A gear is the easy part. You can print a gear, check the mesh, hear it run quietly, and congratulate yourself. Then you bolt it onto a motor shaft and the first real load strips the bore in about four seconds. The teeth were never the problem. The problem is the inch of plastic between a polished steel shaft and everything the part is trying to move, and almost nobody designs that inch deliberately.
This article is about the joint between a shaft and a printed part: the hub, the bore, the feature that transmits torque, and the collar that keeps it from walking off. It is the part of a drive train that fails first, fails quietly, and fails in a way that looks like the plastic was too weak when the plastic was fine and the joint was geometry-blind.
The teeth carry the load. The bore carries the torque.
A printed gear tooth sees a bending load at a tooth root, spread over a contact patch, on a part with a fillet designed to soften it. A printed hub sees a much nastier load: torque crammed into a thin annulus of material through a feature measured in tenths of a millimetre, with no fillet to help and no way to redistribute. The tooth is a beam. The bore is a stress riser that happens to spin.
That asymmetry explains why the failure is so abrupt. PLA and PETG have almost no ductility, so once a keyway corner starts to yield, there is no plastic zone to arrest it. The feature shears, the hub spins, the shaft polishes itself round, and the part is scrap. The load path logic in load paths, not shapes applies here with unusual force, because the entire load enters through one small hole.
First, the number most people skip: what torque are you actually transmitting?
Torque is force times radius, which means a printed hub on a small shaft driving a big gear is in far more trouble than the same hub driving a small wheel. Do this arithmetic before choosing a joint, because everything downstream depends on the answer.
| What you are driving | Typical torque at the hub | What that means for the joint |
|---|---|---|
| A knob, dial or handwheel | Under 0.1 N·m | Friction alone is enough. A plain press fit will outlive the printer. |
| A small fan, a lightweight wheel, a toy axle | 0.1 to 0.5 N·m | A D-flat or a pair of grub screws is plenty. |
| A NEMA 17 stepper at its rated holding torque | 0.3 to 0.6 N·m, more when it stalls | Stall torque is the design case, not running torque. Design for twice rated. |
| A geared motor with a 1:50 reduction | 5 to 20 N·m | Do not transmit this through plastic alone. Use a metal hub or a clamping collar. |
| A printed robot arm shoulder | 3 to 10 N·m plus shock | The joint must be metal-to-metal. Plastic is the adapter, not the coupling. |
Two rules follow from that table. Design for stall and shock, not for the smooth number on the datasheet, because a stepper that misses a step and a gearbox that jams both deliver far more than their rated torque for a fraction of a second, and that fraction is when things strip. And remember that a printed part rated for 5 N·m statically is not rated for 5 N·m hot and loaded; the discussion in creep, fatigue and preload is directly relevant, because a loaded hub creeps on its shaft for months before it finally slips.
Friction alone is a bad idea, and the arithmetic says why
Everyone's first instinct is a nice snug bore, relying on interference to grip. Run the numbers and the idea collapses for anything but the lightest loads.
A press fit transmits torque roughly as torque equals friction coefficient times interface pressure times contact area times shaft radius. With a friction coefficient of about 0.2 for plastic on steel, a 5 mm shaft, 15 mm of engagement and an interface pressure you can actually generate in a printed bore without cracking it, you land in the 1 to 2 N·m range at best. And that assumes the bore is perfectly round, which is exactly the thing FDM is worst at.
Worse, the pressure decays. Printed plastic relaxes under sustained load, so the grip you had on day one is a memory by month three. This is the same viscoelastic mechanism that lets a bolted printed joint lose its preload, and it is why a friction-only hub on a constantly loaded shaft eventually spins free.
That leaves the honest conclusion: friction is a fine retainer and a terrible transmitter. Use it to keep a part located, and add a positive feature to carry the torque.
The six joints people actually print, compared
Every option below is legitimate. What is not legitimate is using one of them for a load it cannot carry.
