AI-enhanced clustering of mine tailings using Geostatistical data augmentation and Gaussian mixture models

Area of Science:

• Earth and Environmental Sciences
• Data Science
• Materials Science

Background:

• Mine tailings reprocessing is crucial for recovering critical raw materials vital for green technologies and sustainable resource management.
• Conventional mineral resource estimation methods struggle with tailings due to their lack of geological structure and irregular spatial patterns.
• Developing effective domain modeling techniques for tailings is essential for optimizing reprocessing strategies.

Purpose of the Study:

• To propose and evaluate an artificial intelligence (AI)-based framework for domain modeling in mine tailings.
• To integrate geostatistical data augmentation with Gaussian Mixture Models (GMM) for defining compact and spatially contiguous estimation domains.
• To enhance the spatial resolution and continuity of geochemical data in tailings deposits.

Main Methods:

• Spatially augmented geochemical data from drillholes using Ordinary Kriging on a 3D grid.
• Applied Gaussian Mixture Models (GMM) with various covariance structures to define estimation domains.
• Evaluated GMM performance using spatial metrics like Moran's I and Silhouette Index, comparing with conventional clustering.

Main Results:

• The AI framework significantly improved spatial coherence (Moran's I = 0.52) and cluster compactness (Silhouette Index = 0.59) compared to borehole-only clustering (0.29 and 0.41, respectively).
• The augmented GMM approach, especially with tied covariance, demonstrated superior performance in domain modeling for tailings.
• Enhanced data resolution and continuity were achieved through geostatistical augmentation.

Conclusions:

• The proposed AI-driven framework offers a robust and scalable solution for domain modeling in mine tailings.
• This method facilitates the design of selective reprocessing strategies for sustainable resource development.
• The integration of geostatistical augmentation and GMM provides an effective approach for analyzing Earth system residues.