Gearbox Backlash: How to Measure It and What It Means

Key takeaways

• Backlash is the small rotational play between meshing gear teeth. Every gearbox has some by design; the question is how much, and how fast it is growing.
• It is specified in arcminutes (1 degree = 60 arcminutes) or as millimetres of movement measured at a known radius.
• Typical values: precision planetary units for servo axes run from under 1 to about 3 arcmin at the low backlash end, standard industrial helical gearboxes commonly sit in the tens of arcminutes. The number on its own means little; the trend against the unit's own baseline means everything.
• You measure it with a dial indicator at a known radius with the input locked: rock the output gently both ways and convert the reading to an angle.
• Rising backlash is a wear signal, and it shows up alongside vibration and temperature changes long before teeth fail.

What backlash is and why it exists

Backlash is deliberate. Gear teeth need clearance to accommodate manufacturing tolerances, thermal expansion and a film of lubricant between flanks. A gearbox with literally zero clearance would bind, overheat and destroy itself. So the designer builds in a small gap, and the drive only feels it when torque reverses: the driving flank has to cross the gap before it picks up the opposite flank.

That is why backlash matters most on reversing and positioning axes: servo driven machine tool axes, indexing tables, robot joints. On a conveyor gearbox running one direction at constant load, generous backlash is harmless. On a positioning axis, it is lost motion the controller cannot see.

Units: arcminutes and millimetres at a radius

Catalogues state backlash as an angle at the output shaft, in arcminutes. One arcminute is 1/60 of a degree. Field measurements produce millimetres of movement at some radius, and the two convert directly:

angle in radians = movement / radius, and 1 milliradian = 3.44 arcminutes.

As orientation across the market: low backlash planetary units for servo applications are typically specified in the 1 to 3 arcmin range, standard planetary units around 3 to 9 arcmin, and general purpose industrial helical or bevel gearboxes are looser still, often in the range of tens of arcminutes. Worm gear units are usually the loosest, and adjustable to a point. Always take the specific unit's datasheet as the truth, and note whether the figure applies to the output flange under a defined measuring torque.

How to measure it: the dial indicator method

• 1. Lock the input shaft. Hold the motor with its brake, or clamp the input coupling. All lost motion you measure must come from the gear train, so the input cannot move.
• 2. Mount a dial indicator against a point on the output flange or a lever fixed to the output shaft, at a measured radius from the shaft centre. The larger the radius, the better the resolution.
• 3. Rock the output by hand or with a small, defined torque: take up the play in one direction, zero the indicator, then move to the stop in the other direction. Use a light, consistent torque; heaving on it deflects shafts and housings and inflates the reading.
• 4. Convert. Movement divided by radius gives the angle in radians; multiply by 3438 for arcminutes.
• 5. Repeat at three output positions roughly 120 degrees apart. Gears wear unevenly, and a single position measurement can sit on the best or worst spot of the wheel.

A worked example with real numbers

An indexing table gearbox is checked during a shutdown:

• Dial indicator mounted at a radius of 100 mm from the output shaft centre.
• Total rocking movement between gentle stops: 0.15 mm.
• Angle: 0.15 / 100 = 0.0015 rad = 1.5 mrad. In arcminutes: 1.5 × 3.44 = about 5.2 arcmin.
• The unit's commissioning baseline, measured the same way three years ago, was 2.1 arcmin.

The absolute number still sits inside what many standard gearboxes tolerate. The finding is the 2.5x growth against its own baseline: flank wear is under way, and on an indexing axis it is already visible as positioning scatter. The right response is to shorten the measurement interval, check lubricant condition and analysis, and plan the rebuild on evidence instead of waiting for tooth breakage.

Reading backlash as a condition signal

Backlash growth rarely travels alone. Worn flanks change the mesh dynamics, so expect rising gear mesh vibration alongside it; our reference on ISO 20816 vibration severity zones covers how to judge the overall level. Bearing wear adds its own lost motion, which is one reason to know your bearing internal clearance classes before blaming the gears. Trended together with temperature and oil analysis, a backlash log turns gearbox replacement from a surprise into a planned line item, which is precisely the job of predictive maintenance tools.

The measurement itself is worthless if nobody can find last year's value. Record the radius, the measuring torque, the three positions and the result in the machine's history, so the next technician measures the same way. Machines with a written backlash baseline and interval measurably beat their MTBF siblings that get opened only on failure. If your gearbox history currently lives in a binder, book a short demo and see how Fabrico keeps measurement history per asset.

Common mistakes

• Measuring through the coupling. Coupling play and key clearance add to the reading. Measure as close to the gearbox output as possible, with the input locked at the gearbox, not at the far end of a drivetrain.
• Confusing backlash with bearing play. Radial and axial shaft movement is a different defect with a different fix. Check shaft float separately before rocking the gears.
• Using heavy torque. Elastic deflection of shafts and housing reads as extra backlash. Light, repeatable torque, same every time.
• One position, one direction. Wear is eccentric; measure three positions and always report the total both ways.
• Judging by the absolute number without a baseline. A loose but stable gearbox can run for years; a tight one growing fast cannot.

Frequently asked questions

What is an acceptable backlash for an industrial gearbox?

Whatever the manufacturer specifies for that unit and duty, measured their way. As rough orientation, low backlash servo planetaries run about 1 to 3 arcmin, standard planetaries about 3 to 9, and general industrial helical units are looser. Trend against the unit's own baseline rather than chasing a universal limit.

Can backlash be adjusted out?

On most fixed centre helical and bevel gearboxes, no: growth means wear and the fix is rebuilding. Worm drives and some bevel sets allow limited adjustment, and servo applications use preloaded or dual mesh designs to cancel it at the cost of efficiency.

How often should backlash be measured?

At commissioning to set the baseline, then yearly on critical reversing axes, and immediately whenever positioning accuracy degrades or gear mesh vibration rises.

Does more backlash always mean the gearbox is failing?

No. A generously specified single direction drive can live with it indefinitely. Fast growth, noise on reversal and vibration changes together are what signal active wear.

What is the difference between backlash and lost motion?

Lost motion is everything the output fails to do when the input reverses: backlash plus shaft twist, coupling play and elastic deflection. Backlash is only the gear tooth clearance component.