0:25 Sorry, not on my person. Thanks, TJ, for for the introduction to the to the problem. We will continue on the same topic. Let’s bring it even further, moving from brackets to some complicated engineering components, also. So, in today’s world, a failed part isn’t just a bad quarter. It’s a grounded fleet, or even worse, a failed mission. When stakes are high, we cannot rely on black box AI.

1:00 I will start my presentation with a simple statement. You cannot certify what you cannot reproduce. Maybe this is maybe a very European approach, in a way to to to see where you want to control everything, but this is a reality in engineering world. In real engineering, determinism is a must-have. Generative AI are stochastic by nature. Nothing wrong with it, but we need same input equal to same output.

1:34 On top of it, we need accountability over every design decisions made to identify root cause in case of failure. And most important in today’s world is sovereignty. On-premises and model-agnostic implementation is key to own the data properly. I’m Rushik Matroja, CEO and co-founder of Cognitive Design Systems. Together with two other co-founders, I founded this company in 2021. In my 15 years of career, together with my co-founders, we worked on we solved 150 engineering challenges of Airbus, Thales Alenia Space, Safran, Amidea, Toyota, Mitsubishi Electric, etc.

2:29 We are a software company and we we create software on generative design and design exploration. Our vision is to transform engineering into a cognitive creation. We empower engineers to become more productive and drive innovation by automating automating the mundane. This vision is no longer very far off. It’s more a short-term roadmap now. And that’s what I will show you today. To position the software itself, I represent you I think you have seen this kind of graph over and over the this morning.

3:18 But what we have seen in the in in our experience is that when you are creating a complex or optimizing a complex mechanical or thermo-mechanical component, you will go through concept creation, simulation, manufacturability check, design feasibility, business feasibility, sorry, and with a lot of different iterations. You might explore two, three, four concepts and this all this long iterative loops could be really expensive. So we are positioning our software at very beginning front loading exploration of the of the design space, where we will be able to explore with generative design algorithms such as topology optimization, but also topology weaving and topology enclosures, new new algorithms we developed.

4:11 Hundreds of different designs, and then you can rank them, you can filter out the best suited design which fits the manufacturability, which fits performance, which fits cost, and which fits also the carbon footprint. And that best suited design that you can bring it back into your conventional CAD environment, and then you can either fine-tune it or you can do also the validation with certified solvers like ANSYS, Abaqus, etc.

4:46 With this methodology, we have seen 5x in term of time gain with our customers. So, the whole idea is to to accelerate the time to market, but also capture the company know-how. It’s a part that I’m I will be talking about with a real use case today. And the third one is of course to to design better and sustainable products. To create this, we started with three different core technologies.

5:24 First is our implicit modeling kernel. We connected it together with analysis engine. We have an FEA engine, which is provided by Intact Solutions. We have also our manufacturing analysis engine, cost analysis, carbon footprint analysis, and everything is packaged in a parametric low-code environment. So, you can optimize for performance, manufacturability, sustainability, and cost. And on top of it, we have a layer of AI orchestration. So, you can actually ask AI in natural language, it will do the orchestration for you.

6:01 You just have to take the right decisions. Engineer, instead of being pushing on the buttons, he will be the decision maker. That way it should be. So, the software itself does generative design, as I mentioned, topology optimization, topology weaving, topology enclosure, also lattice and TPMS structures, but also geometry morphing. On top of it, we have added manufacturing driven design, so you can morph the design for specific manufacturing process, either it’s being machining, die casting, injection molding, additive manufacturing, etc.

6:41 Simulation driven design to fine-tune your geometry to match the performance that is required for that geometry. And also special tool for design to cost, so you can understand what’s the connection between the design parameters and the cost cost itself. With all these different tools, you’ll be able to perform DOE and do design exploration, create various different components, and then filter out the right one for you. Once you have find the the workflow that matches your requirement, you can extract that workflow, and you can reuse it for various different components.

7:25 That is something that we have done with Thales Alenia Space, and I will show you just after. So, the software itself input is the design intent, design space, 3D model, load cases, and the the specification in natural language. We will work on mechanical and thermal mechanical components and we will optimize the geometry for machining die casting additive and injection. And at the end we will bring you we will export the the optimized CAD model, design exploration results such as 3D model, simulation results, manufacturability analysis, etc.

8:00 Etc. And you will have also the reports that could be completely automated. That’s what we did with with a European satellite manufacturer who had a specific problem. So they used our software to first of all create a specific geometry here which they wanted to optimize. But they had 80 different variants of this geometry for the all the for the whole constellation. So instead of designing one by one on CAD, what they did is they found the right workflow for it.

8:39 And then they just changed the input parameters. Once the input parameters has changed, they can rerun the whole thing and they have the right geometry in minutes. Something like that has saved them a lot of time and of course they were able to to match the the performance required for that. It’s already in the space. Another example with one of our customer Safran. They they make a lot of different aerospace components including aircraft engines but also landing gears, nozzles, etc.

9:16 We are working with seven different entities of Safran and we have done more than 30 different projects and they we have seen we have perceived and on an average 5.5 X of design acceleration and the weight reduction of 18% which is already quite big for them because these components have been optimized over the years for machining and we still were able to find way to reduce the weight.

