Under-bed load cell probe on a CoreXY with 2 leadscrews: 4 × 5 kg cells, HX711, Octopus Pro (guide + configs)

Hi all,

I’ve finished an under-bed load cell Z probe on my DIY CoreXY using Klipper’s built-in [load_cell_probe], and wrote the whole build up as a step-by-step guide with ready-to-use config files, so others can reproduce it.

Setup

  • DIY CoreXY, enclosed, 2 Z leadscrews (left and right), 350 × 380 mm glass bed with a 220 V heater
  • 4 × 5 kg bar load cells in parallel → 1 × HX711 (80 Hz mod, powered at 3.3 V)
  • BTT Octopus Pro V1.1, Klipper v0.13
  • trigger_force: 75, probe speed: 1.5

Results

  • PROBE_ACCURACY (10 samples, bed at 100 °C): range 0.0019 mm, standard deviation 0.0005 mm
  • Known weight at the centre and over each cell: within −2.7 % to +1.6 %
  • Z_TILT_ADJUST: 0.006 mm after 2 rounds

Lessons that cost me the most time

  1. “No samples returned from sensor” while the console said “no errors”. The cause was long SCK/DT wires: klippy.log was full of Forced sensor restart. Moving the HX711 next to the Octopus on 10–20 cm wires fixed it at once. The long shielded cable now runs on the analog side, from the cells to the HX711.
  2. Saturated samples after rewiring: a loose A+ joint, plus flux residue on the HX711. Solder everything, clean with isopropyl alcohol, and wiggle the cables during LOAD_CELL_DIAGNOSTIC.
  3. My first calibration was made while the signal still had faults and came out at 210 counts/g instead of 107.5. Calibrate last, with 1–2 kg.
  4. With the Z motors on the left and right, Z_TILT_ADJUST only fixes left-right tilt. Front-to-back has to be shimmed.

Guide and configs (parts, wiring, config, test order, troubleshooting):

Feedback is welcome, especially from anyone running an HX717 or a different board.

JFYI,
I guess 10 sps “should” work, but there is a “bug” in the code, so at too-low sample rates, the sample timestamps are screwed up. Some info here: Load cell probe: use collector per session by nefelim4ag · Pull Request #7417 · Klipper3d/klipper · GitHub

Unfortunately, I’ve never used load cells, but as far as I understand, parallel connection of “resistors” is not ideal. Under ideal circumstances, there should be a 4-channel load cell, but the current code lacks support for 4-channel load cells.

Where I guess the output of the channels should be summarized.

(I guess there are different ways to tackle that, or maybe tackle support for several sensors and/or channels).

*The link to GitHub is broken now, btw.

Regards,
-Timofey