If the extractor bearing goes, the booth loses negative pressure — and the oven goes down with it.

The problem

The extractor bearing is the cheap part that drags down the most expensive asset on the line.

On a PVD line, the paint booth sets the pace: if it loses negative pressure, particles escape into the room, the parts that follow no longer come out in condition, and the downstream curing oven goes into a controlled stop (or its cycle breaks, which is worse). And that negative pressure depends on one small part — the extractor motor — which in turn depends on two bearings. The road to the stop works like this:

1. Lubrication degradation — the bearing climbs a few degrees above its own history. Nothing trips, because it stays below the thermal protection relay.
2. Thermal asymmetry — the motor-side bearing rises relative to the fan-side one (or the other way round). The gradient opens up.
3. Aerodynamic overload — if the filters are saturated on top of that, the motor compensates with more current, the bearings climb faster, and degradation accelerates.
4. Thermal trip or seizure — the negative pressure drops, the booth falls out of class, the oven has to stop.

The classic setup — a handheld pyrometer on the round, a thermal protection relay, the maintenance manager's eye — only reads the snapshot, not the film. And the film runs for days. When the extractor goes down, it is not one motor that fails: half the line goes with it, and the oven right behind.

How it fits into the IRIS system

iLEAN Edge does not replace the maintenance technician — it gives him the permanent eyes he cannot have on his own.

The problem is not a lack of instrumentation: there is a thermal protection relay and there is a round. The problem is that nobody can be watching the extractor bearings 24/7, and the thermal signature lives on an island — the motor housing — between two rounds. iLEAN acts as the filler that closes that gap, without asking you to change the booth, the oven or the line SCADA.

Edge watches the bearings continuously, with a thermal camera and a local CNN. Connect cross-references the filter cycle, the paint recipe and the shutdown calendar. The agent proposes intervening in the next window. The person signs off — never the other way round.

The three iLEAN pieces applied to the PVD booth extractor bearing:

• Edge — a terminal with a fixed thermal camera aimed at the body of the extractor motor (motor side + fan side). A CNN trained to separate the "filter saturation" pattern from the "incipient bearing failure" pattern. It works without a network: if the plant loses WiFi, Edge keeps seeing and keeps firing the alert to the control panel.
• Connect — captures the context: the paint recipe running (more solids = more aerodynamic load), the hours since the last filter change noted in the booth supervisor's spreadsheet, the shutdown calendar of the downstream oven, the shift production report. All cross-referenced at second zero.
• Agent — cross-references the anomalous thermal signature with the shutdown calendar and the criticality of the recipe running. It proposes an intervention in the next window — before the thermal trip, without urgency. And it prepares the work order with the right part and the right procedure. The person validates and signs off.

See the full IRIS architecture →

Before and after

Pyrometer round vs. continuous thermography with a CNN

Aspect	Classic extractor maintenance	With iLEAN Edge thermal on the motor
Temperature reading	Handheld pyrometer + thermal protection relay	Continuous thermal camera, 24/7
Filter vs. bearing separation	Manual inference by the technician	CNN trained to tell the patterns apart
Early detection	Only if the round catches it	Pattern recognised against history, days ahead
Alert to the downstream oven	Reactive, after the thermal trip	Proactive, scheduling the window
Operation without a network	n/a	Edge keeps running on the panel supply
Traceability for an IATF/ISO auditor	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 actual data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• A PVD paint plant or a liquid paint plant in automotive/appliances with 1-2 booths, an extractor motor on each one, and a curing oven downstream.
• Edge pilot on the extractor of the most critical booth (thermal camera + trained CNN + dry contact to the panel). First expected value in a few weeks: baseline thermal history, first useful degradation alert.
• Expected reduction of unplanned stops caused by the extractor of ≥ 30% against the baseline of the last few years.
• Indicative payback between 4 and 9 months, depending on the documented historical frequency of stops and the average cost of a line stop (oven + booth + backlogged hangers).
• The hard lever is a single unplanned stop avoided: a downstream oven that does not go down, a production window respected, a bearing change scheduled in the next shutdown. One single non-urgent intervention pays for the pilot.

And the maintenance manager's fair objection

"What if the CNN confuses filter saturation with a bearing failure and orders a healthy extractor to be stopped?" — hallucination is a problem of free generation, not of anchored tasks. On tasks where the AI is limited to recognising a pattern on the real image from the camera, the best models brought the error below 1.5% ^[1]. And even so, nothing critical is decided alone: Edge alerts in time, maintenance signs off 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.