Issues with pressure advance since RRF 3.4

sebkritikel

@Phaedrux I see some discussions on it at the bottom of Klipper's Kinematics page
https://www.klipper3d.org/Kinematics.html

Some other pages of interest
https://klipper.discourse.group/t/pressure-advance-smooth-time-on-direct-extruders-with-short-filament-path/1971
https://github.com/Klipper3d/klipper/issues/4442

richfelker created this issue in Klipper3d/klipper

closed Pressure advance smoothing induces e-axis position swings that scale with acceleration #4442

dc42

@Heartleander81 said in Issues with pressure advance since RRF 3.4:

@R4ffers oh yes.

@dc42 is there also something like pressure advance smooth time under RRF? A few clippers have the problem with the high smooth time, but those who have problems are mostly DirectDrive extruder users. Maybe that's an approach you can follow.

No, there isn't anything similar. I will look into that Klipper feature.

The reason I have made no progress on this is that while a number of users seem to think that pressure advance doesn't work as well in RRF 3.4 as it did in 3.3, and I am prepared to believe that there may be a difference because the relevant code had to be rewritten to accommodate input shaping, nobody has been able to provide a simple GCode script and machine settings that demonstrate a difference when input shaping is disabled. That includes me - I have compared RRF 3.3 and RRF 3.4 prints.

Heartleander81

@dc42
@Phaedrux

I couldn't find any differences between 3.3 and 3.4.x, but I still have a picture of a test where the corners became nice with PA. As now with klipper. The Artikel from sebkritel is that who I read.

Argo

@Heartleander81

So altering the smooth time value did improve your corners and PA alone did not?

petriheino

Hi folks, my guess is OP simply did finer calibrations after update and found out underlying RRF issues regarding PA, Retraction, Extruder steps and acceleration relation... I got similar results (improvements) with Klipper conversion. I did some testing and posted it here earlier.

Heartleander81

@Argo said in Issues with pressure advance since RRF 3.4:

So altering the smooth time value did improve your corners and PA alone did not?

PA alone didn't make the whole corner look nice. See the picture with the corner sticking out in the direction of travel. Only with the change of smooth time and renewed PA, which was then higher, did the corners become so clean.

Argo

@Heartleander81

I see.

I know from another Klipper user that he also only could get nice corners by playing around with the smooth time setting. Pressure advance alone did not help much.

Maybe extruders / hotends are so particular nowadays that we need this setting to get nice corners?

Here is the example from the user I'm talking about:

He has a bowden setup and needed to increase the smooth time in order to get nice corners. PA alone did not the trick.

@dc42
How high are the chances we might get this feature for RRF? I think this might very much help those who are having problems with PA at the moment.

Heartleander81

@Argo ok. Nice test from him. I find for my Hevort 0.02 smooth time and 0.044 PA. Worked nice. I hope my Info can help Duet.

CR3D

@dc42 Hi David,

I have to join this conversation because through my contact to @Heartleander81 i noticed this effect with "round corners" at some of our machines to.

I think this comes with newer firmware and if you made some changes into PA there would be an issue. But how can we test it that it would be helpfull for you?

one of our customers has now contacted us explicitly about this effect and is looking for a solution. I can't compensate for it by increasing the PA value. We've tested a lot here.

Thank you Regards Christian from @CR3D

dc42

@CR3D said in Issues with pressure advance since RRF 3.4:

But how can we test it that it would be helpfull for you?

Find a print where there is a clear visible difference between the effect of pressure advance (e.g. on corners or infill) between RRF3.3 and RRF3.4.4 with the same configuration for both, and no input shaping when running RRF 3.4.4.
Edit the GCode file to reduce it to a short print, preferably just 1 or 2 layers.
Post that file along with your config.g and other relevant macro files.

gnydick

@dc42 why don't you just generate the move list from a 3.3 system and from a 3.4 system for the same gcode and output it to text files and diff them?

dc42

@gnydick said in Issues with pressure advance since RRF 3.4:

@dc42 why don't you just generate the move list from a 3.3 system and from a 3.4 system for the same gcode and output it to text files and diff them?

Because none of the prints I and others have tried show any difference between 3.3 and 3.4. Nobody has produced a print and machine settings that shows a difference. If somebody ever does, then I will be able to use the method you suggested or another method to see what has changed.

oliof

I personally believe that the changes must have happened before 3.3 ... or that the observed issue isnt due to a change in firmware but a change in perception of acceptable corner bulging due to the performance of non-RRF systems.

A decent way to bisect this is to take a Duet2 board, and go back all the way to RRF 2.0.3, and then work through at least 3.0, 3.1.1, 3.2.2, 3.3 to see whether my hypothesis holds.

But I wouldn't ask anyone to lose a week of their life to do this...

dc42

@oliof I think you are likely correct. Pressure advance is not a compete solution to corner bulging because there are other factors at play. I have to compensate for one of those other factors separately from PA.

Accuracy of perimeters in general (including reduced corner bulging) is better if you select "External perimeters first" in the slicer. Doing so might make printing steep overhangs more difficult, but I am not aware of any other disadvantages. I have been printing external perimeters first for years.

gnydick

@dc42 why wait? I would do what I'm suggesting over and over with lots of different gcode files.

In fact, I would make that part of the release process to validate.

jay_s_uk

I've printed test files with 3.2.2, 3.3 and 3.4.4 and have no been able to see a difference

oliof

@dc42 thanks for the additional insight! It may also be a bit slicer dependent. I remember KISS Slicer for example is doing a slight inwards bend before steep angles; other slicers may not do this.

Argo

I already tested different slicers and settings also different RRF versions.

Only thing that helped with my machine was connecting Klipper (Pi).
I suspect that PA smooth time is the factor (feature) that is RRF missing to improve this.

But the reasons why my bedslinger (also RRF with Duet 3 Mini) produces better corners? I don't know. My Voron (printer with the issues) uses a different kind of hotend which is also not compatible with the PID algorithm RRF uses as PID tuning only works when calibrating the hotend as heater and not as a tool But that's another issue I already reported in another thread.

gnydick

I can conclusively report that Input Shaping interferes with Pressure Advance. Didn't know if anyone else has been able to confirm on their prints.

evan38109

@gnydick I've also seen issues with input shaping + pressure advance, though I'm always open to having done something wrong on my end. At the risk of hijacking a closely-related-but-not-quite-the-same thread, you can see the artifact show up as an indentation before and after turns in the photo below. I wasn't able to remove it with any amount of tuning.

gcode here.
Ringing tower model here.
Config files for the machine when the above was printed in this commit. (Newer repo state is a bit different; I'm converting to closed loop.)
If you want to re-slice for your machine, the PrusaSlicer layer change script, designed for 0.25mm layer height, is:

{if layer_num== 1}
M593 P"none"                           ; no input shaping
M572 D0 S0                             ; no PA
{elsif layer_num== 60}
M593 P"ei3" F42 S0.1                   ; enable input shaping
M572 D0 S0                             ; no PA
{elsif layer_num== 120}
M593 P"none"                           ; no input shaping
M572 D0 S0.09                          ; enable PA
{elsif layer_num== 180}
M593 P"ei3" F42 0.1                    ; enable input shaping
M572 D0 S0.09                          ; enable PA
{endif}