0:15 I’m going to be talking today about how you can bridge data to geometry using implicit modeling. And my name is Wesley Essing. I’m the lead on the implicit modeling and poly nerves modeling teams at Simcenter Inspire. For those people who might have seen the name Inspire before but with a different branding, it’s Altair. Inspire was the the previous name, but Altair was acquired by Siemens last year.

0:45 So, if you’re looking for Inspire, it’s now under the name Simcenter Inspire. And let’s see if this actually changes. There we go. So, obviously I mentioned, you know, bridging data to geometry with implicit modeling, but let’s start at the start and just give a very brief introduction to give everyone an awareness of what implicit modeling is and a very quick idea about Inspire. So, the best way to describe implicit modeling is to look at sort of a more traditional modeling approach that you might be more familiar with like boundary representation or mesh geometry, which is usually described with, you know, set of stitched together surfaces with a number of features bounding it like edges and vertices with its very specific features to the geometry.

1:41 Before implicit modeling, we like to look at it more like a function or a set of equations where essentially you can ask the geometry at any point in space, how far are you from the surface? And so, you can query various point in space and get different values for how far away you are from that geometry. And if you do this for every single point in space, you essentially create a field of values where typically where all those values are zero is where we can say is the surface.

2:16 And, you know, what does changing the way in which you model allow you to do? And really, it’s a really flexible and robust way of modeling geometry because we no longer have to track individual features. Doing Booleans and offsets and shelling is incredibly efficient cuz we’re not having to calculate all the little intersections between different features and different surfaces. It’s a much, much faster approach and robust approach to modeling.

2:49 Because we don’t really care about individual features and topology, we really create some really complex models and complex topologies, really porous style geometries. And the main part of my talk today is it allows us to link data to the way in which we generate geometry. And finally, we can evaluate all of this on the GPU. It’s a highly parallelizable system of evaluating geometry, so we can make it incredibly fast.

3:20 And to give you a look at the way that we’ve implemented this in SimSolid Inspire, there’s a quick demonstration workflow for a heat exchanger with a a lattice core. And essentially, what we can do is in Inspire, we can start with a dimensioned sketch, completely parametric. We can then do a variety of modeling operations and section out the core that we want to put a lattice inside.

3:44 We can then define the lattice core that we want with two fluid domains with a variety of different parameters for the lattice. We can then block the individual fluid domains for the hot and the cold side with baffles, union it all together, and then define what the fluid domain is, add all the boundary conditions on to the inlets and the outlets where we can specify which fluid is where, and then we can say the lattice of the embedded solid and run that on the Inspire fluid solver, which is a GPU-based solver, incredibly fast.

4:14 But once you have that parametric model, it’s very easy to make adjustments to it either with variables and rerun a complete study and evaluate a really, really wide set of of varieties of models. So, we’ve been working on implicit modeling in Inspire for sort of the last 4 years, and it’s a really feature-rich set of capabilities today. We have about 34 now, I think, since I even made this context with about 150 different implicit functions that you can choose from, and we can make all sorts of queries to the geometry, you know, like the gradients and field values and closest points.

4:57 But we can also convert in and out of implicit so that Inspire can work with a variety of geometry types. And we have this full end-to-end digital thread from the sketch all the way to the simulation and even the manufacturing solvers inside of Inspire, all with a single workflow, all captured through a history, which is all editable and runnable. And with the links to simulation, so Inspire has motion solvers, fluid solvers, and structural solvers, all which are meshless and incredibly fast to set up and solve.

5:33 And then this is all wrapped in a Python layer, so this is customizable and able to create repeatable workflows incredibly fast. So, just a quick overview of, you know, some of the different categories of operations that we can do inside of Inspire with implicit modeling, but I’ll show some of these in more detail. But, the bit that I think is really cool about implicit modeling, and it’s the fourth column, fields, is being able to define parameters not as constant values everywhere, but complete spatial control of those parameters.

