Lightweighting? Topology Optimization? Generative Design? Where Do I Start!?

KENNY CORNETT: All right, can everybody hear me OK? Cool. Good morning. Welcome to Lightweighting Topology Optimization Generative Design-- Where Do I Start?

My name is Kenny Cornett. I am a generative design and topology optimization consultant based in Northeastern Tennessee. I've been using Autodesk Generative Design since the first public facing beta, I suppose, about 18 months ago. I'd been using it on a lot of interesting case studies and things and had a lot of experience with it, did the Pier 9 residency earlier this year as well with generative design. So let's go over the learning objectives just real quick here.

I want to show you the landscape of what software options Autodesk is offering for lightweighting components. I want to talk about the difference between topology optimization and generative design. Picking the right process for your project, which one of these various strategies is the right one to go with, and what's the bottom line, what happens in your business when you use these. So the first thing we need to do is clear the air a little bit.

There's some confusion about what generative design is when you just hear this out in the industry, and it can mean one of I say five things. Other people say four things. It can mean topology optimization. It can mean lattices.

It can mean trabecular structures. It can mean computational design, or it can mean topology synthesis. And we're going to look at each of these a little bit in detail as far as what they are and how they affect your business and how they can affect your designs.

I will say that trabecular structures and lattices are more or less the same thing, just at different scales. It's probably a fine way to say that. So those are going to get combined in the way that we talk about them.

So computational design is where we'll kind of start with. And in the Autodesk world, Dynamo is really the king of this particular area. So computational design is algorithmic design. That is, you go through a series of logical steps to create the geometry.

It's a scripting language, really. And I also added fractal to that. Fractal is a project that's associated with Dynamo that allows you to extrapolate options out of the designs that you create.

Some other things in computational space, you can use Macs and Maya. You're going to have to do a lot of the scripting and coding yourself, though-- so a little harder to do there. The thing about computational design, though, is it occurs fairly early in your design process because you use this to explore options. You use this to generate concepts and ideas.

So it's not really something that occurs down the line when you have a product in motion. Next, we'll look at topology optimization. In the Autodesk world, all of the topology optimization is ruled by the Nastrtan solver. So in better shape generator, the straight up Nastran Fusion 360 and Netfabb all run off of the Nastran engine in the background.

Now, each one of these four options has a varying degree of controlability of the solver. Obviously, Nastran itself has the most control, and shape generator has probably the least control of all of the options. Fusion doesn't have a lot of control either, but it has more than Inventor does.

Netfabb, you can actually feed it the Nastran files that you can feed Nastran. So you can play with that a little bit further. If you're not familiar with topology optimization, we can actually see the process across the top here.

So we start with a model, and this is kind of the hallmark of this is that there's an existing design space. So we apply loads and constraints, and the solver spits out optimized geometry. In this particular case and all of these Nastran solves, it's a mesh file that we get out the other side. So we overlay that mesh on our original part, and that gives us kind of a template for where we can remove material from.

So this is a very manual process to get from one end of this all the way to the other. And this occurs pretty late in your process. And the reason I say it's late is because you already have a space claim figured out. Next is the lattice and trabecular stuff.

So these are little tiny struts in your models. There's three tools that do this. Netfabb is an intelligent lattice tool. So you can apply loads, and it will change that lattice or generate that lattice based on the loads that are involved.

Within Medical, I call it semiintelligent because it's very, very granular. You can make a lot of changes to the lattice. And the mesh mixer, I say is dumb. And I say it's dumb because you have almost no control, but you can generate some really interesting things like the bunny rabbit there with the oiling lattice all over the outside of it.

Latticing in particular is generally very, very, very, very late in the process. This is like right before you're going to go to the printer kind of thing typically. Very rarely would you design a lattice early on, particularly with these tools because the output from here is mesh. So you can't do a whole lot with it without a lot of headache.

So topology synthesis-- the only tool in the family is Fusion 360 right now. And so topology synthesis allows us to look at not just one space claim, but lots of space claims.

