The Art of CNC Processing: How Tolerance Is Actually Won
This page explains what really decides the outcome of a machined part: the interaction between tool path, fixturing, thermal drift and inspection. It is written for design engineers and buyers who need to judge whether a drawing is machinable, and where the risk sits before a quote is signed.
What the art of CNC processing really controls
CNC processing is subtractive: a rotating cutter follows a programmed path and removes material until the remaining solid matches the model. The machine does not know what the part is for. It only knows coordinates, feed rate and spindle speed. Everything the drawing asks for, from a bore position to a sealing face, is the result of those three variables staying inside a window long enough to finish the cut.
That is why the art of CNC processing is less about the controller and more about what happens around it. A 0.01 mm error in a fixture can double the position error of a hole. A 2 °C shift in the shop changes the size of a 300 mm aluminum part by roughly 0.005 mm. Neither shows up in the G-code. Both show up on the CMM report.
The practical consequence is that tolerance is not a number you buy from a machine spec sheet. It is a budget you spend across setup, cutting, heat and measurement. When a drawing asks for ±0.005 mm on a bore and ±0.1 mm on the bolt circle, the tight callout gets the rigid setup, the probe check and the temperature-controlled window. The loose callout gets a soft jaw and a roughing pass.
Material choice sets the ceiling. Aluminum 6061 and 7075 cut freely and hold size well. Stainless 316 and 17-4PH work-harden if the feed is too light, so the finishing pass must stay above the work-hardened skin. Titanium Ti-6Al-4V and Inconel move more under heat and dull tools faster, which pushes cost up and pushes the achievable surface finish toward Ra 1.6 μm unless the shop slows down and accepts longer cycle time.
Tool path, stepover and the finish you can promise
Surface finish in milling is geometric before it is metallurgical. The cutter leaves scallops whose height depends on the corner radius of the tool and the stepover between passes. A Ø10 mm ball nose at 0.2 mm stepover leaves a scallop height near 1 μm. Push the stepover to 0.5 mm and the scallop grows roughly six times, which lands you in the Ra 1.6–3.2 μm band. No amount of spindle speed fixes geometry.
Contouring strategy matters just as much on walls. Constant-engagement paths keep the radial cut width steady, so the tool load does not spike in corners. That is what allows a thin wall to stay straight. On a 1.5 mm aluminum wall, a full-width radial cut in a corner will push the wall outward and leave it bowed. A trochoidal path with 8–10% radial engagement holds the wall within a few hundredths of a millimeter.
Five-axis motion earns its place when the feature cannot be reached in three axes, or when a single setup removes more error than it adds. Drilling a pattern of holes on a compound angle from one orientation avoids four re-clamps. Each re-clamp adds its own position error. Fewer setups is usually tighter, not looser.
Roughing and finishing want different tools. A Ø16 mm end mill at 2 mm axial depth clears material fast and leaves a stepped surface. A Ø6 mm finishing tool at 0.3 mm depth follows the final shape with less deflection. Skipping the finishing tool to save a setup is the most common reason a part arrives at Ra 3.2 μm when the drawing asked for Ra 0.8 μm.
Fixturing and setup: where most tolerance is lost
A vise is a spring. Clamp a thin plate on two edges and the middle lifts; machine it flat and release it, and the part springs back into a bow. Soft jaws bored to the actual workpiece diameter distribute the load around the circumference instead of pinching two points. For thin plates, vacuum chucks or a bed of sacrificial material under the part does the same job on the Z axis.
Datum selection decides how errors accumulate. The best datum is a machined face or a bored hole that the machine itself produced, because it shares the same coordinate frame as the features cut later. A raw cast surface is a poor datum: it may vary by 0.5 mm from part to part, and every downstream dimension inherits that variation.
Workholding also sets the access problem. A part held in a three-jaw chuck cannot be cut on the jaw side without a second operation. When a feature wraps around the part, a mill-turn center or a five-axis setup with a Ø400 mm rotary table keeps the part in one frame and eliminates the re-fixture error entirely.
Tool length adds a quieter error. A long tool in a deep cavity deflects under cutting force. The deflection shows up as a taper in the wall, not as a random error, so it can be predicted and compensated. Keep the tool as short as the geometry allows, and add a spring pass at the same depth when the wall tolerance is tight.
Thermal drift, measurement and what the report proves
Metal grows when it warms. Aluminum expands about 23 × 10⁻⁶ per °C, steel about 12 × 10⁻⁶. A 200 mm aluminum part that warms 5 °C during roughing grows 0.023 mm. If it is measured hot, the operator chases a number that will not exist at 20 °C. The fix is boring but effective: finish the part, let it stabilize, then measure.
In-process probing closes the loop between the model and the actual part. A touch probe can find a cast boss within a few micrometers and shift the work offset before the finish pass. That is how a shop holds ±0.005 mm on a feature whose blank position varies from part to part. It is also how a first article gets verified without cutting a second one.
