Extruder Linearity

Phaedrux

I just came across this from Prusa Research about calibrating the extruder motor for linearity using a feature of the Trinamic drivers they use on the MK3. They claim to be able to reduce step variation and consequently the moire pattern that can sometimes show up. I've been trying to get rid of it at low layer heights myself, which is how I came upon it.

Found here: https://github.com/prusa3d/Prusa-Firmware/wiki/Extruder-linearity-correction-calibration
and here: https://github.com/prusa3d/Prusa3D-Test-Objects/tree/master/MK3/ECOR_TOWER
and the gcode file here: https://raw.githubusercontent.com/Prusa3D/Prusa3D-Test-Objects/master/MK3/ECOR_TOWER/PLA_MK3_ECOR_TOWER.gcode

So my question is, do the Trinamic drivers on the Duet support the same command they use?

TMC_SET_STEP_E#

wilriker

@phaedrux From the very vague description of what they are doing I can only assume that they are playing with TBL value in the CHOPCONF register. You can use my chopconf-generator to modify the value and get the appropriate M569 Cnnn parameter.

In Parameterization of spreadCycle Trinamic recommends to increase TBL to improve the waveform but note that Duet already runs at the second highest value for this paramter so you also might want to try the two below (there are only four different settings in total for this parameter).

EDIT: I only now looked into the GCode file and saw they are using custom commands in there. Will have a look in the source code what they do in detail.

EDIT2: So, after looking into the source code of their Marlin fork it seems to me that they are using a feature that is not present in TMC2660 as used on the Duet. I looks to me as if it is possible to manipulate the internal sine wave table of TMC2130

Phaedrux

@wilriker Thanks for digging into this. Drivers sure have come a long way from the A4988s.

dc42

I can believe that linearity could be an issue with ungeared extruders, but I've never encountered a problem that I can put down to motor linearity when using 3:1 geared extruders, even though all my printers use 1.8deg extruder stepper motors not 0.9deg. However, I rarely print with nozzles smaller than 0.2mm or layer heights below 0.2mm (occasionally 0.1mm).

Any linearity correction applied to the motor would probably need to be varied according to the filament, extrusion temperature and extrusion speed, because it will be affected by the amount of back pressure.

So if you think you have a problem with linearity of extrusion, I suggest you look at changing from a 1.8deg to a 0.9deg motor first, and higher extruder gearing second. You may also need to increase extruder microstepping.

wilriker

@dc42 said in Extruder Linearity:

I can believe that linearity could be an issue with ungeared extruders [...]

This would at least explain why Prusa is looking into this kind of stuff as the MK3 uses an ungeared Bondtech extruder.

deckingman

@dc42 said in Extruder Linearity:

So if you think you have a problem with linearity of extrusion, I suggest you look at changing from a 1.8deg to a 0.9deg motor first, and higher extruder gearing second. You may also need to increase extruder microstepping.

I would add increase nozzle diameter if you possibly can. The difference between a 0.4mm nozzle and a 0.5mm nozzle is about 50% greater area so much reduced pressure for not too much loss of resolution (IMO).

Phaedrux

Well I was originally looking into this because I was seeing some tiger stripe like surface finish on lower layer heights below 0.2 when the layer lines are less dominant of a surface feature. However, I think that is resolved now after replacing some bad v wheel bearings.

Left is the issue I was seeing. Right is after replacing the wheel bearings.

I'm already using a Titan Aero (3:1) with a 0.9 stepper. And I will be giving 256 microsteps a try, though I'm a little concerned about 12883 e steps. So far so good though.

deckingman

@phaedrux said in Extruder Linearity:

Well I was originally looking into this because I was seeing some tiger stripe like surface finish on lower layer heights below 0.2 when the layer lines are less dominant of a surface feature. However, I think that is resolved now after replacing some bad v wheel bearings.

Left is the issue I was seeing. Right is after replacing the wheel bearings.

I'm already using a Titan Aero (3:1) with a 0.9 stepper. And I will be giving 256 microsteps a try, though I'm a little concerned about 12883 e steps. So far so good though.

The thing to watch out for with 256 micro stepping of the extruders is how fast you can retract before running into problems. My Bondtech BMGs have the same gearing and roughly the same steps per mm as the E3D Titans and the best I can get is about 1800 mm/min (30 mm/sec) with 1.8 degree motors so I'd guess with 0.9 degree motors you'd be looking at half that (or use 128x instead).

Phaedrux

@deckingman Thanks for the tips.

I haven't noticed any retraction issues yes in the limited testing I've done so far (it's 3 hours through a 4 hour print) and I'm retracting 0.8mm @ 7200mm/min.

Just checking M122:

Slowest loop: 121.14ms; fastest: 0.08ms
=== Move ===
Hiccups: 3995951, StepErrors: 0, LaErrors: 0, FreeDm: 152, MinFreeDm: 120, MaxWait: 139866ms, Underruns: 0, 0
Scheduled moves: 143654, completed moves: 143624

I'm not sure if that's a lot of hiccups and missed moves or not. @dc42?

Phaedrux

And by the end of the print

Slowest loop: 156.84ms; fastest: 0.08ms
=== Move ===
Hiccups: 1478554, StepErrors: 0, LaErrors: 0, FreeDm: 240, MinFreeDm: 150, MaxWait: 172ms, Underruns: 0, 1
Scheduled moves: 3, completed moves: 3

dc42

Yes that is a lot of hiccups, which are caused by the step pulse generator being unable to generate steps fast enough. When hiccups occur, step pulses are no longer generated at uniform intervals, making missed steps more likely. I suggest you reduce microstepping or maximum speed.

