Enhance a Generative Design Model with Event Simulation in Autodesk Fusion 360

JOHN HOLTZ: Thank you for joining us for this Autodesk University presentation called Enhance a Generative Design Model with Event Simulation, both in Fusion 360. My name is John Holtz. I'm a senior technical support specialist with Autodesk. I've been with them for six years now supporting Inventor Nastran and Fusion simulation, and I have over 30 years of experience with finite element analysis. And I'm joined today by Jaesung Eom.

JAESUNG EOM: Thank you, John. I'm Jaesung, and I'm an active developer of a simulation group inside of Autodesk. I've been working on barriers to finite element simulation, and recently, I'm mainly working for the finite element solver for the Fusion 360 Event Simulation and the levelset optimization scheme underneath Generative Design. John, now you can continue.

JOHN HOLTZ: Thank you, Jaesung. Before we get started, just have to show the safe harbor statement, which basically says that if we make any comments about the future versions of the software, those are not a guarantee that anything we talk about in the future actually makes it into the future version of the software.

What we want to accomplish today is to provide some tips on creating a Generative Design model and Event Simulation analysis and be able to use those two software's capabilities to improve the designs, to identify potential problems in which candidate solutions are better than others, be able to tune the parameters, and to evaluate the results to identify the good features in the model. Now, Jaesung, over to you.

JAESUNG EOM: Thank you, John. Then let me first appraise you on what is Generative Design and, essentially, what is the Topology Optimization scheme we use underneath Generative Design. This is the symbolic image of a Mars lander, and this showed what quality Generative Design is looking for in the simulation world and the Topology Optimization world.

And this model basically is composed with three different parts, and each part is targeting a deeper-end cost [? train. ?] And some part is supposed to build through the CN simulating machine, and another part is more than the manufacturing scheme like, 3D printing. And the NASA JPL designer balanced their design goal perfectly well with the new Generative Design platform that Autodesk provides.

So immediately after Generative Design was released to the market, the people were quite excited about it since this is a new, sure way to guarantee that they're reducing the time to design the new product and the cost of the production and that more and better designs with less material is much more accessible in various different objectives and constrained settings. And in short, we say that a more complex engineering problem has opened the door to a broader audience with this new feature.

So Generative Design has a couple of different qualities, but today, I'd like to focus more on the Topology Optimization part of Generative Design. And under the hood of this Topology Optimization, we are using the levelset to represent the geometric operation, and this level set is not only representing geometry, but also represents a physical, structural elasticity underneath. And using this levelset representation, we realized the various boundary conditions as a traditional finite element, and the CAE community has gotten used to it.

And so far, we've enable a bunch of different objectives and constraints. For example, like, a volume reduction target constraint or objective and a maximum stress and additive manufacturing constraint, and a bunch of other constraints are available through this Topology Optimization.

Then let's jump into the little bit theoretical side of this levelset-based Topology Optimization. Of course, I'm trying to minimize the number of equations as much as I can, but essentially, this levelset is not an explicit surface mesh kind of object you're familiar with. It's more like an implicit function called f, and this fx is evolving over time or iteration by the certain normal velocity is described as a V. And essentially, how we determine V to harden our new shape with our design goal and the constraint is the essence of this whole shape optimization.

And to minimize the formula, I tried to make very schematic sketches of how this levelset-based Topology Optimization is working. So as I said in the previous slide, determining this velocity, V, is the most essential step for the Topology Optimization, and this V is composed of two key factors.

The first thing is a compliance derivative, and this compliance derivative is trying to maximize the steepness of any given structure or minimize the compliance. The second thing is that we like to reduce the quantity of a material as much as we can, but we still obtain the higher steepness for the structure performance.

So these two apparently conflicting or competing goals are combined together, and that will generate this new velocity field into the number three. And with this new velocity field driving this implicit levelset [INAUDIBLE] over the iteration, then eventually, before we're repeating this whole process approximately 100 times for the most typical Generative Design model, we landed that proper shape-- what we're looking for.

But here's the catch. So the one thing is, as I explained earlier, this shape is evolving by the velocity field, and this velocity field is essentially determined by strain energy density. And what strain energy density implies is that since this is the energy, this is not a factor quantity. So this is scalar. So this is one of the-- might be counterintuitive for the traditional mechanics user because if we apply that equal amount of a load to a different direction, we will get, still, the same outcome. This is because we are evolving the shape by the energy, not the force, as a vector.

