Carbon Fiber CNC Machining

Abrasion is the only active wear mechanism when a tool cuts fibre-reinforced plastic: the flank polishes rather than cratering, and average roughness rises roughly linearly with cutting length—the tool is consumed on a schedule, not by sudden failure. That happens because the reinforcement fibre is harder than the hard phase in the carbide itself, so wear proceeds by microcutting. The practical mitigation is to run a carbide grain size larger than the reinforcement, and it is also why HSS and conventional twist-drill geometries are effectively obsolete on CFRP—short tool life and a tendency to delaminate the material on exit have practically eliminated them from this work.

Turning trials across carbide, ceramics, SIALON, CBN and PCD on carbon/glass-fibre stock found that only diamond tools are suitable for finish turning, while cemented carbide remains the best cost/benefit choice for rough turning despite its higher wear rate—which is why we route roughing and finishing to different tool materials rather than run one grade through the whole job.

Peel-up happens at hole entry, where tool geometry lifts the top plies. Push-out happens at exit, and it is the more common failure: the thrust force needed to cause delamination collapses as the uncut depth of material below the drill point shrinks toward the last ply, which is exactly why exit damage concentrates there rather than mid-laminate. Counterintuitively, controlled drilling trials found that drill geometry dominates delamination risk far more than cutting speed does—geometry and feed were statistically significant for maximum thrust force, cutting speed was not. Geometry beats parameters, which is why our process controls target the drill and the backing, not just the feeds and speeds.

On thin laminates—2.5 mm and under are especially vulnerable—the outside plies are always backed externally, and feed is reduced specifically on the exit layer rather than held constant through the stack: a practical split runs roughly three times slower on the last ply than the first. Standard twist-drill geometry (118° point, 30° helix) is not the composite answer; a point angle in the 130–140° range with helix above 24° both reduces delamination risk and minimises surface roughness. Where the design allows it, trimming with compression (opposed-helix) router geometry directs cutting forces inward and gives quality edges on both faces at once, which removes the peel-up/push-out choice entirely for that operation.

ISO 2768-1—the standard behind the general tolerance table on our CNC machining page—states plainly that it applies to parts produced by metal removal or formed from sheet metal, with a note that its classes may be suitable for materials other than metal. That is the honest, standards-grounded way to say the general tolerance class on a composite drawing is an assumption carried over from metal, not a laminate-specific figure. The same standard is explicit that exceeding a general tolerance does not automatically make a part rejectable if function is not impaired—which is exactly why the features that carry function on a laminate need their own individual tolerances rather than a blanket class.

The divergence is measurable. On a hole drilled through a titanium/CFRP stack with the same tool, published aerospace acceptance criteria run about twice as coarse on the composite side as on the metal side of the same hole—and measured results follow the same pattern:

The mechanism is fibre pull-out: at the instant of chip separation the brittle fibre fractures cleanly while the softer resin continues to deform, leaving micro-cavities that a metal cut does not produce. On the same part machined with the same tool, one published study measured the deepest surface valleys on the composite side at nearly four times the depth of the metal side—that is the pull-out craters, not measurement noise.

There is also a direction-dependent thermal-expansion problem that a metal drawing never has to solve. Carbon fibre itself has a coefficient of thermal expansion close to zero along its axis—occasionally slightly negative—while the resin around it expands closer to the rate of aluminum or faster. A laminate does not have a coefficient of thermal expansion; it has one that depends on direction and lay-up:

Moisture adds a second variable a metal drawing does not carry: composite dimensional change is a function of both temperature and moisture condition, and moisture diffuses through a laminate many orders of magnitude slower than heat does—so a part measured immediately after machining and the same part measured after conditioning can disagree for reasons that have nothing to do with the machining itself. The tolerance conversation on a laminate therefore needs three things on the drawing that a metal drawing does not: the general tolerance class, with an acknowledgement that it was written for metal; the measurement temperature; and the moisture condition the part will be measured in. Without those three, a tight tolerance on a composite drawing is not a requirement anyone can actually verify. See plastic machining for the same thermal-expansion argument applied to unreinforced engineering plastics.

