Duet Auto-Calibration results - More Data Needed!

[kraegar]

I'm trying to gather some info to compare data and figure out an issue, and I'd like to see if I can get more data to shed some light on things. If you have a duet on a rostock max, and are willing to help, please do the following!

1 - Fire up your printer, and auto-calibrate as usual, doing as many rounds as you normally do to get dialed in.
2 - Edit your bed.g, and on the last G30 line, replace your S6 or whatever factor of calibration you use with S-1, and run one more auto-calibration
3 - Reply to this post with the following:

a. The results of M665
b. The results of M666
c. The results of the S-1 Auto-Cal
d. How many points of calibration you usually use, and how many points (IE, 10 point, 6 factor)
e. build info:
Machine (RMv2, v3, etc)
Arms (ball end, CF tricklaser ball end, CF tricklaser traxxas, etc)
Carriages (Wood, injection molded, tricklaser, etc)
Effector
Probe Type
Any other major modifications that could affect the geometry between the carriages & effector.

[kraegar]

a. Diagonal 300.150, delta radius 143.921, homed height 373.207, bed radius 150.0, X -0.095°, Y 0.482°, Z 0.000°
b. Endstop adjustments X-0.61 Y1.57 Z-0.96, tilt X0.00% Y0.00%
c. Bed Probe Heights: 0.193 -0.246 0.175 -0.208 0.278 -0.206 0.024 0.082 -0.108 0.010 -0.001 0.176, mean -0.001, deviation from mean 0.176
d. 10 point, 6 factor
e.
Rostock Max V2
Arms - Trick Laster CF ball end, 300.15mm
SeeMeCNC injection molded carriages
713 Maker Aluminum Effector
dc42 Mini IR Probe

[kraegar]

a. Diagonal 291.060, delta radius 143.464, homed height 385.225, bed radius 150.0, X 0.520°, Y 0.992°, Z 0.000°
b.Endstop adjustments X-0.07 Y1.05 Z-0.98, tilt X-0.38% Y-0.04%
c. Bed probe heights: 0.125 -0.114 0.128 -0.136 0.102 -0.091 0.053 -0.016 0.034 -0.006 0.008 -0.007 -0.056, mean 0.002, deviation from mean 0.083
d. 13 point, 8 factor
e.
Rostock Max V2
Arms - SeeMeCNC injection olded arms
SeeMeCNC injection molded carriages
713 Maker Aluminum Effector
HE280 Accelerometer probe

[shodapublicity]

a. Diagonal 290.00, delta radius 143.06, homed height 385.11, bed radius 140.0, X -0.60°, Y -0.25°, Z 0.00°
b. Endstop adjustments X0.01 Y0.11 Z-0.11
c. Bed probe heights: 0.154 -0.113 0.101 -0.149 0.100 -0.108 -0.073 -0.013 0.090 0.067 0.015 -0.058 -0.032, mean -0.002, deviation from mean 0.093
d. 13 point, 6 factor
e.
Rostock Max V2
Arms - SeeMeCNC Stock arms (The new ball cup arms)
SeeMeCNC injection molded carriages
SeeMeCNC Injection Molded Effector Plate
713Maker E3D V6 Standard Mount (2 washers on the three mounting points or else peek fan hits effector plate)
JohnSL FSR Endstop Controller (bought from Ultimaker)

[kraegar]

Thanks shodapublicity. I have results from two other users, and a couple more configurations as well. I could really use more data, if anyone else is able to contribute!

[kraegar]

So, the pattern I have, and see in all the other data contributed to me so far is this.

After a full round of calibration, if you re-run the calibration script changing your S6 / S8 to an S-1, there is a wave pattern to the results. It measures higher at the towers, and lower opposite them. What I'd like to find is why - is there something mechanical going on that's causing this? Is it a particular part that is behind the result? Is it just an anomoly that is only happening to some of us?

I have 6 results I've seen now where this is the case. I have swapped out arms and effectors on my own printer, and it still shows. It seems worse the longer your arms. (I used both the injection molded arms, and a set of Carbon Fiber arms that are longer)

I see the same results using both the mini IR probe and the accelerometer probe.

I plan to try different carriages soon, but I don't expect to see a change in this pattern from that. Just trying to narrow down or rule out any variables I can. Again, if others can contribute data, it would help to establish or eliminate the pattern.

