A servo axis on a pick-and-place machine started losing position at the end of every cycle. The encoder read zero at home, the axis moved to a target, and the part placement was off by 0.3 mm. The error was always in the same direction — the axis always stopped short. The customer checked the encoder, the ballscrew, the motor. All fine. The coupling between the motor and the encoder was flexing.
The encoder coupling is a measurement error, not a mechanical part
The encoder is mounted on the motor shaft (or on the ballscrew end). The coupling between the encoder and the shaft transmits the rotation. If the coupling twists under load, the encoder reads a different angle than the actual shaft position. The twist is proportional to the load torque — a servo motor reversing 20 times a minute puts a cyclic load on the coupling. The coupling winds up under load, reads the encoder, and the control system thinks the shaft is where the encoder says. It isn’t.
The failed coupling was a beam-style flexible coupling (helical, 6 mm bore). It’s cheap ($15), it accommodates misalignment, and it winds up under load. The windup at the motor’s rated torque was about 0.1 degree. For a 5 mm lead ballscrew, 0.1 degree at the motor is 0.0014 mm of travel — negligible. But the coupling was 0.3 mm of placement error at the end effector. The error scaled because the end effector was 200 mm from the axis pivot. 0.1 degree at the motor became 0.35 mm at the gripper. The coupling’s windup was the error.
The fix: a torsionally stiff coupling
The beam coupling was replaced with a metal bellows coupling. The bellows coupling has near-zero windup (torsional stiffness 10x the beam type) while still accommodating 0.2 mm of parallel misalignment. The placement error dropped from 0.3 mm to 0.02 mm. The machine placed parts within tolerance again. The bellows coupling costs $60 — $45 more than the beam type. The alternative was a $2000 encoder upgrade that wouldn’t have fixed the root cause.
The rule for encoder couplings: torsional stiffness is the first spec, misalignment capability is second. A flexible coupling that twists 0.1 degree at rated torque is fine for a handwheel, wrong for a servo encoder. The encoder coupling should have a torsional stiffness high enough that the windup at peak torque is less than 0.01 degree. Check the datasheet: the torsional stiffness is usually listed in Nm/rad. For a 5 mm lead axis, you need at least 10,000 Nm/rad on a 6 mm bore coupling.
The coupling types compared
| Type | Torsional stiffness | Misalignment | Cost | Use for |
|---|---|---|---|---|
| Beam (helical) | Low | Good | $15 | Handwheels, light encoders |
| Bellows | Very high | Good | $60 | Servo encoders, precision axes |
| Disc | High | Fair | $50 | Servo motors, moderate precision |
| Oldham | Low | Excellent (parallel) | $20 | Misalignment-heavy, low torque |
| Rigid | Infinite | None | $10 | Only when perfectly aligned |
Don’t confuse encoder windup with backlash
The error was in one direction only, which is the signature of windup, not backlash. Backlash shows up as a reversal error — the axis loses position when it changes direction, in both directions equally. Windup shows up as a one-sided lag under load. The customer had been checking for backlash (they shimmed the ballscrew nut, checked the preload). None of it helped because there was no backlash. The coupling was winding up. If the error is one-sided and grows with load, look at torsional compliance, not clearance.
Also check: the coupling was on the encoder, not the motor. If the encoder is mounted on the motor tail, the coupling error is doubled by the control loop — the servo compensates for the encoder reading, then the actual shaft is where it should be but the encoder says otherwise. The error compounds. Put the encoder as close to the load as possible, with a torsionally stiff coupling, and the windup error disappears.
The encoder coupling is part of the measurement, not part of the drive. A beam coupling that winds up 0.1 degree under load put 0.3 mm of error at the gripper. Replace it with a bellows coupling, check torsional stiffness first, and remember: one-sided error under load is windup, not backlash. The 0.3 mm drift wasn’t the encoder — it was the flexible coupling telling the encoder a lie.
