0:00 So, hi everyone. My name is Marco, I’m the CEO and co-founder of, ToffeeX pleasure to be here in New York. Thanks. And today I’m going to talk about generative design, in particular physics driven generative design for, with a strong focus on flow dynamics. I will put some emphasis also on the design of the product itself that we are going to present, in a way, what are the key aspects that allow us a smooth integration of this new technology into the workflow, so the ro of the software to be very human centric inside the design workflow. So let’s dive in.

0:35 So mostly, most of us knows exactly what is the, the background, the pressure on developing new manufacturing technique, and also the best way to use this new manufacturing technique, especially additive manufacturing. There is a strong pressure on innovate, on sustainability, this translate really on a pressure, innovation in new news cases. I will be, I will simplify here, but today, if you look at the current aircraft, if we look at all the components involving FL dynamics inside the engine, there may be tens of use cases that have been designed by engineers in the last century. Really, in the next 15 years we need to design hundreds of new test cases, because it’s not about the combustion, it’s not about the exchange, will be also about the production and the storage of the hydrogen, the device for the absorption of the carbon dioxide, all the components related with electrification and the coolant system.

1:40 So flow dynamics is the key, flow dynamics, I’m talking mostly about internal dynamics, heat exchange, chemical reaction, and I’m going to focus about this. This physic is the most complex, and the key for the energy transition, for net zero, and it’s also the physics that the human mind struggle to understand, the struggle to to fore. So here is where all the design for additive manufacturing, all the technique, comes into the place, and this is really the situation. There are new materials ad manufacturing is unlocking, so much today, that in theory, again simplifying a bit, the net zero, a sustainable word, would be feasible today, but there is a delay given by, what is the best way to use manufacturing, what is the best way to use this material.

2:39 And we saw, especially in this conference, all the solution that are arising. Most of the solution we saw, for example, the parametric solution, allow a better exploration of complex design, but we believe that still the main restriction is still the human mind, the intuition and the experience of the mind is strongly dependent on the expertise of the engineer, the specific knowledge on the process and the field of application is, yes, a key driver for the new designs, but it’s also a constrain of the optimality of the outcome of the new component. So the C that we see comes with some limitation. So our question was, what could we add to collaborate, to make this device, even this, this technology, even stronger?

3:38 The problem of getting solution that are unbiased and are quick by the human mind has been only partially solved by all the methodology, I would say, data driven, so the pure, because whereas they give strong speed, so the possibility to the human to iterate faster and to reach to most optimal outcome, they lack of generality. So if you have a neural network trained on a specific problem, it fails on a slightly, even a slightly different problem. So here it comes the physics. What do we want to add, we want first of all focus on what is good and what should be on the humans, and onead we want to leave to the machine.

4:26 So we want to leave to the human the abstract thinking, the understanding on the context, and the critical analysis on result. We want to leave to the machine the high speed optimization, maximizing the search space, which is both key aspects already solved by all the parametric optimization we have been seeing, but we want to add something more, the unbi physics based design. So we want the physics, we want the simulation driving the optimization, and we are here, we out from Imperial College of London, and in a nutshell, the, the core technology works by topology optimization.

5:11 You might, you might be aware of topology optimization being a very well established technique for structural problem, where you have, I would say, a block of solid, and you carve, you remove all the part of the solid that are not strictly necessary to maintain the compliance, the stiffness of the solid. So it’s a way to optimize a structure for compliance, for stiffness, and reduce the weight. We do the same for flow dynamics, the equation behind are way more complex, we are talking about the Nav stocks equation, including heat equation, including chemical reaction equation. This is our, I would say, bread and butter, and I would like to show with a couple of examples now, and then more use cases, how, by governing the flow dynamics into the optimization, we can get a large variety of use cases, and give much value, getting very optimal design for our user.

6:11 So this is a snapshot of our user interface, so imagine this disk is a drag and drop digital file you can upload in our, in our platform. I want to optimize a exchanger for a CPU, so imagine you have this, in this case it’s a disc, where I impinge a flow, in this case is water, that is entering from the inlet and is leaving this part from the external wall, from the sides, and on the bottom there is my worm component, in this case a CPU that I want to cool down through the platform. You can set what are all your physical parameter literally, this is my inlet, this is my outlet, this is the material I want to use, this is the fluid I want to use, no much more, the rest is done by the software.

7:06 The software start simulating the physics, so the motion of the fluids, and use that information live to update the optimal solution. You would see in the next slide how the solution grows and reach what is the optimal structure to guide the fluid, to certain, to maintain, to optimize within certain objective, in this case is a coolant system. So we want to extract heat, so we will increase the heat extraction by measuring the energy, the heat energy out from the outlet, but at the same time I want to use the least possible amount of mechanical energy to cool down the device, I want my pump flowing the water to be as small as possible, or at least to consume as little as possible.

7:56 So this is multiobjective optimization, and this is what happens behind the scenes. This is takes one hour, we speed up a lot here, and as you can see, the fluid found is on best puff to reduce pressure losses, so the mechanical losses, and to increase the heat extraction. So on the, on the left you see, is divided into, the white part is the actual solid structure, in this case we pick aluminum, and the rest is the, the channels, how the fluid is channeled out of the component. It’s interesting to see the bottom part, how the fluid is added through the, the surface for a uniform extraction of the heat.

8:48 What is very important to emphasize is the value we are giving to the customer is not in one single design. This is of key importance in the, in the much wider discussion of maintaining a human centric, let’s say, feature of the software, the user have to maintain, let’s say, the critical thinking, the possibilities of choices and plug and play within a bigger system. So here the shift has to be on being a designer of a single component for months, into being a designer with the possibility of plug and play among many possible different optimal outcome. I say many optimal outcome because when it’s multi physics you can have many optimal solution, one is bit better for cost, another one is better for mechanical losses, another one is better for re ex struction.

