Field note ·
Belt tracking is a geometry problem, not an adjustment problem
Every plant has a conveyor that runs off to one side. Almost every attempt to fix it adjusts an idler. Almost none of those attempts work, because tracking is decided by the frame, the pulleys and the load - not by the idlers.

Why belts run off to one side
A belt runs off because the sum of the forces across its width is not symmetrical. Something in the system is pulling it one way: a pulley that is not square to the frame, a crown that is worn unevenly, a frame that has settled, a belt that has stretched more on one edge, or a load that arrives consistently off-centre. Adjusting an idler changes the local path of the belt, which is why it appears to work for a week and then stops working.
The tell is in the pattern. A belt that runs off in the same place regardless of load is a geometry problem. A belt that runs off only when a certain product is running is a loading problem. A belt that runs off after a stop and comes back after ten minutes is a temperature or tension problem. Those three patterns have three different fixes and the same symptom.
Before touching anything, run the conveyor empty for a full circuit and watch the belt edge against the frame at four points: after the head pulley, after the tail pulley, at the middle of the carry side and at the middle of the return side. The point where the deviation starts is where the cause is. A belt that is centred at the tail and off at the head is a head pulley or a take-up problem, not a mid-frame problem.
The four measurements that find the cause
Square the frame. Measure the diagonal from head to tail on both sides; the two diagonals should be equal within a few millimetres on a short frame and proportionally on a long one. A frame that is racked by 10 mm will throw a belt on a long centre distance, and no idler will correct it.
Level the pulleys against the frame, not against each other. A head pulley that is square to a frame which is itself skewed still throws the belt. Use a laser or a taut string along the frame edge, then measure the pulley face to that line at both ends.
Check the crown. A crowned pulley steers the belt by generating a tension difference across the width; a worn crown, a flat spot or a crowned pulley that has been recovered without re-crowning stops steering. On a tracking-critical conveyor, a crown of about 1-2 mm per 100 mm of pulley width is typical, but the number matters less than the crown being even.
Check belt condition at the edges, not in the middle. A belt that has been running off for months has a frayed or compressed edge on one side, which changes its length on that side, which perpetuates the drift. At that point the geometry can be perfect and the belt will still run off. The belt is now part of the cause.
What adjustment actually does
Moving an idler changes the belt path locally. On the carry side, shifting a troughing idler forward on the side the belt is drifting toward increases tension on that side and pulls the belt back. It works, briefly, because it is fighting the geometry with a local force rather than removing the cause. The belt then stretches, the idler settles, and the correction decays.
The legitimate uses of idler adjustment are small and late: a belt that is otherwise geometrically correct but has a slight, stable bias, or a belt that needs a training tweak after a splice. If the correction needed is more than a few millimetres of idler movement, the problem is somewhere else.
Training idlers and self-aligning return rollers are useful devices with a bad reputation, because they are usually installed to compensate for a fault nobody diagnosed. A self-aligning idler on a square frame and a straight belt will do nothing. The same idler on a skewed frame will oscillate. Fix the frame and the self-aligning idlers become unnecessary.
Loading, and the illusion that it was fine last week
A belt that was fine last week and is off this week has usually changed its load, not its geometry. A new product with a different footprint, a chute that now drops product closer to one edge, a hopper that discharges unevenly, or an operator who now feeds from one side - each shifts the centre of gravity of the load and therefore the tension distribution.
The check is to run the belt empty and then with a representative load, and compare. If the empty belt tracks and the loaded belt does not, the fix is in the loading: a centring chute, a skirt that guides the product, a feed that is centred rather than aimed. This is the cheapest fix in the whole list and it is the one most often missed because the belt is blamed first.
Belt speed and loading interact. A belt running near its capacity with an off-centre load will drift more than the same belt at half load, because the material's resistance to being carried straight is higher. If the drift appears only at high rates, the answer may be a wider belt or a slower one, not a straighter frame.
The maintenance that actually prevents drift
Daily: look at the belt edge at the head pulley. A 10 mm drift that appears in a shift is a developing fault; a stable 10 mm that has been there for a year is probably the belt's own bias.
Weekly: clean the return side and check the take-up travel. A take-up that has run out of stroke is a belt that has stretched past its allowance, and a stretched belt tracks differently along its length. Material build-up on a return roller is the single most common cause of a sudden drift after a weekend.
Quarterly: re-check the four measurements. Frames settle, foundations move, pulleys wear. A conveyor on a slab that has been loaded by a neighbouring machine will not stay square, and the quarterly check catches it before the belt edge is destroyed.
Annually: inspect the pulley lagging and the crown. Worn lagging changes the effective diameter, which introduces exactly the kind of asymmetry that tracking cannot tolerate.
What to specify on the next line
Specify tracking as a performance requirement, not a feature: the belt must run centred within a stated tolerance over a stated load range for a stated period, verified at acceptance. That converts a maintenance argument into a commissioning test.
Ask for the frame tolerance, the pulley crown and the belt specification in writing. A supplier who can answer those three has designed for tracking; a supplier who answers with the idler type has not.
For the systems themselves, the reference we work to is set out in our conveyor systems range, which lists the frame, pulley and take-up details that make tracking a design outcome rather than a weekly adjustment.
Finally, keep a tracking log. Date, load condition, deviation in millimetres, action taken. After six months the log will show whether the conveyor has a geometry fault, a loading fault or a belt fault, and it will do so without anyone having to remember.
Splices, and why a bad splice mimics a bad frame
A splice that is not square introduces a permanent bias. The belt is a loop, and a joint cut at an angle makes one edge longer than the other by the width times the tangent of the angle. On a 1,000 mm belt, a 5 mm offset across the splice is a bias that no idler will remove, because the belt is now genuinely asymmetric.
The symptom is characteristic: the belt runs centred for most of the circuit and deviates as the splice passes the head pulley, then returns. If the drift is periodic with the belt revolution, stop measuring the frame and measure the splice.
The fix is a re-splice, cut square with a guide, and a check of the belt length on both edges afterwards. Finger splices and vulcanised splices both work; a splice made in a hurry at 2 a.m. does not, regardless of the method.
The economics of tracking badly
A belt that runs against a frame destroys its own edge. The edge is where the fabric or the steel cords are exposed, and once it frays the belt starts to shorten on that side, which increases the bias, which increases the contact. The process is self-accelerating and it ends with a belt replacement that costs ten times the inspection that would have prevented it.
The secondary damage is to the structure: a belt rubbing a frame wears the frame, a misaligned belt wears the pulley lagging unevenly, and a drifting belt spills product that then has to be cleaned by somebody. None of that appears in the maintenance budget as 'tracking'; it appears as structure, lagging and housekeeping.
The arithmetic that justifies the quarterly check is simple: one belt replacement plus the downtime plus the spillage cleanup is usually several times the cost of four inspections a year, and the inspection is a fifteen-minute task with a tape measure and a string.
What to check
- Run empty, then loaded - compare the two before adjusting anything.
- Measure frame diagonals and pulley squareness against the frame, not against each other.
- Check pulley crown and lagging wear; a worn crown stops steering.
- Inspect belt edges - a damaged edge keeps the drift alive after the geometry is fixed.
- Check take-up stroke: a belt stretched past its allowance tracks differently along its length.