KTL/cataphoresis control with AI — the bath drifts hours before the paint fails; finding out in the touch-up booth is too late.

The problem

The bath drifts inside the band — and the defect shows up two hours later.

The paint line manager tells it with the same frustration in every plant: the SCADA panel shows pH 5.98, temperature 30.1, conductivity 1,450. All green. And in the touch-up booth, two hours later, bodies appear with craters, pinholing, uneven thickness. The bath was not out of spec when those bodies went through — it was drifting.

Three things happen at once in a real cataphoresis line, and almost nobody looks at them together:

1. The bath signals each go their own way — pH, temperature, conductivity, solids. Each with its own threshold. Combined drift (pH creeping down, conductivity creeping up) does not trigger an alarm because no single one steps outside its limits.
2. Paint consumption is not cross-referenced with quality — the top-up is done by recipe, without checking whether the ratio is pulling away from that tank's history for that part reference.
3. A large tank is not homogeneous — the gradient between the ends, uneven recirculation, dead spots in parts of the cage do not show up in the central reading. The veteran operator knows it; the system does not.

The outcome is always the same: the defect is spotted in the touch-up booth, the plant pays the rework cost, and the veteran operator once again puts out the fire with his eye and his notebook. The knowledge that prevents the problem lives in his head; the day he retires, it walks out with him.

How it fits the IRIS system

iLEAN does not replace the SCADA — it joins up what the SCADA keeps apart.

The KTL problem is not a lack of instrumentation: most lines already measure pH, temperature and conductivity. The problem is that each signal lives on its own island and combined drift triggers no individual alarm. iLEAN acts as the putty that binds those signals, paint consumption, line output and the history of your specific tank, without asking you to change the SCADA or the rectifiers.

Edge reads the bath signals in real time. Brain cross-references them with history and consumption, detects in-band drift and warns with hours of margin. The person decides what to correct.

The two iLEAN pieces applied to KTL control:

• Edge — a local terminal connected to the bath probes (pH, temperature, conductivity, ultrafiltration conductivity) and to the rectifiers. If the plant has modern probes, direct integration; if the probes are old and isolated, reading through the tank's local computer or complementary instrumentation. It runs locally: if the plant loses the network, Edge keeps reading, recording and firing alerts on the shop-floor panel. What is critical does not depend on WiFi.
• Brain — an agent that cross-references the bath signals with paint consumption (actual vs. expected top-up), the line's body output and the history of that specific tank for each part reference. It detects in-band drift (every signal inside its individual spec but combining toward a defect) and anticipates what to correct before rework appears. The brain does not act alone on anything critical — it proposes, the plant manager decides.

See the full IRIS architecture →

Before and after

Classic KTL control vs. control with iLEAN

Aspect	SCADA + threshold alarms	With iLEAN Edge + Brain
Cross-check of pH, temperature, conductivity	Each on its own individual threshold	Live cross-check, combined drift detected
Paint top-up	By recipe, never cross-checked with quality	Ratio cross-checked with line output
Homogeneity of a large tank	Central reading, gradients invisible	Real map from multiple cross-referenced points
Anticipating the defect	Alarm once it is already out of spec	Trend warning, hours of margin
Capturing the veteran's knowledge	Lives in his head	Pattern learned by the brain, repeatable
File for IATF / OEM audits	Rebuild signal by signal	Cross-referenced history, 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 specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Automotive KTL line, large tank (tens of thousands of liters), classic pH/temperature/conductivity instrumentation, rectifiers with SCADA, rework for cosmetic defects documented over recent months.
• Edge pilot on the existing instrumentation + Brain learning the tank's pattern over 4-6 weeks before it starts anticipating. First value expected within a few weeks: the cross-check dashboard and the detection of the first in-band drifts appear early, before the predictive model is fully tuned.
• Expected reduction of ≥30% in rejects for cosmetic defects attributable to KTL in the first months, a defensible floor.
• Indicative payback between 4 and 9 months. The hard lever: every body that does not enter rework is direct cost removed, and the touch-up booth is one of the most expensive positions on the line.
• A recurring benefit that does not enter the ROI but carries weight: the pattern learned for the tank stays as a permanent capability of the plant, not of the person who retires.

And the quality manager's reasonable doubt

“What if the AI infers a deviation wrongly and triggers an unnecessary change?” — the iLEAN brain does not act on the bath by itself. It proposes; the plant manager decides; the correction is applied or it is not. Hallucination is a problem of free generation, not of anchored tasks: in tasks where the AI cross-references bath signals with history and output, the best models brought error below 1.5%^[1]. And even so, what is critical goes through the safety rings — the brain lives in the outer ring, proposes inward, and the correction to the bath is signed by a person. Never the other way round.

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