pictures, but you could see even intricate cold spot having been detected by the AI.

15:13 So the funny thing is sometimes these sensing are not working, there is dust or something like that, so the image is wrong, but the AI is not wrong. So, and I think there are some examples like this where, you know, if you have a validated physics based engine, well sometimes it’s better to trust the physics. So as a consequence, how do we use this, how do we invisage to use this concretely, this is a fun example where we have done some work with the DLR, gave us amazing data, and you could see that this impeller, CLA impeller, was printed, was analyzed, and you could see that there’s Alpha Beta decomposition, which basically tells you that the thermal signature led to a an undesired micr structure, so they need to do heat treatments to solve this. But what we were able to do is to predict precisely that upf front, here is the region where you will have overheating, those are the type of melt pool you will get, those are the temperatures etc, etc, and we are working with them to see how can we homogenize this and create a uniform micr structure.

16:24 So, so as a conclusion, I think that process is very important for you guys, if we are able to unlock the process, your components and your designs going to be seen more in our cars and planes and trains. I hope so, and yeah, I’m here if you have more questions, I look forward to engaging with you. Thank you for your attention.