Vertical Banding Revisited

[KAS]

Increased the speed to 50 mm/s, I think I found my new large part print settings.


[img]http://i482.photobucket.com/albums/rr189/Onuxis/20150209_201656.jpg[/img]

[Jimustanguitar]

Mine got better at higher speeds too... I'm starting to wonder if adding weight to the effector platform would help. If this really is a mechanical resonance or vibration from rough motor steps, it would make sense that you'd see the ridges less at higher speeds because the platform had more inertia to it and it "wanted" to travel in a straight line. I might have to give that a try.

Otherwise, what else could one do to try and decrease vibration? I could try to put the machine on isolation feet. Weight on the top of the machine itself. Those little bowstring whiskers on the drive belts? What do you think?

[DavidF]

The spacing is too consistant for it to be resonance or vibration. I was measuring them on my prints and they are very close to being a perfect 1mm spacing at lower speeds. Im thinking the reason it is less noticeable at higher speeds is that there is some lash in the machine that covers it up. Real question is would going to geared steppers on the axis allow for more accurate positioning, or is what we are running into the positioning accuracy of the electronics and generated g code from the model file? I wonder if viewing the triangles in netfabb could tell us anything?

[Jimustanguitar]

I wonder if viewing the triangles in netfabb could tell us anything?

Not many triangles at all on the purely flat surfaces of the part I've been taking pictures of.

[img]http://i.imgur.com/Rt7hcgE.png[/img]

[DavidF]

Yep, scratch the triangle idea...

[Nylocke]

What I'm thinking it is is the delta segments per second thing. Since its a per second setting rather than per print line (I think I've seen this in smoothie?) The faster you go the fewer segments it generates in the firmware, so the more spaced the banding is. You could probably get some predictable spacing by scaling the speed, I'd guess it'd be linear. David measured a perfect 1mm spacing at X speed, I'd be willing to bet that at double that speed the spacing might be 2mm? Looking a a few of the previous prints done, this may not appear to be the case (at least from how I'm interpreting it) but the slop in the system thing would probably be causing that, inadvertently clearing up the "sloppy arduino math".

[DavidF]

ok here is a 25mm cube printed at 100 microns and 20mm/sec. The banding seem to be about 1.5mm, but with 3 distinct lines in each band. Maybe im seeing what I want to see, but thats what it looks like to me.. Now what is the diameter of the timing pulleys? I cant quite get my caliper in there to measure..
[img]http://i1102.photobucket.com/albums/g446/davidflowers1/Mobile%20Uploads/20150209_232123.jpg[/img]

[Tonkabot]

Are you guys using the stepper dampers? I think that might effect it, too. I thought that my banding might have INCREASED when I put the dampers on, but I am not sure.

When I had the dial indicator instead of the printhead, I reached down and twisted a stepper back and forth - the damper lets it turn some - and I got a good +- 5 thousands of up down on the head. of course dependent on where the head is in relation to the tower your fudging.

[Nylocke]

I took mine off. They were screwing with print quality in weird ways. I'd rather have nice prints and a louder printer

[KAS]

Mine are on. I've never printed with them off to have a comparison.

[JFettig]

I definitely don't have nearly this much banding - no dampers here. Its actually really difficult to see any banding in my parts but it does exist.

[DavidF]

no dampners here....there is a difference when printing in pla vs abs. the banding seems much more pronounced when printed in pla....

[bvandiepenbos]

The pulleys are 20 tooth GT2 belt.
The belt tooth spacing is 2 mm.
So 1 revolution of the pulley is 40 mm of belt travel.

Stock stepper motor is 200 steps/revolution
X 16 micro steps = 3200 steps/revolution

40 mm / 3200 = .0125 mm

If my math is correct then the position resolution of each truck movement is .0125 mm

Does that tell us anything?

This banding issue has drove me crazy for years.

[Tonkabot]

I think a change that would be very instructive is to switch to 0.9 deg step steppers (400steps/rev) and see what happens.

And it is incorrect to believe that a stepper moves in 16 equal microsteps per full step. Microstepping smooths out the stepper motion,
but as someone here posted, http://www.micromo.com/microstepping-my ... -realities , points out that it doesn't work that way.

