Zero exclusion-zone intrusion incidents

The problem

One person in the swing radius is the exposure the whole company is measured on.

Heavy-tonnage machinery risk is the sector's biggest reputational, human and financial exposure. Every serious incident costs, beyond the human toll, penalties, work stoppage and possible loss of future contracts — especially on public or high-profile projects.

• The exclusion zone around a heavy-tonnage machine or a demolition robot is defined on paper, marked out with cones and tape, and reconfigured by hand every time the work zone advances — which on a live demolition is daily.
• The crew rotates, subcontractors come and go for a morning, and the person who walks into the radius is very often the one who was not on site when that zone was briefed.
• Beyond the human toll, a serious incident means penalties, a stopped site and the loss of future contracts, which on public and high-profile projects is the part the company does not recover from.
• And a near-miss leaves no trace at all: it is told in the site office at the end of the day and never becomes a record anybody can act on before the next one.

How it fits the IRIS system

Agents orchestrating four rings — permit, voice, camera and evidence.

Ring 1 (Connect): the Work Permit defines which machine operates in which zone and with what exclusion radius. Ring 2 (Connect): the foreman alerts in seconds via earpiece if they see someone approaching. Ring 3 (Edge): cameras on every large-tonnage machine detect intrusion into the swing radius and trigger JIDOKA AI — automatic lockout. Ring 4 (Agents): every intrusion and every safety sign-off is documented in the evidence pack. SMED AI speeds up reconfiguring the exclusion zone as the work zone progresses.

No single ring solves it. Each of them already exists today in some form — a permit, a shout on the radio, a spotter watching the swing — and people still end up inside the radius. The strength is in the cross-check, and above all in the fact that the third ring does not ask a human being to see the intrusion and react in time.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · Permit	Work permit (1) + tablet validation (5)	Which machine works which zone, and with what radius
2 · Voice	Foreman earpiece (4) + supervisor alert (3)	The warning given in seconds, hands free
3 · Machine	Edge on every heavy machine	Intrusion into the swing radius, and automatic lockout
4 · Evidence	Evidence pack (11)	Every intrusion and every sign-off, documented
Reconfiguration	SMED AI	The exclusion zone follows the work zone
Who decides	The site manager	iLEAN proposes, the machine stops

Before: incidents or near-misses with no objective evidence. After: automatic detection and lockout in the moment, fully documented.

Impact estimate

Impact estimate — to be validated with 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 6-12 months, measured against the cost of a single serious incident avoided.
• Detection and automatic lockout in the moment, instead of a warning that depends on somebody seeing the intrusion and reacting in time.
• Near-misses stop being site-office stories told at the end of the day and become documented events with their own evidence.
• It is a case to size with your safety officer and general management, because the exposure it covers — human, financial and reputational — is theirs.

Estimated payback of 6-12 months against the cost of a single serious incident avoided. Estimate to validate with the safety officer and general management.

And the fair question from the production manager

"Do we have to deploy all twelve cases to get this?" — no, and framing it that way would be the way never to start. The rings are built in layers and each case pays on its own. On ring 3, detecting a person entering the swing radius is an anchored task, where the best models drop below 1.5% error [1], and JIDOKA AI locks out on a verified detection, not on model uncertainty.

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