CNC machining basics: understanding the process

• 1
  How a CNC machine turns a CAD file into a part
• 2
  What tolerance and surface finish you can actually hold
• 3
  Matching material and cut strategy
• 4
  Where CNC machining stops being the right answer
• 5
  Inspection, repeatability and what the certificates mean
• 6
  3-axis, 4-axis or 5-axis: picking the right setup

CNC machining is subtractive. A rotating cutter removes material from a solid block until what is left is the part you drew. The control side means the motion of that cutter is driven by numbers, not by a handwheel. Once the numbers are correct, the same program runs the same way on the next part, and the part after that.

The chain has four links. A designer saves a 3D model, usually as STEP or IGES. A CAM programmer chooses tools, cut depths and toolpaths, then posts the result as G-code. The machine controller reads that G-code and drives ball screws and servo motors. The cutter follows. Everything downstream of the model is repeatable, and everything depends on the model being clean.

That last point matters more than most people expect. Open surfaces, duplicate edges, zero-thickness walls and unmarked units will all survive a CAD review and fail at the machine. We run a DFM check before quoting, so a wall that is 0.4 mm thick on a 30 mm deep pocket gets flagged while it is still a line on a screen, not a scrapped block of 7075.

The cutting itself is simple mechanics. The tool edge shears material away in chips. Heat leaves mostly with those chips. If the chip cannot leave, heat stays in the part and the tool, and you get a poor finish, a worn cutter, or a dimension that drifts after cooling. That is why chip evacuation, coolant and toolpath direction drive quality as much as the machine does.

• 1
  Model typeSTEP or IGES solids convert cleanly; mesh files need repair first.
• 2
  CAM outputG-code plus a setup sheet listing tools and work offsets.
• 3
  DFM checkWall thickness, deep pockets and tool reach reviewed before quoting.
• 4
  First articleCut, measured against the drawing, then released for the run.

Tolerance is where drawings and reality negotiate. On a rigid setup with the right tool, we hold ±0.005 mm (±0.0002 in) on critical features. On a long slender shaft, or a thin floor, the same machine cannot promise that, because the part deflects under cutting force. The number on the drawing is not the number the machine delivers by itself. The setup decides.

A practical rule: keep tight tolerances on the features that touch something else, and let everything else breathe. If a bolt hole pattern needs ±0.02 mm, say so. If a clearance hole does not, leaving it at ±0.1 mm saves cycle time and money. Drawings that call out ±0.005 mm everywhere force slow feeds, extra passes and more inspection, and the part is rarely better for it.

Surface finish follows the same logic. As-machined surfaces land around Ra 1.6–3.2 μm. A finer finish, Ra 0.8–1.6 μm, is common for sealing faces and bearing seats. For optical or sliding contact surfaces, Ra 0.2–0.8 μm is achievable, but it costs time, and it often needs a dedicated finishing pass rather than a slower roughing cut.

Finish and tolerance interact. A polished surface measured with a caliper tells you very little, because the caliper touches peaks. A tight bore needs a bore gauge or a CMM, not a set of digital calipers. If your inspection method cannot resolve the tolerance you specified, the specification is decorative.

Aluminium is the default for prototypes. Grades like 6061 and 7075 cut fast, hold good finishes, and take anodizing well. 7075 is stronger but less weldable and more prone to stress movement when you remove a lot of stock. 6061-T6 is the safer choice when the part is large and mostly machined from plate.

Stainless grades behave differently. 303 machines cleanly with good chip control. 304 and 316 work-harden if the tool rubs instead of cutting, so the feed has to stay aggressive enough to stay under the hardened layer. 17-4PH machines well in the annealed condition and gains strength after heat treatment, which is why it shows up in pump and valve parts.

Titanium, Inconel and magnesium each demand their own approach. Ti-6Al-4V has low thermal conductivity, so heat stays at the cutting edge; sharp tools, high pressure coolant and moderate speeds keep it under control. Inconel is worse on tool life. Magnesium AZ31B and AZ91D cut beautifully but need chip handling discipline, since fine magnesium chips are a fire risk.

Plastics are not automatically easier. POM and PEEK machine well but move with temperature. ABS and PC can gum up on a dull tool. Carbon fibre is abrasive and the dust needs extraction. The right question is not 'is this material machinable' but 'what cut strategy does this material need', and that answer changes the quote.

CNC is strong from one part to a few thousand. It is weak when the part is a thin shell, when the geometry is a lattice, or when the same shape is needed in the tens of thousands. At that volume, die casting or injection moulding wins on unit cost once the tooling is paid for. For thin walls under about 0.5 mm in metal, the cutting force will move the part before the tool finishes the pass.

Deep pockets are another boundary. A pocket 8× deeper than the cutter diameter needs a long, thin tool, and long tools chatter. You can relieve the chatter with light passes and a slower feed, but the cycle time climbs. Sometimes the better answer is to split the part into two pieces that bolt together, or to change the corner radii so a larger cutter can reach in.

Sharp internal corners are impossible by definition. A cutter has a radius, so a square internal corner becomes a fillet of that radius. If the drawing calls for a true sharp corner, the designer either needs a smaller tool, which is slower, or an EDM step, which is a different process and a different quote.

Finally, consider what happens after machining. Hardcoat anodizing adds a few micrometres and can round a sharp edge. Black oxide changes almost nothing dimensionally. Plating can build up on threads. If a thread must gauge after finishing, say so, because the machinist will cut it to a different size on purpose.

A first article proves the setup, not the run. After that, in-process checks catch drift before it becomes scrap. We check raw material on receipt, monitor during cutting, and inspect 100% before shipment, with reports available on request. The point is not paperwork. It is knowing whether a dimension moved at hour six of a twelve hour cycle.

Repeatability depends on rigidity and thermal stability. A machine that has been running all day is warmer than one that just started. Warm spindles and ball screws grow, and a tight tolerance cut at 08:00 may drift by mid-afternoon. Shops that hold ±0.005 mm consistently manage temperature, not just feeds and speeds.

Certifications describe the system around the part. ISO 9001:2015 covers quality management. IATF 16949:2016 adds automotive requirements. ISO 13485:2016 covers medical devices. ISO 27001:2022 covers information security, which matters when your drawings are confidential. A certificate does not make a part good, but it tells you the process is documented and auditable.

For a prototype run, the useful question is simpler: can this shop hold the tolerance on this geometry, in this material, and show me the numbers? Ask for the inspection method, not the certificate number. That is where the real answer lives.