Where machine learning fits in mining

Article #2: Where machine learning fits in mining

As my last post pointed out, there is a whole new world of data now available to inform and enhance mining processes, and the next step for the mining industry as a whole has to be figuring how to turn all that new, raw data into useful intelligence — which is where machine learning comes in.

Machine learning is a branch of artificial intelligence (AI) that uses data and algorithms to imitate how we as humans learn: through experience and adaptation until gradually we get better and better at whatever task or skill it is we want to master. Put another way, it is a set of methods for creating models that describe or predict something based on data.

Multiple machine learning pilot projects have already demonstrated that AI techniques based on machine learning can help the mining industry make better, more informed decisions, which means the question now becomes not whether to adopt it, but how to integrate it into our regular processes.

To help with that, it might be good to start with a common understanding of machine learning basics, including more about how it works, as well as how it can be used in mining processes, and its effect on mining professionals.  

Two approaches

Machine learning algorithms can approximate an underlying process that is assumed to exist, and that is responsible for patterns or regularities in the data, using mainly two approaches: supervised or unsupervised learning.

  • In the supervised learning approach, the machine learns from labelled data (data with a known input and output). Using the relationship between input and output, the algorithm generates a function capable of making predictions for future input data, something that is particularly useful for problematic variables that are not easy to model by conventional methods. Supervised learning techniques include:
    • Regression models, a conventional predictive modelling technique that analyses how a target or dependent variable relates to an independent variable in a dataset
    • Bayesian learning, to create models that estimate the likelihood that a given sample belongs to a specified class (category), being fast, robust and easy to use, and
    • Deep neural networks that mimic the human brain, where each neuron is responsible first for solving a small part of the problem, then for passing on what they have learned to the other neurons in the network, until the interconnected nodes are able to solve the problem and give an output. 

In the unsupervised learning approach, there is no need to have labelled data because here the objective is not to make predictions but rather to perform an exploratory study to recognize global and local structures in the data. Techniques used in this approach include:

  • Clustering algorithms, which help to speed up the organisation of large datasets, without any human input, by grouping similar data together, and
  • Principal Component Analysis, which reduces a large dataset into a smaller one while maintaining most of its variation information, making it both more manageable and simpler to use.

By combining techniques from both approaches we can generate ensemble models that can recognize patterns in disparate databases from a wide variety of sources and provide rapid interpretation of large data sets that no single person or team could match.

Machine learning and mining processes

With major discoveries of near-surface mineral deposits declining globally and deep underground mineral deposits becoming harder to reach and smaller in size, machine learning — with its ability to take advantage of large amounts of (often-unexploited) data — can generate faster and more accurate geology and resource assessments than traditional methods.

Its predictive abilities mean it can also be used, for example, to quickly develop accurate models predicting both rock type and the economic feasibility of drilling in specific locations, or to collect geological data from the extraction point to predict the effect it will have on the recovery of the flotation process.

The result is increased knowledge of our mining processes, faster decisions with less risk, higher return on investment, better health and safety, greater efficiency, and lower environmental impact.

Machine learning and miners

Machine learning does not replace mining professionals. Rather, it is a way to help mining professionals make as accurate decisions — as certain — as possible.

In Article #3, I will present some use cases for applying these solutions along with successful applications in big mining to illustrate how mining professionals can apply machine learning to improve the reliability and efficiency of mining processes, thus saving time and money. The final article in this series will include detailed tips for using machine learning to solve real mining problems.

Related posts: