CNC Machining on Demand: How It Works and When to Use It

• 1
  Key takeaways
• 2
  What CNC machining on demand actually changes
• 3
  From CAD file to shipped part: the real sequence
• 4
  Where tolerance is won and lost on the shop floor
• 5
  When five-axis earns its setup, and when three-axis is enough
• 6
  Materials and finishes that behave differently on demand
• 7
  On-demand machining against the alternatives

On-demand machining is not a new cutting process. A rotating tool still removes material from a solid block under programmed coordinates. What changes is who owns the queue. Instead of buying a machine, hiring an operator and waiting for a slot in your own shop, you send a CAD file and a tolerance callout to a shop that already has the spindles running.

That shift matters most in the early weeks of a program. A design revision on Tuesday can be cut on Wednesday and inspected on Thursday. The same revision inside a captive shop usually waits behind whatever is already fixtured on the machine, which is often the real bottleneck, not spindle speed.

The trade is control for speed. You give up direct eyes on the setup and accept the shop's process choices. In return you skip capital, floor space and the hiring cycle. For most teams the decision is not philosophical. It is a scheduling question: is an internal machine free this week, and is it free for the whole job?

There is a second effect that is easy to miss. Because the shop runs many customers, it sees failure modes you have not hit yet. Thin walls that spring, deep holes that drift, threads that gall in 316. That pattern recognition is part of what you are buying, and it shows up in the DFM notes attached to the quote.

The sequence starts with a model and a drawing, not with a machine. A STEP or native CAD file gives geometry. The drawing carries what geometry cannot: datum scheme, critical dimensions, surface callouts, material condition and any fit requirement. Quotes go wrong when one of those is missing.

Next comes DFM review. An engineer checks wall thickness, tool reach, corner radii, thread depth and whether the datum you picked is machinable in one or two setups. A datum that only exists on a casting face is a common problem. So is a 0.5 mm internal corner that needs a 0.4 mm cutter with almost no rigidity.

Then programming and fixturing. This is where the schedule is actually decided. Soft jaws, a custom plate or a vacuum fixture all take time to make. On a one-off part, fixture time can exceed cutting time. Shops that quote fast have usually seen the geometry before and can reuse a known workholding approach.

Cutting follows, then inspection, then finishing if specified. Anodizing, plating or powder coating adds days and can move dimensions. Hardcoat anodizing builds roughly half the coating thickness into the surface, so a press-fit bore machined to nominal before coating may not fit afterward. Flag any coating on the drawing so the shop can compensate.

• 1
  Send both model and drawingGeometry alone does not carry datum or fit intent.
• 2
  Nominate one primary datumA datum the shop cannot reach in the first setup costs you accuracy.
• 3
  Call out coatings earlyPlating and anodizing change dimensions and add lead time.

A tolerance figure on a drawing is a promise about the whole system, not about the machine alone. Machine geometric accuracy is one link. Fixture stiffness is another. Thermal growth during a long cut is a third. Probing and compensation close the loop if the shop uses them.

Aluminium moves about 23 μm per metre per degree Celsius. A 300 mm part that warms 5 °C during roughing grows roughly 35 μm before the finishing pass even starts. On a ±0.005 mm callout that is most of the budget. Shops that hold tight limits either control coolant temperature or let the part stabilize before finishing.

Fixtures matter more than most drawings admit. A part held only by its outer edge will deflect under cutting force, and the deflection appears as a taper or a bowed face. Supporting under the cut and clamping near the machining zone reduces this. Thin-wall parts often need light finishing passes at reduced radial engagement rather than a heavy final cut.

Inspection closes the chain. Calibrated micrometers and CMM checks catch drift, but they cannot fix a setup that was wrong from the start. Ask for dimensional reports when a feature is critical, and say which dimensions actually matter. A drawing with 200 toleranced dimensions and no priority forces the shop to treat every one as critical.

Axis count is a fixturing decision, not a quality badge. A prismatic bracket with features on three orthogonal faces machines fine on a three-axis mill with two or three setups. Adding rotary axes does not automatically improve the part; it changes how many times you touch it.

Five-axis simultaneous motion pays off when the feature is genuinely unreachable otherwise. Impeller blades, contoured ports, angled bosses on a curved surface and undercut geometry all qualify. So does a part that would need five separate setups on a three-axis machine, because each setup adds stack-up error.

There is a cost side too. Five-axis programming takes longer and simulation is mandatory. On a simple part that overhead is wasted. On a complex part it usually beats the alternative of building three fixtures and hoping the stack-up stays inside tolerance.

A practical test: count the setups a three-axis route would need. One or two, stay three-axis. Three or more with tight positional relationships between them, evaluate five-axis. If the part is a rotational form with milled flats or cross-holes, a mill-turn center often beats both.

Material choice drives both cutting strategy and risk. Aluminium 6061 and 7075 cut cleanly and hold tight tolerances with modest effort. Stainless 316 work-hardens under a dull tool, so feed rates must stay aggressive enough to cut under the hardened layer rather than rub on it. Titanium Ti-6Al-4V conducts heat poorly, so the tool takes the temperature and tool life drops.

Plastics behave in the opposite direction from metals. POM and PEEK machine well but move with temperature and absorb moisture, so a dimension measured right off the machine may not be the dimension tomorrow. PMMA chips and crazes if the cutter dwells. ABS is soft enough to smear unless the tool is sharp and the feed is steady.

Finishes add their own constraints. Bead blasting and tumbling round edges slightly, which matters on a sealing face. Laser marking needs at least 1.5 mm character height to stay legible after anodizing. Electroless nickel adds a thin, uniform layer and holds tight tolerances well. Powder coating is thicker and much less dimensionally predictable.

The engineering point is simple. Pick the material and finish together, at the drawing stage, not after the quote. A finish chosen late can force a re-cut, and a recut on a finished part is often a new part.