In this last article from the “AI for Resource Estimation” series, we examine hurdles and opportunities mining companies face implementing Machine Learning for resource estimation.
The challenge for mining companies is not about adopting the latest technologies. The challenge is retooling organizations to use new solutions.
Currently, Machine Learning (ML) techniques for resource estimation are complementary tools to improve prediction confidence. ML modeling requires volumes of data from multiple sources, which requires companies to have a robust digital backbone in place to handle the data requirements.
Even with digital infrastructure, acquiring data is time-consuming and expensive, and gathering consistently high-quality data needed to train models is difficult. Also, there may simply be not enough data available, say for example, at the beginning of exploration programs or greenfield projects. A company may decide that only brownfield projects are best for early ML applications.
Moreover, because ML models “learn,” this unique field of Artificial Intelligence (AI) creates new information continuously, which presents mining companies with an ongoing challenge of integrating ever-expanding data-driven insights into operational decisions. A company might then ask, “What does our organization need to do to fully leverage this asset?”
Training the Data
Today, most mining companies still use legacy technologies and methods to drive operations deeper underground for scarcer ore reserves of declining grade, even as productivity slumps in the face of continual improvement. Machine Learning and other advanced technologies can help, and while promising, these digital technologies have only been developed for mining applications in recent years, and companies are just beginning to learn how to implement them effectively.
Early on in this series on Artificial Intelligence (AI) for Resource Estimation, we discussed the general field of AI and its sub-category of Machine Learning, numerical methods (algorithms) that facilitate predictions for a set of data based on example sets of data. These example sets of data are used to teach, or “train,” the algorithm how to make the predictions. A trained algorithm can process large amounts of data rapidly and provide an interpretation in minutes in a process that would simply be impossible for a human to accomplish.
The challenge, however, is that ML is dependent on the training dataset, which requires large volumes of data that are clean and well-organized in a true digital format. Most datasets in mineral exploration are not set up as such. Using ML methods also run the risk of introducing human bias if the data is not adequately understood and predictions are not properly interpreted, Antoine Cate, explains in Machine Learning And Artificial Intelligence For Mining Geoscience. For example, what might happen if earth science data from satellite imagery were used without fully understanding how to train the algorithm to differentiate between images of porphyry deposits in a cordillera and images of already exploited deposits in an open pit operation?
Continue reading the full exclusive article in the GEOVIA User community
The GEOVIA User community: a community by and for GEOVIA users
Discover the community of reference for GEOVIA peer-to-peer support, materials and expert insights!
- Get started and make the most of your GEOVIA software
- Become an expert in your domain of interest
- Share your knowledge and experience
- Get help from experienced users
- Discover the Virtual Mine Experience
New member? Create an account, it’s free!
Learn more about this community HERE.
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.