CNC machining delivering high quality parts: the 5 basics that decide fit and finish

• 1
  Key takeaways
• 2
  Why CNC machining delivering high quality parts is a stack of errors, not a single operation
• 3
  Tolerance, datums, and how much accuracy a part really needs
• 4
  How setup count and axis choice change the quality of a machined part
• 5
  Tool path, chip load, and the surface finish you can actually expect
• 6
  Material condition: why the same alloy can machine two different ways
• 7
  Inspection and the feedback loop that keeps parts in tolerance
• 8
  Where quality meets cost, lead time, and order size
• 9
  Which process to choose for a given part feature

A machined part is good when it fits the assembly, survives the load, and looks the way the print says. That sounds simple, but every cut adds error. The spindle has runout. The tool wears. The fixture moves a few microns under load. Thermal growth pushes the part as the spindle warms. None of these is large alone. Together they decide whether a bore lands at Ø12.000 mm or Ø12.012 mm.

Engineers often treat quality as a property of the machine. A good machine helps, but the process decides. Two shops can run the same part on the same model of machining center and ship different results. The difference is usually setup strategy, tool path, and how the first part is checked against the drawing.

So the useful question is not "is this shop accurate?" It is "where does this shop spend its accuracy?" A shop that holds ±0.005 mm on critical bores and leaves cosmetic faces at ±0.1 mm is using its budget well. A shop that chases tight tolerance everywhere is adding cost without adding function.

For buyers, this matters before the order. When you read a quote, ask which features carry tight tolerance and why. The answer tells you whether the shop read your print or just ran the clock.

• 1
  Error sources add upMachine, fixture, tool, material, and thermal effects each contribute part of the total deviation.
• 2
  Function sets the targetOnly features that mate, seal, or locate need the tightest tolerance.
• 3
  Process beats machineSetup and tool path choices usually matter more than the machine brand.

Tolerance defines the allowed band. A Ø20 H7 bore accepts 20.000 to 20.021 mm. A shaft at Ø20 h6 runs 19.987 to 20.000 mm. The fit between them is a clearance that a machinist can hit with a boring head and a reamer, then confirm with a bore gauge. When the print says ±0.005 mm instead, the same feature needs a different plan: rougher stock, a smaller finishing allowance, a spring pass, and a temperature-stable measurement.

Datums are the second half of the problem. A position tolerance of Ø0.05 mm means little if the datum faces are not flat or not square. On a part where face A is the datum, the first operation must establish A cleanly. If A is cut after the holes, the holes move with the second setup. This is where most "mystery" out-of-tolerance reports come from.

A practical rule: give tight tolerance to the features that locate or seal, and standard tolerance to everything else. On a typical aluminum housing, that might be two bores at ±0.005 mm and a dozen cosmetic features at ±0.13 mm. The shop can then run the tight features on a finishing pass and the rest on a faster cycle.

GD&T is not decoration. Flatness, perpendicularity, and position callouts tell the machinist how to hold the part and how to inspect it. When a print has no datum scheme, the shop invents one, and the part may fit your fixture but not your assembly.

• 1
  H7/h6 fitsA clearance fit that a boring and reaming sequence can hold and gauge.
• 2
  ±0.005 mmGreatLight's stated machining tolerance for critical features.
• 3
  Datum firstCut and verify the datum surface before any feature measured from it.

Every time a part comes off the fixture, it can shift. A three-axis job on five faces can need five setups. Each refixturing adds position error, and the errors stack in the same direction on every face. A five-axis machine tilts the tool and the table instead, so the part stays clamped while the spindle reaches the other side. Fewer setups means fewer chances to lose position.

GreatLight runs 16 simultaneous 5-axis machining centers, 12 four-axis mills, 27 three-axis machines, and 16 mill-turn centers. That mix matters because axis count is a cost decision, not a quality badge. A simple bracket with holes on one face runs faster and cheaper on a three-axis mill. A manifold with ports on four sides is a five-axis job, because the alternative is four setups and four chances to drift.

The same logic applies to turning. A mill-turn center can produce a shaft with cross-holes and flats in one cycle. The alternative is a lathe operation, a mill operation, and a fixture that must hold concentricity between them. For parts with a tight runout callout, one cycle is usually the safer route.

Setup also includes workholding. Thin walls deflect under clamping pressure. Long parts need support. Soft jaws and custom fixtures cost money up front, but they hold the same position for the whole run. For 10,000-piece runs, a modest fixture often pays back in scrap reduction alone.

• 1
  Three-axisBest for prismatic parts with features on one or two faces.
• 2
  Five-axisBest when features sit on many faces or at compound angles.
• 3
  Mill-turnBest for round parts with cross features and tight concentricity.

Surface finish is the visible result of cutting mechanics. Ra 1.6–3.2 μm is a normal as-machined finish on aluminum with a sharp end mill. To reach Ra 0.8–1.6 μm, the shop reduces stepover, raises spindle speed within the tool's limit, and keeps the cutter engaged. Ra 0.2–0.8 μm usually needs a finishing pass with a smaller tool or a different operation, such as fine boring or grinding.

