Japanese CNC accuracy: how ±0.005 mm is actually held

• 1
  What Japanese CNC accuracy means on a machine tool
• 2
  Thermal drift: the largest error source nobody quotes
• 3
  Tool wear, runout, and the real cost of a dull cutter
• 4
  Probing and in-process measurement
• 5
  When five-axis geometry helps accuracy, and when it hurts
• 6
  Materials and geometry that will not hold ±0.005 mm
• 7
  Five process controls behind the number
• 8
  Which process holds which tolerance

Japanese CNC accuracy is not a brand of machine or a certificate on a wall. It is the practice of accounting for every error source that can push a cutter off its nominal path, then measuring the result instead of trusting the setup. A machine tool repeats to a spec sheet number in a controlled room. In a shop running three shifts, the same machine carries a thermal load that changes its geometry through the day.

The published figure is positioning accuracy: how close the slide arrives to a commanded point. It says nothing about the part. Part accuracy is the sum of positioning error, spindle growth, tool deflection, clamping distortion, material springback, and the datum you chose for the fixture. A machine with ±0.002 mm positioning can still produce ±0.02 mm features if the tool pushes off the wall on a thin rib.

This is why we treat ±0.005 mm as a process capability rather than a machine property. It is the tolerance we can hold on stable materials, compact geometry, and processes that are monitored in-process. It is not a blanket promise on every feature of every part. Knowing which features fall outside that envelope is the more useful engineering skill.

Japan's contribution to this discipline is mostly documentation. Cutting parameters, tool life, and inspection results get recorded per job, so the next run starts from data rather than from memory. That habit travels well. It does not require a Japanese machine, only the discipline of writing down what happened.

• 1
  Machine spec ≠ part resultPositioning accuracy is only one of six or seven error sources.
• 2
  ±0.005 mm is conditionalIt depends on material, feature stiffness, and geometry access.
• 3
  Records beat memoryTool-life and inspection data carry the process between runs.

Cast iron grows about 11 μm per meter per degree Celsius. A spindle that warms 8 °C over four hours of cutting moves a 400 mm workpiece by roughly 35 μm at the far end. That single effect is seven times the tolerance we are discussing. No control system compensates for it unless the machine is instrumented and the compensation is running.

The practical controls are boring and effective. We warm up spindles before the first cut, keep coolant at a stable temperature, and avoid running a finishing pass on a cold machine after a long idle. For tight work, the roughing and finishing operations are separated in time so the part and the fixture reach the same thermal state before the final cut.

Heat also travels with the chips. A pocket machined dry in aluminium can push local temperature far above ambient, and the wall closest to the tool grows more than the wall far from it. Roughing with coolant and leaving 0.3–0.5 mm for finishing keeps the final pass light and the heat input small.

If a customer measures a part in a 20 °C inspection room immediately after it comes off a warm machine, the reading will drift for the first hour. We let parts stabilize before final inspection. That wait is part of the tolerance, not a delay in the process.

• 1
  Warm up before finishingA cold spindle cuts a different size than a warm one.
• 2
  Separate rough and finishLet the part reach thermal equilibrium before the last pass.
• 3
  Stabilize before measuringCMM readings on a hot part move during the first hour.

A carbide end mill does not fail suddenly. It wears on the flank, the cutting edge radius grows, and the forces pushing the tool away from the workpiece increase. On a 6 mm tool with 40 mm of overhang, a few hundredths of a millimeter of radial wear can move a wall by 10–15 μm. The part still looks fine. The dimension does not.

Tool runout matters more than most shops admit. Ten microns of runout on a finishing tool produces a two-flute pattern where one edge does most of the cutting. That edge wears twice as fast, and the surface finish goes from Ra 0.8 μm to Ra 2 μm over the course of a batch. We measure runout at the holder taper, not just at the tool tip.

Tool life is a number we set per material and operation, not a feeling. A finishing tool in 6061 aluminium might run 90 minutes between changes. The same tool in 17-4PH stainless might run 25. When the life limit is reached, the tool comes out even if it still looks serviceable. Pushing past the limit is how a batch of 200 parts develops a slow dimensional drift that only shows up at final inspection.

Coatings and geometry help at the margin. AlTiN coatings extend life in stainless and tool steel. Polished flutes move chips out of deep pockets in aluminium. Neither one replaces a tool-change schedule.

