BRENNEN

Industries / Mining and materials / Pulp and paper

Quality control

Quality control measures pulp and paper during and after production so that each roll meets its specification for weight, moisture, strength and defects.

3

Milestones achieved, from 1915 to 2020.

2

To watch for: announced, not yet confirmed outside.

1

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1915

    A shared technical body

    In February 1915, thirty-five industrialists formed a technical session "for the purpose of exchanging views and mutual knowledge and to advance the art of papermaking", and by September they voted to become the Technical Association of the Pulp and Paper Industry, TAPPI. TAPPI now publishes standards and test methods used across the industry.

    tappi.org

    Credit

    TAPPI's founding members

  2. 1950

    Measuring paper on the run

    Industrial Nucleonics of Columbus, Ohio, sold a beta gauge that measured paper as it ran through the machine by how much of a radioactive krypton-85 source's radiation passed through it. By ABB's account the sensor had been on sale for about 19 years by 1969; it first measured one fixed point, and was then mounted on a scanner that moved back and forth across the sheet.

    LessonBefore they could sell on accuracy, its salespeople had to convince mill managers and machine crews that the radioactive source was safe.

    abb.com

    Credit

    Industrial Nucleonics, later AccuRay, bought by ABB in 1990

  3. 2020

    Cameras find where breaks begin

    At Torraspapel's mill in Leitza, Spain, part of the Lecta group, an eight-camera web monitoring system on a 1,000 m/min coating machine records each break. A mill manager reported that "almost all the breaks are caused by the incoming material and not caused by the machine", and the mill now sends images to its suppliers.

    LessonRecording each break showed the fault lay upstream, in the paper delivered to the machine.

    valmet.com

    Credit

    Torraspapel (Lecta); Valmet, the supplier

Watch for

  1. Strength predicted between lab tests

    Valmet says its IQ Virtual Sensor software predicts tensile, tear, burst and compression strength every minute from a mill's existing data, and keeps learning from lab results. The release names no customers and gives no accuracy or savings figures.

    Achieved whena mill or an independent party publishes the prediction error against lab tests over months of production.

    valmet.com

    Claimed by

    Valmet, March 2026

  2. Weighing tissue without radiation

    Valmet says it has launched the world's first fully non-contact, nuclear-free quality control system for tissue, measuring weight, moisture, softness and creping (the scraping of tissue off the drying cylinder, which gives it softness) without touching the sheet or using a radioactive source.

    Achieved whena mill reports the system in routine use, with its readings checked against lab measurements.

    valmet.com

    Claimed by

    Valmet, June 2026

Open

  1. 01

    A public paper-defect image set

    An openly licensed set of camera images from running paper webs, with labelled defects such as holes, spots, wrinkles and edge cracks and the marks that come before a break, plus a baseline detector and its scores.

    The brief
    Credit · unclaimed

Open 01 · the brief

A public paper-defect image set

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An openly licensed set of camera images from running paper webs, with labelled defects such as holes, spots, wrinkles and edge cracks and the marks that come before a break, plus a baseline detector and its scores.

Why it is still open

A 2024 study by Tampere University and the camera maker Procemex trained its detector on images from two cameras on one paper machine, with 19 and 35 defect images in its two test sets, and does not describe releasing them. The best-known public dataset from a paper mill, released by Ranjan and colleagues in 2018, holds sensor readings, not images.

scitepress.orgarxiv.org

Built with what exists

Cameras already installed on mill machines, or two industrial cameras on a university pilot paper machine; Intel's open-source Anomalib library for anomaly detection; a small open vision model fine-tuned on one graphics chip (Note: "One GPU to fine-tune").

  1. First step

    Write to two university pilot paper machines and one mill quality manager, offer to strip mill names and settings from the images, and agree a shared list of defect types.

  2. Done looks like

    At least 10,000 images from at least two machines, with at least 500 labelled defects across at least five defect types, published under an open licence with baseline scores for each defect type.