Backlash

Why does a machinist wind a handwheel past the mark and creep back to it rather than stopping on it?

A machinist moving a table to a dial reading does not wind straight to the number. If they overshoot, they do not simply wind back the few thousandths — they wind well back, then approach the number again from the original direction.

That looks like a nervous habit, or a superstition inherited from an older machinist. It is neither: it is the correct response to a specific property of the mechanism, and a machinist who stops on the number after reversing will produce a part that is wrong by an amount they cannot see on the dial.

Start with how the table is moved. A leadscrew turns in a nut fixed to the table; turning the screw drives the nut, and the nut drags the table. The dial counts screw rotation, and the machinist reads it as table position.

Now ask what has to be true for that reading to be honest. It requires that every degree of screw rotation produce a corresponding movement of the table. And that requires the screw thread to be actually touching the nut thread on the side that is doing the pushing.

Here is the problem. A screw and nut cannot be a perfect fit. There must be clearance between the flanks, or the pair would seize — and beyond the designed clearance there is wear, which opens the gap further, fastest in the middle of the screw's length where most of the work is done. So the thread sits in the nut with a small gap, and at any moment the screw is pressed against one flank of the nut, with the whole gap on the other side.

Follow what happens on a reversal. The machinist turns the handwheel the other way. The screw rotates. The dial counts. But for the first part of that rotation the screw is merely crossing the gap — travelling from contact with one flank to contact with the other — and the table does not move at all. Only once contact is made on the new side does the table begin to follow.

So the dial has counted movement that did not happen. The reading is now wrong by the width of the gap, and nothing on the machine indicates this. That lost motion is backlash.

Now the practice makes sense, and so does its precise form. If the machinist always approaches a position from the same direction, the screw is always pressed against the same flank when they stop, and the gap is always on the same side. The error is therefore constant rather than varying, and a constant offset that is present for every setting simply disappears — it is absorbed when the dial is zeroed against the work. What ruins a job is not the gap itself but a gap that is sometimes on one side and sometimes the other.

That is why the remedy is not "be careful" but "always arrive the same way". Overshoot and you must go back far enough to take up the whole gap and then come forward again, re-establishing contact on the working flank. Winding back the few thousandths only moves the screw within the gap, so the table does not move and the dial lies about it.

Two related practices follow. Cuts are taken so the cutting force presses the table against the loaded flank rather than letting the cutter snatch the work into the gap — the deep reason conventional milling was standard on older machines, and why climb milling a worn manual machine is dangerous. And the gib and screw are adjusted to minimise the gap, not eliminate it, since zero clearance means binding.

Backlash is not wear, though wear makes it worse. A new machine has clearance by design, because a screw and nut with no gap cannot be lubricated or turned. What wear does is enlarge that clearance unevenly along the screw, so backlash on a worn machine varies with position — which matters, because it means a single measured backlash figure cannot simply be dialled in as a correction across the whole travel.

Ballscrews reduce it but do not abolish the reasoning. A preloaded ballscrew can have very little backlash, which is why computer-controlled machines can reverse direction mid-cut without ruining the surface. Even there, the control system usually carries a backlash compensation figure, and the deeper point survives: any mechanism with clearance has a direction-dependent position, and if you cannot eliminate the clearance you must eliminate the direction changes.

Digital readouts change the problem completely, and it is worth being clear why. A readout measuring the table directly — with a scale mounted along the axis — reports where the table actually is, so the lost motion never enters the reading. What it does not do is remove the mechanical looseness: the table can still be snatched by cutting forces within the gap. So a readout fixes the measurement error and not the machining hazard, and a machinist who trusts the display and climb-mills a worn machine has solved the wrong half.

The falsification test. If the mechanism is clearance between flanks, then reversing the handwheel should produce a measurable amount of dial rotation with no table movement at all — readable directly with an indicator against the table — and that amount should be repeatable at a given position and larger where the screw is most worn. If the table began moving immediately on reversal, there would be no backlash and the one-direction rule would be pure ritual.