And so right here in the center at the bottom you actually see a whole bunch of-- these are cell phone holders that have the same setup geometry, the same loads. Not the same geometry-- I guess there is one difference in that the starting shape is different. So the space claim, each one of those is different. And that's the power of topology synthesis.

This is very early in the design process. So this is all about creating conceptual feedstock for you to pull things that you like from one design and pull things from another design and kind of mix them together so that you get somewhere in the future with it.

So for the purpose of this course and from this point forward, whenever I refer to generative design, I don't mean this greater industry sort of paradigm. I actually mean a very specific piece of software here at Autodesk. So we're going from the term topology synthesis to generative design. So if I say generative design, I mean topology synthesis. So light weighting, here are some approaches to light weighting parts.

The first is the human best guest approach. So this is you the designer, you the engineer use your experience, your intuition to design things. You cut big pockets. You put ribs where you think it's going to be stronger.

You do things like thin sections, or you try and use different materials and things. This is all about you, the designer, trying to come up with interesting geometry. This also includes things like the computational design because that's still you putting your information into the system. And it's biased.

So everything that comes out of that process came from you and your experience. And so if you and another designer were tasked with the same thing, odds are you're not going to come up with the same design. That's just kind of the nature of this.

This next sort of approach is what I call single domain optimization. So this is topology optimization and lattice optimization. And what I mean by single domain is there's already an established base claim.

There's one area that we're going to work within. And the problem with this is it's biased also because I the designer created that space claim. So it's already set up in some design that I came up with, and it's just going to optimize within that same volume.

So my last approach is multi-domain optimization or generative design. In this case, I can consider constraints like manufacturing multiple materials at the same time, even multiple processes at the same time. And the great thing about this one is it's not biased because I'm not giving it any starting information necessarily.

It's going to generate based on the functional information that I give it. So we intuitively kind of know that the gut check or the engineer's best guess is not always best. So we're going to focus on these other two paradigms.

So engineering outcomes-- the ultimate goal here is to strike that balance between performance and cost. We're limited in our resources in order to accomplish that. And we really need to explore all of the options that we can given that trade-off that we have to make.

So if we look at this on a cost to produce versus performance curve, it's this S shape. And you've probably seen this before if you're in any of the hands-on labs for generative design, you'll see it again. I imagine they'll see it. Lots of places won't be there-- yeah.

And so the concept here is that on the bottom left low cost, low performance, this is like the RadioShack Tandy ADComputer. And somewhere in the middle is a Surface laptop. It's kind of a generic laptop. And on the high end is a server. It's very expensive, but it's very powerful, high performance.

So that's kind of the curve of product design. And when we make those trade offs between cost versus performance, we almost inevitably end up somewhere in the middle. So let's talk about a sample design problem.

In this case, GE aerospace and grab had created this challenge that they crowdsourced this challenge. And throughout, they wanted people to try and take this bracket that had established loads on it. It had established manufacturing criteria, and they wanted to make it as light as possible.

And so when we look at the designs that came out of this, you can kind of see where they fall. In the bottom left, there's a sheet metal part. It's probably not very strong, but it was crazy cheap to manufacturer.

On the way far end, there's an additive part that's got an internal lattice in it. Somewhere in the middle right here, there's a couple of like CMC illets parts. So let's talk about how each individual person worked.

So each individual person came up with maybe a couple of concepts. But when you do your work, you probably come up with three, maybe five at the most. And you run with one until you hit a wall, or you reach the finish line.

And so there's this little iterative process where we validate, and we evaluate. And we see if we can actually make this thing. And yeah, there's some back and forth. But in the end, it's kind of a long process to go from just a few concepts to figuring out what's going to work with them and what's not going to work with them.

So this time to market at the bottom is kind of a lengthy process. And what we get in the end is that looks like the engineer's best guess. There's pockets in that have been milled out. It's really just the engineer's best guess, and it's been through some FEA of some kind to validate this. But it was a long process to get there.

