Generative Design in Autodesk Fusion 360: 2022 and Beyond

MIKE SMELL: Hi, everyone, and welcome to Autodesk University 2022. In this session, Generative Design in Fusion 360-- 2022 and Beyond, we will be taking a look at the advancements that we've made in the past year, looking at a customer spotlight and moving ahead with our future development plans. My name is Mike Smell. And I'm a senior product manager on the Fusion 360 team, responsible for all of our generative design technologies.

Before we start off, we have to acknowledge that we will be making forward-looking statements in this session, as we will be talking about our future roadmap. These statements are not guarantees of business results, future availability, and functionality and should not be used to drive any purchasing decisions with our software.

So, again, to talk about what we're going to look at today, we're going to start by talking about our development progress in 2022. It's always good to recap all the things that have happened throughout the year. Fusion updates quite regularly and makes it easy to miss some of the key functionalities that we may be delivering every few weeks.

The next is we'll take a look at a customer spotlight. I've had the pleasure of engaging with a number of customers throughout the year. One of those has been pretty special [INAUDIBLE] think the greater [? community ?] will benefit from seeing, and we'll [? take ?] that today. And then, lastly, we'll take a look over the horizon to see all of the things that we will be looking at developing in the near future.

Let's get started with our development progress in calendar year 2022. Starting off in March 2022, we released multi-export of generative design outcomes. This is something that will enable you as the user to create four editable outcomes in a single job submission. So rather than going independently from outcome to outcome that you're interested in, we can select them all at once and start those jobs.

This will ultimately speed up how you get to comparing your designs and taking the next steps in your downstream workflows. Let's take a look at that in action with a video. So here we are inside the Explore environment, which should look very familiar to the generative design users.

You will notice that we've put our create panel on the outcome-- or on the scatterplot-- or the main view here and in addition to our outcome view. Here, you see we create [INAUDIBLE] multiple outcomes. And we're now able to start the creation process at once.

Just as it was if we did this manually, you'll see that the outcomes enter the job status window, separated by the studies that they are related to. You'll also notice that we've added some new feedback into the UI with the spinning wheel. Previously, if you'd start the process of creating an outcome, you would have to check back on it intermittently to know that it was actually complete.

So, here, you are. You see we've isolated the files where we've started to create designs. And, when we go to processing status, you can see that four of those five outcomes are still in progress.

The next thing that we did in 2022 was work on our experimental solver technologies that have been in preview. So, in both our March 2022 update as well as our May 2022 update, we've taken some major strides in what we've been delivering with our experimental solver technology. First, we've focused on bringing a more modern solver technology architecture and performance to our existing functionality. All of the analytics that we've been tracking so far show that our run times are up to about two times faster.

Additionally, you will have noticed with the May update that, when you're running a basic setup-- so functionality that does not require the experimental solvers and features to be turned on. So perhaps a setup with a basic force and a basic fixed constraint, you would now get three outcomes-- one, with our commercial solver and, two, from our experimental solver technologies that we're working on to advance our tech stack.

The next one that we'll talk about here is simulation contract forces. Now, this is specific to the simulation workspace and the linear static stress study type. But it is extremely relevant to the generative design process.

In the May 2022 update, we added the ability to look at reaction force and moment data at contact regions. So now if we're looking at a multipart assembly where we're trying to do a generative design, this can be really useful to gather the force data at those interfaces so that we can more accurately represent the response of the system versus using, say, fixed or pin boundary conditions. Let's take a closer look at this.

In this example, we're going to look at doing a generative design exercise on this modular motorcycle swing arm. So, as we go through the parts, you can see that the outer sections are individual. And then there is a center section that keeps those-- or puts those together.

In this scenario, we know that our generative design exercise is going to look at each one of those components as individual pieces, as we want to keep the modularity of this design. So here we are in the simulation environment. And we've done an entire assembly analysis where we're looking at the loads and constraints as they would be applied.

You can see here we've added contact force into the menu-- or we've always had that. But now if we isolate the components and we look at, say, the left half of the swing arm, when we go to the reaction force command, traditionally, we would only see results here if we were picking faces that had boundary conditions applied. Now you'll see, at the contact phase where the fastener from the mid section connected to this outer section, we're able to look at the resultant force and moment, based on that connection type.