So that's why you don't have to worry. You don't have to add this kind of computing. At the end, it will be equal to the boundary condition for the Generative Design. So this is one-- be mindful to care for the better num-- you can set up the minimized, the completed, or the redundant number of the load case for the Generative Design model.

So then, John, can you continue?

JOHN HOLTZ: Yes, thank you, Jaesung. So the next topic here is how does the Event Simulation help with the Generative Design? The thing to keep in mind is that the Generative Design is a linear analysis, and a lot of the Generative Design models have thin structures, which often have some kind of nonlinear effect. And that's where the Event Simulation comes into play.

Just to clarify, a linear behavior, as seen here on the left, just says that if you double the applied load, the displacement's going to be twice as much, the stress will be twice as high, and so forth. Most things in real life have some kind of nonlinearity if the displacement is large or if the material properties change due to the displacement. So the actual load versus displacement graph's maybe like this on the right where doubling the force causes more than twice the displacement. That is an example of nonlinear behavior.

In the Event Simulation, it is a time-dependent-- it's a dynamic event where we typically use it to model things such as impacts. So it generally involves very small time increments and a short overall duration. And here's a typical example of some kind of protective eyewear with some piece of debris impacting the lens, and you run the analysis to see is the lens going to break? For obvious reasons, that would not be a good option.

With Event Simulation, it is a nonlinear solver that has many nonlinear capabilities, such as large deformation and large rotation. It can have nonlinear material properties, whether it's because the material goes plastic when you exceed the yield strength whereas, shown in this dog-bone shape, this is a rubber seal that, even though it remains linear-- when you remove the load, it's going to go back to the original shape-- it has large displacement and the stress strain curve is nonlinear. It can also include rigid bodies in the analysis.

One of the advantages of the Explicit Solver inside the Event Simulation is that, just based on the way the mathematics are handled, the solution matrices are very small, so it requires only a very small amount of RAM compared to other types of analysis. So what that means is you can solve larger models with the same amount of RAM.

Another nice feature is that, because of the mathematics, they always converge, unlike-- you may be familiar with a nonlinear static analysis in Fusion or a similar product. And inevitably-- not inevitably, but depending on the model, you may run into problems with it converging. You can get up to a 55% load, but it won't proceed beyond that. The Explicit Solver converges without much problems. And this doesn't apply to Fusion, at least not currently, but the mathematics can be applied to other simulations as well, such as fluid, structural interactions, and so forth.

This is just a quick example of what the traditional calculations look like. This is just the spring equation. The spring rate times displacement is equal to the force. Well, the reason that the traditional implicit methods takes so much time to solve is that inverting the stiffness matrix is very computationally-intensive, whereas with the explicit dynamics, you're just solving this equation directly, so that makes it very fast. The explicit dynamics or Event Simulation excels at solving extremely large models with nonlinear materials-- no problem-- large deformation, as I mentioned.

Another thing-- we'll see a couple of examples here-- is that the model can break apart and create new contact surfaces. And another advantage is the solver picks the time step automatically. So you may be asking yourself, essentially, what's the catch? Why aren't all analyses using the explicit dynamics?

The catch is that the solver takes steps automatically. That time step is typically on the order of 1 Eden -6 to 1 Eden minus 8 seconds. So if you are trying to get to an event that occurred over 1 second, you're looking at doing between 1 million and 100 million times steps with the time step size being that small, and of course, that leads to very long runtime.

And I should mention that these are the main things that contribute to the calculation of the time step-- the mass of the material, the stiffness, and the minimum dimension of an element. So essentially, it's how long it takes the speed of sound and the material to cross the element-- that controls the time step. Another thing that limits this from being used in all situations is that because the Event Simulation is a cloud-based solution, we limit the runtimes to 12 hours currently.

So in summary, the things that you want to do for the Event Simulation is to get a good quality mesh. You don't want any real skinny, sliver elements, so you may have to do a little bit of work on the surface model-- the faces. You don't want them coming together at a real sharp intersection-- in a tight corner, for example.