[DeltaCon]

Could these high points be at mount points? I think thoise differ from printer to printer. I have FSRs in between the towers. I think the resting points for the V3 bed are also at the towers. The V2 style snowflake has 6 mountpoints, so maybe there is an other pattern visible? I will try to post data this weekend.

[kraegar]

I have a v2 bed, the high points don't correspond to the mount points of the bed. I've also experimented with moving my binder clips around, and even added a aluminum spacer between the bed and the glass, no change.

[kraegar]

Oh, and I can also confirm I've seen data for both v2's & v3's that show the same result.

[DerStig]

Hey here are my numbers

Diagonal 300.000, delta radius 133.470, homed height 356.904, bed radius 150.0, X -0.029°, Y 0.258°, Z 0.000°
Endstop adjustments X0.36 Y-0.05 Z-0.31, tilt X0.00% Y0.00%

RepRapFirmware height map file v1, mean error -0.01, deviation 0.05
xmin,xmax,ymin,ymax,radius,spacing,xnum,ynum
-105.00,105.10,-105.00,105.10,110.00,15.00,15,15
0, 0, 0, 0, 0, -0.098, -0.113, -0.095, -0.076, -0.025, 0, 0, 0, 0, 0
0, 0, 0, -0.062, -0.111, -0.135, -0.099, -0.127, -0.111, -0.045, -0.022, 0.012, 0, 0, 0
0, 0, -0.072, -0.104, -0.099, -0.077, -0.099, -0.074, -0.104, -0.036, -0.012, 0.032, 0.041, 0, 0
0, -0.066, -0.037, -0.073, -0.090, -0.097, -0.114, -0.106, -0.061, -0.076, -0.076, -0.022, 0.015, -0.003, 0
0, -0.054, -0.072, -0.039, -0.049, -0.054, -0.074, -0.037, -0.064, -0.024, -0.021, -0.011, 0.033, 0.023, 0
0.027, -0.008, -0.014, -0.048, -0.039, 0.000, -0.036, -0.071, -0.051, -0.076, -0.023, -0.015, -0.011, -0.053, 0.030
-0.001, -0.016, -0.036, 0.003, -0.039, -0.039, 0.001, -0.021, -0.026, 0.010, -0.034, -0.039, 0.009, 0.000, -0.035
0.011, 0.027, 0.011, 0.014, -0.004, 0.029, 0.027, -0.011, -0.012, -0.022, -0.012, -0.002, -0.018, -0.041, -0.072
0.003, 0.024, -0.011, 0.055, 0.050, 0.047, 0.040, 0.048, 0.001, 0.034, 0.047, 0.027, 0.010, -0.034, -0.078
-0.009, 0.051, 0.052, 0.052, 0.051, 0.009, 0.036, 0.060, 0.072, 0.075, 0.011, 0.036, 0.005, -0.046, -0.106
0, 0.003, 0.052, 0.051, 0.036, 0.078, 0.086, 0.039, 0.060, 0.075, 0.033, 0.028, 0.003, -0.042, 0
0, 0.013, -0.017, 0.055, 0.037, 0.037, 0.047, 0.048, 0.034, 0.024, 0.075, 0.017, 0.026, -0.013, 0
0, 0, -0.040, -0.000, 0.015, 0.073, 0.087, 0.052, 0.060, 0.049, 0.063, 0.034, -0.012, 0, 0
0, 0, 0, -0.047, -0.014, 0.037, 0.020, 0.078, 0.059, 0.036, 0.048, 0.018, 0, 0, 0
0, 0, 0, 0, 0, 0.026, 0.017, 0.066, 0.043, 0.064, 0, 0, 0, 0, 0


Rostock Max V2
Trick Laser 300mm Carbon arms with Traxxas ball ends
Trick Trucks
Trick Laser Effector plate (Old Style) 713 Accelerometer mount E3Dv6 HE
HE280 Accelerometer

[kraegar]

Instead of the heightmap, can you run your bed.g auto-cal, but change the S6 to an S-1?

I've been hoping someone with the traxxas rod ends would reply, I'd be interested in seeing a picture of the arms you have on, and if you have springs between them.

[kraegar]

Also, if anyone has mag ball ends, PLEASE test and post your results!

