Key Takeaways
- Calculated mineralogy models and geometallurgical domains are, in practice, regression and classification problems trained on automated-mineralogy data. They are not distinct disciplines requiring separate tooling.
- QEMSCAN, MLA and TIMA supply the ground-truth mineral abundance and texture data that feeds the models.
- scikit-learn and XGBoost are the standard open-source libraries for the modelling layer. Teams commonly build recovery, hardness, throughput and mineral-abundance predictions as gradient-boosted regressions.
- Leapfrog Edge, Seequent’s resource estimation extension to Leapfrog Geo, folds these predictions back into the same 3D block model as your geology, so geomet variables interpolate alongside grade.
- This is confirmed, current practice, not speculation.
TL;DR
To predict geometallurgical domains, train a gradient-boosted model (XGBoost or scikit-learn) on your automated-mineralogy results and metallurgical tests. Then interpolate its predictions into your block model through the Leapfrog ecosystem. Planners then see hardness or recovery next to grade in every block.
What Are Geometallurgical Domains?
Geometallurgy links the geology of an ore body to how it behaves in the plant. Two rocks with the same copper grade can grind, float and recover very differently. The difference usually comes from mineralogy: which minerals carry the metal, how fine they are and what surrounds them.
Geometallurgical domains group the deposit into volumes that behave consistently in processing. Each domain has its own expected hardness, recovery and throughput. As a result, mine planners can schedule ore by plant performance, not just by grade. Calculated mineralogy models feed those domains by estimating mineral abundance everywhere, not only where samples were scanned.
How Do I Predict Mineralogy and Geometallurgical Domains With AI?
The standard workflow is a pipeline with three steps.
Collecting automated-mineralogy training data
First, assemble labelled data. The labels come from automated mineralogy with QEMSCAN, MLA or TIMA (mineral abundance, texture, liberation). Add whatever comminution and flotation testwork you have run, such as BWI, DWI and recovery tests. These results exist only at a limited set of drillhole and sample locations.
Training models for geometallurgical domains
Next, train a supervised model, typically a gradient-boosted regression using XGBoost or a classifier from scikit-learn. The model learns how cheaper, more abundant inputs relate to the expensive targets. Those inputs include assay grades, alteration codes and hyperspectral signatures. For example, a model might learn that a certain clay signature predicts low recovery.
Carrying predictions into the block model
Finally, interpolate the predictions spatially. Leapfrog Edge can estimate geometallurgical variables through the same block-modelling pipeline as your grade estimates. A mine planner therefore sees predicted hardness or recovery right next to Cu grade in the same block. The same approach underpins block model machine learning for geotechnical and recovery variables.
How Reliable Are Geometallurgical Domains From ML?
This is one of the more mature AI applications on this site. It is documented, current practice with real published case studies. Industry literature reports R² figures around 0.89 for recovery prediction models. The practical payoff is that you no longer need to physically test every block.
The main caveat is data quality. These models are only as good as your automated-mineralogy sample coverage. Sparse or unrepresentative sampling produces a model that extrapolates poorly. Therefore, check that samples span every lithology and alteration type. Clustering-based geological domaining can help you spot gaps before you train.
FAQs
How are geometallurgical domains defined?
Teams define them by grouping rock that behaves similarly in processing, based on mineralogy, hardness and recovery. Machine learning predicts those properties across the deposit from denser data such as assays and alteration logs. Geologists and metallurgists then review the resulting zones together.
What data do I need to predict geometallurgical domains?
You need automated-mineralogy results (QEMSCAN, MLA or TIMA) and metallurgical tests such as BWI, DWI and recovery. You also need dense covariates at the same locations, such as assays, alteration codes and hyperspectral data. Representative sampling across all rock types matters most.
Which ML models work best for mineralogy prediction?
Gradient-boosted regressions (XGBoost) are the most common choice, alongside scikit-learn classifiers and regressors. They handle mixed numeric and categorical inputs well. Always compare models on held-out samples before choosing one.
Is predicting geometallurgical domains with ML proven?
Yes. It is established practice, with published case studies reporting strong accuracy for recovery prediction. Results still depend heavily on sample coverage and data quality at each site.
Try It With Geocluster
Wiring together mineralogy data, ML regression and your block model is a multi-step workflow. It benefits from a dedicated research harness rather than a pile of disconnected notebooks and scripts. Geocluster helps you reason across that whole geomet-modelling pipeline. Read how the Geocluster Research Harness works to learn more.