• 1
  Measure runout at the holderTaper error adds to tool error.
• 2
  Set tool life in minutesReplace on schedule, not on appearance.
• 3
  Match coating to materialAlTiN for stainless and steel, polished flutes for aluminium.

A spindle probe turns the machine into a measuring device between cuts. We use it for three jobs: finding the actual stock position before the first cut, checking a critical feature after roughing, and confirming the datum before finishing. On a casting with ±0.5 mm stock variation, probing the real surface prevents a scrapped first article.

The limits are real. A touch probe on a machine tool resolves to roughly 1–2 μm under good conditions, and it is affected by the same thermal drift as the cutting process. It is not a CMM. What it does well is catch a trend early: if a bore measures 8 μm oversize after roughing, the operator can adjust the finishing offset before the part is finished.

Probing also exposes fixture problems. If a probe finds the part has moved 20 μm between the roughing and finishing setups, the cause is usually clamp pressure or chip packing under a locator, not the machine. Finding that during the cycle is much cheaper than finding it in the inspection room.

For high-volume work, we log probe results per part. The trend line shows tool wear before the dimension leaves tolerance. That is the practical version of Japanese process control: a number that tells you to act before the part is bad.

• 1
  Probe the stock, not the drawingCasting and forging variation is real.
• 2
  Probe is not a CMMUse it for trends and offsets, not final acceptance.
• 3
  Log results per partTrend data catches tool wear early.

Five-axis machining improves accuracy mainly by reducing setups. Every re-clamp introduces a new datum error, and a part that moves through four operations accumulates four chances to be located slightly wrong. A simultaneous five-axis machine that reaches five faces in one setup removes three of those opportunities. That is the strongest accuracy argument for five-axis work, stronger than the kinematic capability itself.

The trade-off is stiffness. A trunnion table holds the part away from the machine bed, and a long tool reaches around a rotating workpiece. Both reduce rigidity. A feature that would hold ±0.005 mm on a three-axis machine with a short, rigid setup may only hold ±0.015 mm on a five-axis machine with 120 mm of tool overhang. The number of axes is not the accuracy number.

The right question is how many setups the part needs and how much tool overhang each one requires. A housing with bores on four sides and a tight true-position callout is a five-axis job. A flat plate with a few holes is faster and more accurate on a three-axis machine.

We run 16 simultaneous five-axis machining centers, 12 four-axis mills, 27 three-axis machines, and 16 mill-turn centers. The mix exists because the correct answer changes part by part. Quoting everything on the most capable machine is not accuracy, it is overhead.

• 1
  Fewer setups, fewer datum errorsFive-axis wins on multi-face parts.
• 2
  Stiffness drops with reachLong tools and rotary tables cost rigidity.
• 3
  Match the machine to the partNot every tight tolerance needs five axes.

Tolerance is a property of a part, not just a shop. Some combinations simply will not hold ±0.005 mm in production, and quoting them that way sets up a failed first article. Knowing the boundary saves both sides a revision cycle.

Thin walls are the most common problem. An aluminium wall under 0.8 mm deflects under cutting force and moves again when the clamps come off. It can be machined, but the as-machined dimension will scatter by 20–40 μm across a batch. If the design needs a thin wall at tight tolerance, the answer is usually a stress-relief step and a light finishing pass, or a design change.

Soft or gummy materials behave differently. Magnesium AZ31B and AZ91D cut freely but generate heat and can move after machining as internal stresses release. Copper alloys such as C110 and beryllium copper hold dimensions well but often need a post-machining stress relief before final finishing.

Deep bores with a high length-to-diameter ratio are another limit. Beyond about 4:1, a boring bar deflects enough that the bore tapers. We can hit ±0.005 mm on a 3:1 bore in 6061 aluminium. At 8:1, the realistic number is closer to ±0.015 mm without a reaming or honing step.

Titanium TC4 (Ti-6Al-4V) and Inconel add their own issues: low thermal conductivity, high cutting forces, and work hardening at the surface. They can be held to ±0.005 mm on compact, well-supported features, but cycle times are long and tool life is short. That is a cost discussion, not an accuracy one.

• 1
  Wall under 0.8 mmDeflection and clamp release scatter the dimension.
• 2
  Bore beyond 4:1 L/DBoring bar deflection shows up as taper.
• 3
  Magnesium and copper alloysMay need stress relief before final machining.