A bellows coupling is a flexible shaft coupling that joins two rotating shafts using a thin-walled metal bellows — typically multi-ply stainless steel — bonded between two clamp hubs. It transmits torque with zero backlash while flexing axially, angularly, and laterally to absorb shaft misalignment. The bellows replaces the slop and lost motion of jaw or Oldham couplings, which matters in servo and CNC drives where every arc-minute of error shows up at the tool tip. Typical units handle 0.5 to 500 Nm at speeds up to 10,000 RPM with positional accuracy under 1 arc-minute.
Bellows Coupling Interactive Calculator
Vary torque, torsional stiffness, shaft misalignment, and speed to see torsional wind-up and operating-limit utilization.
Equation Used
The calculator treats the bellows coupling as a linear torsional spring: applied torque divided by torsional stiffness gives elastic angular wind-up. It also compares angular misalignment, parallel offset, and speed against the practical limits stated in the article.
- Bellows coupling behaves as a linear torsional spring.
- Zero backlash is assumed; output error is elastic wind-up only.
- Angular limit is 2 deg and parallel limit is 0.25 mm.
- Speed limit is taken as 10000 rpm from the article.
How the Bellows Coupling Works
A bellows coupling does one job: pass torque from a motor shaft to a load shaft without losing a single degree of angular position, even when the two shafts are not perfectly collinear. The bellows itself is the heart of the part — a thin-walled stainless steel tube, formed hydraulically into a series of convolutions, with each ply usually 0.1 to 0.2 mm thick. Two or three plies stack together to raise torsional stiffness while keeping bending stiffness low. The ends of the bellows are welded or soldered into machined hubs, and the hubs clamp onto the shafts using either a split clamp screw or a single set screw. The split clamp version is what you want on a servo — it grips concentrically without denting the shaft.
Why build it this way? Because torque travels along the corrugations as a near-pure shear path, while angular and parallel misalignment loads bend the convolutions like a tiny accordion. That decoupling is what gives you zero backlash coupling behaviour with high torsional stiffness — usually 1,000 to 50,000 Nm/rad — alongside soft response to misalignment. A jaw coupling cannot do both. An Oldham coupling cannot do both. The bellows can.
Get the tolerances wrong and the part fails fast. If parallel misalignment exceeds about 0.2 to 0.25 mm on a typical 25 mm OD bellows, the convolutions yield plastically on every revolution and the bellows work-hardens, then cracks at the weld root within hours. Angular misalignment above 2° does the same thing. If the clamp hub bore is oversized — say a 6.1 mm hub on a 6.0 mm shaft instead of a proper 6.0 mm H7 fit — the hub slips under torque reversal, you lose your zero-backlash promise, and you'll see position error growing over a shift. Get the bore right, get the alignment right, and the coupling will outlast the machine.
Key Components
- Stainless Steel Bellows: The flexible torque-transmitting element, formed from 304 or 316L stainless sheet 0.1 to 0.2 mm thick. The number of convolutions sets the misalignment capacity — a 6-convolution bellows takes roughly twice the angular offset of a 3-convolution unit at the same torsional stiffness.
- Clamp Hub: Aluminium or stainless hub machined to an H7 bore tolerance, typically with a longitudinal split and one or two M3-M5 clamp screws. The clamp must compress the bore by 0.02 to 0.05 mm onto the shaft for full torque grip. Set-screw hubs exist but slip under reversing servo loads.
- Bellows-to-Hub Joint: Either laser-welded (high-end, fatigue-rated to 10⁸ cycles) or soldered (lower cost, limited to about 100°C continuous). The joint is the most common failure point — a poor weld root concentrates stress and cracks long before the bellows itself fatigues.
- Shaft Interface: The bore must match the shaft to within H7/h6. On a 6 mm servo shaft the bore is 6.000 to 6.012 mm — not 6.1, not 6.2. Anything looser and you lose concentricity, which shows up as runout at the load and accelerated bellows fatigue.
Real-World Applications of the Bellows Coupling
Bellows couplings live in any drive train where positional accuracy matters more than cost. They dominate servo and stepper applications because the alternative — a rigid coupling — demands shaft alignment tighter than most assemblies can hold, while jaw and Oldham couplings introduce lost motion that ruins closed-loop control. You see them on ballscrews, encoder shafts, rotary tables, and lab instruments. They tolerate parallel misalignment, angular misalignment, and small axial offset all at once, which is why CNC builders default to them for ballscrew coupling duty.
- CNC Machine Tools: Ballscrew-to-servo coupling on Haas VF series mills and Tormach 1100MX machines, where a typical 25 mm OD bellows transmits 12 Nm at 3,000 RPM with under 1 arc-minute of windup.
- Semiconductor Equipment: Wafer-handling robot joint drives on ASML and Applied Materials platforms — zero backlash is mandatory because the indexing accuracy budget is sub-micron at the end effector.
