CNC machining an accurate step: a proven 4-stage process

• 1
  What decides accuracy before the spindle turns
• 2
  Why CNC machining an accurate step starts at the datum
• 3
  Choosing the axis count for your geometry
• 4
  Tool selection and cutting parameters that hold tolerance
• 5
  Inspection methods and when each one applies
• 6
  Where this process stops being the right answer
• 7
  Step by step: from CAD to a verified part
• 8
  When to choose 3-axis, 4-axis, or 5-axis

Accuracy is a chain, and the first link is the datum. If the CAD model references a face that the fixture cannot locate against, every downstream dimension inherits that error. We ask for a datum callout on the drawing before programming starts, and we check it against the stock we actually received. A 0.02 mm mismatch between drawing datum and raw stock face is common on castings and forgings.

The fixture is where most shops lose tolerance quietly. Soft jaws bored in place to the finished diameter hold roundness better than a vise with parallel jaws, because the clamping force spreads around the circumference instead of pinching two points. For thin-wall parts, we reduce clamping pressure and add a sacrificial support ring so the wall does not spring back after unclamping.

A useful rule: the fixture should be at least as stiff as the part. When a 6061 bracket is held in a light vise on a thin plate, chatter shows up as a rippled finish and a dimension that drifts 0.03–0.05 mm between the first and last part of the run. Bolting the fixture directly to the machine table removes that variable.

• 1
  Datum firstAgree the datum on the drawing before programming. Never assume the stock face is the datum.
• 2
  Bore soft jaws in placeBoring jaws on the machine keeps concentricity tied to the spindle, not to the vise body.
• 3
  Control clamping forceThin walls move under clamping. Support them or reduce pressure before the finishing pass.

Three-axis machining is still the right answer for a large share of parts. Flat plates, blocks with features on one or two faces, and simple housings run faster and cheaper on a 3-axis mill with a good vise setup. Our 27 three-axis machines handle work up to 4,000 mm in the long travel, which covers most fixture plates and base plates.

Four-axis work enters when a part has features on four sides of a prismatic shape. A rotary table lets the part index 90° between operations without a second setup, which removes the re-zeroing error that comes with flipping a part by hand. Our 12 four-axis mills and Ø400 mm rotary tables suit valve bodies, manifolds, and connector housings.

Five-axis is a different decision. It is not about speed. It is about reaching features on compound angles in a single setup. If a part has an angled port, an undercut flange, or a curved surface that would need three or four 3-axis setups, the accumulated setup error usually exceeds the ±0.005 mm tolerance. That is where our 16 simultaneous five-axis centers earn their cost. If the part is a simple plate, five-axis adds nothing but cycle time.

• 1
  3-axisFlat and prismatic parts, features on one or two faces. Lowest cost per part.
• 2
  4-axisFeatures on four sides. One rotary setup replaces multiple manual flips.
• 3
  5-axisCompound angles and undercuts. Justified when setup count would otherwise exceed two.

For aluminum, a 3-flute carbide end mill at 12,000–18,000 rpm and 0.05–0.10 mm per tooth feeds cleanly on 6061-T6 and 7075. The trap is finishing with the same tool used for roughing. A roughing tool with a corner radius leaves a witness mark and a dimension that reads 0.01–0.02 mm off. We switch to a fresh, sharp finishing tool for the last 0.2–0.3 mm of radial stock.

Stainless 304 and 17-4PH need lower surface speed and more coolant. Running 304 too fast work-hardens the surface, and the next pass cuts through a hard skin that pushes the tool off line. We keep surface speed around 80–120 m/min on 304 with high-pressure coolant and avoid dwelling in the cut. Titanium TC4 is slower still, and we never let a tool rub.

Heat is the quiet enemy of an accurate step. A spindle running for hours grows a few hundredths of a millimeter, and the part moves with it. For tight runs we warm the machine with a 20-minute cycle before the first cut, then check the first article against the drawing. If a feature reads 0.01 mm over, we adjust the wear offset rather than the program, so the correction is traceable.

• 1
  Separate rough and finish toolsFinishing with a worn rougher is the most common cause of an out-of-tolerance step.
• 2
  Watch stainless speedToo fast on 304 work-hardens the skin and deflects the following pass.
• 3
  Warm up the spindleA 20-minute warm-up cycle removes thermal growth before the first article.

Calipers and micrometers are fine for a quick read, but they cannot confirm a true position. For a hole pattern with a ±0.05 mm position tolerance, a CMM or a vision system is the only honest check. We use a CMM for first articles and for any feature called out with geometric tolerancing. The report goes to the customer before the batch runs.

For high-volume runs, the check frequency matters as much as the method. A feature with a 0.02 mm window needs checking every 10 parts. A feature with a 0.10 mm window can run 50 parts between checks. We set the interval from the tolerance width and the material's tendency to move, not from a fixed schedule.

Surface finish is measured separately. Ra 0.8–1.6 μm is the normal range for a machined sealing face. Ra 0.2–0.8 μm needs a finer finishing pass and often a different tool, which adds cycle time. If a drawing calls for Ra 0.2 μm on a non-functional face, question it. That callout can double the cost of the operation for no benefit.

• 1
  First articleCMM against the drawing, report sent before the batch. Non-negotiable on tight tolerances.
• 2
  In-processFrequency set by tolerance width, not by habit. Tighter window, more checks.
• 3
  FinishMeasure Ra where it functions. Do not specify a fine finish on cosmetic-only faces.

CNC machining is subtractive, so internal cavities with no tool access are out. A closed internal channel or a sharp internal corner with a radius smaller than the tool cannot be cut. If a design needs those, it belongs in die casting or 3D printing, or the design needs to change. We flag these in DFM review rather than quoting a part we cannot make.

Very hard materials change the economics. Tool steel and Inconel cut slowly, and the tool wear per part is high. Tolerances still hold at ±0.005 mm, but cycle time and cost rise. If the part does not need that hardness in service, a 4140 or 17-4PH part with a surface treatment is usually cheaper and just as functional.

Volume is the other boundary. Above roughly 10,000 parts a year, die casting or injection molding usually wins on unit cost. CNC stays competitive for low and mid volume, and for any part where the design may still change. We run from one prototype to 10,000+ part runs, so the crossover is a conversation, not a hard rule.

• 1
  No closed cavitiesIf a tool cannot reach it, it cannot be machined. DFM catches this before quoting.
• 2
  Hard alloys cost moreInconel and tool steel hold tolerance but cut slowly. Confirm the hardness is needed.
• 3
  Volume crossoverPast roughly 10,000 parts a year, casting or molding usually beats CNC on unit cost.