From Failed First Article to Closed Corrective Action
Run a formal first article, write up a failure with facts, size the containment, feed a disposition, and close a corrective action that stays closed.
Table of contents
A first article that passes is a form to file. A first article that fails is where an inspector earns their pay. Now you need the right facts on paper, a clear answer on how many parts are in doubt, and a root cause that holds up. This follows one job from the formal inspection, through the failure and the nonconformance report (NCR), to a corrective action that stays closed. Reminder: a nonconforming part is tagged and held, and you do not decide alone whether it can be used.
What You Will Be Able to Do
- Plan a formal first article, including how each feature will be measured.
- Fill in the AS9102 forms correctly, and know when a partial first article may apply.
- Write an NCR that a review board can act on.
- Work out containment across machines, operations and lots.
- Find both why a defect happened and why it got out.
- Prove that a corrective action worked.
Planning a Formal First Article
Start before the part arrives. A good plan stops arguments later.
- Confirm the print revision, the purchase order, and any customer flow-down notes. A flow-down is a customer rule passed along through the contract.
- Balloon every requirement, including notes and the parts list.
- Pick a measuring method for each balloon, and check that it is good enough for the tolerance. A common rule is 10 to 1: the gauge should resolve one tenth of the tolerance.
- Flag any key or critical characteristic. These are features that affect safety or function. They usually need every part checked, with records you can trace.
- Gather the material certificates and the special process certificates, such as heat treat or plating.
Print symbols matter here. The diameter symbol is a circle with a diagonal line through it (Ø). A number in a box is a basic dimension: it is exact, and its tolerance comes from the feature control frame that uses it. If you treat a basic dimension as a plain dimension and apply the title block tolerance, you have inspected to the wrong limit.
| Feature type | Usual method |
|---|---|
| Length, step, general size | Caliper, micrometer, height gauge |
| Holes | Pin gauges, bore gauge |
| Threads | Go/no-go thread gauges |
| Position, profile, 3D shapes | Coordinate measuring machine |
| Surface finish | Profilometer |
| Notes such as burrs or marking | Visual check under good light |
A coordinate measuring machine (CMM) uses a touch probe to find points on a part in three dimensions. It suits complex shapes and geometric callouts. Its report is only as good as its program, so check that the datums and the alignment match the print before you accept the numbers.
AS9102 Forms
Aerospace first articles follow AS9102, which sets out three forms. The current Revision C was released in 2023. Your customer's contract may add requirements, or may still name an older revision, so check it first.
| Form | Covers | Common errors |
|---|---|---|
| Form 1, Part Number Accountability | Part, revision, organization, reason for the FAI | Revision does not match the print |
| Form 2, Product Accountability | Materials, special processes, functional tests | Missing certificate or wrong processor |
| Form 3, Characteristic Accountability | Each characteristic and its result | Missing notes, no real values |
Under Revision C, the reason for a full or partial first article is a required field on every report. The person who reviews the report should not be the person who prepared it. Software that fills in these forms is common, and it does not replace reading them.
A partial first article repeats only the characteristics that a change could affect. It is used when a change is limited, such as a new revision that moves one hole. The reason goes on the report, along with what was covered. Some customers demand a full report after certain changes, or after a long gap in production, so check their rules before you decide.
The Failure
Worked example. You inspect the first article of a machined bracket, part 44871 Rev C. The report has 44 lines, and 43 pass.
| Balloon | Requirement | Actual | Result |
|---|---|---|---|
| 17 | Hole location 2.250 +/- 0.003 | 2.2548 | Fail |
| 18 | Hole size Ø 0.2510 +/- 0.0005 | 0.2512 | Pass |
| 19 | Hole location 4.750 +/- 0.003 | 4.7522 | Pass |
Balloon 17 is over its 2.253 upper limit by 0.0018. Balloon 19 passes, but it is also high. Both holes are drifting in the same direction, and that is worth writing down. Five good minutes here can save an afternoon later.
Before you call it a failure, confirm the measurement. Clean the part, check the setup and repeat the measurement. Confirming a reading once, by a proper method, is good practice. Measuring over and over until you get a pass is falsifying a record.
One failed line fails the first article. The job does not start, or it stops.
Writing the NCR
The NCR records what is wrong. It is not the place for guesses or blame. A review board has to be able to read it without calling you.
| Field | Good entry |
|---|---|
| Identification | 44871 Rev C, job 7712, first article, machine 4 |
| Requirement | Balloon 17, hole location 2.250 +/- 0.003 from datum B |
| Actual | 2.2548 on CMM, 2.2546 on repeat, program 44871-C |
| Quantity | 1 confirmed, 11 suspect |
| Where found | First article inspection |
| Evidence | CMM report 3381, tool calibration current |
Weak: "Hole is out."
Strong: "Balloon 17 measures 2.2548, over the 2.2530 limit by 0.0018. Repeat reading 2.2546. Balloon 19 is high by 0.0022 but within limits. Both locate from datum B."
Leave the cause blank if you do not know it. Write what you saw, such as "chips noted on the fixture stop," without claiming it is the cause.
Sizing the Containment
The operator ran 11 more parts while the first article was at the CMM. That should not have happened, but it did, so the 11 parts are suspect. Containment means finding every part that could share the defect, and it has to look wider than one machine.
Worked example. After you tell the lead, you learn more. Machine 4 ran the job, and machine 6 ran the same program last week.
| Group | Count | Why it is suspect | Action |
|---|---|---|---|
| Machine 4, this run | 12 | Same setup as the failed part | Hold, measure every part |
| Machine 6, last week, in stock | 60 | Same program, same fixture design | Hold, sample as quality directs |
| Machine 6, shipped | 40 | Same program, already gone | Quality decides how to tell the customer |
The count of parts held is 12 + 60 = 72, with 40 more at the customer. Now you have a much clearer picture than "one bad part."
