Defects on melamine board — the scratch you see after the cut is a good part thrown away.
A melamine board with a scratch, a pore or a stain enters the beam saw and comes out as eight finished parts — all cut, all edge-banded, all headed for the bin. iLEAN Vision sees the defect before the cut and diverts the board to the review belt. The good part stays; the scrap is stopped at source. The person signs off — not the line.
The defect that travels from the warehouse to the cut is a multiple of scrap.
The melamine board enters the plant from the supplier, passes through the automated warehouse, is moved with suction cups and ends up on the beam saw's infeed belt. And between one step and the next:
- Nobody records the suction-cup scratch. One handling too many, a rub against a stacked board, and the face side loses quality. The operator who catches it against the light sometimes says something, sometimes not — and the board enters the beam saw with its scratch intact.
- The pore from the supplier's press arrives dissolved in the lot. A pressure change at the board maker's press produces a variation you can see on the face — but not on every sheet of the pallet, only on some. Incoming inspection is done by sampling: it passes.
- The stain shows up after the cut. A board with a glue stain in the middle drops into the beam saw and comes out as eight parts: two have the stain visible and are scrap; the other six have the stain covered or outside, but the operator throws them all away because "we don't trust the lot".
The quality manager knows this, but cannot inspect 100% of the boards before the cut. The classic system (operator at the belt + sampling + the veteran's eye) works 90% of the time. That 10% is finished-part scrap, which weighs several times the cost of the original board, plus the operator hours, plus the service damage to the customer's order.
iLEAN does not replace the veteran's eye — it replicates it, and keeps it available 24/7.
The problem with the melamine defect is not a lack of judgement: the judgement exists but lives in the veteran's head, who is at their station a few hours a day and retires in a few years. If nobody captures that judgement as data, it leaves with them. iLEAN is the filler that captures that veteran's eye and puts it on the line permanently, without replacing whoever had it: it amplifies them.
Vision sees the board before the cut. Connect cross-checks with the lot and the supplier. The agent investigates root cause and proposes the claim. The person signs off — never the other way round.
The three iLEAN pieces applied to melamine board defects:
- iLEAN Vision (Edge) — a terminal with machine vision (CNN) over the beam saw's infeed belt, with low-angle lighting to reveal scratches and pores. It processes the image locally, detects the defect, measures its surface and position on the board, and diverts to the review belt (light + actuator). It works with no network: if the plant loses its WiFi, Edge keeps seeing and diverting.
- iLEAN Connect — reads the board's lot from the ERP, the supplier's delivery note, and the automated warehouse data (did this board come in yesterday or has it been on the pallet three weeks?). And it captures what comes from outside: the supplier's email confirming a recipe change, the operator's WhatsApp about a break in this morning's pallet. Everything comes in at second zero.
- iLEAN Agents — cross-checks the defect photo with the lot, the supplier, the warehouse history and the beam saw's rhythm, and proposes a root cause. "70% of today's detected defects come from supplier A's lot 2025-XYZ — we recommend filing a claim and pausing that lot's entry into production." It never claims on its own. It proposes, the plant manager signs off.
Sampling inspection vs. 100% inspection with iLEAN Vision
| Aspect | Sampling + the veteran's eye | With iLEAN Vision + Connect + Agent |
|---|---|---|
| Inspection coverage | One sample per lot | 100% of boards before the cut |
| Defect detection | After the cut, on the finished part | Before the cut, on the whole board |
| Capturing the veteran's eye | Only when they are on shift | Replicated in the model, available 24/7 |
| Root cause | Discussion with no photo, weeks later | Photo cross-checked with lot and supplier, continuously |
| Claim against the supplier | With a returned pallet and a fight | With a photographic dossier and data |
| Operation with no network | n/a | Edge keeps diverting on cabinet power |
Impact estimate for your plant — to be validated with your numbers.
The block below is an estimate to be validated with the specific figures of your plant. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.
- Modular melamine furniture plant, beam saw with automatic feeding from the warehouse, edge-bander downstream. A documented history of scrap from melamine defects in the last financial year.
- Edge pilot over the beam saw's infeed belt, with camera + low-angle lighting + diversion actuator. First expected value within a few weeks: defective boards detected before the cut, first root-cause analysis by lot and supplier.
- A reasonable reduction in scrap from visual board defects of ≥ 30% in the first quarter — and a reduction in finished-part rework, which weighs more than the board itself in operator hours.
- Indicative payback between 4 and 9 months, depending on the current cost of scrap and end-customer returns. If your claims against the board supplier improve with the photographic dossier, the payback shortens.
And the operations director's reasonable doubt
"What if the model confuses the decor's texture with a defect and diverts good boards?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI classifies an image against a reference trained on your own boards, the best models brought the error below 1.5% [1]. And even so, the diverted boards go to the review belt: a person signs the discard. 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.
What people ask about melamine board defects
What typical defects does a melamine board carry before the cut?
The usual ones are: surface scratches from rubbing during storage or internal transport, pores and bubbles in the melamine layer from uneven pressure at the press, glue or press-wax stains, tone variation between boards of the same lot, and chipped edges from bad handling with the suction cups. All of them are discovered late: when the board has already been cut into finished parts, already edge-banded, already packed, and the good part has been thrown away together with the bad one.
How does iLEAN tell a surface scratch from a printed line in the board's decor?
By training the model with real images from your own warehouse — the convolutional network learns the difference between the texture of your finish (wood print, matte white, decorative) and the superimposed defect (scratch, stain, pore). Training is done with your boards, not with a generic model, because every plant has its own catalogue of finishes and every defect has its own pattern. This is what sets iLEAN apart from an off-the-shelf solution: no single model is expected to fit everything; it is trained on your reality.
Where is the camera placed on the line?
At the point of maximum usefulness before the cut: at the exit of the automated warehouse, over the beam saw's infeed belt, or over the board transfer towards the edge-bander. iLEAN Vision needs to see the whole board under controlled lighting (typically low-angle LED light to reveal scratches and pores, and direct light for stains). The Edge terminal processes the image locally and flags the defective board for manual or automatic diversion depending on your configuration.
What if the defect only shows under certain light?
That is why lighting is part of the pilot. One of the embedded engineer's first tasks is to study the angle and intensity of light that reveals the defect that costs money — because every plant has a defect that shows in one light and hides in another. What the veteran's eye catches against the light, iLEAN Vision learns with low-angle LED lighting and detects continuously, including when that veteran is on their day off.
How long until it delivers first value in a furniture plant?
An Edge pilot over the beam saw's infeed belt can deliver first value within a few weeks: defects detected before the cut, first boards diverted, first root-cause analysis (is it the supplier?, the internal handling?, the press?). Indicative payback between 4 and 9 months depending on the current cost of finished-part scrap, customer returns and rework hours — estimate to be validated with your real numbers.
Tell us your case and within 48h we will send you the estimated ROI of this AI project for your furniture plant.
We work on your plant's real data, not on ours. Diagnostic with no commitment.
Request estimated ROI in 48h See iLEAN Vision