Do You Really Understand These Things About Machining Accuracy?

• 1
  Key takeaways
• 2
  Accuracy and precision are not the same number
• 3
  How tolerance stack-up eats your machining accuracy
• 4
  Thermal drift and cutting force: the quiet accuracy killers
• 5
  Where surface finish and form error fit in
• 6
  How to read a print before you send it for quote
• 7
  Typical machining accuracy by process and material

Accuracy describes how close a finished feature sits to the nominal dimension on the drawing. If a bore is called out at Ø25.000 mm and the CMM reads 25.008 mm, the error is 0.008 mm. That single number is accuracy. It says nothing about the next part in the batch.

Precision describes repeatability. Run twenty parts, measure the same bore on all of them, and look at the spread. If every bore lands between 25.006 and 25.010 mm, precision is tight even though the group is offset from nominal. A machine with good precision and poor accuracy is a calibration problem. A machine with good accuracy and poor precision is a mechanical or thermal problem, and it is much harder to fix.

The distinction matters because the two failures need different responses. Accuracy errors get corrected by touching off tools, updating work offsets, or re-probing a fixture. Precision errors come from spindle runout, backlash, loose clamping, chip recutting, or heat moving the frame. You cannot dial out a spread with an offset change.

On the shop floor we track both. A capability study (Cpk) on a critical feature tells you which one you are fighting. If the mean is off but the range is small, adjust. If the range is wide, stop and find the mechanical cause first.

• 1
  Accuracy errorMean shifted from nominal; spread is small and stable.
• 2
  Precision errorMean may be fine, but the spread is wide and unpredictable.
• 3
  Both offFix the mechanical cause before touching offsets.

Every dimension on a print is a chain. The final feature position is the sum of the machine positioning error, the fixture locating error, the tool deflection, the material springback, and the measurement uncertainty. When a designer puts ±0.005 mm on a hole-to-hole distance, they are asking the whole chain to fit inside that band.

A typical chain on a 4,000 mm gantry part looks like this: machine positioning ±0.010 mm, fixture repeatability ±0.008 mm, thermal growth over a 6-hour run ±0.015 mm, and probing uncertainty ±0.003 mm. Root-sum-square gives roughly ±0.020 mm. That is four times tighter than a ±0.005 mm callout can absorb. The print is not impossible, but it needs a different process plan.

This is why we ask for the datum scheme before quoting. A part dimensioned from a single primary datum with a clear feature-of-size reference is much easier to hold than one dimensioned from three different faces that each get machined in a separate setup. Fewer setups means fewer stack-up contributors.

If a feature really needs ±0.005 mm, the usual answer is to machine it in the same setup as its datum, probe it in-process, and inspect on a CMM with a stated uncertainty budget. Everything else is hope.

• 1
  Count the setupsEach new setup adds locating error to the chain.
• 2
  Watch the datumA single primary datum cuts stack-up sharply.
• 3
  Budget the heatLong runs drift; short runs do not.

A spindle running at 12,000 rpm for three hours will grow. The frame, the ballscrews, and the workpiece all move at different rates. On aluminium, a 5 °C rise over a 300 mm length adds about 0.018 mm. On steel it is closer to 0.009 mm. That is not a machine fault; it is physics.

We manage it with warm-up cycles, coolant temperature control, and by scheduling the tightest features early in the run when the machine is stable. For long parts we sometimes probe a reference feature every 90 minutes and update the work offset. That keeps the mean on target even as the frame grows.

Cutting force is the other quiet factor. A Ø12 mm end mill with a 40 mm gauge length will deflect under load. Roughing with a 0.5 mm radial engagement and finishing with a 0.1 mm pass keeps the deflection under a few microns. Climb milling, sharp tools, and a rigid setup all push the same direction.

Thin walls are the worst case. A 1.5 mm aluminium wall will move when you release the vise. The fix is not a tighter tolerance; it is a better sequence, lighter finishing passes, and sometimes a support fixture or a low-melt alloy fill.

• 1
  Warm up firstRun a 20–30 minute spindle warm-up before the first tight cut.
• 2
  Keep coolant stableA ±1 °C coolant loop cuts thermal drift noticeably.
• 3
  Finish light0.1 mm radial engagement keeps tool deflection low.

Machining accuracy is not only size. Form error (roundness, flatness, cylindricity) and surface finish are part of the same story. A bore can be dead on diameter and still be 0.02 mm out of round. That part will fail in a bearing fit even though the size gauge says it is good.

Finish is usually quoted as Ra. As-machined aluminium sits around Ra 1.6–3.2 μm. A careful finishing pass gets to Ra 0.8–1.6 μm. Fine finishing with a sharp tool and a small stepover can reach Ra 0.2–0.8 μm. Going below that on a mill is slow and often unnecessary; a ground or honed surface is a different process.

Form error comes from spindle runout, tool runout, and the number of teeth engaging the cut. A 4-flute cutter in a long reach holder will chatter before it loses size. Reducing radial engagement, increasing rpm within the tool's limit, and shortening the gauge length all help.

When a print combines a tight size, a tight form, and a fine finish on the same feature, the cost climbs fast. It is worth asking whether the function really needs all three at that level, or whether one of them can be relaxed.

• 1
  Size is not formA roundness callout needs its own inspection plan.
• 2
  Finish has levelsRa 1.6 is normal; Ra 0.2 is a deliberate, slower cut.
• 3
  Shorten the toolLess gauge length means less chatter and better form.

Start with the datum scheme. If the primary datum is a face that gets machined in setup two, and the tight feature is cut in setup three, you have already spent a lot of the tolerance budget. Ask whether the feature can move to the first setup.

Next, sort the tolerances. Most features on a typical bracket are ±0.1 mm or looser. A few are ±0.02 mm. One or two might be ±0.005 mm. Those one or two drive the process, the inspection, and the price. Mark them clearly.

Then check the units and the reference. A decimal like 25.4 could be inches or millimeters depending on the title block. Mixed units on the same print cause real errors. We see it maybe once a month.

Finally, ask what the part does. A ±0.005 mm bore for a press-fit bearing pin is a functional need. The same tolerance on a clearance hole for an M6 bolt is waste. Engineers who explain the function get better quotes, because the shop can propose a process that actually holds the tolerance that matters.

• 1
  Check datums firstThe datum scheme drives setup count and stack-up.
• 2
  Find the tight onesOne or two features usually govern the whole process.
• 3
  Explain the functionIt lets the shop choose the right process, not the safest one.