High-precision machining of small parts: a guide
Small parts fail on stiffness, not on machine accuracy. This guide covers the mechanics behind high-precision machining of small parts, the tolerance floor you can realistically hold, and the point where a process change beats another pass.
Why high-precision machining of small parts is harder
A 40 mm bracket and a 400 mm bracket can be cut on the same machine, but they do not behave the same way. Cutting force is roughly constant for a given chip load, while the part's resistance to that force falls with the cube of its smallest dimension. Halve the wall thickness and the deflection under the same load goes up eight times. That is the whole problem in one sentence.
Small parts also have fewer places to clamp. A large casting can be bolted at six points and stay flat. A 12 mm connector body may only offer 3 mm of land. Every clamp you add distorts the shape you just cut, so the sequence of clamping and releasing matters as much as the toolpath.
Thermal effects scale in the opposite direction. A small part heats fast and cools fast. A 5 mm end mill at 18,000 rpm dumps most of its heat into a chip that weighs almost nothing, and the remaining heat goes into a workpiece with very little mass to absorb it. Growth of 8–12 μm during a roughing pass is normal, and it disappears again before the finishing pass.
None of this is a machine specification problem. A simultaneous 5-axis machining center holds its own geometry to microns. The error budget for high-precision machining of small parts is dominated by the part, the fixture and the tool, in that order.
Tool runout, edge radius and the real cutting edge
Runout is the first number to check. A 3 mm carbide end mill with 10 μm of runout cuts with one flute doing most of the work. Feed per tooth then varies from near zero to double the programmed value, and the surface finish shows it as a repeating pattern at the flute frequency.
The edge radius matters more as the tool gets smaller. A 6 mm tool with a 5 μm edge radius still shears material cleanly. The same edge on a 1 mm tool is a large fraction of the diameter, so the tool rubs before it cuts. That rubbing generates heat, work-hardens stainless and pushes the part away from the cutter.
Spindle speed is not free. Below about 12,000 rpm a small tool has enough surface speed to cut aluminium well but not enough for titanium. Above 20,000 rpm the tool holder and the balance grade become the limit. HSK-E25 and shrink-fit holders hold runout under 3 μm at those speeds.
We keep 16 simultaneous 5-axis machining centers and 16 mill-turn centers for this reason. Turned features on a small part usually belong on a lathe with live tooling, not on a three-axis mill with a rotary table bolted on.
Workholding: where small-part accuracy is won or lost
The fixture is usually the largest single error source on a small part. A vise with 2 kN of clamping force will close a thin-walled aluminium ring by 20–40 μm. Cut the bore, release the vise, and the bore is no longer round.
Five-sided machining in one setup removes most of this. If the part is held on a dovetail or a sacrificial stub and all critical features are cut before the stub is removed, there is no second datum to re-establish and no second clamping distortion. This is the main reason we run small parts on 5-axis rather than as a sequence of 3-axis operations.
Vacuum chucks and fixture plates with M6 grid holes work well for flat parts down to about 0.8 mm thick. Below that, the plate itself starts to bow under vacuum and the flatness spec has to be renegotiated.
For parts under 5 mm in any dimension, a sacrificial carrier is often the only practical answer. Machine the carrier and the part together, then cut the tabs. It costs one extra operation and removes an entire class of scrap.
The tolerance floor and how to read it
A general-purpose CNC shop can hold ±0.05 mm on a small part without much thought. Holding ±0.005 mm on a 10 mm feature is a different job. It requires a temperature-controlled room, a machine that has been running long enough to be thermally stable, and a metrology loop that feeds back into the offsets.
The floor is not a single number. It depends on feature size, material and aspect ratio. A 10 mm aluminium bore at ±0.005 mm is routine. A 2 mm deep slot in 17-4PH at the same tolerance is not, because the tool deflects more than the tolerance.
Surface finish follows the same logic. Ra 0.8–1.6 μm comes off the machine with a correct feed and a sharp tool. Ra 0.2–0.8 μm usually needs a finishing pass with a smaller stepover or a light burnishing pass, and it costs time.
