[BUG] Numeric Error Accumulation Causing Layer Shifts

[murk-sy]

Printer model

Mini+

Firmware version

6.2.0

Upgrades and modifications

Filament sensor

Printing from...

USB drive

Describe the bug

This is a known issue from release notes, but I believe it is a regression as it started appearing on known good gcode.
Was not an issue on 6.2.0-alpha2, can't say for sure about 6.2.0-RC1.

On 2 different gcode files so far I've encountered consistent layer shifting, of course using gyroid (that I usually use).

Gcode B (on Mini 3)
This one was sliced recently and failed both times it was printed.
The shift is by exactly the same amount.

Gcode A (on Mini 2)
This one was sliced on Prusaslicer 2.8.1 and was printed a dozen or so times on both Mini 2 and Mini 3 before the current stable version, without any layer shifting. This is the main reason why I believe it is a regression of some sort.

How to reproduce

Gcode in question can be provided privately to Prusa devs if needed

Expected behavior

No response

Files

No response

[dolezelk]

Same nasty issue on Mk4S. 6.2.0 - firmware
Started a big 16hours print, after it was 40% finished I noticed massive layer shift.
I canceled the print, did check the gcode (was fine)
Started the print again and layer shift comes at the same Z height.

How could be something like that even released for public? is the Alpha 0.1 release ?
Wasted a lot of filament worth $20, will Prusa cover the cost?

[CZDanol]

The accumulation bug is not a regression. You're possibly attributing it something it didn't cause.

[danopernis]

@murk-sy Sorry to hear about that. As a workaround, try using different infill if possible or downgrade the firmware. Yes, we would like to receive your G-Code for our internal testing. Did you observe this on other printers, or just MINI?

[murk-sy]

@CZDanol I can't say for sure it's the issue mentioned, but considering it happens on previously working gcode, I think it's safe to say it's a regression of some sort.

@danopernis Sent an email with both gcodes to [email protected], title "Prusa-Firmware-Buddy Issue 4465: Numeric Error Accumulation Causing Layer Shifts". Total 24MB but hopefully there won't be issues receiving it.

I've only noticed the issue on Minis, haven't had much prints on the XL with any decent amount of gyroid infill since 6.2.

After cancelling the print I believe there were skipped steps on both axes, but I don't remember 100% for sure.

Side note: If there will be an alpha of this fix sign me up. I print the gcodes in question in bulk anyway so printing a bunch of parts into stock is not an issue.

[pepa1367]

I have also experienced this issue on my MK4S fw6.2.0. In the second try the layer shift happened exactly at the same layer. Do you want my gcode as well?

[murk-sy]

Gcode B with cubic infill works without any issues (everything else including versions being the same).

@pepa1367 I don't believe yours is related, as you're using grid infill and while similar, the offset between failed prints is not the same. If I had to guess it's most likely a nozzle crash due to failing supports/overhangs. But you could post the gcode just in case.

[CZDanol]

@murk-sy it is possible that your issue is a regression. But the numeric error accumulation is not a regression, it has been in the firware for some time, we have just identified it now.