When we go over to generative design, you'll see we have a design space develop that represents that left half of the swing arm. So you can see we've got our obstacles that represent clearances we need for a drive train, as well as our fasteners to our center section. We can then isolate our preserves. And you can see those faces that we selected to extract contact forces are preserved in this design.

Now, if we were to assume that those were a fixed constraint, that may represent a stiffness that's actually far too stiff to get a truly representative design. So what we're going to do is go back to simulation. We're going to extract those contact reaction forces, copy those to the clipboard, and then we're going to use those as loads on our preserves in our generative design study so that we can get a more accurate representation of that interface behavior.

The next thing that we've done is we've started to take some efforts to improve our fluid path study public preview. In this first effort in July 2022, we wanted to focus on the terminology. Now one of the unique things about generative design for fluid path studies is you're working in essentially both domains-- the fluid domain, that's your preserve geometry or your preserve volumes; and in the solid domain, which is all of the obstacles.

And we had a lot of customers tell us that they were really struggling to understand what's what. Is the preserve a fluid? Is the preserve solid? Is an obstacle-- how does it play into that? And what does the starting shape represent?

So, in this effort, we went in and did something as simple as adding that definition to the command names. So now preserves, we designate those as fluid preserve geometry. Obstacles, we designate those as solid obstacle geometry. And for starting shapes, we designate those as fluid starting shapes to make it a lot more clear as to what domain you're working in when you're building out the geometry for the design space.

The other thing that we did in this update in July is we took the opportunity to add some coloring to the boundary condition faces to try to help distinguish where we were defining a flow source and where we're defining a flow opening or a fluid pressure.

So in the image here, you can see that we've added teal or a teal type of color to the face in which we've applied a flow source boundary condition. And then, on the other side of the model where we know that the fluid is going to exit and we have our flow opening defined, we've added a purple color. So now it's very easy when you're looking at the design space to understand, where is the fluid flow defined? And where is it free to be calculated by the solver?

Now moving back to the-- or moving on to the Explore experience and this is common across generative design for structural components as well as generative design for fluid paths. We wanted to start adding some additional functionality to support filtering and navigating the data. So, in the July 2022 update, you now have the ability to filter based on generative models in Explorer.

Now, generative models are very, very convenient ways to expand your exploration space by potentially moving geometry around. And, in this example that we will look at, you'll see that we've built multiple generative models that essentially represent the position of the loading location here in this GE bracket. So we've got one configuration where we've got it in the original position and then one in the new position.

So, when we go to Explore and we have results based on these two setups, you'll notice that just like any of the other filters in Explore, we can then isolate for those. We can isolate generative model one. We can isolate generative model two. This makes it much easier to understand which outcomes are coming from which setup?

We can also then sort by them. And this makes it really easy to then say, OK, well, I'm going to look at my setup that's associated with generative model one, set up associated with generative model two, and then maybe do a comparison. And this is going to be particularly useful when those changes between generative models are maybe more subtle than what we've done in this example.

The next thing that we've done related to our experience in Explore is out of the ability to do real-time study names syncing between the study environment and the Explore environment. Now, historically whatever the study was named at the time you've solved your generative design study is what you would see in Explore. And there was really no way to keep those in sync, especially if you started creating additional studies after that. You had no way to go back and rename the original studies and have that come in to Explore.

In the July 2022 update, we now have the ability to do this real time. Let's take a look. So using that same example where we have two generative models, each representing a unique position. If we go into explore, you'll see that the default study names are both study one structural component, study two structural component.

So if we come back over to the setup environment, we can go ahead and change study one's name to original position as that best represents what we have captured in that study. We can activate generative model two and change that study name to new position. You'll notice that it doesn't out of date the studies. And, when we go back over to Explore, you'll see that those new names are now present in the filters. So we can sort by study. You see the naming carries all the way through all the workflows inside of the Explore environment.

And then the last thing that we delivered in 2022, which was a major project for us in the generative design development group was automated modeling. Hopefully, by the time you all are watching this and spend some time with us at AU, you will learned all about what automated modeling has brought to the market. But, in our July 2022 update, we released a new tool in public preview, the design workspace called automated modeling.

This tool is focused purely on the design exploration process and creating new design concepts based on simple definitions of what to connect and what to avoid. This is really great for the early stages of the design process. And it's very similar and built on foundations of generative design technology but without the burden of requiring you to have all of the load information and all of the material Availability and all of the manufacturing processes that you might want to consider just to come up with a basic design idea to figure out directionally where you're going.

