Large/complicated model - modelling practices

[ctrotterTUYYN]

Ohai folks!  I've been using Fusion for 4-5 years now in a pseudo-hobby capacity.  I say pseudo because I design keyboards/pedals/devices that help me keep working through chronic health issues, probably 2000+hrs a year?; my day job is not product design/cad.  (Intel Macbook Pro, 32gb)

While I've hit limits in the past, I've always come away with 'it's just bad practice on my part'.  And I'm pretty sure that continues to be the case today.

The archive file is too large to upload (145mb), so here's a dropbox link: mk21-components 

I would note that the project is in the middle of a refactor, so there might be a few oopsies - but generally as a rule I aim to keep the design timeline clean of errors.  Also, I'm only halfway through the project at this point - this components assembly will then be inserted into a shell assembly where I'll hash out the aesthetic coverings.

 

Previous examples to give more flavour:

ArcBoard-mk20 

ArcBoard-mk19 

ArcPedals-mk3 

I compiled a list of actions (incomplete...but tells the story) that cause a ~10s spinning beachball.

• sweep using guide-face that is a complex loft, several minutes
• fully constraining a profile and orbiting
• editing a sketch such that more profiles are added
• editing an extrude/join to include more profiles
• hitting escape on a chamfer command that was not computing
• hitting escape on a feature edit
• hitting escape on a dimension that was aligned wrong (so it forces a dimension anyways)
• making a sketch visible
• deleting a horizontal/vertical constraint in a sketch
• hitting escape to clear a selection
• deleting a line in a sketch
• renaming a sketch
• creating a sketch on a plane
• unisolating a part
• deleting a line w. constraints

Of note, the behaviour comes and goes, but every minute or two I'll have to sit for ~10s while it chugs on one particular action.

On this design:

Component.Counts 
With Overrides: LeafOccurrences 798: Bodies 1947: VisibleLeafOccurrences 256: VisibleBodies 568: LeafOccurrencesWithVisualMaterialOverrides 0: OccurrencesWithTransformOverides 0

Some notes on my workflow:

• Lots of inherited assemblies; e.g. the components assembly has a trackball assembly which has its own sub-assemblies including a STEP file of a pcb
• If rework is required, I go back on the timeline and fix it, then fix any downstream explosions (vs. applying patches over patches over...)
• I try to avoid sketch patterns, but lately I've taken to doing things like fillets as arc+tangents+dimension (because I find it easier to update sketches than figure out where that fillet operation was on the body that was radically changed; fusion sometimes loses the ref entirely and then you're guessing)
• Joints for locating sub assemblies in the main assembly
• Rigid groups for holding simple small assemblies together
• 'ground to parent' is on unless there's a joint in play
• Inherited 'cutter bodies' over duplicated sketch/extrude-cut situations (so I can combine/cut once)
• Parameters wherever a dimension is used more than once; or for joint fine-tuning; in sketches I then have dimensions reference other dimensions
• I try to keep sketches locked as a rule
• Where possible, use a single sketch to perform multiple operations (to reduce sketch/plane/etc operations)
• Don't model things like threads

Much of the above is learned through failure and very helpful forum posts.  Certainly more to learn.

Specific things I've tried...

• Fusion is set to 'performance' graphics
• Turning off analytics (didn't think it would help, lol, it didn't)
• Running the Fusion repair/cache delete
• Ran through the performance troubleshooting doc, and the Phil Eichmiller pdf 

Things still on my mind to try... (sourced from other Fusion support threads)

• replacing the pcb step files and key components (e.g. the switch itself is not simple) with simplified stand-ins
• finding base-level (no inheritance) components/assemblies and turning off their timeline
• try to reduce components with only an inserted project
• go back and find any duplicated/overlapping rigid groups
• maybe my emphasis on sketch constraints is a problem? removing built-in fillets maybe?

The project has basically ground to a halt, so any design practice pointers or 'hey dum dum you should/shouldn't be...' nuggets would be deeply appreciated.  Thanks!

[TrippyLighting]

I've spent some time looking through your model and I don't see anything jumping out as a bad technique.

There are only two yellow warning icons in the timeline, which are easy to fix and should not cause such performance issues.

