CNC machining complex parts: where geometry meets the machine

• 1
  What makes CNC machining complex parts different
• 2
  Why setup count drives cost more than cycle time
• 3
  Tool reach, stiffness and the limits of deep features
• 4
  Datum control across multiple faces
• 5
  When CNC machining complex parts is the wrong call
• 6
  How a complex part moves through the shop
• 7
  Which process fits which complex geometry

Complexity is not a shape on a drawing. It is the number of independent things the machine, the fixture and the cutter must get right at the same time. A deep pocket with a 3 mm corner radius is simple until it sits on a 12-degree face that also carries a sealing groove. Then the setup count, not the cutting, drives the cost.

Four properties push a part into the complex category. Features on several non-orthogonal faces. Thin walls that move when the tool pushes them. Tolerances tighter than ±0.02 mm across unrelated datums. Surfaces and holes that must line up after heat treatment or coating. Each one is manageable alone. Stacked on one part, they interact.

The practical definition we use: a part is complex when you cannot reach every feature in one spindle orientation without the tool shank, the holder or the table colliding with the workpiece. That test is geometric, not subjective. It also tells you something useful. If the test fails, you either add setups, add axes, or change the design.

• 1
  Many facesFeatures spread over four or more directions
• 2
  Thin sectionsWall thickness under 1.5 mm in metal
• 3
  Tight spreadTolerances across datums that never share a setup
• 4
  Post-process shiftHoles that move after heat treat or coating

Every new orientation costs a re-clamp, a new zero, and a fresh chance to lose position. On a 3-axis machine, a part with features on five faces needs three or four setups, sometimes five. Each one adds fixturing time, an operator decision, and a stack of tolerance error that accumulates across the part.

Consider a manifold block, 180 × 120 × 90 mm, with ports on four sides and a face seal on the top. Three-axis work needs a vise setup, a side setup with angle plates, and a flip. Total position error across the block can reach 0.05 mm even when each individual setup holds 0.015 mm. The part is fine on paper and marginal in the fixture.

Five-axis work collapses that. A Ø400 mm trunnion table tilts and rotates the part so the spindle reaches five faces in one clamping. Position error stops accumulating because there is only one datum. The trade is different. You now need a CAM strategy that keeps the tool clear of the table through the whole tilt range, and you lose some rigidity when the part hangs out at 60 degrees.

The rule we apply: if a part has features on three or fewer orthogonal faces and fits a vise, three-axis plus fixtures is usually cheaper. If it needs four or more orientations, or if two features on different faces must hold ±0.01 mm to each other, five-axis wins.

• 1
  Count orientations firstThree or fewer: stay 3-axis
• 2
  Watch stacked errorEach setup adds its own position error
• 3
  Check table clearanceLong tools at high tilt lose stiffness

A complex part usually contains a feature the tool can barely reach. A 40 mm deep pocket with a 6 mm cutter needs a length-to-diameter ratio near 7. At that ratio the tool deflects under cutting load, and deflection shows up as taper in the wall, chatter on the floor, and a corner radius that comes out larger than drawn.

The numbers matter. A carbide end mill at 4×D is comfortable. At 6×D you reduce depth of cut and step-over, often to 30–40% of what the same tool does at 2×D. At 8×D and beyond, expect to rough with a larger tool first, then finish with a light pass, and to accept Ra 1.6–3.2 μm instead of a polished floor.

Corner radius is the other hard limit. A pocket with R2 corners cannot be finished by a 6 mm cutter; the tool simply does not fit. Either the design opens the radius to at least half the cutter diameter, or you accept a smaller tool with a longer reach and slower feed. Both are valid. The design decision should be made before the quote, not during it.

Undercuts and internal grooves have a similar boundary. If the feature has no straight-line path to the outside of the part, a rotating cutter cannot make it. That geometry belongs to EDM, casting, or additive work, not to milling.

• 1
  4×D is safeFull depth of cut, normal step-over
• 2
  6×D needs careReduce radial engagement to 30–40%
• 3
  8×D and upRough with a bigger tool, then light finish passes
• 4
  No line of sightUndercuts need EDM or casting instead

A datum is a promise about where the part sits. On complex work the promise breaks when different features are measured from different setups. A bore on face A and a slot on face C may each be perfect to their own zero, and still be 0.03 mm apart in the assembly.

The fix is to design one primary datum and reference everything else to it. In practice that means choosing a face that stays accessible through the whole process, including after anodizing or plating. If the primary datum is a face that gets coated, you lose the reference. Coatings add 5–25 μm per side depending on the process.

In-process probing helps here. We touch off a bore or a boss after each orientation and let the control correct the work offset. That does not fix poor fixturing, but it removes the drift that comes from thermal growth and re-clamping. On a 750 × 1,150 × 550 mm envelope, a warm spindle can move a feature 0.02 mm over a long cycle.

Best practice for the drawing: state the datum, state which features must hold relative to it, and leave the rest at general tolerance. Over-tolerancing unrelated features adds cost without adding function.

• 1
  One primary datumKeep it uncoated and reachable in every setup
• 2
  Coating adds sizeBudget 5–25 μm per side for plating and anodizing
• 3
  Probe after re-clampCorrects drift, not bad fixturing

CNC is a subtraction process with a rotating cutter. That single fact sets the boundary. If the geometry needs material added in a closed cavity, or needs a shape no tool can sweep, another process does it better and cheaper.

High-volume parts are the clearest case. At 10,000 units a year, a die-cast or injection-molded part can cost a fraction of the machined version, even after tooling. CNC makes sense for the bridge volume, the low run, and the revision-heavy stage where the design still moves.

Very thin, very large, or highly porous parts also fight the process. A 0.6 mm aluminum panel over 500 mm will deflect under clamping no matter how carefully it is supported. A part with 60% internal lattice is usually better printed. A part in a material we cannot source as bar or plate has to be cast first.

One more boundary is cosmetic. Machined surfaces carry tool marks. If a class-A visible surface is required, plan for finishing steps and accept that the machined texture underneath still shows through thin coatings.

• 1
  High volumeCasting or molding beats machining past a few thousand units
• 2
  Large thin panelsClamping deflection is the limit, not the cutter
• 3
  Internal latticesAdditive processes handle them natively
• 4
  Class-A finishMachined texture can telegraph through thin coatings