Design Space Exploration with Autodesk Generative Design

KENNY CORNETT: Good morning, everybody. Welcome to Design Space Exploration with Autodesk Generative Design. My name is Kenny Cornett. I am a technical consultant with Innovation Forge. My speaker this morning is Michael from Autodesk, senior QA analyst.

So our learning objectives for this session this morning are to understand where AGD resides in the product developed spectrum, to understand how to build complex input geometry and loading, what happens after we generate our geometry and how to develop some tools to analyze it a little bit further. So let's define generative design, right, because this word has been thrown around a lot for the last few years.

And so if you think about generative design, you probably think about it as a design paradigm. So this is what we've been told. And there's four pieces to this from kind of the Autodesk perspective and from my perspective. And so the first part of this is topology optimization.

So what you see here is a bracket that, on the left, is kind of the traditional lightweight design. In the middle and on the right are shape generator created design. So the one in the middle is designed to be milled in a three axis mill-- it's got two flips in it instead of just one. And the one on the far right would be an additive part, it's got undercuts and ribs and things like that.

So the thing about this, though, is we have to start with a shape. We have to have a design that we apply loads to before we ever get any of these funky geometries. The next sort of pillar of generative design is lattice optimization and skin optimization. So what this is is applying explicit geometry with lattices to move loads in certain directions, or to create air flow or high surface area.

So the example we see here, this quad-copter is designed with a lattice that was built with Within or validated with Within and then built in Fusion 360 so that air can flow through the structure. The third sort of pillar is Tribecular structures. So this is evenly dispersed voids or pores in an object.

The purpose behind this is to generate high surface roughness, high surface area. So a part like this one from Protolab 3D is a surgical implant for bone. So we want to see ostio integration and we want to see the bone grow into all those pores and latch on. And the fourth sort of pillar is geometry synthesis. So this is a Dreamcatcher, right, we've all heard of Dreamcatcher we've seen the news about it.

Last year, Karl announced the commercialization efforts of Dreamcatcher. And so what this is is we tell the computer what we need the part to do, not what the part looks like. And then the computer designs geometry for us. What you see here is input geometry and what came out the other side. So now that we've kind of talked about it as a paradigm, let's talk about it as a standalone program.

So Autodesk Generative Design is the commercialization efforts of Dreamcatcher. So from this point forward in the presentation, when we say generative design, we're referring to this program and how this functions. So a quick crash course-- AGD works by starting with geometry. So we start with preserve geometry, this is things that are mandatory in the end design.

So bolt bosses, machine stock where we might be putting an O-ring groove or a bearing or something like that, and obstacle geometry, so these are areas where we can't have geometry. This might be a fastener, this might be a tool access. On the preserved geometry, this is where we apply our constraints and load. So essentially we say, I want to be bolted to something over here and I want to support a load over here and I want the computer to figure out what goes between them.

Once we have those kind of boundary conditions set up, we apply some additional criteria. The first one, the bottom left here is kind of the objective-- what we're shooting for. This particular screen is set to maximize stiffness. So the point of this is that we're trying to get the most stiffness out of the part while maintaining a factor of safety of 2.0 based on the loads that we've put in there.

We'll go up next to manufacturing. So this allows us to specify overhang angle if we're interested in additive so that we can try and minimize the amount of support that goes into our part. Any support on a 3D printed part means there's post-processing, right, there hands on time that has to go into that. If we specify the right overhang for our particular process, then theoretically, we can avoid all the support.

Next is the synthesis topology slider. So this is fuzzy, but this changes the voxel size that we're dealing with when we start removing material from the part. So the finer the resolution, the finer the contours of the part as they come out.

And then last is material. So you would specify your materials that you'll be making this out of. They're not just additive materials, there's concrete, there's plastics, there's of course, metals. So there's lots of different materials and it's constantly growing.

So once we fed all that information in, we get results. So this is a typical sort of result window that shows several different outcomes from the solver. And we can take a look at some of them closer. So this is a comparison view here. And you can kind of see as we rotate these around what these look like. And we can look at the mass, we can look at the displacement, the volume that's involved.

