FW3.9 versus 3.8 - Yikes!  

  RSS
bobstro
(@bobstro)
Illustrious Member

I've been using FW3.9, both RC and final, for a while and was having some print quality issues. I assumed the issues had to do with the Linear Advance upgrade to LA1.5. I read something about specific LA1.0 values causing problems with the conversion. I decided to do a test with the 20mm XYZ cube on 3.9, and after rolling back to 3.8 with no factory reset.

  • For the prints using FW3.9 (left in pics) I used a LA value of 0.2 (M900 K0.2 per the Marlin test print results).
  • For the prints using FW3.8 (right in pics) I used a LA value of 30 (M900 K30).

The results are more dramatic than expected:

I'm staying on 3.8x until either some additional guidance is available, or a fix is offered. I really am looking forward to the additional features long-term.

This topic was modified 5 months ago by bobstro
My notes and disclaimers on 3D printing and miscellaneous other tech projects
He is intelligent, but not experienced. His pattern indicates two dimensional thinking. -- Spock in Star Trek: The Wrath of Khan...
Posted : 08/06/2020 6:31 pm
Kolbi
(@r-kolbi)
Eminent Member

Indeed, your 3.9 looks rough. I've got mine spot-on using K0.15 - after printing an LA tower of sorts. Also, I found it easiest to enable live-k and make adjustments on the fly.

Something else I experienced was x/y jerk values were baken in at 10. Using terminal I could issue change but M503 reflects that the change was ignored. Since I had some other changes to make to the firmware, I edited the jerk value, live-k, and some other non-related things, recompiled, and have been good to go.

 

Posted : 08/06/2020 6:55 pm
Neophyl
(@neophyl)
Noble Member

and this is why I'm still on 3.7.0.  It works and I dont need any of the new features.  There's a thread on github where some members are printing a test model at around 150mm/s and looking at the ringing and corners etc with 3.9.0 and 3.8.1.  I'm getting better results with 3.7.0 than the pictures they posted.  Not sure if its my geared extruder but I've got less ringing and my corners are sharp so I'm sticking with what I have until there's a reason to upgrade.

Posted : 08/06/2020 7:19 pm
v.wood liked
vintagepc
(@vintagepc)
Noble Member

0.2 seems way too large for a K value, this is probably your issue. 

Stock Prusa PETG values of 45 translate to  0.12 (and I found 0.08 to be even better for PETG). 30 for PLA translates into either 0.04 or 0.06.

 

The default values on the marlin test generator are for bowden setups, it seems, and you really want to check the range ~0-0.1 (or 0.2) in 0.02 steps.  See https://github.com/prusa3d/Prusa-Firmware/issues/2543 for examples of this.

This post was modified 5 months ago 2 times by vintagepc
Posted : 08/06/2020 7:52 pm
friedhelm.prusa
(@friedhelm-prusa)
Active Member

Just to throw in some numbers, i did a LA1.5 calibration with PETG from "dasfilament" with a k-value of 0.08

Posted : 08/06/2020 7:58 pm
bobstro
(@bobstro)
Illustrious Member
Posted by: @friedhelm-prusa

Just to throw in some numbers, i did a LA1.5 calibration with PETG from "dasfilament" with a k-value of 0.08

What were the results?

My notes and disclaimers on 3D printing and miscellaneous other tech projects
He is intelligent, but not experienced. His pattern indicates two dimensional thinking. -- Spock in Star Trek: The Wrath of Khan...
Posted : 08/06/2020 8:31 pm
bobstro
(@bobstro)
Illustrious Member
Posted by: @vintagepc

0.2 seems way too large for a K value, this is probably your issue. 

Stock Prusa PETG values of 45 translate to  0.12 (and I found 0.08 to be even better for PETG). 30 for PLA translates into either 0.04 or 0.06.

I'll re-test, although I also tried leaving the old LA values (e.g. 30) and got similar results.

