SHE traceability of MDI in PU sealant — the isocyanate data cannot arrive after the valve is open.

The problem

MDI data does not come in through a sensor — it comes in through the warehouse manager's inbox.

At a PU sealant and adhesive plant, authorizing the unloading of an ISO tank of MDI or TDI means deciding, at second zero, whether this specific tank fits the day's formula and the area's authorized SHE capacity. That decision crosses three realities that are almost never in the same place in time:

1. The supplier's delivery note — batch, NCO%, manufacturing date, weight. It arrives by email, a WhatsApp from the carrier, a paper signed in the cab. The warehouse manager or the receiving lead gets it.
2. The day's formula and the stock — what is going to be made with this MDI, what stock is already on hand and how it fits the recipe signed off by R&D. It lives in the ERP, in the MES or in a spreadsheet updated in the morning.
3. The authorized SHE capacity — Seveso III, REACH, the area's storage authorization, ATEX conditions, confined space. It lives in the SHE system or in a PDF on the SHE lead's computer.

What usually happens: the truck arrives, the operator unloads because the bay is free and the paper is signed by the carrier, and the delivery note data ends up in the ERP at the end of the shift. If this tank did not fit the formula, or the NCO% was outside the acceptable range, or the stock plus this delivery exceeded SHE capacity, the problem is discovered afterwards. The classic system works 99% of the time. That 1% is a SHE incident, a nonconformity or an exposure list that is hard to explain.

How it fits the IRIS system

iLEAN does not replace the ERP or the SHE system — it captures what comes in through the gate and puts it in front of whoever signs.

The MDI unloading problem is not a lack of protocol: it is that the critical information almost never comes in through a sensor; it comes in through an email, a WhatsApp or a piece of paper. iLEAN acts as the putty that seals that crack between the reality of the plant and the systems you already have, without asking you to change your ERP, your SHE system or your supplier.

Connect captures the delivery note through whatever channel it arrives. The agent cross-checks it against the formula, the stock and the SHE authorization. The person signs before the valve is opened — the tank does not unload on its own.

The three iLEAN pieces applied to SHE traceability of MDI in PU sealant:

• iLEAN Connect — captures the MDI delivery note by email, WhatsApp, a photo of the signed paper, a transcribed call from the carrier. It extracts batch, NCO%, manufacturing date and weight. It enters the ERP, the SHE dashboard and the receiving lead's screen at second zero, with nobody having to forward anything. It is bidirectional: it can also prepare the reply to the supplier through the same channel — always with a human signature.
• iLEAN Agents — cross-reference the delivery note data with the day's formula (from the ERP/MES), the available stock (from SHE) and the area's SHE authorization (Seveso, ATEX). If something does not add up, they do not send an email at 10pm: they alert the SHE lead through whichever channel they use, before unloading. At closing, they leave a dossier per delivery ready for a REACH or Seveso auditor.
• The three rings — the delivery note data enters the outer ring, the agent validates it in the middle ring (was this supplier due today? does this batch exist in the master?), and only then does the signed data enter ring 1 (the OT network where tank control lives). The system makes sure the valve opening can only be executed by a person — not that everything has to go through a person, but that it does where it matters.

See the full IRIS architecture →

Before and after

Delivery note at the end of the shift vs. delivery note before the valve opens

Aspect	Classic receiving	With iLEAN Connect + Agents
Delivery note entry into the ERP	At the end of the shift, manual	At second zero, automatic
Cross-check with the day's formula	After the fact, in the office	Before unloading, on the dashboard
NCO% in-range verification	By the lab, afterwards	Delivery note read + alert if out of range
Stock + SHE capacity	Worked out in someone's head	Validated by an agent, with a SHE signature
Decision to open the valve	Operator alone in the bay	Operator + alert + SHE signature
REACH / Seveso dossier per delivery	Rebuilt by hand	Automatic, with a photo of the delivery note

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.

• PU sealant and adhesive plant with regular MDI/TDI deliveries by ISO tank, several active formulas, a Seveso or equivalent area under SHE authorization.
• Connect pilot (delivery note capture by email/WhatsApp/photo + integration with the ERP and the SHE dashboard + alerts ahead of unloading). First value expected within a few weeks.
• Expected reduction in SHE nonconformities related to deliveries ≥ 30% against the documented history of recent months.
• Indicative payback between 4 and 9 months, depending on delivery frequency, the average cost of a SHE incident and the receiving lead's current administrative load.
• The hard lever is a single SHE incident avoided: the regulatory, health and reputational cost of an isocyanate incident comfortably exceeds the cost of the system.

And the SHE lead's reasonable doubt

“What if the AI misreads a figure on the delivery note and authorizes an unloading it should not have?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely extracts data from a document and compares it against a master record (delivery note batch against planned batch, NCO% against the acceptable range), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN proposes authorization and the SHE lead signs before the valve is opened. The three safety rings exist precisely for this.

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