ΔE color control on multi-customer flexo ink — start-up cannot keep being the scrap of the day.

The problem

Starting a run is paid for in rolls, not in coffee.

In a multi-customer flexo printing plant, start-up involves three realities that almost never come together in time:

1. The customer's color recipe — base ink, Pantone or reference, target ΔE, approved anilox. It usually sits in a PDF from the customer, in the ink supplier's data sheet or in the ink manager's head.
2. The reality of the plant — the day's substrate, speed, temperature, ink batch, additive dosing. It changes with every run.
3. The color coming off the press right now — only the operator's eye sees it (and it gets tired), or the benchtop spectrophotometer does (used after the fact, on a cut sample).

The usual pattern: the press starts, the ink manager opens the first roll, measures, corrects, measures again. One roll, two, sometimes three before the run is inside the customer's ΔE. Multiplied by the changeovers in a day, start-up scrap is not an accident: it is a fixed cost hidden inside OEE. And the day the ink manager is on vacation, the problem takes on a different face.

How it fits the IRIS system

iLEAN does not replace the ink manager — it gives them a camera that never gets tired.

The problem with multi-customer ΔE is not a lack of judgment: it is that the ink manager's judgment cannot be on the line, on every print unit, on every roll, on every customer changeover. iLEAN acts as the putty that seals the crack between the customer's PDF, the ERP and the press, without asking you to change press, ink or supplier.

Vision sees the color unit by unit. Connect reads the customer's recipe wherever it comes from. The agent cross-references it with the ΔE curve and proposes the pump adjustment. The person signs — the machine never decides on its own to change dosing.

The three iLEAN pieces applied to ΔE color control in multi-customer flexo printing:

• iLEAN Vision (Edge specialized in color) — a terminal with a CNN camera calibrated on the real substrate running on the press. It reads the color on the ink unit, calculates ΔE against the customer's target and fires an alarm/actuator if the drift crosses the threshold. It works with no network: if the plant loses WiFi, the camera keeps reading and keeps alerting.
• iLEAN Connect — captures the color recipe whether it comes from the customer's PDF by email, from the ERP, from an ink supplier's data sheet or from a spreadsheet somebody updated this morning. It also captures the customer's response when they reject a sample, at second zero, with nobody having to forward emails.
• Agent — cross-references the inline ΔE reading, the customer's signed recipe and the history of similar runs (which pump, which speed, which additive brought the color in) and proposes the adjustment. It does not move the pump on its own: it prepares the correction and the ink manager signs. At the end of the run, it stores the recipe as the starting point for the next time that customer comes round.

See the full IRIS architecture →

Before and after

Starting by eye vs. starting with a signed recipe

Aspect	Classic start-up	With iLEAN Vision + Connect + agent
Inline color reading	The ink manager's eye	CNN camera unit by unit, continuous ΔE
Customer recipe	PDF, ink supplier's data sheet, memory	Signed, reusable recipe per run
First usable roll	After two or three rolls of adjustment	Expected reduction in start-up rolls
Customer changeover	Downtime + scrap	Recipe loaded + guided correction
Traceability for the customer	A cut sample, filed away	ΔE curve for the whole run, roll by roll
Operation with no network	n/a	Vision keeps reading on the panel's own 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.

• Flexible packaging printing plant, several presses of 4-8 print units, dozens of runs a month with frequent customer and substrate changeovers.
• Vision pilot on one press (CNN camera over one or several units + integration with the customer recipe and the dosing pump). First value expected within a few weeks.
• Expected reduction in start-up rolls of ≥ 30% on multi-SKU runs, depending on your current changeover pattern.
• Indicative payback between 4 and 9 months, depending on run volume, the cost of a roll and your current rate of color complaints.
• The hard lever is rolls recovered + complaints avoided: every run that hits ΔE on the first roll pays for the system.

And the ink manager's reasonable doubt

"What if the AI gets it wrong and approves a run that is out of tolerance?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system into another (reading a color and comparing it against the target curve), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN proposes the dosing adjustment and the person signs. The three safety rings exist precisely for this.

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