CNC Machining Plan: What Engineers Decide Before the Spindle Starts

• 1
  What a CNC Machining Plan Actually Controls
• 2
  How the Plan Drives Accuracy: Datums, Setups and Stack-Up
• 3
  When a Part Needs a Full CNC Machining Plan
• 4
  How Material Choice Rewrites the Plan
• 5
  Where CNC Machining Plans Break Down
• 6
  Writing the Plan: 6 Steps in Order
• 7
  Plan Depth by Part Type

A CNC machining plan is the set of decisions that sit between a 3D model and the first chip. It answers five questions: what stock do we start from, where do we hold the part, which surfaces become datums, in what order do we cut, and how do we prove the result. The model does not answer any of these. Two shops can quote the same STEP file and produce two very different plans, which is why identical geometry sometimes comes back at very different cost and lead time.

The plan is not a G-code file. G-code is the last output of the plan, not the plan itself. Before CAM, someone has to decide whether the part is machined from one block or two, whether it is fixtured once or repositioned three times, and whether a feature is better turned or milled. Those choices set the tolerance stack, the number of setups and the scrap risk. Change the datum after programming and you usually rewrite the program.

A workable plan also names its own limits. It says which tolerance matters, which surface finish matters, and which dimensions are reference only. That last point saves money. A drawing with 60 dimensions at ±0.02 mm is not the same job as the same drawing with 6 functional dimensions at ±0.02 mm and the rest at ±0.1 mm. The plan is where that distinction gets recorded and priced.

• 1
  StockBar, plate, casting or near-net forging, plus 1–3 mm allowance per face
• 2
  DatumsPrimary, secondary and tertiary faces chosen for repeatable location
• 3
  Setup countEach extra setup adds stack-up error and handling time
• 4
  InspectionWhich dimensions are measured, with what instrument, at what stage

Tolerance is not created by the machine alone. It is created by the relationship between the cutting tool and the datum. If a 0.05 mm position callout is measured from a face that was itself milled in a second setup, the error of the first setup is now inside the second setup's tolerance. Machining a part in one setup from a single datum removes that chain. This is the main technical reason five-axis work exists: not to look impressive, but to keep features in one coordinate system.

Setup count is the clearest predictor of achievable accuracy. One setup, with the part held on a stable face, can hold tight position between features. Two setups typically add 0.02–0.05 mm of cross-setup variation on a well-controlled process. Three or four setups, especially when thin walls are involved, can push variation past the print unless the plan adds an in-process check between operations.

Thermal behavior belongs in the plan too. Aluminium 6061 moves about 23 μm per meter per °C. A 500 mm part that warms 8 °C between roughing and finishing shifts roughly 0.09 mm before any tool touches it. Good plans rough, let the part rest or cool, then finish. They also leave 0.3–0.5 mm of radial stock for finishing so the final pass removes a light, even load instead of a varying one.

• 1
  One setupBest case for position between features; limited by part access
• 2
  Two setupsAdds 0.02–0.05 mm cross-setup variation in normal practice
• 3
  Rough then finishSeparate operations release internal stress before final passes
• 4
  Rest allowance0.3–0.5 mm radial stock for a stable finishing cut

Not every part deserves a written plan. A flat bracket with four holes and a ±0.1 mm profile can go straight to CAM. The plan is the drawing, the stock size and the operator's experience. Spending two engineering days on that part is waste. The signal to write a fuller plan is not part size; it is feature interdependence. If feature B is located from feature A, and feature C is located from B, the sequence is now a technical decision.

Deep pockets with thin floors are a classic case. Roughing the pocket before the surrounding profile leaves more material connected to the floor and reduces chatter, but it also means the profile cut happens on a part that has already released stress. The opposite order is stiffer during profiling but risks the floor deflecting during the pocket cycle. Neither order is universally right. The plan has to pick one and state why.

Parts with a hardness step, such as 17-4PH in the H900 condition or 4140 pre-hardened to 28–32 HRC, force a hard-milling or pre-hard-milling decision. Pre-hardened stock removes the heat-treat distortion problem but raises tool wear and cost. Machining soft and sending the part to heat treatment afterwards keeps the cutting cheap but moves distortion into the final geometry. This trade-off cannot be settled inside CAM.

Long parts bring their own boundary. On our 4,000 × 400 × 150 mm travel machines, a long rail will deflect under its own weight when only one end is supported. The plan then needs support positions and possibly a sequence that machines from the middle outward to keep the part balanced on the table.

• 1
  Feature chainsB located from A, C located from B: sequence becomes a decision
• 2
  Thin floorsRoughing order changes stiffness more than cutter choice does
• 3
  Hardened alloysPre-hard versus post-heat-treat changes both cost and distortion
• 4
  Long partsSupport placement is part of the plan, not an operator improvisation

Aluminium is forgiving. 6061-T6 cuts fast, holds ±0.005 mm on stable geometry and takes Ra 0.8–1.6 μm without special effort. The plan for aluminium usually focuses on chatter and chip evacuation, because the metal moves heat into the tool and the chips can weld to the cutter if the toolpath leaves them in the pocket.

Stainless is slower and less forgiving. 304 and 316L work-harden, so a light rubbing pass is worse than a firm cut. The plan should specify a minimum chip load, plenty of coolant and tools kept sharp. 17-4PH adds a second constraint: the H900 condition machines differently from the annealed state, so the sequence must decide where heat treatment sits.

Titanium TC4 (Ti-6Al-4V) and Inconel take the slowest surface speeds and generate the most heat at the cutting edge. Thin ribs deflect more than the same geometry in steel. Plans for these alloys usually reduce radial engagement, increase the number of passes and add a spring pass for the final dimension.

Plastics and carbon fibre flip several rules. POM and PEEK move with temperature, so measurement timing matters; a part measured hot can read oversize and measure correctly an hour later. Carbon fibre wears tools fast and needs dust extraction. The plan should name the coolant, the tool material and the measurement temperature, not just the geometry.

• 1
  Aluminium6061-T6, 7075: fast, stable, watch chip evacuation and chatter
• 2
  Stainless303, 304, 316L, 17-4PH: control work hardening with firm chip loads
• 3
  TitaniumTC4, Inconel: low surface speed, light radial engagement, spring pass
• 4
  PlasticsPOM, PEEK, ABS: thermal growth and measurement timing drive the plan

A plan fails most often at the datum, not at the toolpath. If the datum face is a raw casting surface with 0.5 mm of variation, every dimension measured from it inherits that variation. The fix is to machine the datum first and then re-reference, which costs one setup but removes the error source.

The second common failure is underestimating thin walls. A 1.5 mm wall in aluminium at 300 mm length will sing during finishing unless the plan adds support, reduces radial depth or changes the toolpath direction. Adding a support rib that gets removed later is sometimes cheaper than fighting chatter with slower passes.

A third boundary is measurement. If the drawing calls for ±0.005 mm but the inspection room uses a caliper, the plan cannot be verified. Tight tolerances need a CMM, a micrometer on a controlled temperature part, or a gauge built for that feature. The plan should name the instrument, not just the number.

Finally, plans break when the quantity changes. A plan tuned for one prototype may include extra stock and manual checks that make no sense at 10,000 pieces. At volume, the plan shifts toward fixtures, gauges and cycle-time reduction. Both ends are valid. Mixing them is what causes cost surprises.

• 1
  Raw datumMachine the datum face before measuring anything from it
• 2
  Thin wallsAdd support or change toolpath direction instead of slowing down
• 3
  Measurement gapName the instrument that can actually resolve the tolerance
• 4
  Quantity shiftPrototype plans and production plans are not the same document