15mm kevlar belts

Regarding ballscrew rotation speed, they say <ballscrew diameter> x <RPM> shouldn’t exceed 70000 and that’s about it. So with a 1605 ballscrew, 4375 RPM is my limit.

And I agree that 9mm 2MGT is almost certainly enough but I have dreams of doing crazy things to my machine like sticking on an 80mm spindle with ATC in which case I might want a bit more stiffness in my belts. I couldn’t find backlash numbers in the guide you linked though?

I’ve been looking at giant pulleys out of necessity unfortunately, that’s a big part of the problem. My servo shafts are 14mm and the only pulleys that support bores that large are enormous. But on other hand, I don’t want too large a pulley on the ballscrew because that makes designing the axis as a whole a bit more painful.

I’ve seen a lot of stuff on this forum about tension so I’m very aware of the need. My plan at the moment is to add an idler on top of a stiff spring that can be adjusted by one grub screw and locked into place with another. That should allow me to apply a reasonably precise but consistent force to the belt.

Thanks for the supplier recommendations! I’d looked at POWGE in the past but couldn’t find what I needed from them. I was looking at the 3GT section which looks pretty empty but the 2GT section is apparently pretty full! The others look good too. I like the eBay seller but the problem with most of the European sellers is that they take so long to ship that I may as well just order from China…

You can get pulley bar stock…
Online Product Catalog – Timing Belt Pulleys – Bar Stock | Pfeifer Industries

Here’s the backlash chart

Beware! - the vertical axis on the right hand chart is wrong, it should read 0.001, 0.002, 0.003

I would not expect belt backlash to be an issue for you, assuming it’s tensioned well (meaning consistently not necessarily uber tight and fracturing ballscrew shafts).
If you see the 0.0025mm backlash * 2 for 3MGT on a, say, 40mm diameter pulley, that’s 0.015 degrees on your ballscrew which would come out at 0.0002mm on a 5mm pitch ballscrew, measuring that will require some care to measure at all…

The belt tension extension during acceleration is worth calculating though as this could be a non-negligible number with those beefy servo motors.

I tried a few tensioner pulley options on mine before looking at a bunch of the CNC builder forums and seeing that everyone else just uses slots for the motor position and a screw to push the motor out to get tension with fewer components. If you build your own tensioner you need to be careful to ensure that it’s not going to deflect when it’s on the drive tension side of the belt vs. the slack side and create backlash that way.

1 Like

I can get bar stock but I don’t have a lathe so I’m not confident about boring the inner diameter. Concentricity on the pulleys is important because if they’re a little bit eccentric, the effective diameter is constantly changing as they spin. Plus, you need to hold a tight tolerance on the hole to get a decent fit on the shaft.

Ahhhh, right. There’s going to be a slack and not-slack side…

In that case yeah, I think I’ll see if I can build a slider into my motor mount and adjust it with a grub screw.

They’re not super beefy, I bought them mainly for precision. Rated torque is 1.27Nm, peak torque is 4.45Nm. Looks like the smallest pulley I’ll be able to stick on it is ~30mm diameter, so peak force exerted on the belt would be about 300N? I don’t see any numbers for belt stiffness though…

1 Like

Hmm, it looks like 2GT belts might not make a lot of sense for a CNC. The rated torque for a 45-tooth 6mm-wide 2MGT sprocket at 1000 RPM is 1.58Nm. 9mm is 1.64x that, so 2.59Nm.

However a 3MGT sprocket with the same parameters save for 30 teeth (matching diameter) is 2.44Nm for 6mm or 4Nm for 9mm. That’s 1.54x the torque in the exact same size.

But that brings the original problem back: it’s nigh on impossible to find a 3GT pulley with a 14mm bore…

If only I could just use the regular 5M profile pulleys. I can get those anywhere.

1 Like


For belt stiffness you’re looking for the tension modulus, reported as extension for a given tension force

I found that after reading the excellent thread where the real janderson used some real equipment to measure some belt samples

In my simple deflection model for the Shapeoko (the spreadsheet I uploaded to the deflection backlash and vibration thread) I use a belt tension modulus for 9mm kevlar / steel cored belt of 0.019% extension per kg ‘force’. I apologise for this horrid muddled unit but I was converting back and forth to lbs force which our American friends seem determined to stick to :wink:

The Gates docs do, I think, walk through some calculation of belt stretch on the power (tension) side, you work out the belt distance under tension, not touching the pulley teeth, then apply the tension force and work out the extension, if you start with 0.002% per Newton that’s Kevlar / Steel GT2 9mm.

If we take the Gates sheet data, 5.7 lbs force for 0.1% extension, that’s
5.7 * 0.454 = 2.59 kg
2.59 * 9.8 = 25.4 N
0.1% / 25.4N → 0.0039% belt extension per N tension

Which is about right for the polyester core vs. Kevlar or steel core.

Does that help?


So if my numbers are right:

  • 45-tooth 2MGT sprocket = 90mm circumference = 29mm diameter = 14.5mm radius
  • 4.45Nm x 14.5mm radius = 307N = 69lbf
  • From your sheet
    • 45-tooth 9mm 2MGT sprocket has a working tension of 63lbf, a breaking strength of 190lbf and a modulus of 5.7lbf per 0.1% elongation
  • 69lbf/5.7lbf = 12.1, 12.1 * 0.1% = 1.21% elongation
  • The belts I’m looking at are around 180mm long, the section under tension being around half that at 90mm, so the amount of stretch I’m looking at is 1.21% of 90mm = 1.089mm
  • 1.089mm is 0.00605 of a rotation of a 90-tooth sprocket (this would be on the ballscrew for the 2:1 reduction)
  • 0.00605 of a rotation of a 5mm-pitch ballscrew is 0.03mm

So with the 9mm 2MGT belts I’m looking at ~30µm of peak stretch-induced imprecision for my application.

