Gearmotor noise and backlash

Noise has a frequency, and the frequency names the source.

Six things commonly make a small gearmotor loud. Four of them are not the motor. Changing the motor before you know which one you have is how a quiet motor arrives and the complaint stays.

The failure tree

Work down it with a spectrum, not with your ear. Every source below produces a different signature and responds to a different change.

Common gearmotor noise sources and how to recognize each one
Source What it sounds like Where the frequency comes from What changes it
Gear meshA tone, usually a whine, that tracks speed. Often several tones at once from different stages.Teeth on the mating gear multiplied by that shaft's revolutions per second, plus harmonics.Tooth profile and lead error, runout, center distance, misalignment, backlash setting, lubricant, and the number of stages.
BearingA rumble, or a periodic tick that does not track a tooth count.Ball pass frequencies, which are non-integer fractions of shaft speed.Preload, contamination, brinelling from a press or a drop, and the radial load the mechanism applies.
Brush and commutationA broadband hiss with a tone under it, and sometimes crackle.Commutator segments multiplied by revolutions per second.Brush pressure and wear state, commutator surface and film, current level, and whether the drive is switching.
Drive electronicsA steady whine that does not change with speed, or that changes in steps.The PWM switching frequency and its sidebands.Switching frequency, current ripple, and whether the winding or the laminations can move.
Mounting and radiationThe same tone as one of the above, but far louder once installed.Unchanged. The frequency belongs to the source; the mount changes only the level.Bracket stiffness and mass, panel area, fastener torque, and whether there is an isolating element.
Structural resonanceLoud at one speed, quiet either side of it. Often a rattle or a boom rather than a whine.A structural mode of the host, excited when a running frequency crosses it.Speed, stiffness and mass of the structure, and damping. Not the motor.

f_mesh = z × n / 60

f_mesh
gear mesh frequency, Hz
z
number of teeth on the gear at that shaft
n
that shaft's speed, rpm
Compute one of these for every stage. A tone that lands within a few percent of a computed mesh frequency has named its own stage.

f_comm = s × n / 60

f_comm
commutation frequency, Hz
s
commutator segments
n
motor speed, rpm
A published Micro Motion coreless platform states its commutator segment count, so this frequency can be predicted before anything is measured.

How to measure it, so the number survives an argument

A dB(A) figure without its distance, its background and its operating point is not a measurement. Four things have to be recorded with the number.

  1. The instrument

    A class 1 or class 2 sound level meter for the level. A calibrated measurement microphone into an FFT analyzer for the spectrum, because the spectrum is what names the source. A phone application will not give a defensible level, but it will often give a usable spectrum, which is the part that solves the problem.

  2. The distance, stated

    One meter is the convention for machines. For a motor of this size it puts the source too far below the room. Measure closer, 100 mm or 300 mm, and write the distance next to the number. A level without a distance cannot be compared with anything.

  3. The background, measured first

    Turn the unit off and measure the room. If the background is within 10 dB of your reading, most of your number is the room. Within 3 dB, the reading is meaningless. Find a quieter place or accept that you are measuring the building.

  4. The operating point, and both directions

    Noise is a function of speed and load. Measure at the actual operating point, at no load, and at any speed the mechanism passes through, in both directions. A resonance that only exists during the ramp will never appear in a steady-state measurement.

  5. The mounting, twice

    Measure in the real mechanism, then on a soft mount off the bench. The difference between the two is the structure-borne path, and it is often larger than every other effect on the list.

What backlash actually is

One definition, and five other things that get measured and reported as backlash.

Backlash is the angle the output shaft can be turned, with the input held fixed, before the output starts to move. It is a property of the clearance between mating teeth. It has to be quoted with the torque it was measured at, because gear teeth, shafts and carriers all deflect: apply enough torque and you will read a number that is mostly elastic deflection, and it will keep growing with the torque you apply.

Backlash also adds through the train, but not equally. Clearance in an early stage is divided by the ratio still ahead of it before it reaches the output, so the last stage dominates the number at the output shaft and the first stage barely appears in it. That is why a high-ratio gearhead does not have proportionally more output backlash than a low-ratio one, and why fixing backlash means looking at the output stage and the output bearing rather than at the ratio.

Torsional compliance, or windup
The train deflects elastically under torque and springs back. It is reversible and proportional to load, where backlash is a fixed clearance. Measure at two torques: if the number scales with torque, you are measuring stiffness.
Lost motion in the coupling
A set screw on a round shaft, a clearance-fit key, a pinned hub, or a flexible coupling with a worn insert. This is often the single largest contributor, and it belongs to the machine, not the gearhead.
Friction hysteresis
The output does not return to the same place when the torque is removed, because friction holds it somewhere in the band. It reads as backlash on a dial indicator and disappears on a bidirectional plot.
Transmission error
The output does not track the input smoothly through a revolution, even with zero clearance, because of tooth profile and runout. A position error that varies with angle is transmission error, not backlash.
Endplay
Axial float in the shaft. It has nothing to do with angular position, but it is what people feel when they push and pull the output shaft, and it gets reported in the same sentence.
Encoder-to-output error
If the encoder is on the motor, everything above sits between the count and the load. The control loop can be perfect and the output still wrong. See the feedback page.

Spur against planetary

Both are published across the Micro Motion gearhead platforms, and the choice changes noise, backlash, length and torque density at once.

Spur and planetary gear trains compared for a small gearmotor
Property Spur Planetary
Load pathOne mesh carries the whole torque at each stage.Torque splits across the planets, so each mesh carries a fraction of it.
Torque in a given diameterLower, because one tooth pair carries everything.Higher, which is why the highest rated output torques in the catalog are planetary.
Length for a given ratioLonger. Ratio per stage is limited by the center distance the housing allows.Shorter. A planetary stage reaches a higher ratio inside the same diameter.
Bearing reactionThe mesh pushes the shafts apart, so there is a net radial load on the bearings.The planets balance each other, so the sun and output see far less net radial load.
Backlash at the outputGenerally higher for the same ratio and quality grade.Generally lower, and more consistent unit to unit, because the planets average their individual errors.
Noise characterFewer meshes, so fewer tones, but each tone can be stronger and the radial reaction feeds the housing.More meshes at once, so more tones at lower individual level, and better balance into the housing.
Output shaft positionOffset from the motor axis on most arrangements, which can be exactly what a mechanism needs.Coaxial with the motor.
Gearhead platforms published
16
Of which planetary
9
Of which spur
7
Published ratios
177

What to send us

A recording and a spectrum answer more than a paragraph of description does.

An audio recording, uncompressed
WAV rather than a compressed format, because compression discards exactly the quiet tonal detail that identifies the source. Record the ramp as well as the steady state.
The spectrum, if you have an analyzer
With the speed at which it was taken, so the tones can be checked against computed mesh and commutation frequencies.
The level, with distance and background
As dB(A), with the distance, the background level, and the instrument.
The operating point and the duty
Speed and torque at the load, and whether the noise happens at speed, during the ramp, or on reversal.
How it is mounted
A photograph of the installation, and a note on what the motor is bolted to. This is the single most useful piece of context for a noise complaint.
The backlash measurement, with its torque
The applied torque, the measurement point, and whether the reading was taken at the output shaft or at the mechanism.
How many units, out of how many
A rate separates a design level from a lot problem, and a lot problem from one bad unit.

Send the recording and the failed unit

Noise that is already in the field is a failure investigation. The intake and the plausibility review cost nothing, and the plausibility review often shows the teardown is unnecessary.

Motor failure analysis