Denver CNC machining accuracy: where it comes from, and where it stops

• 1
  What a tolerance number actually describes
• 2
  Why five-axis setups reduce tolerance stack-up
• 3
  Thermal behavior and why it is the quiet variable
• 4
  Surface finish: a separate spec from tolerance
• 5
  Inspection is what turns a claim into evidence
• 6
  Which setup fits which part

A drawing calls out ±0.005 mm on a bore, and everyone nods. The trouble is that the number describes the allowed variation of one feature, not the ability of a shop to repeat it across 200 parts. Those are different claims, and only the second one matters on a production run.

Tolerance is the sum of several independent errors. Machine positioning contributes some. Fixture deflection adds more. Tool wear, coolant temperature, and the thermal growth of the spindle push in their own direction. When a shop quotes a tolerance, it is really quoting the width of that combined error band.

±0.005 mm is roughly ±0.0002 in, which puts the work in the range where a temperature swing of a few degrees Celsius becomes visible on a 200 mm aluminum part. Aluminum expands about 23 μm per meter per degree Celsius. That is not a rounding error.

So the useful question is not "what is your tolerance?" It is "what is your tolerance on this feature, in this material, at this quantity?" A shop that can answer the second question has measured its process. A shop that only answers the first has memorized a brochure.

Every time a part moves to a new fixture, it picks up a new datum error. A three-axis job with six faces can mean five re-clampings, and each one contributes a small offset that lands in the final position of every feature machined after it.

Simultaneous five-axis machining avoids most of that. With 16 five-axis machining centers in the shop and a Ø400 mm rotary table on the smaller platforms, complex contours, undercuts, and organic shapes can be cut in one setup. Features that share a datum stay related to each other.

One setup also means one thermal history. The part is not cooling between operations, being measured, and then re-clamped cold. That alone removes a class of drift that is hard to find after the fact.

The trade-off is real. Five-axis programming and setup take longer, and not every part earns it. A flat bracket with three holes is cheaper on a three-axis machine and will hold the same tolerance. Five-axis pays off when the part has angular features, deep pockets on multiple faces, or datums that must stay tied together.

• 1
  Use five-axis whenFeatures on three or more faces, undercuts, or tight position between angled datums
• 2
  Stay with three-axis whenFlat plates, simple pockets, prismatic parts with one dominant face
• 3
  Add four-axis whenCylindrical parts with cross-holes or axial features that repeat around a bore

A CNC machine is not a rigid, unchanging object. The spindle grows as it warms, the ballscrews extend, and the bed absorbs heat from chips and coolant. A machine that has been running for six hours is a different machine than one started cold.

This is why the first parts off a cold machine are often the loosest in the batch. Shops that understand this run a warm-up cycle before critical work, and they keep the finishing passes on high-value features late in the cycle, once the machine has settled.

Material choice doubles the effect. Aluminum moves more than steel for the same temperature change. Titanium is worse in a different way: it holds heat at the cutting edge, so the tool grows and the surface finish drifts even when the machine is stable.

For work at ±0.005 mm, we measure at 20 °C or note the temperature in the inspection report. If your assembly happens in a different thermal environment, that gap belongs in the tolerance discussion, not in a dispute after delivery.

Ra and dimensional tolerance are independent variables, and mixing them up causes arguments. A part can hold ±0.005 mm with a rough Ra 3.2 μm surface, and a part can be mirror-polished while drifting out of position.

The finish ranges that matter in practice: Ra 1.6–3.2 μm for as-machined functional surfaces, Ra 0.8–1.6 μm for sealing faces and bearing fits, and Ra 0.2–0.8 μm when a surface is lapped or fine-bored. Each step down costs time and tool life.

Finish interacts with tolerance through tool pressure. A finishing pass with a small depth of cut deflects the tool less, so it improves both. A heavy pass that burns the surface also pushes the tool off line.

If a drawing calls for both a tight tolerance and a fine finish, say so early. The process plan changes: rough, semi-finish, stress relief if needed, then finish. Cutting corners on the sequence is how parts end up reworked or scrapped.

• 1
  Ra 1.6–3.2 μmGeneral machined surfaces, non-sealing faces
• 2
  Ra 0.8–1.6 μmBearing bores, O-ring grooves, sliding fits
• 3
  Ra 0.2–0.8 μmFine-bored or lapped faces, optical and sealing interfaces

A tolerance is only as good as the measurement behind it. A caliper with 0.01 mm resolution cannot verify a ±0.005 mm callout with any confidence. The gauge needs to be roughly four times finer than the tolerance it checks.

The inspection plan matters as much as the instrument. Raw material is checked on receipt. In-process monitoring catches drift before a batch is finished. Final inspection confirms the features on the drawing, and reports are available on request.

For first articles, a full dimensional report plus a capability check tells you more than a pass/fail stamp. It shows where the process sits inside the band, which is how you know whether the next 500 parts will still be good.

Sampling is a decision, not a default. On tight features we inspect 100% before shipment. On loose cosmetic features, sampling is reasonable and cheaper. The right answer depends on what a failure costs downstream, not on a habit.