Pulp and paper with AI — a web break gives no warning, but its signature does.

The problem

The QCS keeps you in spec. It was not built to anticipate the break.

The QCS (Quality Control System) on every paper machine does its own job very well: it keeps basis weight, moisture and ash within band by scanning cross-direction profiles. The web break, by contrast, lives at the crossing of four realities that are almost never on the same screen:

1. What pulp comes in — headbox consistency and composition. In the lab every hour, in the DCS in real time, almost never cross-referenced with sheet formation.
2. What happens in the press — vacuums, pressures, dryness at the exit. In the line's DCS.
3. What happens in drying — the tension profile across the sections, the cylinder temperatures. In another system, often on another screen.
4. What the sheet shows — formation defects, pinholes, spots, wrinkles. The operator's visual inspection, one spot camera or the defective roll at the end.

The machine lead knows it, but cannot hold all four sources in one head. When the web breaks it is 30-90 minutes until sheet is running again, plus the repulping. The classic system works 99% of the time. That 1% is where the machine-hours that make the month are lost.

How it fits the IRIS system

iLEAN does not replace the QCS — it seals the cracks between QCS, DCS, lab and sheet.

The break problem is not a lack of information: it is information that lives in islands and that, at the critical moment (the minutes before the break), does not reach the decision-maker cross-referenced. iLEAN acts as the putty that fills those gaps, without asking you to change the QCS, the DCS or the pulp lab.

Edge watches the sheet with deep vision at real machine speed. Connect reads QCS, DCS and lab. The agent cross-references that data against the known pre-break signature and alerts the machine lead before the web goes.

The three iLEAN pieces applied to pulp and paper:

• Edge — a terminal with computer vision (CNN) on the machine. Cameras after the calender or over the forming section, with lighting adapted to classify lumps, pinholes, formation defects, spots and wrinkles. It tags the zone and sends the trace to the agent. Works without a network.
• Connect — captures data from the QCS (basis weight, moisture, ash), from the DCS (press vacuum, drying profiles, tension per section) and from the pulp lab (consistency, freeness, fiber), whether it comes from a modern system, an old server or a lab spreadsheet. And it captures what arrives from outside: a pulp supplier change by email, a recipe adjustment from the production manager.
• Agent — cross-references the visual defect pattern with the process signature. When it recognizes a pre-break, it alerts the machine lead through whatever channel they use, with the probable root cause. The person decides what to touch; the machine does not correct itself on its own.

See the full IRIS architecture →

Before and after

A stand-alone QCS vs. cross-referenced control with iLEAN

Aspect	QCS + DCS not cross-referenced	With iLEAN Edge + Connect + Agent
Formation defects	Operator visual inspection or at rewinding	Continuous, classified vision at machine speed
Web pre-break	Seen once it has already broken	Alert to the machine lead on a recognized signature
Root cause of the break	Post-mortem investigation, days	Automatic cross-reference with consistency, vacuum and tension
Time recovered	Each break: 30-90 min + repulping	Each break avoided: machine-hours per year
Operation without network	n/a	Edge keeps classifying on the panel's own power
File for the customer	Rebuilt by hand per complaint	Per-reel dossier, automatic, with a photo of the defect

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.

• Mid-basis-weight paper machine (graphic paper, tissue, light packaging), running at 800-1,500 m/min, with modern QCS and DCS but not cross-referenced with sheet vision.
• Edge pilot over the calender or the forming section (cameras + lighting + integration with QCS/DCS and the lab). First value expected within a few weeks.
• Reduction of unexplained web breaks ≥ 30% versus the historical baseline — a conservative estimate.
• Indicative payback between 4 and 9 months, depending on the documented breaks/month and your line's machine-hour cost.
• The hard lever is the machine-hour recovered, not the scrap from the roll.

And the machine lead's reasonable doubt

“What if the AI fires false alerts at me every five minutes?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely cross-references a visual pattern and a process signature against historical events, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN alerts and the person decides what to touch. The three safety rings exist precisely for this.

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