Accurate CNC Processing Service: What Actually Holds ±0.005 mm

Every accurate CNC processing service works inside an error budget. On a typical aluminum bracket, the machine contributes maybe 8 μm, the fixture 15 μm, thermal drift 20 μm, and the probe another 5 μm. The tolerance you drew is ±0.005 mm in total, so one weak link breaks the part. Engineers who treat the machine as the only variable usually chase scrap they cannot explain.

The largest single term is often thermal. A 100 mm aluminum part grows about 2.3 μm per 1 °C. Over a 6 °C afternoon swing that is 14 μm of pure growth, before any cutting starts. Steel grows roughly half as much, but a 400 mm steel shaft still moves 24 μm across the same swing. Machine geometry, spindle growth, and coolant temperature add to it. No controller can compensate a trend it cannot see.

Workholding sits second. A vise that pulls a thin wall 20 μm during clamping will spring back when you release it. Soft jaws bored in place, vacuum plates, or a dedicated fixture remove that pull. We machine most thin-wall parts on a fixture that supports the wall from the inside and cuts both sides in one setup.

Tool path and tool wear come last, and they are the easiest to control. Constant-engagement paths keep radial load steady, so the tool deflects predictably instead of jumping between full-width and light cuts. We log tool life per material and replace before the flank wear passes 0.05 mm. A worn 10 mm end mill can push 30 μm of deflection into a deep pocket.

• 1
  Machine geometrySquareness, spindle runout, and ball screw backlash. Verified on a test cut, not just a spec sheet.
• 2
  Thermal controlClimate control plus a warm-up cycle. We log shop temperature through the run.
• 3
  Workholding stiffnessBored soft jaws and dedicated fixtures. The clamp should not deform the part.
• 4
  Measurement loopProbe on-machine, then verify on a CMM before shipment. Reports on request.

More axes do not automatically buy accuracy. A 3-axis machine cutting a flat plate with holes from one side is more rigid and easier to verify than a 5-axis machine doing the same job. The right question is how many setups the part needs. Each setup adds a datum shift, and each datum shift adds error.

A part with features on four sides, or with compound angles, belongs on a 4-axis or 5-axis machine. Cutting those features in one setup removes the re-fixturing error entirely. That is usually worth more than the machine's own positioning accuracy. On our 5-axis centers we hold ±0.005 mm on features reached from a single setup, and the rotary table repeats to within a few microns.

Very large parts change the math. A 4,000 mm frame cannot be measured with a hand tool in a meaningful way, so accuracy depends on machine geometry and a stable thermal environment. We run long parts on machines with travels up to 4,000 × 400 × 150 mm and check them with a laser tracker or a large CMM. The tolerance on a 4,000 mm frame is often ±0.05 mm, not ±0.005 mm, and that is the honest number.

Small parts are a different problem. A 20 mm medical component with a 0.5 mm wall deflects under light clamping pressure. Here the 5-axis machine with a Ø400 mm rotary table and a soft-jaw fixture wins, because we can cut the whole profile without unclamping.

Aluminum moves the most. A 6061 part that measures perfect in the morning can drift 10 μm by the afternoon if the shop is not held at 20 °C. We rough aluminum, let it rest, then finish, so the internal stress releases before the last cut. For 7075 and 2024 the effect is stronger, and we often add a stress-relief cycle.

Stainless and titanium resist thermal growth better but fight the cutter. 316L work-hardens if the tool rubs, so we keep the feed per tooth high and the radial engagement low. Titanium Ti-6Al-4V conducts heat poorly, so the tool edge runs hot. A sharp tool and a rigid setup matter more than the machine's positional spec. On Inconel we slow the spindle and accept lower material removal, because chatter is the real enemy.

Plastics are the opposite case. PEEK and POM cut easily but deflect under clamping and expand with heat from the cut itself. We use light clamps, sharp tools, and air blast instead of flood coolant. On PEEK the tolerance target is often ±0.02 mm, not ±0.005 mm, because the material moves after machining.

The honest answer is that material, not the machine, sets the floor. A 5-axis machine cannot hold ±0.005 mm on a PEEK part that grows 30 μm overnight. We will tell you that at the quote stage.

A tolerance claim is worthless without a measurement plan. We check raw material certificates on arrival, monitor dimensions during the run, and inspect every part before it ships. For a first article we produce a full dimensional report. For production runs we sample the critical features and log the rest.

On-machine probing catches trends early. If the probe shows a bore drifting 5 μm across 200 parts, we correct the offset before the trend reaches the tolerance limit. This is cheaper than sorting parts at the end. We also record shop temperature and note any cut that ran outside the normal window.

Final inspection happens on a CMM in a temperature-controlled room, not on the shop floor. That matters: a part measured at 26 °C against a 20 °C drawing carries an error of 14 μm on a 100 mm aluminum feature. The CMM room sits at 20 °C ±1 °C, and parts soak for at least two hours before measurement.

We do not claim every part hits ±0.005 mm. We claim we know which features can, which cannot, and why. That is what the report shows.