So when we look at this challenge as a cohort, as the entire group, the dynamics change. Now there's a whole bunch of concepts that came out. And all of these concepts were validated because the work was done by an individual, in this case, that said that my design works. That was the whole point of them entering the contest.

And so out of that came a single winner quite rapidly. And the concept that this creates is that we go from a validated concept to a production ready design in a much shorter amount of time. And you can see on the far right, there there's a productivity increase that occurs.

So when do we use what tool? Well, where in the process are you? If you're on the left side, the front end of your design process, then you need a tool that's going to help you come up with those validated concepts in a hurry. So that would be generative design. If you're more closer to manufacturing, you've got a design that you're more or less happy with and you just want to shave weight out of it, that's a topology optimization problem.

So generative design is not an immediate shortcut to production geometry. It's not a silver bullet no matter what marketing has said. All right, so this is not production ready geometry. It's not ready to go.

You have to massage it afterwards. You have to validate it. You have to make sure that everything works the way it's supposed to.

So realizing the true value of generative, if we work with existing parts and just try and remake them generatively, you have to do that with a grain of salt because you may not get a lot of great results. The true value is where we can generate tons of validated concepts and find a natural synthesis of what works and what doesn't work to create the best product going forward.

So show me the money. Where does this actually impact the business? So the biggest reason to leverage these tools is they actually affect your bottom line. And when we fold in tertiary processes is when we start to see the difference.

Now I say tertiary because if we stay within like an RFID group, your secondary processes are going to be things like manufacturing. We know that anything that happens in manufacturing happens in R&D. They talk back and forth.

But what about procurement on the other side of manufacturing that's in charge of getting raw material? What about logistics-- the people that are in charge of shipping your final product? I'd take lessons from racing teams, particularly like Formula One and Le Mans teams.

The basic premise here is that if you took a spec Formula One car from the rulebook of the FIA and you had 10 of them on the racetrack with any 10 Formula One drivers, they would all be within about one fourth of a second of each other. You just would be. So all of these teams work with very minor details to gain hundredth of a second here, here, here, here. And eventually, it becomes a full second, and eventually, that becomes five seconds with all of these little tiny changes.

And so the idea is if a Formula One team can make a tiny change here, there, and everywhere and make a five-second difference, we can do the same thing with our geometry. So maybe it's quality assurance that changes. So we save money by changing our process to where quality works better.

Maybe it's logistics. Maybe our parts weigh less, so we don't spend as much fuel to ship them. Or maybe we can ship more in a truck because they don't weigh as much.

Maybe it's procurement. We don't buy as much raw material. Or maybe we've included pilot holes and things like that in our geometry So That now our machinists aren't digging in with their tools as much. Our tools are lasting longer.

And then there's the environmental impact as well. Maybe our trucks are creating less CO2 now because they're not carrying as much weight. Maybe we're not throwing as much scrap out as we used to. I mentioned there in that environmental impact, imagine a foundry.

If we could cut 20% of their electricity consumption, if it was a electrically fired foundry just because we didn't have as much material in there. How much pCO2 would that remove, just something like that? So let's take a quick recap here and look at our learning objectives again.

So the landscape-- there is Inventor and Fusion and Netfabb. Those are all the topology optimization. Looks like I'm missing Nastran in there also. Within Medical and Netfabb do latticing along with Meshmixer. Dynamo Max and Maya do our computational design.

The difference between topology optimization and generative-- so topology optimization focuses on a predetermined space claim, a part that's already been designed and we just need to make it better. Generative design is a shotgun approach. How many validated concepts can we get out of this process? It's a mass ideation tool.

How many people here are familiar with design thinking? If you people so the concept of design thinking is generate as many ideas as possible. Ignore the laws of physics. Ignore the laws of everything and just come up with ideas.

And eventually, you're going to come up with this great amount of information. Then a natural solution will synthesize itself out of it.

So picking the right process-- so where you are in your project timeline. If you're at the front end, obviously, we need concepts. We need validated concepts.

