Geometallurgy Characterization

AI for geometallurgy turns hyperspectral and mineralogical data into the models a block model actually runs on. The goal is to estimate how each part of a deposit will behave in the plant before it is mined, rather than discover it at the mill. This category covers the full pipeline: measuring the core, interpreting the data and predicting metallurgical response block by block.

What AI Does in Geometallurgy

Geometallurgy links geology to processing. Clays, alteration and hardness all change recovery, throughput and energy use. AI for geometallurgy helps in three ways. It measures mineralogy densely with hyperspectral core scanning. It interprets those spectra automatically, so specialists are no longer the bottleneck. Finally, it predicts recovery, hardness and other expensive properties from cheap, routine measurements.

Where to Start

Each page below answers one “how do I do this with AI?” question.

Scan the core. Hyperspectral clay models start from Corescan and HyLogger scans that measure mineralogy densely along the core.
Model the clays. Clay mineral prediction turns scanned spectra into a clay model for the block model.
Map the alteration. Alteration proxies combine hyperspectral maps with implicit modelling in Leapfrog Geo.
Predict the plant response. Recovery and hardness prediction uses gradient-boosted models on testwork data. More broadly, geometallurgical proxy models estimate any expensive assay from a cheap measurement.
Keep the model current. Dynamic block models re-estimate blocks as new drilling and production data arrive.
Find more ore. Brownfield exploration reuses a mine’s historical archive to rank nearby targets.

What’s Mature and What’s Emerging

Hyperspectral core scanning is mature, deployed technology. Recovery and hardness prediction with gradient-boosted models is well published and reproducible. Proxy models follow the same proven regression approach. By contrast, alteration modelling from hyperspectral data, dynamic block models and ML targeting for brownfield exploration are still emerging. In those areas, the software exists, but you assemble the workflow yourself.

Sep 21, 2026

Alteration/Proxies Model

Model alteration proxies in 3D: Corescan hyperspectral maps feed Leapfrog Geo implicit surfaces, replacing a geologist's logged intensity scores alone.

Sep 21, 2026

Clay/Geomet Model

Clay mineral prediction with AI: scan core with Corescan or HyLogger, automate first-pass logging with CSIRO's MyLogger and feed the geomet block model.

Sep 21, 2026

Dynamic Block Models

Dynamic block models re-estimate your grade model automatically as blastholes, grade-control samples and production data stream in from the mine.

Sep 21, 2026

Exploration Brownfield

Brownfield exploration with AI: load a mine's drillhole and block model archive into Leapfrog, then add ML targeting to flag extensions faster.

Sep 21, 2026

Hyperspectral Input Models (High Resolution Clay Models)

Hyperspectral clay models start with Corescan HCI or CSIRO HyLogger scans, about 800,000 spectral samples per metre, then automated mineral interpretation.

Sep 21, 2026

Proxy Models

Geometallurgical proxy models predict an expensive assay from a cheap measurement. Build one with Cancha for a turnkey path, or scikit-learn and XGBoost.

Sep 21, 2026

Recovery & Hardness Prediction with Machine Learning

Recovery and hardness prediction with ML: train XGBoost, LightGBM or Random Forest on mineralogy, geochemistry and geomet testwork, or use Cancha.

FAQs

How is AI used in geometallurgy?

AI for geometallurgy scans drill core to measure minerals such as clays, interprets the spectra automatically and predicts properties like recovery and hardness for every block. Planners then use those predictions to schedule and blend ore before it reaches the plant.

What data do you need to start with AI for geometallurgy?

You need historical metallurgical testwork paired with routine measurements on the same samples. Typical inputs are assays, mineralogy, alteration logging and hyperspectral scans. The more ore types the data covers, the more reliable the models become.

Which machine learning models work best in geometallurgy?

Gradient-boosted trees such as XGBoost and LightGBM, along with Random Forest, dominate the published work. They handle tabular geological data well and are easier to explain than deep learning. Neural approaches appear more often in image and spectral interpretation.