"Ra 1.6" appears on drawings constantly, but Ra is only one average of the peaks and valleys along a measured line. Two surfaces that share the same Ra can feel and perform nothing alike — one may seal cleanly under pressure, the other may weep at the gasket. Anyone who has had to sign off on a finish callout spends as much time explaining what Ra does not tell you as what it does. In the sections below we walk through how the number is actually calculated, what a CNC mill can and cannot control, where a surface grinder earns its place, and how a face gets measured so the print means what the customer thinks it means.
Ra is an average, not a ceiling
Ra averages the absolute deviation from the mean line over the measured length. That simple definition hides the trap: a single deep scratch, a tool chatter mark, or a porosity pit gets averaged into the surrounding peaks and simply vanishes from the number. Ra cannot see the worst spot. That is why a sealing face with a critical Ra callout also needs Rz or Rmax noted on the print — Rz captures the average of the five deepest valleys and catches the defects Ra hides. A hydraulic manifold can read Ra 0.8 µm clean while Rz comes back at 6.4 µm, because two intermittent burrs are enough to move Rz while barely touching the average. Ra alone would have passed a part that leaked on test.
If your print only says Ra, the first question we ask is whether the function actually sees the average or the worst valley. A bearing race cares about the peaks that touch the roller; a gasket face cares about the deepest groove that breaks the seal line. The same Ra means different things in different places, and naming the second parameter is how you stop that argument before it reaches the inspection room. We would rather add one letter to the drawing than reject a hundred parts over a number that was never the right one.
What the machine actually controls
For milled and turned surfaces, finish is driven by three things we can set at the control: stepover, the tool nose radius (or the corner radius of a flat end mill), and feed per tooth. A ball nose running a 0.2 mm stepover leaves a scallop height roughly equal to the stepover squared divided by eight times the radius. Drop that stepover to 0.05 mm and the scallops shrink by about 16× — visible on the profilometer and on the part in the right light. Feed per tooth sets the spacing of the tool marks; cutting faster than the material likes smears rather than shears, which reads as a false-good Ra but a poor real finish under the finger.
Climb milling helps because the tool loads into the cut instead of dragging, and a sharp insert with low runout keeps the marks even. Spindle thermal drift over a long run grows the scallops on the far side of the part, which is one reason we warm up the spindle and qualify the first piece after thermal settling. Our CNC milling cells hold stepover and feed to the program, but the finish still lives or dies on how the tool was set and whether the insert was changed on the schedule we wrote for it. Program discipline does not machine the surface; the edge does.
Matching the callout to the process
The practical rule is blunt: write the finish the function needs, then pick the process that reaches it. Below about Ra 0.4 µm, no milling pass is reliable — you are fighting runout, thermal drift, and the material itself, and the scatter on the tester tells the story. That is where secondary operations take over. A wheel dressed to a fine grit and a controlled downfeed takes a hardened steel face to Ra 0.2 µm repeatably, and lapping goes further still for valve seats and optical seats where the last tenth of a micron is the product.
| Target Ra | Process we use | Typical settings |
|---|---|---|
| 3.2 µm | CNC mill, single finish pass | 0.10–0.15 mm fz, climb, sharp insert |
| 1.6 µm | CNC mill, tight stepover | 0.04–0.06 mm stepover, Ø6–10 mm ball |
| 0.8 µm | CNC mill, fine second pass | 0.02–0.03 mm stepover, light depth |
| 0.4 µm | Grind or fine mill | Wheel 120–180 grit, 0.005–0.010 mm downfeed |
| 0.2 µm and below | Grind or lap | Diamond lap, cross-hatch, controlled pressure |
When the part also needs a coating, the sequence matters more than people expect. Bead blasting a face to Ra 1.2 µm and then anodizing it changes the final number, and a hard coat grows the surface by a few microns of oxide that the seal has to absorb. We plan the finish on the as-machined condition and let the surface finishing step land inside its own spec — otherwise the anodizer gets blamed for a mill that never hit the number, and the print quietly becomes impossible to meet.
Reading the surface: what shifts the number
Finish is not set once and forgotten. Across a run it moves with tool wear, coolant, and part temperature, and a number taken on Monday is not the number shipped on Friday. The variables we watch most closely on the floor:
- Insert sharpness — a worn edge smears aluminum and leaves a shiny, false-good trace that fails under a fingernail.
- Stepover consistency — a missed pass on a 3D surface leaves a visible band at the boundary nobody can sand out.
- Coolant concentration and nozzle aim — a dry flank work-hardens 6061 and lifts the Ra within a single pocket.
- Workholding rigidity — a part that chatters reads as random peaks no program change will ever fix.
- Cutoff length on the tester — too short measures noise, too long averages out the very feature you care about.
We set a cutoff per ISO 4288 — usually 0.8 mm on machined faces — and keep it fixed so two measurements on two days are actually comparable. A number taken at the wrong cutoff is not wrong, it is simply about a different surface than the one drawn, and that gap is where most finish disputes are born.
"If the print says Ra 0.2 and the process is milling, the drawing is wrong, not the machine. Call it out before cutting, because the fight later is about who pays for the scrap."
Hard materials fight back
Finish gets harder as hardness climbs, and not just because the tool wears faster. Above about 45 HRC the same stepover leaves a rougher scallop for the same insert, because the material springs back under the edge and the cut deflects. We see this clearly on 17-4 PH aged to 44 HRC and on D2 at 60 HRC, where milling to Ra 0.8 needs a fresh insert every few parts and a reduced stepover the program would never need on mild steel. The fix is rarely more speed; it is a sharper edge, a stiffer setup, and often a grinding operation the print should have called for.
This is also where heat-treat-then-grind beats machined-in-one-shot. Hardening first and then grinding the seat holds Ra 0.2 µm without the distortion a finish mill would leave on a warped blank. When a customer asks for a tight finish on a hard part, we quote the grind, because quoting the mill sets everyone up for a number the process cannot hold on a Tuesday.
Measuring it honestly
We use a portable roughness tester with a 0.8 mm stylus radius, taking three traces per face and reporting the mean. The stylus must sit on a representative area, not on a corner the program never touched, and it must run perpendicular to the tool marks so it crosses them — running along the marks under-reads by a wide margin because it never leaves the valley. On a contoured face we probe a flat reference island we leave for exactly this purpose, then remove it in deburr.
For parts that ship to audit, the measurement gets written into the FAI with the cutoff, the direction, and the stylus radius all listed. An Ra with no measurement condition attached is a number nobody can reproduce, and our customers' quality teams know that because they have been burned by it. Documenting the how is the difference between a finish you can defend in a corrective-action meeting and one you have to explain away. We would rather spend the extra line on the report than the extra week on a containment.
Writing a callout a machinist can build to
The advice we give designers is blunt: name the function, then name the number, then name the method. "Ra 0.8, machined" leaves too much to chance — it does not say which process, which direction, or which secondary step follows. "Ra 0.8 (Ø measured perpendicular to tool marks, 0.8 mm cutoff), mill finish, no burrs" tells the shop what to do and tells inspection what to check. Add Rz when the worst valley matters more than the average, and the print stops generating arguments at the CMM.
Most of the finish problems we see on incoming prints are not machining problems at all. They are specification problems — a number copied from a similar part, a tolerance tighter than the function needs, a callout that milling cannot physically reach. Catching those early, before the first block is cut, is the cheapest finish control we own, and it is free if you ask before you release the drawing.
For the tolerance and finish numbers side by side, see our machining tolerance chart.