CNC Accuracy Unparalleled: What Sets Tolerance on a Part

• 1
  What CNC accuracy actually means on the shop floor
• 2
  Machine geometry, thermal drift, and the limits they place on CN
• 3
  Workholding, cutter deflection, and setup count
• 4
  How machining accuracy is measured and reported
• 5
  Which tolerance is realistic for which feature

Accuracy and repeatability are two different numbers, and buyers often mix them up. Accuracy is how close the finished feature lands to the nominal dimension on the drawing. Repeatability is how close part 2 lands to part 1 when nothing on the setup changes. A machine can be very repeatable and still be inaccurate if its geometry or tool offsets are off by a fixed amount.

The second distinction is between machine accuracy and process accuracy. A five-axis center may be positioned to a few microns on a test sphere, but the part still has to survive workholding stress, cutter deflection, heat, and chip evacuation. Process accuracy is always worse than machine accuracy, and it is the number the customer actually receives.

That is why a tolerance callout on a drawing is a question, not an instruction. It asks whether the geometry, the material, the setup count, and the inspection method can all hold that band at a cost that makes sense. When the answer is no, the honest move is to change the design, not to promise the number.

Every machine carries a fixed error budget. Linear axis straightness, squareness between axes, spindle radial and axial runout, ballscrew pitch error, and rotary table backlash all stack up. On a well-kept five-axis center these are small, but they do not disappear. Squareness alone can move a feature 5–10 μm across a 400 mm travel if the machine has drifted since its last calibration.

Thermal growth is the quiet one. A spindle running at 12,000 rpm warms the headstock and pushes the tool tip down by tens of microns over the first hour. Castings and fixtures expand too. Aluminium moves about 23 μm per metre per °C, steel about 12 μm. A 10 °C swing across a 500 mm steel part is roughly 60 μm of length change before a single cut is taken.

This is where the lights-out shops and the job shops separate. A temperature-controlled room at 20 ±1 °C, a spindle warm-up cycle, and a probe check on a known artifact at the start of each shift keep the error budget stable. Without that, the first part of the morning and the last part of the night are not the same part.

Structure matters as well. Cast iron and polymer-concrete bases damp vibration better than welded steel frames. That damping decides surface finish and tool life more than the control does. A machine with a stiff frame and a modest control will out-cut a flexible frame with a fast control almost every time.

Workholding is where most tolerance is lost. A vise tightened by hand can bow a thin wall by 20–40 μm, and the part springs back when the jaws release. Vacuum plates, soft jaws bored to the part, and low-melt fixturing remove that fight. For thin sections, rough machine, stress relieve, then finish in a second setup with light passes.

Cutter deflection follows a simple rule: force over stiffness. A Ø6 mm end mill at 3× diameter stick-out deflects far less than the same tool at 6× stick-out. Cutting forces rise with radial engagement and with harder material. In 17-4PH or Ti-6Al-4V, a 0.5 mm radial stepover and a 0.1 mm finish pass keep deflection inside a few microns.

Setup count is the other silent variable. Each re-fixturing adds a datum shift. On a three-axis machine a part with features on five faces may need four setups, and each one adds its own error. A five-axis machine with a Ø400 mm rotary table can often reach those faces in one or two setups, which removes error rather than adding capability.

Tool wear is the last link. A coated carbide end mill holding ±0.005 mm on a finishing pass will not hold it after 40 minutes of cutting. Tool life monitoring, or simply changing the finisher on a fixed cycle, is cheaper than scrapping a near-finished part.

You cannot ship a number you cannot measure. A CMM with a stated accuracy of 2 + L/350 μm is the usual reference. For features the CMM cannot reach, a bore gauge, micrometer, or optical comparator fills the gap. The instrument has to be at least four times tighter than the tolerance it checks, or the reading is mostly noise.

Temperature at inspection matters as much as temperature at cutting. A part measured hot off the machine reads oversize, then shrinks. Standard practice is to let the part settle to room temperature before the final check, or to record the temperature and note it on the report.

At GreatLight, inspection runs through raw material check, in-process monitoring, and a final check before shipment, with reports available on request. We work to ±0.005 mm and hold a 99.99% qualification rate across production runs. Those numbers are useful only if the drawing and the inspection method agree on what is being measured.

Watch for datum conflicts. If the drawing calls out a datum that is not accessible on the finished part, the inspector has to invent one, and the number becomes a guess. Calling out functional datums early removes that argument.