[DerStig]

G32
Bed probe heights: -0.659 -0.713 -0.809 -0.720 -0.801 -0.837 -0.759 -0.749 -0.775 -0.670 -0.688 -0.796 -0.898 -0.821 -0.725 -0.737, mean -0.760, deviation from mean 0.063
PMM666
Endstop adjustments X0.36 Y-0.05 Z-0.31, tilt X0.00% Y0.00%
PMM665
Diagonal 300.000, delta radius 133.470, homed height 356.904, bed radius 150.0, X -0.029°, Y 0.258°, Z 0.000°

Rostock Max V2
Trick Laser 300mm Carbon arms with Traxxas ball ends
Trick Trucks
Trick Laser Effector plate (Old Style) 713 Accelerometer mount E3Dv6 HE
HE280 Accelerometer

[kraegar]

Derstig:
You have these arms? http://tricklaser.com/300-MM-Carbon-Fib ... FTX300.htm

Do you have any sort of spring or strap between each pair to limit backlash?

[DerStig]

Derstig:
You have these arms? http://tricklaser.com/300-MM-Carbon-Fib ... FTX300.htm

Do you have any sort of spring or strap between each pair to limit backlash?

Yes those arms and no springs and I just realized that those numbers include my Z height offsets

[kraegar]

Thanks for the data points, very helpful!

[nebbian]

So, the pattern I have, and see in all the other data contributed to me so far is this.

After a full round of calibration, if you re-run the calibration script changing your S6 / S8 to an S-1, there is a wave pattern to the results. It measures higher at the towers, and lower opposite them. What I'd like to find is why - is there something mechanical going on that's causing this? Is it a particular part that is behind the result? Is it just an anomoly that is only happening to some of us?

I'll put $5 on it being the steps per mm being wrong. Increase your steps per mm by 0.8% and the wave will be gone, providing that your diagonal rod measurement in firmware matches what's actually on your printer. GT2 belt isn't exactly 2mm between teeth unless you stretch the living daylights out of it.

I've verified this many different ways, it would be nice for someone else to try it as well

[kraegar]

I have a (third) set of arms on the way. Once I've tested with them, I can see how adjusting the steps/mm changes things if I'm still seeing error.

Steps/mm is indeed the other variable we've been debating about.

[DerStig]

So, the pattern I have, and see in all the other data contributed to me so far is this.

After a full round of calibration, if you re-run the calibration script changing your S6 / S8 to an S-1, there is a wave pattern to the results. It measures higher at the towers, and lower opposite them. What I'd like to find is why - is there something mechanical going on that's causing this? Is it a particular part that is behind the result? Is it just an anomoly that is only happening to some of us?

I'll put $5 on it being the steps per mm being wrong. Increase your steps per mm by 0.8% and the wave will be gone, providing that your diagonal rod measurement in firmware matches what's actually on your printer. GT2 belt isn't exactly 2mm between teeth unless you stretch the living daylights out of it.

I've verified this many different ways, it would be nice for someone else to try it as well

So you're suggestion running 86.4 steps per mm in firmware?

[mhackney]

I've been collaborating with kraeger in the background on this for a few weeks now, sharing data, etc. I've also had conversations with David Crocker and he validated that the pattern we see is most likely an arm length or steps/mm (or both) issue. That's what got us looking at the different arm geometries and connection types (Traxxas, mag balls, cup balls, etc) and arm types to see if there is a consistent pattern.

The thing I/we need to sort out is the inter-dependency of arm length and steps/mm and the actual print size. What I mean is, the "common" way to fix X-Y scaling issues on a delta is to adjust the arm length in firmware. I have always thought (based on performing the delta kinematics calculations by hand, etc) that - within a reasonable window (say 10% of actual length) that arm length adjustment ONLY affected the X-Y part scale. I did not appreciate that it also contributes to non-planarity of effector movement in the way kraeger described in the first post. David set me straight on that a few weeks ago and added that steps/mm for the carriage movement also affects movement and linearity in X-Y. David suggested printing long lines from tower base to the opposite side using several different arm length configurations. You will see these printed lines begin to curve. kraeger has been using a printable hexagon (with points aligned to the tower bases and mid-point between towers) to see this conveniently on all 3 axes.

Ideally, you want to physically measure one or both of these parameters (arm length and steps/mm) to use in the configuration so they do not have to be calculated. In theory that should be possible as arms can be measured (and manufactured to specific pivot-to-pivot point lengths) and the steps/mm can be calculated using belt pitch, steps/rotation and pulley teeth number. That's how the original arm length and steps/mm values were determined for the RMax firmware. But with these measured actual values, many of us are seeing this odd movement error in various degrees. In retrospect, this might make sense and be more closely tied to steps/mm due to factors like belt stretch (we don't have a consistent and reproducible way to tension belts from machine to machine for instance), actual tooth spacing (as a function of either stretch or manufacturing) or other factors.

