Consistent crashing on shell attempt: geometry too complex?

[thoreaubakker]

Greetings.
I posted about this in Design/Validate, but was told to post here as it's a crash.

I've imported a 9k quad test model and converted it to T-Splines then Solid.

Solid looks pretty clean to me, but when I try and shell it my Fusion crashes everytime.

Can someone confirm on their end? Fusion .f3d project file attached here.

It could be my PC.. Serious system (i7 / 1080Ti / 32 gigs ram) but been having some crashes in games lately I believe might be related to the video card driver. *oops forgot I was on my girlfriend's rig.. not a beast but still pretty decent.

Solid to shell

[g-andresen · Accepted solution]

Hi,

it works if t ≤ 1.15 mm

günther

[thoreaubakker]

hi @g-andresen 

thanks very much for trying that.

Did you find your PC hang? Did you freeze, or did a bigger shell just give you a warning / failure notification?

When press the shell command part of my screen goes black and I freeze.

[jeff_strater]

Fusion does not "crash" on this.  I was able to shell it to 2mm.  You are asking Fusion to do something complex, it is going to take a long time.  It probably took 10 minutes or so to do this shell:

[g-andresen]

Hi Jeff,

that's interesting, at that level I've a crash or freezing.

How is it possible?

@thoreaubakker  no warning!

günther

[thoreaubakker]

Thanks @jeff_strater  and @g-andresen 

Jeff -- thanks so much for taking that time to download and test. Really appreciate it my man!

It is really interesting.
On my system I am getting the 'Fusion is not responding' and the screen goes partially black on one side.
Perhaps it's a matter of system strength, where Jeff's PC is a beast and Guenther's is better than mine.
I will try again and lave the PC for 20 minutes to see.

[thoreaubakker]

@jeff_strater  

I tried a 5mm shell and left it for 30 minutes Looked like it wasn't responding but I clicked 'wait for program to respond'.

After the 30 minutes, Fusion was back, but no shell (body had red on certain faces?).
Couldn't take a screen shot during freeze.

Interesting that in the task manager PC is only using 15% of the CPU.

Any suggestions on alternative workflow? I wonder how toy designers work with shelling more complex geometry for injection molding.. Would they use a more advanced software like Inventor, or would they just work in the mesh domain?

[thoreaubakker]

Possible alternate workflow for anyone else struggling with this:

- if I duplicate the solid and scale it (say .8 %), I'm easily and quickly able to use the smaller as a cutting tool.
I wonder why it's so much harder for the software to shell it. Would be great to have a 'less accurate' shelling option or something like that. The downside of using a scaled down solid as a cutting tool is I imagine it's not even wall thickness throughout the model.. Maybe would cause issues when molding / casting.

[jeff_strater · Accepted solution]

I was able shell at 3mm.  It took 15 mins. I have a decent laptop (early 2018 Macbook pro), but it is no monster.  When a computer program is very busy, the operating system is not able to tell whether it is still running or not, that's why it enters that "not responding" state.  It could mean that it is truly hung, or it could mean that it is just very busy and will eventually finish.  In a true crash, the application stops running entirely, and usually, in the case of Fusion, you should get a crash report.  Please send those in (with your email address filled in), if you see them.  As to why Fusion is only using 15% of the CPU, that is likely because you have a multi-core processor.  Unfortunately, many of Fusion's geometric modeling algorithms are inherently single-threaded, so it can only use one core.  

Regarding alternate workflows, in this case, because the model is a T-Spline, you can use T-Spline Thicken to do a pseudo-shell.  This is an inherently much simpler operation.  A solid Shell is defined to produce a body which has uniform thickness everywhere, to a high degree of precision.  That is why it takes so long.  A T-Spline Thicken operation, on the other hand, is by definition an approximate operation.  The control cage of the T-Spline is offset and the offset surfaces are re-interpolated.  You don't get the guarantee of uniform thickness, but you also don't have to wait 15 minutes for the operation to complete.  I was able to use T-Spline Thicken to produce a pseudo shell in maybe 20 seconds.  Exiting the Form workspace, of course, will take longer, because there is more geometry to build, but even that was pretty quick.  The downside is that I was unable to do 3mm.  The resulting inner T-Spline had self-intersections that I did not bother to try to figure out.

Also, the T-Spline model here has a LOT of faces.  If you can simplify the original quad mesh, the resulting surfaces will be much less complex, and that will improve your life a lot, I'll bet.

I've attached the version that uses Thicken at 2mm.

[thoreaubakker]

@jeff_strater 

thanks for taking the time and for the very helpful information.
I had no idea about the t-spline offset operation -- I will definetly give that a go as I'm sure that (when it worked without intersecting) it would be more accurate then just the  duplicate/rescale/subtract approach I had mentioned.

- noted about the number of faces.. I was pushing it at 9k.. I'll try some models with 500 - 1k and see how that goes

- will def. send crash reports if I get them, although with the t-spline thicken I may end up avoiding that slower approach all together

Thanks again for the helpful support and information. That's really great.

t.

p.s. interesting about the single core being used. I wonder if that means we could have multiple iterations of Fusion going, to be calculating different projects on different cores etc..

[mcramblet]

I thought I'd chime in, as I do a fair amount of conversions of meshes to solids via Fusion 360. I re-meshed the head into something much more reasonable for the amount of detail the model had (around 3,000) and Fusion 360 was able to shell it instantly. Here is the re-meshed file and the result:

https://a360.co/38nDitb

A high density mesh is only needed when there is detail that needs to be maintained. The lower the mesh, the better everything works and the cleaner the translation to a solid.

[thoreaubakker]

hey @mcramblet 

Thanks very much for trying that! 

I acknowledge that was way more polys then needed for that geometry. I wasn't actually trying to do anything with that mesh, it was just an experiment to get a feel for what's possible.

When modeling more complex things, a 10k poly budget is actually pretty low. Do you have any experience with trying to balance mesh reduction when it comes to more complicated things? I'm interested in toy design, and taking complex mesh geometry from programs like ZBrush and Blender, then bringing them in to shell etc. The desire to have them in the solid space is partly for accuracy, and partly for the speed at which boolean operation happen.

MeshMixer (another Autodesk product) is good at shelling and booleans, but they take a long time to compute. Once I have solids in Fusion, they are a delight to work with.

I'd be curious to hear more about your personal experiences and approaches to turning meshes to solids, especially anything complex!

[mcramblet]

Yes, I've done a fair amount of work doing this type of thing. It's an interesting workflow that requires a lot of trial and error. If you'd like to see examples of my workflow and the level of complexity, look here:

https://grabcad.com/mcramblet-1

All of the models I post here are just experimentation and practice of this type of workflow. These are all examples of taking either 3D scans or mesh files and converting them to true CAD (STEP) solids.

[thoreaubakker]

@mcramblet  that's really neat work!

Cool to see it in the 3D view on GrabCad.

Have you found any max poly count that works for you in general? Do you stick to the 10k limit or go far below it?

[mcramblet]

I stay under 150,000 faces per body. On average I try to stay in the 100,000 to 130,000 face count range.

[thoreaubakker]

@mcramblet 

Wow. Am I reading that right?

You're turning quad meshes into t-splines into solids at 100k?
Or is it just mesh into solid?

I remember reading that only 10k was the limit for conversion, so hearing you're doing 10x the limit (or more) is surprising!