Dimensions that differ between two axes that should match. Shifts that appear at random heights rather than at a repeatable one. A part that is square in the model and a parallelogram on the plate.
This page is about the drive itself. If what you have is a single dramatic sideways step in a print, start with layer shifting — that page sorts collisions from drive faults, and it is worth doing before you start adjusting hardware.
Why a toothed belt lets go
Force is transmitted by teeth sitting in the grooves of a pulley. That engagement depends on the belt being pulled tightly enough against the pulley that the teeth cannot climb out of their grooves.
Under hard acceleration the tension on the driving side rises sharply. If the static tension is too low, the belt lifts slightly away from the pulley on the slack side, a tooth rides up over a crest, and the belt advances one tooth relative to the pulley. The toolhead is now displaced by the tooth pitch, and nothing in the machine knows.
That is why this fault appears on fast prints and disappears on slow ones, and why it can be intermittent for weeks.
Tensioning properly
The target is firm enough that the teeth stay seated, loose enough that the bearings and the motor shaft are not being loaded continuously. Over-tensioning is a real fault with real consequences — it wears bearings, heats motors and can bow a light frame.
Two methods, both usable:
By ear. Pluck the belt between two pulleys like a guitar string. It should give a clear note rather than a dull thud. The two belts on a corexy machine should sound alike; the two ends of a single belt should as well. Phone applications that read frequency turn this into a number you can record and repeat.
By deflection. Press the middle of a free span with a finger. A properly tensioned belt yields a small amount and springs back sharply. A belt you can push a centimetre out of line is loose.
Tension each axis with the toolhead at the midpoint of its travel, and re-check after the first print — new belts stretch a little as they bed in.
The pulley fault that imitates a loose belt
A grub screw tightened onto the round of a motor shaft, rather than onto its machined flat, grips by friction alone. It holds for a while, then starts slipping under acceleration, and the symptoms are indistinguishable from a slack belt.
Check it directly: loosen the grub screw, rotate the pulley until a screw meets the flat, and retighten. Where there are two screws, one should land on the flat and the other on the round, tightened afterwards.
The diagnostic for a pulley that is already slipping is a paint pen line drawn across the pulley onto the shaft. Run a print and look again — if the line no longer aligns, the pulley has moved and no belt adjustment will help.
Telling a worn belt from a loose one
Belts wear, and a belt that has been over-tensioned for a year wears faster. Look along the toothed face:
- Rounded tooth crests rather than square ones — worn, and it will keep slipping regardless of tension.
- Cracks at the base of teeth, usually where the belt passes over the smallest idler.
- Visible fibres or a shiny, polished tooth surface — the reinforcement is coming through.
- Uneven tooth spacing when the belt is laid flat — the core has stretched permanently.
A belt is a cheap consumable and a slipping one wastes prints. Replace rather than nurse.
Check the idlers too
A toothed belt running over a smooth idler pulley should have the smooth back against the idler; if the machine is assembled with the teeth against a smooth idler, the belt wears rapidly and runs noisily. Idlers that do not spin freely — a dry bearing, a bolt tightened until it clamps the race — add drag and heat, and they chew belts.
When it is not the belt at all
If the errors scale with the size of the part rather than appearing as discrete jumps, look at steps-per-millimetre calibration instead of tension: an axis with the wrong steps value produces parts that are consistently a percentage off, which is a different fault with a different fix.
If one axis is dimensionally correct and the other is not, and the difference is consistent, calibrate rather than tension. Slippage produces inconsistency between prints; a calibration error produces the same wrong number every time.
When to check all of this
After moving the machine, after any belt or motor work, and as routine maintenance every few hundred hours on a printer that runs regularly. Machines that live on a bench and get carried to a table for a demonstration lose tension far faster than machines that never move — which is worth remembering before blaming the last thing you changed in the slicer.