AI USE CASE
Crusher and Mill Circuit Optimization
Reduce energy costs and maximize ore throughput in comminution circuits using reinforcement learning.
What it is
Reinforcement learning agents continuously tune crusher and mill operating parameters — feed rate, speed, pressure — to minimize energy consumption while maximizing throughput. Industrial deployments typically achieve 5–15% reductions in energy per tonne and 8–20% gains in circuit throughput. IoT sensor data from the comminution circuit feeds the control loop in real time, enabling autonomous adaptation to changing ore hardness and feed characteristics. The result is lower operating costs per tonne processed and reduced wear on grinding equipment.
Data you need
Historical and real-time IoT sensor data from crushers and mills — including feed rate, power draw, pressure, vibration, and particle size measurements — covering at least 6–12 months of operations.
Required systems
- data warehouse
Why it works
- High-quality, calibrated IoT instrumentation across the entire comminution circuit before model training begins.
- A phased rollout starting with a digital-twin simulation environment to validate the RL policy before live deployment.
- Close collaboration between data scientists and experienced process metallurgists to define reward functions and operational safety constraints.
- Change management programme to build operator trust in autonomous recommendations, with clear override and audit trails.
How this goes wrong
- Sparse or noisy sensor data prevents the RL agent from learning a reliable control policy.
- Process engineers distrust the autonomous recommendations and override the system too frequently, preventing the agent from converging.
- Ore type variability is too high and not captured as a feature, causing the agent to make suboptimal decisions during unrecognised ore conditions.
- Integration with legacy PLC/SCADA systems is too rigid, blocking real-time actuation of agent recommendations.
When NOT to do this
Do not deploy this in a mine that lacks reliable real-time sensor instrumentation or where SCADA systems cannot accept external setpoint commands — the agent will be flying blind and unable to act.
Vendors to consider
Sources
This use case is part of a larger Data & AI catalog built from 50+ enterprise transformation programs. Take the free diagnostic to see how it ranks against your specific context.