The default values on the marlin test generator are for bowden setups, it seems, and you really want to check the range ~0-0.1 (or 0.2) in 0.02 steps.  See https://github.com/prusa3d/Prusa-Firmware/issues/2543 for examples of this.

Good tip. Thanks. Will re-try it with finer settings.

Are you getting better results with 3.9?

 

My notes and disclaimers on 3D printing and miscellaneous other tech projects
He is intelligent, but not experienced. His pattern indicates two dimensional thinking. -- Spock in Star Trek: The Wrath of Khan...
Posted : 08/06/2020 8:32 pm
vintagepc
(@vintagepc)
Noble Member

Yes - I was one of the folks involved in debugging that issue. The problem was much worse in RC1; after the fixes that went into RC3, the prints I was getting were of comparable quality to 3.8.0. 

 

I've been running something close to RC3 on my printer for day to day (haven't gotten around to merging my various tweaks/improvements to the final release yet) and I have not noticed any issues like this after I'd recalibrated in my K values. Mine are probably slightly lower than you would expect because I'm running a skelestruder w/ gearing.

For your re-test, whether you rebooted the printer or not may have had an impact. IIRC when it sees a LA1.5 value, it disables LA1.0 compatibility. There's a sequence of events that re-enables it, but depending how you did that test it might not have gotten triggered. 

Posted : 08/06/2020 8:46 pm
friedhelm.prusa
(@friedhelm-prusa)
Active Member
Posted by: @bobstro
Posted by: @friedhelm-prusa

Just to throw in some numbers, i did a LA1.5 calibration with PETG from "dasfilament" with a k-value of 0.08

What were the results?

0.08 was the k-value for perfect prints. 0.1 (towards higher numbers) and 0.6 (towards lower numbers) had visible deformations in the marlin calibration print.

Posted : 08/06/2020 8:52 pm
bobstro
(@bobstro)
Illustrious Member

Happy to report a full reset and reflash with 3.9 seems to have helped, along with the modified LA values. I printed the Marlin test from 0 to 0.2 at 0.01mm increments and got much more useful results. As expected, M900 K0.08 was the sweet spot. I also verified that using the old LA1.0 value of M900 K30 worked. I thought I'd be OK upgrading between RC and release without the reset. Should have thought of that first!

Left-to-right: Firmware 3.8.1 with M900 K30, 3.9 with M900 K30 and 3.9 with M900 K0.08:

My notes and disclaimers on 3D printing and miscellaneous other tech projects
He is intelligent, but not experienced. His pattern indicates two dimensional thinking. -- Spock in Star Trek: The Wrath of Khan...
Posted : 09/06/2020 5:24 am
bobstro
(@bobstro)
Illustrious Member

For those interested, looking through the 3.9 firmware source, I found (I think) the relevant K factor conversion code:

// Approximate a LA10 value to a LA15 equivalent.
static float la10c_convert(float k)
{
   float new_K = k * 0.004 - 0.05;

    return new_K < 0? 0:
          new_K > (LA_K_MAX - FLT_EPSILON)? (LA_K_MAX - FLT_EPSILON):
          new_K;
}
This will simplify converting filament profiles over to LA1.5. Was not able to easily identify FLT_EPSILON, but this gets me close.
My notes and disclaimers on 3D printing and miscellaneous other tech projects
He is intelligent, but not experienced. His pattern indicates two dimensional thinking. -- Spock in Star Trek: The Wrath of Khan...
Posted : 10/06/2020 8:05 pm
vintagepc
(@vintagepc)
Noble Member

FLT_EPSILON is not important in that regard. 

(It is a measure of the difference to the next smallest value the floating point value can represent with its discrete number of bits - this is a subtlety of comparing floating point values, as rounding errors can mean that 1.0 !=1.0. Using a construct like this helps alleviate that to some degree.)

Posted : 10/06/2020 9:20 pm
bobstro liked
Share:

Please Login or Register