Selection guide
Choosing Servo or Stepper for a 1% Speed Regulation Pinch Wheel
Short answer
Pick the servo: behind a 45:1 planetary reducer, a velocity-mode servo meets a 1% wheel-speed tolerance with margin, because 200-300 lb-in at 80 rpm needs only 4.9-7.4 lb-in at the motor and the loop corrects load-induced error. A stepper qualifies only when its torque-speed curve clears the load with margin at the required motor speed and a measured ripple test holds the wheel inside 0.8 rpm.
On this page
Build it as a servo axis: a velocity-mode servo behind a roughly 45:1 planetary reducer meets a 1% wheel-speed tolerance with margin, and the sizing arithmetic below shows why. The wheel load — 200 to 300 lb-in at up to 80 rpm on a 6 in wheel — is 189 to 284 W of mechanical power, and after reduction the motor sees only 4.4 to 7.4 lb-in. A stepper is not physically wrong for the job, but it qualifies only if its torque-speed curve clears the load with margin at the chosen motor speed and a measured ripple test at the wheel holds inside the tolerance band. The decision, the arithmetic and the acceptance test follow.
The decision, up front #
Speed regulation is a property of the whole axis — motor, feedback, reducer, wheel and the load's disturbance — not of the drive family alone. For this specification the servo with a planetary reducer is the design to build, for one reason that decides everything else: it measures the result and corrects it. A stepper commands a speed and trusts it. Both can hold average speed inside 1%; only one of them reports when it cannot.
| Servo, ~45:1 planetary, velocity mode | Stepper, direct drive | Stepper, ~10:1 reducer | |
|---|---|---|---|
| Regulation mechanism | Encoder feedback closes a velocity loop | Rotor synchronism with the commanded step rate | Rotor synchronism with the commanded step rate |
| Long-term drift | Corrected by the loop; set by feedback and tuning | None while synchronized | None while synchronized |
| Short-term ripple | Corrected within torque headroom and bandwidth | Detent torque and resonance at low speed | Resonance plus reducer transmission error |
| Torque at the motor (calculated) | 4.4-6.7 lb-in ideal; 4.9-7.4 lb-in at 90% efficiency | 200-300 lb-in continuous | 22-33 lb-in near 800 rpm |
| Overload behavior | Speed droop, then a fault or alarm | Silent lost steps | Silent lost steps |
| Verdict | Meets 1% with margin | Not this axis | Qualifies only after the curve check and the ripple test |
The row that decides it is overload behavior: a servo that runs out of torque announces it; a stepper that runs out of torque miscounts and keeps going. On a feeding wheel, silent miscounting is the failure mode you cannot accept unless you test for it, which is exactly what the stepper path below requires.
Size from the wheel first #
Every later number comes from four inputs taken at the wheel, not at the motor: diameter 6 in (0.5 ft), torque 200-300 lb-in (22.6-33.9 N-m), maximum speed 80 rpm, and the tolerance, 1% of commanded speed. Treat the torque figure as continuous unless you know otherwise, and check whether the process also demands acceleration torque; both stack on top of the steady load, never in place of it.
The arithmetic, carried with its units:
- Tangential pull at the pinch line: F = T / r = 200 lb-in / 3 in = 66.7 lbf, rising to 300 lb-in / 3 in = 100 lbf.
- Surface speed: v = π × D × n = π × 6 in × 80 rpm = about 1,508 in/min = 125.7 ft/min (0.638 m/s).
- Mechanical power: P = T × 2πn / 60. At 300 lb-in (25 lb-ft): 25 lb-ft × 8.38 rad/s = 209 ft-lbf/s = 284 W. At 200 lb-in (16.7 lb-ft): 140 ft-lbf/s = 189 W.
- Motor side, with ratio G = 45 and gearbox efficiency η: T_motor = T / (G × η) = 4.44 to 6.67 lb-in at η = 1, or 4.9 to 7.4 lb-in (0.56-0.84 N-m) at an example η = 0.90. Motor speed is 45 × the wheel speed = 3,600 rpm.
- Motor output power must also cover transmission loss: 189-284 W divided by η gives 210-316 W at η = 0.90.
These are calculated values for these inputs, not measured results, and η = 0.90 is a rule-of-thumb placeholder — take the real efficiency, and the torque the motor must hold continuously at 3,600 rpm, from the datasheets of the exact motor and reducer you shortlist. One sizing caution worth stating flatly: do not shrink the wheel torque before calculating. A shortcut that sizes from a smaller typical load understates the motor and the reducer together, and the axis fails at the machine instead of on paper.
What a 1% tolerance really says #
One percent of 80 rpm is 0.8 rpm at the wheel. That single number is three different requirements, and the drive choice hangs on which one the process means:
- Long-term drift: mean speed error over the interval the process cares about — a fill, a cut length, a batch.
- Short-term ripple: peak-to-peak speed variation inside a short window, one wheel revolution or less.
