IQF prawn — the ice crystal decides the quality, not the packaging.

The problem

The damage is not visible in the tunnel — it shows up when the importer's kitchen thaws it.

Premium prawn sells on its texture and on the integrity of the peeled meat after thawing. And that is decided entirely in the IQF tunnel, where three things almost never line up at once:

1. The size grade changes with every batch — and the tunnel setpoint is usually fixed to the typical grade. Anything thicker crosses the large-crystal zone too slowly; anything thinner gets over-frozen and loses moisture.
2. Tunnel load varies with the pace of the plant — higher product density on the belt changes heat transfer; the tunnel performs worse exactly when it needs to perform best.
3. Humidity and inlet temperature move with the day — and the operator cannot recalculate the freezing curve every hour of the shift.

Large ice crystals cannot be seen. They show up as broken peeled meat, drip loss in the tray and a complaint from the importer. The classic system (a PLC with a fixed setpoint + the operator's eye) gets it right 95 % of the time. That 5 % is the batch that arrives as second grade — and in margin terms, that 5 % eats the year's premium.

How it fits the IRIS system

iLEAN does not replace the tunnel — it seals the cracks between the PLC, the probes, the ERP and the shift lead's eye.

The IQF problem is not a lack of information: it is information that lives in islands and that at the critical moment (a size-grade changeover, a load peak) does not reach the decision-maker in time. iLEAN acts as the putty that fills those gaps without asking you to change the tunnel or the PLC.

Edge reads the IQF line in real time. Connect captures size grade, humidity and inlet temperature wherever they live. The agent cross-references them with the small-crystal pattern and proposes a setpoint. The person signs — the tunnel never changes its own recipe.

The iLEAN pieces applied to prawn IQF control:

• iLEAN Edge — a terminal with a CNN and probes in the tunnel's control cabinet. It reads the temperature of every zone, the belt load, relative humidity and the product's outlet temperature. In milliseconds it works out whether the curve is drifting out of the small-crystal range and proposes a speed or thermal setpoint adjustment to the PLC. It works with no network: if the plant loses WiFi, Edge keeps calculating and delivering setpoints; what is critical cannot depend on connectivity.
• iLEAN Connect — captures the batch size grade whether it comes from the ERP, the MES, a quality manager's spreadsheet or an old counting scale, and it also captures what arrives from outside (the importer's specification with the allowable drip-loss range, a supplier alert about a size grade out of tolerance). It reaches the system at second zero, without anyone having to forward anything.
• Agent — cross-references the size grade, the tunnel's real curve, the line's history and the importer's allowable range. If the curve drifts, it does not send an email at 10 p.m.: it proposes a setpoint and alerts the shift lead on whichever channel they use. The person validates and the setpoint goes into the PLC.

See the full IRIS architecture →

Before and after

An IQF tunnel with a fixed setpoint vs. an IQF tunnel with an adaptive setpoint from iLEAN

Aspect	Fixed setpoint + the operator's eye	With iLEAN Edge + Connect + Agent
Adapting to the batch size grade	The operator readjusts by eye if something looks off	Setpoint recalculated for the actual incoming grade
Reaction to a load peak	Noticed in the importer's next complaint	Adjusted on the line, before the crystal grows
Curve in the critical 0 ºC / −5 ºC zone	Assumed by recipe	Measured in every zone and compared with the optimal pattern
Detecting drip loss at the outlet	Visual, at packing	Before packing, on the line
Operation with no network	n/a	Edge keeps running on the cabinet's own power
Per-batch file for the importer	Rebuilt by hand if there is a claim	Per-batch dossier with the tunnel's real curve, automatic

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 diagnostic.

• Plant with one or two prawn IQF tunnels, mixed size grades, exporting to a European or American importer with a strict drip-loss tolerance.
• Edge pilot on one tunnel (probes in the zones + a camera over the outlet belt + integration with the PLC and with the batch recipe from the ERP). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on how often batches get downgraded to second grade and on the average cost of an importer penalty.
• Reduction in second-grade batches ≥ 30 % as a defensible floor; the hard lever is the premium/second-grade price gap per kilo and the export share recovered.

And the operations director's reasonable doubt

"What if the AI proposes a setpoint that wrecks the tunnel?" — the Agents have no hands in critical OT: they propose, the person signs, the PLC executes. And hallucination is a problem of free generation, not of anchored tasks: in tasks where the AI merely recontextualizes measured signals (real curve → allowable range), the best models brought error below 1.5 % ^[1]. The three safety rings exist precisely for this.

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