Backlash and Slack Compensation [SOLVED]

[Jet]

Tried a GT2 Steel belt, didn’t solve it, infact error increased, so we can probably rule out the belt. Checking the idlers again, noticed different behaviour on the idlers since reassembly, before one went up whilst one went down with a change in direction of travel, but on reassembly (with different idler order), only one idler does it so going to scramble the idler order to see if that changes the error amount, maybe one isn’t straight or there’s a bearing problem.

[JuKu]

While at it, check the sturdiness of the frame also, in case something has gotten loose. If you need or want spare idlers, please let me know.

[Jet]

While at it, check the sturdiness of the frame also, in case something has gotten loose. If you need or want spare idlers, please let me know.

Frame checked, everything is okay

[Jet]

Revisited this again.

I’m just concentrating on the X axis (although Y has a similar problem, X is easier to isolate as it’s one belt).

Outruled via testing the following, Cable Carrier, flexure of the motor / idler plates and belt tension, belt vertical positioning on the idlers/gear. The one thing that does make a difference is changing the belt, but doesn’t solve the issue.

The error is variable but maxes out at about 0.085mm (with the Liteplacer shipped belts) which is about 6.8 x 1/8 microsteps. This is with Slack Compensation on.

I’ve noticed that for X positions 300 -> 200 -> 300 ->200 -> 300 etc and measuring a dot at X=300 with the camera, the position of the measurement of the dot increases in X (a large amount on the first move, and gradually increasing until it reaches a maximum after about 30 moves). And then when moving from 300->400->300->400->300 etc. it jumps initially a large amount on the first move, then gradually increases in the opposite direction until it reaches a maximum after about 30 moves. I.E. Behaves as though uneven tension is being removed or the belt is stretching unevenly.

Here are some numbers:

Error between left/right movements at the extremes after settling (30 moves):

White Steel Belt under high tension: 0.362mm
White Steel Belt under medium tension: 0.213mm
White Steel Belt under low tension: 0.352mm

Black (Liteplacer) Belt under high tension: 0.171mm
Black (Liteplacer) Belt under medium tension: 0.085mm
Black (Liteplacer) Belt under low tension: 0.128mm

I’m concluding steel belts aren’t the way to go (although there should be less stretchy, so not sure why).

So I’m going to order a bunch of fibreglass belts and see if I get better results.

[wormball]

I think it is about fitness of tooth shapes on the belt and on the pulley. So maybe you should try various pulleys also.

[Jet]

I think it is about fitness of tooth shapes on the belt and on the pulley. So maybe you should try various pulleys also.

Thanks for the suggestion, I did try another make of pulley recently but it didn't improve it. Theoretically at the correct belt tension, the effect of the tooth backlash should be less
than the larger values I'm seeing.

Infact I see what could be classed a 2 backlashes, at small distances moved I see the regular tooth backlash you're describing (and it's small in value), at larger scale movements I see the
larger scale backlash which increase to a limit and then stops increasing. The tooth backlash is effectively removed by the Slack Compenstation setting, the larger one isn't, it requires a slack compenstation movement that is much larger than practical to remove.

[JuKu]

Your values are much larger than I’m seeing. Of course, this proves nothing. The gradually decreasing error is interesting behavior. Have you tried somehow to measure if the belt tension changes? Or, does the gantry position drift, and not just the camera view (which means that it the camera mount is solid and we are looking at the right direction)? Does using different speeds make any difference?

Also, a couple of things that have made me to look at a wrong thing some time in the past, that might be worth checking, if you haven’t already (although I’m sure you have looked over your machine many times already):
- V wheels: To see that the wheels or wheels adjustment isn’t the reason, you could loosen the bottom wheels (you might need to slow down the machine also) and check if the problem still exists.
- Belt ends slightly slipping.
- Belt teeth rubbing against idlers, effectively shortening the belt a little.

[wormball]

You may also want to test kevlar belts. I ordered this one today: https://ru.aliexpress.com/item/5m-10m-2 ... Title=true

for X positions 300 -> 200 -> 300 ->200 -> 300 etc and measuring a dot at X=300 with the camera, the position of the measurement of the dot increases in X (a large amount on the first move, and gradually increasing until it reaches a maximum after about 30 moves)

Looks like thermal (or other) stretching of part of the belt that is in contact with pulley. I will also measure this on my liteplacer.

