Master CNC precision machining of high-quality parts
This page explains what actually sets the accuracy floor in CNC precision machining of high-quality parts: machine geometry, workholding, thermal drift, tool pressure and metrology. It is written for design engineers and buyers who need to judge whether a drawing is machinable at the quoted tolerance, or whether the tolerance is only a wish. By the end you should be able to read a print and predict which features will be hard.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
- 7
CNC precision machining of high-quality parts: what the process controls
CNC precision machining is subtractive. A rotating or stationary cutting tool removes material from a solid block, and the machine follows a program of coordinates. Precision comes from two separate things that are often confused: repeatability and accuracy. Repeatability is how tightly the machine returns to the same point. Accuracy is how close that point is to the drawing. A machine can repeat perfectly and still cut the wrong size.
For high-quality parts, the drawing tolerance is the contract. If a bore is called at Ø20.000 mm with a ±0.005 mm band, the shop must hold 19.995 to 20.005 mm on every piece, not on the best one. That single number drives machine choice, fixture design, cutting strategy and how often the operator stops to measure.
Three error sources dominate. Geometric error lives in the machine itself: squareness, straightness, spindle runout. Elastic error comes from cutting force pushing the tool and the part apart. Thermal error comes from spindle heat, chips and ambient air changing dimensions during the run. On a 100 mm aluminum part, a 5 °C rise moves the material about 0.012 mm. That is more than the whole tolerance band.
The practical meaning: tight tolerances are not bought with a better machine alone. They are held by controlling the three error sources at the same time. A shop that only talks about spindle speed is usually not the shop that holds ±0.005 mm.
Which machine geometry suits which part
Three-axis machines cut prismatic parts: plates, housings, brackets. Access comes from one direction, so every feature must be reachable from the top. They are fast and rigid, and for flat parts with holes and pockets they are often the correct answer.
Four-axis machines add a rotary table, usually Ø400 mm class. A part can be indexed to four sides without a second setup. This removes the re-zeroing error that creeps in when a part is flipped by hand. Shafts with cross holes, and parts with features on all four sides, belong here.
Five-axis machines tilt the tool or the table on two axes at once. The real gain is not shape freedom. It is that a short, stiff tool can reach a deep feature in one setup. A 3-axis cut of a deep cavity needs a long tool, and long tools deflect. Five-axis access lets you use a stubby tool, which holds size better.
Mill-turn centers combine turning and milling in one spindle. Parts that are round with milled flats, slots or cross holes finish in a single setup, so concentricity between the turned diameter and the milled feature is set by the machine, not by two fixtures. Our floor runs 127 high-precision CNC machines, including 16 simultaneous 5-axis centers, 16 mill-turn centers and a 4,000 mm maximum processing size for long parts.
Workholding is where most tolerance is lost
A part is only as stable as the fixture holding it. The first rule is to clamp onto a surface that does not matter, or onto sacrificial stock. Clamping on a finished face leaves a mark and can distort the wall behind it.
The second rule is support under the cut. Thin floors and tall walls vibrate. Vibration shows up as chatter on the surface and as a size that drifts along the wall. Adding a support jack, reducing axial depth of cut, or changing the tool path entry angle usually fixes it without changing the machine.
Vacuum chucks and soft jaws spread clamping force over a wider area than a vise. For thin plates and rings, that difference decides whether the part stays round. For small runs we often machine the jaws in place on the machine, so the jaw profile matches the part to within the machine's own repeatability.
Never assume a fixture is rigid because it is heavy. Rigidity is about the load path from the cutting edge to the machine table. A heavy block held by two bolts at one end will still move.
How a precision process is run, step by step
A stable process starts before metal is cut. The material is checked against the certificate. Bar stock with internal stress will move after the first cut, so for tight parts we rough, stress-relieve if the material allows, then finish. Aluminum 7075 and 17-4PH stainless are the usual offenders.
Cutting parameters are set from the tool maker's data, then trimmed for the part. High spindle speed with a light radial engagement keeps cutting force low, which reduces elastic error on thin walls. That is the reason high-speed strategies hold size better than heavy conventional passes.
In-process measurement closes the loop. The operator measures a feature, compares it to the nominal, and offsets the tool. On a long run this happens at fixed intervals, not only when a part looks wrong. Tool wear is gradual and predictable, so a scheduled offset beats a reaction.
Temperature is managed, not ignored. Warm-up cycles run before the first cut. Coolant is aimed at the cut, not at the whole machine. For the tightest work, the finishing passes are done after the machine has reached thermal equilibrium, which can take an hour or more on a large machine.
Reading tolerances and surface finish together
A tolerance tells you the allowed size band. Surface finish tells you how the surface was made. They interact. A turned surface has a visible helix, and its roughness depends on feed per revolution and tool nose radius. A milled surface carries a step pattern from the cutter path.
Ra is the arithmetic average of profile height. On a finishing pass with a sharp tool and a light feed, aluminum and brass reach Ra 0.2–0.8 μm. General machined surfaces sit at Ra 0.8–1.6 μm. As-machined surfaces without a finishing pass are Ra 1.6–3.2 μm.
