Every vendor of generative design software knows Andreas Vlahinos. If they don’t, they should. Vlahinos, a bear of a man, is perhaps the foremost authority on the use of generative design. As a practitioner, instructor and consultant of generative design, Vlahinos is a breed apart from the rest of us who rely on demos and vendor claims to assess technology. When Vlahinos talks, design industry vendors listen. If they don’t, they should. Because all his opinions, frustrations and wishes—expressed at industry conferences (including COFES and ASSESS)—are pure gold in a world where vendors, high on their own fumes, maintain that the only reason their technology is not being adopted wholesale is because of ignorance and inertia.
Vlahinos has consulted with PTC, among other generative design vendors, on the development of generative design. PTC may have actually listened to Vlahinos and implemented his suggestions. You sense this from Vlahinos’ recommendation of Creo’s generative design.
Vlahinos considers PTC Creo’s generative design “ready for prime time.” The implication is that other vendors’ products may not be.
What follows is our conversation regarding the state of the art in generative design in general and of PTC Creo specifically. We interviewed Vlahinos via Zoom at his home in Castle Rock, Colo. It’s been a couple of years since we’ve seen each other, thanks to the pandemic.
Engineering.com: Did you go to ASSESS this year?
Vlahinos: I did not. He [Joe Walsh] did not ask me to give a talk. I go when I can add value. I miss COFES.
Me, too. COFES was like a Who’s Who of our industry. And now we have nothing. They're trying to bring it back every year, but the last two years … nothing.
You need a champion. And Brad [Holtz] was a really good champion. After he left.… Anyway, tell me what’s exciting.
We’re doing a series on generative design. And we thought we’d interview some experts—starting with you. Is that okay?
Sounds fine. I use PTC generative design and I teach it, so I’m very familiar with it.
Great. What do you think of it?
There are four big benefits. First, the usability is amazing. It’s very simple. The second big benefit is when you’re done with a generative design, you can get a CAD model.
You get a solid model? With their geometric kernel?
Yes. Granite. You get a subdivisional surface model. The third big benefit is they separate the attachment geometry—what they call “preserve geometry”—from the generative design geometry, and then they Boolean it at the last minute. So that way, if the preserve geometry has holes in it, for example, you can put GD&T [geometric dimensioning and tolerancing] on it. That was a big bottleneck before. The shapes were so sculpted, you couldn’t get it manufactured down the hall. The fourth big benefit is that it is very fast compared to others.
What are the disadvantages?
The disadvantage is that they have very limited physics. Right now, it’s for structural optimization only—no fluids. We would love to have fluids. We find workarounds, but without fluids, there is no thermal. At this point, they have only structural and modal.
The solid model—is it smooth?
Yeah, very smooth. Let me show you….
You don’t have to straighten it out, remove bumps … anything like that?
Give me one second. Let’s see if I can find the presentation. That will answer your question. You will see. You expect that bones won’t have sharp corners, right? Okay, see this. This is from my class. They just finished it. They started with something [that was] a simple structure. This is a mirror for a telescope held on the sides. There is a pressure applied and then a generative design is done. I show them how to set up the generative design, add material limits and sufficient design constraints. Then there is the outcome of the generative design. This is a crazy one. Because we tried to do it with 3D printing.
Have you told the generative design that there will be symmetry planes or axes?
No, I did not. I could have done that. But I wanted to test it. You see it is almost symmetric.
It does look symmetric from here.…
It’s not perfectly symmetric. We did force it. You’re right. Because this was a class exercise.
It does look very smooth, though. No lumps.
Yes, I didn’t have to do anything. You can see as I slice it. Here, see three branches grow from the ground. It was a nice thing when PTC bought Frustrum. Frustrum was very good in generative. They couldn’t do it by themselves. Here is the subdivision on surface modeling in all these fingers, and it finds out the load path along the fingers and subdivides.
And this sectioning—I can see more clearly the asymmetry.
You’re right. But when you educate young kids, you need to go step by step. You can throw them all the stuff at once [but] they get overwhelmed. Let’s try this and then you realize, oh, it’s not symmetric. There is another thing we can teach. Then we can force symmetry.
What about lattices?
Another nice thing about PTC’s generative design is they have really strong lattice capability. So, I have the students do a lattice infill for comparison.
Won’t lattices be hard to manufacture?
Yes, the manufacturing of lattices will be hard. But you need it because you will get support.
Can you force generative design to generate shapes that can be machined?
Yes, here is one that was an extrusion. So, it can be milled on a 2 ½ axis.
Is there a box you can check off to create an extrusion?
Yes. In Creo, they call them design constraints. To have an extrusion, you can specify the extrusion angle, and if you want it to be bidirectional or not. By the way, I did one symmetry plane. I should really have three symmetry planes. But again, this is a class that you know at Vermont Tech, so I need to go slow. Anyway, I hope that answered your questions about smoothness.
Yes, you did. I see you may still have to 3D print the part, but at least it doesn’t have lumps. Is there a way to specifically call out the manufacturing process—CNC, for example, so it will produce a machinable part?
Yes, that was the one I showed you, the extruded part. There is one manufacturing constraint missing, though: sheet metal. It does not have sheet metal. That’s really a limitation for the automotive guys.
What else is missing?
They [PTC] do not have traditional objective functions, such as the ability to minimize the reaction. I don’t know if you want to go this far, but when we design a material like lattices, the lattice overall has weird material properties, like a negative Poisson’s ratio. Or it has orthotropic material properties, like 10 percent stronger in the Z direction than the Y. When we need to design the lattice with zero coefficient of thermal expansion, we have to have the generative design minimize the reaction. We have deflection but we want to minimize the reaction, but if something has a negative Poisson’s ratio and we squeeze it in on all four sides, we impose deflections. It doesn’t have reaction constraints. But then you have to remember the program is designed for designers. Otherwise, the interface would be too intimidating. Keep it simple. Very simple. You specify a design domain. Here is my preserve volume, which you keep the optimization out of, here are the pivots, the exclude domain, so you don’t build anything there. You just click on bodies. Then you say what to minimize, and usually that means make the volume a fraction of the initial volume, subject it to the constraints—either manufacturing constraints or stress constraints.
They try to make it simple, but … there is no displacement constraint. Say I want to deform the mirror but no less than four millimeters?
So, there are some … some limitations, but overall, I think it’s very useful. It’s designed for designers. And it can be used to create innovative designs today. The fact that we get a CAD model is a very big deal. That lets us manipulate the resulting shape. You need to combine human intelligence with the generative design. It comes up with some crazy stuff.
Any other limitations that were made for the sake of simplicity?
Yes. Another limitation is they don’t have a minimum element size constraint. In aerospace, we have rules of thumb. Never have a wall less than three millimeters thin because of corrosion, never have rods less than five millimeters in diameter because of buckling, etc. We have developed all these rules of thumb over the years. We can’t impose them here—not easily. We have to find workarounds that give you limited controls on controlling element size.
I have shared some weaknesses, but where Creo generative design is strong. It is very strong. It gives a robust data reconstruction. It is smooth. It does not look ugly.