Final inspection answers a different question. It tells you whether the parts that shipped match the drawing, not whether the process is stable. A 100% inspection before shipment catches outliers. Dimensional reports on request let the buyer trace a specific lot. Neither replaces the statistical picture you get from measuring the same feature across a run.
Surface finish is measured, not eyeballed. A profilometer gives Ra directly, and the Ra 0.8–1.6 μm band is the usual target for sealing faces and sliding surfaces. Decorative surfaces can run at Ra 1.6–3.2 μm as machined. Where a surface is both cosmetic and functional, agree on the measurement location in writing before the first cut.
Which setup fits which part
Match the feature to the process, not the other way around.
| Part situation | Recommended setup | Typical tolerance | Watch out for |
|---|---|---|---|
| Prismatic block, 3 accessible faces | 3-axis vise with soft jaws | ±0.02 mm | Spring-back on thin floors |
| Compound-angle holes, one orientation | 5-axis with Ø400 mm rotary table | ±0.01 mm | Rotary backlash on reversal |
| Turned shaft with cross holes | Mill-turn center | ±0.01 mm | Tool interference at the cross feature |
| Thin plate, 1.5 mm wall | Vacuum chuck, trochoidal path | ±0.03 mm | Chatter at the unsupported edge |
| Large frame, 4,000 mm travel | Gantry setup, probe datum | ±0.05 mm | Thermal growth over a long cycle |
| Cast blank with variable boss | Probe work offset, 4-axis | ±0.005 mm | Probe stylus wear over a run |
| Prototype, one piece | 5-axis, no dedicated fixture | ±0.02 mm | Setup time dominates cost |
| Hardened tool steel insert | 3-axis, carbide, spring pass | ±0.01 mm | Work-hardened skin from light feed |
When to spend on setup, and when to spend on inspection
If the tight callout sits on one or two features, pay for probing and a rigid fixture and leave the rest at ±0.05 mm. If tight tolerance runs across the whole part, pay for temperature control and a single-setup five-axis process instead. Trying to buy both from a three-axis vise is how a project misses its window.
Questions engineers ask before releasing a drawing
Can ±0.005 mm be held on every feature of a part?
Only where the setup supports it. That tolerance is realistic on a bored hole or a ground face measured at 20 °C, with probing and a rigid fixture. On a long unsupported wall, or across a 4,000 mm frame, thermal drift and tool deflection push the practical limit to ±0.05 mm.
The honest approach is to mark the two or three features that matter and leave general dimensions looser. A drawing that asks for ±0.005 mm everywhere costs more and does not improve function.
Does a smoother surface always cost more?
Yes, past a point. Going from Ra 1.6 μm to Ra 0.8 μm usually means a finer stepover, a smaller finishing tool or a slower feed, all of which add cycle time. Going below Ra 0.2 μm in metal generally needs polishing or lapping, which is a separate operation.
Where the surface is not sealing or sliding, Ra 1.6–3.2 μm as machined is enough and keeps the part cheaper.
Why does my part measure in tolerance at the shop and out of tolerance in my lab?
Temperature and datum are the usual reasons. A part measured at 25 °C reads differently than at 20 °C, and the difference scales with size: roughly 23 μm per meter per 5 °C on aluminum.
The second reason is datum choice. If the shop measures from a machined bore and your lab measures from an unmachined edge, both are correct and both give different numbers. Agree on the datum in the drawing.
When is five-axis worth the cost over three-axis?
When the feature needs more than one orientation, when each re-clamp would add error you cannot afford, or when the part is too heavy to move between setups. On a simple prismatic part with three accessible faces, three-axis is faster and cheaper.
A rough rule from the shop floor: if a part needs three or more setups in three axes, the five-axis version usually lands at similar cost and tighter tolerance.
What does a dimensional report actually show?
It shows the measured value of the dimensions you list, taken with a calibrated instrument, with the measurement temperature noted. It proves the shipped lot against the drawing.
It does not prove process capability. For that you need the same feature measured across many parts, which is a different conversation and usually a different document.
How does material choice change the achievable finish?
Free-cutting aluminum and brass hold a fine finish easily. Stainless 316 and 17-4PH work-harden, so the finish pass must cut under the hardened skin or the tool rubs and tears. Titanium and Inconel heat the edge faster, so finish tends to sit one band coarser for the same cycle time.
If a drawing calls for Ra 0.8 μm on Inconel, expect a longer cycle and a fresh tool for the finishing pass.
Send the drawing and get a machinability read
We review your model, flag the features that carry real risk, and quote from a process plan rather than a guess. Quotation and free DFM analysis within 12 hours.
12-hour quote100% inspectionNo minimum order quantity
More process notes from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