So genitive is the right tool there. If you're further downrange and you're heading towards manufacturing, then topology optimization may be the right tool for you. If you're all the way at the end, getting ready to manufacture, maybe lattices are the right tool for you.

Lastly, the bottom line-- so lightweight parts positively affect logistics. They positively affect things like quality and procurement, maintenance. Other business silos that we haven't even mentioned can all be affected. And it all goes back to that Formula One sort of adage there that any small amount matters when we start piling up small amounts. And any questions so far?

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: So topology synthesis became generative design. So the topology synthesis is where we create that geometry from an empty design space versus optimization, we take an existing space and pare it down. We chuck material out of it.

AUDIENCE: I'm thinking of [INAUDIBLE].

KENNY CORNETT: Yes, right. Yeah, so I guess the topology optimization was top left there, and the topology synthesis was bottom right. So the bottom right was that quadcopter frame that you saw all the different iterations of it.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: So sure, right now, in the generative tool, as it is currently released, really, the only manufacturing constrains that you can do or is it additive or is it milled. And then there's some milling constraints as far as how is it flipped around or what's the specific process.

From there, you can kind of extrapolate as far as how much machining time is that's going to be, how much material are we going to have to work with. So there's some work on your end to try and figure out some of that information. Maybe Doug can speak to some of that, but Doug is the product manager for generative design-- so manufacturing constraints, things like cost of manufacture.

AUDIENCE: [INAUDIBLE]

DOUG: Yes, you can take in things like material costs, build high time, labor, things like that, [INAUDIBLE]. But I [INAUDIBLE] generating a design. I mean, it's only [INAUDIBLE] certain types of sort of [INAUDIBLE].

KENNY CORNETT: So the solver that runs generative design is a level set algorithm, which is different than a topology optimization algorithm, which is they refer to it as simp. So level set tools in and of themselves are kind of an odd thing in that if you had an infinite amount of computing power and you had an infinite number of iterations, they would all reach the same global minimum for that design. But that's an academic exercise or a theory even.

What's interesting about generative design is it's very sensitive to local minima. So you saw like the cell phone holders. I can throw different starting shapes in there, and it'll converge at different places in that if you think of this as a giant three-dimensional graph, where are the local minimums? And that's really all that starting shape does is move us to a different place on that graph, and it'll find that local.

If we take that same part and feed it back in again and again and again, eventually, it'll all come out to be the exact same part. So is it biased? To some extent, yes. When you put a starting shape in it, you intentionally bias it.

If you don't put one in it, though, not so much. The other aspect to it too though is every simulation is garbage. Some of them are useful. That's like the adage.

So garbage in, garbage out-- if you don't put the right loads and you don't put the right constraint and you don't build it correctly, then what you get out the other side is obviously biased in a bad way. So there's that sort of facet to it as well.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: So for example, the cell phone holders. Each one of those was a different study that had a different starting shape to it. And again, if I ran them all an infinite number of times, they would all end up at exactly the same shape.

AUDIENCE: That's just one documentation.

KENNY CORNETT: Sure.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: And you actually see that in generative currently where if you have a design that looks like it should be symmetrical, it won't be symmetrical when it comes out of generative because, say, one side of the part will be at one local minima, and one side of the part will be another local minima. Both of them satisfy the requirements, but they're different.

Is that a fair assessment? Any other questions? Yes, no, maybe? How many of you have seen generative design in person, used it, played with it?

A few people. How many of you are taking a of hands-on labs-- anybody? Couple of you? We want your feedback, or Autodesk wants your feedback on this.

If you'll go to the Idea Exchange, there are multiple studies going on research studies that they're looking for information on some of these topics here. The guided walkthrough for new users would probably a really good one if you've not ever experienced this at all. It'd be something to look at and to play with. If you're not familiar with where the Idea Exchange is, it's just outside breakfast.

AUDIENCE: Your part validation so that you have a model where you look at aggregated [INAUDIBLE].