So where does all of this live in product development? Well, if you follow kind of a linear cradle to grave style of product development path, this probably lives at the front end, right. So this is great for ideation, this is great for concepts, you may even do some prototyping using a pathway like this. But it doesn't necessarily make sense downstream.

However, if you follow more of a cradle to cradle approach in the way that you design your products and the way that you build your things, this actually has a place everywhere. Because we look at projects like the hack rod, who knows what the hack rod is? Anybody? A few of you. The hack rod is a car that was built using generative design tools where we fed in-- not we-- Autodesk fed in loads that were measured using strain gauges and accelerometers after driving the car, and fed that information back into Dreamcatcher.

So it's an iterative process. You can take your existing product that you're getting ready to retire and come out with version 2 or version 3 and feed real usage information back into it and keep using it over and over and over again. So you're refining your product on a long term scale.

So in this kind of circular design workflow, AGD lives anywhere in here. So let's take a look at a case study. This is a hydraulic manifold that would connect a hydraulic pump to a hydraulic motor. The part you see there is cast, that's right here. It weighs a little over three pounds and it would cost something like $12.75 or so to manufacture it. It's made using sand casting.

So what this is though is there's the three bolts here and there's seven on the back of the motor that mechanically hold everything together. What you don't see is there are two flow channels through this where hydraulic oil moves between the pump and the motor. So because there are flow paths inside of here, this is kind of a complicated casting. It's got sand cores that have to go into it. Those sand course have to be removed after the casting process, because we don't want contaminants, we don't want sand inside our pump system.

So our annual volume for something like this is, say 100,000 units, and that will come into play. This is not an aerospace component, this is not a medical component, this is not an exotic race car or something like this. This is the kind of industrial component that you would find at a Home Depot or a Walmart.

So to create our input geometry, we just use our tools that we've got. We use Inventor or Fusion. For our flow paths, if we want it to do simple and match the existing geometry, we can do offset surfaces and patch it together, sculpt it so that we get a solid out of it. If we want to get the wall, we can offset that wall off that flow path.

We use multi-body extrusions in Inventor of Fusion so that we can create all of these individual pieces. We actually don't want them to be conjoined together, we need them to be separate. You can also do things like flow optimization using third party tools.

So you'll notice in the sand cast version of this that there's actually ports on the top of this. You can see them, they come out. They actually don't serve a purpose. They are there to support that sand core in the mold. They are an artifact of the manufacturing process. And so when we look at this from a generative perspective from an optimized perspective, we want to get rid of those, right, they don't serve any purpose. So we can use a third party tool that will optimize the flow for us and we can get a much better picture of what's going on.

So here's our input geometry for this part. It's got a flange that would amount to the motor face with the bolt bosses behind it, the two flow channels and three bolt bosses and a flange to mount to the pump side. Here's obstacle geometry that is inside the flow path. So this prevents the solver from putting anything inside that flow path, we don't want any obstruction.

And you can see here that these were actually optimized using a CVD package as opposed to the other one that it necked down, and we'll actually back up to look at it. So it necks down in weird places and it's got that extra section on it that we don't need. So we've optimized this.

Here are the fasteners that are involved. So you see not only the bolt head but a little bit more. This is a socket clearance. This allows us room to put our driver in when we're assembling this later on. And so here is what it looks like altogether. And if you'll notice, you can still see the flow path obstacles in the model there.

So let's talk about some of the results. We ran 40 studies on this. So over 4,000 results that we evaluated on a lot of different criteria. This is just one result and it's all over the board just to show you kind of what comes out. So there's some that look kind of anemic, there are some that are a little bloated and then there's that one up there that is just not acceptable, right?

So if the purpose here is to remove material and we've got that thing, than we didn't do what we needed to. However, we explored the whole design space and that's the point. So also to throw this out there, this part is A356 aluminum. That particular part up there in the solver was cast iron, because we were exploring other materials. So that's a 17 pound chunk of material. It totally defeats the purpose. However, just looking at it, we can see that there is an envelope there that this is kind of operating in. And then that can drive some further things down range.

