Duet Auto-Calibration results - More Data Needed!

[nebbian]

@DC42
could a function be added to the firmware to allow individual movement of each tower?
Then we could fasten a dial indicator to the carriage and command it to move say 25mm then measure the actual movement to determine your actual steps per mm.
Likewise, measuring each tower individually could help getting belt tensions all the same. That is if the belt tension really does change steps per mm, which personally i think it can.

You don't need a special function -- just jog down 25mm using your favourite computer interface. All three carriages move, but that doesn't matter for the purposes of this test. There's no delta calculations when moving straight down.

[DerStig]

Well changed my steps per mm to 80.64 and here are my results

G32
Bed probe heights: -0.031 0.009 0.002 0.019 -0.012 0.014 0.031 0.051 -0.031 0.075 0.062 -0.025 -0.095 -0.050 0.040 0.027, mean 0.005, deviation from mean 0.043
M666
Endstop adjustments X-0.05 Y0.02 Z0.03, tilt X0.00% Y0.00%
M665
Diagonal 300.000, delta radius 133.021, homed height 356.050, bed radius 150.0, X -0.007°, Y 0.218°, Z 0.000°

[bvandiepenbos]

@DC42
could a function be added to the firmware to allow individual movement of each tower?
Then we could fasten a dial indicator to the carriage and command it to move say 25mm then measure the actual movement to determine your actual steps per mm.
Likewise, measuring each tower individually could help getting belt tensions all the same. That is if the belt tension really does change steps per mm, which personally i think it can.

You don't need a special function -- just jog down 25mm using your favourite computer interface. All three carriages move, but that doesn't matter for the purposes of this test. There's no delta calculations when moving straight down.

I think you do need to directly drive just one motor.
Wouldn't delta kinematics be in effect even when doing a Z only move?
But, that gives me an idea... measure movement of each carriage individually with same z down move, ideally they should all move exactly the same amount

[Xenocrates]

Pretty sure RRfirmware already has that capability, using the S1 and S2 parameters in a G01. See http://reprap.org/wiki/G-code#G0_.26_G1:_Move

[dc42]

@DC42
could a function be added to the firmware to allow individual movement of each tower?
Then we could fasten a dial indicator to the carriage and command it to move say 25mm then measure the actual movement to determine your actual steps per mm.
Likewise, measuring each tower individually could help getting belt tensions all the same. That is if the belt tension really does change steps per mm, which personally i think it can.

I think you mean individual movement of each carriage. The firmware already supports that, see https://duet3d.com/wiki/G-code#G0_.26_G1:_Move.

[kraegar]

I haven't started delving into the belts being the source yet, but I have tried another option for arms. This time I used magball end CF arms. 304.22mm in length (per the manufacturer). I'm only using printed adapters to attach them to the carriages & effector. The same 50mm spacing is in place that was with the stock ball ends. I also have the new aluminum carriages on from Tricklaser. This is with the Mini IR probe:

Diagonal 304.220, delta radius 133.898, homed height 350.899, bed radius 145.0, X 0.261°, Y 0.709°, Z 0.000°
Endstop adjustments X0.28 Y0.13 Z-0.41, tilt X0.00% Y0.00%

Bed probe heights: 0.027 -0.041 0.038 -0.025 0.064 -0.042 0.047 0.041 -0.029 -0.028, mean 0.005, deviation from mean 0.039

The swap to the magball arms greatly reduced the deviation overall for me, but the up/down pattern still remains. Since these arms are longer than either the Tricklaser (300mm) or stock (292mm) it does rule out the idea that longer arms exaggerates the issue. I can rapidly switch between any of the 3 arm types, and the results are very consistent for each.

I think next I want to come up with a way to affix calipers to my tower, and try moving a carriage 50mm to measure any difference.

