Sausage drying room control with AI — a piece that drifts is not recovered, it is anticipated.

The problem

The master reads the chamber in his head, the datalogger records to a memory card, the ERP never finds out.

In a sausage drying room, what decides whether the batch comes out right is accumulated intuition: the master curer walks in in the morning, checks the thermo-hygrometer, touches the casing of three pieces, remembers what happened on Wednesday night, and lowers the humidity half a point or raises the temperature one degree. That knowledge is the most valuable asset in the plant — and the worst protected.

The classic system has three built-in gaps:

1. The datalogger records, but does not warn. The usual Tewes, OMNIA or CFS units store the curve, but nobody looks at the curve until there is already a complaint. Slow drift never fires as an alarm.
2. Pieces are not weighed piece by piece. Weight loss (the ultimate truth of curing) is usually measured by sampling, and by the time anyone detects that a whole room is running off target, the room has been running that way for weeks.
3. The master's judgment is not inherited. That "if this goes up, I bring that down" lives in his head. The day he retires, the plant enters a learning cycle paid for in lost batches.

The result: when white spot, case hardening or out-of-range shrink appears, the problem is not today's — it started three days ago and nobody connected the dots in time. The classic system works 99% of the time. That 1% is where the year's margin goes.

How it fits the IRIS system

iLEAN does not touch the chamber — it seals the cracks between the datalogger, the panel and the master's head.

The drying room's problem is not a lack of sensors: it is information living in islands that, at the critical moment — Wednesday night, the recipe change from salchichón to fuet, an incoming batch of meat with more fat — does not reach the decision-maker in time. iLEAN acts as the putty that fills those gaps without asking you to change the chamber, the datalogger or the ERP.

Connect reads what the datalogger records and what the panel says. The agent cross-checks the curve against the piece's weight history. If the accumulated drift puts the batch at risk, it warns the master. The person signs the adjustment — never the other way around.

The two iLEAN pieces applied to a sausage drying room:

• Connect — captures the data whether it comes from a modern datalogger with an output, from an old, isolated local computer, or from a photo of the analog panel taken by a person with a phone. It also captures what comes in from outside: the casing supplier's email with a change in caliber, the veterinarian's WhatsApp with a note about the incoming batch. Everything at second zero, no forwarding.
• Agent — cross-checks the datalogger's humidity/temperature curve, the piece's weight history, the batch recipe and the product's master curve. If the accumulated drift puts the batch at risk (case hardening, white spot, shrink off target), it warns the master or the shift lead on whatever channel they use, with the proposed adjustment. The person validates and signs; the chamber does not reprogram itself.

Edge comes in here later — vision over the pieces at the chamber exit to validate casing uniformity — but it is not the first step. The first value comes from the data you already have and nobody reads.

See the full IRIS architecture →

Before and after

A drying room run on craft vs. a drying room with the curves cross-checked

Aspect	Today in most drying rooms	With iLEAN Connect + Agent
Datalogger reading	On demand, once there is already a complaint	Continuous, cross-checked with weight and recipe
Slow drift detection	When the defect shows up on the piece	Early warning, with days of margin
Recipe or caliber change	The master recalibrates from memory	Agent suggests an adjustment based on history
White spot / case hardening	Discovered when the piece comes out	Anticipated from humidity/weight deviation
Master curer's handover	Learning curve paid in lost batches	The veteran's judgment captured as data
Traceability for the IFS/BRC auditor	Rebuilt by hand, weeks	Per-batch dossier with the full curve

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with your plant's data. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized plant making cured ham or long cured sausages, 4-8 drying chambers, a heterogeneous datalogger fleet (a mix of modern and old equipment), a master curer with 20+ years of craft.
• Connect + Agents pilot in 2 representative chambers (the oldest and the newest), without touching the chambers' programming. First value expected within a few weeks: early drift warnings on batches in progress.
• Indicative payback between 4 and 9 months, depending on volume, unit value of the piece and current level of waste + grade downgrades. A reduction in waste and B-grade pieces on the order of 30% or more is defensible as a floor.
• The hard lever is not just waste: it is preserving the master's judgment when the handover comes, without the plant losing a year re-learning it.

And the technical director's reasonable doubt

"What if the AI recommends badly and the batch is lost by following it?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely cross-checks a measured value against a historical pattern (datalogger curve vs. master curve), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN proposes, the master or the shift lead signs. The three safety rings exist precisely for this.

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