Packaging Defect on the Fly

The problem

At line speed, a wrong label looks just like a right one.

At line speed, the human eye lets through bags or drums that are poorly sealed, mislabeled or irregularly filled; the error here isn't just cosmetic, it can be a mislabeled allergen.

• The packing line fills bags, pouches and drums of premixes and functional blends at a speed the human eye cannot follow unit by unit, shift after shift.
• Quality samples 1-2% of units, so a poorly sealed bag, an irregular fill or a crooked label goes out on most of the pallet before anyone sees it.
• Here a label error is not cosmetic: the wrong label on the wrong blend can be an undeclared allergen on your customer's production line.
• Complaints about seals and labels arrive weeks later, after the bags have been opened and their content already used in someone else's product, which is the worst moment to find out.

How it fits the IRIS system

Edge on the packing line — every unit inspected, the defective one diverted.

Edge: a camera over the line with a CNN trained on good/bad examples of the specific format infers in milliseconds and diverts the defective unit before palletizing.

A camera over the line and a model trained on good and bad examples of your own formats decide in milliseconds, locally, without waiting for the network. The defective unit leaves the line before the palletizer, not after the customer opens it. And the label check turns the packing line into the last allergen gate before the dock. Every rejected unit is stored with its image, so quality knows why it was pulled and the pattern of defects per format and shift becomes visible.

See the full IRIS architecture →

Before and after

Packing control today versus Edge on every unit

Aspect	Today	With iLEAN Edge
Packed units actually inspected	1-2% by sampling	100%
Seal integrity on bags and pouches	Checked by hand on samples	Checked on every unit
Label matches the blend being packed	Assumed after the changeover check	Verified unit by unit
Irregular or short fill in the bag	Found later, if at all	Seen in the image
Defective bag or drum	Palletized and shipped	Diverted before palletizing
Sealing and labeling complaints	Arrive weeks later	Drastically reduced

1-2% sampling → 100% unit control. Sealing/labeling complaints → drastic reduction.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 5-12 months, depending on line speed, formats and the cost of each complaint.
• From 1-2% sampling to 100% unit control on seal, label and fill.
• Sealing and labeling complaints drop drastically, because the defect is caught on the line and not in the customer's warehouse.
• And a wrong-label allergen risk is intercepted before it leaves the plant, which is worth more than any scrap figure on this page.

Payback 5-12 months · 100% unit control vs. 1-2% sampling — estimate to validate

And the fair question from the production manager

"What if it rejects good bags and slows the packer?" — the false positive is the real risk of any vision system, which is why the model is trained on good and bad examples of your specific formats with that line's lighting, not on a generic model. Reading the printed label text is an anchored task, where the best models drop below 1.5% error [1], and borderline units go to a reject lane for a person to decide, feeding back into training.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.