Empty container inspection — one particle in a bottle contaminates the whole product.

The problem

The empty container is the most underestimated link in bottling.

On a beverage line the focus is usually on the filler, the capper and the labeller. The empty container entering the line is assumed to be clean by default — and most of the time it is. The problem is what happens on the day it is not:

1. Returnable glass — it comes back from the street. The washer does an enormous job, but a small percentage arrives with residue, a badly stripped label, a fragment of a previous cap or, occasionally, with whatever the consumer put inside and the washer failed to remove.
2. New glass and PET — they come from the container supplier. Tracing a supplier defect is a classic problem: by the time the line discovers a defective series, three pallets have already gone in.
3. A transparent particle — the worst case. A human operator, holding the bottle up to the light, catches most of them; a human operator at 36,000 bph catches nothing.

When a particle gets through, the cost is not the bottle: it is the whole batch pulled from the shelf, the brand damage, the AESAN/RASFF notification and the retailer audit. And the quality director's question is not "did it happen?", it is "how do you prove it will not happen again?". End-of-line inspection is necessary, but it arrives too late — the right place is before the filler, not after.

How it fits the IRIS system

iLEAN does not add one more inspection — it puts the veteran's eye where nobody can stand.

The empty container problem is not a lack of method (the classic end-of-washer inspector exists) — it is speed. No human inspects every bottle at line speed. iLEAN acts as the putty that seals the crack between the washer and the filler: the veteran's eye, installed at the exact physical spot, working 24/7.

Edge sees every empty container on the conveyor. Connect reads the washer and warns when something upstream is wrong. The agent cross-references reject rate, pallet supplier and shift, and proposes the root cause. The person validates.

The iLEAN pieces applied to empty container inspection:

• Edge + Vision — a terminal with a CNN over the conveyor at the washer exit (or at the filler infeed). It reads every empty bottle: breakage, particles on the base and walls, label residue, residual contamination. An actuator (pneumatic ejector) fires in milliseconds. It works without a network: critical inspection does not depend on the plant's WiFi.
• Connect — captures the washer readings (temperature, caustic, cycles) — from a modern PLC, an old data logger or a photographed display — plus the pallet quality check at goods-in (a photo of the supplier's delivery note, the container batch). It seals the cracks between the inspection and the upstream systems.
• Agent — cross-references reject rate per minute + shift + container supplier + washer cycle. If the rate steps up, it identifies the common factor (a supplier batch, the washer start time with diluted caustic, one specific format) and alerts the shift lead before rejects become massive. The decision to stop the line is the person's.

See the full IRIS architecture →

Before and after

Classic inspection vs. AI inspection with iLEAN

Aspect	Human inspector + end-of-line control	With iLEAN Edge + Connect + Agent
Coverage per shift	Sampling + the attention of the operator at the head of the line	100% of bottles, without fatigue
Transparent particle detection	Limited to human speed	Vision with structured lighting — at line speed
Traceability of rejects	Aggregate count at the end of the shift	Image + timestamp + reason, bottle by bottle
Detecting an upstream problem	Reactive — once rejects pile up	Anticipated — the agent flags the step change
Operation without a network	n/a	Edge keeps inspecting autonomously
Quality file per shift	Rebuilt by hand	Automatic dossier with the reject curve and causes

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Beverage bottling plant (soft drinks, beer, water, juices), returnable or new glass, 24,000-50,000 bph on the main line.
• Edge pilot at the washer exit or the filler infeed (CNN camera + lighting + pneumatic ejector). First value expected within a few weeks: detection above 90% on the typical defects identified during the diagnostic.
• Reduction in undetected particles / residual contamination of ≥ 30% versus the baseline — the real lever is eliminating the blind zone between the washer and the filler.
• Indicative payback between 4 and 9 months, depending on the frequency of documented contamination incidents and the average cost of a recall or rework.
• Additional hard lever: meeting the retailer's quality standard and IFS/BRC audits — bottle-by-bottle traceability with an image is exactly the kind of evidence being demanded more and more.

And the context that rarely gets quoted

The benchmark quality standard in automotive is of the order of 25 PPM defective^[1]. Empty container inspection in beverages does not need to get there, but the ratio that decides a committee is not that one — it is the cost of a single particle getting through and ending up on social media. On the reasonable fear of AI: hallucination is a problem of free generation, not of anchored tasks — classifying an image of a bottle as "clean / with particle" is exactly an anchored task, and on that ground the best models brought error below 1.5%^[2]. And the critical decision (stopping the line, checking the washer) remains the person's.

^[1] Automotive quality standard of the order of 25 PPM — Symestic.

^[2] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.