Blog · 2026-08-15 · Belts
Belt tension is the most commonly botched adjustment in hobby robotics — partly because "tight enough" is vague, and partly because over-tightening feels like being careful. Here's how to get it right, and how to read what a misbehaving belt is telling you.
A timing belt is a positive drive: its teeth mesh with the pulley's teeth, so in principle it cannot slip. In practice it skips constantly, and the reason is almost always tension. Too little and the belt lifts out of the tooth grooves under load; too much and you overload bearings, stretch the belt and add friction to every move. The correct range is wider than people fear but narrower than guessing allows, and there are two reliable ways to find it.
Think about what happens on the loaded side of a belt. Torque at the driver pulley pulls one span tight and lets the other go slack. If the belt has too little preload, that slack span sags — and near the point where the belt wraps onto the pulley, the sag lets the belt ride up and out of the tooth groove. That's a skipped tooth: a sudden position error, a bang, and a robot that has silently lost its place.
Preload prevents that by keeping both spans in tension even when torque is applied. The requirement scales with torque, which is why a belt that behaves perfectly during gentle moves skips the first time the axis accelerates hard.
This is the practical bench method and it's good enough for essentially every hobby machine. Measure the free span between the two pulleys. Press the middle of that span sideways with moderate finger pressure — roughly the force you'd use to press a doorbell firmly.
target deflection ≈ 1/64 of the span length
100 mm span → about 1.5 mm of deflection
200 mm span → about 3 mm of deflection
300 mm span → about 5 mm of deflection
The belt should feel firm and spring back immediately, with no perceptible sag when at rest. If you can push it far enough to see obvious deflection without effort, it's loose. If it feels like a guitar string and barely moves, it's too tight. The test is crude but repeatable, and its real value is comparative: once you know what a correctly-tensioned belt on your machine feels like, you can reproduce it.
A belt span is a string under tension, so it has a natural frequency — and that frequency is a direct measure of tension. Pluck the span and record the note with any phone tuner or spectrum-analyser app. This is how printer communities standardise belt tension, and it's genuinely more repeatable than finger pressure.
T = 4 × m × L² × f²
T = tension (N) · m = belt mass per metre (kg/m)
L = free span (m) · f = plucked frequency (Hz)
For a 6 mm GT2 belt (roughly 0.0043 kg/m), a 150 mm span at 110 Hz gives about 4.7 N — a sensible figure for a light axis. Rather than chasing an absolute number, the practical use is consistency: tension one axis until it behaves well, note its frequency, and match the others to it. On a machine with two belts that must behave identically — a dual-Y printer axis, a two-sided robot arm joint — matched frequencies are far more reliable than matched feel.
Tension has to be adjustable, because belts stretch slightly during their first hours of use and centre distances are never exactly as designed. Three standard approaches:
Slotted motor mount. The motor slides to increase centre distance, then locks. Simple, effective, and the reason so many motor plates have oval holes. The drawback is that adjusting tension also moves the motor, which matters if anything else is aligned to it.
Idler pulley on a slot or spring. A separate wheel presses on the slack span. Fixed idlers give a set tension; spring-loaded idlers maintain it automatically as the belt stretches and are the most forgiving arrangement for a machine you don't want to keep adjusting. Use a toothed idler if it runs on the toothed side, a smooth one if it runs on the back.
Belt clamp tensioner. On linear axes with open-ended belt, both ends clamp to the moving carriage — often with a screw pulling one clamp. This is the most precise method available and it decouples tensioning from motor position entirely, which is why quality linear axes use it.
Whichever you use, design it in from the start. A mechanism with no adjustment is one where the belt calculator's exact centre distance has to be hit perfectly in the first print — see the Belt & Pulley Calculator for what that number should be.
| Symptom | Likely cause | Fix |
|---|---|---|
| Skips only under hard acceleration | Too loose for peak torque | Increase tension; reduce acceleration |
| Skips at all loads | Too few teeth in mesh | Larger small pulley or longer centre distance |
| Position drifts one direction over time | Repeated single-tooth skips | Tension, then check pulley grub screws |
| Play when reversing | Slack belt or loose pulley | Tension; check grub screw on the flat |
| Belt walks off the pulley | Misalignment; pulleys not coplanar | Realign shafts; add flanged idler |
| Squealing, motor runs hot | Over-tensioned | Reduce tension |
| Frayed belt edge | Rubbing a flange or misalignment | Realign; check pulley flange clearance |
Two of those deserve emphasis. Tooth engagement is a design property, not an adjustment: if fewer than about six teeth are in mesh on the smaller pulley, no amount of tension will stop skipping, and the fix is a bigger pulley or more wrap. And grub screws masquerade as belt problems constantly — a pulley slipping on its shaft produces exactly the symptoms of a slipping belt. Always check that the grub screw seats on the shaft's flat, and use threadlocker.
Belts run true when the pulleys are coplanar and the shafts are parallel. When they're not, the belt walks sideways until it rubs a flange, frays, and eventually climbs off. Sight down the belt run: the belt should sit centred on both pulleys, and the span between them should look straight rather than skewed. On printed brackets, check for the frame flexing under belt tension — a bracket that's square when unloaded and skewed when tensioned is a common and confusing cause of belt walk.
After the first few hours of running (new belts settle and stretch slightly), then whenever behaviour changes. A spring tensioner removes the need almost entirely.
Yes, and most builders end up doing exactly that — but calibrate your feel against the deflection or frequency method first, and re-check with a measurement whenever a mechanism is misbehaving. Feel drifts.
Look elsewhere in the chain: pulley grub screws, shaft couplers, bearing preload, or the mounting brackets flexing. Belt tension only removes belt-related play, and on a well-built axis it's rarely the last remaining source.
New mechanism, new belt, ten minutes of method. Set it loose first and hand-turn the axis through its whole travel, watching the belt seat onto each pulley. Anything that catches, rubs a flange or runs off-centre is an alignment problem, and tensioning will only hide it temporarily. Fix alignment before touching tension.
Bring it up gradually. Tension to roughly the deflection target, then run the axis under power at moderate speed and listen. A correctly tensioned belt makes a soft, even sound; a loose one slaps and occasionally bangs as a tooth skips; an over-tight one whines and the motor runs noticeably warmer. Adjust in small increments — a quarter turn of a tensioning screw is often plenty — and re-test between each.
Then load it deliberately. Command the fastest acceleration the axis will ever see, ideally with the heaviest load it will ever carry, and repeat it twenty times. This is where marginal tension reveals itself, and finding a skip now in a controlled test is far better than finding it mid-job. Finally, re-check after the first few hours of real running: new belts settle and stretch slightly, so the tension you carefully set on day one will have drifted by day two. After that first re-check, a well-built axis usually stays put for months.
Firm, not tight; measured, not guessed; and adjustable by design. Get the geometry right first with the Belt & Pulley Calculator — particularly the teeth-in-mesh figure — and if you're still weighing belt against gears or chain, that comparison is in Timing Belts vs Gears vs Chain.