Here's another way of looking at our results. In this case, we've done this as displacement versus mass. And we like to use this view and get our results to tuck into that corner, bottom left corner as much as possible. The colors are material. So the aluminum is the blue color there. And the one that we've picked is that-- so if this is mass, it's the lightest aluminum part there.

And if we want to look at that one in a little more detail, there it is. So this is not the one that you have, but this is just one of the results that that came out. Now you'll notice that the weight on this one is 1.28 pounds. So right there. 1.28 pounds of material. That's not quite a third, it's about 60% ish less weight than this one. So we're close, we're definitely making headway here.

So here are three other examples that came out of this. And this one is actually the winner that we've kind of run with. I'll let this video cycle again so that we can kind of take a closer look. Each one of these created what we ended up referring to as the bat wing, which is really the geometry that was created between the two sections.

The bat wing type shape is pretty universal to all of these that came out looking really nice. And this one, the one we chose has the webbed section on those extremities there. That's actually the key chain that you have as well. So this would be the finished part, after we machine it, ready to go in that pump and motor assembly.

With that part, we're using the same material. It's not exactly A356, but it's aluminum. Our weight is down to 2.66 pounds, so we've cut nearly 80% of the mass out of this part. At 100,000 units, our cost is $10.49 using laser centered additive. So there's a caveat here that I have to throw out-- we don't include the cost of the machine in the part.

And the reason we don't do that is, if you were say, a startup, and you were choosing between putting your money in casting technology or putting in an additive technology, it's going to be about the same if you're going from the ground up. Now if you're using a house, a firm to do this for you, casting is going to be cheaper.

If you are a mass manufacturer though and you want to control this and you're comparing one to the other, the laser centered is actually a cheaper option, just because of how much material we removed and some of the other features of this. So there are some contributing factors to why this ends up being cheaper. The internal flow passages on the cast version had sand that got trapped in it quite frequently and we'd have to get that sand out.

Typically, an assembly like this will self machine to some extent and any of the steel parts or iron parts that are in it can get caught in a magnetic trap of some kind. Sand, on the other hand, wreaks havoc inside like a positive displacement pump or a motor. So any sand particles that were left from the molding process became a disaster versus the parts from the centering process weren't attached to anything, they're not trapped in there, they come right out.

So the other part of the cores is the molding to make the core. So we not only have the compaction mold and the pattern for the outside of the part, but we also had a compaction tool for the inside of the part. So we had two tools to keep up with. And that means any changes required two operations. Any initial overhead was twice as much.

And then lastly, the cast part could not have pilot holes in it, because we couldn't control the tolerances tight enough. So this was a solid brick other than the two flow channels in it. With laser centering, we were able to get pilot holes quite close to the net shape size. What that means is, we're not throwing material in the scrap bin and paying for it.

So let's talk about some other design considerations. So maybe you're interested in surface area, or thermal transfer, or aesthetics, acoustics, fluid flow. We kind of show one way to look at fluid flow, but there are some others. What you actually see here was a test that we did fairly early on just to evaluate the solver, but the purpose of this is actually to be printed in TPU, or an elastomer of some kind, that would act as a vibration absorber, a shock absorber.

So this is actually for like a acoustic environment. You would mount panels to this to try and absorb sound. So there are lots of ways that we can use AGD to drive other designs. This is one, we were interested in getting a lot of surface area out of a part, because it ended up being a heat sink that was loaded on both sides. So we created very convoluted geometry as our obstacle and the results that we get are really wacky. And they're intentionally wacky.

So here's one of the parts that came out of this. So what happens after we make this part in AGD? So AGD spits out mesh information currently. So in order to take this to the next level, we have to convert it to BRep geometry. So there's several different pathways that you can do this through. The one that we use is we re-topologize the mesh into a quad mesh, take the quad mesh into a T-spline program, either Fusion 360 or Rhino with T-splines, and then convert to BRep from there.

We actually prefer doing this through Rhino, because Rhino allows you to be system limited on the number of surfaces that you bring in versus Fusion has a hard limit that you can reach fairly quickly. So some other things you can do to this downstream would be lattice optimization. So maybe we can continue to further push the weight down on this part by building more lattices inside by skinning it with a attention skin and building a scaffolding inside of it.