Do not specify a fine finish on a face that will be painted, and do not specify a coarse finish on a sealing face. A seal needs a smooth, continuous surface in one direction. A paint key needs a slight tooth. The drawing should say which faces matter.
Tolerance stacking is the other trap. If a part has three dimensions in a chain, each with its own tolerance, the end-to-end error is the sum. Designers sometimes call ±0.01 mm on each link of a chain and expect ±0.01 mm overall. That is not how it works.
Inspection and what the numbers prove
Inspection is only useful if it can see the feature that matters. Calipers read outside diameters and lengths. Micrometers read them more precisely. Bore gauges and pin gauges read holes. For position and form, a coordinate measuring machine or an optical comparator is needed.
A report is not a guarantee by itself. What matters is that the same measurement method is used at the machine and at final inspection. If the operator checks a bore with a pin gauge and the inspector checks it with a CMM, a disagreement between the two methods has to be resolved before the parts ship.
GreatLight measures 100% of parts before shipment, with raw material checks, in-process monitoring and final inspection. Reports are available on request. The historical qualification rate on this process is 99.99%, which is a floor set by the inspection plan, not by luck.
For regulated industries the paper trail matters as much as the number. Aerospace, medical and automotive programs need traceability from the material certificate to the final report. That is why our plants hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022.
Choosing the process for the feature
Match the feature to the method before you ask for a price.
| Feature | Recommended method | Hold on this | Watch out for |
|---|---|---|---|
| Flat plate, holes, pockets | 3-axis milling | ±0.01 mm typical | Reach from one side only |
| Features on four sides | 4-axis with rotary table | ±0.005 mm on position | Indexing error between sides |
| Deep cavity, short tool needed | 5-axis simultaneous | ±0.005 mm, Ra 0.8 μm | Program cost, setup time |
| Round part with milled flats | Mill-turn center | Concentricity held in one setup | Bar size limits |
| Thin wall under 1 mm | Light radial cut + support | Wall thickness ±0.02 mm | Chatter, spring-back |
| Hole under Ø1 mm | Micro drilling, peck cycle | ±0.01 mm on diameter | Tool breakage, chip packing |
| Sealing face | Fine finish pass | Ra 0.2–0.8 μm | Directional lay must be unbroken |
| Long shaft over 1,000 mm | 4,000 mm travel machine | Straightness 0.05 mm/m | Sag between centers |
Where the money actually goes
If the feature is flat and reachable from one side, a 3-axis machine holds your tolerance and costs less. If the feature is deep, thin-walled or needs four-sided position in one setup, pay for 4- or 5-axis access and a machined fixture. Buying a tighter tolerance on a part that does not need it buys nothing.
Questions engineers ask before releasing a print
Can a ±0.005 mm tolerance be held on a 300 mm part?
On a short, stiff part in aluminum or brass, yes, when the machine is thermally settled and the finishing pass is light. On a 300 mm part the same band is much harder because thermal growth and fixture deflection both scale with size.
A practical approach is to keep the tight band on the few features that need it and open the rest. If the whole part is called at ±0.005 mm, the price reflects the inspection time, not the cutting time.
Does a finer surface finish always mean a better part?
No. A very smooth surface can be worse for adhesive bonding or for paint adhesion. For a sliding seal, a controlled lay direction matters more than the lowest Ra number.
Tell us how the surface is used. We will set the finish range around that, for example Ra 0.8–1.6 μm for a general wear face.
How does material choice change the achievable tolerance?
Aluminum cuts freely and moves with temperature, so it reaches tight size but needs thermal control. Stainless 316 work-hardens, so light passes and sharp tools matter. Titanium TC4 and Inconel generate heat at the edge and deflect tools, so tolerances are usually opened.
Plastics such as POM and PEEK move after machining. For those, we often leave stock and do a final light pass after the part has relaxed.
What happens to internal stress when a part is machined from bar?
Removing material releases the stress that was locked in the bar. The part bends or twists. The fix is to rough, let the part settle, then finish. For critical parts we may specify stress-relieved stock.
This is why a part can measure correctly off the machine and be out of tolerance a day later. If the drawing is tight, tell us and we will plan the sequence around it.
How do you handle a part that needs both turning and milling?
A mill-turn center does both in one setup, which protects concentricity between the turned diameter and the milled features. If the part goes to two machines, the second setup introduces a new zero point and that error adds to the total.
For runs above a few hundred parts, a dedicated fixture for the second operation is usually worth the cost.
Can you work from a 3D model only, without a drawing?
Yes, but the model must carry the tolerances. A STEP file with no GD&T tells us nominal geometry only, so we have to assume general tolerances. That is fine for prototypes and often wrong for production.
Send the model with a drawing that marks the critical features. We return a DFM analysis with the quotation within 12 hours.
Send the print and get a real answer on tolerance
Upload your CAD and drawing, and we return a quotation with free DFM analysis within 12 hours. No minimum order quantity, from one prototype to 10,000+ part runs, and an NDA on request.
12-hour quote100% inspectionNo MOQNDA available
More process notes from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
