Key Takeaways
- Grinding circuit optimization matters because SAG and ball mills are among the largest energy consumers on a mine site. Comminution overall is roughly half of total plant energy use, so small efficiency gains are worth real money.
- Closed-loop AI controllers now write mill setpoints directly, rather than just advising an operator. They use deep neural networks trained on historical plant data instead of a first-principles model.
- Vendors report 2 to 5% throughput gains, lower specific energy, fewer liner strikes and shutdowns, and better downstream recovery from AI-driven grinding control.
- This is a mature, commercially available category, with multiple named vendors actively selling into mining today.
TL;DR
Grinding circuit optimization with AI mostly means adopting a closed-loop advanced process control (APC) platform. The main options are ABB Ability Expert Optimizer, Imubit’s Closed Loop AI Optimization, and Metso’s control stack built on the MillSense charge sensor. Each learns your mill’s behaviour and adjusts feed rate, water addition and mill speed in real time.
What Is Grinding Circuit Optimization?
A grinding circuit takes crushed ore and reduces it to the fine particle size flotation needs. It usually combines SAG or ball mills with hydrocyclones that send coarse material back for more grinding. Grinding circuit optimization means getting the right grind size at the highest throughput and the lowest energy per tonne.
The trade-offs are constant. Push more ore through and the grind gets coarser, which can cut flotation recovery. Grind finer and energy use climbs. Meanwhile, ore hardness changes the balance hour by hour. Holding the circuit at its best point by hand is difficult, which is why grinding is one of the most mature areas for AI control.
How Do I Optimize Grinding With AI?
The practical path runs from instrumentation to control to a staged rollout.
Instrument the mill first
Start with instrumentation, not modelling. Any AI system needs reliable sensor data on mill charge, throughput, power draw and particle size before it has something to learn from. Metso’s MillSense is a wireless sensor bolted to the mill liner. It measures the toe and shoulder angle of the charge directly, which is a much better real-time proxy for what is happening inside the mill than power draw alone. It is often bundled with acoustic and vibration sensing (SmartEar) for a fuller picture.
Choose the control approach for grinding circuit optimization
Once the data is flowing, the control layer is where AI earns its keep. ABB Ability Expert Optimizer applies model predictive control (MPC) across the grinding and flotation circuit. It stabilises feed rate, separator speed and reject rate against a process model.
Imubit takes a more explicitly AI-native approach. Instead of building a first-principles model, it trains a deep neural network on years of historical operating data. The network predicts how the mill will respond to setpoint changes. Imubit then closes the loop, writing optimal setpoints to the DCS in real time while keeping the previous control system as a safety fallback. Imubit has 90+ deployments in refining and chemicals and is expanding into mining. It reports throughput gains in the low single digits, alongside energy and downtime reductions.
Roll out grinding circuit optimization in stages
For a mine site, the realistic rollout has four steps:
- Instrument the mill with charge and vibration sensors.
- Historize the data for 6 to 12 months if you do not already have it.
- Run the AI system in advisory or shadow mode and compare its recommendations with operator judgement.
- Hand over closed-loop control progressively, with operator override always available.
This is not experimental. It is a standard APC modernisation project with an AI-based control core instead of, or alongside, a classical model-based one.
Grinding Circuit Optimization in the Wider Plant
Grinding sits between two other AI opportunities. Upstream, crusher optimization sets the feed size the mills receive. Downstream, grind size drives how well AI flotation control can recover metal.
Periodic mill surveys remain the reference for checking that the circuit and its instruments perform as the model assumes.
FAQs
How much energy does grinding use in a mine?
Comminution, which includes crushing and grinding, typically accounts for roughly half of a processing plant’s total energy use, and grinding takes the larger share. That is why even a few percent improvement in grinding efficiency has a noticeable effect on operating cost.
What is the difference between MPC and AI control for grinding?
Model predictive control (MPC) uses a process model, usually built from engineering knowledge and step tests, to choose setpoints. AI control platforms such as Imubit learn the model directly from historical plant data with neural networks. Both can run in closed loop. AI approaches aim to capture non-linear behaviour that simpler models miss.
What sensors does grinding circuit optimization need?
The essentials are mill power draw, feed rate, water addition, mill speed and a measure of product size, such as a particle-size analyser on the cyclone overflow. Charge sensors like Metso MillSense and acoustic sensing add a clearer view of what is happening inside the mill.
Can AI grinding control run without operators?
AI grinding control can run in closed loop, but operators stay in charge. The existing control system remains as a fallback, and operators can override the AI at any time. Most sites reach closed loop gradually, after a period in which the model only makes recommendations.
Try It With Geocluster
Deciding whether closed-loop grinding AI is worth the instrumentation investment at your mill means digging through vendor case studies, published throughput and energy results, and your plant’s data maturity. That multi-source research is what Geocluster is built to accelerate. Point it at your comminution circuit question and let it pull together the evidence before you write the business case.