[DeltaCon]

Here some extra results:

a. The results of M665
Diagonal 293.637, delta radius 144.328, homed height 357.719, bed radius 140.0, X -0.006°, Y -0.017°, Z 0.000°

b. The results of M666
Endstop adjustments X0.14 Y-0.16 Z0.02, tilt X0.00% Y0.00%

c. The results of the S-1 Auto-Cal
Bed probe heights: 0.027 -0.058 0.089 0.003 -0.056 0.070 0.003 -0.074 0.068 -0.011 0.001 0.026 -0.019, mean 0.005, deviation from mean 0.049

d. How many points of calibration you usually use, and how many points (IE, 10 point, 6 factor)
13 points, 6-factor

e. build info:
Machine : RMv2,
Arms : New style original SeeMeCNC ball-cup arms
Carriages : SeeMeCNC injection molded skates
Effector : Original SeeMeCNC that gets deliverd with the arms, only with the ball joints on top instead on bottom
Probe Type : FSR, on Duet 0.6
HotEnd : E3D v6, with a mount printed in ABS (could perhaps have a little more flex in it?)

[nebbian]

Well changed my steps per mm to 80.64 and here are my results

G32
Bed probe heights: -0.031 0.009 0.002 0.019 -0.012 0.014 0.031 0.051 -0.031 0.075 0.062 -0.025 -0.095 -0.050 0.040 0.027, mean 0.005, deviation from mean 0.043
M666
Endstop adjustments X-0.05 Y0.02 Z0.03, tilt X0.00% Y0.00%
M665
Diagonal 300.000, delta radius 133.021, homed height 356.050, bed radius 150.0, X -0.007°, Y 0.218°, Z 0.000°

Looks like you're on the right track. Your deviation has improved from 0.063 to 0.043 (what's that, something like 30% better?)

[kraegar]

So, I kind of went a little overboard. I had adapters for the magball ends, but had tilt I couldn't easily find/eliminate. So I went back to my stock arms, modelled and printed a new effector just for the magball ends. I carefully measured every point on it after printing, and made sure the carriage connectors matched to < .1mm for the spacing between the magball ends to the same connectors on the effector (all spacing is 50mm). The magballs are also all at the same height, giving me a parallel plane to the bed. Did a 10 point calibration, with height offsets for each point.

http://imgur.com/n6vQsYq

I've attached that, and now calibrate at a 0.15 deviation, but still show thee +/-/+/- pattern.

So, the springs seem to make things worse, but are not the only cause. Now I'm working on accurately measuring the steps/mm by moving my carriages 50mm and measuring the actual movement - does it make any difference if I just move down 50mm in Z, or do I need to individually move each carriage to measure this? Do I need to clear any of my settings (tower steps offset? etc) first?

Currently, starting at Z75, If I do a G1 Z25 I'm measuring ~49.9mm actually moved. I need to improve the method of measuring, but I'm positive it's moving < 50mm. I need to get it so I can accurately measure a 100mm move, to get a better measurement difference, but have to find a way to statically mount my calipers... that's still removable.

[bvandiepenbos]

So your deviation has approximately doubled with your newest printed parts setup?
Or was that a mis-placed decimal point?

[kraegar]

Sorry, yes, misplaced decimal point!

Calibrated 6 factors using 10 points, deviation before 0.016 after 0.015

That's from today.

Adjusting the steps/mm I can lower that further.

[bvandiepenbos]

Have you measured the steel balls?
Are they truly spherical and all the same size?
All of the 4 sets of steel ball-studs I got in where somewhat egg shaped and varied in size more than I would like.
So far I have found the white molded plastic barbells made by seemecnc have been better and more consistent than the steel ball-studs.

If your magnetic arm cups are pivoting on a egg that probably would cause weird movements.


.016 mm is really good.
Even a deviation around .05 mm is not bad for these machines, that in inches is...
.002" ....a piece of paper is .003-.004"
The glass is probably not even that flat and HPB moves around also.
I'm all for precision and pushing the limits, but don't get to freaked out by your results.