- Laboratory Instruments: Encoder couplings on Heidenhain rotary encoders mounted to test stand spindles, where a miniature 12 mm bellows handles 0.3 Nm with negligible torsional windup.
- Aerospace Test Rigs: Torque sensor coupling between a Magtrol dynamometer and a turbine shaft on engine component test stands, where the bellows isolates the load cell from misalignment-induced bending.
- Medical Imaging: Gantry drive couplings on Siemens and GE CT scanners — the bellows keeps the rotating ring positioned within the encoder's resolution while tolerating thermal growth across a 1.8 m diameter frame.
- Printing and Converting: Register-roll drive couplings on Heidelberg and Komori web presses where any backlash translates directly to print misregistration in colour passes.
The Formula Behind the Bellows Coupling
The number practitioners actually need is the torsional windup angle θ for a given transmitted torque T, because that windup is the position error you see at the load when the motor commands a move. At the low end of typical operating torque — say 10% of rated — windup is barely measurable and the coupling behaves essentially as rigid. At rated torque you sit at the design sweet spot where the bellows has been optimised for fatigue life. Push beyond rated torque and windup grows linearly but bellows fatigue life drops with the cube of stress amplitude, so a 25% overload halves service life. The formula below tells you exactly where you are on that curve.
Variables
| Symbol | Meaning | Unit (SI) | Unit (Imperial) |
|---|---|---|---|
| θ | Torsional windup angle between the two hubs | rad | ° (degrees) |
| T | Transmitted torque | Nm | lb-in |
| kT | Torsional stiffness of the bellows (manufacturer datasheet value) | Nm/rad | lb-in/° |
Worked Example: Bellows Coupling in a precision optical scanning rig
A vibration analysis lab in Eindhoven is coupling a Maxon EC 60 flat brushless servo to a 16 mm precision leadscrew driving a laser interferometer carriage. The team selected an R+W BK2/60 bellows coupling with a rated torque of 10 Nm and a published torsional stiffness of 13,000 Nm/rad. The carriage normally cruises at 2 Nm load torque, but accelerates briefly at 6 Nm and sees a 9 Nm transient when the carriage hits its end-of-stroke damper. They need to know the windup at each operating point to budget the position loop.
Given
- Tnom = 2 Nm
- Taccel = 6 Nm
- Tpeak = 9 Nm
- kT = 13000 Nm/rad
Solution
Step 1 — at nominal cruise torque of 2 Nm, compute the windup angle in radians:
Convert to arc-seconds for a feel of what this means at the encoder: 1.54 × 10-4 rad × (180 / π) × 3600 ≈ 31.7 arc-seconds. On a 16 mm leadscrew with a 5 mm pitch that translates to about 0.4 µm of carriage position error — well below the interferometer's noise floor, so the coupling is effectively invisible at cruise.
Step 2 — at acceleration torque of 6 Nm:
That's 1.2 µm of dynamic following error during the accel ramp. Still inside the position loop's tolerance for this rig, but you can see it on a high-resolution encoder trace as a brief lag.
Step 3 — at the 9 Nm end-stop transient:
Now you're at 1.8 µm of windup and the bellows is at 90% of rated torque. Run there continuously and fatigue life crashes — fine for an end-of-stroke event a few times per cycle, not fine as a steady-state condition. The sweet spot for this BK2/60 is the 2 to 6 Nm range, and the design is correctly sized.
Result
Nominal windup at 2 Nm cruise is 1. 54 × 10-4 rad, or about 32 arc-seconds — invisible at the carriage. The full operating range runs from 32 arc-sec at cruise, through 95 arc-sec under acceleration, to 143 arc-sec at the end-stop transient, which puts the design comfortably in the linear elastic region of the bellows with margin against rated torque. If you measure significantly more windup than predicted on an actual rig, the cause is almost never the bellows itself — check the clamp hub screws first (an undertorqued M4 clamp on a 16 mm shaft slips at around 4 Nm and adds apparent compliance), then check for a hub bore that was reamed oversize to 16.05 mm instead of held to H7, and finally inspect for a cracked bellows weld root that has reduced the effective ply count from three to two and dropped kT by 30%.
