Precision CNC: What a Stable Output Guarantee Actually Rests On

• 1
  Key takeaways
• 2
  Why output drifts even on a good machine
• 3
  Geometry and the five-axis advantage
• 4
  Thermal state: the quiet variable
• 5
  The measurement loop that closes the guarantee
• 6
  When a stable output guarantee is not realistic
• 7
  Setup and control strategies compared

A CNC machine does not hold a dimension. It repeats a motion. The dimension you measure is the result of that motion plus everything that changed between parts: spindle growth from heat, ball screw expansion, tool edge wear, chip load on the fixture, and the material's own stress relief after roughing. A stable output guarantee is a claim about how well those changes are controlled, not about the machine's brochure accuracy.

Consider a real sequence. The first part of a morning run is cut with a cold spindle. By part forty, the spindle and the ballscrew have grown 15–25 μm from thermal expansion. If the offset is not corrected, every part after that sits on the high side of the tolerance band. The machine is fine. The process was not managed.

This is why a stable output guarantee is usually written as a tolerance plus a batch size, not as a single number. ±0.005 mm on one prototype is a different claim from ±0.005 mm on 5,000 parts spread over three weeks. The second claim requires measurement, correction, and documentation at every step.

• 1
  Thermal driftSpindle and axis growth of 15–25 μm over a few hours is normal and must be compensated.
• 2
  Tool wearA carbide end mill can lose 20–40 μm of edge radius across a long run, pushing dimensions and finish.
• 3
  Fixture repeatabilityClamping force variation of 10% moves a thin wall more than any axis error.

On a three-axis machine, a part with features on five faces needs multiple setups. Each re-clamp adds a positioning error of roughly 5–15 μm, and those errors stack. Five-axis machining keeps the part in one fixture and rotates the tool or the table instead, so the reference datum never moves. That removes one of the largest sources of batch-to-batch variation.

The gain is not only accuracy. Fewer setups mean fewer chances for a chip to land on a locating face or for a clamp to deform a wall. On a complex aluminum housing, going from four setups to one can cut the total position error from around 40 μm to under 15 μm without changing the machine's base accuracy.

Geometry has limits too. A five-axis machine with a Ø400 mm rotary table cannot hold a 4,000 mm part. GreatLight runs a mix of travel sizes: 4,000 × 400 × 150 mm for long parts, 750 × 1,150 × 550 mm and 600 × 600 × 600 mm for mid-size work, and compact 500 × 500 × 450 mm and 500 × 310 × 200 mm envelopes for small, high-feature-density parts. Matching the part to the machine envelope matters more than the machine's headline spec.

• 1
  Single setupRemoves re-clamp error of 5–15 μm per additional setup.
• 2
  Rotary tableØ400 mm table suits compact parts with features on all faces.
• 3
  Travel matchingLong parts need the 4,000 mm machine; small parts hold better on a compact envelope.

Heat enters a machine from three places: the spindle bearings, the axis motors, and the cutting zone itself. A spindle running at 12,000 rpm can rise 8–12 °C above ambient within two hours. Over a 400 mm steel ballscrew, a 10 °C rise translates to roughly 45 μm of growth. That is ten times the tolerance on a precision part.

Shops handle this in different ways. Some run a warm-up cycle before production and then hold the offset. Others use spindle chillers and temperature-compensated scales. The simplest reliable method is to probe a master artifact at fixed intervals and let the control adjust the work offset. That corrects both thermal growth and any slow mechanical shift in one step.

For critical runs, we log spindle temperature and the probe result alongside the part. If the two drift together, the correction is working. If the part drifts while the probe stays flat, the problem is elsewhere: tool wear, fixture, or material. Separating those signals is most of the diagnostic work.

A stable output guarantee depends on measuring the right thing at the right time. Final inspection alone tells you whether a batch passed. It does not tell you whether the next batch will pass. In-process probing is what turns a hope into a control.

In practice, we probe the fixture datum on the first part, cut a test feature, measure it on the machine, and adjust the offset before the run continues. During the run, the operator checks a defined feature every 20–50 parts, depending on the tolerance band and the material. For a ±0.005 mm part, the interval is short. For a ±0.05 mm part, it can be much longer.

Final inspection adds a second layer: 100% inspection before shipment, with raw material checks upstream and in-process monitoring in the middle. Reports are available on request. This is what makes a tolerance claim auditable rather than promotional.

• 1
  First-article probeSets the work offset before the run begins.
• 2
  Interval checksEvery 20–50 parts for tight tolerances; longer for loose ones.
• 3
  Final inspection100% before shipment, with reports on request.

Some parts cannot be held to ±0.005 mm across a long run, and saying so up front saves everyone time. Thin walls under 0.8 mm deflect under clamping and cutting forces. Soft plastics such as PP and HDPE move with temperature and moisture, and they spring back after the tool passes. Deep pockets in titanium accumulate heat faster than it can be removed, so the tool wears unpredictably.

Material choice matters as much as geometry. Aluminum 6061 and 7075 are stable and machine cleanly. 17-4PH stainless in the H900 condition holds tolerance well but wears tools quickly. Inconel and magnesium AZ31B both need slower parameters and more frequent tool changes, which raises cost and shortens the reliable batch length.

The honest answer is that a stable output guarantee is a range, not a constant. For a well-behaved aluminum part with good fixturing, ±0.005 mm across 10,000 pieces is achievable. For a 0.5 mm wall in PEEK, ±0.05 mm may be the realistic floor. A supplier who claims otherwise is not measuring carefully.