Roughness peaks specifically at 45° and 225° to the cutting direction, and in trimmed stacks the 90° and 135° orientations show markedly poorer surface quality and more subsurface fibre/matrix damage than 0° and 45°. None of that is a defect in the part—it is physics that a fixed feed and speed cannot fully compensate for, which is why the routing plan, not just the cutting parameters, changes around the edge of a real part.

Composites are usually moulded, but achieving close fits and tolerances, or reaching true near-net shape, still requires some machining—and because of their non-homogeneity, composites pose real problems that metal-machining data cannot answer. Machining a cured laminate also cuts fibres, which weakens the local structure and creates stress concentrations; a poor cut edge becomes an in-service crack initiator rather than a cosmetic issue, which is exactly why cut quality and edge planning matter as much on a structural CFRP part as the dimension itself.

Moulding is not a free path around tolerance, though—it has its own, wider bands. Normal thickness variation runs from about ±0.13 mm on matched-die processes up to ±0.51 mm on hand lay-up, and a published class table for reinforced-plastic parts puts even the "fine" class at roughly ±0.5 mm for a 100–200 mm dimension—which sits at ISO 2768 class c, not the tighter classes a metal part might carry. That is the honest starting point for a moulded composite, and it is why features that need to hold function on either a moulded or a machined laminate get individually toleranced rather than left to a general class.

The producibility literature makes the opposite point just as plainly: acquiring a part in final net shape reduces machining and fabrication time and cost, but that does not mean net-shape moulding is automatically cheaper—a design guidance handbook puts it directly: do not necessarily assume the near-net-shape process is the most economical; always compare the cost of both methods before deciding. Mechanical joints stay common for exactly that reason—drilling remains the most-used technique for composite assembly despite laser and waterjet alternatives being available—so the real question on a new part is not "moulded or machined" in the abstract, but which one gets your specific geometry and tolerance to the drawing for less. Send both the geometry and the volume (see rapid prototyping for early-stage builds) and we will tell you which side of that line your part sits on.

Dust extraction is required for both reasons at once: carbon dust is electrically conductive, so beyond the respiratory hazard it can affect electrical components and increase spindle wear if it is allowed to migrate through the machine. That conductivity is a real physical property, not a marketing point—graphite and graphite-containing composites are treated as a metal for the purposes of dissimilar-metal contact, which is also the source of the galvanic warnings elsewhere on this page.

The particles themselves are fine enough to matter: individual carbon fibres run roughly 5–10 µm in diameter, and composite machining generates predominantly sub-10 µm dust, with some fragments as small as 0.05–0.1 µm—fine enough to penetrate deep into the lungs. No CFRP- or GFRP-specific occupational exposure limit exists anywhere in the world—general dust thresholds are applied instead, and a stricter airborne-fibre limit has been proposed but is not yet universal. Controls run to that general standard: HEPA filtration, local exhaust ventilation, fine-particle-capture vacuums, and encapsulation of the machining zone, with PPE appropriate to composite dust rather than to a standard that does not exist.

Because composites cut by brittle fracture rather than shear, coolant is often unnecessary—abrasive wear from the fibres is the dominant tool-life mechanism, and cutting fluid has limited effect on either tool life or achievable finish, though it can help with swarf disposal. That is why CFRP work typically runs dry. The honest exception is stacked work: on CFRP/metal stacks, cutting fluid helps prevent chip clogging in the flutes and can improve surface quality on the composite side, and heat generation on some cuts is damaging enough that cooling is genuinely required. Which regime applies is a routing decision made per part, not a blanket rule—and it is never made on the theory that CFRP avoids coolant because the laminate absorbs water; that is not the reason, and we do not claim it.