You want to use the actual mass density and modulus of elasticity, meaning that if you have a part that you want to be rigid, there's an option to make it rigid. Don't change the modulus of elasticity to make it 1,000 times larger because that's what you would do in a traditional analysis. That would reduce the time step greatly, make the runtime longer.

But another nice thing that you can do since it's able to converge regardless of what the duration is is you can start with a very short duration in your setup-- for example, somewhere around 1 Eden -5 to 1 Eden -3 seconds for the entire duration. That will get you the fastest runtime so you can check the results to make sure your analysis is set up correctly.

Of course, if you have a statically-stable model, you can also just run it as a linear static analysis. Even though that will not include any large displacement effects, at least I'll show you if your mass and constraints are set up properly.

So once you run it with a short duration and you see that it runs for x number of minutes, if you want to get closer to a static solution or to reduce the dynamic effects, you can then increase the duration, and you'll know approximately what the runtime will be because the runtime is approximately proportional to the duration. If you make the duration 10 times longer, the runtime is going to be approximately 10 times longer. Now, of course, if you have an actual impact analysis or something that has a real duration, meaning you're not trying to simulate something in a static-like condition, then of course the duration has to be what it is to capture the actual impact.

Another thing you can do if you are trying to capture a static condition is we're really doing a transient analysis, and therefore, all loads will follow a load curve. So instead of having the load be constant or instead of just using a simple linear ramp, if you make a sinusoidal ramp where you start accelerating the load more slowly and then increase the load towards the end of the analysis, if you then level it off, that will help to minimize any shocks that occur in the model due to the sudden loading.

Here's one example-- just a half-model of a lip. You can see when we go to pull it out, it breaks like you would expect it to if you don't release the clip mechanism. A couple other examples-- this is a little Wi-Fi router. So this is the model setup, and this demonstrates the capability to actually break the elements apart.

If I go to the next slide, we can see that in more detail. So there you can see it breaking apart. And in some cases, you can see the parts actually bounce off each other because, as er mentioned, the contact is recreated when it breaks.

A couple other examples here, lots and lots of contacts easily handled by the Event Simulation. Another case here-- I believe the fan blades here may be rigid, maybe not. It's a little hard to tell. I guess they are deforming there-- some kind of bird impact. These are just some of the examples that you can do in an Event Simulation.

Another topic capability that Fusion has that I won't talk about in today's talk other than this slide is that there is an option called Quasi-Static Event Simulation. Essentially, what that does is it runs the regular Event Simulation, a dynamic analysis, but it extends the duration and it adjusts the application of the loads to try to minimize the kinetic energy because, essentially, if you had a model that had zero kinetic energy, then there's no dynamics going on and that solution would be the same as a true static analysis.

But since the kinetic energy is usually not entirely eliminated, that's why we call it a quasi-static solution. It's as close to static as you can get it with a dynamic analysis. And now, Jaesung, if you would, go ahead and talk about the generative design models.

JAESUNG EOM: Thank you, John. So as John covered on the Event Simulation capability, once we get the original structural assessment from the Event Simulation, what can we do? So we found a couple of problems of our Generative Design model. Then I'll show you quick tips and how can we improve after of this kind of design structural assessment.

So the first quick tip is that you'd better start with a coarse resolution with the Generative Design for the synthesis. So if you go to the Study Settings, the button on the top of Fusion, then you can select the synthesis resolution. So this is the traditional [INAUDIBLE] called mesh density. So if you go to the coarse resolution, you are using a lesser number of degree of freedom to solve the problem and also a lesser degree res-- number of a box cell representation is using for the whole other process.

So that provided-- still, this coarse resolution will capture the essence of your problem. And then also, you can see that if there is any mistake during the modeling process. You can quickly identify and fix it.

The second thing is, as John provided on the Event Simulation, an even linear-static analysis can capture what is wrong and what is working on this current design outcome. So then once you've assessed the whole of this process, the next step you can do-- there are a couple of choices.

The first thing is that you can refine the loading and boundary conditions. I'll explain more in the next slide. And another big avenue with which you can intervene the model is applying more constraint. So if you're a commercial user, now this experimental solver capability is accessible through the Fusion Option tab. So the one thing I recommend to you today is using the displacement constraint, and another is Additive Manufacturing constraint-- not just the 3D printing, but it has some quick hack usage of this manufacturing constraint.