We'll take a look at this in more detail in a demonstration. And here we are in the design workspace with a new command called automated modeling on the automate panel. We'll go through, and we'll just start by saying, what do we want to connect?

And then we'll tell automated modeling to go off and generate us some design ideas. And, within a matter of minutes, you'll see that we have multiple design alternatives coming back, some based on different connection schemes or interfaces. So how does it connect on to the surfaces that we've selected? And then we also have different algorithms for how those surfaces are being connected.

So we'll take one of these alternatives. And you'll see it populates the timeline with an editable outcome. And, in this case, we see that it looks like it's going to be taking up too much space in that area of movement. So we have the ability to use a body to avoid as a way to capture some design intent.

So, now, in this example, we've added in that curved body to say, hey, that's the space we need as this suspension rocker moves through its travel. And you'll see that the that the design geometry updates and stays away from that space. So, again, we can say, OK, we can see the type of outcome we get there.

Now, further, we also have the ability to use these bodies to avoid to drive some concept of design intent. So, here we have that very cagey looking design. But we say maybe we want something that's a little bit simpler or that has a less complex space. In the case of a mountain bike suspension rocker, we may want that thing not to be super easy to clog up with mud.

So, here, we've added two additional big cylindrical bodies to avoid. And that's going to drive the shape in a very different direction. So we'll take a closer look at that.

And if we isolate that body you can see what automated modeling has created. Now, one thing to point out is, at the interfaces for the faces to connect, you'll notice that it's a very soft face. And the entire body is made up of a T-spline. You also notice a green line in the middle.

So, for any setups that we recognize that the faces to connect and the bodies to avoid are symmetric, we will automatically enforce symmetry in the shape that we create, as well as the editability.

Now, here you see we're picking another alternative that looks very similar. But this is based on a sharp connection at the interface. So here what we're doing is actually putting a prismatic body over that connection phase, and, instead of the T-spline body being the entire shape, the T-spline body connects up to those prismatic interfaces. Symmetry still applies just as we saw in the other example. So two very unique ways that the connections are established-- and, in further than that, again, automated modeling in just a very quick amount of time can allow you to explore new designs without a significant investment in upfront modeling.

Now, I'd like to transition into the customer spotlight section of our talk. And, here we'll be talking about a gentleman Ryan McClelland from NASA Goddard. So Ryan approached us about a year ago with the goal to create and infuse process into product development at NASA where they could make lightweight spaceflight structures using a commercial software offering.

So let's unpack that a bit. Ryan is someone who's centered around developing and applying digital engineering technologies. He's got experience doing product development and product design. He also has experience doing product simulation and validation.

But he's not a simulation analyst. And he's not just someone looking at say industrial design or basic shape definition. So he's looking at the end-to-end mechanical engineering space for developing these lightweight spaceflight structures. And the progress to date that we'll be talking about is Ryan's been able to successfully use Fusion 360 to develop 20-plus different parts for various spaceflight missions at NASA.

The other thing that I will point out here is you think, well, what is a space flight structure? I think, as we move forward and look at some of the examples, you'll see that almost all of these are basic brackets. Now, they may connect a number of different places in space. But they are still just structural brackets.

So I think it's important that I point that out as we think about this example. And we think about how it may apply to the types of work that you do where you're looking to provide and create structural support applications.

So you might wonder, why did Ryan look to generative design as the way to develop a solution for lightweight space flight structures? There's plenty of manual techniques that could be used. There's plenty of simulation tools on the market. But what Ryan expressed to me is that the current development methods that are being used in his divisions are too costly and time consuming to really optimize designs for performance while also considering new and alternative manufacturing processes and different materials.

So we were really looking at, how do we streamline that process? Well, you've heard us talk for a long time about generative design in Fusion 360 that it's really about automating the design exploration process. And, through that exploration process, we're coupling manufacturing process, materials, and performance to give you a given set of solutions, all of which meet those requirements.

If we look at a high-level example now of, what does it mean to be more effective and efficient during this exploration process, is to get started with generative design and encode all of the product requirements for where my connections exist, where material needs to be avoided for other clearances required in the assembly. And the load and constraints that represent the performance requirements takes Ryan about an hour for these types of designs and for the design that you see here. After that, the system is ready to go wake up and start solving for these different solutions.

