Studies show that Machine Learning (ML) techniques, when integrated with traditional geostatistical techniques, save time, reduce costs, and improve prediction accuracy.
In our last post we examined various approaches data scientists are exploring to develop Machine Learning tools for resource estimation. Today, we'll look at early Machine Learning applications and pilot studies and how they might affect the mining industry in the near term.
Over the past 15 years new mineral deposits have become increasingly difficult, expensive, and risky to discover. In response, engineers and data scientists are developing new geological data-collection tools that generate enormous, continuous volumes of data that geologists and mining engineers can use to improve log drill cores, predict lithofacies from geophysical logs, predict mineralization, and optimize drilling and exploitation.
The challenge is integrating all this data with traditional geostatistical resource estimating techniques.
Hybrid Approach
Geologists currently use two common methods for geology interpretation. One requires interpretation of geology in sections by digitization. From these sections, a 3D model is constructed using triangulation. Another is implicit modelling (based on radial basis function). It has become widely used to construct orebody models by interpolating drillhole data directly.
For resource modeling, most geostatistical algorithms have already been automated in modern mineral resource software packages. The industry commonly uses simple univariate geostatistical models such as Inverse Distance Weighting and Kriging for resource prediction.
Data scientists are developing ML to integrate more data from multiple sources into traditional geostatistical models to find, evaluate, predict, and map subsurface rocks. ML models have the potential to dramatically cut drilling costs, ensure early availability of geology data, reduce human error, increase prediction accuracy, slash overall costs, and shrink exploration lead time.
More Data, Better Models
Machine Learning is beginning to be deployed in some limited mining applications. UK-based Ionic Engineering has deployed machine learning to enhance image recognition used to identify copper grades, a recent Deloitte/NORCAT report said. By training the machine in parallel with a person using neural networks to learn characteristics the operator is looking for, operation time and product quality are improved. Canadian mining consulting firm Shyft Inc. tapped more than six years of data to develop Machine Learning algorithms to forecast energy peaks and artificial intelligence (AI) to enhance automatic ventilation systems control to reduce energy costs in underground mines.
For resource estimation, researchers are developing various ML approaches aimed at integrating more data with traditional geostatistical models.
In one study, Machine learning as a tool for geologists, scientists used six ML algorithms to predict the presence of gold mineralization in drill core from geophysical logs acquired at the Lalor deposit in Manitoba, Canada. Results show that the integration of a set of rock physical properties (measured at closely spaced intervals along the drill core with ensemble machine learning algorithms) allows the detection of gold-bearing intervals with an adequate rate of success. Since the resulting prediction is continuous along the drill core, this type of tool helps geologists select sound intervals for assay sampling and model more continuous ore bodies during the entire life of a mine.
In another study published in August 2021, Quantifying Mineral Resources and Their Uncertainty Using Two Existing Machine Learning Methods, researchers compared two machine learning methods (multiple linear regression and a multilayer neural network) to generate tonnage curves and their confidence intervals (CIs) directly from the data. The results showed no significant differences between the two methods and validated the effectiveness of the approach for tonnage prediction and uncertainty quantification. Also, ML resource predictions outperformed those obtained with ordinary kriging, constrained kriging, uniform conditioning, and indirect lognormal correction.
In, Integration of Machine Learning Algorithms with Gompertz Curves and Kriging to Estimate Resources in Gold Deposits, researchers sought to refine a novel Machine Learning approach (GS-Pred) that incorporates network analysis for geology-based anomalous data detection and outlier removal. In this application, ML algorithms were integrated with sequential-kriging block modeling for high resolution in situ grade estimation using data from the auriferous conglomerates of the Witwatersrand Basin in South Africa.
“Our algorithms feature fast data processing, geology- and assay-based outlier detection, visualization of complex geospatial data, and they open new avenues for intelligent and automated in situ Au-grade prediction,” the researchers reported. “We demonstrate that GS-Pred target predictions are feasible substitutes for assays for the purpose of block modeling under suitable deployment conditions.”
In a 2021 paper published by the open access academic publishing platform MDPI, Machine Learning—A Review of Applications in Mineral Resource Estimation, editors Nelson K. Dumakor-Dupey and Sampurna Arya surveyed scores of peer-reviewed studies of Machine Learning techniques since 1993 and concluded that, while the technology represents a powerful tool for resource estimation, integrating it with traditional geostatistical methods will take time.
It will also require that companies facilitate that integration. In our next and final post in this series we will examine hurdles and opportunities mining companies face implementing Machine Learning for resource estimation.
▶ Share this article on LinkedIn or Facebook.
Min Liang is a Geostatistician and a Data Scientist. She holds a PhD in Geostatistics from the École Polytechnique de Montréal and records 6 publications on geospatial modelling and simulation. Before joining Dassault Systèmes in 2019, Min worked 1 year in Data Sciences and 3 years as an Environmental Consultant
Access all the technical articles:
machine learning resource estimation optimization AI_for_resource_estimation series
