Key Takeaways

  • “Real-time geometallurgy” is the umbrella term for fusing ore characterization proxies (grind/crush hardness proxies, mineral liberation, %solids) into live setpoint recommendations for the plant, rather than waiting for periodic lab testwork.
  • It runs on the same closed-loop AI platforms used for crushing/grinding optimization — this is an application pattern, not a separate piece of software.
  • The core idea: instead of running a static geomet model built from quarterly composite samples, you stream ore-property predictions into the control room as material actually arrives at the plant.
  • This is commercially available today via general-purpose industrial AI/APC vendors, not a mining-specific off-the-shelf product — expect a systems-integration project, not a shrink-wrapped purchase.

TL;DR

Real-time geometallurgy means piping your ore characterization proxies (hardness, liberation, clay content, %solids) into a closed-loop AI/soft-sensor platform like Imubit or ABB Ability for Mining so the plant adjusts itself to the ore in front of it, instead of running on a fixed setpoint tuned for average ore.

How Do I Apply AI to Real-Time Geometallurgy?

Real-time geometallurgy is less a single tool and more an integration pattern that sits on top of everything else in your geomet and plant-control stack. The inputs are the proxy measurements you’re likely already generating elsewhere — blasthole hyperspectral/XRF grade data, hardness and clay proxies from core scanning, ore liberation estimates from automated mineralogy — and the output is a continuously updated prediction of how the ore currently entering the mill or flotation circuit will behave.

The AI layer that makes this “real time” rather than “quarterly report” is the same class of soft-sensor platform used for grinding and flotation optimization: Imubit and ABB Ability for Mining both build models that continuously ingest live plant and lab data and translate it into setpoint recommendations, and vendors are increasingly framing this specifically as “real-time geometallurgy” or “dynamic ore characterization” in their mining pitch decks. The pattern is: characterize ore upstream (drilling, sampling, sensors) → stream those proxies into a data historian → feed a model trained to map ore properties to optimal plant behavior → write setpoints back to the control system, ideally with the same shadow-mode-then-closed-loop rollout used for standalone grinding/flotation AI.

Be honest with stakeholders that this is still an emerging integration pattern rather than a single named commercial product you can buy off a price list — you’re assembling it from your existing geomet data pipeline plus a general industrial AI/APC platform, which means the project looks more like a systems-integration effort than a software purchase.

Try It With Geocluster

Stitching together which of your existing ore-characterization data streams (hyperspectral, XRF, automated mineralogy) are actually good enough real-time proxies — and which AI/APC vendor’s approach fits your plant’s control architecture — is a research problem before it’s an engineering one. Geocluster is built to help you work through exactly that kind of multi-source technical evaluation.