Phaedrux

@dc42 thanks. I'll go back to 16 with interpolation. I wasn't able to notice any difference in print quality one way or the other.

dc42

You may find that reducing extruder microstepping to 128 is sufficient.

Phaedrux

@dc42 thanks. I'll give that a try first.

I understand from some of the work shown by @deckingman that higher microstepping can actually give better extrusion accuracy in certain situations. And if any axis could benefit from it, I would think it the extruder.

T3P3Tony

If noise is not a consideration I assume that "real" microstepping is preferable to interpolation?

deckingman

@phaedrux said in Extruder Linearity:

@dc42 thanks. I'll give that a try first.

I understand from some of the work shown by @deckingman that higher microstepping can actually give better extrusion accuracy in certain situations. And if any axis could benefit from it, I would think it the extruder.

Only if you are using a mixing hot end where one filament could be extruding as little as 1% of the total. I wouldn't imagine there would be all that much to gain by using higher than (say) x32 micro-stepping or 0.9 degree motors and 16x. It's easy enough to check. Just look at the extrusion amount you get for small segments in mm, then divide your steps per mm that to see how many micro-steps you get. IIRC the numbers I was looking at for a small 0.5mm segment gave an extrusion amount of around 0.025mm. Which at 415 micro-steps per mm (16x for a 3:1 geared extruder like Bondtech BMG or E3D Titan) is about 10 micro-steps which ought to be fine. But if you want to extrude 1% of that with a mixing hot end, then it's only 0.1 micro-step and ain't gonna work.

Phaedrux

@deckingman yes those are the specific circumstances I eluded to, thank you for summerizing then far more accurately and concisely than I could have.

I guess another way to think about it isn't that microstepping is providing more accuracy, but rather that in this instance there aren't enough steps to capture the movement slices. Higher stepping means more step pulses and each step can capture a smaller movement slice so nothing gets lost to rounding.

In my case with a single extruder I don't think this applies to me.

deckingman

@phaedrux said in Extruder Linearity:

@deckingman yes those are the specific circumstances I eluded to, thank you for summerizing then far more accurately and concisely than I could have.

I guess another way to think about it isn't that microstepping is providing more accuracy, but rather that in this instance there aren't enough steps to capture the movement slices. Higher stepping means more step pulses and each step can capture a smaller movement slice so nothing gets lost to rounding.

In my case with a single extruder I don't think this applies to me.

Exactly so. But do the maths. My typical usage case is with 0.3mm layer height and 0.5mm nozzle diameter. With smaller layer heights and nozzle widths, the extrusion amount for a given segment length will be less, and there may be a case for using say 32x or 64x micro-stepping in order to capture those small movements. Bear in mind that 0.9 degree motors @ 16x micro-stepping will give the same steps per mm as 1.8 degree motors @32x micro-stepping. But 1/16th micro-step will have more torque than a 1/32 micro-step all other things being equal (like the torque rating of the motors).

Phaedrux

@deckingman Yes I'll have to dig into it a little deeper. I do print at low layer heights down to 0.05 at 0.4 extrusion width.

wilriker

Reading all of this forms the question in me: is there a desirable minimum number of steps/mm for the extruder? Or even better have a calculator that can provide a recommendation based on layer height and nozzle size/extrusion width.

deckingman

@wilriker said in Extruder Linearity:

Reading all of this forms the question in me: is there a desirable minimum number of steps/mm for the extruder? Or even better have a calculator that can provide a recommendation based on layer height and nozzle size/extrusion width.

What I was concerned about was that with very small extrusion amounts (i.e. when using mixing ratios of 1%) then the amount of material to be extruded was less that one micro step at the 16x setting that I was using. So the desirable micro-stepping would be that which allows the smallest extruder move to be accomplished. I did do a spread sheet but it's specific to my printer/layer height/nozzle diameter. Looking through some old gcode files, I found that for a 300mm carriage move, the extruder move was about 15mm so it kind of follows that the extruder moves are about 5% of the carriage move. I used that 5% as a basis and calculated the extruder move for a realistically small segment move of 0.5mm giving me 0.025 mm of extrusion. Then of course with a mixing ratio of 1% of that, it moves the decimal point two places to the left. It's easy enough then to compare that amount of movement with what 1 micro-step will give at different micro-stepping settings.

For a more universal calculator, I'd say that you've talked yourself into a job

wilriker

@deckingman said in Extruder Linearity:

For a more universal calculator, I'd say that you've talked yourself into a job

I saw that one coming. So, I guess I gotta do what I gotta do then.

wilriker

OK, created a spreadsheet to calculate minimum microstepping factor for a given line width, layer height, filament diameter, flow rate and segment size. To get reliable extrusion lengths I sliced a small disc in Cura with varying line width and took their source code as a basis for how they calculate how much the extruder should move to get the required amount of plastic pushed through the nozzle.

My only problem is: somewhere in Cura there seems to be a multiplication factor of slightly more than 57.1% that I cannot find anywhere. I have no clue where it is coming from but it is consistent between line widths of 0.4, 0.45 and 0.5mm.

E.g. for a line width of 0.5mm and a layer height of 0.3mm and a filament diameter of 1.75mm it would take 0.10913mm of extruder movement to extrude 1mm of filament through the nozzle. But instead Cura will write 0.06237mm.

Has anyone any idea? Maybe @burtoogle?

jeanmarc_Scaled

You could re-slice with Slic3r, then use the flow calculations from the manual: http://manual.slic3r.org/advanced/flow-math

wilriker

@jeanmarc_scaled Thanks! I will definitely look at this.

As all slicers do basically the same (don't hit me ) this might already explain where this strange multiplier is coming from.

But still, if anyone else has an idea I am willing to hear that. Especially since I followed the source code trail in CuraEngine on how this amount is calculated - but clearly I missed something.