Going to a .9 deg step motor is much better than switching from 16 to 32 microsteps.

Or going to a fewer tooth pulley should be a good thing too - up to the point that the pulley is too small around. What was the reason seemecnc switched to bigger pulleys?

Another interesting change would be to see what actual servos (not model servos) and encoders (and run the encoders up and down the towers, not on the motors) would do.

The pulleys are 20 tooth GT2 belt.
The belt tooth spacing is 2 mm.
So 1 revolution of the pulley is 40 mm of belt travel.

Stock stepper motor is 200 steps/revolution
X 16 micro steps = 3200 steps/revolution

40 mm / 3200 = .0125 mm

If my math is correct then the position resolution of each truck movement is .0125 mm

Does that tell us anything?

This banding issue has drove me crazy for years.

[KAS]

It appears smooth especially with the potential of three moving at once. Then again, that's where the randomness comes in and doesn't uniformly correlate to the vertical patterns.

Still interesting...Maybe the teeth on the belts are causing some type of issue coming off the top/bottom belt tensioner. Not at home right now, but are the belt teeth rounded or square?


Not conducive to long term although it would be interesting to flip the belt over and run it smooth side down over the pulley.

[bvandiepenbos]

SeeMeCNC changed to 20 tooth to save money. They originally machined custom 16 tooth pulleys themselves.
20 tooth is the smallest standard pulley available.

[KAS]

SeeMeCNC changed to 20 tooth to save money. They originally machined custom 16 tooth pulleys themselves.
20 tooth is the smallest standard pulley available.

https://www.lulzbot.com/products/gt2-16 ... ing-pulley

In your honest opinion, is this better to use?

[Polygonhell]

If I had to guess at causes, it's much more likely to be the linear approximations in the firmware than the resolution of the motion.
If I was really interested in isolating this I'd start by changing the segments per settings in the firmware in an attempt not to remove the issue, but to make it worse, if you can establish a predictable change from the changes in the settings then you can pretty much declare that's the issue, and accept that probably the only real solution is to move to a faster electronics package.
If it's not that, then I'd look at motor resolution next, again the easy test is to turn off uStepping and see what the error looks like, if it's the cause it should suddenly get MUCH worse, as to fixing at that point, you don't really have a lot of options, 9 degree steppers, or geared steppers, but they have their own sets of issues especially with the existing electronics package.
FWIW I'm running an original V1 with 8x drivers and 15 tooth pulleys, I guess my assumption was wrong in that mine always seemed to line up with the infill.

[Tonkabot]

SeeMeCNC changed to 20 tooth to save money. They originally machined custom 16 tooth pulleys themselves.
20 tooth is the smallest standard pulley available.

https://www.lulzbot.com/products/gt2-16 ... ing-pulley

In your honest opinion, is this better to use?

Really cheap ones. How much does seemeCNC save over buying a big batch from China? Or is the quality that bad?
http://www.aliexpress.com/store/product ... 78451.html

[jdurand]

Another thing about stepping motors, only full steps (one coil on at a time) are guaranteed to be equal. Since half and micro stepping has both coils on at once it depends on both coils having EXACTLY the same characteristics. As we live in the real world (says the guy making new-age stuff with two 3D printers and working on kid locks for iPads), the coils are slightly different from each other. This means you'll have a sinusoidal position error as you go from a full step through the microsteps and back to a full step.

[bvandiepenbos]

Actually the pulleys must be 15 tooth not 16 on my original MAX, it also has 1/8 stepping Rambo board.
It has vertical banding also.

Honestly, I do not know if you would see improvement with the Lulzbot 16 tooth, but it should not hurt anything.

