The first article was perfect. Part 180 was not.
That is the short version of a 250-piece run we finished this month: 6061-T6 aluminum housings, 96 mm long, one Ø12 H7 bore, an 8 mm slot, and a four-hole bolt pattern. General tolerance ±0.025 mm, twelve working days, first article signed off at 09:40 on a Tuesday. By the afternoon of day four the slot width had walked past its upper limit on two parts and we stopped the machine.
When a run goes sideways like that, an opinion is not worth much. Data is. So we pulled every measurement we had: four critical features on each of 250 parts, 1,000 values, all logged by the same operator on the same CMM.
What we actually measured
The four features were chosen because each one is produced a different way, and that turns out to matter more than the tolerance number on the drawing.
The Ø12 H7 bore is reamed — a fixed-size tool cuts the last 0.2 mm and leaves the same diameter every time until it wears. The 8 mm slot is end milled with a 4-flute carbide tool and a 0.2 mm finishing pass. The four Ø5.5 holes are drilled on a Ø68 bolt circle, position Ø0.25 to datums. The 96 mm overall length is set in the second operation, on soft jaws, after the part has been flipped.
| Feature | How it is made | Band | Spread we saw | Cpk |
|---|---|---|---|---|
| Ø12 H7 bore | Reamed | 12.000–12.018 | 12.004–12.014 | 1.72 |
| 8 mm slot | End milled | 8.000–8.050 | 8.009–8.046 | 0.91 |
| Ø68 bolt circle, 4× | Drilled | Ø0.25 position | Ø0.04–Ø0.13 | 1.85 |
| 96 mm length | Op 2, soft jaws | 95.975–96.025 | 95.982–96.024 | 1.08 |
Read that table as a ranking of how much each process wants to behave. The reamed bore and the drilled pattern sat comfortably in the middle of their bands. The milled slot and the two-operation length did not. Same machine, same operator, same week — the difference is the method, not the skill.
Three causes, in order of size
Temperature, and this was the big one. Parts came off the machine at 33 to 35 °C — spindle heat, coolant, twenty minutes of handling. Our inspection room runs at 20 °C. 6061 expands 23.6 µm per metre per kelvin, so a 96 mm length grows about 2.3 µm for every degree. Fourteen degrees of difference is 0.032 mm, which is more than the entire ±0.025 mm band. The parts were not drifting in the cut. They were shrinking on the bench, and we were catching them at a different point on that curve depending on how long they sat in the queue.
Tool wear on the slot. Slot width went from 8.012 mm at part 20 to 8.041 mm at part 180. That is 0.029 mm over 160 parts, about 0.00018 mm per part. Perfectly linear, entirely predictable, and invisible if you inspect only the first article and the last one on the same shift. A 4-flute carbide end mill in 6061 at 300 m/min does not fail. It just gets slightly smaller, and the slot gets slightly wider.
Op-2 fixture repeatability. Soft jaws, cut in place, clamped at 40 Nm. We measured 0.008 to 0.015 mm of variation in the length feature across three re-clamp cycles on the same part. Small on its own, but it stacks on top of the thermal number and it is why the length feature finished second-worst on the list.
What we changed
Nothing in the program. All three fixes sit around it.
Parts now sit for forty minutes before final measurement, or get air-cooled to within 2 °C of the room. That one change removed most of the length scatter and costs nothing except floor space.
The 8 mm end mill gets indexed at part 60 and part 120 instead of being run to the finish limit. That adds about USD 9 of tooling to the run and took 0.02 mm of drift out of it.
And we moved the reamed bore target from 12.002 to 12.008. Reaming is stable, so we hold the middle of the band and let the 0.004 mm of tool wear spend itself inside the tolerance instead of against the lower limit.
What this means when you are writing the drawing
Put your tight numbers on the features the process is naturally good at. A reamed bore holds 0.010 mm all day. A drilled-and-reamed hole pattern holds Ø0.10 mm position all day. A milled slot width and a two-operation length do not, because they are open-loop features that drift with heat, tool wear and clamping force.
If a milled slot genuinely has to hold ±0.01 mm, that is a fine thing to ask for — just say so on the drawing and expect in-process checks every 25 parts plus a longer lead time. What actually costs money is not the tight tolerance. It is a tight tolerance on a feature nobody measures, discovered at final inspection when 250 parts are already made. That is also the failure mode our first-article inspection process is built to catch early, and the numbers behind every band in the table above come from the same reference we hand to customers as a tolerance quick reference.
This run was 250 pieces of 6061-T6, twelve working days, about USD 4.80 per part at that quantity. The same housing at ten pieces lands closer to USD 11.20, because programming and fixturing get divided by ten instead of 250. Material pricing moves with the market, so treat that as a budget estimate based on our latest purchase price rather than a fixed number — send the drawing through our instant quote page or straight to our team and we confirm within one business day.
A thousand measurements, one conclusion: tolerance does not get lost in the cut. It gets lost after the cut, in the twenty minutes between the machine and the gauge.