The third but last implicit option you can use is, again, tune the synthesis resolution. So it's not always true depending on the preserving feature or the obstacle feature in your model. So fine resolution tends to generate more of the membrane- and shell-dominant shape, and the coarse resolution is a more beam- or truss-dominant shape as the outcome.

So let me briefly explain a little bit more about the boundary condition inside of Generative Design. So loading is composed of two things-- it's the direction and the magnitude. So if you have any design intention of keeping the model set up, I think that finding a direction is quite intuitive. But sometimes, people have a hard time to pick the right number to start for this loading or momentum-loading magnitude.

So here's some quick table for each loading. So smartphone weight is normally 1 to 2 newtons, and a hatching chicken egg 50 newtons. And the standing human weight is 1 kilonewton, and a 100 kilometer per hour car crash, the result's 100 kilonewtons for the seatbelt and airbag. So this isn't really the lump sum number, but you can scale your load based on this kind of real-life load in the beginning.

So one benefit of having a linear-static finite element is assess the Generative Design Topology Optimization is this is more like a linear-- so you can either double it or triple it, and you can see more scale the result directly. So in that way, you can scale the magnitude over the iteration of Generative Design modeling.

So the next thing is accessing that experimental solver feature, especially for the local and global displacement constraint. And another thing is readily available-- thickness control inside of the Additive Manufacturing constraint.

So Additive Manufacturing constraint essentially offers a 3D printer-ready shape as an outcome. And what its result is, as I highlighted on the bottom of this table, for a real-life 3D-printing experience, you need to add some supporting structure for that 3D-printing process. And this Additive Manufacturing filter or constraint helps to minimize the [INAUDIBLE] support mass.

So in the beginning, without any of the manufacturing constraint, there are still 28 grams of waste that are produced during the 3D-printing process, but it can be reduced to 19 grams with this. But there is one implicit connotation of this Additive Manufacturing filter, which is if you check that dialog box of the Additive Manufacturing filter, you can find out that minimum thickness.

So this might be used in your more direct intervention about the model with the proper setting of the opening angle. And so the one quick hack is that this Additive Manufacturing constraint can be not only used for the 3D printing, but also, you can intervene or you can more directly engage in your Generative Design model.

And the next thing you can consider to improve your Generative Design model is the displacement constraint. So this constraint generates a shape, and that shape is the dictate that prescribed the displacement constraint either globally or locally. So in a local displacement constraint, you can select the point, the edges, or face that you'd like to control and you'd like to give a certain constraint to allow this shape deforming throughout the Generative Design setting boundary condition.

And the one downside is since the sample of the location should be intact throughout the simulation and the Generative Design iteration, this local displacement constraint is only applicable for the preserving geometry for now. But the global displacement constraint can be [INAUDIBLE] wholly in a holistic model anywhere inside a model.

So let's check the effect of a displacement constraint. So without displacement constraint, this [? output ?] bracket can be deformed almost to 2.4 millimeters. But once you dictate and prescribe the allowable maximum displacement in x and y and g directions globally, the model exactly follows this guideline.

So inside of the logic of Generative Design that stops and regulates that the reduction of the shape, don't violate this prescribed constraint. So in that way, you could see that this maximum displacement is exactly matched that the prescribed condition.

And another way to impose this kind of constraint is locally. So in this clutch pedal model, without any constraint, the maximum displacement can be 0.11 inches. But with the proper setting of the local constraint on this face of the pedal, you can also see that it matched exactly.

And so the optimizer guides the shape to reach the allowable maximum displacement on the face of a [INAUDIBLE] geometry, as you see. And as a result, you can see it's a little bit bulkier, but still, it's not violating the shape as an outcome of the Generative Design process. And now, John will show a more holistic end-to-end demo for his model. John, can you continue?

JOHN HOLTZ: Thank you, Jaesung. We'll take a look at two models. I'll start with a compound bow. So this is what it looks like here. And we're going to look at just the bow itself, not the pulleys. In fact, it looks like the pulleys probably may have already been designed with some kind of Generative Design. So we'll leave the demonstration of that to someone to handle next year, or maybe we'll have a design contest to redesign the pulley. But for now, let's take a look at the bow.