9:44 And here is an example, we run these cases three times giving different weights into the pressure losses, or the heat exchange, or the weight, and then we compare it on the, on this graph. As you can see, we shift the role of the user of being weeks and months on one design, that is probably suboptimal because it’s only based on his human experience, into being days, weeks at most, into plug and play with this optimal design and picking the best. When I say picking the best, you see massive difference across aerospace, where compactness and lightweight is fundamental, but they don’t mind much about cost, and automotive, the, when cost is the main driver.

10:32 And talking about cost, you might nice object, yes, those components are very, but they are only, they may have some restriction, they may be too complex, even in the, let’s say, in the space of manufacturing. So big part, a big chunk of our technological development has been on including the manufacturing technique, the manufacturing constraints, inside optimization. Even within manufacturing, the user is able to pick, for example, the resolution of the 3D printers, the material obviously, but also the overhang, for example, and for other more standard manufacturing technique, like the stamping and the milling, parameters are given, choice of parameters are given to the user.

11:21 I’m going to give you a quick example, this is again an exchanger, very similar, just squared rather than circular, where is literally the user can pick from the platform, I want this component to be suitable for meing, and obviously then picking, yeah, but how big is the tool, how big is the drill for my milling machine, and you have different outcome for different choices. But most, very interesting, is also within ad manufacturing, you can leave the optimization completely unconstrained, so creating what is the best outcome, what I would say the nature would do, or you can constrain the optimization, especially for self supporting, or for, I would say, for overhang, as we have here an example.

12:14 It’s hard to visualize, but if you can see, this is a cut, a vertical cut between the first one, where it’s, you even have dome and horizontal dome, and the one on the right you see the structure are more vertical, to respect, in this case, in example, the 45 deg over angle. And again the value we give to the user is navigating this possible design that the user is able to create in one day, because each simulation take about between one and two hours, navigating and checking efficiency also based on manufacturing, automatically also includes a cost evaluation for the overall design workflow.

13:01 So I would like, for the second part of my presentation, just show you a bit of different use cases to emphasize what is the variability, the generality, that a physics driven approach can give to this kind of problem. So this is like a snapshot of a part of all the solution that we have been tackling in the last, in the last months, in the last couple of years. We are dealing with internal flow dynamics, so most of our customers use it for the optimization of he exchangers, and they go from family of heat sinks, more complex he exchangers involving more than one fluid, co plates, but also flow control, so optimizing for flow uniformity, it’s fundamental, especially in manu process.

14:02 First example, from one of our, is the optimization, similar of the one show of ans, you see on the left what was the design, the human design before Tof, and this is what I would do by experience and by intuition, I would put more to increase the surface of contact within this metal structure and the air, P by the F below to co down CPU, that is on to this dis on the right, you see what for exactly the same working condition has been simulated and optimized by T. For comparable, for comparable weight, we got a 31% increase on the heat extraction in the heat efficiency.

14:56 Can much more comp test use case, made Airbus, more complex, here there are two fluids involved, air and oil. As a first S it may look like jid or a parametric optimization, but this is just because it was an optimization of a unit cell and then has been repeated, there is no parametric. Here the optimization started from an empty cube, where, where you, where the user in, let’s say, set the two, the two characteristic of the fluid, so the air and the and the oil, and then getting the optimal design and repeating it within the same domain.

15:43 One of the latest on our development is including chemical reaction, especially reaction of absorption, for direct obvious application, for carbon capture. So the carbon capture, still struggling with problem of scalability, especially in a n energy in is still too high compared to the carbon that we can extract using T. We prove you can reduce the energy you need for the absorption by 60, 70%, obviously the design of the the absorbant may be more expensive, but at least you find a way to make carbon capture scalable.

16:19 Another example, including manufacturing constrain, is this example from Rolls tryce, very classic example, when you have a coolant plate with three batteries, in this case again what human design is putting, a serpentine underneath the three batteries to keep the temperature in a meaningful range for working. We fed into our platform exactly the same problem, so the digital file inputting, this is the inlet, this is the outlet, and let the software run the optimization, and we got in a matter of an hour this distribution of solid, suitable for milling, that was hitting 65% less pressure losses. So you need 13 of the energy to get a function that is even better, because the component was colder and lighter.

17:09 It’s interesting to see, we gave no instruction to the software, and you have, I cannot see here, you have the inlet, this left inlet is where the cold water enter, and and then hitting this warm batteries, you see at the entrance, at the very beginning the water is cold, so you don’t need much surface to extract the heat, so the islands are bigger, while you travel downstream the water tend to be warmer, so to extract heat you see the islands are getting smaller and smaller to increase the surface, this has been done by the physics, no instruction by the user.

17:53 So I been focusing, this is the last one, on using, using methodology to improve the final product, to improve the component, but can be used to improve the actual production method. This is an example, the injection molding, in this case, of a ble, you have a mold where you inject, let say, molten glass, and then you have to cool it down, and the speed you manage to cool down, strongly dependent on the uniformity of the heat and your efficiency of production. Today these MS are simple, again designed by humans, and can increase the efficiency of this M, increasing the uniformity and speeding up also the process of production of plastic and glass, etc.

18:46 So literally meaning with the same resources I can speed up by 10, 20% your production. Rec physics driven generative design, like the one we develop for FL dynamics, allows to explore a vast majority of design, so solving the problem of the human bias, when it’s a matter of designing a complex component with flow dynamics, and also solving the lack of intuition understanding such complex physical behavior, massively reduces the design time. Been talking about hours per mon to ands, optimizing performances, and creates a shift of the engineers between being focused on one single component to the engineer, focus on a system level optimization, navigating many optimal outcomes, and this is what we are pushing progress than.