Model Predictive Control.

Aeterna's MPC is a true multi-input multi-output controller. It does not stack independent PID loops and hope they don't fight each other. It treats the whole circuit as a single optimisation problem: every manipulated variable is moved against every controlled variable simultaneously, with the coupling between them respected explicitly.

MPC is the most widely deployed form of Advanced Process Control (APC) in mineral processing. Where conventional PID loops control one variable at a time, MPC manages the full MIMO (multiple-input, multiple-output) system simultaneously — anticipating interactions between mill load, power draw, sump level, and cyclone pressure before they cascade.

What the controller moves (MVs)

• Feed rate — tonnes per hour into the SAG
• Mill speed — SAG variable-speed drive setpoint
• Water addition — dilution to the mill and the sump
• Pump speed — cyclone feed pump VSD
• Cyclone pressure setpoint — classifier operating point

What the controller holds (CVs)

• Mill load — held tight against the overload limit, not the safety margin
• Motor power — respected as a hard constraint, never breached
• Product P80 — grind size on target while ore drifts
• Sump level — within band, no flooding, no starvation
• Recycle load — pebble return managed, never choking the SAG

Why MIMO matters

Single-loop control fails the moment two loops want the same actuator. Push feed rate up to chase throughput, mill load rises — the load loop pulls it back down. Add water to manage density, sump level rises — the level loop changes pump speed, cyclone pressure shifts, P80 drifts. The loops fight each other and the operator has to referee.

MIMO MPC eliminates the fight. The QP solver computes one move plan per cycle that respects every constraint and pushes towards the objective. The same move can satisfy three objectives at once because the controller sees the whole coupling matrix.

The optimisation

• Objective: maximise throughput, subject to all constraints
• Constraints: mill load ≤ 96 %, motor ≤ 13.4 MW, P80 ≤ 160 µm, recycle ≤ 380 t/h
• Horizon: 30 minutes look-ahead, 60 second control cycle
• Solver: sparse QP solved every cycle; warm-started from the previous solution

Process optimisation, not just stabilisation

Stabilisation is the floor, not the ceiling. Once the circuit is stable, PROCESSOPT pushes setpoints toward the constraint — more feed rate, tighter grind, lower energy per tonne — and holds them there continuously. That is the difference between Advanced Process Control and a well-tuned PID.