Where Coaxiality Is Lost: How Workpiece Location and Clamping Affect Micro-Screw Removal

When a removal station misbehaves, attention goes first to the bit, the torque setting or the program. Yet on micro screws, a substantial share of cam-out, head shear and recess deformation traces not to the bit but to a small offset between the bit axis and the screw axis — coaxiality.

Coaxiality here is not an abstract geometric tolerance term; it converts into a lateral force. With a 0.1 mm offset between bit axis and screw axis, the bit no longer applies pure torque when it enters the recess — it carries a lateral component. That component pushes the bit against one wall of the recess and slowly enlarges it; with more offset the bit climbs out of the recess and cam-out occurs; and where threadlocker or light corrosion has raised the torque requirement, the combination ends in a sheared head. At the M0.8 end the recess is only about 0.2 mm wide, so a 0.1 mm offset already consumes half of it.

The error comes from three places: how the workpiece is located, how it is clamped, and how much error the machine's motion chain contributes. The final coaxiality is the combination, and the common field mistake is optimising only one of the three.

Take location first. Three methods dominate on 3C mid-frames and small structural parts.

Contour location uses stops and edges to push the part against two perpendicular datum faces. It is the simplest, fastest to change over and cheapest, but it depends on the accuracy of the part's own outline. Stampings and die castings typically carry ±0.05 to ±0.1 mm profile tolerance, with batch-to-batch drift from tool wear, and that error transfers one-to-one to screw position. Contour location suits parts with a finished outline or loose positional tolerance on the screws.

Pin location passes locating pins through process holes or the screw holes themselves. Repeatability can be very high because pin and hole constrain each other directly; the cost is that clearance must exist, and the clearance sets the accuracy. Using a screw hole as a locating pin has a side effect: the pin occupies a hole, so that screw must be removed last or handled differently. On densely populated parts the pin layout also conflicts with the bit's travel path.

Surface location with vision compensation is the more common combination on micro screws in recent years. The part is flattened onto a datum by vacuum or magnetism, and XY position and rotation are compensated from the screw positions found by vision. This sidesteps profile tolerance because the part edge is no longer the reference; but vision compensates only in-plane error. It cannot compensate height variation in Z or part tilt. If the part sits tilted in the fixture, vision sees the projected centre of the screw, and the bit still goes in offset.

Now clamping. The clamping method decides whether the part deforms under load, and deformation moves the screw holes.

Single-side clamping is the easiest: one cylinder presses the part against a stop. The problem is that the clamping force creates a moment, and thin-wall or elongated mid-frames bow slightly. The bow is invisible in the free state, then releases as soon as the bit applies downward force, and screw positions shift with it. Multi-point even clamping distributes the contact points onto structurally stiffer areas so the resultant of the clamping forces passes near the part's centre of mass, and deformation drops significantly.

Cantilever clamping is another common trap. One end of the part is held and the other overhangs; when the bit works on the overhanging end, the downward force elastically deflects the part, and after each screw the part springs back and the next position has changed. This error never appears on a static dial gauge — only in the cam-out rate of continuous operation.

Clamping force itself should be calibrated rather than applied until tight. Too little and the part moves; too much and thin walls deform. A workable approach uses a clamping circuit with pressure regulation and pressure monitoring, with the force written into the changeover recipe and bound to the product model.

The third source is the machine's motion chain, and it is the most often overlooked. XY positioning accuracy, Z perpendicularity, spindle radial run-out and the repeatability of the bit chuck all accumulate. Spindle run-out is routinely underestimated: bits have manufacturing tolerance, and a chuck may hold a bit off-centre by a few microns to a few tens of microns, an eccentricity that becomes bit wobble at speed. The practical check is to load a test bar into the chuck and indicate it with a dial gauge near the bit end face while rotating; run-out should stay inside the permitted range.

These three errors combine as vectors with direction, not as a simple sum. A useful troubleshooting order is therefore: measure static coaxiality first (dial gauge on a master part, machine stationary), then location repeatability (load the same part ten times and look at the distribution of screw positions), and finally dynamic deviation (actual removal, watching cam-out rate and torque curve shape). Only by separating the three steps can the specific link be identified; mixing them yields nothing but "sometimes it works, sometimes it does not".

From a design standpoint the basic principle for reducing accumulated error is to shorten the datum chain. The fixture's locating datum, the part's design datum and the head's motion datum should share one set of datum surfaces, avoiding repeated transfers. Every transfer adds its own machining error and assembly clearance to the total. In its combined fastening and removal projects, Shenzhen Honred Technology typically builds locating pins and clamping points onto one base plate on its small screw removal equipment, and mates that plate directly to datum slots in the machine base with no adjustable transition plate in between — adjustable joints look flexible but are in fact a source of error and drift.

A station built around an automatic vision screw removal machine must also manage the calibration relationship between vision and motion. Coordinates from vision pass through a hand-eye transform before they become motion commands, and the error of that matrix goes straight into positioning accuracy. Calibration plate accuracy, number of calibration points, temperature and lens focus state at the time all affect the result. The engineering answer is to make calibration a reproducible procedure: fix the plate position, fix light and exposure, verify immediately against check points after calibration, and recalibrate when deviation exceeds the threshold. Working distance also matters — a change in lens-to-part distance introduces a scale error, and where Z height varies, vision reports XY systematically larger or smaller, which must be considered on projects with incoming thickness variation.

Across M0.8 to M2.0 one fixture set rarely covers every size. Small sizes need higher locating accuracy and lower clamping force; large sizes need more clearance. A workable approach makes fixtures interchangeable modules that mate through one set of datum pins, so changeover swaps the module without touching the machine datum. Repeatability after changeover is then guaranteed by the datum pins rather than by an operator re-aligning. This is why the M0.8-M2.0 screw removal solution treats modular fixtures as standard configuration.

One closing caution: coaxiality is not a case of higher being better, but of matching the process requirement. Holding 0.01 mm means precision-machined fixtures, a stiffer machine and tighter assembly tolerances, and the cost rises visibly. For a product with ±0.15 mm positional tolerance on its screws, chasing that is waste; for densely spaced M0.8 holes, 0.05 mm of deviation can produce steady cam-out. Work out what the process needs first, then decide how tight the structure should be — that is more economical than piling on precision.