The full system against the drum that fails in transit

The problem

One drum that fails in transit can cost the rating itself.

The capital pain of any drum maker for dangerous goods is the same: zero nonconforming UN-rated drums shipped to a customer. A single drum that fails in transit — a closure that opens, a hoop that gives, leak-tightness that fails — can cause a chemical or pharma spill, with regulatory sanctions, loss of the rating and a broken customer relationship.

• The capital pain of any drum maker for dangerous goods is the same: zero nonconforming UN-rated drums shipped to a customer.
• A single drum that fails in transit — a closure that opens, a hoop that gives, leak-tightness that fails — can cause a chemical or pharma spill. The drum maker is the first one asked for evidence.
• The consequences go well beyond a claim: regulatory sanctions, loss of the rating and a broken customer relationship. For a drum maker, the rating is the license to sell to the customers that pay best.
• And the failure rarely has one cause: a wrong closure format, a badly calibrated change and a test signed in a hurry each pass their own control. Together, they make the drum that opens on a truck.

How it fits the IRIS system

Cross-checking rings — the previous pieces, coordinated by Agents.

The iLEAN system coordinates the previous pieces into several cross-checking rings: the ERP's active order defines format and rating; that information travels automatically to the sister plant's closure press; the Edge camera on assembly validates that the fitted closure is correct, and if it doesn't match, JIDOKA AI stops the line; and every batch ships with a dossier cross-referencing the previous rings plus the test photo and quality sign-off. SMED AI speeds up format changes so the coordination doesn't hurt throughput.

No ring is new: each already exists in the earlier cases. The strength is in the cross-check, because a nonconforming drum gives itself away when the order, the closure, the line and the evidence stop agreeing. Each ring alone is a control that already exists somewhere; crossed, they leave a drum nowhere to hide.

See the full IRIS architecture →

Before and after

The rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · Order	Customer alert (3) + batch startup (1)	Format and rating defined by the live order
2 · Closure	Closure press (8)	The right closure format from the sister plant
3 · Assembly	Closure defect (9) + verification tablet (5)	JIDOKA AI stops the line on a mismatch
4 · Format change	Format change validation (10)	No startup without calibrated evidence
5 · Dossier	Evidence pack (11) + lab spreadsheet (6)	Every batch ships with test photo and sign-off
6 · Throughput	SMED AI on format changes	Coordination without losing output

Before: risk of a nonconforming drum dependent on manual controls. After: discrepancies caught on the line, never at the customer.

Impact estimate

Impact estimate — to be validated with quality and management.

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 6-12 months, measured against the first avoided nonconformity. One drum that does not fail in transit is the whole business case.
• It comes from protecting the rating and the customer relationship, which is what a drum maker for dangerous goods sells. Drums are the product; trust in the rating is what the customer buys. The rings protect that trust batch after batch.
• From the risk of a nonconforming drum depending on manual controls to discrepancies caught on the line, never at the customer.
• And SMED AI keeps format changes fast, so the extra checks do not cost throughput.

Estimated payback 6-12 months against the first avoided nonconformity, by protecting the rating and the customer relationship. Estimate to be validated with the quality lead and management.

And the fair question from the production manager

“Do we need all twelve cases before this works?” — no. The rings are built in layers and each case pays on its own; batch startup and the closure press already give you the first two. Cross-checking that the order, the fitted closure and the test record agree is an anchored task, where the best models drop below 1.5% error [1], and every block is confirmed by a person. Agents coordinate; people decide.

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