And, through this example, these designs come out of generative design in about one hour. So pretty fast solve to get some pretty interesting-looking designs. And then, from there, Ryan has looked at both additive manufacturing, as well as CNC machining to produce the bracket here in the image that you see. And, depending on the process, these parts were able to be fabricated within one day to three weeks, depending on the process used.

Now, we'll take a closer look at this and just kind of play back the way in which generative design was used to look at this tilt bracket. So they started off by saying, let's try to do this the traditional way. So Ryan and one of his colleagues said, we'll spend some time-- and in this case, two days-- doing this the old-fashioned way, using our tribal engineering knowledge for, how do we make lightweight designs?

So you can see there in the first column the first attempt was to create basically a weldment of three different plates with a pretty traditional rib-style design. And what we found here is, while some of the requirements were met-- it's relatively easy to manufacture. The first modal frequency target of 100 was exceeded. But the mass, the mass was cost-prohibitive. And that design was far too heavy.

So, as we move along in attempts two and three, you'll see that the design intent is very, very similar. However, they went on the process of making those plates thinner and putting pockets in them, as well as holes and then adding in additional sets of ribbing to then try to add stiffness back into the design after making it significantly thinner. And you'll see in both of these that what happened is the stiffness that's required for this structure was lost.

So now, while we've achieved a much, much lighter design, we've got something that doesn't meet our stiffness and vibration requirements. Excuse me, it's also noted here that these things then became very difficult to machine. You can look at how complex those designs with all of the small pockets and small ribs would've created challenges in machining.

Now, the fourth attempt went a very different direction, adding a trust-like behavior at the bottom. And you'll see, while we've got a very lightweight structure and were able to achieve getting a design that was over 100 Hertz for the first natural frequency, that, based on the way it was developed, that it wasn't easily machinable. Nor was it easily printable.

Now, let's compare that with using generative design. So, given that the set of setup requirements that we saw in the previous slide, we were able to ask generative design to go off and explore both CNC machining and additive manufacturing, as well as different materials and put in the requirement that the design needs to exceed 100 hertz.

And what we were able to achieve is two very lightweight designs, two designs that exceed the minimum first modal frequency requirement, 147 and 177 hertz and also that the stiffness to mass ratio, as well as the Max stress ratio versus their allowable were much, much better.

And I think the interesting thing to point out here is, on the manufacturing side of things, while these designs both look fairly sophisticated for a simple bracket with very organic-looking surfaces, this part was able to be machined by an outside vendor for 1,000 and was able to be done within three days. And, similarly, that part that was additively manufactured, again, through partnership with an outside vendor, was able to be produced for $2,000 in about three weeks.

So, in the context of the traditional costs in manufacturing times that Ryan and his teams were used to, this was viewed as very efficient. Not only did they spend two hours versus two days in the development process. The manufacturing time was also greatly reduced as compared to more traditional approaches.

Now, they've done a number of things beyond just the simple bracket that you've seen there. I had mentioned Ryan's achieved more than 20 different designs out of generative design to solve various applications. And I won't go through all of these with specific detail. But you can see here that, in each one of these, the complexity level is quite different, both from the amount of connection points and the overall sophistication of the resulting design.

All of these have been manufactured and physically tested as you see here. So quite a great showing of progress with generative design for these bracket applications, where we're either trying to create a bracket to span a unique and complex space or a number of these solutions also were areas where a number of traditionally designed parts that were bolted or welded together were able to be replaced with a single component through part consolidation workflows and generative design.

So, in summary generative design has helped Brian and the folks at NASA automate the design process for metallic structural components. For the folks who are now up to speed using generative design, they can get from requirements to parts ready for fabrication in one to two days. These parts can be developed, as we said, for CNC machining or additive manufacturing. And, generally speaking, these parts are coming back about three times stiffer, lighter, and stronger than a typical human-based design, as validated by their physical testing.

So here's a couple additional examples of projects that have been worked on. This one on the top really exotic, impressive looking design, machined with a 5-axis mill. And then, again large part here at the bottom, again, additively manufactured using metal. So, again, coming across both manufacturing methods, multiple different materials, all while being able to meet performance requirements in a much more automated and efficient process.

