OCR of heat numbers on hot aluminum ingots with AI — no reading, no traceability; and no traceability, no serious customer.

The problem

The number is stamped. The hard part is reading it when traceability depends on it.

In an aluminum foundry hall, every heat is identified with a number embossed on the ingot, while hot. That number is the link between the raw material, the chemical composition measured at the heat and the final part that will reach the automotive or aerospace customer. Three realities make its classic reading fail:

1. The heat. The ingot leaves the furnace above 400-500 °C and emits radiation that saturates standard OCR cameras. The stamp, clear when cold, blurs when hot.
2. The environment. Smoke, oxide dust, steam from the cooling systems and the changing light of the overhead cranes make a "studio" shot impossible.
3. The manual record. When the OCR cannot cope, the operator writes it down by hand. One badly copied digit breaks the finished part's traceability weeks later — and there is no way to recover it.

The quality manager knows this, but cannot be in every hall verifying every ingot. The manual record works 99% of the time. That 1% is what surfaces when the customer asks for the material certificate — and the heat shows up as "not located".

How it fits into the IRIS system

iLEAN Vision does not replace the heat leader — it seals the crack between what is stamped hot and what the ERP stores cold.

The heat-OCR problem is not a lack of information: it is information living in islands (the ERP/MES heat plan, the optical reading on the part, the lab's chemical analysis, the shift leader's sheet). At the critical moment — the hot ingot dropping into the cooling area — those data points are never cross-checked in time. iLEAN acts as the filler that closes those gaps without asking you to change the furnace or the mould.

Edge sees the hot embossed number. Connect reads the heat plan and the chemical analysis. The agent cross-checks and, if the reading does not fit, flags the ingot. The person signs off — never the other way round.

The iLEAN pieces applied to heat-number OCR in aluminum:

• iLEAN Vision (Edge) — a terminal hardened for hot environments with an industrial camera, suitable IR filters and structured lighting. A CNN trained on real ingots in plant conditions, not on cold studio parts. It reads, decodes and flags in milliseconds. It works with no network. If the hall loses its WiFi, Edge keeps reading and recording locally.
• Connect — captures the ERP/MES heat plan, the lab's chemical analysis (the analyser's spreadsheet, the chromatograph's sheet) and the mould changes the shift leader decides. The filler works in the foundry too: what sat in a forgotten lab spreadsheet enters the system at second zero.
• Agent — cross-checks the optical reading with the heat plan, validates it if it fits within confidence and, if not, holds that ingot's traceability as "pending re-read". When a customer asks for a material certificate, the per-ingot dossier comes out automatically — a boring, concrete agent, exactly the kind that closes sales.

See the full IRIS architecture →

Before and after

Classic OCR + manual note vs. hot reading with iLEAN Vision

Aspect	Classic OCR + the shift leader's note	With iLEAN Vision + Connect + Agent
Hot reading (400+ °C)	Frequent failure, discard by default	Foundry-trained CNN reads while hot
Number validation	Blind optical reading	Cross-check with the ERP's heat plan
Parallel manual record	Yes, accumulated human error	Automatic signed record per ingot
Material certificate for the customer	Reconstructed by hand per lot	Dossier per ingot, automatic
Operation with no network	n/a	Edge keeps operating on cabinet power
IATF 16949 / AS9100 traceability	Vulnerable to a badly copied digit	Anchored to the optical reading + chemical analysis

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.

• Secondary aluminum foundry with embossed numbered ingots, automotive/aerospace customers demanding heat → part → final product traceability.
• iLEAN Vision pilot at a furnace exit area (hardened camera + lighting, integration with the ERP's heat plan and the lab's chemical analysis). First expected value within a few weeks.
• Indicative payback between 4 and 9 months, depending on the annual cost of heats not located during material certification and the cost of the manual cold re-read of ingots.
• A reduction of heats with incomplete traceability of ≥ 30% over the baseline — the hard lever is a material certificate delivered on time to a key customer. One order not lost over traceability pays for the pilot.

And the quality manager's reasonable doubt

"What if the AI confuses an 8 with a 0 while hot?" — hallucination is a problem of free generation, not of anchored tasks. When the AI simply reads a number whose expected range it already knows (the ERP's planned heat), the best models brought the error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN flags the ingot if confidence does not reach the threshold and the 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.