Slow and weird at 140mm/s on curvy/circular path

dc42

Please run M122 to clear the stats, then run your test print for a little while at full speed. Then run M122 again and look for for the count of Underruns in the Move diagnostics. Repeat using 70% speed. Then report the underrun counts here.

Hugues

Ok, i will do this tomorrow morning.
Thanks !

Hugues

Hi,

At full speed :
Move Diagnostics:
MaxReps: 8, StepErrors: 0. Underruns: 2475

At 70% :
MaxReps: 5, StepErrors: 0. Underruns: 2869

dc42

That's the first time I have seen an underrun count higher than a few tens, and is almost certainly the cause of the problem. Can you confirm that you are printing from SD card?

Hugues

Yes printing from SD card, the file is uploaded by the web interface and launching with it.
I also tried with the 1.13 firmware and same issue.

Now i will try with an other slicer.

dc42

Thanks. The results with another slicer - or with the latest version of S3D - will be interesting.

I have some work scheduled that should allow a greater throughput of gcodes before underruns occur. If it goes smoothly then this will be in release 1,15.

Hugues

The file was sliced with S3D last version.
I made a quick try with kisslicer and Slic3r this last give me the same problem (visual) and MaxReps: 8, StepErrors: 0. Underruns: 1225, more i increase speed (either by the web interface) it also increase the underruns.

Hope it's helping you

Hugues

If you need specific test, don't hesitate to send me them !

roboduet

Just finished printing this model: http://www.thingiverse.com/thing:331035/ (small one), using Slic3r 1.2.9, layer height is 0.2 mm.
Got the following results: MaxReps: 5, StepErrors: 0. Underruns: 443

I can provide gcode file or settings used if this can help.

Hugues

Can you send me a Gcode for this one : http://www.thingiverse.com/thing:30420 .There are more curvy path on it and i will compare the gcode with mine…
Thanks,
Hugues

Hugues

I was worried about a mechanical problem so i did a quick test at 200mm/s with a low polygon model :
https://youtu.be/WnnXbRyDGRo

Really happy of it even i didn't finish the print (children never can wait…). The sound is really different from what i encounter with curve. For reference this is a cylinder printed at 70mm/s and i push the speed to 150 % after 10mm. You can see how many blobs are coming...

bot

Are you running a 12v supply? Seems like you're simply commanding too fast of a feedrate. 0.9 degree steppers require double the number of steps as 1.8. CoreXY requires another doubling of steps compared to cartesian. You're losing a lot of speed to resolution that you may not be using in terms of additional accuracy. I'd suggest going back to 1.8 degree steppers, or upping the power supply to 24v (if not already), or both.

Hugues

I'm still in 24v and i will do a test with a 1.8 motor for extruder but i think that's not the problem cause it work well (previous video) when extruding a straight line. And blods seem to come cause extruder is overextruding compare to the motion

dc42

cubexupgrade, I have put an experimental firmware binary at https://dl.dropboxusercontent.com/u/19369680/DuetWiFiServer.bin. You may find that it allows higher speeds before the quality deteriorates. Or not.

bot

It's not the extruder motor per se that needs to be 1.8, it's simply that you are requesting too many steps in too little time. Try the new firmware, see how it works, and then either dial down speed or resolution. These are your only choices really.

deckingman

@DC42. This isn't something that could be cured by changing the value for M566 is it? I don't really understand how it works except that it is most noticeable when printing cylinders. Just a thought….....

dc42

If it's a regular speed up/slow down when printing "cylinders" with small numbers of facets, then higher M566 values do help to achieve higher print speeds. However, the blobbing in the image suggests something else, which is that the cylinder has a very large number of facets (therefore many gcode commands per perimeter). and the gcode processing and lookahead system isn't filling the pipeline as fast as the motion generation system can empty it when it runs at the requested speed. This is confirmed by the high underrun count. The underrun count is incremented every time the lookahead system finds that in trying to achieve the requested speed for one move, it needs to adjust then profile of some earlier move, that move has already been frozen because it is due to be executed soon.

T3P3Tony

In effect, as bot says, the resolution is too high. This used to be a problem with S3D but using Cura or Slic3r did not have the same issues. If you have tried these and still the same issue maybe try reducing the number of facets of the model?

Hugues

Sorry soccer evening (and we won !! ) !
Thanks for your feedback i will try the experimental firmware tomorow, and it's sure exchanging the 0.9 motor by a 1.8 and if not enough going to a direct extruder, the e3d titan combine to 0.9 motor need lot of signal… I design this bot to be speed but this extruder combo were just here cause my own design are still in work in progress (1.8 motor and geared drive 1 : 2).

I experimented this issue with lot of model, astrobot, morena treefrog... Always when curve or and loop are sharp and short.
I tried kisslicer and it seem to work because the oversampling was set really low but i 've the same issue with slic3r on the same cylinder.

Is there a way to calculate the limit the duetwifi can achieve per configuration ?

bot

dc42 has been making improvements in just that area recently, so the numbers are liable to improve. He has mentioned the latest speeds here: https://www.duet3d.com/forum/thread.php?pid=447#p447

I'm not sure that you can directly calculate the exact speeds you can move, as they depend on the number of individual commands that makeup the steps, how many motors are moving, etc.

You could also try lowering the microstepping to 8x instead of 16x, but I'm not sure the results will be excellent.