A buyer measures a motor at 58 dB(A). The supplier’s datasheet says 62. Both are correct, and neither is lying. The difference is in the method, because a decibel figure is a reading produced by a procedure, not a property printed on the part. Until the procedure is fixed, two people measuring the same motor are not comparing numbers, they are comparing laboratories. That is why a noise specification written as a bare figure cannot be enforced, and why arguments about 2 dB are usually arguments about the method rather than about the motor.

Sound Power Is a Property, Sound Pressure Is a Reading

The first thing a noise number has to state is which quantity it is. Sound pressure level is what a microphone reads at a point: it falls with distance and rises in a reflective room. Sound power level is what the machine radiates, independent of where you stand. They differ by a constant that depends on the measurement surface, so they are not interchangeable, and neither is more “correct” than the other.

A datasheet that says 60 dB(A) without saying which one, and at what distance, has given a number without units in any useful sense. The convention that resolves it is the one in ISO 1680, the test code for airborne noise from rotating electrical machines, which defines both quantities and specifies the operating and mounting conditions under which they are determined.

The Accuracy Grade Sets the Smallest Difference You Can Argue About

Standards do not promise a true value; they promise a spread. ISO 1680 accepts three grades of measurement, and each carries a reproducibility standard deviation for the A-weighted result: about 1 dB for a precision method such as ISO 3745 in a hemi-anechoic room, about 1.5 dB for an engineering method such as ISO 3744, and about 3 dB for a survey method such as ISO 3746.

Those figures are the whole argument. A 1.5 dB standard deviation means two competent laboratories measuring the same motor can land roughly 3 dB apart within two standard deviations, with nothing wrong at either end. On a survey-grade test the same span is nearer 6 dB.

So a purchase limit tighter than the method can resolve is unenforceable. If the acceptance window is 2 dB and the test is survey grade, the test cannot tell a compliant motor from a non-compliant one, and any dispute will be settled by whoever is more persistent. The practical rule is that a limit should be at least as wide as twice the reproducibility of the method written next to it, and grade 2 is the sensible default because grade 3 is often too coarse to compare products at all.

Background Noise Is Where Most Factory Tests Fail

Every reading contains the room. The standard handles this with a correction, K1, applied when the motor and the background together are louder than the background alone by a stated margin, and the correction is only valid down to a floor.

Grade 2 requires the motor to be at least 6 dB above the background, with 15 dB preferred, and the resulting correction is capped at 1.3 dB. Grade 3 allows only 3 dB of margin, with a correction up to 3 dB. Grade 1 wants 10 dB and caps the correction at 0.4 dB.

The size of the correction follows the margin: about 1.3 dB at 6 dB of margin, 0.5 dB at 10, and 0.14 dB at 15. Read those against the reproducibility figures above. At the minimum margin the standard allows, the correction alone is as large as the method’s own uncertainty, and the measurement is at the edge of usability.

This is the part most factory measurements get wrong, not through carelessness but because a quiet room is expensive. A motor at 58 dB(A) measured in a room with 52 dB(A) of background is exactly at the 6 dB limit, carrying a 1.3 dB correction and a wide uncertainty. The honest options are a quieter room or a survey-grade claim, not a tighter-looking number obtained by ignoring the margin.

Mounting and Load Change the Number as Much as the Motor

A small motor bolted rigidly to a steel plate radiates through the plate, and the plate can be louder than the motor. ISO 1680 addresses this by specifying how the machine is mounted and loaded during the test, which is why a figure taken on a compliant test bench and a figure taken in the customer’s housing are genuinely different measurements rather than one accurate and one sloppy.

The same applies to operating point. A motor at no load and rated speed is quieter than the same motor at rated torque, and much quieter than one running at a speed where a gear mesh frequency lands on a panel resonance. A noise figure measured at no load describes a condition the machine never sees.

What This Means for an Acceptance Limit

An enforceable criterion therefore has to carry the method, not just the number: the quantity and weighting, the accuracy grade and the standard it comes from, the mounting, the speed and load, the measurement distance or surface, and the background level at the time. Without those six, the figure is a claim. With them, it is a test anyone can repeat.

Where the limit is also meant to capture how the noise sounds rather than how loud it is, a separate tonal criterion is needed, because a single gear mesh tone can be the most objectionable feature of a machine that passes its overall level comfortably.

What to Ask For

Four items make a noise claim usable. Ask which quantity is quoted, sound power or sound pressure, and at what distance if it is pressure. Ask which standard and which grade produced it, since that fixes the uncertainty. Ask for the background level during the test, which tells you how much of the figure is the room. And ask how the motor was mounted and loaded, because that is the condition the number describes. With those four, two datasheets become comparable, and a 3 dB difference stops being an argument.