Accuracy custom cnc machining: where part accuracy actually comes from

• 1
  Accuracy is a stack, not a single number
• 2
  What machine geometry gives you, and what it does not
• 3
  Fixturing decides the tolerance you really get
• 4
  Heat moves the part before the cutter does
• 5
  You cannot hold what you cannot measure
• 6
  How to write a tolerance that survives production
• 7
  Which accuracy route fits the part

A tolerance on a drawing is a promise about one feature. Accuracy custom cnc machining is the sum of everything between the spindle and the probe: machine geometry, the fixture, the tool, the material, the coolant, the room. Change one and the stack moves.

The stack has a useful property. Errors that repeat can be compensated. Errors that wander cannot. A machine with 0.008 mm of repeatable positioning error will hold ±0.005 mm on a well-fixtured part after calibration. A machine with 0.002 mm of error that drifts 0.01 mm over a warm-up cycle will not, no matter what the spec sheet says.

So the first question on any tight job is not “how accurate is your machine.” It is “what makes this feature move, and does it move back to the same place.” Everything else follows from that.

On our floor the stack is written down per part. Machine, fixture, tool path, inspection method, environment. If a feature sits at the edge of capability, we say so before cutting metal rather than after.

• 1
  Repeatable errorCan be mapped and compensated out.
• 2
  Random errorShows up as scatter between parts.
• 3
  Drift errorGrows during a run, kills the last parts.

A 5-axis machining center with a Ø400 mm rotary table gives you access to five faces in one setup. That matters more for accuracy than most people expect, because every additional setup adds a re-clamping error. A three-setup part on a 3-axis machine can lose 0.02–0.05 mm purely in re-fixturing, even when each individual setup is perfect.

Simultaneous 5-axis motion also keeps the tool normal to a curved surface. On an impeller or a contoured mold insert, that turns a scalloped surface into a smooth one and removes the need for hand blending, which is itself a source of variation.

What geometry does not give you is stiffness on thin walls. Cutting force deflects the workpiece, not the machine. A 1.5 mm aluminum wall on a rigid 5-axis machine will still bow if the depth of cut is too aggressive. That is a tool path and support problem, and it is solved with support, not with a bigger machine.

We run 16 simultaneous 5-axis centers, 12 four-axis mills, 27 three-axis machines and 16 mill-turn centers across 3 plants. The part picks the machine, not the other way around.

• 1
  One setup beats threeEach extra setup adds re-clamping error.
• 2
  5-axis keeps tool normalBetter surface on contoured faces.
• 3
  Stiffness is per workpieceThin walls deflect regardless of machine size.

A vise is a spring. Clamp a thin ring or a thin plate and it deforms; machine it flat, release it, and it springs back to a bowed shape. The drawing says flat. The part is not flat. Nothing was wrong with the machine.

Soft jaws bored in place to the actual workpiece diameter fix most round parts. For plates, we use vacuum chucks or a bed of sacrificial material machined flat in the same setup, then skim the top face. Both approaches make the fixture match the part instead of the other way around.

For a batch, the fixture is the repeatability. A dedicated fixture with hard stops locates part 1 and part 500 the same way. A generic vise with a dial indicator locates them approximately the same way. On a 500-piece run at ±0.01 mm, that difference is the whole job.

The practical rule: if a feature is tighter than ±0.02 mm and the wall is under 3 mm, plan on a dedicated fixture before you plan on a machine upgrade.

• 1
  Thin wallsSupport or sacrificial material, never bare clamping.
• 2
  Round partsSoft jaws bored to the workpiece Ø.
• 3
  Batch runsHard stops beat indicators for repeatability.

Aluminum expands about 23 μm per meter per degree Celsius. A 300 mm aluminum part that warms 8 °C during roughing grows 0.055 mm. If the finishing pass happens before the part cools, the final dimension carries that error. We rough, let the part stabilize, then finish.

Spindle growth is the other half. A spindle warms over the first hour of use and pushes the tool down by 10–30 μm depending on the machine. That is why the first part off a cold machine is often the worst part. Warm-up cycles and in-process probing both address it.

Material stress is quieter but just as real. Extruded 6061 plate carries internal stress from the mill. Machine one side heavily and it releases as warp. Stress-relieved or cast stock behaves better on large thin parts.

On a 4,000 mm maximum processing size part we often schedule a roughing day and a finishing day. It sounds slow. It is faster than scrapping the part.

• 1
  Aluminum23 μm/m/°C; let the part cool before finishing.
• 2
  Spindle growth10–30 μm over the first warm-up hour.
• 3
  Stock stressRelieved or cast stock for large thin plates.

A CMM in a 20 °C room and a caliper on the shop floor do not agree. They are not supposed to. The CMM measures the geometry; the caliper measures the geometry plus the operator's hand pressure plus the local temperature. For anything under ±0.02 mm, the CMM is the referee.

In-process probing changes the economics. Instead of cutting a batch and finding out at final inspection, the machine checks a datum feature every few parts and offsets the remaining work. On a 200-piece run, that catches a worn tool at part 40 instead of part 200.

Surface finish is measured on the same principle. Ra 0.2–0.8 μm is a fine finish that usually needs a finishing pass with a small stepover. Ra 0.8–1.6 μm is the normal high-quality machined finish. Ra 1.6–3.2 μm is as-machined. Calling out a finish you do not need adds cycle time for no function.

We inspect 100% of parts before shipment: raw material check, in-process monitoring, final inspection. Reports are available on request.

• 1
  Under ±0.02 mmCMM, temperature-controlled room, defined datums.
• 2
  Above ±0.05 mmCalipers and gauges are usually enough.
• 3
  FinishSpecify Ra only where it carries function.

Datums first. A feature toleranced from three different references in the same view cannot be inspected consistently, and two shops will read it two ways. Pick the faces that actually touch the mating part and make them A, B, C.

Then decide what is bilateral. A ±0.005 mm callout on a 200 mm aluminum plate is a different animal from the same callout on a 20 mm steel bushing. The first one is fighting thermal expansion and stock stress; the second one is routine.

Then think about the cost curve. Going from ±0.1 mm to ±0.05 mm is often free, because it just means a normal finishing pass. Going from ±0.01 mm to ±0.005 mm can double the cycle time and add a dedicated fixture. The last increment is the expensive one.

Finally, say what the part does. A shaft that spins at 20,000 rpm and a bracket that holds a sensor have different accuracy needs. When we know the function, we can tell you which callouts are load-bearing and which ones are decoration.

• 1
  Datums from functionTolerance from the faces that mate.
• 2
  Scale mattersSame number, different difficulty on 20 mm vs 200 mm.
• 3
  Last increment costs±0.01 to ±0.005 mm can double cycle time.