Accuracy, step 1 Pen

[attachment file=57539]
I am trying to come up with a good pattern for calibration the auto squaring. I will eventually turn this into a page for everyone to follow…I hope. We have arrived at a whole new realm of measurements, and know that 0.1mm is fricken tiny.

Just to keep us all on the same page,https://en.wikipedia.org/wiki/Accuracy_and_precision. Accuracy I will try to use the terms in the wiki context, Accuracy (how faithful to dimensions), Precision (repeatability), Trueness (a combination of the two), and referencing the picture above, X=left to right.

The first problem…the pen mount. The pen mount has a really good amount of spring to it so the people with really uneven surfaces can use it without tearing there paper. It is not accurate enough in the Y direction, to fix that I would have to make a more stiff version. The Pen I am using has the least amount of slop in any pen I have ever seen. Needlepoint and clicked into place it seats itself in the “collet” very well.

Second, measuring two lines on a piece of paper is extremely difficult with any precision.

I can say the machine has stupid good precision. Using as light of pressure possible to avoid divots in the 65lb Cardstock, a 0.3mm ball point pen I can run repeat jobs after moving the gantry and auto squaring so good I can not tell they have been re run. Precision=dead nuts in my book.

Accuracy with the pen mount running repeated tests for me along the x axis has been -0.05mm to -0.1mm over 150mm , and -0.2 to -0.4 over 150mm Y.

So if any of you want to run some tests I would love to see what you get.

The one good thing that has come out of this so far is that, we have a way to tension the belts. Start stupid loose, draw a square and measure it, tighten each belt one click at a time. Stop when the number stops changing for that axis’ corresponding line. It does not have to be very tight to get great results. I kept going tighter and saw no improvements, I also couldn’t mess up the calibration so it seems there is a huge window of just right in terms of tension.

1 Like

Can you make the pen mount bars wider, which would make them stiffer, but mostly in the X/Y and not as much in the Z? Maybe they would have to be thinner as well, to try to make them similar in the Z stiffness…

I have to sidebar and say that this picture makes engineering look very elegant. :slight_smile:

I can try a single wall thickness instead of two and make it twice as wide. It does get mighty finicky at this level of measure though. I ran in to the problem of trying to figure out which axis is skewed. Measuring diagonals only gets you so far, you can’t tell which axis to adjust when it is close. The other test I have to try with my tiny Calipers are different sized squares to see if the same tolerances are there, that would indicate precision is still great and a small test is as good as a large test. The largest diagonal I can measure is about 150mm…Tiny. After that I have to go to a tape measure and that might be horrible, I am about to try them both right now. Maybe through octoprint.

Elegant…that has never been used about anything I have done before. Get a couple fancy lights and all the sudden my snaps are elegant. Score one for Tom telling me to start with lights before worrying about a camera.

[attachment file=57594]
Precision testing the dual endstops. Stupid precise that is three runs on top of each other 35, 55, 65mm/s. Acceleration kicks in above 35 for the corners. No visible difference. The accuracy is about 0 to -0.8mm over 435mm diagonal (measured with a tape) but that was just the test runs, have to try and dial it in better. The pen seems to still have almost 0.3mm slop in the y direction so ±0.3=0.6…hard to say. Need to print that new pen mount.

IMG_20180411_165344.jpg

I would think drawing a circle with a center point would be easier to test for squareness.

Put a dot, draw a circle with a 100mm radius. Any measurement from the center dot to any spot on the circle should be 100mm. If you’re skewed, the larger values should point to the corners you’re skewed to. (basically drawing an ellipse pointing in the direction of skewness)

I think that is the same as measuring a squares diagonals, right? I am seriously not sure any more. I had some friends over yesterday to brainstorm. The numbers are so small it is very difficult to measure. I got the machine to as square as I could with my current pen mount. As soon as a printer gets freed up I will try another design.

It would be the same, but your calipers are good to 150mm.

So a circle of Radius 100mm gives you a nice big 200mm circle. The dot in the center allows you to only have to measure half of that.

I guess you could put a dot in the middle of the square and do it to.

It will be much easier to measure physical objects but I really want to get this dialed in with a pen. I feel like adding a router is just compounding the possible sources of error.

What kind of scenario would the rectangle be off at 30 degrees, but then be right again at 45 degrees?

Since the machine has two major motion dimensions, I think it’s safe to use a rectangle.

What’s interesting is to think of the extreme, if the outside 4 bars were a parallelogram. Totally out of square. If you drew a measurably perfect rectangle, what would that mean? It would mean that the center + gantry was measurably perfectly square, even though the outside wasn’t. That’s exactly what the dual end stops can correct.