6:10 And so, we call this field-driven design. And it allows you to have really, really localized control for a number of different parameters, you know, your lattice parameters. You can have fillet radii, shell thicknesses, offsets. There’s a whole host of different ways of applying fields to your geometry. And this is really where we can take data and map it to the geometry, because a lot of the data that we work with is either volumetric or planar in some way, and it’s really nice to be able to precisely map the data to your geometry.

6:46 So, we’re going to look at data-driven design, which is, like I said, a subset of field-driven design, mapping it to data. And there’s a number of different types of data that we see people use day-to-day when it comes to generating geometry. So, you’ve obviously got your more classic sort of CAD data, sketches, CAD features, edges, and surfaces. But, you’ve also got your simulation data, which can either come in directly from a Inspire sort of solver, or externally from a set of point clouds, you know, could just be a random set of CSV data with a scalar value on every single point.

7:26 Or, it can be image data, which I’ll show a variety of ways of using that in the model. But, it could also be scripted data, so you could create your own custom sort of equations, or your custom data sets that you can import. So, the first one we’re going at is you know, using simulation data inside of your your design, you know, taking something where we’ve done a structural simulation on the components and then we can use the analysis explorer to look at different results and we can take that result and immediately create a field out of it and then we can create a lattice which at first has a constant value and we can map the displacement values to the density of the TPMS lattice.

8:08 Or we can take a velocity field from a CFD result and again map that to a lattice parameter where we can then modify the lattice to look and feel like it’s mapped directly to that CFD result. But we can also take data from other SimSolid products. So, we have Inspire Cast and here we have a shell mold which was also using implicit modeling to create. We can take that to the cast solver to get some results.

8:39 We can look at something like the solidification time and then we can map that result to the wall thickness of that mold so that we can improve the performance of that part. Or a different type of simulation data which is topology optimization. So, unitless density field that you can get out and we can then also map lattice parameters to those fields. But if you want something a bit more sort of fine tune, you can take a point cloud and you can use that to have very precise values for those parameters at different locations.

9:17 Either you can import that as a CSV file or you can by hand use it to locally grade your geometry. But another way of using point clouds is a more sculptural approach. So, you’re trying to reconstruct the geometry in some way. So, you’ve got some topology optimization results. Normally, you can only offset or thicken that with a constant value everywhere, but maybe you want to be a bit more precise and say in very specific areas, I want to smooth out my geometry or rebuild some sort of broken results, or I want to be able to define region where I want to keep some of my original geometry.

9:53 So, you can take the ultimate result and say in certain locations, I want to keep the original sort of definition of the model. Now, creating fields up with CAD features is also something we see happening a lot where we can extract certain p- bits of geometry of interest, and then we can use them directly in a a field where we can again map different types of parameters of a lattice in this case of different densities, and we can smoothly grade away from those features that you’ve defined.

10:27 So, you can have really good control, but you can then go back and edit that and say, “Actually, I want the bottom of the surface to be solid as well.” So, again, you can just go back and edit that history and go forward. Something that I I like that we implemented recently was being able to link your unit cells directly to the sketches inside of Inspire. So, here we’ve just got a regular diamond lattice, but I can go in and sketch whatever type of unit cell that I want in the sketching tool, and I can convert that to implicit and use it directly inside of my lattice as the unit cell.

11:06 And once that have been linked together, I can go in and edit the sketch, and I can change the geometry, and it will automatically update the lattice of the the unit cell that’s being used. Now, image data, there’s many different ways that we can use image data, and the bit that’s always quite tricky is that obviously an image data is 2D, and we want to take that 2D data and map it to something that’s more 3D like a surface.

So, we had a tool called warp map where we can take a sort of transformation from a planar surface to a 3D surface. So, in this case, we can take an image that had the text warp map and we can use it to map it onto a specific area of the model and I can edit that mapping so I can handcraft the positions and make sure that it fits really nicely on the to the surface.

11:58 So, I can use that 2D data on the 3D surface to either emboss or engrave, or we can use it to add some sort of texturing. So, maybe you’ve got a bump map and you want to add some texturing to the bottom of a shoe. So, we can then again map that 2D image to the the bottom surface and we can mix and match that with another lattice geometry where we can then offset the lattice geometry with that image data.