| Joint | How torque gets in | Strength | Where it fails | Use it when |
|---|---|---|---|---|
| Plain press fit | Friction | Low, and it decays | Creep and relaxation | Light loads, non-critical alignment, parts you want to remove by hand. |
| D-flat (D-bore) | Bearing on a flat | Medium | Corners of the flat, where hoop stress concentrates | Motors and shafts that already have a flat. The default for hobby drives. |
| Printed keyway and key | Shear on the key | Medium to high | Sharp internal corners and the key itself | Larger shafts, moderate torque, where you can add material around the hub. |
| Double D or hex bore | Bearing on two or six flats | High for its size | Corner cracking, or the bore going out of round | Knobs, hex shaft drives, printed hex adapters. |
| Split clamp collar | Radial clamping plus friction | High, and adjustable | The screw, or the split line if it is printed in the wrong orientation | You need adjustability, or you are clamping onto a smooth shaft. |
| Metal insert or metal hub | Metal on metal | Very high | Nothing plastic can fix | Real torque. Geared motors, robot joints, anything above a few N·m. |
Why a D-flat fails at the corner and not on the flat
If you print a D-bore and drive it hard, the failure is not the flat wearing away. It is a crack starting at one of the two corners where the flat meets the circle, running radially outward, and splitting the hub. Those corners are stress concentrations, and in a brittle material a sharp internal corner is a crack that has not opened yet.
The fix is the same one that appears in ribs, gussets, fillets and bosses: put material where the stress wants to concentrate, and round the corner you cannot avoid. Two practical moves:
- Extend the flat past the hub face. If the flat stops exactly at the hub, all the bearing load happens at the very edge of the part, on the layer line. A chamfer, or a small extension of the flat, lets the load enter the material rather than the edge.
- Put a filleted boss around the bore. Not a boss sticking out into space; a boss with a fillet at its base, so the hoop stress from the flat has somewhere to travel before it reaches a free surface.
And do not rely on the printer to make a crisp D. An FDM bore comes out slightly out of round, with a seam and a small elephant's foot at the bottom, so a nominally snug D-bore is often loose on the flat at one end and tight at the other. The dimensional budgeting methods in designed in, not sliced out apply directly: give the flat a clearance band rather than a single number, and measure the part before you trust it.
If you print a keyway, design the key to be the cheap part
Printed keys are a good idea used badly. The instinct is to print the key as part of the hub, which makes it a shear tab that either works or tears out a chunk of hub with it. Better: print the key as a separate loose piece, so that when it fails it is a five-minute reprint rather than a new hub.
The geometry rules are unforgiving:
- Keys fail in shear, not in bearing. Size the key by its cross-section at the shear plane, and keep it short and wide rather than long and thin.
- Bottom the keyway with a radius, not a square corner. A square-bottomed keyway in plastic is a stress riser on both sides; a radiused floor spreads the load into the hub.
- Never let the shear plane run along the layers. If it does, the keyway roof delaminates instead of shearing. Orient the hub so the keyway load direction crosses the layers, the same orientation logic used in shell count and wall thickness decisions.
- Make the key slightly shorter than the keyway. End clearance stops the key from acting as a wedge and prying the hub open under load.
Bore clearance is not a tolerance, it is a decision
The most consequential number in the entire joint is the gap between the shaft and the bore, and most of us choose it by feel. Choose it by purpose instead.
| Intended function | Bore versus shaft | Why |
|---|---|---|
| Free-running bearing surface (printed bushing) | 0.2 to 0.4 mm | Needs a film of clearance or it welds itself shut; see bearings you can print. |
| Location only, torque carried elsewhere | 0.1 to 0.2 mm | Tight enough to centre the part, loose enough to assemble warm. |
| Light press fit, retained by friction | 0.0 to 0.05 mm interference | Do not exceed this with PLA; it will either not go on or split. |
| Clamp collar over a smooth shaft | 0.1 to 0.2 mm | The clamp does the work; the bore only needs to locate. |
| Anything transmitting real torque | 0.05 to 0.15 mm plus a positive feature | Snug, but the feature carries the load, not the fit. |
Note the direction of the compromise. As torque goes up you do not want a tighter bore; you want a looser bore and a better feature, because a tight bore under high torque means the shaft is trying to open the hub while the feature is trying to shear it, and the two loads add. This is the same counter-intuitive conclusion that shows up in abrasion and sliding wear, where a looser fit survives longer under grit.
Clamp collars: the joint you can adjust
The split direction decides everything
A split clamp collar is the best printed answer for a smooth shaft with no flat and no keyway, and it has one design rule that decides whether it works: the split must be perpendicular to the layer lines, not parallel to them.
Print a collar with the split lying in the plane of the layers and the two halves are held together by layer adhesion only, the weakest direction in the part. Under clamping load the collar opens along that seam and never grips. Print it with the split cutting across the layers, and the clamping load runs into the strong direction of the print.
Two more details matter. Put the screw as close to the bore as you can without cutting into it, because clamping force is lost to the bending of the collar wall between the screw and the shaft. And give the slot enough width that the collar can actually close; a 0.3 mm slot on a 30 mm collar is decorative, not functional. This is the small-geometry discipline that grip and handwheel design covers from the other side: features sized for the load they carry rather than for how they look.