Ask what the tolerance is doing. If it locates a bearing, it is real. If it is copied from a title block, it may be costing 30 percent of the part price for nothing.
When another process wins
CNC is not always the answer for small parts. If the part is a thin, flat, high-volume stamping, sheet metal fabrication will beat milling on cost per piece by a wide margin, and the tolerance is set by the die, not by the machine.
If the geometry is a lattice or an internal channel that no tool can reach, 3D printing is the practical route. Metal die casting makes sense once a small part is needed in the tens of thousands and a tooling investment is justified.
Vacuum casting sits in the middle for low-volume urethane parts. It gives a cast surface and a moulded shape without tooling, and it is useful when a small enclosure has to look like the production part before the mould exists.
The honest boundary for CNC is this: one to a few thousand parts, tight tolerances on reachable features, and materials that cut. Outside that, we say so and point to the process that fits. We run no minimum order quantity, so a single prototype and a 10,000+ part run both go through the same inspection route.
Choosing the process for a small part
Match the feature size and tolerance to the process before you quote.
| Part feature | Typical tolerance | Process to use |
|---|---|---|
| Ø10 mm bore, aluminium | ±0.005 mm | Mill-turn, one setup |
| 0.8 mm wall, flat plate | ±0.02 mm | Vacuum plate, light finishing pass |
| 2 mm slot in 17-4PH | ±0.01 mm | 5-axis, small stepover, sharp tool |
| Ø1.5 mm cross hole | ±0.02 mm | Mill-turn with live tooling |
| Thin ring, OD ground | ±0.005 mm | Turn oversize, then grind |
| Freeform 5-face feature | ±0.01 mm | Simultaneous 5-axis, dovetail stub |
| Sub-5 mm micro part | ±0.01 mm | Sacrificial carrier, then detab |
The short version
If the tolerance is real and the feature is reachable, machine it on 5-axis in one setup and pay for the fixture. If the tolerance is inherited or the volume is high, change the process instead of buying more precision.
Small-part machining questions
What is the smallest part you can machine?
There is no fixed size limit. The practical limit is the ratio of feature size to tolerance. Parts under 5 mm in their largest dimension are usually cut on a sacrificial carrier and then detabbed.
Send the drawing and we will tell you in the DFM review whether the tolerance can be held on the stated features, and which ones need a note.
How do you control thermal growth on a long run?
The machine runs a warm-up cycle before the first cut, and the room is held at a stable temperature. In-process probing corrects offsets as the spindle grows.
For a 10 mm aluminium feature, growth of 8–12 μm during roughing is expected and is removed by the finishing pass, provided the finishing pass happens after the part has stabilized.
Can you hold ±0.005 mm on stainless?
Yes, on features where the tool has enough stiffness. A 10 mm bore in 316L at ±0.005 mm is routine. A 2 mm wide slot in 17-4PH at the same tolerance is not, because tool deflection exceeds the tolerance.
We would propose ±0.01 mm on that slot or a change to a wider geometry.
Do you inspect every small part?
Yes. We inspect 100 percent of parts before shipment, with raw material checks, in-process monitoring and a final inspection. Inspection reports are available on request.
For high-volume runs we use sampling plans agreed with the customer, backed by the same in-process monitoring.
What lead time should I expect?
Quotation and a free DFM analysis come back within 12 hours. Production can start within 24 hours of approval, and parts ship in 3–5 days.
On our historical data the late-delivery probability is below 2 percent. We do not promise fixed dates before the drawing is reviewed.
Can you work from a customer fixture or datum scheme?
Yes. Send the datum callouts and any existing fixture model. If the datums cannot be reached in one setup, we will propose an alternative and show why.
Uploads are secure and confidential, and an NDA is available on request.
Send the drawing, get a straight answer
Upload your small part and we will return a quotation with a free DFM analysis within 12 hours, including which tolerances can be held and which should be relaxed.
12-hour quote100% inspectionNo minimum order quantity
More from the shop floor
We publish setup notes, tooling trials and inspection data from the factory floor.
