CNC Surface Roughness: What Actually Sets the Finish

• 1
  Key takeaways
• 2
  How the cutter writes the surface
• 3
  Reading Ra, Rz, and the callout on your drawing
• 4
  What each process can hold
• 5
  Why the finish misses the target
• 6
  What a tighter callout really costs
• 7
  How to dial in a finish on a real job
• 8
  Ra ranges and where they fit

A CNC surface roughness value is not a material property. It is a record of the tool path. The cutting edge is not infinitely sharp and the machine is not infinitely stiff, so every pass leaves a groove. On a face mill, the corner radius of the insert sweeps an arc and leaves a scallop between passes. On a lathe, the nose radius of the turning insert does the same thing as the tool feeds along the axis.

The geometry is simple enough to predict. For a turning operation with a round nose, the theoretical peak-to-valley height is roughly the feed squared divided by eight times the nose radius. For a face mill, the same relationship holds with the cutter diameter in the denominator. Feed enters as a square, so it is the strongest lever you have. Tool geometry enters linearly.

Real parts do not reach the theoretical number. They land somewhere better or worse depending on what else is happening at the cutting edge. A sharp carbide insert in 6061 aluminium on a rigid machine can beat the formula because the material shears cleanly and the built-up edge is small. A dull tool in 304 stainless will miss it badly.

That gap between the formula and the part is where most finish problems live. When a shop reports Ra 2.5 μm on a job that should run at Ra 1.0 μm, the cause is usually one of four things: tool wear, chatter, chip recutting, or material that does not machine cleanly. Each has a different fix.

• 1
  Feed is squaredCut the feed per tooth in half and theoretical roughness drops about 75 percent.
• 2
  Nose radius is linearA larger tool radius spreads the same chip load over a longer arc, lowering peak height.
• 3
  Depth of cut is quietWithin reasonable limits, axial and radial engagement barely change Ra on a stable setup.
• 4
  Speed changes the chip, not the pathHigher surface speed helps most when it stops built-up edge or reduces cutting force.

Ra is the arithmetic mean of the profile height over the sampling length. It is the value most drawings use because it is easy to measure and easy to compare. The weakness is that it averages. A surface with a few deep grooves and an otherwise smooth field can post the same Ra as a uniformly scalloped surface, and the two will behave differently under a seal.

Rz is the average of the five highest peaks and five deepest valleys in the sampling length. It is more sensitive to the worst features. If a surface has to hold a gasket, an O-ring, or a hydraulic seal, Rz is the number that tells you whether the seal will weep. A common working rule is to keep Rz below four times Ra when the surface is critical.

The cutoff length matters as much as the value. A 0.8 mm cutoff filters out long-wavelength waviness and reports the short-wavelength roughness. An 8 mm cutoff pulls that waviness back in. Two labs can report different Ra on the same part if they use different cutoffs, so the drawing should state the cutoff or reference a standard such as ISO 4287 or ASME B46.1.

Direction also matters. A milled face has a strong lay in the direction of the passes. Measuring across the lay and along the lay gives different readings. Specify the measurement direction when the finish is functional, or the inspection result may not match the intent of the design.

• 1
  Ra for general calloutsGood for wear surfaces, cosmetic panels, and anything where the average texture drives performance.
• 2
  Rz for sealingUse it when a single deep valley can cause a leak.
• 3
  Rmax for extreme casesThe single worst valley in the sample. Rare, but used in some aerospace and medical specs.

Milling and turning start at the as-machined range. A normal face mill or turning pass lands between Ra 1.6 and 3.2 μm. That is fine for brackets, housings, and most structural parts. Tightening to Ra 0.8–1.6 μm is routine with a fresh insert, a smaller feed per tooth, and a rigid setup. It is the working range for bearing bores and gearbox faces.

Below Ra 0.8 μm the process changes character. You are no longer just dialing down the feed. You need a sharp, coated insert with a small nose radius, a balanced tool holder, and a machine that is not fighting thermal drift. Reaming and fine boring help on holes. On flat faces, a fly cutter with a large radius and a light finishing pass can reach Ra 0.4 μm in aluminium.

