This is a ready-to-use qualification protocol. Replace every <<FILL: ...>> placeholder with your own specifics and route it through your normal validation review and approval. A worked filled specimen follows the template. It is an educational aid to adapt and verify, not a substitute for the standards it references or your validation program.
Approval page
| Field | Entry |
|---|---|
| Protocol title | Automated Visual Inspection Qualification by the Knapp Test |
| Protocol number | <<FILL: protocol ID>> |
| Machine / system | <<FILL: AVI machine ID and product in scope>> |
| Version | <<FILL: version>> |
| Author | <<FILL: name, role>> |
| QA / Validation approver | <<FILL: name, role>> |
1. Objective
Demonstrate that the automated (or semi-automated) visual inspection machine performs at least as well as qualified manual inspection for the product-and-container combination in scope, using the Knapp method, sometimes called the Knapp-Kushner method, which compares machine detection against a manually characterized test set.
2. Scope
This protocol qualifies inspection performance. The machine is also a computerized and measurement system, so it additionally requires equipment qualification (IQ/OQ/PQ) and computerized-system controls (audit trail, access control, data integrity) covered by <<FILL: reference to EQ/CSV documents>>. This protocol addresses the detection-performance comparison specifically.
3. Prerequisites
- The machine has completed installation and operational qualification, and the inspection recipe (lighting, camera, thresholds, rotation) is defined and under change control.
- A characterized Knapp test set exists (section 5.1) and is within its validity period.
- Qualified manual inspectors are available to establish or confirm the manual benchmark.
4. Roles
| Role | Responsibility |
|---|---|
| Validation lead | Executes the protocol, records results, dispositions deviations. |
| Qualified manual inspectors | Provide the manual characterization runs that set each unit’s detection probability. |
| QC / QA inspection owner | Reviews the comparison against acceptance criteria. |
| Engineering | Maintains recipe, lighting, and reject-mechanism configuration under change control. |
5. Procedure and test cases
5.1 Build and characterize the test set (Test case K1)
- Assemble a test set of units spanning the full range from clear reject-zone defects, through the gray zone, to accept-zone (undetectable) units and good units.
- Characterize each unit by manual inspection over many repeats (a defined number of qualified inspectors, each inspecting each unit several times) and assign each unit a detection probability from the fraction of times it was correctly rejected.
- Group the units into zones by their manual detection probability (reject zone, gray zone, accept zone) per your defined POD bands.
| Item | Detail |
|---|---|
| Manual repeats per unit | <<FILL: e.g. 10 inspectors x N reads>> |
| Zones assigned | Reject / gray / accept per <<FILL: POD band definition>> |
| Expected | Every unit carries a manual detection probability and a zone |
| Actual / Pass-Fail | <<FILL>> |
5.2 Run the machine against the same set (Test case K2)
- Run the identical test set through the machine (or the semi-automated setup) a defined number of times, in randomized order, under the qualified recipe.
- Record, for each unit, the fraction of machine runs that rejected it (the machine detection probability).
5.3 Compare performance (Test case K3)
- Compute the reject-zone efficiency for the machine (the aggregate detection across the reject-zone units) and compare to the qualified manual benchmark.
- Also examine the gray-zone and accept-zone performance, so the machine is not merely matching humans on easy units while doing worse on the harder ones.
6. Acceptance criteria
| Metric | Acceptance criterion |
|---|---|
| Reject-zone efficiency | Machine efficiency >= the qualified manual benchmark |
| Gray-zone performance | Machine not materially worse than manual across the gray zone <<FILL: define the tolerance>> |
| Accept-zone / good units | False-reject rate on good units at or below <<FILL: limit>> |
| Recipe under control | The qualified recipe is the version tested and is change-controlled |
The machine is qualified when its reject-zone efficiency meets or exceeds the manual benchmark and it is not worse than manual across the gray and accept zones, at an acceptable false-reject rate.
7. Periodic re-challenge
Even a qualified machine is challenged on a defined frequency (commonly each shift or batch start, and periodically with a full defect set) using a known challenge kit, to prove it still rejects what it should. Record the re-challenge and act on any drift. A creeping false-reject rate signals drift, a new component lot, or a degraded sensor.
8. Deviation handling
Any departure from this protocol during execution is recorded, assessed for impact on the qualification decision, and dispositioned before qualified status is granted.
9. Summary and conclusion
| Field | Entry |
|---|---|
| Reject-zone efficiency: machine vs manual | <<FILL>> vs <<FILL>> |
| Gray-zone comparison | <<FILL>> |
| False-reject rate on good units | <<FILL>> |
| Overall result | Qualified / Not qualified |
| Scope | <<FILL: machine, product, container, recipe version>> |
| Approved by (name, date) | <<FILL>> |
10. References
USP General Chapter <1790>, Visual Inspection of Injections (probability of detection, comparison of automated to manual inspection). USP General Chapter <790>, Visible Particulates in Injections. EU GMP Annex 1 (2022), on validation of the inspection process and periodic challenge of automated systems. Your equipment qualification and computerized-system validation procedures for the IQ/OQ/PQ and data-integrity controls.
Confirm the current version and clause numbers of each reference before issue.
Revision history
| Version | Date | Author | Summary of change |
|---|---|---|---|
<<FILL: 1.0>> | <<FILL: date>> | <<FILL: author>> | Initial issue. |
Filled specimen
An illustrative comparison for a vial line AVI machine. Numbers are examples.
| Zone | Units | Manual benchmark efficiency | Machine efficiency | Result |
|---|---|---|---|---|
| Reject zone | 50 | 0.94 | 0.97 | Machine >= manual, pass |
| Gray zone | 40 | 0.55 | 0.58 | Not worse than manual, pass |
| Accept zone / good | 60 good | 3% false reject (manual) | 2.5% false reject (machine) | Within limit, pass |
Decision: qualified for the vial product at recipe version 4, subject to per-shift re-challenge with the defect kit. The machine matched or beat manual in the reject zone (the primary criterion) and was not worse in the gray zone, so it did not simply excel on easy units, and it did not over-reject good product.
Common inspection findings this protocol prevents
- Claiming an automated machine is equivalent to humans without a Knapp-style comparison.
- A machine that matches humans on easy units but is worse in the gray zone, passed on reject-zone data alone.
- An inspection recipe changed without change control, so the qualified state is not the running state.
- No periodic re-challenge, so drift or a degraded sensor goes unnoticed.
How to adapt this protocol
- Size the test set and set the number of manual characterization repeats for a stable per-unit detection probability.
- Define your POD bands and the gray-zone tolerance and false-reject limit for acceptance.
- Tie the recipe version to change control and set the re-challenge frequency in the parent inspection SOP.
- Reference your EQ and CSV documents for the machine’s qualification and data-integrity controls.
- Confirm each reference in section 10 against its current published version before issue.