Bed compensation on Delta printers

I will soon start work on adding delta support to the firmware for the Duet electronics. However, I have a question about automatic bed compensation. The Marlin firmware for the Mini Kossel probes the bed in a 7x7 array (omitting points that are outside the physical limits of the bed); then for each move it makes, it adjusts the Z position by linear interpolation between the 4 probe points closest to the XY position concerned.

My question is: why such complicated bed plane compensation? I can see a few reasons why bed compensation may be needed:

1. The bed might not be flat. But this should not be a significant issue if the bed is standard float glass and is not too thin. It's not a significant issue on my Ormerod. So I don't think we should need to provide any compensation in this respect.

2. The XYZ endstops may be set at different heights. To compensate for this, we may have tilt the printing plane in 2 directions. But we need only 3 bed probe points to establish the required tilts, not 49.

3. The ratio of delta radius to diagonal rod length configured in the firmware may be slightly out. In that case, the bed will appear to be slightly convex or concave. But if the bed is probed at 4 points (one close to each tower and one in the centre), then the firmware can work this out and make a small correction to the diagonal rod length.

So it seems to me that only 4 probe points should be needed. Am I missing something? Is the reason for probing more than 4 points to allow for the possibility that the diagonal arms are not quite all the same length?

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Hi,
the printer should print flat if you calibrate it at four points (in front of each tower and in the center) - when it is build accurately and without geometrical errors by calibration Delta Radius and the Endstops (Rod lenght is for the dimension of the parts only)
But if you have tower position errors/twist etc. you have to modify Delta Rod length and Radius for each tower. The Rich Cattell Marlin Firmware can do this and I guess it needs these amount of points to calibrate the Tower Position (Distance between AB, AC, BC)

Manu

add 1) I too believe that glass is planar enough.

add 2) Repetier does this (probes only 3 points). Though adjusting top end stops cannot perfectly compensate for tilted bed. So this is more about compensating a tilted bed.

add 3) It is used primary for this and all the other kinds of calibration errors (e.g. also wrong tower positions, tilted towers).

Thanks to both of you. I've taken a look at Rich Cattel's version of Marlin, and I'll consider using a similar auto calibration method.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Just to throw in a data point here... I've built a mini kossel, my first delta. I started out driving it with a printrboard I had left over from another printer with marlin and Johan's grid bed probing.
I could get decent prints almost immdiately (once some minor stickiness on the probe was sorted out).
Since then I have graduated to a smoothieboard and have been trying manual calibration and also the "discover the geometric errors by probing at the towers and centre" style auto calibration.
I've not yet discovered where the problem lies ( are't deltas difficult to measure! no nicely aligned flat edges to measure distance between the towers for instance) but I can't get a decent print out of it. I've tried manual calibration too, again to no avail.

If I stick the printrboard back on and do the grid probe thing again, bingo, decent prints (but no support for network, LCD, faster printing etc I get from the smoothieboard).

If I change the version of Marlin to Rich Cattels's auto calibrator (which breaks the grid probe compensation) I can't get it to converge reliably or give decent results.

So I think the grid probe basically bypasses a lot of the complex analytical questions regarding exactly how well centered / twisted / measured all the towers and arms are to give "good enough" results. It's more forgiving, but possibly not quite as accurate as doing it "properly".

Cheers,
Robin.

You might also want to check out this thread where there is some good discussion of this sort of thing going on:

[forum.seemecnc.com]