0:00 Awesome. Thank you to So like Professor Ali, I also had cold so my voice might just go in and out. Please bear with me. I’m Karthik. I’m a specialist engineer from Eaton. And yeah, I’ll be talking about whatever you see here plastic to metalto plastic conversion using AI implicit CAD and meshfree simulation methodologies. I come from Eaton Research Lab which is a central R&D solution for all of Eton North America and and part of global but I don’t think people are really aware of who Eaton is.

0:32 I’ve not had the opportunity to network with most folks because I’ve been trying to save my voice for this talk but I I noticed something. You know Eaton is historically known as an electrical company or power management company. So, it was interesting that I was waiting at the men’s restroom this afternoon and there was this eaten circuit breaker protection panel over there. So, I got curious and I started snooping around the building and the security guy was following me because I think I went to spots that I’m I’m not supposed to be yet.

1:01 I I found another transformer which is also made by Eton. Whoever finds this Eton transformer might I don’t have t-shirts to give away for the winner but you know this building alone uses two of these EN products but you know over the decade Eon has grown more than just an electronic or electrical company it has been a you know transition into a global power management company but also an industry technology leader mainly through targeted acquisitions.

1:28 So we now have portfolios spanning beyond just electrical. We have solutions in technology solutions in the aerospace sector. Air filtrations for industrial applications and e-mobility and vehicle applications as well. So that is Eaton in a nutshell and like I said I’m a specialist engineer from the R&D sector of eaten. So we we kind of you know innovate on materials manufacturing and design and cater to all these different business divisions within Eon.

1:56 What I’m interested right now or at least my group is currently interested which is I come from the materials and manufacturing group within the EEL R&D center is how do we use high performance polymers and composits for as a metallic replacements for all of these different divisions aerospace you know aerospace is a big problem for weight reduction and cost cost reduction. So they are very interested in using high performance and high strength polymers for some of those applications and we also have a lot of applications in the power electronics and data center cooling as well.

2:24 So how do we make heat spreaders with thermally conductive polymer composits that that’s lightweight that’s also sustainable and more cost effective and then you can see a bunch more electrical and automated applications in engineing housing casing small interface materials and sort of so wherever we can think of deploying multi-functional polymer composits that can do just more than you know providing strength but also can be electrically conductive electrically insulated in some of the applications but also has strength and thermal management solution capabilities as well.

5:12 So that is where we’re interested in high performance polymers for metallic replacements. But if you look across all these applications, the common challenge is how do we go from the traditional or existing legacy metallic components to the polymer counterpart without losing its functionality, manufacturability and the the structural integrity or the safety of these components. So that has been the underlying challenge and you know a lot of metallic replacement is not a new topic.

It’s been over from many decades. But how can we innovate on the materials and manufacturing side and use digital technologies to accelerate the conversion process is really what I wanted to you know the story is on that and I want to share that with you folks. So the approach to metal to polymer conversion you know it it just boils down to how can we integrate materials the physics of the material the physics of the manufacturing and the design aspect.

5:24 How do we couple and tightly couple these three aspects together from a digital standpoint so that they can you know automate and iterate within themselves and give us the best outcome that is not only lightweight but is also functional and manufacturable and you know meets the performance CTQ of whatever the operating conditions are you know traditionally metal design process we have predefined manufacturability rules we have deterministic FBA solvers that we can rely on we have very known understanding of metallic process properties which are you know primarily isotropic in nature.

So it is some somewhat easier but we can apply those same principles when we come to plastic or polymer composite lightweighting. So how do we capture for instance the materials anisotropy you know especially if you use carbon fiber reinforced polymer composits? How do we capture the anisotropy or the direction dependent properties and that sort of physics into the into the design process. And then you know we do a lot of composite manufacturing.

We do injection molding. We do robotic carbon fiber tape layer printing. So all of these have their own manufacturability tool tooling or process inspired or induced manufacturing constraints. So how do we take that into into the picture as well and then coupling just these materials and manufacturing problems unlocks the design space but also makes it complicated in exploring the design space and coming down to the to the best solution.

So that is sort of the underlying or the hard problem to solve. And then on the simulation side you know based on the the performance constraints we might have the design subjected to static load dynamic load impacts. How do we capture those multi-ysics multi boundary conditions into the into the design iteration process. So those were some of the challenge challenges that we were trying to address. And you know it all comes down to the the design and simulation bottleneck.

5:37 You know, I kind of copied some of these slides from nTop and a bunch of other presentations, but you know, the traditional CAD modeling and FBA loops are really slow and they don’t really enable rapid design iterations. So, you know, we using traditional CAD packages. There are chances that the CAD updates might break down during the the automated iteration process. There might be machine failures during the simulation workflow.

And then just the two things together add up a lot of simulation and computation time. And then organizationally you know we have different experts working on different problems. So we have like a design engineer working on the design update. We have machine specialist who is an expert on a specific software and we have a simulation specialist and we need to bring all of these folks together and that adds a lot of lead time integration of multiple expertise into the conversion process.