Fine finishing below Ra 0.2 μm is a different operation. Grinding, lapping, honing, or polishing takes over. These processes remove the machining marks and replace them with a finer texture. They also add a step, a fixture, and inspection time. If a drawing calls for Ra 0.1 μm on a milled face, the shop will almost certainly grind it, and you should budget for that.

Material choice shifts every number. Aluminium 6061 and 7075 machine cleanly and take a fine finish easily. Stainless 304 and 316 work-harden and tend to tear, so the same parameters give a coarser result. Titanium and Inconel are worse. Plastics smear, and a POM or PEEK surface that looks smooth can measure rougher than it feels. The callout has to account for the material, not just the geometry.

• 1
  Aluminium and brassEasiest to finish well. Ra 0.4 μm is reachable on a mill with a good setup.
• 2
  Carbon and alloy steelPredictable. Ra 0.8 μm is routine turning; harder grades need more attention to tool wear.
• 3
  Stainless and titaniumProne to built-up edge and work hardening. Expect one step coarser for the same parameters.
• 4
  PlasticsSmear and spring back. Sharp tools and high speed help, but Ra readings can be misleading.

Chatter is the most common cause of a finish that is worse than the formula predicts. It shows up as a regular pattern of marks with a spacing that matches a tooth-passing or spindle frequency. The fix is not always to slow down. Often it is to change the tool overhang, add a support, or move to a different spindle speed where the machine is stiffer. Reducing radial engagement also helps.

Tool wear is the second cause. A coated carbide insert that has run for an hour on 4140 steel will not produce the same surface as a fresh edge. The wear land rubs instead of cutting, which raises cutting force and leaves a torn surface. Monitoring the finish on a first article and then rechecking after a set number of parts is more reliable than trusting a tool life chart.

Built-up edge is the third. It happens when the workpiece material welds to the cutting edge and then breaks off, taking a chunk of the surface with it. It is common in aluminium at low speed and in stainless steel at almost any speed. Raising surface speed and using a sharper, more polished tool usually clears it. A larger rake angle helps in soft materials.

Chip recutting is the fourth. When chips are not evacuated, they get dragged across the finished surface and scratch it. This is a programming and coolant problem more than a cutting-parameter problem. Through-spindle coolant, air blast, or a different tool path that lets chips fall away will fix it. The scratch it leaves is often deep enough to raise Rz even when Ra looks acceptable.

• 1
  ChatterRegular marks, frequency-linked. Fix the setup stiffness before touching the feed.
• 2
  Tool wearGradual drift over a run. Track first-article finish and recheck on a schedule.
• 3
  Built-up edgeRandom tearing and smearing. More speed and a sharper edge usually clear it.
• 4
  Chip recuttingScratches that raise Rz. Fix evacuation first.

Surface roughness is one of the few drawing features where the cost curve is steep and non-linear. Going from Ra 3.2 μm to Ra 1.6 μm is almost free. You change the feed and run the same program. Going from Ra 1.6 μm to Ra 0.8 μm costs a finishing pass, a fresh insert, and a little cycle time. Going below Ra 0.4 μm usually means a second process, a second setup, and a second inspection.

That second process is where the money goes. Grinding needs a machine, a wheel, a fixture, and a skilled operator. Polishing is manual and hard to keep consistent across a batch. Honing is fast on bore diameters but limited to that geometry. Lapping is slow and usually reserved for flat sealing faces on small parts.

Inspection adds its own cost. A portable roughness tester is quick, but it only samples the spots the operator chooses. A full surface map on a complex part needs a profilometer or a white-light system and takes time. When a drawing specifies Ra 0.2 μm on ten surfaces, every one of them has to be measured, and the measurement time can exceed the machining time on small parts.

The practical answer is to specify the coarsest finish that meets the function. Put a general Ra 3.2 μm note on the title block and call out tighter values only on the surfaces that need them. This is the single biggest cost lever a designer has on a machined part, and it costs nothing to use.

• 1
  Title block noteSet the default at Ra 3.2 μm and override only where function demands it.
• 2
  Functional surfaces onlySeals, bearing fits, sliding contacts, and optical faces.
• 3
  Avoid blanket fine finishesOne Ra 0.4 μm callout across a whole part can double the price.