CNC Machining Small Batch Parts: Where the Process Actually Breaks

• 1
  Why CNC Machining Small Batch Work Is a Different Job
• 2
  How Fixturing and Thermal Drift Move Your Tolerance
• 3
  When 5-Axis Pays Off in Small Batch, and When It Does Not
• 4
  Sampling Plans That Catch Drift Before the Batch Is Gone
• 5
  Where Small Batch Machining Stops Making Sense
• 6
  Setup Rules That Keep a Small Batch Repeatable
• 7
  Choosing a Process for CNC Machining Small Batch Parts

A prototype is one part. You can shim it, tap it into place, adjust the cutter offset by hand and still ship something that fits. A production run of 10,000 parts spreads the setup cost across so many pieces that a dedicated fixture and a long cycle time are easy to justify. Small batch sits between the two, and that is exactly where most tolerance failures happen.

The number that governs everything is the ratio of setup time to run time. In a 50-piece order, a fixture that takes three hours to build adds over three minutes of cost to every part. In a 5,000-piece order the same fixture adds under two seconds. So the shop has to choose between a cheap setup that drifts and an expensive setup that does not pay for itself.

There is a second ratio that matters just as much: the number of distinct operations divided by the number of parts. Twenty parts that need milling on four faces, two drilled hole patterns, a reamed bore and a tapped end face carry more fixturing than 200 parts with one operation. Batch size alone tells you very little. Operation count per part tells you almost everything.

This is why a quote for 30 parts can come back higher per piece than a quote for 300 parts from the same shop, on the same drawing, in the same material. It is not a pricing trick. The 30-piece order simply carries more setup and more program proving per part, and no amount of volume discounting changes that arithmetic.

• 1
  Setup dominates below 100 partsProgram proving and first-article checks are a fixed cost.
• 2
  Operation count beats part countFive operations on 20 parts is heavier work than one operation on 200.
• 3
  Per-piece price falls in stepsNot a smooth curve, because fixtures are bought in whole units.

Every batch starts with a cold machine and a cold fixture. As the spindle runs, the headstock, ballscrews and fixture all grow. On a 750 mm travel machine, a 5 °C rise over a four-hour run can move the tool point by 15 to 30 μm in the worst direction. That is three to six times the ±0.005 mm we hold on a good day, and it is the single most common reason a first article passes and the fortieth part does not.

The fix is not exotic. Run a warm-up cycle before the first cut, keep the same spindle speed across the batch instead of optimizing each part, and check the critical dimension on parts 1, 10 and 30 rather than only at the end. If the trend is drifting one way, the operator adjusts the offset once instead of scrapping a batch.

Workholding matters more in small batch than in mass production, because you cannot amortize a dedicated hydraulic fixture. A vise with soft jaws machined in place will hold 20 μm repeatability on a part with a good clamping face. The same vise on a thin-walled part will deflect the wall and spring back after unclamping, and the part will measure oversize on the outside and undersize on the inside.

Thin walls are the classic failure mode. Below roughly 1.5 mm wall thickness in aluminium, or 1.0 mm in stainless, the cutting force pushes the wall away from the cutter, the tool takes a lighter chip, and the finished wall is thicker than the program says. Rough the wall leaving 0.3 mm, let the part cool, then finish with light radial engagement. Two passes cost less than one scrapped batch.

• 1
  Warm up before part oneA 15-minute spindle warm-up removes most of the first-hour drift.
• 2
  Measure at intervals, not at the endParts 1, 10 and 30 show the trend while you can still correct it.
• 3
  Rough, cool, then finishSpringback in thin walls is a heat and force problem, not a tool problem.

Five-axis machining looks expensive per hour, and it is. A simultaneous 5-axis center costs more to run than a three-axis mill. The reason it wins on small batch is that it removes setups. A part that needs four faces machined goes from four setups on a three-axis machine to one on a five-axis machine with a trunnion table. Fewer setups means fewer fixture builds, fewer datums to chase and fewer places for a stack-up error to enter.

The break-even is roughly three faces. Below that, a three-axis machine with a simple vise is faster and cheaper. At four or more faces, or when the part has features at compound angles, five-axis usually wins even at 20 pieces. The deciding factor is not the geometry itself but how many times the part has to be released and re-clamped.

There is a real limit though. Five-axis work on a part that fits in a 500 × 500 × 450 mm envelope is comfortable. Push toward the 4,000 mm maximum processing size and the machine choice narrows, the fixture gets heavy, and thermal growth over a long cycle becomes the dominant error source rather than the cutter path.

