Qualicoat powder coating on aluminum profiles — a cured lump is a 7-meter bar sent to rework.

The problem

A cured defect is not rework — it is sanding, labor and a degraded profile.

On a powder coating line for architectural profiles, the critical moment is before the oven. Up to that point, a lump or a sag is uncured powder: you blow it off, it goes back into the circuit and the bar returns to the booth. After the oven, the same incident moves into a different cost category:

1. Manual sanding of the profile — sanding an architectural profile to remove a cured lump is slow, it leaves a mark and it takes the thickness below the target Qualicoat class.
2. Re-application + second cure — gas, kWh, line occupancy, and the risk of not reaching the required thickness because the substrate has already been sanded.
3. Risk of failing Qualicoat — if the final thickness falls outside the class, the bar is no longer that product. Either it drops a class or it goes to scrap.

The booth operator knows this — the problem is that he cannot see the whole bar in time. A 7-meter bar with several visible faces moves at line speed, and the eye covers the nearby zone, not the rest. The classic system works 99% of the time. That 1% is the bar that reaches the oven with a lump in the middle of the profile.

How it fits the IRIS system

iLEAN does not change the booth or the oven — it provides the eyes the operator cannot have on the whole bar at once.

The problem on profiles is not a lack of skill: it is that the human eye does not scale to long bars with several faces moving at line speed. iLEAN acts as the putty that seals that crack between the booth and the oven, without asking you to change the powder, the hoist or the burner. It is the natural extension of the powder coating case already documented in automotive to the architectural sector.

Edge sees the powder before the oven and rejects the bar. Connect reads the dispenser and the booth however they come. The agent cross-references the bar's recipe and the Qualicoat class. The person signs whether the bar goes back to the booth or on to rework — the line does not decide alone.

The three iLEAN pieces applied to powder coating on architectural profiles:

• iLEAN Edge — one or more vision terminals (CNN) over the conveyor at the booth exit, before the oven. They recognize the lump and sag patterns, fire an actuator (light stack + signal at the station) and allow the bar to be rejected in milliseconds. It works with no network: if the plant loses WiFi, Edge keeps detecting and stopping defective bars.
• iLEAN Connect — reads the powder dispenser and the booth control by whatever route they allow: modern integration if there is one, a panel reading or a photo of the display if it is old. And it captures the recipe record (Qualicoat class, color, target thickness) from the ERP, the MES or the spreadsheet where the coating manager keeps it.
• Agent — cross-references the detected defect, the bar's recipe and the history (what changed since the last good bar: powder, humidity, gun settings). It proposes a root cause to the booth manager, not just the isolated defect. At the end of the lot, it leaves a per-bar dossier for the Qualicoat auditor.

See the full IRIS architecture →

Before and after

Visual inspection at the end vs. detection before the oven

Aspect	Classic inspection	With iLEAN Edge + Connect + Agent
Detection point	Oven exit or shipping	Before the oven, with the powder still uncured
Powder recovery	No — the powder is already cured	Yes — it goes back to the booth circuit
Type of rework	Sanding + re-application + re-curing	Blow-off + return to the booth
Qualicoat class risk	High if sanding was needed	Low — the profile keeps its base thickness
Root cause of the defect	“The gun, I suppose”	The agent cross-references history and recipe
Operation with no network	n/a	Edge keeps running on the panel's power

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.

• Qualicoat plant coating architectural profiles, one or several booths, long multi-face bars, a mix of classes (Cl. 1, Cl. 2, Cl. 1.5).
• Edge pilot on one booth (CNN cameras before the oven + rejection actuator + integration with the dispenser and the recipe). First value expected within a few weeks.
• Expected reduction in bars needing heavy rework (sanding + re-application) ≥ 30% against the documented frequency of recent months.
• Indicative payback between 4 and 9 months, depending on the average rework cost per bar and the percentage of bars currently downgraded to a lower class because of sanding.
• The hard lever is changing the category of the defect: from deep rework to a clean re-application. The powder is recovered, the bar keeps its class, and the oven does not burn a profile's cycle on rework.

And the booth manager's reasonable doubt

“What if the camera rejects a good bar because of a reflection?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI simply recognizes a visual pattern on a controlled substrate (powder on aluminum inside the booth) and compares it with the learned pattern, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the bar and the booth manager signs whether it is re-applied or goes to rework. The three safety rings exist precisely for this.

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