CV joint: constant velocity through an angle

A simple Cardan joint transmits rotation between two shafts that meet at an angle — true — but it lies about the speed: even though the input shaft turns at a steady rate, the output speeds up and slows down twice per revolution. At a few degrees nobody notices; at thirty degrees the speed ratio already swings between cos(30°) ≈ 0.87 and 1/cos(30°) ≈ 1.15 — that is, ±15% — which shows up as vibration, noise, and wear, and at forty-five degrees the amplitude jumps to nearly ±41%. The constant-velocity joint — the CV, for constant velocity — solves exactly that: it delivers the same angular velocity at the output as at the input, whatever the angle, with no fluctuation. It is the wheel-side joint on the front halfshafts of your car: the rotation has to reach the wheel while the wheel steers and rides up and down with the suspension — and there, a swing in speed would be intolerable. Reproducing it in FDM is possible, but it is one of the most demanding joints you will ever print, because its virtue depends on contact surfaces that your printer cannot make smooth on its own.

Why the Cardan oscillates and the CV does not

The difference lives in one plane. In any joint that transmits rotation between two angled shafts, there is a point — or a set of points — where one shaft hands the torque to the other. What decides whether the speed comes out constant is where that transmission point sits relative to the two shafts. The constant-velocity condition is geometric, and it can be stated in a single sentence: the contact point must always lie in the bisecting plane of the two shafts, the plane that splits the angle between them in half.

When the contact lives in that plane, its linear velocity makes the same angle with the input shaft as with the output shaft — that is the symmetry of the bisector — so the velocity component that actually carries torque looks identical from both sides, and that is why the speed ratio is one-to-one at every instant. The simple Cardan does not meet this: its cross forces the two yokes to rotate in planes fixed relative to their shafts, not in the bisector, so the projection of the rotation shortens and lengthens twice per revolution — which produces the angular acceleration you feel as a jerk. The CV is built precisely to pin the contact in that plane, whatever happens to the angle, and to kill that oscillation at the root.

This has a practical consequence before you draw anything: a CV joint is not an "improved" Cardan, it is a different machine. You cannot start from a cross and refine it; you have to start from the bisector condition and build the geometry that holds it.

Tripod and ball: two ways to pin the bisector

There are two classic architectures for sustaining that condition, and it pays to understand how they differ, because they do not behave the same way when printed.

The ball joint (Rzeppa type, the wheel-side one) houses a ring of balls in grooves cut into both an inner part and an outer one. The grooves are laid out so that, whatever the angle, the balls line up in the bisecting plane and that is where they carry the torque. What forces them to stay coplanar in that plane is an intermediate cage, and that cage — not the outer bell — is the part that physically realizes the bisector condition: if it yields, the balls stop sitting in the bisector even if everything else is rigid. It is the cleanest solution kinematically, exactly constant-velocity, and the one that takes the largest angles — real Rzeppa joints reach about 45–50° — at the cost of a groove and cage geometry that is among the hardest there are: three-dimensional curved surfaces whose finish decides whether the joint turns smoothly or grinds.

The tripod uses three trunnions fixed to the inner shaft; on each one rides a roller — supported on inner needles that let it spin freely on the trunnion — and that outer roller rolls and slides inside one of the three longitudinal tracks of the outer bell. The contact stays near the bisector because the rollers travel along the track as they turn, and, as an added benefit, the joint absorbs axial displacement: the shaft can plunge in and out a little without binding. But note the qualification: the tripod is only quasi-constant-velocity. It has a small residual speed error and, above all, it generates an axial force that pulses at three times the rotation frequency — the classic source of halfshaft shudder. It is more forgiving to make and so reaches a smaller angle (on the order of 26°), but it is not "better" than the ball type: it trades that ease for axial vibration. In both architectures the challenge is the same: torque is not transmitted by meshing teeth, but by precision sliding-contact surfaces, and that precision is exactly what FDM delivers least readily.

The sliding contact: what FDM does not provide on its own

Here is the bottleneck of printing a CV. The tracks, the grooves, and the rollers are surfaces over which another surface slips under load, and they depend on two things that layered printing does not provide for free: smoothness and hardness. A deposited bead leaves a stepping — the stair-stepping — on any surface that is not perfectly vertical or horizontal, and that microrelief is precisely what turns a smooth slide into a coarse scrape. A CV track printed "as is" does not turn smoothly: it clatters over the terraces of its own layers.