Bellows Coupling vs Alternatives
Pick a coupling by what the application punishes you for. If you cannot tolerate backlash, your shortlist is bellows, disc, or beam. If cost rules and a degree of lost motion is acceptable, jaw and Oldham come in. The table below covers the engineering dimensions that decide the call.
| Property | Bellows Coupling | Jaw Coupling | Disc Coupling |
|---|---|---|---|
| Backlash | Zero | 0.1° to 0.5° (worsens with elastomer wear) | Zero |
| Torsional stiffness (typical 25 mm OD) | 10,000 to 30,000 Nm/rad | 200 to 2,000 Nm/rad (elastomer dependent) | 20,000 to 80,000 Nm/rad |
| Max speed | Up to 10,000 RPM | Up to 8,000 RPM | Up to 30,000 RPM |
| Angular misalignment capacity | 1° to 2° | 1° | 0.5° to 1° |
| Parallel misalignment capacity | 0.2 to 0.4 mm | 0.1 to 0.2 mm | Near zero (<0.05 mm) |
| Cost (25 mm OD, 10 Nm class) | $80 to $250 | $25 to $80 | $200 to $600 |
| Fatigue lifespan | 10⁸+ cycles within rated misalignment | Elastomer changes every 1-3 years under reversing load | 10⁸+ cycles, brittle failure mode |
| Best application fit | Servo and stepper drives, encoders, ballscrews | General-purpose pumps, fans, conveyors | High-speed turbomachinery, dynamometer test stands |
Frequently Asked Questions About Bellows Coupling
Almost always misalignment, not torque. The bellows weld root is the highest stress concentration in the part, and the stress there comes from bending — angular and parallel misalignment — not from transmitted torque. A coupling rated for 10 Nm at 1° misalignment fatigues fast at 2° even if you're only sending 2 Nm through it.
Put a dial indicator on each shaft and measure runout against the other shaft's centreline. If parallel offset exceeds the manufacturer's spec (typically 0.2 to 0.25 mm on a 25 mm bellows) the coupling is being killed by bending fatigue, full stop. Reset the motor mount or use a longer bellows with more convolutions.
More convolutions buy you more misalignment capacity at the cost of length and torsional stiffness. A 2-convolution bellows is short and stiff — good for an encoder coupling where the shafts are aligned within 0.05 mm and you want maximum positional fidelity. A 6-convolution bellows is long and compliant — what you want on a ballscrew where thermal growth and machine flex push parallel misalignment toward 0.3 mm.
Rule of thumb: if your alignment is held tight by design (rigid bracket, machined-in concentricity), pick fewer convolutions for stiffness. If alignment drifts during operation, pick more convolutions to absorb it without cooking the weld.
Probably not the bellows itself, but the shaft fit. As the motor and screw warm up they expand axially and radially. If the clamp hub bore was reamed slightly oversize — or worse, a set-screw hub was used on a hardened ballscrew end — the hub starts to creep on the shaft once thermal expansion lifts clamp pressure. You lose concentricity and gain apparent backlash, which the servo sees as drift.
Check this by marking the shaft and hub with a paint pen at machine-cold and reading the marks at machine-hot. Any rotation between them means the hub is slipping. Replace with a split-clamp hub torqued to spec on a properly H7-bored hub.
You can, but you have to size for the shock peak, not the running torque. Bellows are elastic, not damped — they do not absorb energy the way a jaw coupling's elastomer spider does. A shock load that briefly exceeds rated torque will yield convolutions plastically and you'll see permanent angular set, then rapid fatigue.
For impact-prone applications either oversize the bellows so the peak transient stays below 80% of rated torque, or pair it with a torque limiter upstream. On indexers we generally recommend a disc coupling with a slip clutch over a bare bellows.
Install at a fraction of rated capacity, not at the rated number. The catalogue value is the absolute maximum the bellows will tolerate without fatigue damage — running there continuously means you've used your entire fatigue budget on alignment alone, leaving none for thermal growth, vibration, or bearing wear.
Target install alignment of 0.05 mm parallel and 0.2° angular on a coupling rated for 0.25 mm and 2°. That gives the bellows headroom to absorb operational drift without ever hitting peak stress. A laser shaft alignment tool pays for itself the first time it saves a coupling on a servo.
You're hitting the torsional natural frequency of the motor-coupling-load system. The bellows has a finite torsional stiffness kT and the load has rotational inertia J — together they form a mass-spring system with a natural frequency fn = (1 / 2π) × √(kT / J). When motor commutation harmonics or load disturbances hit fn, the system rings.
Either shift fn by changing coupling stiffness (a stiffer bellows raises it, a softer one lowers it) so the resonance lives outside your operating speed range, or add damping. On servo drives, notch filters in the velocity loop tuned to fn kill the ring without hardware changes.
For most printer Z-axes and hobby CNC ballscrews, yes — but not always for the reason builders think. Steppers don't need zero backlash to position correctly because they're open-loop, but they do need misalignment compliance. A rigid coupling on a slightly misaligned stepper-to-leadscrew joint loads the motor bearings sideways, and you'll kill the front motor bearing within a few hundred hours.
The bellows isolates the motor from that side load. A flexible jaw or Oldham gives you the same protection at lower cost, so unless you're running closed-loop steppers or a precision rig, a $5 jaw coupling is genuinely good enough. Save the bellows for the servo upgrade.
References & Further Reading
- Wikipedia contributors. Coupling. Wikipedia
Building or designing a mechanism like this?
Explore the precision-engineered motion control hardware used by mechanical engineers, makers, and product designers.