Chip load is the other lever. If the feed per tooth is too low, the tool rubs instead of cutting. That work-hardens stainless and burns the edge. If the feed is too high, the tool deflects and the wall goes out of tolerance. For 6061 aluminum, a 12 mm end mill might run at 0.05–0.10 mm per tooth. For 316 stainless, the same cutter runs much slower with a lower chip load and more coolant.

Tool path strategy matters for thin features. A trochoidal path keeps radial engagement low, which reduces cutting force and heat. A conventional slotting pass loads the full width of the cutter and pushes the wall away. On a 1.5 mm wall, that difference decides whether the part stays flat or bows.

The practical point for engineers: specify the finish you need, not the finish that sounds best. A sealing face may need Ra 0.8 μm or better. A cosmetic panel may look fine at Ra 3.2 μm after bead blasting.

• 1
  As-machinedRa 1.6–3.2 μm, normal for most milled faces.
• 2
  Fine finishRa 0.8–1.6 μm, needs a controlled finishing pass.
• 3
  Mirror-levelRa 0.2–0.8 μm, limited to specific faces and operations.

Two bars of 6061 aluminum are not always the same part. Temper matters. 6061-T6 is strong and machines cleanly with sharp tools. Annealed 6061 is gummy and builds up on the edge. The same program produces different finishes and different dimensional drift. This is why the material certificate and the temper callout belong on the drawing.

Stainless is the classic problem. 304 work-hardens quickly, so a light finishing pass on a previously cut surface can be harder than the first cut. 316L behaves similarly and adds a corrosion requirement. 17-4PH in the H900 condition machines very differently from the solution-treated state. The shop has to know which condition arrives at the door.

Titanium Ti-6Al-4V (TC4) has low thermal conductivity, so heat stays in the cutting zone. Tools need lower speeds, higher coolant pressure, and sharp edges. Inconel is harder still. These materials are chosen for function, not for machinability, so the process must be planned around them rather than corrected later.

Plastics bring their own rules. POM and PEEK machine cleanly but move with temperature. ABS and PC can chip or melt. Carbon fiber wears tools fast and creates dust that needs control. The finishing pass on a plastic part often decides whether the edge is clean or frayed.

• 1
  Aluminum6061-T6 and 7075 cut cleanly; annealed grades can smear.
• 2
  Stainless304 and 316L work-harden; keep the cutter engaged.
• 3
  TitaniumTC4 holds heat; plan for lower speeds and more coolant.

A tolerance claim is only as good as the measurement behind it. GreatLight inspects 100% of parts before shipment, with raw material checks, in-process monitoring, and a final inspection. Reports are available on request. That sequence matters because catching a drift at the machine is cheaper than catching a full lot at final inspection.

In-process checks close the loop. The operator measures the first part, records the result, and adjusts the offset before the second part. On a tight bore, that might mean measuring every fifth part and watching the trend rather than the absolute number. Tool wear shows up as a slow drift. If the trend is visible, the offset can be corrected before the part leaves tolerance.

Final inspection uses the method the drawing implies. A caliper is fine for a ±0.13 mm feature. A ±0.005 mm bore needs a bore gauge or a coordinate measuring machine. Surface finish needs a profilometer, not an eyeball. If the measurement method does not match the tolerance, the inspection report is a guess with a number on it.

For regulated industries, the paperwork matters as much as the part. ISO 9001:2015, IATF 16949:2016, ISO 13485:2016, and ISO 27001:2022 cover quality management, automotive, medical devices, and information security. These certificates tell a buyer which systems are in place. They do not replace part-specific inspection data.

• 1
  First-part checkVerify the setup before the run continues.
• 2
  Trend monitoringWatch drift, not just the last number.
• 3
  Method matchUse a gauge that resolves at least 10% of the tolerance.

Tight tolerance costs money, but not in a straight line. Going from ±0.1 mm to ±0.05 mm may add a finishing pass and a few minutes per part. Going from ±0.05 mm to ±0.005 mm may add a temperature-controlled measurement step, a dedicated fixture, and a slower cycle. The second step often costs more than the first, and it only makes sense on features that need it.

Order size changes the math. A single prototype is programmed, set up, and machined once. A 10,000-part run spreads the setup and fixture cost across every piece, so the per-part price drops. GreatLight has no minimum order quantity and runs from one prototype to 10,000+ part runs. For prototypes, the goal is to prove the design. For production, the goal is repeatability.

Lead time depends on the same factors. Quotation and a free DFM analysis come back within 12 hours, production can start within 24 hours, and parts ship in 3–5 days. That timeline holds when the material is in stock and the print is clear. A missing datum or an ambiguous finish callout can add a review cycle before the first chip is cut.

The cheapest way to improve quality is often a DFM review. If a wall is too thin, a corner radius is too small, or a tolerance is tighter than the function needs, the shop can say so before the run. That conversation costs nothing and often removes a problem from the process.

• 1
  DFM firstFree analysis within 12 hours, before the run starts.
• 2
  No MOQOne prototype or 10,000+ parts, same process discipline.
• 3
  3–5 day shippingWhen material is in stock and the drawing is clear.