DFM, DFA-- designed for assembly and manufacturer. That can be a secondary process to this. And we kind of looked at the fasteners and the clearances and things, and it becomes abundantly clear when you get in some of the more complicated shapes that you need to be very specific with how you say you want to assemble something fairly early on. If you need to be able to get an impact wrench into a tight spot, it's probably not going to happen.

So you have to be very cognizant of that upfront. You really need to look at it on the back end too. So cam-- for this particular part, we had to machine and grind several surfaces as well as drill the holes for the bolts. On the back side of the part, this face had to be very, very smooth, because it was an O-ring face as well.

So once we got the BRep model, we could take this into our cam package and get everything nice and flat, get it sliced off correctly. If we had been using a mesh file, I don't know that we would have got the right accuracy out of it. And last would be, rendering and visualization.

So this looks really good when we render it. We can use this in marketing, we can use this and brochures and flyers and videos and things like that. So the mesh file works great for that, the BRep also works great for it. If you want to show the machine surfaces, you pretty much have to do to the BRep, unless you've got a Max or Maya person on staff that can take it and massage it quite a bit.

So as sort of a recap here-- AGD lives everywhere in the cradle to cradle cycle. If you are cradle to grave, it's probably on the front end for you. This has the potential to disrupt traditional mass manufacturing. So the reason that we showed this part is it's kind of a run of the mill, dumpy part, right? This isn't a motorcycle frame, this isn't a sports car, this isn't the next prosthetic limb that's going to change the world. This is the sort of thing that gets created all day everyday all over the world in millions of units.

One thing that we showed you just briefly was, you can trick the solver into giving you different kinds of results. The purpose behind like that convoluted geometry was so that we could get more surface area out of it. So there's other tricks that you can employ using the loading to get different kinds of geometry out.

And then the last part is, it really takes a lot of downstream work to take something out of Dreamcatcher or AGD and turn it into a useful part. This is again, not going to replace you as an engineer or a designer, this is really only to give you a starting point I think. And it's more than just the mesh to BRep, there's a lot of constraints that were on this part that we couldn't put into the system yet.

And so it took a designer to go back and massage this and to make sure that the bolt bosses were correct, to make sure that we had enough meat in the right places for machining and things like that. So this isn't going to take anybody's job away from them, this is only going to make your job better. So that's really where we are with AGD on a product like this.

If you've got any questions about this process, or you'd like to see more of the economic side of this, get with me afterwards and we can have a deeper discussion about it. OK,

MICHAEL: So you're welcome everyone. My name is Michael Musiol and I'm senior QA analyst for Autodesk Generative Design product working on this software almost from the beginning of development. And today, I would like to show her what we achieved in cooperation with anybody and company. So maybe a quick agenda of the beginning.

So I would like to start with AnyBody Technology company profile. After that, go through project subject, measurement methodology and creating problem definition in AGD and results. So basically, about project subject, probably you saw on the leaflets that we will try to optimize bike frame, [INAUDIBLE] bike frame.

OK, so AnyBody technology is a company from Denmark. And they deal with musculoskeletal analysis. So basically, they have body models. These models are a diverse, systematic validated-- systematic validated, sorry. And these models are open, it means that user has an access to all variables.

Next position on this customizable model. So as you can see, motion capture-- imagine patient specific anatomy, accurate motion environment. On the right side, you can see the guy who is getting in the car. And here we can see the body model for this but particular behavior. Last information-- we can do a lot of things with outputs from simulations like that, I will go into details in a couple of seconds.

We can export results to FE Software, so basically, it does what you guys did for us. Quick information about the application. So they operate in sports, orthopedics, universities, consumer products, automotive, assistive devices, aerospace and defense.

OK, let's talk about modeling system they use. The first thing is input. So the input for the modeling system is motion and forces, body model and environment model. So having that, we are able to build input for the simulation. And what the simulation does. As a result of the simulation, you will get internal body loads, as a muscle forces, joined forces, or interface forces.

So basically, there should be an animation, but it's not working-- I don't know why. You will get internal forces from this body model, basically, that's how it works. You can get to those forces and use it for optimization. And that's exactly what we did. There is one slide about model repository.