The belt drive is likely the least consistent and precise part of the whole machine. Getting all 3 belts the same tightness and measuring the actual movement would go along ways in getting machine to be as precise as possible.

[kraegar]

I used a digital micrometer to measure the steel balls the best I could, and all were spherical.

I think the next step will be improving the way I'm measuring the carriage movement, so I have a repeatable and accurate test at a movement distance of 100mm. That should be enough movement to give me an accurate indication of how far off the steps/mm actually are, and correct them for each tower.

It will also show differences well enough I should be able to use it as a way to measure the belt tension at each tower and get them equal.

[kraegar]

Well, back to the drawing board.

X tower: 99.98 - 100.01mm measured, for a 100mm move
Y Tower: 99.85 - 99.90mm measured, for a 100mm move
Z Tower: 100.09 - 100.10mm measured, for a 100mm move

All measurements repeated at least 5 times.

Even at the worst, that puts steps/mm at 80.1. I need to get around 80.7 to cancel out the up/down motion in the calibration.

Putting in a longer arm length can also cancel it, but then putting in a longer arm length causes dimensional issues in prints.

[martins]

little bit outdated thread i put some my findings
stock rostock max v3 + duet wifi +teensy to run probe on H280


G32
Calibrated 6 factors using 13 points, deviation before 0.089 after 0.089
M665
Diagonal 291.600, delta radius 144.247, homed height 405.333, bed radius 140.0, X -0.001°, Y -0.123°, Z 0.000°
M666
Endstop adjustments X0.22 Y0.15 Z-0.37, tilt X0.00% Y0.00%

after callibrated with 7 factors

G32
Calibrated 7 factors using 13 points, deviation before 0.026 after 0.026
M665
Diagonal 295.304, delta radius 145.403, homed height 405.547, bed radius 140.0, X -0.010°, Y -0.162°, Z 0.000°
M666
Endstop adjustments X0.16 Y0.13 Z-0.29, tilt X0.00% Y0.00%

[geneb]

You want to avoid the 7 factor calibration. It fiddles with the arm length and that can mess with your printed part scaling.

g.

[martins]

true thx
goes back to 6factor as the print was to small 0.5mm

[clytle374]

Sorry to bring this thread back from the dead, but having serious problems with calibration. I know my printer has been out for a while after upgrading the firmware in a failed y adapter install.(never was reliable). I don't know how bad as I have only printed small parts in the center of the bed.. I just installed a duet WiFi on my Max v2. I put a 1" dial indicator in place of the hot end and got very bad results. Especially in the y- part of the bed. I ran through the least squared calibration several times and would move the bad spots around the bed.

I haven't had a chance to work on it for a few days. But I found this post about an hour before I gave up last time and changing steps per mm from 80 to 80.6 was an immediate and drastic improvement. I have a few days off to work on it and thought I'd ask if this issue was ever really sorted out, or even understood. Been thinking about this for a few days now and want to get this printer working right finally

Thanks in advance
Cory

[clytle374]

https://youtu.be/dRv_20ABC_g I am pretty sure there must be a error in the duet firmware. I don't have a probe so all the calibration was done with an indicator and the least squared calculator on the website. I have been through it countless times trying to get it correct without changing the arm length, and fiddling with the bed tilt to get it as close as possible. My steps per mm is set to 80.3 as that seemed to produce correct travel distances. Here is my Delta numbers.

Code: Select all

 M665 R144.80 L293.40 H367.25 B140.00 X-0.20 Y0.51 Z0.00		
M666 X0.50 Y0.03 Z-0.53 A-.12 B-.38 

I have to admit that even after levelling and squaring up countless machine tools, this was the worst, by a large margin.

I haven't had a chance to print and check the dimensional accuracy yet. But after a lot of thought, there can not be 2 solutions to get the results I'm getting. This is a 1 inch dial indicator, so the divisions are .001". No bed mapping, just calibration.

Cory