FAI to Production: A Practical Inspection Plan for 5-Axis and CNC Batch Runs
The most common quality failure I see in CNC production isn't a bad machine or a bad machinist. It's an inspection plan that doesn't match the volume. Either every part gets treated like a first article, which destroys the price, or a 50,000-piece run gets checked like a prototype, which produces a recall.
Here's the framework we use, and the reasoning behind each step.
Stage 1 — First Article Inspection (FAI)
Purpose: prove the process, not the part.
- Every dimension on the drawing measured, referenced to stated datums
- GD&T evaluated functionally, not just as numbers (a true position callout requires the actual mating feature as a datum simulator)
- Material certification and hardness verified against the purchase order
- Surface finish measured where specified
- Results documented against the drawing with a serialized report
FAI answers one question: can this process make this part? If the answer is anything other than a clear yes, do not proceed to volume. Fix the process first.
Stage 2 — Process Capability
Before releasing a run, measure capability on the critical characteristics — typically 5 to 15 features, not all 200.
- Take 25–30 consecutive parts from the same setup
- Calculate Cpk on each critical feature
- Cpk ≥ 1.33 is the usual production floor
- Cpk ≥ 1.67 for safety-critical or medical features
If a feature lands at Cpk 1.0, the process will produce scrap on normal tool wear. The fix is usually shifting nominal to the middle of the tolerance band, or adding an in-process check — not tightening the machine.
Stage 3 — Production Sampling
The sampling plan must reflect two things: how much the process drifts, and what a failure costs.
| Volume | Sampling approach |
|---|---|
| 1–50 | FAI + 100% visual, dimensional spot check at start/middle/end |
| 50–1,000 | FAI + capability study + 1 in 20 dimensional, 100% critical features |
| 1,000–20,000 | FAI + capability + hourly SPC on critical features + 100% critical features |
| 20,000+ | All of the above + in-process probing on the machine |
Two principles matter more than the exact numbers:
- Sample the drift, not just the parts. Tool wear, thermal growth, and fixture settling produce trends. Sampling at the start of a shift and again at the end catches what random sampling sometimes misses.
- 100% inspect the features you cannot recover from. A safety feature that occasionally fails must be checked on every part, regardless of volume.
Stage 4 — In-Process Probing
On 5-axis work, in-machine probing is often cheaper than post-process inspection because it catches the error before the part leaves the fixture. A probe cycle that takes 40 seconds and corrects a work offset can save an entire scrapped part.
Good uses of probing:
- Verifying stock position on castings and forgings with variable material
- Re-establishing a work offset after a fixture change
- Checking a critical bore before the finishing pass, so you can adjust
Bad uses of probing:
- Substituting for a real measurement of GD&T relationships between features
- Probing on a machine whose thermal state hasn't stabilized
What a supplier's inspection report should contain
- Drawing revision the inspection was performed against
- Actual measured values (not just pass/fail)
- The datum reference frame used
- Measurement equipment and calibration status
- For capability studies: sample size, Cpk, and which features were studied
- Deviation log for anything outside nominal but inside tolerance
If you receive a report with only checkmarks, you don't have an inspection report. You have an assertion.
The document you should ask for before placing a volume order
Ask for the Control Plan. It answers, in one page: what is checked, how often, with what instrument, by whom, and what happens when it fails. A supplier who cannot produce a control plan on request is not ready for your volume.
About the author
I work with CHX Precision Metal, a CNC machining manufacturer in Shenzhen, China — 220 CNC machines including 100 Taiqun T600 5-axis machining centers and 70 Tsugami Swiss-type lathes — operating under ISO 9001:2015 and ISO 14001:2015. Prototyping in 3–7 days, production in 7–15 days, MOQ from 1 piece. More on our CNC machining and quality process: chxmetal.com
If you have a Cpk problem on a specific feature, describe the callout and the observed spread in the comments — it's usually solvable without changing the process.
Quick answers
Short, self-contained answers to the questions this article is usually asked. Full reasoning is in the sections above.
What should a first article inspection plan include for 5-axis parts?
A workable FAI plan covers every drawing dimension with a stated method, the datum reference frame actually used in machining, true-position and profile results reported as deviation from nominal rather than pass/fail, material and finish certificates, and a ballooned drawing that maps each callout to its measurement. On 5-axis parts the critical addition is recording the rotary axis positions used, because a feature measured away from its machining orientation can pass FAI and still fail in assembly.
How many parts should I inspect during a production run?
Start with a capability study of at least 30 consecutive pieces on the tightest callout to establish Cpk, then move to a sampling frequency driven by drift rate rather than a fixed percentage. A practical default is 3–5 pieces per shift plus first-off and last-off, tightened to 100% for any feature whose Cpk is below 1.33 or whose tolerance is consumed mostly by tool wear.
What Cpk should I require from a CNC supplier?
Require Cpk ≥ 1.33 for standard production callouts and ≥ 1.67 for safety-critical or assembly-critical features. Below 1.33 the process is not capable — it is being sorted into compliance, and the escape rate will rise as soon as attention moves elsewhere. Ask for the study data, not the number: 30 consecutive parts, same lot, same operator, with the measurement method stated.
Why do parts pass inspection at the supplier and fail at incoming?
The usual causes are different datum simulation (the supplier measured in the machining fixture, the customer measures free-state), thermal differences between the two metrology environments, and gauge disagreement beyond the rule of 10. Fix it by agreeing the datum scheme and the measurement method in writing before the first batch, and by exchanging a certified reference part between both measurement systems.