Now, for the last section of our talk today, I'd like to look over the horizon at where we're going with our generative design technologies for Fusion 360.

We'll start off by looking at what we're doing in the structural component study. As I already talked about in the things that we've accomplished in this calendar year, we're going to continue the pursuit of commercializing our experimental solvers. Now, what does that mean to you? First off, that means that we're going to bring modern solver technology and performance to our existing functionality.

So I had mentioned that we were already seeing run times of up to two times faster. This will be the mainstream solver going forward. So if you're not using Experimental Solver in [? future ?] technology today, in the very near future, you will start to see the benefits of this technology and the run times to get your solutions.

The next phase of this process of commercializing this preview will be to bring remote forces, remote moments, and remote constraints along with symmetry and displacement constraints to the mainstream product. Again, what does that mean to you? That means that to use this functionality, you won't have to go turn on the previews. These things are very well proven today. It's just a matter of making sure that we've got all of our i's dotted and t's crossed that these different setups are working across all manufacturing methods and things like that.

And then the last phase of this commercialization effort will be around bringing the modal frequency constraints, the rigid body mode support, or inertial relief, and the buckling constraints to the mainstream offering, again, so that we're not having to go turn on a preview or having functionality that only works with certain load types or certain manufacturing methods.

The next thing that we want to do is around lightweighting of existing designs. And I'll talk a little bit about this one in detail because you might think, well, generative design already can create lightweight designs. And, absolutely it can. And it is quite good at creating lightweight designs.

But a piece of feedback that we've gotten from our customers is that if they have an existing design where a lot of design intent needs to be maintained and they're just looking to remove material, then the workflow of modeling preserves, modeling all of the obstacles to ensure that that design intent is maintained, sometimes it's overwhelming. So what we want to look at is, how can we streamline the setup workflow when we work with these existing designs? The existing design itself has a lot of that design intent already built into it.

So we want to reduce the modeling requirements associated with the scenarios where you're working with an existing design versus the scenario where you may have a lot more design flexibility to where you may be consolidating parts or coming up with a net new design incomplete.

So, for us, we see this as quicker time to result because, in the example we just looked at with Ryan, that example said they were spending an hour to set up their design study. Now, there may be scenarios where folks have to spend even more time capturing all of that design intent, our hope is that we can bring you a quicker time to results, again, through reducing that upfront setup process maybe from hours down to minutes.

And the way that we're thinking about doing that right now-- and we will be looking to engage with some of our users and get more feedback on this as we progress forward on this project-- is around allowing you to say that the starting shape potentially acts as the boundary of the space that can explore.

So, essentially, all the starting shape that you can only take material away from it. Whereas today the starting shape is an influence point for where the solver will go. But it absolutely can go outside of that starting shape and reform the design space if it is required to meet your performance requirements.

The other thing we're thinking about here is, how can we allow the user to work directly on the starting shape? And one way that we're thinking about that is by allowing you to then define preserves as surfaces rather than bodies.

So, again, you can imagine in certain setups, if we have a lot of bolt holes, if we have potentially an internal flow channel that [INAUDIBLE] pressure, today we need to spend a lot of time modeling to extract basic minimal thicknesses out of that to provide the preserve bodies. Whereas here, if we can work directly on those surfaces, again, we have an opportunity to really streamline what's required to set those designs up.

The next thing that we will do-- or we hope to do this year looking forward is around Casting. We've had die casting in public preview for a while. We've collected a large amount of feedback on that. We want to continue to work to improve our die casting solution, as well as to expand out to support sand casting with the hopes that, as we make progress in this area, that we will have a much better path for users who are looking at high-production volume applications because today, in a lot of scenarios, additive manufacturing, really advanced, multi-axis milling may not meet requirements for the highest-volume applications.

Another area where we will be looking at to extend our concept of manufacturability is around weldments and how we can support users in lightweighting those. So, again, you can see a theme here of, how can we continue to expand generative design into some more traditional, higher-volume manufacturing methods that more folks are commonly using to produce products today. This is something where we will look at adding the ability to both create these types of, weldment-style designs, as well as provide a result that has the optimal thickness and profile.

So, similar to our current workflow, we have either an existing design, like we just talked about, or we have an open design space, how, rather than fill that design space with a very organic shape, can we fill it with a much more simplistic network of planes or plates? And then, for that given network of plates, how can we make sure that the thickness of that plate, as well as the profile of that plate is the most efficient way to achieve those design objectives.

