Most CNC machining accuracy: where scale sets the limit

• 1
  Most CNC machining accuracy is a ratio, not a number
• 2
  How part size changes what the machine can hold
• 3
  Mass, clamping and the proportion of part to machine
• 4
  Heat is the largest uncontrolled axis
• 5
  Which machining route fits which part

Ask two shops for ±0.005 mm and you may get two different parts. The tolerance value is the same. What differs is the length over which that tolerance has to hold. On a 40 mm bracket, ±0.005 mm is routine. On a 3,000 mm frame, the same callout means the machine has to control position along a much longer loop of axis motion, thermal growth and material movement.

This is the proportion problem. A linear error of 0.02 mm is 0.05% of a 40 mm feature and 0.0007% of a 3,000 mm feature, but both read 0.02 mm on the CMM report. Engineers who read a tolerance as an absolute value get surprised at assembly. Engineers who read it as a ratio plan the datum strategy first.

So when we talk about most CNC machining accuracy, the useful question is not "how tight can you hold?" It is "how tight can you hold over this length, in this material, at this mass?" Everything below answers that second question. If your part fits in a 500 × 500 × 450 mm envelope and weighs under 50 kg, most of this page is background. If it does not, it is the whole job.

• 1
  Small parts, tight numbers±0.005 mm is normal below 500 mm of travel.
• 2
  Large parts, same numberThermal and mass effects start to dominate.
• 3
  The real questionTolerance per unit length, not tolerance alone.

A CNC machine holds position with a closed loop: encoder, drive, ballscrew or linear motor, then the structure that carries the tool to the work. Every element of that loop has an error budget. Over 500 mm the errors stay small and mostly cancel. Over 4,000 mm they accumulate, and the structure itself starts to bend under its own weight and the cutting load.

Consider a 4,000 mm gantry traveling along X. A temperature difference of 2 °C between the top and bottom of the beam bends it. In steel, thermal expansion is about 11.5 μm per meter per °C. A 4 m beam that warms 3 °C grows roughly 0.14 mm. That is 28 times the ±0.005 mm callout, and it happens before the tool touches metal.

Material choice changes the arithmetic. Aluminium expands at about 23 μm per meter per °C, roughly double steel. A 6061 frame that warms 4 °C over a 4 m length moves about 0.37 mm. On a 200 mm feature the same warming moves 0.018 mm. Same alloy, same shop, very different risk.

This is why we ask for the drawing envelope before quoting a tight tolerance. A ±0.005 mm hole pattern on a 300 mm plate is a normal five-axis job. The same pattern on a 3,200 mm base needs a different plan: staged machining, thermal soaking and a datum scheme that survives the growth.

• 1
  Rough ruleTolerance per meter matters more than tolerance alone.
• 2
  SteelAbout 11.5 μm per meter per °C.
• 3
  AluminiumAbout 23 μm per meter per °C.
• 4
  Long beamsBend before they stretch. Check both.

A heavy part does not just need a bigger machine. It changes the stiffness chain. The workpiece becomes part of the loop between the spindle and the table. A 1,200 kg casting sitting on four jack stands flexes under its own weight as the tool pushes down. The deflection may be only a few micrometers, but it moves between roughing and finishing as the material is removed.

Clamping is where most large-part errors hide. Bolting a thin wall down flat to a fixture flattens it for machining, then it springs back when released. We see 0.05 mm to 0.15 mm of springback on large aluminium housings when the fixture is not planned around the wall thickness. The fix is not a tighter tolerance. It is a better sequence: light roughing passes, stress-relief where the material allows, and finishing after the part has settled.

Five-axis work adds one more variable. The further the tool reaches from the rotary center, the more the machine's own geometry error is multiplied. On a Ø400 mm rotary table, a 300 mm tool offset doubles the effect of any angular error in the trunnion. On a small part the same error is negligible.

Practical proportion: when the part mass passes about one tenth of the machine's table capacity, deflection and clamping start to matter as much as the control loop.

• 1
  Springback0.05–0.15 mm on thin-wall large housings is common.
• 2
  Reach mattersLong tool offsets multiply angular error.
• 3
  Mass ratioPast 10% of table capacity, plan for deflection.

Machine tools grow as they run. A spindle that has been cutting for two hours is warmer than one that started cold, and the frame it sits on has followed. On short parts the drift is absorbed by the tolerance. On long parts it shows up at the far end of the travel, exactly where you can least afford it.

The control side is straightforward. We soak the machine, run warm-up cycles before critical cuts, and keep the shop at a stable temperature. Spindle and ballscrew cooling hold the loop closer to nominal. None of this removes the effect. It only keeps it predictable enough to compensate.

The part side is harder. A large casting pulled from a cold truck into a warm shop will move for hours. We let it equalize before the first finish pass. For aluminium and magnesium, that can mean overnight. For cast iron and steel, a few hours is usually enough if the shop is stable.

In-process probing is the practical answer. Instead of trusting the machine to arrive at the nominal position, we measure the feature and adjust the offset before the finishing pass. That is how a 4,000 mm part keeps a ±0.005 mm callout on a critical bore: the tolerance is closed around the measured reality, not the theoretical zero.

• 1
  Warm-upRun the spindle before any tight cut.
• 2
  SoakLet large parts equalize to shop temperature.
• 3
  Probe and offsetMeasure, then correct, before finishing.