How to make your STL not take 3 days to calculate tool paths for


(Jonathan K) #1

So far I’m really enjoying working with my Nomad, and I’m learning a fair bit of best-practices I thought I’d share. Today’s tip?

Make sure you’re using optimal mesh resolution.

Last week I exported a sculpt from 3D Coat, did a mesh reduction in Rhino, and then brought that into MeshCAM and proceeded to set it up for tool-paths. I originally used too small of a tool step-over for finish passes, with too small of a tool on the first round, and in calculating the paths it took three days on my core i7 workstation notebook, and the resulting paths were going to take an extremely long time as well (calculated at over 30 hours, but likely realistically much more than that).

I knew I could get a result more reasonably calculated result, and in less time. So I stepped back and re-visited the mesh in Rhino and reduced resolution considerably further to make sure the mesh couldn’t have more detail than the machine could possibly put into it—I put the minimum segment distance at 0.001", and the maximum deviation from the original surface at .001", with the maximum angle deviation per facet at 5 degrees to ensure smooth curves.

Then I took that result, and ran the mesh reduction to reduce co-planar triangulated areas, and then ran it again to cut the mesh density to 1/2 of what it was from that result.

I then exported that, and re-calculated paths. It was done calculating in under an hour, and I was done cutting three hours after that.

Moral of the story folks, is reduce your meshes.

Ideal STL parameters are to match the resolution the machine can make, and set the accuracy in MeshCAM to .0005" or less accurate for most things, it’ll still be just as good as the machine can be expected to actually perform in most materials.

I cut the first round of that file in foam, so I’ll be re-cutting it in something that can fully hold the details after I’ve made some design changes, and I’ll try to post that in the next few days.


#2

UnionNine, I’ll start by admitting that I don’t have a Nomad, so haven’t had a chance to use the Nomad release of MeshCAM. But I have been using MeshCAM since Beta 5 in 2003. Your experience does not seem right to me, unless Robert is doing something radically different with the Nomad release of MeshCAM.

I dusted off one of my largest STL’s from 2009. It is 49MB and contains 1,004,957 surface facets. I generated it in SolidWorks with .0005" deviation and 2 degrees angle. I modeled the rawstock along with the geometry, and it is 4" square and .550" tall.


My PC has a 5-year old Pentium Core Duo at 3GHz and 4GB memory, running 64-bit Windows 7. Here is the toolpath parameters screen. The actual finish machining back in the day was with a .094" ball endmill, but the stepovers and stepdowns are representative of “back in the day”.

The toolpaths took 39 minutes to calculate, and they are beefy! :smile:

This resulted in a 695,000 line gcode file.

This is probably a little steeper than I’d expect of a Nomad, but I’d like to be able to downsize to a Nomad or similar when the day comes I sell my Tormach mill.

UnionNine, would/could you upload the STL as you originally output it? I’d like to give it an independent try with off-the-shelf MeshCAM as a test. If it’s too big, PM me and I’ll let you know my Gmail address.

Randy


(Jonathan K) #3

By all means, I’d love to have my case be the odd one, because that was nuts for it taking what it took. I will say I had the paths calculating at too fine a resolution, but it was “offsetting mesh” and “refining toolpaths” for an especially long time in particular.

Here’s the STL original, the first refinement, and the most reduced version. Have at it!

I believe this was from the simulation of the “first refinement” version:


#4

Thanks, UnionNine. That is a big model! 2 million facets, almost twice as large as my largest… Is this a mm model? I have MeshCAM set for “Always use inches” and get

I forgot to also ask for a screenshot of your Toolpath Parameters screen, so I know how to set up the job.

Randy


(Jonathan K) #5

That was in MM, yes. Should be about 7.5"x3" on the X&Y, and I had set it to only cut a partial depth (no need for the whole block there).

Like I said the tool-path settings were unrealistically off on the first run, which also contributed substantially to the calculation time. The roughing tool was the 1/8" flat-end mill, then I had programmed in a tapered conical bit that I’d put in too small of a point @0.127mm for the tool (.005") for the waterline and pencil-finish, with a step-over which I think was .002" or .003". Yet again, this is assuming I remember correctly b/c I didn’t grab the settings/save the job after it was done calculating.

The waterline finish is the vast majority of the program, and the program didn’t include regular parallel finish, so as you can imagine waterline on the geometry above is a LOT for it to think about with such an non-optimal file.

Here’s the actual NC code if you want to pull it into something to simulate/examine actual step-overs or other details.


#6

OK, I’ll play with the file tonight after work. When I imported your first STL into MeshCAM (screenshot above) even if the units would be mm it is still 137 x 252 x 52 roughly, or 5.4 x 10 x 2 inches roughly.

But I will say that the contours of your part are totally inappropriate for waterline + pencil. I would finish them with X-Y parallel finishing, and then mask off the top surface with a keepout region and use pencil+waterline for the perimeter only.

I learned that lesson a few years ago when I tried using waterline on a lithophane I was doing. I agree, it will churn for hours with no progress. Waterline likes smooth continuous curves like boat hulls or my locomotive wheel. Lots of little undulations like your door carving drive MC crazy figuring out how to do a continuous pass at a given depth.

Parallel finishing just takes a cross-section of the part and follows it. Much simpler on undulating terrain like your door (did that come from a point cloud scan or did you literally 3D sculpt it?)

Another factor if you’re using mixed measurements, MeshCAM might interpret the calculation tolerance as mm even though it says inches. I.e. typing in .001" was really .001mm… That’s what it seemed like from the behavior in the past.

Randy


(Jonathan K) #7

Hi Randy,

I agree wholeheartedly with your assessment of appropriateness of waterline pathing—I had put a fairly aggressive slope filter on it so that I thought it would only go around a few features that it thought would be worthwhile to go around, in essence filtering out a lot of the small-amplitude detail. I believe I may have also scaled it in MeshCAM prior to running it.

I just didn’t anticipate 1) screwing up the tool-path size as I had or 2) it calculating things out the way that it did, trying to evaluate everything as vigorously as it had. I also could have had mixed measurements, so who knows at this point—so the additional important lessons beyond just making your mesh resolution reasonable are to check units and the appropriateness of the path strategy to the piece!

I had sculpted that door in 3DCoat, which allows you to sculpt in voxels without worrying about mesh topology and then you can do a dense meshing or more sparse re-topology from there. It’s really powerful and a lot of fun, I highly recommend it!

-Jonathan


#8

Jonathan, my emojii next to the “inappropriate” remark didn’t come through. It was supposed to be like :grimacing: --not a rebuke but a warning from experience. And you are building up experience–the best teacher. And that is really nice sculpting. If you’d have told me it came from a point-cloud scan of a 17th century monastery door I’d have believed you.

Randy


(Jonathan K) #9

I didn’t take it as a rebuke, but as you meant it, because that’s what I too found out through experience :wink:

No harm, no foul, and definitely some different tacks on tool-paths for the next project.