Now, I'd like to transition over to the fluid path study. You've seen that I just talked about some of the basic terminology improvements that we made earlier in the year. Probably, by the time that you're watching this and we're talking about this together at Autodesk University in person, we will be very close to having some boundary condition refinements out in the commercial market as part of the public preview.

So here we're doing a couple of things. One, we've tried to be a little bit more clear with our naming and also with our command isolation to just better align to the user's intent. So, in the current shipping product you're probably used to seeing that we had a command called inlets and outlets.

And, in that command, you were able to define flow velocities, flow rates, as well as pressures. Now, that required the user to understand a few different things. One, they had to understand that a non-zero pressure represented essentially a free boundary condition where flow could exit or enter the system.

So what we've done is we've tried to be more intentional to say flow sources represent locations where both the magnitude of the flow as well as the direction of the flow are defined by the user. And those are the ones that will have the teal coloration like we saw earlier in the presentation.

And then we're going to separate out flow openings, which is essentially a zero pressure-- or a zero-gauge pressure boundary condition, where we're saying that flow is the magnitude of the flow rate is undefined the magnitude the velocity is undefined. That is to be calculated by the solver, as well as the direction.

And then we're now also making a fluid pressure command that basically allows you to prescribe a gauge pressure value. So the pressure is fixed. But, again, the flow rate or velocity magnitude will be calculated by the solver, as well as the direction of the flow.

So we hope that, with these changes we will have a much more streamlined intent for what you're trying to set up. And it's also made it easier for us to define, what are the proper combinations of boundary conditions to drive a solution? So, one, we can have single or multiple flow sources going to single or multiple flow openings.

We can have single or multiple flow sources going to single or multiple flow pressures. And then we can have a combination of all three, where we've got single or multiple flow sources going to single or multiple flow openings or fluid pressures. So a lot of flexibility there but also simplification in the nomenclature that we're using to capture the requirements for the design exploration.

The other thing that we will do is around shape quality. And this is about improving the geometry transitions at the interface boundary between the preserves that you define and the shape that we generate. So, in this example here, you'll notice that, on the left side, we've got a pretty abrupt transition on this right most preserve and, similarly, at the upper preserve.

And, in this case, that's going to be a lot of modeling that's required to smooth that out right. And, in this scenario, the solvers thought, well, you probably don't need the inlet size to be that big. Now, in a lot of cases, folks have the inlet sizes or the outlet sizes through the preserve volumes that they defined as fixed because they're either needing to fit into a specific type of standard fitting. Or maybe they're just looking to smooth out or optimize the internal flow domain of an existing design, again, which connects into an existing system. So we don't have the flexibility to change everything.

So, in those scenarios, you'll notice here, on the right, we're going to try to do a much, much better job of putting a gradual transition to those preserve volumes to really try to help you guys with reducing the amount of downstream editing that you have to do and any sort of weird anomalies that may be encountered when you go try to validate this in, say, Autodesk CFD to make sure that not only is the pressure drop minimized but, also, any of the other flow characteristics that you might be designing for are achieved.

Going up one step further in the workflow, once we have the fluid domain, we often get asked, well, how can I easily make the encasing volume? Maybe it's a ductwork. Maybe it's a thickness that we need to go build around to actually build out a valve body or a manifold.

And we're looking at, how can we automatically generate this as an editable solid body? When you go to the Explore environment, you've chosen an outcome and you say create, not only are we going to create the fluid domain, which we do today, we're going to start to enable workflows to create this solid encasing volume based on an approximate specified defined by you, the user, that stays in contact with all of the wetted surfaces. Now, we see this as a way to set up downstream workflows to generative design for structural components. Maybe we have a basic thickness here.

But we say, hey, there's going to be a burst pressure that we need to respond to. There's going to be structural loads that we need to respond to. So, for this given optimized fluid path, how do we build a structure around that that can handle the actual installed application of that design.

And then the last thing that we will look to tackle in fluids in the near term is around flow-opening biasing. So, today, the working assumption is if you define a flow source and you define multiple fluid pressures or flow openings, we will balance the flow uniformly across those outlets or those regions where flow has the freedom to come in or out of the system based on a pressure load that you've defined.

