How Machine Learning is Disrupting the Mining Industry

March 2, 2018

Examples of leading companies’ application of AI-based mining technology

A report from McKinsey & Company calculated that, by 2015, worldwide mining operations had become as much as 28% less productive than in the previous decade. On a worldwide scale, the McKinsey report concludes that the general impact of moving material more productively could reach an annual $370 billion by 2025. Other experts agree that productivity is now a major risk: EY’s 2016 risk report now lists productivity as the third most critical risk for mining operations.

It has become well known that the mining industry is facing challenges such as declining ore grades in mature mines, longer haul distances, and increasing development times for new mines. At the same time, we’ve entered an era of data abundance, low computing costs, and open access to the equipment and literature that allow mining operations to take advantage of machine learning (ML), often in tandem with other forms of applied artificial intelligence. Advances in computing technology (GPU chips and cloud computing, in particular) are enabling engineers to solve problems in ways that weren’t possible before. With the rise of big data, large unstructured datasets are abundant, with ML models growing more accurate as data grows richer.

Because mining companies produce commodities in large volumes, small improvements in efficiency can make substantial impacts on a company’s bottom line. Today, the mining industry is heavily focused on efficiency, applying ML to enhance operations at all levels. Here are several current examples:

Avoiding overload events

PETRA Data Science first teamed up with Newcrest Mining in 2016 to solve a problem with the mills in Newcrest’s gold mining operation in Lihir. In this case, PETRA used machine learning algorithms to predict and subsequently avoid overload events in the semi-autogenous grinding (SAG) mills, with the end goal of reducing mill downtime.

To create their model, the data scientists started by looking at one year’s worth of data from the mill (with variables like power, speed, energy consumption, and control parameters) in order to find indicators of an overload event. They then created a machine-learned model—a set of algorithms that they trained to describe in a mathematical formula the indicators of an overload. The model predicts the probability of an overload event an hour in advance, sending a warning to operators so that they can take action to prevent it.

Dr Penny Stewart, managing director and principal at PETRA Data Science, explains the impact of the project: “Obviously I can’t give exact figures relating to this case, but I can say that preventing one or two downtimes paid off the cost of the project.”

Predicting mineralization

Machine learning is a powerful tool for prediction tasks: machine learning software learns from an ongoing feed of new data that (ideally) enables the model to make more accurate predictions over time, even in the face of changing conditions. Several companies are using machine learning software to make predictions on mineralization. For example, Goldspot Discoveries Inc. has been testing machine learning software in new exploration. Their first test predicted 86% of the existing gold deposits in the Abitibi gold belt region of Canada using geological, topography, and mineralogy data.

Another notable case comes from Goldcorp, who has teamed up with IBM Watson to review large quantities of geological information to find better targets as the software learns to understand millions of core samples, 3D models, seismic surveys and geological data.

Sorting raw materials

As part of the EU-funded I²Mine project, TOMRA Sorting Mining has developed an underground mineral sorter that uses a combination of X-ray transmissions or near-infrared sensors examine materials moving through the line, sorting them based upon desired criteria. This color-sorting equipment resulted in 12% less mass needing to be moved (this means less fuel and fewer truckloads are required).

Sensor-based sorting (SBS) technology is not new to the mining industry; mining operations already employ many varieties of sorting technology that use optical color sensing, near-infrared spectrum classification, x-ray transmission, electromagnetic sensing, and laser scanning. Now, with the addition of machine learning, sorting systems can be enabled with self-learning software that will grant them versatility for sorting diverse materials and the ability to self-optimize as the software fine-tunes its parameters.

Increasing automation and autonomy

In a progression towards fully automated, intelligent mines, several companies (including Rio Tinto, BHP, Stanwell, Suncor, and Fortescue) have begun using autonomous haul trucks at their mines. Industrial vehicle manufacturers Komatsu, Caterpillar, and Hitachi have been developing these driverless haul trucks in close collaboration with mining operations, employing a combination of wireless communication, object-avoidance sensors, on-board computers, GPS systems, and artificial intelligence software that enable the trucks to operate autonomously (and in some cases, to be controlled by an operator from a remote control panel).

Rio Tinto, who has employed a fleet of roughly 400 Komatsu haul trucks in its Pilbara mine, explains that the autonomous trucks have improved safety and cut costs by nearly 15 percent, partially due to the fact that the vehicles can be operated 24/7.

Volvo has also recently announced testing on a fully autonomous underground truck in the Kristineberg Mine. The new truck navigates the very narrow tunnels of the underground mine. This is especially innovative because GPS doesn’t work underground as it does for autonomous trucks on surface mines. Volvo explains that these trucks increase productivity and safety: as they are driverless, they can work directly after blasting occurs, instead of having to wait as required by the current protocol.

©2010 - 2018 by Tribal Interactive Inc.