G32 vs G29 – substantial differences in readings ...



  • I'm not sure I understand this, so I'm hoping someone can educate me a little.

    I've been doing some final tuning in a new CoreXY printer, and I'm having dramatically different results from running G32 vs G29. I'm using a BLTouch.

    I've got the bed calibration set up with information about where the four screws are on the bed, and the recommendations for adjustments have seemed to be consistent (meaning, if it says to turn 1 turn down, when I run G32 again, its close…)

    My last G32 was pretty good -- in fact, no amount of adjusting the corners seems to be reducing the errors:

    G32
    Manual corrections required: 0.08 turn up (-0.06mm) 0.02 turn down (0.01mm) 0.06 turn up (-0.04mm) 0.08 turn down (0.06mm)

    But what's weird is immediately after that I ran a G29, and it did its thing (196 points) for a while...

    Number of probe points: 196

    Probe area: 676.0 cm²

    Maximum deviations: -0.129 / 2.218 mm
    Mean error: 0.775 mm

    RMS error: 0.542 mm

    What it shows me is enormously different than what G32 seems to be reporting. 15,15 is coming up at -0.29 (which is the first number in there, so it seems to know (although disagree on how much) that its a little low. What's bizarre though is that it shows the map dramatically skewed. 275,275 shows as 2.218mm high! That's roughly at the 0.06 point where G32 suggests its low. (G32 probed 285,285)

    The mesh is also somewhat twisted, and while at the micron scale its possible a quarter inch thick slab of aluminum is twisted, its showing nearly a millimeter twist along the axis between those two points.

    Anyone have any clues why these two may be reading so differently? Running G32 after the G29 shows nearly the same values:
    Manual corrections required: 0.08 turn up (-0.06mm) 0.02 turn down (0.01mm) 0.06 turn up (-0.04mm) 0.08 turn down (0.06mm)

    So it doesn't look like something literally skewed during the G29.

    I have no idea how to even tell which is actually accurate.



  • Just a thought but do you disable any bed compensation (M561) before probing with either method?

    Can you test the repeatability of your bltouch, this thread https://www.duet3d.com/forum/thread.php?id=1330#p12742 has two simple tests which probe 30 times at a single point consecutively or moving the head around between probes. Be worth seeing if the sensor is accurate enough to be ruled out as the source of error.


  • administrators

    One possible source of confusion is that G32 leadscrew or manual levelling adjustment tries to minimise the errors at the points you specify - which are normally as close to the leadscrew or adjusting screw positions as you can get. So the bed centre could be a long way off. Whereas when you use a Z probe for Z homing, you normally choose to probe at the centre. If the bed is dish- or dome-shaped, or (more likely) the gantry sags slightly when the print head is close to bed centre, then you can get a discrepancy between the height error at the centre and the height error near the leadscrews/bed levelling screws.



  • The bed is a brand new sheet of machined aluminum, it should be perfectly flat. (And a straight edge confirms that). Plus, the height map shows its just skewed, its high in the max-X/max-Y corner, not the middle.

    Like I said, its reading as a bit twisted (which I can see with a straight-edge it isn't), but its generally "flat", just with one corner 2+mm higher than the other.

    Bed.g does disable the bed transform:
    [c]; Clear any bed transform
    M561

    ; Home all axes
    G28

    ; Probe the bed at 4 points
    G30 P0 X15 Y15 H0 Z-99999
    G30 P1 X15 Y285 H0 Z-99999
    G30 P2 X285 Y285 H0 Z-99999
    G30 P3 X285 Y15 H0 Z-99999 S[/c]

    I don't see a comparable file for G29, so I don't know what it is or isn't doing.

    I'll have to look at the spread of readings from the BLTouch, but the fact that G32 reads pretty consistently suggests it can't be much more than a dozen or two microns of inconsistency. Certainly not as much as 50, and definitely not 2mm! I can't find anything specific, though – is there a GCode that'll probe a spot and spit out the deviation there? G32 doesn't spit that out, it only spits out the adjustments. I can't find anything that does that. People in the thread linked are doing it, so I'm sure I'm just not spotting it... 🙂



  • Yeah use G30 S-1

    That will spit out the height at that point.

