When is a circle not a circle?

Talk about all things CNC
Carl Fisher
Posts: 127
Joined: Fri Mar 07, 2014 1:36 pm

Re: When is a circle not a circle?

Post by Carl Fisher » Fri Sep 26, 2014 1:33 am

Well, for comparison and if it makes you feel any better MakerCAM actually did worse than jscut on the holes. Instead of the problems above, it actually dug out a starting "bump out" for lack of a better term in each hole. When it was done with the circle you could see clearly where the bit entered as a bump out in the circle. And to top it off, the holes are actually too small even though I was using the correct 90dpi setting when I opened the svg. Not sure why the undercut.

Carl Fisher
Posts: 127
Joined: Fri Mar 07, 2014 1:36 pm

Re: When is a circle not a circle?

Post by Carl Fisher » Fri Sep 26, 2014 1:37 am

Oh, and I'm not using GRBL if it matters. I'm on LinuxCNC using a Probotix PBX-2 which is Xylotex based best I can tell

tbfleming
Posts: 89
Joined: Sat May 31, 2014 12:04 pm

Re: When is a circle not a circle?

Post by tbfleming » Fri Sep 26, 2014 1:58 am

Carl Fisher wrote:Oh, and I'm not using GRBL if it matters. I'm on LinuxCNC using a Probotix PBX-2 which is Xylotex based best I can tell
Right now jscut produces gcode which should work on any CNC controller. My original idea for Z tool-height probing just flew out the window; LinuxCNC doesn't use gcode to change configuration settings like GRBL does. Actually, the jscut side becomes a lot easier; LinuxCNC is designed for end users to handle that issue themselves by writing and debugging complex macro code in either Ladder or Python :twisted: . All jscut would have to do is issue a T# command and your macro would take over :D

Carl Fisher
Posts: 127
Joined: Fri Mar 07, 2014 1:36 pm

Re: When is a circle not a circle?

Post by Carl Fisher » Fri Sep 26, 2014 2:30 am

I'm the exception rather than the rule really. I know most SO users are GRBL.

I just want something that produces a part that will match on the drawing I created. I never realized how many ways there apparently are to interpret a drawing.


Oh, I'm not sure if you're aware of this one but I do have to edit all of my jscut generated gcode files. Something about the way you close out your file LinuxCNC doesn't like. I just grab the last few lines from another file and replace your last line and all is well. I can add that as a bug if you want.

tbfleming
Posts: 89
Joined: Sat May 31, 2014 12:04 pm

Re: When is a circle not a circle?

Post by tbfleming » Fri Sep 26, 2014 2:45 am

Carl Fisher wrote:Oh, I'm not sure if you're aware of this one but I do have to edit all of my jscut generated gcode files. Something about the way you close out your file LinuxCNC doesn't like. I just grab the last few lines from another file and replace your last line and all is well. I can add that as a bug if you want.
Please add it. Include the error message and detail what changes you have to make.

CastIrony
Posts: 650
Joined: Mon Jul 21, 2014 10:21 pm
Location: Portland, OR

Re: When is a circle not a circle?

Post by CastIrony » Fri Sep 26, 2014 7:04 am

tbfleming wrote:Other circle shape issues: I might have to dump snap.svg; it turns circular curves into bezier curves before my code gets to take a peek. Bezier curves only poorly approximate circles. I didn't realize how insane I was to trust snap.svg until just now. I fell for the clean API and the flashy website.
No idea what algorithm jscut uses for circle bezier fitting, but the trivial case with four cubic beziers (one on each quadrant) has very low maximum error, IIRC on the order of 0.1% of the circle radius.

I can pretty much guarantee that your machine has other sources of error that make bezier error insignificant.
Shapeoko 3 #192 - Complete! Needs limit switches.

Caesar S
Posts: 197
Joined: Mon Sep 29, 2014 7:19 pm
Location: Seattle, WA

Re: When is a circle not a circle?

Post by Caesar S » Wed Nov 05, 2014 12:01 am

Did you ever find out what the cause was? I've run into the same problem.

