Strictly Controlled CNC Part Tolerance: The Mechanics Behind ±0.005 mm

• 1
  What a strictly controlled CNC part tolerance depends on: the machine loop
• 2
  Workholding, datum and tool path: where tolerance is lost before cutting
• 3
  Material behavior and how it moves a strictly controlled CNC part tolerance
• 4
  How the tolerance is verified, not just claimed
• 5
  Where ±0.005 mm stops being the right answer
• 6
  Which control matters most for which part

A tolerance number on a drawing is a wish until the machine loop can hold it. That loop has four parts: the frame, the drives, the tool, and the feedback that tells the control where the cutter actually is. If any one of them drifts, the part drifts with it, no matter how tight the callout reads.

Rigidity sets the floor. A 5-axis trunnion cutting a titanium bracket sees different deflection than a 3-axis mill roughing an aluminium plate. The heavier cut pushes the tool and workpiece apart, so the control has to compensate for a spring-back it cannot see directly.

Thermal growth is the slow drift nobody notices at 8 a.m. A spindle running for two hours can move 10–20 μm in Z on a machine without a chiller loop. Two hours later the same offsets cut oversize. Shops that hold ±0.005 mm all day let the machine warm up first and keep the coolant at a set temperature.

Feedback closes the loop. Glass scales or encoder feedback tell the control where the axis really is, not where the motor thinks it is. On a 4,000 mm travel machine, ballscrew growth alone can eat 30 μm over a long part. Scales read the table, not the screw, which is why long parts stay parallel.

• 1
  RigidityHeavier cuts deflect more; light finishing passes protect the last 20 μm.
• 2
  Thermal stateWarm-up cycles and chilled coolant keep offsets stable through the shift.
• 3
  Feedback typeScales read the slide; encoders read the motor. Long parts care about the difference.

Most out-of-tolerance parts are lost in setup, not in the cut. If the vice jaw lifts the part 5 μm on one side, the first face is flat but the second is not. Six setups can stack six small errors into one large one. That is why a strictly controlled CNC part tolerance usually starts with fewer setups, not a slower spindle.

Datum choice matters as much as clamp force. Referencing from a machined face gives the operator a true surface. Referencing from raw stock gives the operator a surface with saw marks and scale, and the error transfers straight into the part. On second operations, a soft jaw cut in place holds round parts without crushing them.

Tool path strategy decides how much the last pass has to remove. A constant-engagement path keeps radial load steady, so the cutter deflects the same amount all the way around a pocket wall. A sharp corner in the path spikes the load, and the wall bulges. Leaving 0.2–0.3 mm for a finishing pass at Ra 0.8–1.6 μm is a common shop habit that pays off on the wall straightness.

Five-axis parts add one more variable. The rotary table has its own center offset and its own backlash. If those are not measured and stored in the control, a hole drilled at B 45° lands somewhere other than the CAD position. Shops that run a rotary table of Ø400 mm check that offset on a schedule, not once at installation.

• 1
  Fewer setupsEvery re-clamp adds an error that no tool change can remove.
• 2
  Machined datumReference a cut face, not raw stock.
• 3
  Constant engagementSteady radial load keeps wall deflection even around a pocket.

The same program cuts differently in 6061-T6 and in 7075. Aluminium moves with heat, so a heavy roughing pass can leave a thin wall bowed until it cools. The finishing pass then cuts a shape that is not the final shape. Shops that hold tight walls rough, let the part rest, then finish.

Stainless 316L work-hardens. A cutter that rubs instead of shears hardens the surface and pushes the next pass away. Sharp tools and a feed that stays above the rubbing range keep the cut clean. 17-4PH in the H900 condition cuts more predictably than in the annealed state, but it is harder on the tool.

Titanium TC4 ( Ti-6Al-4V ) conducts heat poorly, so the heat stays in the cutter. That heat grows the tool, and a grown tool cuts oversize. Keeping the tool cool with high-pressure coolant is not only about tool life; it is about the dimension at the end of the pass.

Plastics bring a different problem. POM and PEEK expand with the heat of cutting and shrink back after. Measure them hot and they read oversize; measure them cold and they read true. A part held to ±0.005 mm in PEEK usually needs a temperature soak before final inspection, not just a caliper check off the machine.

• 1
  Rough and restLet the part cool before the finishing pass on thin walls.
• 2
  Sharp toolsRubbing work-hardens stainless and pushes the next pass away.
• 3
  Coolant on titaniumTool growth from heat shows up as an oversize bore.

A tolerance you cannot measure is a guess. Calipers read to 0.02 mm on a good day, and that is already four times the ±0.005 mm band. Bores go to a bore gauge or an air gauge. Outside dimensions on a 4,000 mm part go to a CMM or a laser tracker, because a hand tool cannot span that length and stay honest.

Sampling strategy decides what the number means. One part measured at the start of a run says almost nothing about part 4,000. In-process checks catch drift while there is still time to correct the offset. Final inspection on a CMM confirms the last parts match the first. Both are needed if the callout has to hold across the order.

CMM temperature matters. Aluminium grows about 23 μm per metre per °C. A 300 mm aluminium part measured 5 °C warmer than the reference reads roughly 35 μm longer than it is at 20 °C. That is several times the tolerance band, so a shop without temperature control cannot honestly certify ±0.005 mm on aluminium.

Rounding is the quiet error. A drawing that says 0.05 mm is not the same as ±0.005 mm, and the two get mixed up in conversation. Before quoting, read the callout, the datum scheme and the inspection method together. If they disagree, the part is not ready to cut.

• 1
  Right instrumentBore gauge, air gauge or CMM; match the tool to the feature.
• 2
  In-process plus finalDrift correction during the run, confirmation at the end.
• 3
  20 °C referenceThermal expansion on aluminium can exceed the whole band.

Not every feature needs the tight band. A tolerance of ±0.005 mm on a mounting hole that only has to clear an M6 bolt adds cost and inspection time for nothing. Designers who mark a whole drawing tight usually get a slower quote and no better function. Put the tight band on the features that locate or seal, and let the rest run at general machining tolerance.

Size limits the claim. A small part in a rigid fixture is a fair fight for ±0.005 mm. A 4,000 mm frame with thin walls is a different problem, because temperature alone moves the material more than the band allows. On very long parts, the honest move is to agree on a datum and a measurement temperature before quoting, not to promise the number.

Surface finish and tolerance trade against each other. A Ra 0.2–0.8 μm seal face takes light finishing passes, which is fine. Demanding that finish on every face doubles cycle time and can pull the part out of tolerance through repeated spring passes.

Finally, the number has to survive the process after machining. Anodizing adds a few micrometres per surface. Plating adds more. If a ±0.005 mm bore is anodized after cutting, the shop has to cut it undersize on purpose. That decision belongs in the drawing notes, not in a phone call after the parts ship.

• 1
  Callout only what mattersTight bands belong on locating and sealing features.
• 2
  Size changes the claimLong parts need a datum and a temperature agreement.
• 3
  Finishing adds thicknessAnodizing and plating move dimensions after cutting.