The first lever is orientation, and it has two sides you must not confuse. One is the contact surface itself: leave it as aligned as possible with the layer plane or with the vertical, where the terraces are smallest, and never at a shallow intermediate angle, which is where they come out most pronounced. The other is the direction of sliding relative to the layer lines: a surface that minimizes the stepping may at the same time leave the lines running crosswise to the motion, and a roller that slides against those lines drags more than if it runs along them. You won't always be able to leave all three tracks in their best orientation at once — they are spaced 120° apart — so prioritize the face that really bears under torque and watch both effects. The orientation of each functional surface matters here as much as in any other moving part; Layer orientation for motion develops it.

The second lever, and the harder one to accept, is not asking the plastic for what it cannot give. Rolling or sliding contact under load is exactly where plastic against plastic wears fast, and where a printed finish will never reach a machined one. The professional solution for a functional printed CV is not to print everything: it is to print the structure and embed the metal contact elements — real bearing balls in the Rzeppa version, steel rollers on needles in the tripod. A steel ball on a well-oriented plastic track rolls far better than a printed ball on a printed track, and it confines the wear to the plastic track, which you can redesign and reprint. Designing the seat for those balls or rollers so they drop into place and stay retained is the same problem of budgeting geometry around a bought component: Embedded hardware: magnets, bearings, and inserts covers it.

Track clearance: neither binding nor rattling

All of a CV's smoothness depends on the clearance between the rollers or balls and their tracks, and it is a tighter balance than that of an ordinary pivot. Too tight and the joint binds: the roller can't slide along the track to hold the bisector, and what should be a fluid turn becomes a struggle against the walls that heats, scrapes, and ends up seizing. Too loose and the joint rattles: each reversal of the torque takes up that play — the backlash — all at once, and the output arrives with a lag and a knock that ruin precisely the speed uniformity for which you chose a CV over a Cardan.

The trouble is that the effective clearance is not the one you model. As with any printed fit, the grooves come out narrower than nominal and the rollers thicker, so a track modeled at zero clearance comes out as interference, bound from the start. You have to open the track on purpose, budgeting your machine's dimensional offset, and do it on the measured dimension, not the nominal one. And since the margin here between binding and rattling is a few tenths, don't trust a table: the real clearance comes from your printer, with your material, in that specific orientation. The method for finding it — printing a series of stepped clearances and testing by hand which one slides without seizing and without rattling — is the same one in Tolerances for moving parts, applied to the track instead of to a round hole.

When it is worth it and how it fails

The CV is the answer when you need smooth rotation between two angled shafts and a Cardan's speed fluctuation is not acceptable: a halfshaft, a precision drive working at an angle, any transmission where the oscillation translates into vibration, noise, or position error at the output. If the angle is small and constant and you don't mind a slight oscillation, a simple Cardan — useful up to about 30–35° — is much easier to print and to make reliable; save the CV for cases where uniformity is genuinely required, because its advantage is paid for in manufacturing complexity.

And when you print it, watch its characteristic failure modes. The first is binding or roughness from poorly finished tracks: if the contact scrapes, it is almost always the layer stepping or a clearance that is too tight, and you address it with orientation and clearance, not with force. The second is fast wear of the plastic contact: a printed roller or track rolling under load wears and rounds off within relatively few revolutions, which is why embedded metal elements are not a luxury but the difference between a demonstration joint and one that lasts. The third, the most insidious, is loss of the constant-velocity condition through deformation: if the Rzeppa's cage yields, the balls stop being coplanar in the bisector; if the bell or the tripod flexes under torque — because the walls are thin or the material creeps over time — the contact point slips out of the bisecting plane and the joint goes back to oscillating like a Cardan — exactly what you wanted to avoid. A CV that starts smooth and vibrates after a few weeks hasn't simply "worn out": it has lost the geometry that made it constant-velocity. Design it stiff where the torque bears — and the cage stiff above all — and treat it as what it is: the most delicate of your joints, one that won't forgive a tenth of a millimeter out of place.

If you are going to combine the printed structure with metal balls or rollers — and for a functional CV you should — the next step is to size those seats well: Embedded hardware: magnets, bearings, and inserts takes you from the bought part to the cavity that receives it.