Precision custom CNC machining parts: how tolerance is actually held

• 1
  Key takeaways
• 2
  What precision custom CNC machining parts really depend on
• 3
  Why setup count decides the tolerance you can hold
• 4
  Feature geometry sets the accuracy floor
• 5
  How datum strategy connects machining to inspection
• 6
  Material behavior changes the tolerance you keep
• 7
  What actually drives cost on tight-tolerance parts
• 8
  From one prototype to 10,000 parts without losing the tolerance
• 9
  How a precision custom CNC machining parts order moves through the shop
• 10
  Choosing the machining route by part geometry

Accuracy on paper and accuracy in a production run are different things. A machine that positions to ±0.002 µm on a warm-up test can still produce a part 0.03 mm out of true if the stock moves between operations. When engineers ask us to quote precision custom CNC machining parts, the first questions are never about spindle speed. They are about how many setups the geometry needs, where the datums fall, and what the part will do after machining.

Subtractive machining removes material, and removal releases stress. A 7075 aluminum block machined from solid will move as the internal stress redistributes. The same block cut from pre-stress-relieved plate moves less. That is why the process plan matters more than the tolerance number on the title block. A ±0.005 mm position tolerance on a hole is reachable. The same callout across a 400 mm part face, held only by a vise, is a different problem.

Heat is the second variable. Spindle growth, coolant temperature, and chip load all shift the tool relative to the work over a shift. On a 2-hour cycle, the first article and the last article can differ by more than the tolerance band if no thermal compensation is applied. This is not a machine defect. It is physics, and it has to be managed with warm-up cycles, in-process probing, or both.

• 1
  Material removal releases stressPre-relieved stock and balanced roughing passes limit post-machining movement.
• 2
  Each setup adds errorEvery re-clamp stacks fixture error onto the feature it locates.
• 3
  Thermal state driftsWarm-up cycles and probing keep the tool-to-work relationship stable.

A three-axis machine cutting five faces means five setups, or four plus a re-fixture. Every time the part leaves the vise, the operator re-establishes a datum. The error does not add linearly, but it compounds. Typical re-clamp repeatability on a good vise and stop is around 0.01 mm. Do that four times and the last feature carries all of it.

Five-axis simultaneous machining changes the arithmetic. A trunnion table with a Ø400 mm rotary table lets the tool reach five faces without releasing the part. Features that were separated by three setups now share one coordinate system. On a medical manifold with ports on four sides, that single change often moves the achieved true position from 0.03 mm to under 0.01 mm without touching feeds and speeds.

There is a boundary. Five-axis does not help a part that is mostly a flat plate with one critical bore. A well-trammed three-axis mill with a rigid fixture will hold ±0.005 mm on that bore all day. Five-axis adds cost and programming time for no gain. The decision is geometric, not prestige.

• 1
  Three-axis is enoughSingle-face or two-face prismatic parts with one critical feature.
• 2
  Four-axis earns its keepCylindrical parts with cross holes or slots around the axis.
• 3
  Five-axis pays offComplex angles, deep cavities, or features on five sides in one setup.

Tolerance interacts with shape. A 2 mm diameter hole, 20 mm deep, has a length-to-diameter ratio of 10:1. A standard carbide drill will walk. You need a pilot, a peck cycle, or a reamer, and you may still see 0.02 mm of drift at the bottom. The drawing says ±0.01 mm. The geometry says otherwise. Good process planning flags this before the first chip.

Thin walls behave the same way. A 0.8 mm aluminum wall on a 50 mm tall pocket will deflect under cutting force, then spring back and leave a tapered wall. Reducing radial engagement, using a smaller tool, and taking a finishing pass at low load all help. None of them make a 0.3 mm wall on a 100 mm tall part a stable production feature. That part belongs in a different process or a different design.

Surface finish also has geometry limits. Ra 0.2–0.8 μm on a flat face is routine with a fine finishing pass. The same callout inside a deep rib or on a curved blend requires a smaller tool, longer cycle, and more tool wear. If the finish callout covers the whole part, the quote reflects the hardest feature, not the average one.

• 1
  Deep holes driftPast 5:1 L/D, plan for peck drilling and reaming, not a single pass.
• 2
  Thin walls deflectBelow roughly 1 mm, expect to slow down and take spring passes.
• 3
  Finish scope mattersA blanket Ra callout is priced at the hardest surface on the part.

A drawing datum is a promise about what the part locates against. If the machine holds the part by the bottom face and the drawing calls the side face as datum A, the inspector will measure a different relationship than the machinist built. That gap shows up as a rejected part that is actually good, or an accepted part that will not assemble.

