Know what a setpoint change will do before you make it. Ceptron learns the physics of cooking, washing, recovery and the paper machine from millions of simulated mill histories, then reads your mill's own history to answer for your mill. No bump tests, and no experiments on the running mill.
The physics hereChemistry and heat. Mass, energy, sodium and sulfur balances for every unit; cooking, bleaching and causticizing kinetics; dewatering and drying; delays from tanks and residence times, seconds at the dryer to months in the recovery loop.
A mill answers in seconds at the dryer, hours at the digester and months in the recovery loop. The model is trained on simulated histories, not snapshots, so slow effects are part of every answer.
Operators and controllers react to what they see, which makes mill history misleading. The model is built to tell what the controllers did from what the process did.
Physics sets the ceiling, and each mill sits below it in its own way: dryers typically use 1.2 to 1.35 tonnes of steam per tonne of water, against about 1.0 in theory. The model infers where your mill sits from its history.
Ask: What is the lowest steam per tonne at 7.5% sheet moisture, with strength in spec? Answer: raise press load 5%; dryer steam falls about 4% (± 1), moisture holds, break risk unchanged. Illustrative example.
The forecast, the recommended setpoints and how sure it is.
Regenerates white liquor and reburns lime. The equilibrium sets a ceiling; the kiln sets fuel cost.
How the mill is actually run: DCS loops, advanced control and operator habits. The reason mill history cannot be read naively.
When the mill does not do what was forecast, find the cause and add the missing mechanism.
Turns chips and white liquor into pulp: how far lignin is removed (kappa), yield and alkali consumption.
Concentrates weak black liquor so it can burn. Steam economy and scaling decide capacity.
Given every parameter, it computes the outcome: mass, sodium, sulfur and energy balances per unit, kinetics where they exist, and the lags between units.
What goes in: wood species and chips, recycled fiber, cooking liquor and purchased chemicals. Changes here show up hours to days later downstream.
What the mill must make: grades, specs, rates and prices. The same machine is easy or hard to run depending on the slate.
The historian, lab results and event records. It tells the model which simulated mill yours is.
What is installed: the flowsheet, unit types and sizes, and how they are connected.
Removes water: mechanically in the presses, then with steam in the dryer section, where most of the machine's steam goes. Every point of water the presses leave behind costs steam.
The forward model plus a range for everything unknown. Each sample is one complete simulated mill, run for weeks of simulated time.
What you ask: a setpoint change, a goal, or a plan over a horizon.
Burns concentrated liquor to make steam and recover chemicals. Usually the mill's bottleneck.
The knobs. The search space Recommend explores; each one is a mechanism in the forward model.
Ties the mill together and turns it into dollars: steam headers, turbines, chemicals, energy and fiber.
One network learns, across all simulated mills, how mills behave and what changes do.
Separates pulp from spent liquor, then brightens it. Washing efficiency decides how much soda and organics go forward instead of back to recovery.
Refines and blends fiber, adds chemistry, and feeds the headbox. Fast loops and the least transferable models.
The trained network. Every question is one forward pass over the mill's recent history.
Millions of synthetic mill histories. Each is a different simulated mill, run for weeks.
Numbers are illustrative.
Each trap is a test file with the answer planted first, told the way it shows up in a mill. Click a card for the trap, its impact, what the model did, and the test files.
Historian tags, lab results and event records, exported as they are. We return first answers in a working session. Nothing is installed on the line.
The test files are synthetic, with the answers planted in advance. Nothing here is a result on mill data.
Tell us which units and questions matter most and we will be in touch.