[Jet]

The gradually decreasing error is interesting behavior. Have you tried somehow to measure if the belt tension changes? Or, does the gantry position drift, and not just the camera view (which means that it the camera mount is solid and we are looking at the right direction)? Does using different speeds make any difference?

It's decreasing error in that it stabilizes, i.e. when direction change happens, there's an initial jump, then the error increases with repeats until it stabilizes.
It's the gantry moving, not the camera and happens at all "reasonable" speeds including very slow. The is uneven tension in the belt, as in, if the gantry is at the mid point of the axis, and I pull equally on both sides of the belt, the gantry centers to a center point that is matched to moving in the opposite direction and doing the same.

- V wheels: To see that the wheels or wheels adjustment isn’t the reason, you could loosen the bottom wheels (you might need to slow down the machine also) and check if the problem still exists.

Just tried, doesn't make any difference

- Belt ends slightly slipping.

Checked this, they're teeth locked and the X axis are also locked on the edge of the plate, cable ties are tight.

- Belt teeth rubbing against idlers, effectively shortening the belt a little.

Not sure what you mean on this one. Are you talking about the edge of the belt touching the flanges on the idlers? The idlers are larger than the belt, so it shifts up and down slightly during movement. I've tried adjusting the gear to force it to the top or bottom consistantly, but it doesn't change the position.

Now onto the bizarre. I left it for 3 days, and checked again today, hadn't changed anything, and suddenly X has fixed itself, and is down to a consistent 0.2-0.3mm error, I cannot tell why unless the belt had stretched to a fixed length, but that doesn't make sense as I've been using this belt since the beginning. 3 days ago I did restring the belt during test, and the error was there afterwards, but now it's not.

Y hasn't however. Here's a chart of Y. Starting at the center point, Y=200 (slack compensation on), and I do the following movements:
Index 1->14, Y=100, Y=200, Measure Dot (repeat to index 14)
Index 15->28, Y=300, Y=200, Measure Dot (repeat to index 28)

Note the jump between 14-15, this is the backlash is not compensated by slack compensation: 0.063mm in this case.
Also note decreasing values to 14, it plateaus at around 30 repeats and increasing values from 15 onwards (again plateaus around 30 repeats).

Screen Shot 2018-11-18 at 9.30.14 PM.png (81.52 KiB) Viewed 6858 times

[wormball]

I measured difference in coordinates while moving from (38, 18) forth and back by 500 mm (X) or 300 mm (Y).

Slack compensation off, vigorous homing on.

X coordinate difference after (38, 18) > (538, 18) > (38, 18) moves (micrometers) (it's what "measure" button says with reverse sign):

21, 61, 57, 61, 57, 57, 61

Now i moved (38, 18) > (38, 318) > (38, 18) and registered Y coordinate difference:

0, 73, 73, 73, 76, 76, 82, 82, 97, 103, 82, 76, 73, 76, 79, 82, 73, 70, 76, 73, 73, 73, 73

Now X coordinate difference after X moves again:

79, 76, 73, 70, 70, 70, 73, 73, 73

Then i performed (38, 18) > (538, 18) > (38, 18) about 15 times and got difference (73, 30).

Then Y difference after Y moves again:

-27, 42, 57, 57, 64, 67, 67, 67

After another 15 (38, 18) > (38, 318) > (38, 18) moves the difference was (39, 45).

Another X difference after X moves:

57, 57, 57, 57, 61, 57, 57, 61, 61, 61, 57, 64, 61

Another Y difference after Y moves:

24, 33, 36, 42, 45, 42, 48, 48, 45, 57, 48, 33, 42, 39, 39, 39, 42, 42, 42

Now i turned slack compensation on.

X difference after X moves:

-6, -9, -9, -6, -9, -12, 3, -3, 3, -3, -6, -3, -3

Y difference after Y moves:

-15, 24, 21, 21, 21, 21, 21, 21, 21, 18

So the maximum difference is (58, 130) without slack compensation and (15, 39) with slack compensation.