Because of symmetry, we only need to model half of the bow-- from the area where you grip it with the hand up to the upper pulley. In Generative Design, when you're creating the model, there's basically three types of volumes that you have to create-- the volume shown in the green is called a preserve region. Those are volumes that you want to keep. For example, up at the top of the bow, there's a hole where the pulley is acting. So that's where we want to put forces.

Well, in order to keep the force in the model, we have to keep this geometry. We're going to constrain where you hold it with the hand, so you have to keep that geometry.

The second type of geometry, which is optional, is called the starting geometry, and I have that shown by this yellow section here. It's kind of a suggestion to the generative solver as to where to generate the volume.

As I said, it is optional. It provides guidance to the solver. The solver will generally keep the mesh inside this volume because, after all, it is trying to reduce the volume, and it's probably not going to reduce the volume too much if it goes outside that volume. So you can set up the analysis so that you only have your preserves, and the solver will figure out some kind of optimized shape to get from the constraint to the load.

The other type of geometry that I do not have shown here-- Jaesung will show it in other models-- is an obstacle. And that is where the Generative Design cannot create any elements because you're going to have some other kind of structure there that would interfere if we created geometry there.

The Generative Design is a static analysis. So the types of loads you can apply are similar to static analysis of force, pressure, moments, and so on. Types of constraints are similar-- fixed constraint, pinned constraint, frictionless, and so on. And as Jaesung mentioned, you can create multiple sub-cases, but you don't need an excessive number of load cases because you really only need the loads that would cause the maximum stress and strain in the analysis.

And then we get into the input that's a little bit different from a regular static analysis, and that is to tell the Generative Design what you're trying to do. So the first one you can get from the model tree or from the ribbon is to define the objectives. You want to minimize the mass, you want to maximize the stiffness. And then what limits do you want to have on those, such as the safety factor or the target mass? If you're doing a modal analysis, you can base it on the frequency, you can use displacement. If you're doing a checking for buckling, you can have it so that the buckling load factor of safety is greater than some value.

The second input specifically for Generative is the type of manufacturing method that you want to use. Jaesung talked about some of these. I typically begin with the unrestricted. As the name implies, that gives you the most free-form shapes. Sometimes, it's nice to see those shapes, even though some of them may be a little bit exotic. But it may give you some ideas of what you can do to the analysis, even if you're going to use a more traditional method. And of course, if you are limited to a specific type, such as if you have to do die casting, then you're going to select that as the manufacturing method.

For the material properties, the entire model will have the same material throughout. It's not like an assembly where you can have different material in different parts. However, you can select different or multiple materials from the library, drag and drop those into the current study, and then each one of these different materials-- such as aluminum, magnesium in this example, a PEKK plastic-- will be used to create different generative designs. And you'll see how the stronger material can generate thinner shapes and so forth.

Once you start the analysis, you'll begin to see the results up here in what we call the Outcomes in the Explore window while the analysis is progressing. So you can see here there's some recommended shapes, and then there's a bunch of additional shapes that are occurring. To see the details on any of these particular ones, you just click on it, and that gets you into the Outcome Viewer where you can see the details.

There's basically three details to look at here. One is that there is a tabulation of the results for the stress and the displacement. And let me just mention that those results are approximate because the stress in particular, the Generative Design will filter out stresses-- the high stresses-- that it thinks are caused by the mathematical singularities or the mathematical stress concentrations that occur in this type of analysis.

The second thing to take note of down across the bottom of the screen is the Iterations. So even though this is called one Outcome, there can be several different Iterations or variations on this outcome. By using the slider, you can scroll through here and look at the different shapes, and maybe one will fit your needs better than the final Iteration.

Once you decide which one of these you like, step number three up on the ribbon is that you can click the Create command, and that will create a solid model that you can then use to perform an analysis on. The thing to keep in mind is that the exported model can also be edited as desired-- for example, you can trim away material. For example, in my model, what I had to do is I had to drill the hole where the shaft of the pulley is going to go because the Generative Design kind of filled that in. So you can trim material, you can add material. It's a regular CAD model, solid model that you can then make changes to.

In the Generative Design, you can look at a stress plot, but as you can see here, it shows minimal details. Basically, it's just a scale from low to medium to high stress to give you an idea. Once you create the actual model and perform a simulation, then, of course, you're going to be getting all the detailed results of stress displacement everywhere in the model.