    There is a gcode in the thread I linked to just download and run it, quickly look at the gcode and just check the coordinate it will probe at is reachable.



  • Ah I missed those links, they got truncated and I didn't even see they were gcode. I'll run them later and see what I see. I'll be surprised if there's much variance, though, given how consistent G32 is. And a reading inaccuracy wouldn't be consistent and even across a 300mm print surface like that.



  • No I suspect it won't be, but it needs to be eliminated. Also above I think you missed what David was saying about flatness. Its not your tooling plate that isn't flat, depending on the printer its the x-axis itself that sags. 8mm rods especially can sag a good 0.2-0.4mm in the middle, more with a heavier printhead attached.

    I saw the effect of this directly, my corexy was twin vertical 8mm rods and I switched it to linear rail, the "hill" in the middle of the bed disappeared.



  • So I ran both of those scripts. The one that moves the head around results were:

    [c]Bed probe heights: -0.010 0.006 0.006 0.007 0.010 0.034 0.054 0.067 0.064 0.068 0.070 0.065 0.071 0.074 0.074 0.084 0.082 0.078 0.074 0.081 0.080 0.081 0.078 0.076 0.076, mean 0.058, deviation from mean 0.029[/c]

    Definitely some variance, but pretty small.

    The one that did 36 tests in the same point failed until I reduced the tests to 30, not sure why. Its results:
    [c]Bed probe heights: 0.058 0.072 0.057 0.067 0.055 0.060 0.060 0.072 0.066 0.057 0.056 0.047 0.065 0.074 0.054 0.052 0.050 0.064 0.060 0.051 0.050 0.049 0.067 0.045 0.047 0.049 0.070 0.053 0.075 0.047 0.047 0.047, mean 0.058, deviation from mean 0.009
    [/c]

    Considering BLTouch claims 0.01 deviation from the mean, at least without moving the head around they seem to meet it.

    I guess what it boils down to is if I'm seeing different results from G32 and G29, does it matter? Its not super obvious how the output of one relates to the other, its just weird that G32 suggests such smaller adjustments that G29 suggests are needed.



  • @DjDemonD:

    No I suspect it won't be, but it needs to be eliminated. Also above I think you missed what David was saying about flatness. Its not your tooling plate that isn't flat, depending on the printer its the x-axis itself that sags. 8mm rods especially can sag a good 0.2-0.4mm in the middle, more with a heavier printhead attached.

    I saw the effect of this directly, my corexy was twin vertical 8mm rods and I switched it to linear rail, the "hill" in the middle of the bed disappeared.

    I guess I'm confused, how does your Z rods impact the sagging of your X-axis? In this case, the printer is a D-bot inspired, so the X gantry is 2040 extrusion. I doubt its sagging.

    Rather than screwing with the two calibrations back and forth, I've been tweaking focused just on the G29 calibration. For whatever reason, its starting to flatten out. Its still showing me a .2mm sag in the middle of the bed, though, which I'm just not understanding. Its consistent enough to make the doubt the aluminum plate or the PEI on top of it. I didn't double check the extrusions before assembling the printer, but its hard to imagine they're not straight…



  • So after the last round of tweaks, G29 seems to be leveling out:

    I think I'm going to just stick with what its telling me, as opposed to G32 until I see some indication from prints that its not correct…



  • I have a aluminium heat bed painted black and with a 6mm sheet of glass on top of it so it is 100% flat.
    just did a mesh grid using the BLTouch and another with the DC42 IR Probe. can anyone help with it?why both of then are giving bad results?
    BLTouch

    DC42 IR Probe


Locked
 

Looks like your connection to Duet3D was lost, please wait while we try to reconnect.