Tissue and towel mills run on a tradeoff that never fully resolves: use enough chemistry to hit strength targets, and no more, while raw material quality drifts, machine speed changes, and grades shift throughout the day. Get it wrong in one direction and you waste chemistry and margin. Get it wrong in the other and quality slips. For years, the only lever mills had was operator experience — skilled, but inconsistent shift to shift and impossible to scale.
One tissue manufacturing facility set out to change that by putting chemistry dosing under autonomous, self-learning control.
The challenge
The mill’s dry-strength and wet-strength chemistry programs were run the way most are: operators adjusting dosage based on lab results that lagged the process by hours, plus judgment built up over years on the floor. That works, to a point. But it also means every shift change reintroduces variability, every raw material shift requires a manual correction, and the mill has no way to distribute its best operators’ instincts to the other two shifts.
The mill needed a way to hold chemistry dosage on target continuously — not just when someone was watching closely.
The approach
ProcessMiner deployed its adaptive, self-learning platform directly against the mill’s chemistry program. Ahead of go-live, the system loaded roughly a year of historical process data — physical tissue property measurements, papermaking process knowledge, and machine metadata — to build its initial models. From there, it took over as a continuously learning system, not a static one.
A few things set the deployment apart from a conventional control scheme:
The models retrain with every new lab value that comes in, so the system’s understanding of the process keeps current with the mill’s own data rather than drifting out of date. When the process shifts — a raw material lot changes, a grade transition happens — the platform detects it and retunes automatically instead of waiting for an operator to notice and intervene. And a built-in error-isolation step uses statistical F-tests to catch and discard bad sensor readings in real time, so the models aren’t learning from noise.
The result is a system that predicts quality outcomes every 30 seconds and adjusts chemistry dosage in a closed loop — holding the process to target without an operator manually recalculating dosage every time something changes upstream.
This is the same trend-based, closed-loop approach ProcessMiner uses across pulp and paper deployments: predictions run continuously in the background, and the system acts only when the process needs correcting — not as a constant stream of alerts operators have to sort through.
The results
Once live, the platform’s autonomous control held the mill’s chemistry usage to a meaningfully lower level than manual dosing had achieved, while meeting the same wet- and dry-strength specifications:
- Dry-strength chemistry usage down 25% — at the proven ceiling ProcessMiner has demonstrated across pulp and paper deployments, not a one-off best case.
- Wet-strength (Kymene-grade) chemistry usage down roughly 30% for the grade mix run at this mill — a deployment-specific result tied to this facility’s grade profile, not a general claim across all wet-strength programs.
- Real-time predictions every 30 seconds, sustained through the deployment, giving the closed-loop controller a continuously current read on the process rather than a lagging one.
- Improved adherence to target parameters across both strength specifications, with the system holding the process closer to its optimal operating range than manual dosing had.
None of this came from a single tuned moment in time. The models kept adapting as raw materials, grades, and machine conditions changed — which is the part that let the mill sustain the reduction instead of losing it back to variability within a few weeks.
Why it worked
The mechanics matter here more than the headline number. Most chemistry-control efforts fail not because the model is wrong on day one, but because the process moves and the model doesn’t move with it. This deployment’s advantage was a modeling approach built to keep pace with a live process: continuous retraining, automatic detection of process shifts, and real-time error filtering, all running underneath a 30-second prediction cycle.
That’s also why the reduction held at the proven 25% ceiling rather than a smaller, temporary dip. ProcessMiner’s conservative floor across pulp and paper chemistry deployments is 10% — a minimum commitment mills can expect even in a first phase. This deployment reached the high end of that range because the closed loop stayed accurate as conditions changed, not because of a single well-timed adjustment.
For mills evaluating whether autonomous control is worth the change-management effort, this is the case for it: chemistry cost is one of the largest controllable line items in a tissue or paper operation, and the system that manages it needs to adapt as fast as the process does. See how autonomous, closed-loop control works across pulp and paper deployments, or explore the ROI case for chemistry optimization to estimate what a similar reduction could mean for your mill.