So they have a kind of model library with muscles, ligaments, bones, joints. These models are detailed, validated and published. And you can personalize the model using anthropometric scaling and morphing.

Before I will go to measurement methodology, I would like to say that guys from AnyBody are here on this class. They are sitting here-- [? Sorent ?] and [? Urgin. ?] So if you have any question about methodology, about modeling system, you can ask them-- whenever you want-- after this class or during Q&A section about details.

OK, so what we did basically. We decided to optimize the bike frame. So we asked guys from AnyBody to provide interface forces acting on the bike. They were able to provide this kind of information using their model repository. And as you can see, there is a body model cyclist on the bike and interface forces can be calculated using their modeling system.

So basically, what we get? We get crank loads, saddle mounting loads, headset loads and wheel bearing loads. So these forces can be exported and be used as an input to Autodesk Generative Design. One slide about measurements-- so basically, we compute or measure forces in 6 points. Two points on saddle, [INAUDIBLE] saddle [INAUDIBLE], two points on the headset and two points on the pedals. And the first question is, why two points for one component?

Because when we started this project, it was some time ago-- moment load was not implemented in AGD yet. So we had to find a workaround to implement moment loads into our program to finish them. So we decided to compute, or measure forces for 2 points and incorporate moment load into force. That's the reason.

Of course, when we are thinking about bike frame, we cannot use only one load case as steady riding a load case, because we need to create bike frame, which is appropriate for sprint, braking, steady riding case, et cetera. So we request that to anybody to provide forces for this free scenario-- steady riding case, sprint and braking.

Let's start with the first scenario-- steady riding case. So basically, cyclist is in the aero position. The power generated by cyclist-- 340 watts. Seated posture, arm support. That's the first scenario. Second scenario-- maximum effort case-- sprint. And what is interesting here is this very high power 1,300 watts-- this value is quite big, I would say. Standing posture, hands on grips.

And the last load case-- braking. So basically, it's a kind of emergency case. Braking is very, very hard. So full break decceleration-- near rear wheel lift off. Zero pedal power, seated posture, arm support. OK, now I would like to show you how the measurement looks like and what kind of forces we have here.

So basically, this slide is about pedals. So you can see that for sprint scenario, the force acting in the vertical direction is over 3,000 newtons-- it's really a lot-- two times bigger than for steady, right? And for the braking, the force is very, very small, because as we said, they cyclist is not pushing pedals during this scenario.

The same set of data for headset. What is interesting here that the maximum force is for braking scenario, because cyclist during braking is pushing handlebar. And that's why the maximum force is generated for breaking scenario.

And the last scenario-- and the last component-- saddle. Here you can see that for sprint force equal zero, because cyclists is standing on the pedals. And the forces acting for study ride and braking are quite the same.

So summarizing, 3,000 newtons on pedals-- max force, 800 on headset, 400 newtons on saddle. Why I'm showing this kind of information for you? Because basing on that, we can try to guess what will be generated shape? Let's take a look on the pedals.

3000 newtons acting here on crank. I will tell you that this model will be constrained here on the fork interface in here, where we have rear wheel brackets. So basically, disconnection. And there will be a very big connection, very strong connection between crank and the rear brackets and the crank and the fork interface.

Headset-- I predict that we will have something like that. There is a force acting on the headset. And again, this force will be transmitted to the constraints like that. And last one, saddle, this force is much more smaller than done for previous scenarios. So I expect this connection will be not very strong here.

OK, so having all the forces measured, we can start to define problem definition out of this generative design. I know that Kenny already showed this information, but I will tell that once again. We have two types of geometry in AGD-- preserve geometry-- so something we want to include in the final design. So you can see that we have a lot of cylinders in this example on the right side. And all these cylinders are included in the final part.

And the second type of geometry-- obstacle geometry-- kind of geometry we would like to exclude from the final shape. So everything what is marked in red is exclude-- this area is excluded from the design space. So that's why all the cylinders are clear inside, et cetera.

OK, so we can start to define our problem definition AGD. And I said already, we don't have any cut preprocessor in AGD, so we need to prepare geometry and extend our cut software. I used Fusion 360 for this purpose. And that's my preserve geometries. So basically, what should be included in this geometry?

