topic Re: Generative Design Failing without Giving Reason in Fusion Support Forum

Generative Design Failing without Giving Reason

avoss2J5KM — Tue, 29 Apr 2025 14:53:38 GMT

Hello Folks. I'm trying to create a generative design for a thrust structure on a rocket. I did 1 round that worked, but the safety factors were still really high and so I wanted to thin it out more. I took the result from the first round and started a new generative design with that as the starting shape, but it no longer works. It continues to fail and not say why. Wondering if someone could offer some ideas as to why it is failing, I have tried a few different things on the setup to no avail.

"centering ring bio-opimization setup" is the first on that worked

Re: Generative Design Failing without Giving Reason

henderh — Tue, 13 May 2025 23:37:25 GMT

Hi @avoss2J5KM,

Thank you for attaching the setups and a detailed description of the issue. Unfortunately, I wasn't able to tweak the setup to arrive at successful outcome generation in the setup with the starting shape from outcome as well.

I've posted a request for our Generative Design experts to investigate further, via an internal discussion forum. I'll definitely circle back here with any news to share.

We appreciate your patience, and apologize for the inconvenience this has caused.

Best regards,

Re: Generative Design Failing without Giving Reason

henderh — Wed, 14 May 2025 22:30:31 GMT

Hi @avoss2J5KM,

I have a minor update...the Generative Design Engineering team has determined the cause of the issue.

We have yet to determine if there is an easy temporary workaround which we can provide, until a holistic solution can be formulated that can be included in an upcoming release.

Re: Generative Design Failing without Giving Reason

henderh — Sat, 17 May 2025 00:24:57 GMT

Hi @avoss2J5KM,

In a nutshell, Development says the primary reason for the failure is caused by the relative thinness of the shells, especially the outer one (0.027mm) in comparison to the overall model size.

Under the hood we use a Voxel mesh to define the initial convex hull shape. I've heard we can visualize a Voxel mesh as creating the overall shape using stacks and stacks of millions of sugar cubes. The element size can be changed from the default setting using the Outcomes resolution accuracy slider in the Study Settings. However even at the highest resolution, the smallest voxel mesh element size is still to large to "fit inside" the super thin walls.

i.e. '...we cannot voxelize the mesh according to the preserve shape(s) that are specified.'

Possible Workaround: Increase the thickness of the Preserve body shells.

The minimum thickness threshold to allow a successful outcome iteration is 0.7mm. However, if you can increase it to 1.0mm or more, we can yield even better outcome shapes:

Hope this helps!

Best regards,

------

ps: We tried pushing the mesh to its limits, and still couldn't maintain fidelity of the original 0.027mm thick preserve geometry, even with 100 million elements! These images are from an internal development viewer, I thought you might also be interested see "behind the curtains" what was tripping up the solver. If we encounter a disconnected shape, it's a detrimental problem.

Original setup, with highest accuracy setting (smallest voxel element size)

"Turn it up to 11" (somehow via an internal development / research tool that is outside of my knowledge domain)