6:28 So how do we mitigate this bottleneck from a design simulation which is the iterative overload and also organizationally how do we consolidate these expertise into one unified conversion workflow is is the challenge and that is currently adding weeks and months of lead time to the conversion process and that can go beyond you know you know just from the R&D standpoint taking to the production level it can span over many months to years getting that product out to the customers.

6:55 So how do we cut down the lead time? How do we cut down the cost has always been the challenge with the with the traditional process and that is where you know we really wanted to move more into implicit CAD and mesh fe as as the backbone in in accelerating and mitigating all of these issues right now with the traditional simulation approaches. So for starters like professor Ali mentioned to be able to update the design iteratively without having failures is moving from traditional CAD B representation to more of an implicit modeling.

So we we adopted nTop into our workflow for these lightweing process and then most importantly which is kind of like the core essence of this topic and Neil will speak more on the intact solution. So they offer meshfree alternative to to structural FE analysis and that alone has allowed us to save time from you know almost 40 minutes on this on this box. I’ll talk about the box later to cut down to 2 minutes in just simulation time.

7:56 So being able to same predict the same stress field at that time is a significant boost in our in our iteration process. And then the other thing you know just wanted to spice things up. How do we go beyond the traditional sequential based optimization and how do you bring AI into the picture? So I I we were interested in maybe looking into algorithms like Beijian optimization. How do you have Beijon optimization?

8:20 Use these tools through API communicate with them and train an AI model in the back end and can then also accelerate the design conversion process from metal to plastics and that just integrating these three tools has allowed us to go from weeks of design optimization to few hours and I’ll I wanted to talk I’ll show you some some use cases in leveraging these three tools together. So that’s just on the on the design you know side of things like we have the Beijian optimization that you know it starts with the design bounds it speaks to NTOP through Ntop automate through Python coding so it can update the design during the iteration process and then we call intact through PI intact which is another API through Python.

9:07 So we can quickly evaluate that design you know in two minutes or so. But rather than going through a sequential optimization on the back end we decided to take those data and teach an AI model like something like a gshian process model so that we can train an AI model make some best guess on the design and then check for feasibility whether it’s through strength or thermal conditions that whatever the the operating scenario is.

9:31 If it doesn’t meet, we go back, take the data, retrain itself, and then that just goes on in an in a continuous loop. This is not something new. I mean, none of what I’m presenting is new. It’s just more of an app application of different application and integration of different tools. But what was more interesting is how do we incorporate the material and manufacturing physics into this design exploration pipeline and that is the the left branch of of things, you know, for for composite injection molding applications.

9:58 Now we started bringing in other commercially available tools into the picture. So for instance we have autoes mold flow which also offers API. So we are able to bring in the geometry and we are able to you know u apply the specific injection molding constraints. For instance based on how we put the nozzle into the mold cavity that will determine how the the plastic flows into the cavity and that will determine the fiber orientation and things like that.

10:25 So to capture that and map it onto the structural geometry we use Digimat’s API as well. So they talk to each other and it gets the mold flow results and Digimat helps create the fiber mapping information and we take that fiber mapping information and further map it through API with Intax FE meshless FE simulation. So now we can apply the boundary conditions while capturing for the material and isotropy within the geometry.

10:49 And then how we bridge the two branches is we can now you know sort of predict the stress field point cloud map that is being result that is a resultant of coupling the manufacturing and material physics on the back end and nTop has this fielddriven design approach. So we able to take that stress field and link it to the the AIdriven pipeline. So now we don’t have middleman speaking on way to add or remove material in the design optimization process.

11:18 So we have this field-driven design dictating and telling nTop and the AIdriven pipeline on where to remove material, where to add material and then that just feeds the the iteration on its own. So that’s just kind of the the workflow that we built that allowed us to go from weeks to a few hours in in the design conversion process. So one SL on one of the applications.

11:42 So this is a a box that houses some electrical components. So we need it to be electrically insulated but also strong enough to contain any gas leaks from those components. So it needs to be explosion proof ceiling explosion proof not in the in the boom kind of explosion just to contain the gas leakages from the components. So how do we go from a box looking to a slightly cooler box that is lighter but also meets the the material and the manufacturability and the and the safety requirements of that and we were able to achieve that in few hours.

12:17 And that’s just one single iteration. But the integration of the mesh free and the implicit CAD allowed us to rerun this many times. So we have to speak with our external manufacturing supplier who had a lot of issues. So we go back, we add those things and you know we can come up with a new design in in less than 2 hours. And we were able to redo the design until we met both the the customer requirements and the manufacturability requirements as well which wouldn’t have been possible.

12:44 That would have taken us like 6 to 8 months. We were able to get down to like one to two months with this entire workflow and now this is being produced right now externally with with a manufacturing supplier and yeah like I said this is a box I remember when we first showed this to Duon he was like oh this is all cool guys but it’s it’s a box like what what why do you need such a you know sophisticated AI implicit CAD mesh free and all of this stuff I agree like I said the win for us is being able to cut down the development time from months 6 months to less than 2 months.

13:16 That is a huge win. We saved a lot on cost and the time and we’re able to get to manufacturing duty quicker than we were able to before. But then most importantly, we were able to build this workflow put together a more robust workflow of integrating implicit CAD mesh free analysis and the AI as a backbone that learns from its solution and gives iterates through the design optimization.

