Visit CNC Machining Precision Expert: What Tolerance Really Costs

• 1
  What the word precision covers in CNC machining
• 2
  How a five-axis setup removes stacked error
• 3
  Workholding is where tight tolerance is won or lost
• 4
  Heat, chips, and why the tenth part moves
• 5
  How a precision expert proves the number
• 6
  When a part does not belong on a precision CNC
• 7
  Matching the process to the part geometry

Precision is not one number. It is the smallest feature the process can repeat, the surface it can leave behind, and the stability it keeps across a full run. A drawing that says ±0.005 mm is asking for all three at once.

Three variables decide the answer. Machine geometry sets the floor. Workholding decides how much of that floor you keep. Heat decides whether the last part still matches the first. A shop that only talks about spindle speed is talking about the easy third.

GreatLight runs 127 high-precision CNC machines across 3 wholly-owned plants, with 16 simultaneous 5-axis machining centers and 12 four-axis mills. The mix matters more than the total. A 4,000 mm gantry job and a 40 mm medical fitting do not belong on the same spindle.

So when someone asks whether we can hold ±0.005 mm, the honest reply is a question back: on which feature, in which material, at what batch size? Precision is a property of a setup, not a slogan.

On a three-axis mill, a compound angle needs the part tilted on an angle plate. Every re-clamp adds a new datum error. Two setups can easily cost 0.02 mm before the cutter touches metal, and that is before the operator reads a dial.

A simultaneous five-axis center moves X, Y, Z plus two rotary axes at once. The tool approaches the feature from one direction and stays there. Complex surfaces, port faces, and deep pockets get cut without the part ever leaving the fixture.

That is the mechanism, not magic. Fewer datums means fewer places for error to enter. On a 5-axis part with five angled faces, we often see the setup count drop from four to one. The tolerance budget then goes into the cut instead of into the fixture.

The trade-off is real. Five-axis cycles are slower to program and need more clearance, so shallow flat parts are still faster on a three-axis machine. We route work by geometry, not by which machine sounds more advanced.

A perfect machine cannot save a part that moves. Thin walls, long shafts, and thin flange plates deflect under clamping force. The cutter then removes material from a shape that springs back after unclamping.

The fix is usually soft jaws machined in place, vacuum plates for thin sheet, or a low-profile fixture that supports the part near the cutting zone. For a Ø400 mm rotary table job, we balance the fixture as well, because an unbalanced load shows up as chatter at the outer edge.

Clamping pressure is a number, not a feeling. On aluminium 6061 or 7075, light hydraulic clamping holds a part with far less distortion than a hand-tightened vise. On stainless 316L and 17-4PH, the same part may need more pressure and a different cutter path.

If a feature is critical, say so on the drawing. We can leave a finishing allowance of 0.2–0.3 mm, relax the clamp, and take a light pass. That single pass is often what makes a ±0.005 mm callout achievable.

Metal grows when it warms. Aluminium expands about 23 μm per meter per degree Celsius. A 500 mm part that runs 5 °C warmer than the gauge room is roughly 0.06 mm longer than it measures cold. That is twelve times a ±0.005 mm band.

So in-process measurement has to match the final condition. We let parts stabilize before final inspection, and we keep the inspection room at a controlled temperature. Measuring a hot part and shipping it cold is how good shops create bad reports.

Chip evacuation belongs in the same conversation. Recut chips grind the surface and push the cutter off line. Through-spindle coolant and programmed chip breaks keep the pocket clean on deep features.

Spindle growth is the third piece. A spindle that has run for two hours is not the same length as one that just started. For long runs we warm up the machine and re-probe the tool, which is why a first-article check matters more than a sample from mid-run.

A tolerance claim without a measurement method is a guess. CMM reports, roundness testers, surface profilometers, and gauge pins each answer a different question. The drawing should say which one applies.

Surface finish is the easy one to misread. Ra 1.6–3.2 μm is a normal as-machined finish. Ra 0.8–1.6 μm needs a finer step-over, a sharper insert, and often a finishing pass. Ra 0.2–0.8 μm is a different process window and a different cost.

We inspect 100% of parts before shipment, with raw material checks, in-process monitoring, and a final inspection. Reports are available on request. If a feature needs SPC data, tell us at quote stage so the inspection plan is built around it.

The four certificates behind this work are ISO 9001:2015, IATF 16949:2016, ISO 13485:2016, and ISO 27001:2022. They cover quality systems, automotive, medical devices, and information security. They do not replace a first-article report on your part.

CNC is a subtractive process. It cannot beat a die casting on unit cost at 50,000 pieces, and it cannot match injection molding on a thin cosmetic shell. If the volume is high and the geometry is simple, the correct advice is to change process, not to quote harder.

Very large thin panels are another poor fit. A 4,000 mm part is machinable on our gantry travel of 4,000 × 400 × 150 mm, but a 2 mm wall over that length will chatter no matter how good the fixture is. Sheet metal fabrication is usually the better route.

Hardened tool steel above roughly 45 HRC also changes the plan. It can be cut, but tool life drops and the finishing strategy changes. Sometimes the right answer is to machine soft, heat treat, then grind the critical faces.

The last boundary is time. Quotation and free DFM analysis come back within 12 hours, production can start within 24 hours, and parts ship in 3–5 days. Those windows assume the drawing is frozen. A moving revision is the most common cause of a late delivery, and our historical late-delivery probability sits below 2%.