12:26 And so, you can then create really nice sort of texturing effects onto the geometry. But, we can also use it for measured data. So, in this case, we had a pressure sensor. So, we took a static reading of someone’s pressure on the on the mat and we can map that to the geometry of the of the shoe sole. And then, what we can do is take a regular lattice and use that field from the pressure data to map the thicknesses of the the strut lattice to it.

13:01 So, many, many different ways of creating and using this sort of data. And something that we’re seeing now being used. So, Simcenter has a tool called physics AI where we can use Inspire to generate a lot of design alternatives using the design explorer tools where we can generate lots of different lattice structures, analyze them all, and then we can send this to Physic AI, and we can train the Physic AI model to either generate the parameters of the lattice, but the latest thing that’s is it’s coming out is actually using it to generate the lattice geometry directly for a specific response.

13:42 So, you want a certain displacement, and it can come up and generate the actual geometry, and not just the parameters for you. So, obviously taking data in is great, but you know, you want to be able to take the data out as well. So, obviously implicit modeling is is a different style of data structure, but there are lots of different forms of data that it can take out of the the tool.

14:09 So, one of them is is volumetric data. So, that can be either a 3MF file or VDB file. It could be mesh data using the adaptive meshing strategies that we have with different file formats. It could be in a B-rep itself. So, with Parasolid, we could save that as a a convergence facet body, or we can use the Polynurbs tool in Inspire to wrap implicit bodies with B-rep or Nurb geometry.

14:36 It could be beams. So, if you’ve got a strut lattice, and you just want to take the the 1D elements out, you can do that through either the FEM and take it to a different solver, or 3MF and go and print them. There’s slice data as well, so we can just natively slice the implicit models and export them out. Or what we can do now as well is automatically generate the Python scripts for generating these these geometries.

And shout out to 3MF if you can see how many different formats are are captured in in the different file formats. So, obviously you have a design workflow that you have, and you want to potentially automate this or repeat it or customize it in some way. So, what we’ve introduced is the mechanism for automatically generating a Python script for any implicit part that you’ve generated. So, in this case we’ve got a strut lattice with a lot of field effects with a shell that’s variable as well.

15:37 And we can just right click and export to a Python script and then we can take a look at the code and it gives you all the exact API calls inside Inspire that you would need with all of the non-default that you may have applied or even all the field driven effects that you have in there. And you can just copy and paste that back into Inspire and rerun it and basically regenerate your geometry exactly as it was on the screen.

16:01 But you can also create your own custom extensions. So, let’s say there’s a workflow that you do a lot and you also maybe want some UI to help you have some parameters that you want to play with when it comes to the script. So, there’s a UI designer inside of Inspire as well, which actually the implicit modeling tools have all been created using the UI designer inside Inspire.

16:23 So, you can very quickly mock up or generate a panel with whatever types of variables that you want exposed and then you can link that to the Python script that you had and you can have that as your own tool inside Inspire to repeat the same workflow but with varieties in it. And we have basically a demo and an API for every single operation that you can do inside of Inspire.

16:52 You can see in the demo browser you can see there’s a demo for every single operation including all of the ones for implicit and it even has a Python interpreter and debugger inside of the program so you can just go off and run and test different sort of varieties. And it’s a really easy way then to learn all the different API calls that we have in here.

17:14 And the latest thing that we’ve done is create an extension where you can set up an FTP server and you can expose all of those API calls to some external agents or some low code platform like Mendix or AI studio or Senara for example and have an external source be able to run your your workflows. And so you can see AI studio is going to be generating some the calls inside of Inspire.

17:48 Not the most complex model but this is pretty new so you can see that in a couple of weeks we’ll be showcasing a pretty interesting workflow at the Hanover trade show. So if you’re there come visit the Siemens booth. But other than that I’ll be around. If you want to get in touch please you know scan the QR code or come see me after the talk. But thank you very much for your attention. To learn more about the CDFAM Computational Design Symposium, access the archive of previous presentations, interviews with speakers, and information about future events around the world, visit CDFAM.com.