Absolute accuracy in CNC machining

A machine tool has a commanded position and an actual position. The gap between them is positioning error. When a builder quotes a tight positioning figure, that number usually comes from a laser interferometer test on a new machine in a temperature-controlled room, with no cutting load and no chips on the table. It is a useful baseline. It is not what you get at 3 pm on a Thursday.

Repeatability is the spread of results when the machine returns to the same point twenty times. A ball screw with 5 μm of lead error can still repeat to 1 μm if the error is consistent, and the control can compensate for a consistent error. Random error cannot be compensated. That is why a machine can hold a boring diameter all day and still miss a true position callout on a bolt circle.

The practical question is never the number on the certificate. It is whether the error stays inside your tolerance band across a full run, in a shop that is not air-conditioned to 20 °C, with tools that wear. We measure both and report the spread, not just the best reading.

Thermal growth is the largest single source on most jobs. Cast iron and steel expand roughly 10–12 μm per meter per °C. An aluminum 6061 part grows faster, about 23 μm per meter per °C. A 400 mm aluminum housing that warms 5 °C during roughing moves about 46 μm before finishing even starts. On a ±0.005 mm callout, that is the whole budget.

Tool deflection behaves differently. A Ø12 mm carbide end mill at 3× diameter stickout will bend under cutting force. Light finishing passes at 0.2 mm radial engagement keep that bending small. Heavy radial cuts do not, which is why we leave 0.3–0.5 mm of stock for the finishing pass on tight features rather than cutting to size in one go.

Geometric error sits underneath both. Squareness between axes, spindle tilt, and ball screw pitch error are built into the machine and change slowly with wear. They are mapped during calibration and compensated in the control. If the compensation table is never refreshed, a machine that held ±0.005 mm when new may drift to ±0.02 mm after a few years of hard service.

Fixturing closes the loop. A part held on three points and clamped at one end will lift as material is removed. We check the datum after roughing and re-clamp before finishing on parts with thin walls or long unsupported spans.

The first thing we do on a tight part is decide what not to machine in one setup. Five-sided access on a 5-axis center removes the repositioning error that comes from flipping a part across three vises. We run 16 simultaneous 5-axis machining centers for exactly this reason: fewer setups means fewer chances for the datum to move.

In-process probing is the second lever. A Renishaw-style touch probe measures a datum or a bore after roughing, and the control adjusts the finishing offsets before the finish pass runs. This catches thermal drift and tool wear without pulling the part off the machine. On a 20-part run, that is the difference between the first part and the last part matching.

Temperature control matters more than most buyers expect. Our inspection room and the tight-tolerance cells are held near 20 °C. A part measured straight off a warm machine will read smaller than it is, especially in aluminum. We let parts stabilize before final inspection and record the temperature on the report.

Finally, we measure the process, not just the part. If the first three pieces drift in the same direction, the offset is adjusted and the run continues. If they scatter, the process stops and the cause is found. That is what keeps a run predictable instead of hoping the last part passes.

Aluminum is the hardest common material to hold a tight tolerance on, because it moves the most with temperature and cuts freely with little damping. A 7075 aerospace bracket and a 6061 housing behave differently under the same cut. Thick sections are stable. Thin ribs on a 2 mm web will chatter and spring regardless of the machine.

Stainless and tool steel are more dimensionally stable once cool but push cutting forces up. A 17-4PH or 4140 part needs more finishing stock and more attention to workholding. Titanium TC4 (Ti-6Al-4V) is worse: it work-hardens, so a dull tool raises force and deflection at the same time. Tool change intervals get shorter, not longer.

Plastics are a category of their own. POM and PEEK move with humidity and heat, and the finished part may measure differently a day later. If a plastic part carries a ±0.02 mm callout, it needs to be measured under the same conditions it will see in service. Otherwise the number is meaningless.

The practical rule: pick the material for the function first, then tell us the tolerance. We will say whether it is realistic on that geometry, or what it would cost to get there. Pushing a soft, thin part to ±0.005 mm is possible but slow, and slow means money.