Pigment dispersion in a bead mill — stop at the right fineness, without over-milling.

The problem

"One more pass, just in case" — the rule that runs half the bead mills out there.

Fineness gets measured properly (Hegman gauge, NPIRI grindometer for inks), but the measurement lives outside the batch:

1. Manual, one-off readings — the operator draws down on the gauge, eyeballs where the coarsest particles start to scratch, and writes it down — or doesn't — at the end of the batch.
2. The stop is decided by habit — "this red always takes four passes." If the premix came in better dispersed, passes are wasted; if it came in worse, they fall short.
3. Over-milling is not free — every extra pass is kWh, occupation of the plant's bottleneck and heat in the millbase; on sensitive organic pigments, heat and excess shear shift the shade and can destabilize the dispersion. It also wears beads and liners.
4. Falling short is found out late — a coarse grind shows up at application QC (tint strength, gloss, specks) once the batch is already discharged: rework or downgrade.

The production manager knows it: the mill is one of the most expensive machines in the plant to run, and the decision of when to stop it rides on the least-traced data of all.

How it fits the IRIS system

iLEAN does not change your gauge or your mill — it ties every reading to its pass and its batch.

Dispersion does not fail for lack of a gauge; it fails because the reading never gets tied to the pass, the batch and the mill parameters. iLEAN acts as the putty between the Hegman gauge, the mill PLC and the ERP, without changing how you measure.

Edge identifies the batch and the pass. Connect captures the mill parameters. The agent calls the moment the batch is in fineness — and the person signs off the discharge.

The iLEAN pieces applied to dispersion fineness:

• Edge — a terminal at the mill. It reads the batch label (QR), and the operator dictates the reading by voice ("pass three, Hegman six and a half") or photographs the gauge with the drawdown on it, and the system ties it in. It works without a network. If the plant loses its connection, it keeps capturing.
• Connect — hooks into the mill PLC (or the drive) and captures agitator rpm, flow rate, millbase temperature and specific energy per pass. If the mill is old and exposes nothing, it captures what there is: hours and temperature.
• Milling agent — cross-references fineness + pass + energy + temperature against the product spec. When the batch reaches fineness, it calls it: "batch in spec — discharge." If the millbase runs hotter than the pigment's limit, it warns before that. The person signs — the mill never decides alone.

See the full IRIS architecture →

Before and after

Fineness on paper vs. milling traced with iLEAN

Aspect	Gauge + paper record	With iLEAN Edge + Connect (pass by pass)
Fineness record	One note at the end, if at all	Every reading tied to batch, pass and time
Stop decision	By habit ("always four passes")	Against the spec, batch by batch
Extra passes	Invisible — nobody counts them	kWh and extra passes visible per batch
Coarse grind	Shows up at application QC	Warning before the mill is discharged
Millbase temperature	Checked "just in case"	Captured and tied to the batch's shade
Milling dossier	Does not exist	Fineness-energy curve per batch, 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.

• Manufacturer of pigment pastes, inks or coatings running bead mills (horizontal or vertical) with fineness control by Hegman/NPIRI gauge.
• Edge pilot on one mill (terminal + batch identification + PLC capture). First value expected within a few weeks.
• Indicative payback between 5 and 10 months, depending on energy cost per batch, mill occupation and how often batches get reworked for coarse grind.
• Hard levers: killing the "just in case" pass where it adds nothing; ≥ 20% fewer mill hours per ton on mature products (estimate to validate); fewer reworks for out-of-spec fineness.

And the quality manager's reasonable doubt

"What if the system mishears a reading dictated by voice?" — it can happen, and that is why the system never decides alone. Hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely records a reading and compares it against a specified range, the best models brought error below 1.5% ^[1]. And even so, the mill discharge is signed by the operator or the quality manager — the three safety rings exist precisely for this.

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