13:38 And now we are able to apply it to more different more complicated and expensive problems out there within Eaton. So this was just sort of like a proof of concept within our Eaton research lab and now we have the buy in from our leadership on deploying this tool on on other applications as well. For example, now we are we use this lightweing or the you know this is not really a metal to plastic conversion workflow.

14:01 It’s a it’s a lightweighting problem. It’s metallic 3D printing. How do we use meshfree and implicit CAD to lightweight this aerospace accumulator? So we have you know this vessel is going to be 3D printed using robotic carbon fiber AFP process but then the endcap or the enclosure they have many names for that but that needs to be tightly tightly put into the vessel and we wanted to 3D print that and explore opportunities for lightweing.

14:29 So we were able to use nTop and intact integration the API integration to you know within few hours we were able to go from that design to the one on the right which is 45% lighter but also 20% stiffer and now we are using power bit fusion to 3D print that as we speak and then there are other ongoing use cases I already talked about this we have you know another thing Eton research lab works with a lot of government funded programs so we work with DOE arl and engineering research development center.

15:00 So that one is a 9 ft long sector gear that they use that to open water dams. So they engineer research and development center they came back to us and they’re like hey we want to 3D print that using DD process but we want to cut down the cost on that. So we using the same approach to lightweight this for them. Then we have like cold spray thermal management.

15:20 So how do we come with the best spray pattern that uses the least material but maximizes the the thermal dissipation? It’s the same problem in sort of structural we solving for thermals. So we are able to deploy the end top and intact workflow here and coming up with the best pattern and then there are ongoing interest in how do we do the set to thermoplastic conversions as well.

15:41 Yeah we have some more time I want to give the presentation to Neil who will speak a little bit more on I can skip the takeaway and insights on the intact. Thank you Karthik. So basically as Karthik showed what a simulation which is behind the scene available on demand and reliable can do to your design process. I just wanted to talk a little bit about Intex simulation the simulation platform that’s was behind that.

16:16 So he already touched upon all the problems that he was facing in meshing, cleanup, geometry and all these problems essentially creates this big barrier to automation for traditional finite analysis and makes it not automatable and the kind of workflows you want to build today will not work if your simulation is not fully automate automatable and reliable behind the scenes. Right? So you want to do large scale design exploration optimization at scale building digital threads where you new data comes in and your thread just updates itself or now we have better and better physics models.

16:50 So how do we which depends on lot of data simulation data. So how do we generate these data reliably for to train these models. So our approach to solving this automation problem is essentially separating geometry from the simu simulation space or the grid or the elements what we whatever you want to call it. So instead of trying to mesh the geometry we immerse our geometry in this simulation space and at runtime we fuse the two and compute all the important physics quantities you need to be able to solve.

17:24 And this has two very important implications for us. On the left side since you we’re not meshing the geometry anymore. So basically you can just feed it any geometry where you can compute its volume we can simulate it right and on the right side because this grid space is essentially classical fundamental analysis. So any CA solver that you already use is easily pluggable to this. And the the implications of having this type of sim meshing free simulation is that now you’re able to analyze very complex models like these implicit lattises and rib structures some from Anttop again very easily which will take days to mesh and clean up and and and be able to simulate.

18:09 And not only that now you can easily mix and match representations. Sometimes you have bre and implicit inside latises. You can do that and sometimes there are non-traditional representations which represent a valid volume but to be able to simulate that you’ll have to create a surface mesh of this city scan data. But with with meshless you can just do the simulation right away on the representation. We are already available in many of the computational design platforms out there because our goal is to be where you are.

18:44 So we are in Rhino directly simulating implicits in Anttopa the process automation. So we are available natively in in that platform as well. CDS uses in their design pipelines and we’re also on the cloud and have integration with on shape and finally really excited to talk a little bit about PI intact. So we released earlier this year PI intact which is essentially a very native crossplatform Python API for our meshless simulation.

19:20 And what this does is now you can basically program programmatically set up simulations. I think the animation did not run. Yeah, very easily set up these kind of simulations. Set up once and just throw hundreds and thousands of designs at it and it’ll simulate it for you. You just leave your computer overnight. We also have basically un level up optimization built in and we provide you with the sensitivities datas, the gradients, whatever you need to drive your design decisions.

19:51 All of those are available natively inside Python very quickly. You can run thousands of simulations to do your design space exploration and just to stress like it’s very native to Python meaning you can use all your existing modules ML tools and all our data types are in memory so you’re not doing some hacky way of calling external applications and doing file based inout reading writing it’s all natively in Python and again available both on Linux and Windows and can be easily deployed cluster and with that I’ll wrap up.

20:28 So, Intact is still young but we have great leading partners now who are working with us and getting great success in different types of application areas. You saw we are they are seeing u quite a bit speed up in their design iteration being able to do automated design of experiments at scale from concept to validation and while staying in their design workflows. So with that I’ll I’ll wrap up this session and we can all head for coffee or Yeah. We have a guest on that side not enough.