Key Takeaways
- Crusher optimization is one of the more measurable AI wins in mining: reported gains of 1 to 5% in throughput and 10 to 15% in energy use from AI setpoint optimization.
- Imubit’s Closed Loop AI Optimization uses reinforcement learning to write optimal setpoints directly to control systems in real time, rather than relying on a static linearized process model.
- ABB Ability for Mining is the more established alternative. It depends more on linearized process models, which can struggle with the genuinely non-linear behaviour of crushing circuits.
- This is a production capability with measured results at deployed sites, not a lab demo.
TL;DR
Crusher optimization with AI means closed-loop control. Imubit uses reinforcement learning to adjust setpoints continuously against live plant data. It reports real throughput and energy gains over traditional control approaches such as ABB Ability.
What Is Crusher Optimization?
Crushing is the first stage of size reduction. Primary, secondary and sometimes tertiary crushers break run-of-mine rock down to a size the grinding mills can handle. Crusher optimization means running that circuit to get the most useful product for the least energy and wear. The levers include feed rate, crusher gap setting and how the circuit is loaded.
Crushing circuits are hard to control well by hand. Feed rate, gap setting and ore hardness all interact non-linearly. Operators have historically tuned setpoints from experience and periodic manual adjustment, not continuous optimisation. The gap between what the circuit could achieve and what manual tuning achieves is where AI earns its keep.
How Do I Optimize Crushing With AI?
There are two main approaches, and the choice between them comes down to how each one models your circuit.
Reinforcement learning with Imubit
Imubit’s Closed Loop AI Optimization is built around reinforcement learning. The system learns from actual plant data and writes optimal setpoints directly back to the control system in real time. It keeps adapting as ore conditions change, rather than running a fixed rule set. This matters because crushing circuits drift. Feed hardness, particle size distribution and equipment wear all shift the best setpoint over time. A reinforcement-learning agent that keeps interacting with the live process can track that drift in a way a static model cannot.
Model-based optimisation with ABB Ability
ABB Ability for Mining is the more established incumbent. It offers broader digital mining services beyond crusher optimization. However, its process-optimisation approach tends to lean on linearized models of the circuit. That is a reasonable simplification for many control problems. In crushing and grinding circuits, though, the underlying physics is genuinely non-linear, so a linear model can leave performance on the table.
Choosing a crusher optimization approach
The practical question is less “which brand” and more “does this vendor’s approach actually model the non-linearity in my circuit?” Ask each vendor how their model handles changes in ore hardness and liner wear. Then ask for results from circuits similar to yours.
Results and Rollout for Crusher Optimization
Reported results from concentrators using this approach include throughput increases of 1 to 5% and energy cuts of 10 to 15%. At plant scale, those are meaningful numbers. As with any closed-loop system, expect a staged rollout. The model first recommends setpoints in shadow mode, and operators take over less often as trust builds.
Crushing does not work in isolation either. Crusher product size sets the feed for the mills, so gains here compound in grinding circuit optimization. Changes in ore hardness also show up in crushing first, which is why real-time geometallurgy often starts at the crusher.
FAQs
How does AI crusher optimization improve performance?
AI adjusts crusher setpoints such as feed rate and gap continuously as ore hardness and wear change, instead of relying on periodic manual tuning. It learns from plant data how each setting affects throughput and energy. As a result, the circuit stays closer to its best operating point for longer.
What data does crusher optimization need?
It needs continuous plant data: feed rate, crusher power draw, gap setting, product size where it is measured, and equipment status. Ore hardness information from upstream, such as blast or geomet data, makes predictions better. The data has to be logged in a plant historian so the model can learn from it.
Is AI crusher control better than traditional process control?
It depends on the circuit. Model-based control works well where behaviour is close to linear. Crushing circuits are often strongly non-linear, which is where data-driven, reinforcement-learning approaches report gains. Run a trial or shadow-mode comparison on your own circuit before deciding.
How long does it take to see results from AI crusher optimization?
Most projects start with a shadow-mode period, where the model only recommends setpoints, before moving to closed-loop control. The timeline depends on data quality and how quickly operators come to trust the system. Measured gains usually come after closed-loop control is running, not in the first weeks.
Try It With Geocluster
Benchmarking reinforcement-learning process control against traditional linearized optimisation for your specific crushing circuit is worth doing with real data before committing. The Geocluster Research Harness helps teams run that research systematically. Read more in Meet the Geocluster Research Harness.