So once you generate your model, now, you can do an analysis on it, as we were mentioning. In this case here, the bow has large displacement-- probably going to have some nonlinear force versus displacement results. So that's why you want to use Event Simulation.

This is just a conceptual drawing of what we're doing. Again, we're using symmetry, so we're only modeling the top half of it. The loads on it are-- we have a force where the pulley is representing the tension in the bowstring when you pull back, and we have a Prescribed Translation-- a PT-- here that pulls the two-- the top and bottom section of the bow-- together due to the cam action of the pulleys. So those are basically the two loads.

If you're new to Event Simulation, there's basically four things under the Settings that you need to specify for the analysis. One under General is the total duration, and as I explained, that controls the runtime of the analysis.

The number of results output-- a lot of users think that is somehow affecting the time step size, but as we described, that is controlled by the solver itself. All this does is it just tells it how many of those millions of results it's calculating-- how many times do you want to output the results? So changing this from 1 to 100 is not going to change the overall runtime at all.

The second thing is the mesh. Remember that the time step size the solver calculates is related to the smallest elements in the model, so what you want to avoid doing is if you do create an assembly from that model, sometimes, if you have a small part in it, if you're using the model base size, a small part will get a very small mesh size, which would create a small time step, which would result in a long runtime. So often, it's nicer to use an absolute mesh size and just apply the same mesh size to the entire model where, if there are regions where you absolutely need a smaller mesh, you can use a local mesh control to get smaller elements in that particular area.

If you have to use damping, I suggest that you use the mass proportional damping only. According to this equation, you can calculate it. What happens if you use these stiffness proportional damping is that that reduces the time step size. Of course, once you reduce the time step size, the analysis runtime is going to get longer. The mass proportional does not change that time step size.

And the last setup under the Settings is just to request what output do you want. For example, do you want to view the contact forces? You have to do that. If you're not interested in velocity or acceleration, you can uncheck those, and it'll just save a little bit of disk space somewhere, which isn't too important because it's all being saved on the cloud.

Remember that the loads are transient because we're doing a transient analysis. In my example, I used a sine curve to ramp the prescribed translations more slowly at the beginning to avoid any kind of impact-type effect and to ramp it down towards the end, again, to try to minimize any dynamics ad get a little closer to the static analysis.

And the main reason is that the duration here of 5,007 seconds-- 0.005 seconds-- that's pretty fast for pulling the bowstring back. Perhaps a more realistic time would be a half a second, but just because of the number of calculations, if I were to have run this at half a second, the analysis would probably run for a week, and that's not supported currently.

So once you get the results, you can review them. Of course, the results are more accurate in the Event Simulation because it's based on an actual mesh which you have control over as opposed to the voxel-sized elements that are used in Generative Design. Plus, the fact-- as you can see here, we showed the results-- everything, displacement, stress, whatever you choose, the output. Of course, you can ignore the stress concentrations here.

Sometimes, we get a customer saying that the safety of factor in their stress analysis is lower than in Generative Design. That's because in the analysis, we don't filter out the results of the stress singularity concentrations.

So in this first model that I did, the displacements looked pretty good. But when we look at it from the side, notice right here that it is buckling out of the plane, and that's not very good for a design. So we now know that the outcome that I chose is not a very good design. And because all the outcomes looked very similar, it's time to do something different. Question is, what do you do?

Well, we go back to the Generative Design and we create a new study, and we can enforce a limit on the minimum thickness. In this case here, go to Additive Manufacturing, for example, and the original parts were around between 1/8 of an inch and 1/4 of an inch thick, so I wanted to make them thicker-- specified 3/8 of an inch. Regenerate the model and then choose one or several of the new outcomes for further simulations-- for example, this outcome here.

So I'll put that, set up the Event Simulation the same way, run the analysis-- displacement looks good from the front view. From the side view, we see that it does not buckle, so that is good.

But there is one other problem that I hadn't considered when running the analysis, and you might be able to see it here. Let me bring in a close-up. And the problem is that that section-- this section here-- looks like it's probably hitting the other side, and I didn't specify any contact in the analysis.

So it's time to make another change to the Event Simulation. Specify the contact-- it's a one-step or a two-step procedure. Step one is the command Global Contacts. That gives you this dialog where you can specify what type of contact you want to create in the model-- either separation or, if this is unchecked, then you would be creating bonded contact. If you're using separation contact, the second option is do you want to allow the part to make contact with itself? And in this case, the answer is yes, so I checked that.

