A seized conveyor bearing at 250 °C is not a spare part — it is eight hours of oven downtime.

The problem

The bearing warns you with heat. Nobody is reading that heat.

A biscuit tunnel oven cannot be stopped halfway — and precisely for that reason, a cheap part can turn into a whole-shift incident. The conveyor drum bearings (head, tail, intermediate) work in a hostile environment: high ambient temperature, airborne flour, continuous vibration, long cycles. When they start to fail, they tell you in three thermal phases:

1. Lubrication degrading — the bearing rises a few degrees above its own history, with no alarm because it is still inside the oven's absolute range.
2. Asymmetric micro-friction — the bearing on side X rises relative to its counterpart on side Y. The gradient opens up.
3. Exponential heating — minutes before seizure. By the time the manufacturer's spot thermocouple (if there is one) notices, it is already too late.

The classic system — a maintenance round with a handheld pyrometer, the veteran's eye, a calendar based on running hours — only reads the snapshot, not the film. And the film runs for days, not minutes. When the oven goes down, the cost is not the part: it is the 6-8 hours of controlled cooling, the raw dough trapped in the tunnel and the batch in progress lost.

How it fits the IRIS system

iLEAN Edge does not replace maintenance — it gives it the permanent eyes it cannot have on its own.

The problem is not a lack of instrumentation: any technician with a pyrometer reads the bearing temperature in 5 seconds. The problem is that nobody can be reading the oven's 12 critical bearings 24/7, and the thermal signature lives on an island — the bearing housing — between two rounds. iLEAN acts as the filler that closes that gap, without asking you to change the oven or the line's SCADA.

Edge watches the critical bearings continuously, with a thermal camera and a local CNN. Connect cross-references the shutdown calendar and the history of replacements. The agent proposes intervening in the next window. The person signs — never the other way round.

The three iLEAN pieces applied to the tunnel oven conveyor bearings:

• Edge — a terminal with a fixed thermal camera focused on the critical bearings (head, tail, intermediate). A CNN trained to recognise the abnormal heating pattern against the hot oven background and the belt cycle. It works with no network: if the plant loses WiFi, Edge keeps seeing and keeps alerting the maintenance panel.
• Connect — captures the context the camera alone cannot resolve: the planned shutdown calendar, the last bearing replacement noted in the maintenance manager's spreadsheet, the special grease batch received from the supplier by email, the shift production report. All cross-referenced at second zero.
• Agent — cross-references the abnormal thermal signature, the shutdown calendar and the criticality of the bearing (head, tail, intermediate). It proposes an intervention at the next planned stop — before seizure, with no emergency. And it prepares the work order with the right part. The person validates and signs.

See the full IRIS architecture →

Before and after

Pyrometer round vs. continuous thermography with a CNN

Aspect	Classic oven maintenance	With iLEAN Edge thermal on the bearing
Temperature reading	Handheld pyrometer, daily round	Continuous thermal camera, 24/7
Early degradation detection	Only if the technician catches it on the round	The CNN recognises the pattern against history
Side-by-side comparison	Only if the technician does it explicitly	Automatic gradient between symmetrical bearings
Useful warning window	Hours, sometimes minutes	Days, in most cases
Operation with no network	n/a	Edge keeps running on the panel's power
Traceability for IFS/BRC	Manual round sheet	Thermal history per bearing, 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 from your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized biscuit plant with 1-2 tunnel ovens, continuous belt, multi-recipe, scheduled deep-cleaning stops.
• Edge pilot on the 4-6 most critical bearings of the main oven (head, tail, intermediate drive bearings). First expected value within a few weeks: baseline thermal history recorded, first useful alert on lubrication degradation.
• Expected reduction of unplanned stoppages caused by conveyor bearings ≥ 30% against the baseline of the last few years.
• Indicative payback between 4 and 9 months, depending on the documented historical frequency of incidents and the average cost of oven downtime (cooling hours + dough lost + batch in progress).
• The hard lever is a single seizure avoided: 6-8 hours of oven saved, batch in progress recovered, a non-urgent stop. One single scheduled intervention instead of an emergency one pays for the pilot.

And the maintenance manager's reasonable doubt

«What if the CNN gets it wrong and orders the oven stopped over a thermal pattern that was just the normal belt cycle?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI limits itself to recognising a pattern on the camera's real image, the best models brought the error below 1.5% ^[1]. And even so, the critical call is never made alone: Edge alerts in time, maintenance signs the intervention in the next window. 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.