Moving forward, we want to give you the ability to control that further. And if we want to distribute the flow in a non uniform way-- you can see here in the image. Instead of 25, 25, 25, 25, we want to do something like 20, 20, 40, 20. We will be able to optimize for the minimum pressure drop while maintaining that flow distribution.

The last area that we will talk about advancing our generative design technologies is around automated modeling in the design workspace. So, as I had mentioned, automated modeling is built on foundations of our generative design technology, simplified in a way to make it feel a lot like a modeling tool that's super fast. And it enables this rapid design exploration.

But we've heard from our users in the process of bringing that to public preview that they still want more control over the design. How can I create some additional variation? How can I easily take this design that looks almost like what I want. But maybe I want it a little bit thicker. Or maybe I want it a little bit thinner.

So this first project is around adding a thickness or volume control slider to the alternatives. And we see this as a way to really reduce the amount of potential downstream editing that's required. So let's take a look at that.

So here we are. I've got a bunch of examples just so you can see how the system is starting to respond. But here we've got some finished outcomes. And you'll notice now, under each alternative, we've exposed the slider.

And what that slider is going to let us do is exactly what you would expect. You move the slider to the left, the design gets thinner. You move the slider to the right, the design gets a little thicker.

So again we're not having to go into T-spline environment and try to do any weird complex scaling operations or heavy remodeling to get to a very similar design that's just thinner or thicker. We'll just go through a couple more of these. We'll pick one of the other alternatives where we've got a very different style of shape to see how the thickness slider influences those designs.

We'll jump over to another example. If you were part of our private preview and the insider program, you probably recognize this design from some of our early demonstrations. Again, simple model, three connections.

First outcome here is very tubular looking. So, again, you can see what that slider does pretty natural, kind of changing of the design. Similarly, here in the one that looks a little bit more plate-like, kind of moving both the inter-- or the exterior profile, as well as the thickness.

Now, over to the GE bracket, which I'm sure everybody who's been involved with generative design over the past few years recognizes immediately. Again, seeing some of the shapes that are coming out of automated modeling, as well as how the thickness slider is helping you manipulate those designs, again, before getting straight to the B wrap and having to do all of this manipulation manually.

So we see this as a really great way to add some additional influence into automated modeling so that you can really put your influence or your style on it and also just give you more to consider. Maybe you know that you're going to need that thinner design, so we'll just move the slider all the way to the left.

Or maybe you say, hey, I love this shape. It's too thick, so-- or it's too thin. We're just going to make it thicker.

The next area that we're going to work on in automated modeling is around context switching. And this was something that we didn't expect to hear from our users because we really made a big effort to make automated modeling fast. And, in almost all scenarios, it runs in just a few minutes. But what we found is when users would click away to a new tab or would try to start another command, thinking that, hey, I'll go do something else while it's computing because it's on the cloud, they were inadvertently canceling the operation.

So we're doing some design work here. And this is a prototype that shows how we intend to prevent unintended canceling of the alternative generation. This is going to enable some multitasking. So if you wanted to go explore another design with automated modeling while you're working on-- you get one started, and you go to the next one. That's a workflow that will be enabled by this.

And it's going to be a clear indicator that automated modeling, as the command, needs to be finished before we can move forward with any additional timeline operations. Let's have a look. So here we are. We'll go ahead and build out a definition for automated modeling.

And we'll go ahead and start generating alternatives. And you might say, well, all right, it's generating. It's going. I know it's going to take a minute or two to get to a result. I want to switch to my other document.

Well, again, in the current solution, that would cancel the operation. And you'd have to set up again. What we're thinking is that, when we move to the other tab, we'll quietly close the command in the background. But we'll put an automated modeling feature in the timeline in a sixth state.

What this is going to do is allow you to go do that other work but come back and still have access to the results that you were generating. So what you'll find is basic things moving around the canvas, moving around the browser you won't be interrupted. But, as soon as you try to start a new operation, we'll tell you, hey, you need to go take an alternative.

So you can either choose to discard. And that will let you move forward with the command. And it will delete the automated modeling session that you started in the future. Or you can simply choose to select. And that will open up the automated modeling command for you.

You can choose an alternative that meets your desires. And you can continue on with your workflows. The other thing is, like any other feature that's sick in the timeline, the most natural thing might be to right-click and edit and figure out what's going wrong.