ProductionNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.

AI warns paper machine operators before a sheet break and points to the likely cause, so mills lose fewer hours to breaks and rethreading (Sappi, Georgia-Pacific, a Voith paper line)

Added 2026-11-18 · Last updated 2026-11-18

Manufacturers in this case: Sappi · Voith · Georgia-Pacific · all manufacturers using AI

Role

Production and process engineering in pulp, paper, tissue and board mills (paper machines and converting lines) · Typical users: paper machine superintendent, process engineer, production manager, converting manager, mill manager, operations data or digital lead

The problem

When the paper web breaks on a paper machine, production stops without warning, the machine must be cleaned and rethreaded, and the paper made in the meantime becomes broke or waste. Breaks have many interconnected causes (moisture, draw, vacuum, chemistry, raw material), and they are hard to explain even for experienced people, although thousands of sensors are already recording the process. Vernaio, which worked on a paper line with Voith, describes the problem as one that has resisted all attempts to solve it for decades. The authors of the Sappi pilot write that manual analysis and traditional methods did not prevent breaks from coming back, and that the rapid changes in the wet web are very hard to detect in real time. At Georgia-Pacific, tears and breaks on parent rolls and converting lines can cost millions of dollars per year per line, and the company has more than 150 converting lines; the data on material quality, moisture, temperature and machine calibration sat in disparate sources, and a small team of site experts held the knowledge of each site's machines, many of them about to retire.

How AI helps

Machine learning models are trained on the historical process data of the machine (quality control system, distributed control system and machine sensors), with the moments of past breaks marked in the data. In real time, the model scores the risk of a break and shows which signals are drifting, so operators can act before the break and engineers can work on root causes. In the Sappi pilot, the BIG-AI platform from Turilytix, with the paper process consultancy Optipid, watched variations in draw, moisture, vacuum and other parameters, and gave warnings up to 30 minutes before a break with the likely root cause. On a Voith paper line, Vernaio's aivis engine was fed the raw historical data, used a reflection sensor in the drying section to mark breaks, and sorted the breaks into several root cause scenarios, each with its own signals; the process engineer used them to design countermeasures. At Georgia-Pacific, models built with Amazon SageMaker on an operations data lake give operators real-time feedback on the best machine speed and other settings, and predict, from the quality of each parent roll, how fast a converting line should run to avoid tears.

Step by step

  1. List every break of the last six to twelve months with time, position (wet end, press, dryer, reel), grade, speed and the cause written in the log.
  2. Export the main process signals (moisture, draw, vacuum, consistency, pH, steam, speed, basis weight) at one-minute resolution for the hour before each break and for calm periods.
  3. Group the breaks by position and grade, and look with the process engineer for the signals that move before each group.
  4. Test a model or tool in advisory mode: it gives warnings but operators decide, and every warning and break is reviewed with the crew.
  5. Turn the root cause scenarios into countermeasures (setpoints, chemistry, maintenance, raw material) and standard responses for operators.
  6. Track breaks per day by position, lost hours, broke tonnage and false warnings, and retrain the model after grade or equipment changes.

Tools

  • Turilytix BIG-AI sheet break prediction, with Optipid: used at Sappi
  • Vernaio Process Booster with the aivis engine: used on a Voith paper line
  • Amazon SageMaker models on an AWS operations data lake: used at Georgia-Pacific
  • Other tools to evaluate: the break analysis and advisory features of your paper machine or quality control system supplier, other industrial AI process analytics platforms, your historian's anomaly detection
  • To start without new software: a historian export around past breaks and any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below

In the AI tools index: Turilytix BIG-AI · Vernaio aivis (Process Booster) · Amazon SageMaker · all AI tools used by manufacturers

Result

Sappi (paper by the vendor Turilytix and Optipid in the IPPTA journal, 2025): on a pilot with 45 days of test data, prediction accuracy of 83.53%, sheet breaks down 35% in the first three months after commissioning, unscheduled downtime down 28% and material waste down 15%; the authors' seminar slides also mention a 50% break reduction, so take the 35% figure from the case study table. Voith paper line (Vernaio success story, the vendor): sheet breaks reduced by an average of 55% and up to 87% (from 16 to 2 per week), with warnings as early as 60 minutes before a break. Georgia-Pacific (AWS case study): 40% of parent-roll tears eliminated on one converting line, an increase in profits of millions of dollars for that line, and failure predictions 60 to 90 days ahead for selected assets.

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

  • The Sappi and Voith figures come from the vendors (the Sappi paper is written by the vendor and its partner, and its slides give a different reduction figure); the Georgia-Pacific figures come from AWS's case study and cover one converting line.
  • The Sappi pilot used 45 days of test data; check the results over a full year, with grade changes and seasonal raw material variation.
  • A warning only helps if operators know what to do: write standard responses and review false warnings, or crews will start ignoring them.
  • Breaks have physical causes (felts, wires, doctor blades, raw material): the AI points to signals, but the countermeasures still come from process engineers and maintenance.
  • Historian data is often poor around breaks (missing tags, wrong timestamps, breaks not logged); plan time to clean it before judging any model.

Try it yourself

You are a paper machine process engineer. I want to understand our sheet breaks and decide where an early warning would help most.

I will paste: (1) our break log for the last months [date, time, machine, grade, speed, break position (wet end, press, dryer, reel), duration until rethread, cause written by the crew]; (2) optionally, a summary of key signals in the 60 minutes before each break [moisture, draw, vacuum, consistency, pH, steam pressure, basis weight].

Do this:
1. Count breaks and lost hours by machine, grade, position and shift, and show the top five patterns.
2. Clean up the cause text into a short list of cause categories and say how many breaks have no usable cause.
3. If signals are provided, list which ones moved before each group of breaks, and say clearly where the data is too thin to conclude.
4. Propose a trial for an AI early-warning tool: which machine and break type to start with, how to measure warnings, false alarms and breaks avoided, and what operators should do when a warning comes.
5. Draft a one-page summary for the mill manager with lost hours and tonnage at stake.

Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Want more like this? Join the newsletter

Related cases

All 21 Production use cases →

Sources

Weekly: 1–2 practical AI use cases for your plant

Newsletter in

Free, one email a week. By subscribing you agree to receive the weekly newsletter (and its sponsor message) from AI Agents for Manufacturing. Every email includes an unsubscribe link, and you can unsubscribe anytime. See our Privacy Policy.