Advanced Delta Printer Calibration Question

Question

Specs: Printer Configuration: Delta Firmware: Marlin 2.x Probe: Conductive nozzle (Verified repeatable 0.01 mm accuracy)

I'm working on calibrating my delta printer to 0.01 mm +/- 0.02 accuracy. Every guide I see either says "run G33 that should be good enough" (news flash it isn't lol) or they reference this Advanced Delta Calibration tool.

Which relies on your filament/nozzle diameter/flow rate/temperature to all be exactly precise (which isn't physically possible if we're talking about differences small enough to affect dimension calculations in some way).

I'm essentially looking for a precision tuning guide for a "delta robot with a probe" rather than one for a "delta 3D printer". Does a guide like this exist?

If not, maybe you guys have calculations that may help. Right now I'm working on tower angles and individual rod offsets using this script.

It probes towers and opposite towers in the order of: Z-ZY-Y-YX-X-XZ-Center.

With a loosely leveled bed (+/- 0.05 mm @ towers) and a G33 P2 I get probe values:

Z: -0.08

ZY: -0.21

Y: -0.01

YX 0.23

X: 0.02

XZ: -0.66

C: 0.01

Which is essentially a "tilted pringle" shape which can be visualized here with this input.

Just wondering what to do about inaccuracies in the diagonals. Is there a formula for tower angle/rod length based on probe points? And if not is there a more accurate way of doing this that doesn't involve printing something?

Answer 1

I designed and built both a full Stewart platform, 6-dof router, and a "delta" style, constrained Stewart platform 3D printer with 3-dof.

For the 6-dof machine, I postulated a set of error variables and used simulated annealing to find a good estimate of the error values. Those error values were inputs to the kinematics and gave much better results.

For the 3-dof machine, I found it difficult to calibrate love the whole 18" diameter print bed, and decided that a real fit may not be worth the trouble. Instead, I mapped the surface using a probe, and fit an error function to that data.

After trying a few methods of measuring the bed surface, I settled on using a strain gauge to measure nozzle contact. I had tried electrical contact, IR parallax sensors, and inductive sensors, but all had unfortunate ways to give bad measurements. A strain gauge worked better, although it, too, is subject to problems if there is debris on the bed, or plastic that flows through the nozzle.

When I make my next set of changes to the 3-dof delta 3D printer, given the superior sensing of the strain gauge, I may go back to a scheme of estimating error variables.

There are a lot of potential errors, all of which affect a delta machine's ability to know where the Z=0 cartesian plane lies. Any positioning error of the joints, any error in the spacing between joints, any racking of the printer structure, and any non-orthogonal-to-the-bed movements of the top joints will give you various compound cup and disk errors. Even the uncertainty of the home sensors adds an expected curved error to what should be the Z=0 cartesian plane.

My experience has been that the most important thing in 3D printing -- knowing exactly where the bed is and maintaining a uniform height for the first layer -- is the hardest thing to get right with a delta machine.