- Disturbance recovery: how far speed dips, and how long it takes to return, after a load step.
Write the acceptance criterion as all three, with numbers attached: an averaging window (ten wheel revolutions is a workable default), a ripple window (one revolution), and the load step you will apply. A percentage tolerance also shrinks in absolute terms as speed falls — the same 1% is 0.4 rpm at half speed — and it is undefined at standstill, so write a separate stopped-axis criterion if the wheel dwells.
Wheel speed is not material speed. A pinch wheel drives product by friction; slip, contact-force variation and a worn or compressed running diameter all change material speed while shaft speed holds perfectly. If material speed is the controlled quantity, measure the material and close the loop there — the approach behind matching a stepper feed rate to conveyor speed with an encoder — and treat the wheel loop as the means, not the proof.
One distinction before choosing hardware: holding an absolute speed is not the same as holding speed relative to another axis. If this wheel must track a master axis — a feeder, a knife, a conveyor — the specification is synchronization, and the setup described in following a master axis with a stepper traverse is the shape of that solution; absolute regulation is then the wrong number to chase.
Why the servo wins this axis #
A velocity-mode servo measures shaft speed and adjusts torque-producing current to drive the error toward zero. Its regulation is limited by torque headroom, loop bandwidth and feedback resolution — not primarily by how often the loop computes. Update interval is how frequently the math runs; bandwidth is how fast the axis actually corrects, and it is capped by mechanics, feedback counts and tuning. Faster updates without torque or resolution buy nothing, which answers the usual doubt about servo speed regulation directly.
Resolution is a non-issue once reduced. An illustrative 2,048-count-per-revolution motor encoder behind 45:1 reports 2,048 × 45 = 92,160 counts per wheel revolution. In a 0.1 s window at full wheel speed the wheel turns 0.133 revolutions, or 12,288 counts, so one count of quantization is 80 / 12,288 ≈ 0.0065 rpm — better than a hundred times inside the 0.8 rpm budget. That is a calculated example; any mid-resolution feedback multiplied by the ratio clears the bar.
Running the motor at 3,600 rpm puts it where small servo motors are comfortable and where cogging and torque ripple are a small fraction of load torque. When the loop does hit its limit — current saturation on a hard load step — the drive behaves visibly: speed droops, then a tracking fault or alarm fires. It does not miscount quietly. On a used or rebuilt motor, confirm what feedback it actually carries before promising a regulation number; identifying geared resolver and encoder feedback on old servos works through that check.
For this wheel the commitment is easy: a correctly sized velocity-mode servo behind 45:1 holds wheel speed inside 1% with margin. Anything tighter than that is a datasheet-and-machine question, not a topology question.
When a stepper still qualifies #
None of this makes the stepper physically wrong. A synchronized stepper follows its commanded step rate on average with no long-term drift — the rate derives from a digital oscillator whose stability is orders of magnitude better than 1% — and load torque changes the rotor's angular lag, not the average speed, until the torque limit is crossed. The question is where the motor must operate to make the wheel torque:
- Direct drive at the wheel: the motor supplies the full torque continuously, which means a very large frame running slowly, exactly where mid-band resonance lives. Reject.
- Through 45:1 at 3,600 rpm: deep in torque rolloff — hybrid stepper torque falls steeply with speed, and by a few thousand rpm only a small fraction of the low-speed torque remains. Reject.
- Through about 10:1 at 800 rpm: 22-33 lb-in at the motor at the example efficiency, a speed where useful torque still exists. This is the one configuration worth testing.
Then apply two gates. The paper gate: the published torque-speed curve — for that exact motor, drive and supply voltage, because the curve belongs to the combination, not the motor alone — must clear the required torque with margin; 50% over the running load is a common rule of thumb for open-loop steppers. The machine gate: the ripple and recovery test below, passed at the wheel. Fail either and the stepper is out; it is a servo job.
Two more stepper truths to carry. Microstepping smooths excitation and reduces resonance, but it neither adds torque nor makes each microstep an equal, load-independent angle. And an encoder on a stepper means different things in different drives — position verification, stall detection, or genuine closed-loop current control — so read the drive's functional description before treating an encoder-equipped stepper as a velocity-regulated servo.
The reducer does more than multiply torque #
The 45:1 planetary does three jobs at once: it divides torque demand by 45, so the motor is small; it divides reflected wheel inertia by 45² = 2,025, so the wheel looks like almost nothing to the loop; and it multiplies feedback resolution by 45, as the encoder arithmetic showed.