With 9mm 3MGT belts, I’d reduce the belt-induced imprecision to about 20µm, with 15mm 3MGT belts, to 10µm, with 9mm 5MGT belts to 6.24µm and with 15mm 5MGT belts to 3.36µm.

In the thread you linked there’s talk of much better performing steel-cored belts and talk elsewhere about Kevlar/aramid but Gates doesn’t seem to manufacture steel/kevlar-core GT3 belts, so how does one obtain them?

Bringing us back to the original topic of the thread :smiley:

1 Like

I still haven’t found anything new to report. @LiamN, you mentioned that you have done a reduction pulley system, did it really improve anything? I am looking into doing something similar to gain resolution while dropping as much micro-stepping as possible.

1 Like

It seemed to improve the stepper deflection under loads applied along the X rail. I only did it initially because I wanted to use larger pulleys for the 3MGT belts.

Once it was assembled I found that the Hiwin linear rails had a lot more friction than the V wheels though and I was trying to reduce backlash so I geared down a little more.

I would not recommend this as a good trade off for effort vs. benefit, as an experiment it was a success, I learned things, I learned I wouldn’t do it again :wink:

Trying to get extra resolution with the belt system is, IMHO, not a good use of your efforts, if you really want that, go linear rail and ballscrew and accept that you’re changing the motors, controllers and PSU at the same time.

I have less deflection now and can cut faster but I have no more precision than the machine when it was stock.

Yep, I get the same so far

I think I fell in this hole too. It’s the unsupported straight length of belt whose ends are the tangents to the pulleys that stretches, the teeth of the pulley hold the belt to the nominal pitch where it is wrapped on the pulley.

If anyone wants the Fusion models for the 2MGT and 3MGT pulleys let me know.

You have about 36mm (a bit less) on the 45T pulley (under 1/2 circumference) plus about 107mm on the 90T pulley. Drawing that out I get about a 240mm 120T belt of which about 48mm is unsupported between the pulleys on each side.

Then apply the 1.21% elongation to that 48mm section → 0.6mm

Carrying on with your numbers,

0.6 / 180 = 0.0032 rotation
0.0032 * 5mm pitch = 0.016mm

So ~16µm of peak stretch-induced deflection when your servos are loaded to their breakaway (or static stall?) torque during acceleration on the machine.

You could go to the 3MGT, they are certainly heavier duty belts, especially if you go to 15mm width.

Remember though that this is not your sole source of deflection in the machine, we are calculating the peak deflection under dynamic loads right at breakaway torque for your servos and you should not be doing anything remotely close to a finishing pass at those forces.

You are probably looking at 1/10th this force as a dynamic ceiling for a low-engagement finishing pass with sensible accelerations on the machine so you’re down to ~1.6µm just by disengaging Cleetus mode and not sending it.

I would be willing to start putting out the plate, cutlery and ketchup to assist me in eating my hat if the rest of your machine doesn’t have at least that deflection in it and we may well be below the accuracy of the ballscrews by this point too.

I have a roll of the Powge 15mm steel reinforced 2MGT here, but that’s open, not the closed loop. You can buy the 3MGT Gates belts, such as this one about the right length


I got all the pulleys from Powge on AliExpress or eBay, they seem perfectly decent as well as cheap enough to modify on a lathe.

1 Like

Ok when’s the tutorial going to be published?!? haha

1 Like


I was hoping that Lucas would do all the hard work and then make a video…



I am really just trying to improve the surfaces on faces that aren’t parallel to X or Y. Right now there is a visible stair-step pattern even in very light finishing passes. I also wonder if the reduced torque at micro-steps is part of the issue. Hence my desire to do a pulley reduction since it is relatively low backlash and not very expensive compared to ballscrew. I won’t be going ballscrew on this machine. The cost and effort to change everything out and do it the right way is high. At that point I would rather save the money for a little longer and get a tormach 440 which is way more rigid and capable. I will be converting to linear rail relatively soon though.



I have that issue too, curved faces on Aluminium have all the staircase on them. I’m not entirely sure if that’s controller or motor limits though. There was a long thread about wall finish and step size a little while ago that left me with more questions than answers.

Bear in mind that as you gear down the steppers you’re going to lose rapids speed on the machine, without more Volts the motors won’t go any faster so the trade off, in addition to all the work of belts for the steppers, is to lose speed to gain precision.

Thinking about it, if this were my goal, I’d start with replacing the steppers with 0.9 degree step motors (as Will suggested recently) to halve the step size. From there I’d replace the controller with a better microstepper which can cope with more PSU volts before trying to get enough step down in the belt transmission to make that difference.

1 Like

I started a new thread on this over here:

Switching the steppers is certainly a lot less effort than gearing down the existing stepper, not as much excuse to machine stuff though…


Don’t tempt me… I do really like how the whole idler/pulley/idler chain is supported on both ends.

1 Like

Well if you do decide to do something like that I’ll happily share the CAD as well as the somewhat irritating aspects of my design when it comes to assembly and operation so your 2.0 can be better.


I may have found the perfect project for when your Nomad upgrades are done

Now that is small scale accuracy