Fish cold room temperature control with AI — the range is not a chart, it is a dossier per batch.

The problem

The recorder draws a pretty line. The fish does not.

The room's panel reads 1°C and the recorder draws a horizontal line with no jumps. The shift lead sleeps soundly. And yet four things happen on the floor that the chart never captures:

1. The door opens all day long. Fifteen, twenty times per shift: dock unloading, order picking, a warehouse hand going in for a crate. Each opening lets in warm air that the wall probe takes a while to notice.
2. Warm load arrives. A pallet from the dock that did not spend the time it should have in the refrigerated truck. The load pushes up the real curve of its surroundings even if the average ambient does not move.
3. The evaporator frosts up. Before the room warms, the evaporator loses performance. The ambient temperature holds for a few minutes on inertia, and by the time it rises, product is already out the door.
4. The probe sits in the convenient spot. Next to the office door, not at the back where the oldest stock accumulates. The probe measures what it sees, not what happens.

The health inspector knows it, and when auditing asks for more than the curve: opening frequency, average duration, defrost cycles, the correlation with the batches that were in the room at every moment. Rebuilding that by hand takes a week — and by then the fish is already sold.

How it fits the IRIS system

iLEAN does not buy you a new room — it puts eyes where there were none and joins what lives in islands.

The room works. The probe works. The panel works. What is missing is for the four realities (probe, door, load, batch) to be seen together, at second zero, on the shift lead's phone. iLEAN is the putty that fills the joints between the refrigeration PLC, the wall recorder, the warehouse hand's notebook and the warehouse ERP.

Edge sees the door and the load. Connect captures probes, recorder and batch. The agent anticipates the deviation, gets the right technician on it first, and closes the per-batch HACCP dossier. The person validates — the room warns, it does not go silent.

The three iLEAN pieces applied to fish cold room temperature control:

• Edge — a machine-vision terminal (CNN) over the room's door. It counts actual openings, measures opening time, identifies the pallet coming in and which batch it belongs to. It works without a network: if the warehouse loses WiFi, it keeps recording on the panel's power.
• Connect — captures the distributed probes (existing or supplementary), reads the analog recorder's curve from a photo if there is no integration, and pulls the batch record from the ERP or from the dock delivery note. Capture gradient: if the room has a modern PLC, direct integration; if it only has a ticket recorder, Edge digitizes it.
• Agent — cross-checks distributed temperature + openings + load + the room's history, anticipates a frosting evaporator or a short defrost, and alerts the refrigeration technician before the ambient curve reflects it. It closes the per-batch HACCP file without anyone gathering spreadsheets on audit day. The person decides — the AI does not open the room, does not stop the compressor unit, does not sign alone.

See the full IRIS architecture →

Before and after

A room with a recorder vs. a room with an iLEAN nervous system

Aspect	Classic recorder	With iLEAN Edge + Connect + Agent
Temperature reading	One wall probe, one curve	Distributed probes + correlation with openings and load
Evaporator frost detection	Once the ambient temperature has already risen	Anticipated from the performance drop
Record of door openings	"Whoever was there remembers"	Count and duration by vision, automatic
Per-batch traceability	Rebuilt by hand for the inspector	Per-batch dossier with the curve actually lived
Alert to the refrigeration technician	When someone notices	Before the thermometer reflects it
HACCP file	Paper folder, weeks of preparation	Ready at any time, signed by the person

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 on-site diagnostic.

• Fresh fish wholesaler or processor, 2-5 fresh-fish rooms, analog recorder or a mixed PLC setup.
• Pilot in one critical room: Edge over the door + 2-3 distributed probes + integration with the batch record in the ERP. First value expected within a few weeks — the first early warning for frost or excess openings.
• Indicative payback between 4 and 9 months, depending on the historical frequency of incidents and the documented cost of cold chain breaks (destroyed product, complaints, health alerts).
• Measurable hard lever: ≥30% reduction in quality hours spent on HACCP reconstruction, plus early warnings on refrigeration performance before the incident.

And the warehouse manager's reasonable doubt

"What if the AI raises a false alarm and makes us start the backup generator for nothing?" — the AI here does not start the generator. It alerts the person, proposes a probable cause, and lets the manager decide. The reliability of AI on anchored tasks (reading a probe and comparing it against a history) brought error below 1.5% in the best models^[1]. Even so, no critical decision is executed without your signature. The three safety rings exist precisely for this.

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

Related: dairy cold chain · frozen, pallet by pallet · cold room inventory