The flats happen at NW and NE directions as oblongs.

I've ruled out software as I'm using the same software as before ("ShapeOko" hello world plotted just fine when I completed assembly a month ago).

I've tightened up the frame X, Y, Z axis. Somewhere there is mention of belt pulleys, so I'm going to look at that and the X belt tension too, it seems dirt also gets in the X belt groove (I have skirts covering the Y axis on both sides).

I'm also cleaning dirt from the rails and v-wheels. I've had dirt there that caused bumps in axis movement vs smooth travel.

I'm glad I ran into this problem before attempting the 1000 x 1000 mm build and 3D printer conversion.

Help?

Caesar S
Posts: 197
Joined: Mon Sep 29, 2014 7:19 pm
Location: Seattle, WA

Re: When is a circle not a circle?

Post by Caesar S » Thu Nov 20, 2014 5:34 am

So, I've experimented a little which made the oblongs better. Note that I have "flats" on the diagonals of the circles. Which upon closer inspection is what Carl is experiencing.

The troubleshooting steps here are from easiest to most effort (i.e. try software then mechanical fixes).

1. Reflash to factory shipped grbl, just in case the configs are shot.
2. Run HelloWorld again in case the mechanicals are shot.
3. Check $0 is the same as $1, if not it will cause an oval shape in the x or y axis.
4. Reduce acceleration $8=25 down to $8=10 (this helped me as I noticed the carriages tend to overshoot when decelerating and caused oblong corners)
5. In combination with #4, reduce the feed rate to some 400 mm / min (avoiding deflection of mill), and plunge rate (to avoid the mill oscillating while drilling down), this also helped me
5. Square the x with y axis, then with z axis. I found my z axis teetering at 2 degrees forward (takes more effort, 2 hours, this helped me greatly)
6. Tram the surface of the waste board, that is put on a square endmill and run a pocket operation across the working area to level the work area relative to the endmill and spindle. Note that if you loosen the spindle for whatever reason you have to re-tram, pain in the buttocks.
7. I'm upgrading to Nema 23 and 9mm belt next as I'm scaling it out anyways.

I plan to take jobs with this so a perfect circle is absolutely necessary :!:

My circles are now tolerable on paper, will try on hdpe next. If that shows oblongs, I'm shooting the machine :lol:

I can see the oblongs but my wife can't!

twforeman
Posts: 1347
Joined: Tue Jan 29, 2013 4:51 pm
Location: Minneapolis, MN
Contact:

Re: When is a circle not a circle?

Post by twforeman » Thu Nov 20, 2014 2:21 pm

I've been chasing out of round holes on my machine since I built it.

I'm pretty sure my issue is flex in the Z axis.

However, you should fully calibrate your X and Y axis to make sure they are moving the distances you expect. The belts can stretch a little, and longer belts stretch more.

This is the previous method I used to test the accuracy on my machine and calibrate it: http://timf.anansi-web.com/wp/shapeoko-accuracy-really/ (and if you search for "accuracy" on my blog you can see some other stuff I did.)

I have a different method I use now, that I thought I had written up, but apparently not. I'll have to do that soon.
Ender 3 3D Printer
ShapeOko v3 serial #0004 - upgrade thread
All of my ShapeOko related blog posts

Caesar S
Posts: 197
Joined: Mon Sep 29, 2014 7:19 pm
Location: Seattle, WA

Re: When is a circle not a circle?

Post by Caesar S » Thu Nov 20, 2014 5:16 pm

What I've also noticed is flex between x and y axis. When you twist the x axis horizontally around the z axis you'll notice the wheel mounts on the y axis move about 3mm either direction.

It's noticeable in small circles, not so in large ones.

My solution when cutting small holes is to drill / file them round. Using slow feed rate also for small holes.

Works for me now.

I'm going to tighten up the y motor plates and see if I can reduce the flex and metal wheels if necessary.
Last edited by Caesar S on Fri Nov 21, 2014 5:41 am, edited 1 time in total.

Post Reply