A couple of ideas

[WayneManion]

First of all, cool project.

Comment/idea #1: I have asked around the Klipper Discord about this idea and the reaction has always been something to the effect of "I don't want to ping my bed with the nozzle because it will damage the nozzle/the heatbreak/the POM nuts." I was thinking maybe your technique could mitigate some of these concerns if it worked in concert with another probe to measure z-offset? For the sake of this example, let's say a lightweight inductive probe with only three wires and no moving parts.

• Since the other inductive probe has "some idea" of where the bed is, you aren't stabbing the nozzle blindly at the bed.
• The z-offset only needs to be calculated once every nozzle/hardware change, so there are far fewer "accelerometer-based probing" events than there are when building a bed mesh using only an accelerometer
• Since you only really need a single sample, the "accelerometer-based probing" can take place at the corner of the bed where models usually aren't printed anyway.

Possible work flow for accelerometer-based z-offset measurement:

1. Home the printer using the inductive probe.
2. Move to an extreme corner of the print bed. Probe z with the inductive probe.
3. Move the probe down until a tap is detected.
4. There's your z offset.

Comment/idea #2: years ago I built an ESP8266 project to use the tap detection feature of a LIS3DH sensor to determine if my dryer was running. If there were lots (50+) of tap events within a rolling 10-second window it meant the dryer was running. Otherwise it must not be running. I got this project to work with just a little calibration using LIS3DH sensors. I tried to re-implement it later using ADXL345s because they were cheaper and more widely available at the time. I never got it to work reliably with ADXL345. Long story short: have you ever tried this technique with LISx sensors? Klipper recently added support for these and there is at least one pre-made RP2040/LISx input shaping board available.

[jniebuhr]

I've separated idea 2 into #4

[jniebuhr]

For Idea 1: Voron famously uses 800g of downforce on the bed now for probing. We're closer to 150-200g. But of course you could use multiple probes to measure offsets. It's probably doable in G-Code as well. See this: https://github.com/protoloft/klipper_z_calibration

[WayneManion]

Do you happen to know how much force the Bambu printers exert on the bed-mounted load cells in the X/P printers?