What CNC experts recommend before you release a part file

• 1
  CNC experts recommend reading the cut path, not just the drawing
• 2
  Setup count and datum choice
• 3
  Material behavior sets the finishing window
• 4
  Where tight tolerance pays and where it wastes money
• 5
  Inspection closes the loop on the process
• 6
  How to tell whether a shop actually follows these rules
• 7
  Which machine class fits which part

A drawing shows the finished shape. A machinist sees the path that creates it. Engineers with shop-floor time recommend reading a part as a sequence of cuts, because every feature needs a tool that can reach it and a setup that holds it rigidly. A wall of 0.5 mm or a pocket 12 mm deep with a 6 mm corner radius will not run the same way twice, no matter how clean the model looks.

The physics is boring and it never changes. Cutting force pushes the tool and the workpiece apart. Long tools bend. Thin walls deflect away from the cutter and spring back after the pass, so the measured dimension drifts from the programmed one. Chatter starts when the tooth frequency matches a natural frequency of the setup, and the surface turns into a pattern of waves that no finishing pass removes.

This is why two shops can quote the same STEP file and land 40% apart. One plans three setups on a three-axis mill; the other plans one setup on a five-axis center and skips two re-fixturings. Same part, different number of chances to lose datum alignment. The file did not change. The process did.

The practical takeaway: model the part, then model the process on paper. Ask which face locates the part, which tool reaches each corner, and where the tool exits. Most of the redesign work happens in that ten-minute review, not in the machine.

Every re-fixturing adds error. The part comes off, chips are cleared, the vise is re-clamped, and the new zero is set from a face that was itself machined in the previous setup. Errors stack. A three-setup job on a three-axis machine can hold ±0.005 mm on a single critical feature only if the datums are chosen with that stack in mind.

Five-axis work changes the arithmetic. One rotary table holds the part while the spindle reaches five faces, so the relationship between features is set once. On our Ø400 mm rotary table, a 300 mm aluminum bracket can be finished in one setup where a three-axis route needs four. The gain is not speed. The gain is that the bolt-hole pattern and the bearing bore stay concentric by construction rather than by careful re-indication.

Datum choice is the part engineers most often get wrong. A datum should be a machined face, large enough to locate, and stable enough not to move when the clamps tighten. A raw casting surface is none of those things. If the drawing calls for a cast face as datum A, expect to add a machining pass to create a real one.

Clamping force is the quiet variable. Aluminum 6061 tube walls at 2 mm will ovalize under a standard vise. A soft jaw bored to the part diameter, or a low-pressure hydraulic clamp, keeps the round within tolerance. For titanium and 17-4PH, springback after unclamping is small but real, and a finishing pass with light depth of cut is cheaper than a rework loop.

Material decides the surface finish you can hold, not the other way round. Aluminum 6061 and 7075 cut freely and reach Ra 0.8–1.6 μm with a sharp carbide tool and a clean finishing pass. Stainless 316L work-hardens at the cut, so a dull tool or a dwell in the pass raises the local hardness and the next pass cuts worse. Feed through the work-hardened layer, never rub it.

Titanium TC4 (Ti-6Al-4V) burns tools at the edge because heat does not leave with the chip. Rigid setups, sharp uncoated carbide, low surface speed and generous coolant are the standard answer. Inconel is worse. Both cut well on a five-axis center where the tool stays engaged at a constant angle, and poorly on a light three-axis mill where the tool exits and re-enters on every pass.

Plastics follow different rules. POM and PEEK hold tight tolerances but move with temperature, so measure after the part cools, not at the machine. ABS and PC soften under frictional heat and need sharp single-flute cutters and air blast. Carbon fiber delaminates if you push a two-flute cutter through a laminate; diamond-coated tooling and a backing plate are the difference between a clean edge and a frayed one.

Hardness itself is a routing decision. Above roughly 45 HRC, cutting gets slow and grinding or EDM becomes the cheaper answer. Below that, carbide milling is still the fastest way to a finished face. Tell the shop the heat-treat state of the incoming stock, because 4140 pre-hard and 4140 annealed are not the same job.

Tolerance is a cost curve, not a quality badge. General dimensions on a milled part hold ±0.1 mm without extra effort. Below ±0.05 mm the shop starts adding finishing passes, temperature control and more inspection time. At ±0.005 mm, the achievable limit on our machines, you are buying a process, not a feature. Apply it only where function demands it: a bearing bore, a sealing face, a mating dowel.

GD&T communicates intent better than a blanket tolerance block. Position tolerance on a bolt pattern tells the machinist the pattern is a group, so the whole pattern shifts together and the mating part still fits. The same pattern dimensioned as independent ±0.05 mm coordinates can be in tolerance feature by feature and still fail assembly. The drawing that reflects how the part works costs the same to make and fails less often.

Surface finish follows the same logic. Ra 1.6–3.2 μm is a normal as-machined finish and suits most structural faces. Ra 0.8–1.6 μm needs a controlled finishing pass. Ra 0.2–0.8 μm needs a specific tool, a light depth of cut and often a separate operation. Call it out only on sliding, sealing or optical faces. A cosmetic Ra 0.4 μm on a hidden internal wall adds cost and changes nothing.

One more habit worth copying: put the critical dimension on the drawing with the datum that controls it. If a bore must be coaxial to a shaft journal, dimension it that way. Machinists and inspectors both read the drawing literally. What is not stated is not controlled.

A first article inspection proves the setup, not the run. Once the process is proven, in-process checks catch drift from tool wear and thermal growth. On a long run, hole diameters creep as the drill wears; a check every 50 parts catches it before the batch is scrap. This is standard practice here, with 100% inspection before shipment and reports available on request.

CMM reports are the usual evidence for tolerance-critical features, but they are not the only tool. Bore gauges, micrometers and pin gauges are faster for simple features and just as reliable when the operator is trained. What matters is that the measurement method matches the tolerance. A ±0.005 mm bore cannot be verified with a caliper; it needs a bore gauge or a CMM with a known probe calibration.

For regulated industries the paperwork is part of the process. Our quality system is certified to ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022, so automotive, medical and defense programs can be supported with the documentation they expect. Material certificates, inspection reports and traceability records travel with the parts.

The last inspection question is often the first one a buyer should ask: who signs off, and against what drawing revision. A part measured against rev B when the customer released rev C is scrap with a nice report attached. Revision control on the shop floor is unglamorous and it prevents the most expensive kind of rework.

Ask for the process plan, not the price. A shop that can name the machine, the setup count and the datum for your part has already done the engineering. A shop that answers only with a number is guessing. The answer to "how many setups?" tells you more about the quote than the total does.

Ask what changed after the first article. Good shops send a DFM note before cutting, listing the features that will be hard to hold and suggesting a change. If the first you hear about a 0.3 mm wall or a 4:1 depth-to-diameter hole is a failure report, the review never happened.

Ask how they measure the tight feature. The method should match the tolerance. If the answer is a caliper for a ±0.005 mm bore, the number in the report means less than it looks.

These are cheap questions. They take one email and they separate a shop that machines to a drawing from one that machines to a process. The second kind delivers the same part on the second order, and that is the one worth keeping.