I know this was discussed previously, but any idea how to fix unequal X and Y axis cuts? The two panels for this cabinet are 1/4" off over 24", or about 1%, which doesn’t seem like much but is pretty wonky in the world of woodworking. The precision is touted to be +/- 0.5mm 
Pictured are the two panels on the bottom right. Please don’t ask to rotate the pieces, it’s the only way they fit and not really a solution. FWIW, those two panels were cut together from one NC program.
Out of curiosity can you mark on the full panel layout where these two corners are located?
Edit:
Oops I can see them now. One is on outside right edge. Is your frame large enough that all the cutting is in white area?
It all stays within about 2” from the edge. I also have supports that extend beyond the workpiece so the sled stays stable.
I’ll measure the last one I did, but pretty sure it came out the same.
the machine is very repeatable, we are still working to figure out the accuracy
issues.
which version of maslow do you have (2 chains or 4 belts)? if this is the 4 belt
version, do you remember what fitness score you got during calibration?
if you can map the distortions, you can correct for them in the cad.
unfortunantly, it’s not as simple as ‘scale X by 1%’ the errors are in curves,
not a simple linear percentage
David Lang
you may end up finding that the best thing to do is to find an area that is
accurate and shift the workpiece around. say cut 1/2 or 1/3 of the width at a
time.
David Lang
Version 4 with JSTs, extra thick 3D corner anchors, on a vertical wood frame, and software version 0.87. Fitness was about 0.75 when I calibrated a few days ago.
Any idea if this is just the reality of how accurate the M4 is these days, or maybe something wrong with my setup? Seems like others have flagged the same issue.
I think that you are on the bleeding edge of testing the full sheet accuracy of
the system, so if you can help us here, we can make significant progress is
figuring out what the accuracy limits are.
since you just did a calibration recently, I assume you are on the current
firmware.
0.75 fitness is pretty good right now.
I would check that your Z offset values are correct (you mentioned extra thick
anchors) and check if you have flex in your frame (a camera recording video
pointed at the anchor from close, then watch the video at high speed will help
you see the flex)
what are your frame dimensions?
can you measure your parts in several places to map where the distortion is?
(also check with a straightedge to see how curved the cuts are)
if you look at my frame imager http://lang.hm/maslow/maslow4_frame.html I would
be curious to learn if the parts of your cuts that are most distorted out
outside the white area
David Lang
I’d be happy to help. Considering loading a fresh sheet and plotting a grid of dots and measuring where and how it skews. I’ll need to print a pen holder though.
To answer your questions:
The leftover ply is still attached to the spoil board, so I’ll go measure and post how much off everything is from my design.
Here’s my frame if you’re curious. It’s got about 20° of angle on it now, not as steep as this photo, but you can see the extra diagonal supports.
Don’t think it’s related, but had some errors when cutting the last job:
[MSG:INFO: Tool No: 2]
[MSG:WARN: Encoder read failure on Top Right]
[MSG:WARN: Bad connection on Top Right encoder, failed to read 1 times in the last second]
[MSG:WARN: Position error on Top Right axis exceeded 15mm while running. Error is 33.655mm Counter: 1]
[MSG:WARN: Previous error was 33.655mm]
[MSG:WARN: Position error on Top Right axis exceeded 15mm while running. Error is 33.735mm Counter: 2]
[MSG:WARN: Previous error was 33.735mm]
[MSG:WARN: Position error on Top Right axis exceeded 15mm while running. Error is 33.789mm Counter: 3]
[MSG:WARN: Previous error was 33.789mm]
[MSG:WARN: Position error on Top Right axis exceeded 15mm while running. Error is 33.852mm Counter: 4]
[MSG:WARN: Previous error was 33.852mm]
[MSG:WARN: Position error on Top Right axis exceeded 15mm while running. Error is 33.888mm Counter: 5]
[MSG:WARN: Previous error was 33.888mm]
[MSG:INFO: Program End]
[MSG: /sd/doors.nc file job succeeded]
@dlang Measured the leftover board and here’s how everything deviated from what it was supposed to cut. I broke up the cuts into multiple jobs, and labeled what was cut together in the purple letters, in case that helps at all. The order is also how they were cut. Might have reset the machine after cutting group A (can’t remember) but all the others were done w/o resetting the Maslow.
can you take a straightedge and see which of the edges are curved instead of
straight? I’m guessing that the outside edges aren’t going to be straight.
it’s looking to me like the outside edges are distorted more than the inner
edges of the same parts.
David Lang
I measured away from the corners a bit to avoid any deviations from the straight cuts. I can try to measure the actual cut parts tomorrow if that would be better, but might be difficult to know which of a pair came from where.
even if you do the checking on the frame still on the machine.
I am thinking that the outer edges are going to be significantly more distorted
than the lines slightly further in (the white area of my frame graph)
David Lang
even if you do the checking on the frame still on the machine.
Not sure I follow…
What would you recommend as next steps? All I can think of that I can control is the corner z offset and making sure the frame doesn’t twist. I was considering making custom anchors so they all have the same angle. Pretty sure the sled lifting issues I’ve encountered are largely due to the TL arm having much more angle than the other arms.
Suppose I can optimize that and try to run some sort of grid with a marker, then manually check the spacing. Anything else seems like it would require software fixes.
sorry, for some reason the message got clipped
can you check with a straightege to see if the lines out nearest the edges are
straight or bowed? I’m guessing that they are bowed. I was attempting to say
that you could take the straightedge to the frame that was left on the machine
since the parts may be confused with each other.
If it is possible for you to mark a grid, that would be fantastic.
It may be wishful thinking on my part (partly wanting to be right, mostly hoping
that this is identifying a real problem 
), but I think I’m seeing
confirmation that as you move out of the green area of my frame plot that there
are distortions. If I am correct, the answer is to use a smaller portion of the
work area.
It also looks in your picture like the bottom right belt is a bit looser than
the others. setting the correct Z offsets (including for your workpiece
thickness) will hopefully help that.
We still need to improve the software. A calibration fitness of 0.75 means an
average error of 1.3mm (and I believe that is ± 1.3mm) in the anchor
posistions.
Please save your current maslow.yaml file and then when you do a new calibration
with the correct Z offsets, please save the log data, especially the large array
that shows all the points it’s using for the calibration. I’m hoping that by
examining those, we will be able to spot any other issues.
I’d also like to have you do some careful measurements between the anchors (all
6 distances, including diagonals) so that I can plug them in to the manual
calibration calculator and see what anchor positions it comes up with compared
to the automated calibration.
I’ll send money for materials if needed if you are willing to spend the testing
time.
David Lang
Great write up, I’ll tackle this as I’m able but heading out of town later this week until the new year. Appreciate the offer, but I have a good job and happy to contribute to the project!!
After I give you the data you need, I’m thinking I’ll make a 1/4” awl and a test pattern for the matching to punch in some butcher paper. Seems like the easiest way to repeatedly check the actual accuracy.
Kyle wrote:
After I give you the data you need, I’m thinking I’ll make a 1/4” awl and a
test pattern for the matching to punch in some butcher paper. Seems like the
easiest way to repeatedly check the actual accuracy.
sounds good. No rush on this.
David Lang