When a flexible coupling is the right answer
Sometimes the best joint is deliberately soft. If you are connecting a stepper to a lead screw, or a motor to a shaft that cannot be perfectly aligned, a rigid printed coupling transmits every misalignment as a bending load into both the shaft and the motor bearing. A flexible coupling absorbs it.
The printed versions worth knowing:
- Oldham-style coupling. Two hubs with a sliding centre disc. Handles parallel offset well, tolerates almost no angular misalignment, and the centre disc is the wear part. Print the disc from a low-friction material and expect to replace it.
- Beam coupling. A single piece with a spiral cut that flexes. Excellent for small angular and parallel misalignment, but it is a spring, so it stores energy and it will fatigue. Print it in a material with real fatigue tolerance, and keep the spiral cut wide enough to survive flexing.
- Spider or jaw coupling. Two jaws and an elastomer insert. The best compromise for hobby drives: the jaws carry the torque, the spider absorbs shock. Print the jaws in a stiff material and the spider from TPU at a hardness that matches the load, which is the whole subject of choosing flexible filament by Shore hardness.
Flexible couplings move the failure mode from shear to fatigue, and fatigue failures are patient. A beam coupling that has been flexing for six months fails without warning in the middle of a job, the same mechanism described in rattles, hum and fatigue.
Round parts do not print round
Why a 5 mm bore is not 5 mm
This is the quiet reason so many hub designs disappoint. An FDM printer builds a circle out of chords, so a nominal 5 mm bore comes out as a polygon with flats where you wanted contact and gaps where you wanted fit. On a small bore the error is a meaningful fraction of the clearance you budgeted.
Three compensating moves:
- Oversize the bore and use a feature. If the bore cannot be accurate, stop asking it to be the interface. Open it up and let a flat, a key or a clamp do the work.
- Print the bore vertically if you can. A bore printed in the Z direction is a stack of near-perfect circles. A bore printed lying down is an arch that sags at the top. The difference is often larger than the entire clearance budget.
- Ream it, or drill it, after printing. A printed hole is a blank. Running the correct drill through it, or a reamer if you have one, converts a polygon into a hole. It is thirty seconds of work that replaces a great deal of guessing.
And use a hardened nozzle if you drill, ream, or run abrasive filled filament through the same machine. Carbon fibre PETG and glass-filled nylons will eat a brass nozzle long before your hub design matters. A tungsten carbide nozzle is cheap insurance if you print abrasives regularly, and for the H2 class machines the high flow nozzle exists precisely because high-flow profiles and abrasive material are a demanding combination.
Orientation: the single decision that beats every material upgrade
You can double the strength of a printed hub without changing filament, by rotating it on the plate. A printed part is strong in-plane and weak between layers, and every torque joint has a shear plane that either crosses the layers or runs along them.
| Hub printed | Layers relative to the bore axis | Result under torque |
|---|---|---|
| Flat on the plate, bore vertical | Perpendicular to the axis | Best. Hoop stress runs in-plane, keyway walls are solid, bore is round. |
| On its side, bore horizontal | Parallel to the axis | Worst. The hub splits along a layer line, and the bore is an oval. |
| At 45 degrees | Oblique | Compromise: no support needed on the bore, but both failure modes are partly present. |
The clear winner is printing the hub with the bore vertical, which usually means the gear or wheel face is horizontal and needs no support at all. That single choice does more than moving from PLA+ to a filled engineering polymer, and it costs nothing. It is the same conclusion as in shaping a part so it breaks where you want: orientation sets which failure mode you are even eligible for.
Material choice for hubs and couplings
Stiffness transmits, toughness absorbs
Stiffness transmits torque; toughness absorbs shock. A hub needs both, in that order, and the materials that offer both are not the ones people reach for first.
