Change the Accuracy of CNC Machining: Where the Real Limits Come From

• 1
  What Change the Accuracy Really Means on a Machine
• 2
  Machine Geometry and Ball Screw Condition
• 3
  Workholding: The Most Underrated Variable
• 4
  Tool Wear, Deflection and Chip Load
• 5
  How Probing and Inspection Close the Loop
• 6
  Which Parts Hold Tight Tolerances and Which Do Not
• 7
  Tolerance and Finish by Process and Feature

Accuracy of CNC machining is the difference between the dimension on the drawing and the dimension on the finished part. It is a result, not a setting. Nothing on the control panel says "accuracy". The operator sets feeds, speeds, tool offsets and work offsets; the machine turns those into a cutter path; the material pushes back. What comes off the table is the sum of all of it.

Three words get mixed up in purchasing conversations. Accuracy is how close you land to the nominal dimension. Repeatability is how close the second part lands to the first. Resolution is the smallest step the control can command. A machine can have fine resolution and poor accuracy, and that combination is common on older equipment with worn thrust bearings.

So when someone asks how to change the accuracy of a job, the honest answer starts with a question: which of the three numbers is failing? A part that is consistently 0.03 mm oversize on one bore is an offset problem. A part whose bore wanders ±0.02 mm around the mean is a stiffness or thermal problem. The fix is different, and so is the cost.

We work to ±0.005 mm (±0.0002 in) when the geometry and the material allow it. That figure is a capability, not a promise for every feature on every drawing. A 300 mm long aluminum housing with thin walls will not hold the same tolerance as a 40 mm stainless shaft, even on the same machine, in the same shift.

• 1
  AccuracyDeviation from nominal on one part
• 2
  RepeatabilitySpread across a run of parts
• 3
  ResolutionSmallest commanded step, not a quality guarantee

The frame, the guideways and the ball screws set the floor for everything else. Squareness between X and Y, straightness of each axis and the perpendicularity of the spindle to the table all fold into the part. A machine that is 0.01 mm out of square over 500 mm will produce that error in the part no matter how carefully the program is written.

Thermal growth is the second half of the geometry story. A spindle running at 12,000 rpm heats up over the first 60 to 90 minutes of a shift. Ballscrews warm from their own motion. Castings absorb heat from the shop. On a 400 mm travel, a 2 °C rise in the frame can move the tool a few micrometres, and that drift does not announce itself.

This is why warm-up cycles matter. A machine that is pushed straight into tight work at 07:00 behaves differently at 09:30. Shops that hold ±0.005 mm routinely run a 20 to 30 minute spindle warm-up and then check a known artifact before touching a tight job.

Ball screw wear shows up as backlash: the axis lags when it reverses direction. Laser compensation and pitch error mapping can correct a repeatable error, but they cannot correct a screw whose preload is gone. When backlash grows past the control's compensation range, the axis needs mechanical work, not a new program.

A part is only as stiff as the way it is held. A thin-wall aluminum housing clamped in a vise will deflect under the clamping force before the tool ever touches it. The cut releases that stored energy and the wall springs back, so the finished wall is thick where the clamp pushed and thin where it did not.

The usual fix is to stop fighting the part. Support it underneath, clamp down rather than sideways, or use a soft jaw machined to the part profile. For thin floors and webs, we take light finishing passes at 0.1 to 0.2 mm radial engagement instead of one heavy pass, and let the material relax between passes.

For parts with a flatness or parallelism callout, the fixture is often more important than the machine. Vacuum plates and dedicated soft jaws hold the reference face flat during the whole cycle, so the second op starts from a true surface. Clamping on a rough cast face and expecting a 0.02 mm parallel result is a coin flip.

Five-axis work changes the equation again. A part held in a rotary table sees gravity and cutting force in different directions as the table indexes. If the fixture is not balanced and rigid, the part moves between operations even when the machine itself is fine. That is a fixture problem, not a machine problem.

A carbide end mill is not rigid. A 6 mm tool hanging 40 mm out of the holder will deflect under normal cutting force. Long reach and small diameter are the two things that hurt most. If a finishing pass is expected to hold ±0.005 mm on a deep pocket, the tool needs to be as short and as large as the geometry allows.

Wear changes the effective diameter. A tool that has cut 200 parts in 6061 has a slightly different radius than a fresh one, and the difference shows up on a critical bore. Production runs therefore track tool life and change tools on a count, not on a feeling. On tight work, we offset for the measured wear rather than waiting for a visual sign.

Chip load per tooth is the practical knob. Too light a load rubs the material and work-hardens stainless; too heavy a load pushes the tool and the part apart. In 304 stainless, a common starting point is 0.03 to 0.05 mm per tooth with plenty of coolant, then adjust from the sound and the chip form.

Roughing and finishing belong in different passes. Roughing removes bulk with a tough tool and generous load; finishing takes 0.2 to 0.5 mm radial and 0.05 to 0.1 mm axial with a sharp tool. Trying to hold a tolerance in a single heavy pass is where a lot of scrap is made.

In-process probing changes what a shop can promise. A touch probe measures a reference feature on the actual part after the first op, and the control shifts the work offset for the next step. That corrects setup error and the part-to-part variation in a casting in one move, without a human touching the offsets.

The measuring side matters just as much. A part can only be inspected to about one fifth of the accuracy of the tool doing the measuring. Checking a ±0.005 mm feature with calipers is not a measurement, it is a guess. A coordinate measuring machine, a micrometer on a controlled temperature, or a bore gauge gives an answer you can act on.

Temperature matters on both sides of the process. A part that is 5 °C warmer than the inspection room will shrink as it cools. Aluminum moves about 23 μm per metre per °C, so a 300 mm aluminum part inspected hot can read 0.03 mm large and measure correct an hour later.

We inspect 100% of parts before shipment, with raw material checks, in-process monitoring and a final inspection, and we can supply reports on request. For safety-critical work in aerospace or medical, the inspection plan is agreed before the first chip is cut, not after.

Material behavior decides how much of the machine's capability you can actually use. Aluminum 6061 and 7075 cut clean and hold ±0.005 mm well on rigid features. Stainless 316 and 17-4PH hold tolerance but move more from heat and work hardening, so the process needs more care and more time.

Titanium TC4 (Ti-6Al-4V) is a different game. It conducts heat poorly, so the cutting edge runs hot and wears fast. It also springs back against the tool. Tight tolerance on thin titanium sections is possible, but it needs reduced radial engagement, sharp tooling and patience. Inconel is harder again.

Thin walls and long slender features are the usual failure point, not the material grade. A 1 mm aluminum wall 80 mm tall will move as the internal stresses from the plate release. Stress-relieved stock and symmetric material removal on both sides help; so does accepting that the wall tolerance has to be looser than the bore tolerance.

Plastics are their own case. POM and PEEK have high thermal expansion and low stiffness, so the same feeds that work in aluminum will push them around. They also relax after machining. A plastic part that measures on size at the machine may be 0.05 mm different the next morning.