// Key takeaways
- Acceptance criteria are written into the proposal before the station is built.
- False rejects and escapes are two separate figures, each measured on stated samples at a stated speed.
- FAT proves the inspection; SAT proves the reject, signals and fault responses on the real line.
- Records and repeated challenge tests keep the station proven after handover.
What does the sample test prove?
It comes before any design or quotation. You provide good and defective samples that represent real production, including borderline cases judged by your quality team. We image them under candidate cameras, lenses and lighting and record what is and is not visible. The result is a feasibility note with example images, limits and open risks.
A sample test proves that the faults can be seen and told apart from good product. It does not yet prove how the finished station performs at line speed; that is what FAT and SAT are for.
What happens at the factory acceptance test?
The FAT runs on the built station before it leaves the workshop, using an agreed FAT sample set at an agreed speed. It includes challenge samples: marked defective products of each defect type, fed at known positions, alongside good products that show normal variation. The tests cover detection of each defect, false rejects on good products, recipe changeover for every product, no-read and wrong-content code packs, and the station’s outputs. You may witness it. The result is a FAT record with any open points agreed in writing.
Most tuning happens here, before site, so the installation shutdown stays short.
What changes at the site acceptance test?
Image processing that works on the bench does not prove that the reject works on the line. At SAT, after installation, mounting, focus and lighting are re-checked, because site light and vibration differ from the workshop. Real production runs for every product, with seeded defect samples, and the tests extend to the reject and its confirmation, the PLC signals, fault behaviour including restart after a power interruption, and the operator workflow. Limits are fine-tuned with your quality team, each change recorded. The signed SAT report is the basis for handover. The delivery stages page compares FAT and SAT side by side.
How are false rejects and escapes agreed?
Two figures pull against each other. Tightening a limit catches more defects but rejects more good product; loosening it does the opposite. Statistics knows the same pair: NIST’s handbook describes the risk of rejecting something that is in fact fine and the “error of the second kind”, failing to reject something that is not, and notes that one risk rises as the other is pushed down (NIST/SEMATECH e-Handbook).
| Figure | What it counts | How it is measured |
|---|---|---|
| False reject rate | Good products rejected, as a share of good products inspected | A known number of good products, with normal variation, at production speed |
| Missed defect rate (escapes) | Defective products passed, as a share of defective products inspected | Marked defective samples fed at known positions |
| Reject confirmation | Failed products confirmed as removed | Every failed test product checked at the chute or bin |
| Changeover | The right recipe for each product | Each product selected from the PLC or HMI, first product checked |
The sample set, speed and duration are written down with each target. A short test with no misses shows the station caught every defect in that set; it cannot prove the station will never miss one. For critical faults the target and the sample size are set with you. We do not quote typical figures, because they depend on the product, the defect and the samples. How to lower false rejects without letting defects through is covered in reducing false rejects.
What records should a plant keep?
Keep the signed requirement and sample set, the camera settings and reference images, each recipe with its limits and the reason for each, the tested I/O list, the FAT record, the SAT report and the handover set. After handover, log every change to limits and recipes: who changed what, when and why. The commissioning procedure lists these records step by step.
Proof does not stop at handover. BRCGS guidance on clause 6.2.4 expects online label and print verification equipment to be tested at a frequency based on risk, with incorrect packs where practical, checking detection, the reject or line stop, the authorised restart and the no-read alarm; it does not prescribe the method (BRCGS Issue 8 FAQ). Marked test packs and a record sheet, such as those in the inspection checklists, keep the test the same each time.
Design acceptance around defect families and failure modes
Acceptance starts before tuning. Assign sample identities and reference judgements, include realistic acceptable variation, and reserve some parts for the final locked-recipe test. Repeatedly presenting one defective item explores presentation stability but does not create a diverse defect population. Keep the number of distinct products and the number of presentations as separate fields in the report.
| Requirement or condition | Meaning or risk | Practical check |
|---|---|---|
| Faulty items accepted | Missed defect count / faulty items tested | Report separately for each critical defect family |
| Good items rejected | Unnecessary reject count / good items tested | Include normal production variability |
| Measurement agreement | Difference from a reference and its uncertainty | Do not replace with pixel resolution |
| Removal confirmation | Correct item physically removed when commanded | Include adjacent items and restart cases |
Worked example — hypothetical: In a hypothetical trial, all seeded missing-cap samples are rejected but a wrong-label sample passes. A combined headline percentage could conceal the identity failure. Report both groups independently and correct the wrong-label path before re-running a reserved test. The plant decides whether a remaining risk is acceptable; the supplier must not hide it by adding easy samples to the denominator.
NIST’s proportion guidance explains the uncertainty in finite pass/fail trials. JCGM 100 sections 4–6 address uncertainty for measured quantities, while NIST gauge R&R addresses measurement variation. Select the method that matches the actual claim. (NIST e-Handbook, 7.2.4.1 Confidence intervals for a proportion) (NIST e-Handbook, 2.4 Gauge R&R studies) (JCGM 100:2008, Guide to the expression of uncertainty in measurement)
Keep the original image together with the active recipe and the reference judgement when investigating a disagreement. Change one controlled setting at a time and re-run the relevant challenge samples. Once a correction works, verify the neighbouring requirements too: improving contrast for one feature can hide another. Record the revised configuration and the evidence for release so maintenance does not have to reconstruct the decision from memory.
Frequently asked questions
Who signs off FAT and SAT?
The plant does, against the criteria agreed in the proposal: detection on defect samples, false rejects on good samples, changeovers, reject confirmation and fault responses. Open points are listed and agreed in writing rather than left verbal.
Can a vision station be guaranteed to catch every defect?
No. A finite test shows the station caught every defect in the challenge set; it cannot prove it will never miss one in production. That is why the missed defect figure is stated with its sample set and why challenge tests continue after handover.
How often should challenge tests run?
As often as your own risk assessment sets. We hand over marked test packs and a record sheet with the station, so each test is run and recorded the same way.
Sources
- BRCGS Food Safety Issue 8 FAQs (clause 6.2.4, online verification equipment) (BRCGS guidance, checked 2026-10-03)
- NIST/SEMATECH e-Handbook of Statistical Methods: What are statistical tests? (US National Institute of Standards and Technology, checked 2026-10-03; the two kinds of error)
- NIST e-Handbook, 7.2.4.1 Confidence intervals for a proportion (binomial interval methods; read 2026-10-27)
- NIST e-Handbook, 2.4 Gauge R&R studies (measurement study design; read 2026-10-27)
- JCGM 100:2008, Guide to the expression of uncertainty in measurement (sections 4–6: uncertainty evaluation and combination; read 2026-10-27)