Size limits to Generative Design?

[JustaBill]

Is there  / what are the size limits to the Generative design workspace? Everything I've seen has been smaller parts/brackets etc with the largest part being a motorcycle swing arm. I imagine this is all because it's more intended use of Additive manufacturing so people only use it for parts that would fit, but is the generative work space itself limited , or is it that just no one has used it for something bigger?  3 story jib crane as an example.

[bmfrank · Accepted solution]

There is no "limitation" to the size of the solution, but you are correct in noting that the solutions that are put forth are dependent on the technology you have available to make or fabricate the item.  For instance, if you had the ability to cast a very large item like this building structural component there is no reason you couldn't use Generative Design to determine what is should look like and which material to use. 

We are working on additional synthesis methods that would allow for things like a "kit of parts" approach to determining solutions for a given engineering problem, and would be interested to speak to you more about your needs.   

[I_Forge_KC]

I'll add that size becomes a relative metric once you understand the concept of voxel density.

If you're working on a design for a structural member of a bridge (huge) but include details like a plate with machine screws in it (tiny), you're in for a bad time. I say that because one of two realities must be at play:

1) The voxels are tiny enough to capture the small details but in order to capture the entire design space, there must be an absolute ton of them (which would make performance suffer).

2) The voxels are large enough to maintain performance but therefore are too large to capture the small details.

In the current iteration of Generative Design, there is not a parameter for voxel size (either absolute or relative). We must assume that the balance is thereby pushed towards the performance end of the spectrum and the voxels are sized relative to the overall design space volume. Since the voxel size becomes some (unknown) relative size, the overall design space size becomes irrelevant.

(we assume performance because Autodesk has a flat fee for solves - if your solve took days or weeks, then the computing cost would outweigh the cloud credit fee)

[JustaBill]

That " kit of parts " sounds interesting, is that going to be something along the lines of splitting the part into a number of components to be joined via different processes ? ( welding, rivet, end flange nut & bolt, male/female epoxy, bailing wire and duct tape? )

[JustaBill]

I do believe that is the first time I have ever heard of a voxel.  So, it's like facets when working with a STL file but with a finite number of voxels that have to be spread through out entirety of the part ? Or to explain like I'm 5 , the legos can be any size I want but I can only have 100 to build a house.  

[I_Forge_KC · Accepted solution]

Voxel = Volumetric (3D) Pixel

Think of it like this. If you had to carve a 3D cat out of this block but you can only remove whole boxes, how cat-like would it be?

Ok, now try again on this one.

The idea here is that there is math performed on each little box in order to get your shape when you use Generative. The more boxes, the more detail can be obtained. The downside is that more boxes mean more calculations. 

So back to my original thought... the size of these two cubes are irrelevant but the number of little boxes inside are what really matter. Since we don't have control over how many boxes are there, we can assume they are probably larger rather than smaller.

Don't be fooled though, there are still lots and lots of voxels being used. You will just hit a brick wall when trying to generate designs that have a huge size with small details.

[bmfrank]

That's the point... we don't want you to have to worry about voxels.  😉

To use your analogy, you need to have the right size legos to capture the details of your design space.  if you have fine details in very large design space, you need to use a lot of voxels...

But you don't need to concern yourself with voxel counts or grid sizing. 

[JustaBill]

Voxel knowelege bomb was definitely a trip down the rabbit hole but, good stuff to be aware of. And glad I'm not the guy that needs to be concerned with it.  Thanks again 

[Anonymous]

How does a voxel differ from a brick  element (fea) in this case? The only other time I've seen that term used is with autodesk flow design. 

When I run a topology optimization on ansys (which I know isnt necessarily generative design) I can select refinement regions to solve the exact problem you've described. An intuitive mesh algorithm may even do it on it's own at times. I havent tried this with a local refinement in f360 but I would imagine it's the same as well.

What is the technical difference between generative design and topology optimization that prevents these work arounds (local mesh refinements)? Or is it just a state of technical maturity that will eventually occur as the product is refined?

[I_Forge_KC]

@Anonymous

The technical difference comes down to the underlying math model.

"Typical" FEA methods for topology optimization use what you would consider to be a normal mesh (which may have refinements both globally and locally). This is the SIMP method.

Generative (in this context) uses a level set approach. This is where the iterations come from that Generative creates. Levet set methods use a global system of voxels to move the surface boundary normal to the load path. This is why it gives the impression of melting through the iterations.

For a slightly more detailed (but not too deep) explanation, check this out:

https://forums.autodesk.com/t5/fusion-360-design-validate/what-is-the-difference-between-generative-design-fusion-and/m-p/8163616/thread-id/164959

***

I should mention that you could certainly have voxel refinements but it's definitely a more complicated measure and a more mature solution than what is currently implemented.