The fix is alignment between three things: the CAD model's datum reference frame, the fixture's locating scheme, and the drawing's callouts. When all three agree, a CMM report reads cleanly. When they disagree, every measurement becomes a debate. On parts with tight true position, we ask for the datum scheme before quoting, because it changes the fixture and the setup sequence.

In-process probing closes the loop. On a long cycle, touching off a critical bore after roughing lets the control adjust the finishing pass before the part is finished. That is how a ±0.005 mm bore stays in band across a 10,000-part run without scrapping the tail of the batch.

• 1
  Match datums end to endCAD, fixture, and drawing should share one locating scheme.
• 2
  Probe before finishingIn-process measurement corrects for drift mid-cycle.
• 3
  100% inspection on critical featuresFinal reports available on request for the features that matter.

Aluminum 6061 and 7075 machine differently. 6061 cuts clean and holds form well. 7075 is stronger but more prone to movement after heavy stock removal, especially in thin sections. For a tight tolerance 7075 part, pre-stress-relieved plate and a rough-then-finish sequence with a rest between passes reduce distortion. That rest period is not wasted time. It lets the part settle before the finishing cuts define the geometry.

Stainless 304 work-hardens. A light finishing pass at low feed can harden the surface and dull the next tool, which then pushes the wall. The usual fix is to stay under the hardened layer with a consistent depth of cut and avoid dwelling. 17-4PH behaves better when machined in the solution-treated condition and aged afterward, but that requires a heat-treat step the print may not mention.

Titanium Ti-6Al-4V and Inconel generate heat at the cutting edge. Thermal growth on the tool and the part is significant, and coolant delivery matters more than on aluminum. Tolerance on these materials is often held with in-process probing rather than relying on the machine's thermal stability alone. Plastics like PEEK and POM move with temperature and moisture, so they are usually measured after a stabilization period.

• 1
  Rough, rest, finishOn 7075 and thin sections, let stress release before final cuts.
• 2
  Avoid work-hardeningOn 304, keep depth of cut consistent and do not dwell.
• 3
  Probe on hot materialsTi and Inconel need in-process correction more than aluminum.

Cost scales with the number of operations that touch the part, not with the tolerance number alone. A ±0.005 mm callout on a single bore adds a finishing pass and a probe cycle. The same callout on eight bores across four faces may add two setups and a fixture, which is a different order of cost. Buyers who understand this can often save money by tightening only the features that function.

Lead time follows the same logic. Standard parts ship in 3–5 days. A part needing a custom fixture, a heat-treat step, or a specific surface finish adds time for the fixture build or the outside process. Quotation and free DFM analysis come back within 12 hours, and production can start within 24 hours once the plan is agreed. The DFM step often catches a feature that would have added a week.

Surface finish is a quiet cost driver. As-machined Ra 1.6–3.2 μm is standard. Ra 0.8–1.6 μm adds a finishing pass. Ra 0.2–0.8 μm may need a different tool, a slower feed, or a secondary operation. If the finish callout is only functional on a sealing face, mark that face and leave the rest as-machined.

• 1
  Tighten selectivelyApply the ±0.005 mm callout only where function demands it.
• 2
  Fixtures add timeCustom workholding is a real cost item, not overhead.
• 3
  Finish is a line itemBlanket Ra callouts are priced at the hardest surface.

The first article proves the process can make a good part. The run proves it can make the same part 10,000 times. Those are different tests. On a prototype, an operator can adjust a tool offset by hand. On a run, the process has to hold without that attention. That means tool wear compensation, scheduled insert changes, and a documented setup sheet that the next shift can follow.

Tool wear is the main drift source on long runs. A carbide end mill that cuts a slot at 10.00 mm when new will cut 9.98 mm after 200 parts if wear is not compensated. Setting the control to adjust the offset on a fixed interval, or using a tool presetter with wear tracking, keeps the slot in band. For holes held to ±0.005 mm, a reamer with a known wear curve is more predictable than an end mill interpolation.

We run 127 high-precision CNC machines across three plants, including 16 simultaneous five-axis centers and 16 mill-turn centers. Maximum processing size is 4,000 mm. That range covers most custom parts, but the limit that matters on a long run is not machine size. It is whether the process can repeat without an operator watching every cycle. Qualification rate across inspected features runs 99.99%.

• 1
  Setup sheets travelWritten parameters let the next shift repeat the first shift's result.
• 2
  Compensate for wearScheduled offset changes keep tight holes in band over thousands of parts.
• 3
  Inspect before shipmentRaw material, in-process, and final checks with reports on request.