Ever think what is an interface? So basically, I need the head tube to assemble this frame over the fork. I need the seat post and maybe place for the seat post to assemble saddle. I need crank to assemble pedals. And I need wheel bracket gets to assemble rear wheel.

After that, I can add obstacle geometries. So that's quite clear that we need to secure area for the front wheel, rear wheel and some area for turning, because of course, the front wheel will turn. All internal tubing areas are also secured, because with that solver for sure will generate something inside tube, so we need to secure everything as on the slide.

And the second part, because that's not all, as you can see, I created obstacle geometry also for cogset. You can see it here. Of course, cogset should be on the right side, but when I'm defining problem like that, I'm trying to keep all obstacles symmetrical to get symmetrical frame. So that's why you see this obstacle on the left side also.

Fork interface-- there is obstacle for fork interface. Because this interface needs to be clear without any material generated in this area to assemble everything without any problems. So you will see that the only surface available for the head tube is here. And here, probably [INAUDIBLE] will add the material.

And the last obstacle in this problem-- definition-- seat post side surfaces-- they are here. And the reason why I added that is because I wanted to have this frame connected to the seat post from the bottom, not from the side. And that's the only reason why it looks like that. Having preserve and obstacle geometry is defined. We can start to define constraints.

For example we used free constraints. So basically, you can see that rear wheel is fully fixed for the rear brackets. And there is also a constraint on the head tube. And here, one direction is not fixed. So basically, vertical and side directions are fixed and the horizontal direction is not fixed. It corresponds to the test conditions we had during measurements in AnyBody, because basically this bike frame was installed on some kind of training support. So that's why those supports looks like that.

And the last thing-- forces. As I said, we had forces from AnyBody. So all the positions, all the values, and all the directions were known, I just added to existing geometry. And that is how the fully defined model looks like. So we have loads, we have constraints, preserve geometries, obstacle geometries.

Last thing, before we will click Generate, is our objectives. So I select to minimize mouse and various factor of safety, because I run a lot of simulations. So for some of them, it was 1.5, for some of them, it was 6. Manufacturing, I selected option, unrestricted, because I didn't want to create any additive manufacturing constraints. And material-- aluminum 6061.

Now we are ready to click Generate and see some results. OK, so that's the first thing. Let's take a look for the frames generated for particular load cases. On the left side, you can see sprint scenario. As I said, there is no load acting on the saddle here, so that's why this frame-- there is no connection between saddle and the rest of the frame.

On the right side, there is a breaking scenario. I said that there is quite a big force acting on the handlebar, so that's why you can see this nice organic structure here near the handlebar, because of this big force acting for this scenario. And the middle one-- steady ride.

Having that and combining all these frames, we can get the final generated frame. On the right side, you will see how the optimization process looks like, how [INAUDIBLE] removes material. And now I would like to go to the final section on my presentation, to the renders, because I created many, many different frames. That's not the only one type.

OK, so let's take a look on this one. As I said, there was a quite big force acting on the crank, so that's why this connection here is quite big and that's why there is quite nice organic structure here, quite a big organic structure here. That's how it looks like from the front and from the back. That's the first type of frame I got.

The second type of frame-- this frame is slightly different, because we don't have direct connection from the handlebar headset to the seat post, but I think it's also quite interesting, something like that. And at the very end, I would like to show you the most optimized frame.

There is one question, why it's so thin? And I will explain that. For now, we don't have buckling analysis in AGD, it's only static analysis. So does the first reason. And the second reason is that in a real bike, we would probably use tubes. And AGD for now generates [INAUDIBLE]. And that's why this part looks like that. However, the weight of this part is quite good. As I remember, it's 1.3 kilogram, or something like that.

And basically, that's all from my side. We have still-- we have still a lot of time, so maybe Q&A session. So Kenny.

KENNY CORNETT: OK.

AUDIENCE: [INAUDIBLE]

MICHAEL MUSIOL: Good question.

KENNY CORNETT: Well.

MICHAEL MUSIOL: [INAUDIBLE]

DOUG: By the way, this is Doug, Connect product manager over Generative Design Strategies. What was that?

