CNC Machining Beginners Guide: How a Model Becomes a Part

• 1
  Key takeaways
• 2
  What actually happens inside the machine
• 3
  Three, four and five axis: what each one buys
• 4
  How the workpiece fights back
• 5
  Where the tool cannot go
• 6
  Building a tolerance budget that survives inspection
• 7
  Why the last 0.1 mm is a separate decision
• 8
  When CNC is the wrong answer
• 9
  Matching the part to the process

A CNC machine does one thing: it moves a cutting tool to a coordinate and holds it there while spinning. The controller reads G-code, which is a list of positions, feed rates and spindle speeds. Nothing in that list knows what the part looks like. The shape is an emergent result of thousands of small moves along a path called a toolpath.

The CAM programmer takes the solid model and decides which tool cuts which region. A 12 mm flat end mill clears the bulk. A 3 mm cutter reaches into corners. A ball nose cutter follows a curved surface in small stepovers. Each tool change costs time, so the order matters as much as the geometry.

Accuracy is not a property of the code. It comes from the machine frame, the spindle bearings, the workholding and the cutter itself. A machine that deflects 0.02 mm under load will produce a part that measures 0.02 mm off, no matter how clean the program is.

This is why a shop asks about function before it quotes a tolerance. A bracket that locates a sensor needs a tight bore. The same bracket's outer profile can sit at ±0.2 mm and work fine.

A 3-axis mill moves X, Y and Z. The tool always points straight down. Undercuts, side holes and compound angles need a second setup on another face. Each setup re-datums the part, and every re-datum adds stack-up error.

A 4-axis machine adds a rotary table, usually turning about X. Now you can drill around a cylinder or machine four faces in one setup. The Ø400 mm rotary table covers most shaft and housing work, and the setup count drops from three to one.

A 5-axis machine tilts the tool as well. That lets a short, stiff cutter reach a deep pocket wall at an angle, and it machines a curved surface in one continuous pass instead of many parallel ones. Surface finish improves because the tool stays in contact.

The trade is programming time and machine cost. If your part is a flat plate with holes, 5-axis buys nothing. If it is a turbine housing or a medical implant with blended surfaces, it is often the only practical route.

Aluminium 6061-T6 cuts fast and holds tolerance well, which is why prototypes are usually made from it. It also moves when you remove material, because residual stress in the plate releases as the section thins. Rough the part, let it rest, then finish it.

Stainless 304 work-hardens. If the cutter rubs instead of cutting, the surface gets harder and the next pass is worse. Sharp tools, a positive rake and a feed that stays above the work-hardening threshold solve this. 316L behaves the same and is common in medical and food-contact parts.

Titanium Ti-6Al-4V conducts heat poorly, so the cutting edge absorbs it. Speeds drop, coolant flow rises, and tool life becomes the cost driver rather than cycle time. Inconel is worse. Both are machinable, but the quote will reflect the tooling.

Plastics are the opposite problem. POM and ABS cut easily but deflect under clamping force and melt if the feed is too slow. Sharp single-flute cutters, light depths of cut and air blast instead of flood coolant keep the edge clean.

Every cutter is a cylinder with a length and a diameter. The length-to-diameter ratio sets how much it bends. A 6 mm cutter sticking 60 mm out of the holder will chatter and leave a rough wall, even at conservative feeds. Keep reach under four times the diameter when you can.

Internal corners are always rounded by the tool radius. A pocket designed with sharp internal corners cannot be milled, because the cutter spins. Either specify a corner radius at least equal to the cutter radius, or accept that the shop will leave the sharpest corner the tool allows.

A blind hole needs a drill point at the bottom. A flat-bottom hole needs an end mill, and the end mill leaves a small radius at the floor. Put that radius on the drawing so the inspector checks the right thing.

Threads reach a limit too. A tapped hole close to a wall needs a tap wrench or a tap holder that clears the wall. Thread milling avoids that, but it costs more cycle time.

A tolerance is a permission, not a target. If a bore is called out at Ø20 +0.02 / 0, the shop will aim for the middle of that band and the inspector will check the whole band. If every dimension on the drawing carries the same tight band, the part becomes expensive and the yield drops.

Stack-up is the part beginners miss. If three features locate a sub-assembly and each one carries ±0.05 mm, the assembly can be 0.15 mm off before anyone measures the mating part. Work out the chain first, then assign the bands.

Surface finish and tolerance interact. A Ra 0.8–1.6 μm finish on a bearing seat usually comes from a fine finishing pass, which is a separate operation. Asking for Ra 0.2–0.8 μm on a non-functional face adds cost for no benefit.

GreatLight holds ±0.005 mm on critical features and inspects 100% before shipment, with reports on request. That capability is for the features that need it. Loose features stay loose, and the quote stays sane.

Machining leaves tool marks. Bead blasting hides them and gives a uniform matte look. Tumbling rounds edges and removes burrs on small parts. Brushing leaves a directional grain that hides scratches on flat panels.

Anodizing adds a hard oxide layer and changes the dimension slightly. Hardcoat anodizing grows more than clear anodizing, so a tight bore should be masked or machined undersize before the coating. Tell the shop which surfaces are functional.

Laser marking is a common way to add a part number or a traceability code. The minimum character height is 1.5 mm. Below that, the mark becomes hard to read and the laser settings get fiddly.

Finish choices are visible to the customer but rarely affect function. Decide them after the tolerance and material are locked, not before.

CNC removes material from a solid block. If the part is a thin-walled enclosure the size of a shoebox, sheet metal fabrication is faster and cheaper. If it is a hollow shape with internal channels, die casting or vacuum casting wins at volume.

For a one-off bracket, CNC is usually the fastest route to a real part. For 10,000 identical simple parts, casting or stamping beats it on unit cost. The crossover point depends on geometry, not on a fixed number.

CNC also struggles with very soft, gummy materials and with parts that need a mirror finish over a large curved area. Those go to polishing or to a different process entirely.

The useful question is not which process is best. It is which process delivers the features that matter, at the volume you actually need, without a tooling investment you cannot amortize.