It also adds error of its own, and the part that gets missed is that transmission error and eccentricity at the output reach the wheel undivided, one for one. Convert a datasheet angular error into speed ripple like this: a sinusoidal transmission error of A arc-minutes at full wheel speed appears as a peak speed ripple of 2π × 80 × A / 21,600 ≈ 0.023 × A rpm. One arc-minute is 0.023 rpm; five arc-minutes is about 0.12 rpm. A planetary in the few-arc-minute class sits comfortably inside a 0.8 rpm budget, but take the actual figure from the datasheet of the exact ratio you buy, and remember that wheel eccentricity and contact-force variation add ripple at the same one-for-one rate — which is why the acceptance test watches ripple phase against wheel rotation. When the ratio itself is still open, the page on torque-speed trade-offs in planetary and harmonic drives compares those drivetrains on exactly this axis of choice.
The last joint in the chain is the reducer output to the wheel shaft: put a flexible coupling there, aligned to the runout you can afford, because coupling eccentricity reaches the wheel undivided too.
Prove it at the wheel #
Before the test, one safety point: a running pinch wheel is a nip point. Test with guards in place and the speed sensor on a free shaft — never a handheld tachometer at the pinch line — and take guard and interlock decisions through the machine's risk assessment; ISO 12100 structures that assessment, from hazard identification through risk estimation to risk reduction. Lock out the drive before touching anything on the shaft.
The protocol:
- Fit an encoder or tachometer to a free shaft geared or belted from the wheel, not the motor encoder alone, and log speed.
- Long-term drift: mean error over ten wheel revolutions, inside ±0.8 rpm at full speed.
- Ripple: peak-to-peak error in a one-revolution window, same band.
- Recovery: apply the worst credible load step; record dip depth and return time, and agree both with the process owner.
- Phase check: ripple that repeats once per wheel revolution points at the wheel or coupling — runout, eccentricity, contact force — not at the drive. Fix mechanics before touching gains.
- Thermal: repeat the drift measurement at the end of a warm run. A machine that holds 1% cold and 2% hot has a mechanical or thermal problem no drive retune will fix.
- If material speed is the controlled quantity, measure the material independently during the same test.
A stepper passes or fails this protocol exactly as a servo does. That symmetry is the point of writing the criterion before choosing the hardware.
The decision by case #
New axis, this specification — 6 in wheel, 200-300 lb-in, 1% tolerance: build the servo. Continuous motor torque at least the calculated 4.9-7.4 lb-in plus acceleration, velocity mode, 45:1 planetary with datasheet efficiency and transmission error in hand. It meets the tolerance with margin and reports when it cannot.
Stepper forced by budget or by what is already on the shelf: about 10:1 reduction, motor near 800 rpm, curve margin of 50% over the 22-33 lb-in requirement, then the full wheel test. Pass both and the stepper is a legitimate answer; fail either and stop spending on it.
Material speed is the controlled quantity: close the loop on the material with its own sensor, and move the acceptance number to that measurement; wheel regulation becomes an inner means, not the proof.
Motor already in the building: identify its feedback first — a resolver and an encoder support different loops — then run the same sizing against its datasheet before reusing it here.
Questions asked
Can a stepper motor meet a 1% speed tolerance at the wheel?
For average speed, yes: while the rotor stays synchronized, the wheel follows the commanded step rate with no long-term drift. The failure is lost steps from a load transient or resonance, and the error they leave persists until the axis is re-referenced, because nothing in an open loop corrects it. So a stepper passes only after the curve check and the measured ripple test, and the machine needs a homing step after any stall event.
How much torque does the motor need behind a 45:1 reducer?
Calculated example: 200-300 lb-in at the wheel divided by a 45:1 ratio is 4.4-6.7 lb-in at 3,600 rpm with an ideal gearbox, or 4.9-7.4 lb-in (0.56-0.84 N-m) at 90% efficiency. Size to the drive's continuous rating at that speed using the datasheet efficiency, and add acceleration torque. Acceleration from the wheel itself is small once reduced, since its inertia is divided by 2,025, so check motor and gearbox inertia in the sum.
Does a faster servo loop update rate improve speed regulation?
Not by itself. Update interval is how often the loop computes; bandwidth is how fast the axis actually corrects, and it is capped by feedback resolution, tuning and available torque. A 1% tolerance at this wheel speed needs torque headroom and enough counts per wheel revolution, not an exotic update rate. If regulation is marginal, the fixes in order are more torque, more feedback counts, then retuning; raising the update rate comes last.
Should the servo drive the wheel directly, without a reducer?
No. Direct drive makes the motor supply the full 200-300 lb-in continuously at 80 rpm, in the speed region where low-speed cogging is strongest, and it passes every wheel disturbance straight to the shaft. The 45:1 reducer cuts torque demand by 45 and reflected inertia by 2,025, which is why the reduced servo is the configuration that meets the tolerance with margin.
Does gearbox backlash affect speed regulation?
Not while the wheel turns one direction under one-sided load: backlash shows up at reversals and load direction changes, not in continuous speed. The ripple terms that matter here are transmission error and eccentricity, which reach the wheel one for one. If the wheel reverses, or product tension can reverse the load, backlash becomes a positioning error you must budget separately.