Aero paint & coating with AI — the bath drifts in silence and you find out 200 parts later.

The problem

The tank runs fine until it doesn't, and the sampling never arrives in time.

Every chem proc manager tells the same story: "my line meets mil-spec as long as the parameters stay in range. The problem is that sampling runs every 8 hours and the bath drifts in 3." And underneath, the reality:

1. The bath is titrated at 7 and known at 4 pm — the operator takes a sample, the lab titrates, the result comes back. If in that window a run of parts went through dragging chemistry with it, the drift appeared and was seen afterwards.
2. Thickness is measured at the end, not in line — the inspector pulls the part, measures with a handheld probe, jots it on a sheet. If it's out, curing is already done, tank time already spent, paint already used.
3. The cure curve lives in another system — the oven records, the batch is in the MES, tying them to the part's dossier is manual work.
4. The per-part NADCAP chem proc dossier — bath composition, thickness per point, cure curve, paint batch, certs — gets assembled when the OEM customer asks for it. A sprint for the quality team.

And the worst-case scenario: discovering the bath was contaminated retroactively, with 200 parts already made, already painted, already delivered. Customer claim, rework, program delay.

How it fits the IRIS system

iLEAN does not replace the tank or the lab — it seals the cracks between bath, thickness and dossier.

The problem in chem proc is not a lack of control: it is that the control lives in islands and on a delay. The tank has its sensors, the lab has its titration, the inspector has their probe, the oven has its curve — but joining them per part and in the moment requires a person cross-referencing screens and filling in sheets. iLEAN acts as the putty that joins tank + lab + thickness + curing + MES batch into a living per-part dossier.

Edge measures thickness in line and sees the part. Vision reads the serial number. Tracer joins bath, curve and certs. The agent holds the part if a reading falls out, holds the batch if the bath approaches the limit. The person signs.

The three iLEAN pieces applied to aero coating:

• iLEAN Edge — an in-line terminal with a thickness probe (eddy current / XRF / interferometry depending on the process) and an industrial camera (CNN). It measures N points per part defined by geometry, links them to the part's serial, and holds before the next step if a reading falls out of range. It works without a network — if the plant loses WiFi, the Edge keeps measuring and holding.
• iLEAN Vision — assisted visual inspection on the coated part (cosmetic defects, missing coverage, visible contamination, out-of-range color). The CNN learns the defects your inspector sees. It does not replace the inspector — it gives them a first filter and takes the bulk of trivial inspections off their plate so they can spend time on the real doubts.
• iLEAN Tracer (Connect) — capture from the bath sensor (pH, conductivity, temperature, concentration of the key chemical) continuously, from the lab titrator when the result arrives, from the curing oven (curve per load), from the MES (which part, which batch, which SKU), from the PLM (which specification applies). The agent assembles the per-part NADCAP chem proc dossier, ready when the customer asks.

See the full IRIS architecture →

Before and after

Manual chem proc vs. with iLEAN

Aspect	Periodic sampling + handheld probe	With iLEAN Edge + Vision + Tracer
Bath control	Titration every shift, result hours later	Continuous in-tank sensor + agent that alerts
Thickness measurement	Handheld probe at the end of the line	Online, per part, in line
Out-of-range detection	After curing, with the part already made	Before the next step, in milliseconds
Curing-part-bath association	Manual cross-reference across three systems	Living per-part dossier, self-generated
Visual inspection	Inspector goes over every part	Vision filter + inspector focuses on doubts
NADCAP chem proc dossier	Assembled by hand when the customer asks	Generated automatically by the Writer

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 line. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Aero NADCAP chem proc line with anodizing or plating + paint, a mix of Boeing/Airbus/Safran/RR programs.
• Edge pilot at the thickness measurement point + continuous bath sensor. First value within a few weeks: the first bath drift caught before it affects parts pays for the pilot.
• Indicative payback between 4 and 9 months, depending on parts volume per year, frequency of recent chem proc incidents and average rework cost.
• Rework reduction ≥30% once Edge and Tracer are on the line. The hard lever: one single bath drift caught in time saves the annual cost of the investment.

And the chem proc lead's doubt

"What if the camera mistakes a cosmetic defect for dust on the part?" — hallucination is a problem of free generation, not of anchored tasks. Identifying a defect on a painted part is exactly what computer vision does best: it learns from your historical rejects. In anchored tasks, the best models brought error below 1.5% ^[1]. And the line does not approve on its own: Vision flags the doubtful, the inspector decides and signs. What is critical always has a human in the loop.

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