Could you record a screencast of some of the issues you describe above?

I use OBS Studio. It is free and open source software and works very nicely.

[ctrotterTUYYN]

Ah!  Very generous of you, and good to hear haha!

I've been meaning to get OBS running anyways, will do after work today.

[TrippyLighting]

While nowadays I work with Fusion mostly on two different Windows 11  computers ( office office and home office ) I opened your model on my mid 2017 MacBook Pro.

I don't see any of the severe delays you are experiencing. The timeline is pretty long so that takes time to recompute after edits, but nothing out of the ordinary.

[ctrotterTUYYN]

Ok, so I went to record and two things happened:

First, OBS doing screencap just _crushes_ the CPU on this machine - Fusion performance absolutely craters - I can barely move around.

Second, while recording the issue only cropped up once.  😞

(i rebooted and tried again, fresh, and OBS recording 4k is just too much to ask)

But it's really as simple as, 'perform any action from the list above (and more), and get the spinny beachball for about 10s'.  That's it.

I suspect I just have let this file get too long in the tooth.  Normally I would have kept the 'keywell' part as another component, but didn't do that for some reason.

I've tried to avoid very long chains of inheritance, and it's awkward to rapidly iterate with that workflow.

e.g. in this context, you can have a serial order of operations until you get to tying the 'thumbcluster' (w. the ball/5 keys) to the ring gear assy and the 'keywell' (the normal keyboardey keys); and also to the lcd & encoder(roller) mount.  That interaction of complex parts is tricky b/c they all refer off of each other - i.e. 'which order' is not an easy answer.

Are there debug procedures or ways to see what operations it's hanging on when the beachball comes up? (when timeline recompiles I can see things, and in the past I've diagnosed problem areas - but not a thing w. beachball)

[TrippyLighting]

I don't see any of these issues on any of the three machines I've run this model on.

The timeline is pretty long and takes a few seconds to recompute, but that is to be expected. I don't ever see a beach ball. I can navigate through the model in the viewport without issues.

I can select features in the viewport and timeline. I can edit features and sketches and complete the edits fine. Actually, except for the timeline recompute, the model is snappy on all of the three machines.

I shared a screenshot of the 2017 MacBok Pro in my last post. My computer at work is a 6-year-old Dell Precision Notebook with 32 GB and an old "pro" NVIDIA graphics card. My machine at home uses a Ryzen 5950X with 64GB and an Nvidia RTX 3070 Ti. But even on my aging MacBook Pro, the model edits were snappy. No wait, stuttering in the viewport. Zero Beach Balls.

Something you can try is to upload your own design from the .f3z you share to a new folder and see how that behaves.

Also, could you share which hardware is in your MacBook Pro?

I assume it uses a dedicated Graphics card.

[ctrotterTUYYN]

Very strange!!   I'll try the export archive -> import to new section, see what happens.

Also, you're on MacOS Ventura, which is pretty old, wonder if that's a factor.

 

 

[ctrotterTUYYN]

Ok, it's been over an hour and no sign of the issue.

Hindsight note - I had been seeing data panel weirdness, like the thumbnail renders would be glitched or missing (like just a red box).  I was using the default folder for my active stuff, and like once a year i'd cull into an archive folder. 

So, theory - either the default folder has hidden limitations, or there is some sort of general constraint related to UI rendering that is a blocking operation, and the regular/timely beachballs were due to UI rendering jobs getting stuck.  Perhaps it's something that is triggered on job completion or cancellation - i.e. many people report issues when hitting escape in sketches, could just be having too much in the folder list that's crushing that one task?

Going forward I'll keep each keyboard iteration in its own folder!!!

So thankful for your help, @TrippyLighting !!  I'll report back if things break, but this is extremely encouraging.

[TrippyLighting · Accepted solution]

I am glad I was able to help!

In (very) rare cases, after many changes, projects get corrupted in some way. When you export such a design, that strips it of the design history.

Then, when you upload it, it looks like a new design to Fusion, with no links to older versions of the linked components.

[ctrotterTUYYN]

Ahhh, I'd only experienced 'corrupt' in much more obvious and dramatic ways before.  Good to know!