Check the time this will take. If the CMM needs 6 minutes a part, 72 parts is 432 minutes, or about 7 hours. That tells your lead whether to add a second gauge or a simple check fixture for this one feature. Quick, honest numbers help people decide.
Report shipped product at once, even before you have all the answers. Only quality or management contacts the customer.
Feeding the Disposition
The disposition is the formal decision on the parts. It belongs to the people with authority, often a Material Review Board (MRB). Your job is to give them what they need.
- The measured values for each suspect part, not just pass or fail.
- What the feature does. Is the hole a clearance hole, or does it locate something?
- Whether rework is possible, and what it would change.
- Whether the customer must approve it. Many contracts do not allow a supplier to accept "use as is" or "repair" without customer approval.
If the board chooses rework, the part is corrected by a written method and inspected again, and it is released only if it meets the specification. If the board chooses scrap, the part is defaced so that it cannot return to stock.
Root Cause: Two Questions
A corrective action has to answer two questions.
- Why did the defect occur? This is the occurrence cause.
- Why did we not catch it sooner? This is the escape cause.
An is / is not table is a good place to start. It sets what you know against what you might have expected, and it narrows the search.
| Question | Is | Is not |
|---|---|---|
| What | Hole location high in X | Hole size, other axes |
| Where | Holes located from datum B | Holes located from datum A |
| When | After the fixture was reinstalled | Prior runs on this fixture |
| How much | About 0.002 to 0.005 | Random scatter |
All the bad readings run the same way, and they started when the fixture came back. So the fixture and its stop are the first place to look. Then use five whys, with evidence at each step.
| Why | Answer | Evidence |
|---|---|---|
| Why is the hole high in X? | The part sits away from the stop | Reading shifts when the stop is cleaned |
| Why does it sit away? | Burr on the part edge holds it off | Burr found on 9 of 12 parts |
| Why is there a burr? | Saw cut edge is not deburred | Traveler has no deburr step |
| Why no deburr step? | Removed when the routing changed | Routing Rev D change note |
| Why did that pass review? | Routing changes do not get a quality review | Procedure has no such step |
That last answer is the true root cause. Fixing only the burr will not stop the next one.
The escape cause gets its own line. Why did the operator run 11 more parts during the first article? Because the job's hold point was only verbal. The fix is a system hold, so the next operation cannot start until the first article is signed.
Many companies use the 8D format (eight disciplines), which holds this same logic: team, problem, containment, root cause, chosen action, carried-out action, prevention and closure. Customers often require it for a supplier corrective action request.
Good and Weak Corrective Actions
Some actions look like progress and change nothing. Rank what you propose by how much it depends on people remembering.
| Strength | Example |
|---|---|
| Strong: makes the error impossible | Fixture stop relieved so a burr cannot hold the part off |
| Good: detects the error at once | Go/no-go check at load, or a probe cycle |
| Fair: changes the system | Routing changes need a quality review |
| Weak: depends on memory | "Retrain the operator" or "Be more careful" |
Retraining alone is seldom a real answer. If a trained person could make the same mistake tomorrow, the cause is still there.
Proving It Worked
An action is not closed when it is done. It is closed when the data shows that the problem is gone.
- Set the measure before you start. For example: no location failures on this feature across the next three lots, with Cpk at or above the plan's target.
- Run the check. Say balloon 17 on the next three lots averages 2.2502, with no failures.
- Compare that with the old data, which ran about 0.004 high. The shift is gone.
- Look for side effects. Did the new stop change anything else?
- Update the control plan, the work instructions, and any risk documents.
- Close the record, or keep it open if the numbers say so.
After the fix, a partial first article covers the features the change could affect. Record the reason on the report.
Common Mistakes
| Mistake | What it costs |
|---|---|
| Measuring until a line passes | False records and a defect shipped |
| Containing only one machine | Bad parts at the other machine or at the customer |
| Writing the cause on the NCR too early | The search stops at the first idea |
| Skipping the escape cause | The next defect also gets out |
| Closing on "retrained" | The same NCR returns in six months |
| Closing with no data | No proof the fix worked |
| Using a partial first article without checking the customer's rule | A report the customer rejects |
Check Your Understanding
1. A first article passes 61 of 62 lines. The failed line is a general note about burrs. Can the job start?
No. A note is a requirement, and one failed line fails the first article. Fix the cause, then recheck that line under your plan.
2. A print revision moves one hole. The contract is silent on partial first articles. What do you do?
Check your procedure and ask quality. A partial may fit, covering what the change could affect, with the reason recorded. If you are not sure the customer allows it, get an answer before you rely on it.
3. A corrective action reads, "Operator retrained on setup." What do you ask?
What in the process allowed the error, and what will stop it next time if the operator is new or rushed? Look for a change that does not depend on memory, and for the reason it was not caught sooner.
4. A location error shows on the first article from machine 3. Machine 5 uses the same fixture design and program. What does containment include?
Parts from both machines, including stock and anything shipped. A common cause in the fixture or program would affect them both.
5. Your corrective action is finished, and three lots have run clean. What else closes it?
Compare the new data with the old, confirm no side effects, and update the control plan and work instructions, then close the record.
Key Takeaways
- Plan the first article before the part arrives, including how each feature will be measured.
- One failed line fails it. Confirm a reading properly, and never measure until it passes.
- An NCR holds facts and real numbers, not guesses or blame.
- Containment reaches across machines, stock and shipped parts.
- Find both why it happened and why it got out.
- Prefer fixes that do not depend on memory, and close only when data proves it worked.