| Material | Why it suits a hub | Watch out for |
|---|---|---|
| PPS-CF10 | Stiff, dimensionally stable, holds torque at temperature | Expensive, needs a hot nozzle and an enclosure, brittle in shock |
| PA612-CF15 | Excellent balance of stiffness and toughness; low moisture pickup for a nylon | Still needs drying, and it is abrasive |
| PA6-GF25 | Very stiff for the money, good creep resistance | Absorbs moisture, so bores move over time |
| PETG-CF | Stiff, cheap, prints easily, good enough for most hobby hubs | Brittle in shock; no good as a spider or a beam coupling |
| ASA and ASA-CF | UV stable if the part lives outdoors, decent toughness | Warping, fumes, needs an enclosure |
| ABS | Tough and cheap, forgiving of shock | Creeps more than the filled grades; poor for sustained torque |
| PETG | The pragmatic default: tough, printable, cheap | Creeps under sustained load; not for preloaded joints |
| High-speed PLA+ and standard PLA+ | Stiff and dimensionally accurate, prints fast | Brittle and heat-soft; keep torque joints away from motors that get warm |
| Matte and silk PLA | Cosmetic surfaces for visible knobs and dials | Additives reduce interlayer strength; not a torque material |
| TPU 95A and TPU 95A high speed | Spider inserts, dampers, flexible couplings | Too soft to transmit torque through a key; use it as the elastic element only |
| TPU 85A and TPU-LW 90A | Soft grippers and dampers where compliance is the point | Not a drive material at all |
If you take one line from that table: for anything carrying real torque, move from unfilled to fibre-filled and accept the nozzle wear, the drying and the slower speeds. This is the same logic that makes spars and hollow sections work; stiffness per gram comes from the material modulus and the geometry together, and a hub is mostly geometry.
Drying is not optional for a torque joint
Wet filament prints with micro-voids and weak layer bonds, and a hub with weak layer bonds is a hub that shears at the feature instead of bearing on it. The difference is invisible on the outside and decisive on the inside. Nylons are the worst offenders and the most tempting hub materials, so run the drying box and then keep the spool dry with a FilaBox drying box, and read moisture by material before blaming a design that was actually printed wet.
If the filament path itself is the weak link, with a spool that drags, a bend that kinks or a run that jerks at the end of the spool, an auto feeder does for the feed path what a good hub does for the shaft: it takes the shock out of the system. Both are the same engineering instinct in different places.
A five-minute test that tells you whether the joint will survive
Mark it, load it, watch it
You do not need a torque wrench or a test rig. You need to know which failure mode you built, and a bench test will tell you.
- Assemble dry and mark the joint. Draw a line across the shaft and the hub. That witness mark will tell you about slip long before you feel it.
- Load it by hand to roughly the working torque. A lever of known length and a luggage scale is enough. Watch the witness mark.
- If it slips, do not tighten anything. Redesign the feature. Slip at hand torque means friction was doing the work, and friction is the part that decays.
- Run it under load for twenty minutes and check temperature. A joint that is quietly slipping gets hot. Warmth at a hub with no load is a diagnosis, not a nuisance.
- Re-check the witness mark after a week of real use. Creep is slow, and the mark is the cheapest instrument you have for it.
Measurement discipline from measured, not guessed applies here too. The bore, the shaft diameter and the D-flat width are three numbers worth writing down, because a joint that has drifted 0.15 mm has a different failure mode than the one you designed.
When the answer is a metal part
Some joints should not be plastic at all, and the honest list is short:
- Anything above roughly 5 N·m sustained, or any joint that sees stall torque from a geared motor.
- Any shaft under 3 mm, where the annulus of plastic is too thin to hold a feature.
- Any joint that must locate precisely and repeatedly, because printed bores wear and creep out of tolerance.
- Any joint near a heat source, where a PLA or PETG hub softens and loses the interference that was holding it.
In those cases the right design is a metal hub or clamp collar, often a cheap off-the-shelf part, and the printed part becomes an adapter that carries the gear or the wheel. That is not a failure of 3D printing. It is the same division of labour that pins, pivots and bushings and threaded connections arrive at: let metal carry the concentrated load, and let plastic carry the shape.
The rules, in order of how much they matter
- Design for stall torque, not rated torque. Double it, then add shock.
- Print the hub with the bore vertical. Nothing else you do is as cheap or as effective.
- Never trust friction to transmit torque. Feature for the load, friction for location.
- Round every internal corner at a keyway, a D-flat or a bore boss. Sharp corners in brittle plastic are pre-cracks.
- Make the key a separate, replaceable part. Shear it, reprint it, keep going.
- Open the bore as torque goes up. Tight bores and high torque fight each other.
- Put the clamp collar split across the layers, not along them.
- Size couplings for fatigue, not just for strength; flexible joints fail months later.
- Drill or ream printed bores when the fit matters, and give the flat a clearance band instead of a single number.
- Move to metal above 5 N·m and stop asking a printed hub to do a clamp collar's job.
The gear gets the attention because it is visible and pretty. The hub is what makes the machine run, and it is a design problem with a short, knowable list of answers. Work out the torque, pick a feature that can carry it, print so that feature is not sitting in the weak direction, and keep the rest of the line quiet with a dry spool and a feed path that does not fight back. The gear will follow.