Optionally, you can go to Manage Contacts to see what is created. And here, it's showing that the bow can contact itself using separation contact.

Now, there was another block that I put into the model in order to apply the prescribed translation to the model. The block is not important for the results, per se, it's just a way to apply the load. So I don't really need to have separation contact between that block and the bow because even if it would happen, it's not important. And I know it's not going to happen just because of the geometry. So you can right-click on any of these entries and change it. In this case, I suppressed that and I suppressed the block from detecting if it were to somehow contact itself.

With that change, we have the new results, and you can see that it does come down, it touches, but it is not penetrating. So that is a good type of result that we're looking for. Jaesung, why don't you continue with your bike saddle analysis?

JAESUNG EOM: Thank you, John. I'll go over. So yeah, the second example we'd like to tackle is this bicycle saddle. And so thanks to John, it's already elaborated that it's a necessary component for the Generative Design, and I can jump into the most interesting part. So to design a new bike saddle, I know there is already the 3D-printed bike saddle that is available commercially, but I'd like to try that my own [INAUDIBLE] way.

So first, the choice is to put the material since this bike saddle needs to absorb the shock and also sustain the rider's weight. And I chose the [? hyper ?] [? foam, ?] and in this model, I chose the polyethylene. And for the boundary condition, it is sure to sustain the body weight, so I apply this contact point of the pelvic bone to the saddle as the pressure [INAUDIBLE]

And also, the rider can change the position on and off to the tangential direction against this saddle surface, so I applied two-directional movement. And the bike saddle also has a groove to protect our organ, so to realize that groove intact, I added the obstacle.

So with this setting, I ran the basic variable Generative Design model, as I explained earlier-- coarse resolution with the unrestricted case. And depending on the boundary condition is active or not, we can get a couple of exotic shapes. So then after having this kind of outcome, I ran each one of them in the Event Simulation. So I want to cover that part.

So after that, as I explained earlier, there's two routes I can enhance or I can apply to both of these constraints simultaneously. And one is displacement constraint, and another is the Additive Manufacturing constraint that John showed in his compound bow model.

So here's an outcome. So with a naive Generative Design model, the maximum displacement was 10 millimeters. And I think that that's a little too much, so I'd like to-- I hope this is less than 2.8 millimeters around. And this one perfectly matches, but it still doesn't violate that constraint I applied, and, as a result, you can see the much thicker beam is connecting from the back part of the saddle to the lower part of the saddle.

And the more interesting outcome I can get from the Additive Manufacturing constraint and that [INAUDIBLE] thickness constraint in this case. So I dictate that-- I'd like to have 8 millimeters for this minimum thickness for this model. It can generate a more elaborate connection between the other part of the pre-jumping geometry, and more uniform thickness around the whole saddle is connected with the beam. And then also, it matched as I expected for this saddle design's supposed to be function.

And here's the Event Simulation outcome comparing side-by-side. So initially, without any changes on the Generative Design-- so even though it has the 10-millimeter maximum displacement, it could go to a 20-millimeter displacement because this includes all the nonlinearity that Generative Design cannot capture. And again, for this new shape I can get with the thickness controlled by the Additive Manufacturing constraint, it's also deformed to twice larger than the linear-static presumption of the Generative Design. But still, the shape is more aesthetically-appealing, and also, the displacement is matching what I expected.

So to sum up the whole talk, we tried to share that these two [INAUDIBLE] coupled-- the Event Simulation and the Generative Design-- can be a really powerful tool. So once it is set up properly and they use one other iterative way, they can read a better and more out-of-box design from the traditional design process it can achieve.

So quick summary of the tips is you'd better start small. In the case of Generative Design, you can start with the coarse resolution with a minimum of basic boundary condition and the unrestricted shape generation, and then you evaluate any design outcome from the Generative Design with the Event Simulation. That will open a huge variety of choices whether you can expect it or you are not expecting it. And then you can go back to Generative Design with a bunch of the experimental features or the couple of constraints and you can make your model much more interesting or much more profitable.

So long story short, do more experiments with this new fascinating tool. And this is all we prepared. Thanks for listening.