Seal and code verified on 100% of bags

The problem

Closure defects arrive in bursts. That is exactly the pattern sampling does not see.

Control is by sampling and by the attention of an operator who is also doing other things. Closure defects are intermittent: they correlate with sewing head wear or with jaw temperature, so they arrive in bursts. That is exactly the pattern periodic sampling does not see. When a bag bursts there is spillage and lost product, but there is something more serious in a plant handling sugar: there is dust in suspension where there should not be any, with everything that implies for housekeeping and combustible material control. And if what fails is the printed code, nobody on the floor catches it: the customer catches it at receiving, and then it is a lot you cannot trace.

• Control is by sampling and by the attention of an operator who is also doing other things.
• Closure defects are intermittent: they correlate with sewing head wear or with jaw temperature, so they arrive in bursts.
• When a bag bursts there is spillage and lost product.
• But there is something more serious in a plant handling sugar: there is dust in suspension where there should not be.

How it fits the IRIS system

Edge — computer vision running on the line itself, with local inference.

Edge — computer vision running on the line itself:

Inference in milliseconds per bag on local hardware. At this line speed, the latency and bandwidth of a cloud solution simply do not allow it.

• An overhead camera and a side camera over the bagging and palletizing line.
• A neural network trained on good and bad images of this plant's specific format: stitched 50 lb bag, heat-sealed bag, big bag, retail bag.
• Inference in milliseconds per bag on local hardware (such as NVIDIA Jetson Orin): at this line speed, the latency and bandwidth of a cloud solution do not allow it.
• Verification of two things in the same pass: closure integrity and printed code legibility — which closes the loop with case 8. The defective bag is rejected before it goes onto the pallet.

See the full IRIS architecture →

Before and after

Closure control, before and after

Aspect	Today	With iLEAN Edge
Coverage	Sampling	100% of bags
Burst-pattern defects	Invisible to sampling	Caught as they start
Who watches	An operator doing three things	Two cameras, doing one
Formats covered	—	Stitched 50 lb, heat-sealed, big bag, retail
Dependence on network or cloud	—	None: local inference
Sugar dust in suspension	A consequence of every burst bag	One less source

Impact estimate

Impact estimate — to validate against 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 current defective bag and complaint rates.
• Counting lost product and avoided cleanups.
• And a dust-in-suspension exposure that goes down, which in a sugar plant is not a quality matter but a safety one.
• Estimate to be validated.

estimated payback of 5 to 12 months depending on the current defective bag and complaint rates, counting lost product and avoided cleanups. *Estimate to be validated*.

And the fair question from the production manager

"How many bad bags does it need to be trained?" — fewer than feared, because the network is trained on good and bad images of this plant's specific format, not a generic catalogue. And it does not decide: it marks, and the line operator resolves. That is the difference between a system that helps and one that gets switched off the first busy day.

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