AUDIENCE: [INAUDIBLE]

DOUG: Like how you would put it together?

AUDIENCE: [INAUDIBLE]

AUDIENCE: [INAUDIBLE]

MICHAEL MUSIOL: Basically, I've run four simulations. I can run a simulation with all scenarios together, but also remove some scenarios to have only one-- braking, screened, et cetera. So that's how it was done. We don't need to mix all scenarios together, we just run four separate simulations-- one for braking, one for sprint, one for normal, steady ride and one, which is a combination of all scenarios in the same time. Yes?

AUDIENCE: I'm assuming based on the outcome that there's no size limit?

DOUG: Can you repeat?

AUDIENCE: I'm assuming based on the outcome of that model that there's no size limit increase? Obviously, there's [INAUDIBLE].

DOUG: So the frames are upright, there's no side loads on it as if it were turning or anything.

MICHAEL MUSIOL: That's partially true, because we had some side loads, but as I said, the bike was mounted on the kind of support.

AUDIENCE: So when you set up your obstacles, I noticed that they said [INAUDIBLE], but you didn't put anything in for the pedals in the rotation. [INAUDIBLE]. Did you run into a problem when you ran it because--

MICHAEL MUSIOL: To be honest, I worked with the software for a while and I know where obstacles are necessary. But to be sure that everything is created OK, we should add obstacles also for crank, for the side surfaces of crank.

DOUG: But you kind of have to think about where the preserved geometry is and where the loads are traveling. So if they're in the same plane, you probably don't need to go too far out of that plane to keep it from getting all crazy.

AUDIENCE: [INAUDIBLE] correct, but I was just curious why you'd put-- you did an obstacle for the back side of the front wheel so it would turn, but you didn't do it for the pedals or anything else.

MICHAEL MUSIOL: Yes, there should be an obstacle, but I knew that in the final solution, this area would be OK. Sorry for that.

AUDIENCE: [INAUDIBLE]

DOUG: Yeah, one of your objectives can be stiffness versus strictly mass reduction, you can do the stiffest part.

AUDIENCE: Yeah, what's the time frame for actually running simulations?

MICHAEL MUSIOL: It depends on the problem definition, but for frame like that, I had over 20 load cases, because I had three scenarios about many load cases for one single scenario. It's to [INAUDIBLE] one day, but for simple model, it should be about two hours to get to the final solutions.

AUDIENCE: [INAUDIBLE]. What are the driving forces to make it longer versus--

MICHAEL MUSIOL: Number of load cases-- that's the first thing, the resolution settings, because as Kenny said, you can adjust the voxel size, so having some smaller voxel size, of course, the competition time will be longer. And that was the second thing. And what else?

AUDIENCE: [INAUDIBLE]

DOUG: Yeah, so things like material choice. So a part made of titanium can have a different size and shape than a part made out of iron. So the more material that you include in this, the more outcomes you get out of it. And then--

MICHAEL MUSIOL: Factor of safety also.

DOUG: Factor of safety and minimum element size. So what's the smallest component inside the part. So it's kind of one of these delicate balances where if we raise the resolution of the way up and we make that minimum member size really, really small, you end up with paper thin surfaces all over the place. And so you can get diminishing returns on trying to be really accurate and really powerful, and on the other side, you end up with really blocky shapes, because you don't have enough resolution and you don't have a small enough member.

And so there's like a balance that you have to strike and you really need to be aware of what your design space size is and then tweak those values so that they fit within a range that's appropriate for the size of the object you're working with.

AUDIENCE: [INAUDIBLE]

DOUG: Correct. Yes.

AUDIENCE: [INAUDIBLE]

DOUG: And by the same token, a lot of times, you'll find results that are not what you expected at all and so you'll say, we need to stop this one and backpedal and figure out what we did and change it. Other questions?

AUDIENCE: There was one thing I'll add before we leave is this product is available at [INAUDIBLE] if you want to try it. It's right across the hall. [INAUDIBLE].

DOUG: We'll let you go a little bit early. We threw a lot at you, but it's early in the morning so go get coffee.