Materials shift the decision too. Aluminium 6061-T6 and 7075 cut fast and forgive a light fixture. Titanium TC4 and Inconel 718 do not. They generate heat at the cutting edge, work-harden if the tool rubs, and need rigid setups and conservative parameters. On those materials, a small batch is often better split into two operations on separate machines than run in one long unattended cycle.

• 1
  Count faces, not featuresThree or fewer machined faces: stay three-axis.
• 2
  Compound angles favor 5-axisA trunnion removes the trigonometry from the setup.
• 3
  Rigid materials need rigid setupsTitanium and Inconel punish any flex in the fixture.

100% inspection before shipment is the floor, not the strategy. The useful question is what you measure during the run, because a batch of 300 parts that all fail final inspection is a total loss. In-process monitoring is what turns a bad batch into a corrected one.

For a typical batch we check the first article fully against the drawing, then watch one or two critical dimensions every tenth part. If the dimension is inside a third of the tolerance band, the operator keeps running. If it drifts past half the band, the offset gets adjusted. That gives room for the drift that happens over the next hour.

Reporting is on request. A raw material certificate, a dimensional report with the actual measured values, and a first-article report are the three documents most engineers ask for. For medical work under ISO 13485:2016 or automotive work under IATF 16949:2016, the documentation expectations are higher and should be stated on the drawing or the purchase order, not agreed verbally after the fact.

One practical point about tight tolerances. ±0.005 mm is achievable on a rigid part with a clean datum and a stable material. It is not achievable on a part with a 0.8 mm wall and a 200 mm unsupported length, no matter how good the machine is. If the drawing calls for both, expect the shop to come back with a design change rather than a promise.

• 1
  First article in fullEvery dimension on the drawing, before the run continues.
• 2
  Critical dimensions every tenth partCatches drift while there is still time to correct it.
• 3
  Ask for the documents up frontMaterial cert, dimensional report and FAI on request.

CNC machining is a subtractive process, so the cost per part tracks the volume of material removed and the number of operations. For a simple bracket in 6061 aluminium at 500 pieces, machining is usually still competitive. For the same bracket at 5,000 pieces with a 40 mm thick blank, the material waste alone starts to argue for die casting or extrusion.

The transition is not a fixed number. It depends on geometry, wall thickness, surface finish and whether the part can be redesigned for a mold. A part with one drafted face and no undercuts will move to injection molding or die casting far earlier than a part with machined pockets on six sides.

Surface finish sets a second boundary. As-machined finishes land around Ra 1.6–3.2 μm. A high-quality machined finish is Ra 0.8–1.6 μm, and fine finishing reaches Ra 0.2–0.8 μm on the right material. If the drawing calls for Ra 0.4 μm across a large curved face, expect polishing or lapping time to be a real line item, not a rounding error.

There is also a hard ceiling on size. Our maximum processing size is 4,000 mm, and that envelope exists on specific machines with specific travels. A 3,000 mm part is not a small batch problem, it is a large-part logistics problem. Discuss the handling, the fixture weight and the inspection method before the quote, not after.

• 1
  Material removal drives costA thick blank at high volume argues for casting.
• 2
  Finish is a line itemRa 0.4 μm across a large face is slow work.
• 3
  Size changes the riskAbove 1,000 mm the fixture, not the cutter, sets the error.

The first rule is to design the fixture around the datum the drawing uses, not around the easiest face to clamp. If the drawing dimensions from a machined bore, that bore should locate the part in every operation. Clamping on a raw cast surface and dimensioning from a machined bore guarantees a stack-up error that grows with batch size.

The second rule is to keep the tool list short. Every tool change is a chance for a chip to sit on a holder taper and shift the tool by a few micrometres. A batch that runs with six tools instead of twelve has half the opportunities for that error, and it usually runs faster too because the non-cutting time drops.

The third rule is to separate roughing and finishing in time, not just in the program. Rough the whole batch, let the parts reach room temperature, then finish. This is slower on paper and faster in practice, because it removes the thermal variable from the finishing cut.

The fourth rule is to write the offset adjustment into the setup sheet. Tell the operator which dimension to check, how often, and what to do when it drifts. A batch with a written plan produces fewer surprises than a batch where the operator is guessing whether the trend matters.

• 1
  Locate on the drawing datumEvery operation should use the same locating feature.
• 2
  Fewer tools, fewer chancesSix tools beat twelve on repeatability and cycle time.
• 3
  Rough, rest, finishLet parts cool before the finishing pass.
• 4
  Write the offset plan downWhich dimension, how often, what action.