[FR] (Practical algorithm provided) Vibration compensation

[nalimcos]

Hello!

'Ringing' (wavy surface pattern) is a common problem when attempting higher printing speeds; vibration compensation has been brought up here before, but this approach seems new, and I believe it would require little enough processing power to comfortably run even on 8-bit controllers.

Quick description

• No acceleration, high jerk - any speed change is 'instantaneous' here. We can tone that down later on.
• Roughly sinusoidal pattern -> print head motion can be approximated as a simple sprung mass system.
• Let's call T the vibration period of the system when disturbed. (/!\ T is something else in the article)

<some math here, check the article for details>

By cutting each corner with an additional segment, along which the speed vector is the average of the speed vector for the previous and next segments, and starting T/4 before reaching the initial corner, and ending T/4 after when it would have been reached (i.e. T/2 after the start of the corrective segment), we can replace the usual oscillation by two waves which cancel each other, resulting in strongly reduced artefacts. This image was output by a simple simulator I wrote to test my idea.

• Grey lines: uncompensated and compensated 'ideal' trajectories
• Red line: actual print head motion, no compensation, slight damping.
• Blue line: actual print head motion, compensation calculated without accounting for damping, slight damping applied anyway. i.e. non ideal conditions but a nice demonstration (No damping in the model leads to a slightly wilder red line and an even better fit of the blue line)

Full explanation with lots of math in attached PDF: vibcomp.pdf . Old work in progress so a bit messy though.

A simple acceleration-induced-vibration correction method as a by-product While extending this approach to compensate for linear-acceleration-induced vibrations is not as easy, acceleration can be replaced by segments of movement at varying velocities, such that each segment has the duration of half a vibration period of the uncorrected system. Incidentally, this requires much less processing power than even true linear acceleration.

I will add more information and further ideas later on, time for bed here!

Milan

EDIT:

• Requires one printer-specific correction factor per corrected axis (or one for all of them in case of a symmetrical design)
• Factors easily measured on a test cube printed without compensation, at suitable settings
• Article provides bits of formulas linking max acceleration, average acceleration, max uncompensated deviation, max compensated deviation...
• Problem when working with files sliced into short segments: no solution for adjacent segments; solution: look for first further segment for which a suitable end point exists within its first half? needs more thought

[jrabkid]

Hey folks, I don't suppose there has been any update on this project by any chance? Seems this one is the only project that seems to be focusing on this topic.

[AapoTahkola]

https://github.com/MarlinFirmware/Marlin/commit/232a104a927988c63f8c0c53a8c2e26005166e2d fixed lin advance + s curve combo :) @tombrazier I salute you!

[thisiskeithb]

See https://github.com/MarlinFirmware/Marlin/pull/24797

[github-actions[bot]]

This issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.