Label verification SOP: checks at line clearance, start-up and during the run, and when to automate

What should a label verification SOP check?

One practical way to structure the procedure is three checkpoints, each with its own record. Some sites call this a three-point label verification; the name matters less than having each point written down, signed and kept.

The weak point is the third row. A spot check proves the label was right when someone looked. If a wrong reel is loaded just after a check, every unit until the next check carries it.

What do FSSC 22000 and BRCGS expect?

FSSC 22000 version 6 expects line start-up and changeover controls that ensure labelling and packaging from the previous run have been removed from the line (FSSC 22000 v6, 2.5.9(c)). For packaging manufacturers it also expects print control, including approval of each run against the master sample and a process to detect and identify printing errors during the run (FSSC 22000 v6, 2.5.2(d)).

BRCGS guidance on clause 6.2.4 says that a site using online label or print verification equipment must be confident it rejects non-conforming packaging, and that good practice is to challenge it with incorrect packs and test detection, the reject or line stop, the authorised restart and the no-read alarm (BRCGS Issue 8 FAQ). A South African consultancy summary of BRCGS Issue 9 lists label reconciliation, changeover checks and in-run checks under clause 6.2 (consultancy summary of BRCGS Issue 9); check the standard itself for the exact wording that applies to your site.

None of these requires a camera. They require control, and records that prove it.

When should label checks be automated?

Manual checks struggle in a few predictable situations. Fatigue on long, repetitive checks is one of them: a check sheet records that a check was done, not what was missed between checks.

How a label inspection station is built, and what it can and cannot check, is on label inspection system. For date and batch codes on the same label, see OCR/OCV inspection.

What stays in the SOP after a camera is fitted?

• Line clearance. A camera checks the label it sees, against the recipe it is given. Old reels on the line are still a human check.
• Recipe selection. If the wrong recipe is selected, the camera checks against the wrong label. Selection from the PLC, and a check of it at changeover, belongs in the SOP.
• Artwork approval. Whether the approved label is legally correct is decided before the run, not by the camera.
• Challenge tests. Wrong and missing labels are passed through at start-up and after changes, and the result is recorded. The challenge test record template covers this.

Make the SOP record the comparison and the disposition

An effective record identifies the approved artwork, product order, expected variable data and camera recipe revision. It then records what the challenge actually tested. “Label checked” does not establish whether a wrong variant was rejected, whether a missing label was detected or whether a failed pack was physically removed. Keep those outcomes separate so the next shift can reproduce the test without guessing what the previous operator meant.

Hypothetical SOP example: at a product change, present a correct first-off pack, a pack with an old but readable product label and a pack with the new label but an obsolete batch. The first demonstrates the intended configuration; the others test identity and variable-data comparison. Record the failure reason and follow each challenge to its controlled disposition. This example supplies a procedure pattern, not a universal audit frequency or a claim about a particular factory.

The dated R146 text gives the South African batch/date labelling context in regulations 11 and 12, while GS1 General Specifications section 3 defines the relevant encoded fields when GS1 identifiers are used. (South African food labelling regulations R146, published 1 March 2010) (GS1 General Specifications, Release 26.0) Neither means that decoding any symbol is sufficient. The SOP must identify the source of the expected product, batch and date values and show how that source is approved.

Keep the current certification scheme and customer requirements in the controlled procedure. The older BRCGS FAQ and FSSC version cited earlier are dated supporting references, not a claim that they are the current version for every site. Where a referenced document has moved or an audit scheme changes, review the site’s applicable requirements rather than copying a historical clause number uncritically.

After a challenge fails, retain the image, expected data, recipe and physical sample where practical. Identify the potentially affected product since the last valid check, apply the site hold procedure and investigate the cause. Repeat the failed challenge after correction; a new good pack passing is not enough to establish that the wrong-label path works. Changes to a learned classifier require independent evaluation data, following the training/validation/test separation described by Google. (Google Machine Learning Crash Course: dividing datasets)

This SOP article defines the operating evidence and escalation path. The label-inspection application page addresses camera design, and the changeover article addresses the transition between runs. Together they support a controlled process, while food-label approval, quality release and any certification decision remain with the responsible business functions.