What if one side was just longer, like a trapezoid? Say the X axis was parallel, but the Y was bowed out to the +X side. As the gantry moved toward the longer side, the ends would stretch out. Hmmm. What would that do to the drawing? It sort of depends on where the slack is when it’s stretching. Maybe half the error goes to either side, so you would get a similar shape to the outside edge, but maybe less dramatic (by a factor of 2).

So then, you need to measure the corner to corner distances, and also the width of each side. That’s what you’d do if you were gluing up a wooden box. If each side is the same size, and the corner-to-corner distances are the same, and the ends are straight, it’s a perfect rectangle, right?

It would definitely be cool if you could draw a circle (or a square) and rotate it around the center and draw it again in the same place. I don’t know how you would get it to perfectly rotate around the center, but maybe if you got it a little off, it would still tell you something.

You need some kind of highly detailed material, with close to no strength. Like plaster or something. Is there a more atomic foam?

The hard part is figuring out which axis is off, or if they both are. I drew two test patterns yesterday, 4 squares inside of a large square and all the small squares were long on the opposite diagonal of the large one…HUHHHH how does that happen?

This is such a brain bender. I was doing all sorts of geometry yesterday to try and come up with a formula, or pattern. I guess I should google cnc calibration or skew? It is just more fun to use the machine and think about it. I am going to try one more pattern right now and then I will reluctantly go to google.

If there’s hysteresis in your pen mount’s error that can be trouble too.

Were the smaller rectangles drawn in a different rotation?

What happens of you draw one 100mmx100mm rectangle clockwise, and a 98mmx98mm rectangle counter clockwise? Do you see the constant 1mm gap everywhere?

Really, coming up with ways to show errors is a lot easier than telling why it’s off. That’s certainly true. There’s no “ground truth” everything has some kind of error.

Wouldn’t you be able to rotate it with the pen down, holding the center in place? Print dot, draw circle print dot again, stay put. Then rotate/repeat as necessary.

Right now I’m getting a bunch of slop in my center gantry, I’ve reprinted a couple of parts that I never liked in the first place, and will be re-assembling as soon as the time is available. I hope that gets it tight but still freely moving, which seems to be my general issue. Either I can take out the slop, and then there’s too much drag… or it’s sloppy but moves freely.

[attachment file=57688]

Full set of notes, starting points are circled, direction of travel with an arrow. It looks to be all down to the pen/mount combo slop. Direction of travel does seem to have an effect.

The large square seems to be off by ~ +0.5mm and the large circle (opposite direction of travel) is off in the exact opposite corner by about ~ +0.34mm.

New pen mount should be printed and tested in 30 minutes or so.

All I have to say is I feel this is pretty dam good. If this is coming down to what was considered an amazing pen mount doing a freaking crazy good plot two weeks ago to …“dam this mount has 0.3mm slop…” I never thought I would be chasing millimeter decimal places on a machine I build at home!

Agreed.

Well now you have a quantifiable number. Is a bunch +/-0.3mm or like 3mm? I honestly would have no issues with an accuracy of 3mm over a giant diagonal before …not anymore.

This is really making my head spin (in a good way). So interesting.

If the pen was perfect (Where’s your laser?):

In that picture, the circle is longer in the upper-left to lower-right direction than in the lower-left to upper right direction.

So either:

  • The lower endpoint needs to move left
  • The upper endpoint needs to move right
  • The left endpoint needs to move down
  • The right endpoint needs to move up

Any one of those would give you a square center gantry.

But… What about the outer rails, is the gantry parallel with them? If you think of an extreme case, where the center was a perfect ‘+’ and the outside was a perfect square, but it was off from the center by 5 degrees, The square or circle that it drew would still be the right size and the perfectly square, but the drawing would line up with the orientation of the grantry, and not the orientation of the outside frame. You would lose a small amount of travel, but that’s the biggest consequence.

If you move one of those corners and managed to get a measurably perfect square, then you could tell if there was some orientation to the outside error if a purely X motion cause the gantry to move sideways (in the middle) along X pipe.

Me too.

 

Laser is a fat 0.4mm oval shaped dot. I think it might be just as off if not more. I should look that up to verify the number.

 

But the large square is the exact opposite by an even larger amount?

I have used the offset and thin tape to get the rollers distances from the corners but the sources of error in the printed parts surfaces can easily be pretty high. I will test this again, and try to find a consistent surface to measure from that is no the bottom or top of the print.

 

Agreed. The gantry rails need not be perfectly perpendicular to the outer rails I do not think. This also brings up how dam square is my machine. The corners are with 2mm, even less I hope.

I think this is going to boil down to at some point the numbers will flip flop do to frame squareness, rail straightness, print quality. I just want to see how small of a number I can get before that happens. for me anything over 150mm and under +/- 0.35mm (+/- 1/64") would be an insane win and it looks like I am close already. Can’t wait to see how this translates into cut material. With a known machine accuracy seeing what a router, collet, bit, CAM adds will be interesting.