9:45 And all these workflows, once we have created them, we are reusing them again and again. Let’s see how it works. So, I have a small video, a demo that shows how it would be doing. So, this is a typical kind of a component of turbo after that you can bring in. So, I will bring it in in our software. So, it it it looks like this. You can bring in the the CAD model.

10:17 You once you are satisfied with your CAD model of the design space, you will be able to apply load cases on top of it. There you go. You can apply load cases. I have put some images, so we we have it for reference, so it’s easier to to understand. You can apply mechanical and thermal mechanical load cases. You perform simulation on it. There you go. This is powered by index solutions.

10:45 We see that the the the von Mises stress here is 193. We perform topology optimization on this this geometry. Same here. We are using index solutions level up to to perform that. So, we have this geometry. All the the post-processing could be done completely automatically. So, here the post-processing was done with a component called topology optimization post process, also available in Simcenter. Then we added some ribs for structural integrity.

11:22 That could be added with implicit modeling kernel. You can do Boolean operations very easily. You perform manufacturing driven design. Here I chose additive, but I could have chose I could selected some other manufacturing process. We go to simulation driven design. We see that the the result of the simulation the stress is pretty high. So, what I will do is that I will do a filter and design operation.

11:49 I will increase the thickness where the stress is high. I can remove also the thickness where the stress is low. I can fine-tune the geometry this way. I can also do various kind of variable offsets, lattice, shells, etc. And at the end I will be validating the geometry with at the end to to ensure that it is it is good enough. I validate the manufacturing analysis also.

And then I will be I will be calculating the cost and carbon footprint. There you go. Once everything is calculated, I can send it that this is one of the the iteration. So, I can send it to to my reporting. And it will come in a in a design exploration window, which looks like that. You can keep adding new iterations. You can do the comparative analysis, stress versus weight.

12:48 You can also do cost versus displacement, other kind of thing. You can change the the axis, and you can filter out the the component that is most suited for you. You can of course also extract the report, which is generated automatically, which is a very simple report, but very with all the the KPIs that are listed. Now, this work of course of course you can bring it in in in CAD as an export.

13:18 You can also create an FEA mesh, so you can bring it into Ansys together with your load cases. So you can perform validation. That’s it. That’s one of the the requirement. So this is cool. This is how our current software works. But what if I ask AI to do the same job? What would happen? The results were quite interesting to say the least. We went from the original part when they were conceived it it was 200 hours of of CATIA V5 I think at that time that they used.

14:02 Instead of that in cognitive design it was 40 hours. But with with connect when it’s connected with with Claude or with Mistral or or OpenAI. I know nobody talks about Mistral here, but yeah, that’s a company in Europe. And they we you can actually run the same kind of orchestration pretty easily. So how it works is pretty simple. I will be able to do the the natural language command.

14:33 AI will I can do the brainstorming with the AI on what are the parameters, what is the material, what I want to explore, what I don’t want to explore, what are the priorities. Once the exploration is done, you launch the the generative design design exploration design automation. You can either launch it in headless with an MCP or you can launch you can create a local execution with by exporting a cognitive design file.

15:04 So you have the whole workflow available to you. Let’s see how it works. There you go. So here is somewhere you can do brainstorming with our with any of the the LLM model that you can choose. Yeah. Once you have done the brainstorming, you’ll be able to it’s creating right now the the workflow itself. This is in real time, not accelerated. So, you can see how fast it is.

15:38 There you go. The workflow workflow was created. I can check the workflow. I can iterate on it. I can modify it if it’s necessary. It took all the load cases. It took all the It took the geometry and based on that it created this workflow. What I’ll be doing here is I will I will export the cognition design file, JSON file, that you can import in you know in the in the software so you can execute it locally without sharing any data with on cloud.

16:14 There you go. In couple of seconds, you’ll be able to to create a new project. You can import a macro block like this one. And once the macro block has been imported, you bring it here. There you go. And you open it. And you can see that the whole workflow with all the annotations, everything is created automatically. I will just have to to run click the the run button and I’ll be able to perform similar kind of analysis automatically.

16:47 So, all the work that we were doing previously by hand can be completely automated. And that’s what I’m talking about that okay, cognitive creation is that you don’t really need to know which in which where is the menu, which button lies. That’s not a skill. For me the real engineering is knowing mechanical engineering, mathematics, physics, and that’s what will be the most important here. Last thing here is that by doing this, we wanted to separate the the stochastic nature of the of the LLMs with deterministic nature of of the workflows of the the physics of the geometry and the the validations that we are applying on it.

17:40 By doing so, we will be able to control completely what’s being executed, how, which decisions are made, and also will be have will have a complete transparency over over the workflow. So, we are able to go back to the workflow, and we are able to to see what could be changed in the future if there is any issue. So, maybe few few of people here know me, but if you don’t know me, you can of course come check me.

18:16 We can help you identify the the benefit. First of all, it could be a simple interactive workshop. After that, we usually run a benefit validation proof of concept. And once you have done the proof of concept, we’ll be able to deploy together with your IT team either locally or on your private cloud pretty easily. Thank you very much for your time. And happy to talk to you.