KENNY CORNETT: So there's a whole different world of metrology and scanning and probing and this, that, and the other. And it's really interesting when you look at companies like Renshaw that are doing both the additive side and the inspection side side by side. And so they actually have a pretty complete workflow as far as being able to take a part of a build tray, machine it correctly, probe it correctly, and then figure out if the part is in spec or not.

But as far as how that interfaces with this, there's a total disconnect for right now. One other thing is Autodesk has recently launched the Services Marketplace where there's curated providers for services. My business is one of them on the manufacturing side doing generative design.

So if you are interested in running a project with us, you can get in touch. There are other ATC providers as well and other manufacturing providers that are doing training. We're doing project consulting, anything that you can really imagine. And we're all approved by Autodesk to do these things. So you can find us at the answer bar inside the expo as well.

I want your feedback. Everybody wants your feedback. In the app, there is a survey tool. So be sure and leave your feedback. And that's important because that actually drives AU for next year.

So tell us what you like and what you don't like. And then you'll see those changes occur next year. So I've actually cut this a little bit short because a bunch of people asked to go to the AEC keynote, which starts in 15 minutes, I guess. So we did go through this really fast.

If you've got additional questions, you want to talk some more, feel free to reach out. If you're going to go to the AEC show go ahead, the keynote. Anybody else have any other questions or anything, feel free to stay and talk and chat.

AUDIENCE: I thought it was fascinating, by the way, thank you. [INAUDIBLE].

KENNY CORNETT: Oh, I'll let Doug answer that one.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: And Dynamo with fractal is kind of where that is. So essentially, what fractal does is it let's say you drew a simple cylinder inside Dynamo. What fractal says, I have a range of diameters from, say, 1 inch to 10 inches.

And I have a resolution of 0.25 inches. So it's going to kick out everything that satisfies that requirement. So you start to grow these huge data sets that way.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: Sure, so Dynamo is a BIM tool, technically, but it outputs step files and output t spline files. So you can do all kinds of mechanical design stuff with it also. And it's not marketed towards mechanical design whatsoever, which is interesting. But it will do some really interesting things as far as like parametric textures on a part that would be really hard to do in a CAD software. But you could program them in Dynamo to create like a randomized texture on a part that is true b rep geometry.

AUDIENCE: You also need to have an idea of what you're designing.

KENNY CORNETT: Well, sure, you can't just go into it blindly.

AUDIENCE: You can't just, like, say, give me something here. You have to [INAUDIBLE]. So you have predefined this, and you're just trying to figure out what the optimal thing is. And there are [INAUDIBLE]. Are you working on merging those [INAUDIBLE]?

KENNY CORNETT: There's a great deal of discussion in the Fusion user community about how that will play out. There's no official answer from anybody on it, but the more voices we have yelling for it, the more likely it is to occur.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: And if anybody is familiar with Grasshopper, that is what Dynamo is from Autodesk. The Grasshopper is the McNeil product at this point but same thing.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: What do you mean?

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: So in generative currently, under tech preview, I guess you'd say, is there's a space frame solver, which is an interesting approach. So it makes small beam structures everywhere as opposed to these organic flowing shapes. It's a whole bunch of small struts everywhere.

So that's another type of solver that's available. There's lots of solvers out there that can do different geometry types. It's just a matter of how soon do they get plugged in and which ones get chosen and what have you.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: And that's not to say too that you can take a result that you generated in generative design and pushed it into Dynamo to use that as like the boundary to do some other crazy thing inside of it. So you can definitely start stacking these on top of each other to get different results out-- yeah.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: So if the model properly converges, there shouldn't be a lot of like little tails and things left on it. The other side to combat that is they're not exposed to the user at this moment.

AUDIENCE: [INAUDIBLE]

KENNY CORNETT: And as a secondary line of defense against that, now that we can export ts splines from generative, you can go in and massage it and smooth it and do all of that stuff after the fact. So which is, I mean, it's a consolation prize, but it's not a bad one.

Just get out, get out. [LAUGHS]

Anything else? I realized this was like crazy fast, but we had like 10 people ask if I could do that.