ATEX PPE in a refinery zone — the spark that must never come off the clothing.

The problem

Four pieces that must all be there, every single time — and HSE cannot look at thirty people at once.

Access to a refinery's Ex zone depends on every operator wearing, simultaneously:

1. Antistatic footwear approved for the ATEX zone, not generic safety footwear.
2. A dissipative coverall or workwear (EN 1149) — compatible with the substance present in the zone.
3. Anti-spark gloves compatible with the operation to be performed.
4. Helmet and complementary protection (face, respiratory) when the operation demands it.

The airlock guard knows this, but at 06:50 thirty people cross in a hurry to clock in at 7:00, and the human eye cannot keep up. The operator is not doing anything strange either: they changed in the locker room, and sometimes do not realise that the coverall they put on was the other zone's, or that the right glove was not approved. The classic system works 99% of the time. That 1% is the accident that ends up in the press — and by itself it costs several years of reinforced operation of the unit.

How it fits into the IRIS system

iLEAN does not replace HSE — it puts eyes where HSE physically cannot be at once.

The problem with PPE verification in ATEX is not a lack of rules: it is information living in islands that, at the critical moment (the seconds before the access turnstile), is never cross-checked against reality. iLEAN acts as the filler that closes that gap, without asking you to change the access control, the HSE ERP or the locker rooms. See the category: full IRIS architecture.

Edge sees the operator before the turnstile. Connect reads the zone, the authorised operation and the approved PPE catalogue. The agent cross-checks and, if something is missing, keeps the turnstile closed. The HSE person signs off — never the other way round.

The three iLEAN pieces applied to ATEX PPE control in a refinery:

• iLEAN Vision (Edge) — a terminal with one or two cameras in the turnstile's airlock. The CNN is trained on the plant's specific approved PPE catalogue (models, labelling, reflective bands) and verifies every piece in milliseconds. It looks at what it has to see, not at the people — there is no facial recognition. It works with no network. If the line goes down, Edge keeps verifying against the last synchronised set.
• Connect — captures the ATEX zone being accessed, the authorised operation from the work order (which comes from the HSE ERP or from the work permit the supervisor sends by email/WhatsApp) and the prevention manager's approved PPE catalogue. It digitises it at second zero.
• Agent — cross-checks the zone, the operation, the required PPE catalogue and what Edge has seen. If a piece is missing, it shows the operator on screen what they lack and alerts HSE through whichever channel they use. The turnstile never opens on its own. The person validates and signs off.

See the full IRIS architecture →

Before and after

Manual ATEX PPE control vs. verification with iLEAN Vision

Aspect	Guard + checklist	With iLEAN Vision + Connect + Agent
Verification at peak time	30 operators in 10 minutes, the guard's eye	Every operator verified individually
Approved vs. generic coverall	By sight, with normal human error	CNN trained on the real catalogue
PPE compatibility with the operation	The operator's assumption	Cross-check with the day's work permit
Detection of incomplete PPE	Reactive (sometimes never)	Before the turnstile, in milliseconds
Traceability for the audit	A sheet signed by the guard	Dossier per entry with photo and sign-off
Operation with no network	n/a	Edge keeps verifying on cabinet power

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 figures of your plant. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Refinery / petrochemical plant with 2-4 airlocks into ATEX zones and several hundred personnel movements per day.
• Edge pilot at one airlock (cameras + integration with access control + work order). First expected value within a few weeks: a verification rate genuinely useful for HSE.
• Indicative payback between 4 and 9 months, depending on the number of airlocks equipped, the frequency of accesses and the cost of audits / shutdowns from previous incidents.
• Expected reduction in accesses with incomplete PPE of ≥ 30% versus the previous manual control, with a floor that is reasonably defensible before the committee.
• The hard lever is the avoided cost of an ATEX incident: unit shutdown, investigation, personal injury, insurance premium, reputational impact.

And the IT manager's reasonable doubt

"What if the AI confuses an approved coverall with one that is not?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI simply classifies an image against a defined catalogue (this coverall is / is not the approved model), the best models brought the error below 1.5% ^[1]. And even so, what is critical is never decided alone: the turnstile stays closed and the HSE person signs off. The three safety rings are there for exactly this.

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