XY skew measurement with an eddy probe and aluminum plate

Oct 18, 2026 Last reply: 5 hours ago 4 Replies
Jump to replies (4)

Once everything is dialed in, this setup lets a printer measure its own XY skew without printing a single test object. Uwe Damm built MEASURE_SKEW, a Python extension for Klipper that scans an aluminum plate with an inductive probe or eddy current sensor, a Cartographer for instance, and then spits out correction values that drop straight into printer.cfg. On the demonstrated Voron 2.4 350 with its CoreXY frame, the macro runs right through the KlipperScreen touchscreen.



The workflow is easy to follow. After homing and quad gantry leveling, Z gets adjusted so the plate fits under the toolhead, the un-machined aluminum plate of roughly 100x100mm goes on the build plate under the probe, and homing in Z happens on top of the plate, preventing collisions. The toolhead scans positions in X and Y and measures edges along both axes. Corners are numbered clockwise 1 to 4 beforehand, and after each pass the plate gets turned 90 degrees counter-clockwise, four cycles in total, with the next numbered corner facing southwest.



The detail that stands out is the fourfold rotation: plate imperfections cancel out mathematically, and an un-machined plate is all the reference object this needs. Nothing gets printed, no filament is wasted, and there is no cooling time between passes. With a plain 100x100mm plate, accuracy lands around 0.2mm/m, claimed to be around twice as precise as older approaches such as Califlower. Corner angles and the skew angle appear in the console once the four passes finish.



There are open ends. Only aluminum has been tried, other materials remain untested, and since only flat plates were used, the effect of plate flatness is hard to judge. Beacon compatibility is assumed but unverified, and documentation is still being improved. Even so, for a 0.35m range printer, the reported statistical error of around 2.0mm/m seems acceptable to me for a tool this convenient.



Project by Uwe Damm. Watch it on their YouTube channel:

formatting link


I am curious how exact the rotation between passes needs to be. Does the plate have to sit square against a fence or reference edge before every scan, or can it simply be turned by hand and the math takes care of the rest? And does corner 1 need to point toward x=0,y=0 precisely, or is a rough placement enough?

Using an un-machined plate strikes me as the smartest part of this. Since the four rotations cancel plate imperfections mathematically, there is no need for precision-ground stock, which keeps the cost of the reference object close to nothing. The trade-off is handling: the plate gets picked up and turned three times, and every pass still needs the edges inside the probe's detection range while the nozzle stays clear of it.

The maker tested a 10mm thick aluminum plate and said thinner stock should also work, so it would be worth trying a thinner sheet on a smaller printer, where a 10mm block may not clear under the toolhead at all. If accuracy holds up, the method would open up to machines with less Z clearance, and the plate would cost even less.

The maker reported that the rotation does not need to be very exact: simulating a 90-degree violation of plus or minus 5mm, about 3 degrees, barely affected the results, which read 1.801mm/m instead of 1.800mm/m. The statistical measurement error was actually larger, around 2.0mm/m instead of 1.8mm/m. Placing the plate within 1-3mm of orthogonal only slightly decreases accuracy.

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required