Add a real-time inspection layer from the pulp bale to the finished reel with vision AI for pulp and paper defect detection. Built for the mills where a hole the size of a fingernail hides in a nine-meter web running at 1,800 meters a minute, a dirt speck in the pulp becomes a customer complaint on a coated sheet three days later, and the wrinkle that started at the dryer section is discovered when the roll unwinds at the printer. Roboflow inspects every meter of the sheet, every bale and reel end, and every finished roll on the cameras you already have or the ones a single camera adds, and turns the findings into per-reel records, in real time.
Sheet and Web Defects on the Machine:
- Detect holes, pinholes, and thin spots across the full web width at machine speed, from the wet end to the reel, so a break is predicted or a hole is mapped before it reaches the winder
- Catch wrinkles, creases, calender cuts, and edge cracks at the dryer section, calender, and reel, where a fold becomes a break or a rejected roll
- Flag dirt, shives, slime spots, stickies, and fiber lumps on the sheet, so the contaminant is traced to the stock, the felt, or the headbox instead of found by the customer
Pulp, Coating, and Finishing:
- Count dirt and shives on pulp sheets and bales against ISO 5350 and TAPPI T 437 dirt-count methods, and detect bale wire, wrapper, and contamination at the pulper feed
- Detect coating streaks, skips, scratches, blade lines, and orange peel on coated grades, and measure defect size and position so a streak is matched to the blade or the applicator
- Inspect reel and roll ends, core condition, wrapper, and label at the winder and roll handling, so a damaged or telescoped roll never leaves the mill
Reels, Root Cause, and Systems Integration:
- Build a defect map for every reel and every wound roll, each flag logged by cross-direction position and machine-direction meter, so the winder set and the converter know what they are unwinding
- Match periodic defects to roll, felt, and cylinder circumference, so a spot repeating every 4.3 meters names the press roll or dryer cylinder that made it
- Push alarms, downgrade decisions, and reel records into the QCS, DCS, and MES through PLC and API integration, alongside the basis weight, moisture, and caliper data the machine already logs
Bring intelligence to every reel today. Stop holes, wrinkles, dirt, and coating streaks from becoming sheet breaks, downgraded reels, and the customer claim that arrives from the printer with a photo.
More About Pulp and Paper Defect Detection
What is pulp and paper defect detection with Vision AI?
Pulp and paper defect detection with vision AI uses cameras along the paper machine, coater, calender, winder, and pulp line and deep-learning models to detect, classify, and locate defects in the sheet and in the pulp: holes, pinholes, thin spots, wrinkles, creases, calender cuts, edge cracks, dirt, shives, slime spots, stickies, coating streaks and skips, and damaged reel and roll ends. Each finding is logged by cross-direction position and machine-direction meter, so the mill gets a defect map for every reel and every roll, periodic defects are matched to the roll or cylinder that made them, and downgrade and cull decisions are made with an image behind them. It runs alongside the machine's existing QCS scanners and web monitoring, and feeds the DCS, MES, and quality systems the mill already uses. For film, nonwoven, and general converting webs, see web and film inspection.
Can it handle machine speed, web width, and the dust and steam on a paper machine?
Yes. Line-scan and high-speed area-scan cameras already run on paper machines at full speed and full width, and models are trained on footage from those cameras in the real machine environment, so steam at the dryer section, dust at the winder, broke on the floor, and the daylight change through the machine-room windows are the training data rather than an exception. Inference runs on an edge device at the machine, on-prem, inside the mill's control network, within the decision window a break-prediction or cull decision needs. The honest constraints are optical: a defect smaller than what the camera resolution can resolve at machine speed will not be seen, and a camera position determines what part of the sheet it can see, which is why a deployment starts with the defects that cost the most and the positions that catch them.
Can it integrate with our QCS, DCS, and mill systems?
Yes. Roboflow Inference (opens in a new tab) runs on an edge device at the machine and exposes a standard API and common industrial protocols, so defect detections, reel maps, and downgrade decisions flow into your existing systems: quality control systems from Valmet, ABB, Voith, and Honeywell, distributed control systems from ABB, Emerson, Siemens, and Honeywell, mill MES and reel tracking, winder and roll handling controls, PLCs from Allen-Bradley and Siemens, and quality lab and LIMS systems, through REST, MQTT, OPC UA, discrete I/O, and direct database writes. PLC-level integration triggers a winder slowdown, a cull, or a reel-end mark on the spot, and every event carries reel and roll ID, grade, cross-direction and machine-direction position, defect class, timestamp, and imagery, with a full record behind every reel that ships.
