UV varnish defect detection on lacquered doors — a crater spotted after the UV tunnel is another full booth cycle.

The problem

UV varnish is irreversible — before the tunnel everything is fixable; after it, everything is rework.

UV lacquering on a premium door is the finish where a single speck of dust, a silicone crater or the smallest sag is not an incident: it is a rejection from a meter away. The line usually runs: preparation → spray or curtain coating → flash-off table → UV tunnel → inspection. On the table between coating and the tunnel, the varnish is still wet: a blast of air removes the speck of dust, a scraper corrects the sag, a touch-up compensates for low film build. Any defect is recoverable.

The problem starts when the door enters the UV tunnel. The lamps cure the varnish in seconds and the film ends up like glass. After the tunnel, recovering a crater or a sag means full sanding, putty where needed, repainting and another complete booth cycle. The compound cost:

1. Booth labor — the operator who sands, putties and repaints an already cured door.
2. UV lamp life and the consumables of the second cycle.
3. The risk of going too far with the sanding: if you break through the base coat, the whole door is scrap — and on a premium lacquered door with fine veneer underneath, that scrap is expensive.
4. Delay to the customer — booth capacity is limited and reworks eat whole shifts.

Classic inspection at the end of the line (an operator with a lamp after the tunnel) works 99% of the time, but when it does catch a defect it is already too late to avoid the rework. The useful inspection is precisely the one done before the tunnel.

How it fits the IRIS system

iLEAN Edge is not one more final inspection — it is the eye right before the point of no return.

The problem is not a lack of inspection: your quality team knows perfectly well what counts as a rejectable crater and what is a normal pattern of the lacquer. The problem is that the useful inspection has to happen before the UV tunnel, at conveyor speed, in a booth full of varnish fumes, and no person can hold that for eight hours. iLEAN acts as the putty between the booth lead's criteria and the reality of each door, without asking you to change the booth or the UV cycle.

Edge sees every door before the tunnel, with your team's criteria. Connect reads the batch destination and the color. The agent alerts the operator, logs the defect and proposes the pre-UV rework. The person signs — never the other way round.

The iLEAN pieces applied to UV varnish defect detection on lacquered doors:

• iLEAN Edge — a terminal with machine vision (CNN) and an industrial camera with combined lighting (grazing light to pick up craters, sags and dust + reflected light to pick up low film build and gloss variation). Positioned between the flash-off table and the entrance to the UV tunnel. It alerts the operator in milliseconds, without stopping the conveyor: the operator steps in, blows off the dust, sands if needed, and the door carries on. It works without a network: as long as the terminal has power, the CNN keeps classifying.
• Connect — it reads the batch destination (premium high-gloss white vs. standard matte lacquer, retailer A vs. retailer B) and the booth's drift history (aging UV lamp, new varnish batch, booth temperature). It also captures what comes from outside: the kitchen distributor's specification, the demands of a bespoke furniture manufacturer.
• Agent — it cross-references the image of each door, the batch destination and the booth history. If the booth builds up an abnormal pattern (more craters than usual, gloss drift), it does not wait until the end of the shift: it alerts the booth lead on whatever channel they use, with a proposed check (paint filters, lamp, temperature). The operator validates the proposal and signs.

See the full IRIS architecture →

Before and after

Final inspection after the UV tunnel vs. in-line detection with iLEAN Edge

Aspect	Manual final inspection	With iLEAN Edge + Connect + Agent
Moment of inspection	After the UV tunnel, varnish cured	Before the UV tunnel, varnish still wet
Action possible on a defect	Sanding, putty, repainting, second UV pass	Air blast, scraper, touch-up — no second UV pass
Coverage	Whatever the operator happens to see	100% of the doors, both faces
Relief defects (crater, sag)	Only those visible from a meter away	CNN over dedicated grazing light
Film build and gloss defects	Subjective	CNN over reflected light
Operation without a network	n/a	Edge keeps alerting on the power in the cabinet
Traceability per door	Batch level, no detail	Before/after photo + classification + reason

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.

• Premium lacquered door manufacturer with one UV lacquering booth, a mix of finishes (high-gloss white, matte lacquer, color), rework rate documented over recent years.
• Edge pilot between the flash-off table and the entrance to the UV tunnel (camera with dedicated lighting + Edge terminal with CNN + alert to the operator + integration with the booth SCADA). First value expected within a few weeks, with the first doors detected and recovered before the UV tunnel.
• Indicative payback between 4 and 9 months, depending on volume, the historical frequency of booth reworks and the average cost of a full repainting cycle.
• Expected reduction in post-UV reworks ≥ 30% over the baseline. The hard lever is every booth cycle saved + every door not delivered defective to the end customer.

And the booth lead's reasonable doubt

"What if the AI flags a normal variation of the lacquer as a crater and we stop the line for nothing?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares an image against closed criteria learned from your quality team, the best models brought error below 1.5% ^[1]. And even so, the line never stops on its own: the system alerts the operator, the operator decides whether to step in or carry on, and 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.