After a gear set leaves the inspection bench and is fitted into a gearbox, backlash is rarely exactly what the drawing predicted. Machining stack-ups, bearing clearances, housing bore tolerances and thermal gradients all shift the real flank clearance away from the design target. That final gear backlash adjustment on the assembly floor is what converts a set of dimensionally correct parts into a quiet, durable mesh. Get it wrong and the pair either binds under warm-up, hammers during reversal, or runs with an oil film that a loose mesh can no longer hold. Backlash on paper is a range, not a number; the assembly step is where the pair lands inside that range, and the installer is the last person who can tune it without pulling the box apart again.
The most common in-service method is the shim pack between the bearing outer race and the housing shoulder, or between a motor flange and the gearbox adapter. Adding shims pushes the pinion deeper into the wheel and reduces backlash; removing them opens the mesh. Shims work best on parallel-shaft spur and helical sets where one shaft is fixed and the other can be shifted relative to its bearings, and on worm gears where the worm position is moved axially. Start with the manufacturer's baseline pack, assemble dry, measure, then add or remove 0.05 or 0.10 millimetre layers one at a time. Always retorque the housing bolts before remeasuring, because clamp-up itself changes the centre distance slightly. Shims are cheap, repeatable and leave no running wear surfaces behind, which is why they remain the default on reducers, conveyor drives and elevator gearboxes.
Where shims are impractical because the housing is a rigid casting with no axial float, designers use an eccentric bushing or an adjustable bearing cartridge. The eccentric sleeve sits between the bearing outer ring and the bore; rotating the sleeve by a few degrees moves the shaft centre along a small arc, opening or closing the mesh in tenths of a millimetre. Lock nuts or set screws hold the sleeve at the chosen angle. This method is standard on small gear motors, machine tool heads and some tractor transmission layshafts. It gives finer control than shimming on compact units, but it is sensitive to installation: the sleeve must be fully seated and the lock tightened in the correct direction, otherwise vibration will walk the setting during the first hours of operation. A useful trick is to scribe a line across the flange and the housing before adjustment, so that the original factory setting can be restored if the tuning goes the wrong way.
It helps to remember that moving the centre distance does not only change backlash. Close the centre distance and the teeth run on the inner portion of the profile, the root fillet carries more contact and the tip clearance shrinks; open it and the contact patch shifts outward, the teeth run thinner, and backlash grows faster than most engineers expect. For spur gearing, a 0.01 millimetre change in centre distance typically changes backlash by roughly 0.02 to 0.03 millimetre, because backlash scales with the pressure angle on both flanks. Tip clearance is the other limit: once it falls below about 0.25 module, you risk interference between the tip of one gear and the root fillet of its mate under load. So the tuning loop is not reduce backlash until it disappears, but walk the mesh inward until backlash is at the lower spec, then verify the contact patch with layout compound.
Measuring backlash on the bench is done three ways, depending on access. A dial indicator on the pitch circle of the fixed gear, with the mating gear held still, gives the cleanest reading of tangential lash. A feeler gauge slid between tooth flanks works through an inspection cover when the pair is already bolted in. The lead-wire method remains the fastest for large, slow gears: a soft lead wire is nipped between meshing teeth, and the flattened thickness at the non-contact side is the backlash. Take at least four readings around the circumference, because a single reading hides runout. Record the average and the spread; if the spread is more than thirty percent of the average, suspect eccentricity or a bent shaft before touching the shims. Digital dial indicators with a peak-hold function make the measurement faster because the installer does not have to watch the needle while wiggling the gear by hand.
Once the backlash is in spec, do not hand the gearbox straight to production. Run it under no load for an hour, then at a quarter load for another hour, listening for roughness and watching the sump temperature. Touch the housing by hand; it should warm gradually but never become too hot to hold. Recheck backlash after this break-in, because seating of bearings, run-in of tooth surfaces and minor thermal contraction can move the value by a few hundredths of a millimetre. Acceptance should be based on three things at once: backlash within the agreed band, contact patch covering at least sixty percent of tooth height and sixty percent of face width on the working flank, and oil temperature stabilising within the manufacturer's limit under rated load. A box that passes backlash but howls at speed is not finished.
Adjustment has a limit. If even the thickest shim pack or the eccentric sleeve at its end stop leaves backlash above the upper specification, the problem is no longer on the assembly floor. It points to undersized tooth thickness on one member, an oversized housing bore, a warped casting, or a centre distance machined beyond drawing tolerance. Forcing the mesh closed at that stage by over-tightening bearings or shimming the wrong flange will only weld the pair on the first run. At that point the right move is to stop, photograph the contact pattern and the measured values, and send them back to the gear manufacturer with the original drawings. A replacement pinion at the correct tooth thickness, or a bushing machined to the offset that the box actually needs, is cheaper than a gearbox that comes back on the warranty truck.