G32 bed level correction converges very slowly

sverreb

I have added a 3rd leadscrew to my experimental 300x300 corexy printer in order to test automatic bed levelling. I do however find that corrections done are way to small and I only approach a level bed by calling G32 multiple times:

See log below

I also tried a g29 probe to see the shape and it was clearly tilted around the x axis.

Firmware:
Firmware Name: RepRapFirmware for Duet WiFi
Firmware Electronics: Duet WiFi 1.0 + DueX5
Firmware Version: 1.19 (2017-08-14)
WiFi Server Version: 1.19
Web Interface Version: 1.19

leadscrew settings: (from config.g)
M584 X0 Y1 Z7:8:9
M671 X-37:150:337 Y150:329.5:150; locate leadscrews

log:
8:09:58 PMG32
Leadscrew adjustments made: -0.073 -0.085 -0.086, points used 5, deviation before 0.087 after 0.036
8:08:39 PMG32
Leadscrew adjustments made: -0.090 -0.127 -0.090, points used 5, deviation before 0.099 after 0.039
8:07:43 PMG32
Leadscrew adjustments made: -0.121 -0.186 -0.113, points used 5, deviation before 0.126 after 0.036
8:06:39 PMG32
Leadscrew adjustments made: -0.170 -0.308 -0.180, points used 5, deviation before 0.189 after 0.039
7:59:50 PMG32
Leadscrew adjustments made: -0.240 -0.270 -0.223, points used 3, deviation before 0.241 after 0.000
7:58:47 PMG32
Leadscrew adjustments made: -0.260 -0.343 -0.262, points used 3, deviation before 0.281 after 0.000
7:55:53 PMG29
169 points probed, mean error 0.041, deviation 0.103
Height map saved to file heightmap.csv
7:43:35 PMG32
Leadscrew adjustments made: -0.306 -0.431 -0.298, points used 3, deviation before 0.335 after 0.000
7:42:44 PMG32
Leadscrew adjustments made: -0.335 -0.490 -0.334, points used 3, deviation before 0.374 after 0.000

As expected the middle-rear leadscrew is moved a bit more than the others to level the tilt, but not nearly enough.

dc42

The leadscrew adjustment algorithm assumes the ideal case of the bed remaining perfectly flat and gimbaling about the leadscrew nuts. If the leadscrew adjustment causes the bed to flex then the algorithm won't be accurate.

It might be instructive to probe the set of 3 points twice in bed.g: once with S-1 on the last G30 command to print the height errors, and once with S3 to apply the corrections.

sverreb

I think you are right. I tried moving the leadscrew by hand and only got 20% of the movement on the bed. Likely the bed resists torque around the x axis enough that the motor mounts (which are printed) deflects more than the bed rotates.