[bvandiepenbos]

If I had to guess at causes, it's much more likely to be the linear approximations in the firmware than the resolution of the motion.
If I was really interested in isolating this I'd start by changing the segments per settings in the firmware in an attempt not to remove the issue, but to make it worse, if you can establish a predictable change from the changes in the settings then you can pretty much declare that's the issue, and accept that probably the only real solution is to move to a faster electronics package.
If it's not that, then I'd look at motor resolution next, again the easy test is to turn off uStepping and see what the error looks like, if it's the cause it should suddenly get MUCH worse, as to fixing at that point, you don't really have a lot of options, 9 degree steppers, or geared steppers, but they have their own sets of issues especially with the existing electronics package.
FWIW I'm running an original V1 with 8x drivers and 15 tooth pulleys, I guess my assumption was wrong in that mine always seemed to line up with the infill.

I agree, it is more likely to be in fw.
You have some great ideas there on how to test.

[lordbinky]

It helped when I went to 0.9° steppers, I also moved to the smoothie-board at the same time. Here's the weird thing though. I still get banding along the one of the axis occasionally and not the other.

It is likely a combination issue, where the most prominent cause is going to change depending on various factors like speed making it hard to pin down the exact cause. For example if the issue is really in the bearings or the bearing cover on carriages, then watching the banding pattern expand and magnitude decrease makes sense, you have intertia between steps HELPING you overcome either a friction source (ie the carriage doesn't move at all until the stepper force exceeds the static friction. and since each microstep isn't the same in amount of force it will leads to a pattern based on this) or there is a non-uniformity in the bearings (ie flat spot in something). If it's back to mechanical, go back through unhook the carriages from the belts and do your tightness checks to minimize their friction without letting them wiggle, and don't forget to move them the full length of the tower by hand and feel for any vibration that would indicate a flat spot. If you eliminate that as prominent banding force, then you're left with electrical or software.

I'll refrain from touching on the electrical since I go overboard on it and I'll try to do a software test prints later today. I'm thinking four prints, a slow (~30mm/s) and fast (60mm/s) hollow box each done with the Smoothieware's delta_segments_per_second set as high as I can get it (at least 300/350, maybe 400-500?) and a set at a low value (100). This hopefully will provide some kind of decent feedback. Also, if the switch to 0.9° step steppers does fix the issue for anyone, it still doesn't say if it covered up or fixed the prevailing mechanical/electrical/software issue, or point to which one it is. It could still be mechanical for example, for two identical motors one with .09° steps and the other 1.8°, the force applied on each microstep will be different, changing how many microsteps it takes to overcome the breakpoint of friction for movement over the same distance. This is still true if all other things such as the coordinate's resolution of the software is the same.

[Tonkabot]

I just had a thought, on any given line segment that the head is moving along we have the three steppers all going at once.
Now if we have a move that stays at a constant Y position, and moves from X=-100 to X=100, then for that move the 'Z' tower sits idle.
So for this move, 'z' doesn't move, but 'x' will start high and go down, while tower 'y' will start low and go high.
If the head is moving at constant speed from -100,0 to 100,0, then I believe 'x' will start slow and increase in speed, while 'y' starts fast and decreases in speed.
so for the example I am thinking of I just set up the exact case that it doesn't happen.

So now take the same as above and instead of going from -100,0 to 100,0 we add a small y change: -100,-1 to 100,1.
for this, the 'x' and 'y' towers do almost exactly what they did before, but also the 'z' tower gets in the act and slowly increments while 'x' and 'y' are doing their thing.

It seems to me that as 'z' is making it's very slow move during the 'x' and 'y' moves it would be 'stair stepping' ie: it cannot make a perfect diagonal, it will output a slowly changing position increment that consists of starts and stops - which seem like they would be making some kind of banding on each step change.

Of course almost every move is a diagonal, and also they are changing speed as they move [ unlike a XYZ (can't spell cartesion) printer.
Every diagonal will have the stair-steps [except for the three diagonals parallel to the lines between towers]. Unless you have an infinite number of microsteps per step, and microsteps actually were perfect (have the same torque as a full step), which we know is not possible or true.

[lordbinky]

I didn't bother changing my filament from the clear I was using...so pictures aren't turning out well. Still, preliminary result is that at least for me, it appears to be mechanical since there is little difference between 100 and 300 segments per second, but going faster increases the distance between bands. The print was a 100mm square 10mm high. With a single wall it's a quick print.