Can You Use Set Screw Pinions on a CNC Machine?

A pinion is a small gear that meshes with a rack or a larger gear. In a set screw version, the hub bore slides over a motor or gearbox shaft and the only thing stopping rotation is friction. One or two set screws run radially through the hub and bite into a flat ground on the shaft. That is the entire torque path.

Three things decide whether that path holds: screw preload, contact area between the screw tip and the flat, and the friction coefficient of the mating steel. Preload comes from a screw that is tightened to its rated torque. Too little and the hub creeps. Too much and the hex socket strips or the screw yields before it ever reaches full clamp load.

Radial clamp load is small. An M5 set screw torqued to about 3 N·m produces a few kilonewtons of point contact on a flat maybe 6 mm wide. Compare that to a keyed joint, where the key carries load across its full side face. The set screw is a friction drive pretending to be a positive drive.

Slip does not announce itself. The pinion rotates a fraction of a degree, then grips again. On a positioning axis that offset shows up as lost motion, and the servo does not know it happened. The encoder is on the motor, upstream of the joint.

Rapid traverse is the first place trouble shows. A G00 move accelerates the gantry or table hard, then stops hard. The screw tip digs a groove in the shaft flat, and each reversal widens it. After a few thousand cycles the screw no longer seats where it was set, and the pinion walks along the shaft.

Contouring loads the joint in a second direction. Cutter force reverses with each change in travel direction, so the hub sees alternating torque rather than steady torque. Alternating load loosens a screw that would hold perfectly under one-way load. Vibration from an interrupted cut speeds that up.

Thermal growth matters on a long axis. A rack that runs 2 m or more grows a few tenths of a millimeter over a warm shift. If the pinion is rigidly clamped to a fixed shaft, that movement goes into the mesh as a change in backlash. A small amount is normal and the preload spring absorbs it. Excess shows up as noise and tooth wear on one flank.

Part quality is the cost. A 0.05 mm slip at a 30 mm pinion radius is roughly 0.1° of lost rotation. On a 5 mm pitch rack that is about 8 µm of position error, and it appears at random. Nothing in the control can compensate for a joint that only sometimes slips.

• 1
  Good fitLow-speed indexing tables, conveyor drives, feeder axes that only need repeatable stop points.
• 2
  Marginal fitLight gantry axes under 20 kg moving mass with low acceleration and short travel.
• 3
  Poor fitHigh-speed contouring, 5-axis heads, heavy Z axes, any axis where lost motion costs scrap.
• 4
  NeverVertical axes without a brake. A slipping joint plus gravity is a crash.

A keyed pinion with a set screw as a retainer is the common middle ground. The key carries torque through its side faces; the screw only stops axial movement. This is cheap, standard, and it removes the slip mode entirely. The trade-off is backlash from the key-to-keyway clearance, usually 0.02–0.05 mm, which you can reduce by fitting the key by hand.

A tapered or split clamped hub is the better answer for a servo axis. The clamp squeezes the whole bore onto the shaft, so friction acts around the full circumference instead of at one point. Runout stays low because there is no screw pushing the hub sideways. Expect to pay more and to need a shaft with the right taper or a plain cylindrical fit.

For rack-and-pinion axes, the pinion is often mounted on a reducer output flange rather than a bare motor shaft. That changes the question. If the reducer output already has a keyed or flanged interface, use it and skip the set screw entirely. Bolt circles and clamp collars handle torque far better than a radial screw.

On our own builds we keep set screw pinions for setup tooling and test fixtures. Production axes get a keyed or clamped hub. The cost difference is a few dollars per joint. The cost of a scrap batch from lost motion is not.

Machine the flat properly. A ground flat, not a milled notch, gives the screw a predictable contact face. Keep the flat at least as long as the hub and about 1.5 times the screw diameter wide. A screw bearing on a curved shaft surface has almost no contact area and will loosen within a shift.

Use a cup point screw only on a hardened shaft. On soft steel a cup point cuts a crater instead of gripping, and the crater grows each time the screw is retightened. A flat or cone point spreads load better on soft shafts. Replace screws at every service, not every other service.

Torque to spec and record it. A torque wrench on the set screw is the only way to know the preload is repeatable. Mark the screw and hub with a paint line so a quick visual check catches movement before it becomes a quality problem.

Check the joint on a schedule. Push the pinion by hand with the motor disabled and watch for rotation relative to the shaft. A dial indicator on the pinion face shows axial walk and runout in one setup. If you see either, the joint has already moved and the flat is damaged.

Not every CNC axis is a servo positioning axis. A bar feeder, a chip conveyor, or a parts catcher does not care about lost motion. It only needs the pinion to stay on the shaft. Here a set screw pinion saves money and installs in a minute.

Setup and prototype work is another good case. When you are testing a gear ratio or a rack length, you want to swap pinions fast. A set screw lets you do that with one hex key. Once the design is frozen, switch to a keyed or clamped hub for the production build.

Low-speed indexing with hard stops is borderline. If the axis drives into a mechanical stop and the motor stalls against it, the joint sees a torque spike every cycle. Add a second screw at 90° and check torque weekly, or accept that the joint is a wear item.

The honest rule: match the joint to the duty. If the axis has to hold position under varying load, use a positive drive. If it only has to turn something, a set screw is enough.