Now that we understand the cause and effect more clearly, the next step is to design the tests to vary these independently and then together - along with verification of the print size accuracy - to collect the data needed to understand and configure the firmware properly. I've started to work on that but I've been traveling for work this and last week.

In retrospect, it makes a lot of sense. I built 2 CNC mills that have a beefy belt to drive the Y axis. Even though I calculated the steps/mm for the Y movement, I also actually calibrate my mills by measuring the actual number of steps required to travel a target distance. And these belt driven Y axes always need to be tweaked a little as compared to the direct screw driven X axis.

[mhackney]

So, the pattern I have, and see in all the other data contributed to me so far is this.

After a full round of calibration, if you re-run the calibration script changing your S6 / S8 to an S-1, there is a wave pattern to the results. It measures higher at the towers, and lower opposite them. What I'd like to find is why - is there something mechanical going on that's causing this? Is it a particular part that is behind the result? Is it just an anomoly that is only happening to some of us?

I'll put $5 on it being the steps per mm being wrong. Increase your steps per mm by 0.8% and the wave will be gone, providing that your diagonal rod measurement in firmware matches what's actually on your printer. GT2 belt isn't exactly 2mm between teeth unless you stretch the living daylights out of it.

I've verified this many different ways, it would be nice for someone else to try it as well

Nebbian, see my reply above. I believe, and this needs to be validated, that there is more variation in steps/mm (due to belt tension, etc) than the manufactured arm length from printer to printer and that the fudge factor to compensate the steps/mm will be slightly different and unique for each machine. So a sweeping "increase steps/mm by 0.8%" will likely not give the best results. Let's do the tests, collect the data and figure out what's really going on.

[dc42]

There's a simple test you can do to see whether it is just a steps/mm issue. After running 6-factor calibration, print 2 parallel lines gong from top to bottom of the bed, about half a bed width either side of the centre. Check them for straightness against a straight edge such as a metal ruler.

Then run 7-factor calibration so that the diagonal rod length is adjusted to remove wave pattern. Print those two lines again, and once again check them for straightness.

If the product of the steps/mm and diagonal rod length used by the firmware matches the correct values for your printer, the lines will be straight (this assumes that the delta radius is correctly calibrated too). So if the second set of lines in the above is straight and the first isn't, the problem is incorrect steps/mm. Print a test cube, measure it, then scale your steps/mm by the factor needed to get the correct dimensions. Then run 7-factor calibration again.

OTOH if the first set of lines is straight and the second one isn't, then incorrect steps/mm is not the issue. If both sets of lines bend but in opposite directions, then incorrect steps/mm is partly to blame assuming you configured the diagonal rod length correctly.

[nebbian]

Nebbian, see my reply above. I believe, and this needs to be validated, that there is more variation in steps/mm (due to belt tension, etc) than the manufactured arm length from printer to printer and that the fudge factor to compensate the steps/mm will be slightly different and unique for each machine. So a sweeping "increase steps/mm by 0.8%" will likely not give the best results. Let's do the tests, collect the data and figure out what's really going on.

Agreed that it will be different for each machine. However I have calibrated this with three different batches of belts, on two machines, and the final value was always around 0.8% higher than calculated. I don't know of anyone else that calibrates their steps/mm, hence my request for more people to try adjusting this value instead of blindly trusting the manufacturer.

I calculated that moving the tensioner about 0.9mm results in a steps/mm change of 0.1% on my mini. So obviously if you change the belt tension, you'll change the steps/mm value. To get the correct 2mm/tooth you'd have to move the tensioner 7.2mm -- this would make the belt unacceptably tight for me, but maybe that's the tension that they manufacture the belts for.

All I'm really saying is that steps/mm is something else that needs to be measured and accounted for if you want perfect dimensions on your print. I've found 0.8% higher to be a very good starting point.

[nebbian]

So you're suggestion running 86.4 steps per mm in firmware?

Assuming you're running 80 steps per mm now, I suggest 80.64 steps/mm. As mentioned above, this value still needs tuning but I've found this to be a good starting point.

[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.