AI in mining: how does it actually work?

AI can make better decisions faster, but still needs humans to pose the question. Credit: Parilov via Shutterstock.

Artificial intelligence (AI) has in recent years disrupted global markets and industry in profound and perhaps unexpected ways. The exceptional speed with which generative AI in particular has advanced since last autumn, marked by the launch of ChatGPT in November 2022, has caused industries worldwide to reconsider their assumptions about how they do business. The technology now also sits at the heart of a lot of software development and has, as a result, started to seep into all corners of business management. 

The use of AI in mining specifically, for planning, maintenance, and operations, has capitulated over the last decade as mines become increasingly autonomous to keep up with demand, safety and environmental requirements, and the tireless pressures of efficiency. 

The use of AI-based software comes hand-in-hand with advancements in physical operations, with predictive maintenance and enhanced production capabilities possessing the ability to save miners considerable amounts of money in both the near and long-term. 

So: AI is fast becoming critical to all aspects of mining, from exploration to processing, but how does it actually work? And what are the tangible, quantifiable benefits of these advancements?

In its most basic form, AI is a tool woven into various software platforms used primarily by companies. Here, it makes sorting through the ever-growing amounts of data produced from machinery and equipment operating within mines easier, quicker, and more accurate. It can be used for a wide variety of purposes, and as the technology gets more advanced, real-time data analysis can aid activities such as mine planning, predictive maintenance, energy management, and health and safety improvements in increasingly sophisticated ways. 

“Let's say that you have an underground drilling machine,” says Perry Zalevsky, senior director of industry at software developer Aveva, as we sit down in a boardroom in downtown San Francisco; the setting for our interview. “That machine has sensors, and some computer control on it. Those can collect data in real-time: the temperature of the drilling part, the pressure it is exerting, the vibration of the machine and a number of other data points,” he explains. 

“Now, let's say the machine is being operated by someone remotely in an operating centre. That person is looking at all of this data [within an AI-powered software system], which takes all of this live data being sent from the sensors and computer controls and aggregates it into one place, like some screens, that the operator can see.” Some companies, such as Microsoft or Aveva, have also begun to build language models like ChatGPT into their software platforms, allowing human users to quickly receive written or spoken answers to queries about datasets.

In some instances, Zalevsky says, a new machine might start to fail or malfunction after a few weeks, and the operator can then begin to notice a pattern, relating to pressure change and failure, for example. However, they would also see huge streams of other data coming in constantly. 

The operator “can't really think of five, or 10, or 50 different pieces of data at one time. It's too much for a human,” Zalevsky continues. “So, once they get enough data, and it could be six or 12 months’ worth, or more, they could use an AI-based algorithm to look for patterns, conditions, scenarios to run through millions of possibilities.” Over time, AI-based algorithms can use continual inputs of changing data to learn exactly how a machine or system works and accurately predict outcomes and help make decisions on a huge variety of issues, but humans are still needed on both ends of the process.

“It's still good to have a knowledgeable person on hand to see if the AI tools' suggestions are good,” he goes on, although human error and limited understanding can complicate this. Sometimes, software can detect things that a human cannot, but human judgement is required at the data selection level, and in the programming of AI software itself.

“It might be than an operator needs more data to increase the types of failures the tool can help with. It may be a continuous loop, and this is where you need human decision-making and process engineers.”

So the relationship between human and artificial intelligence relating to the modernisation of mining management remains synergistic; naturally one cannot exist or productively develop without the other. “I really see AI as an assisted way of helping people solve these problems, just much faster and much better than they could otherwise,” Zalevsky concludes.

Safety is often hailed as key reasons for, and benefits of, automation in the mining industry. This trend has become the subject of increasing criticism as labour workforces are removed or replaced by automated machines and AI-based technology, but despite this, examples of AI’s tangible benefits on the safety of workers who do remain within mining can be seen. 

“When a driver goes up and down dips in the road, they rev the engine, and in many cases the driver actually hits their head on the roof of the cab,” says Zalevsky, discussing AI’s role in increasing the accuracy and speed with which roads within mines are improved or repaired. Principally, these improvements come through analysing real-time data from heavy haul trucks as well as geo-data. “There were safety implications to it, and some miners have reduced those types of injuries by around 30% by not having the driver rev the engine, the engine being smoother and not going up and down on the roads so much”. 

Research suggests that advancements in and implementations of digital technologies, including AI, within the global mining industry has saved 1,000 lives and prevented 44,000 injuries. This equates approximately to a 10% decrease in lives lost and a 20% decrease on injuries over recent years. 

The role of AI in mine planning has also fundamentally changed the processes by which drilling areas are selected. Zalevsky says that he used to write mining software, and that data would be painstakingly collected by a person on-site, testing samples every 20 feet within a specified area. “He’d then send me the data and I’d write a model to say ‘here's where we should dig today, and here's tomorrow, and here's the day after’.” In other words, he wrote software that would predict days or weeks in advance where the best places to mine might be. 

“Now, there are companies that have cameras and sensors everywhere, and they can do real-time analysis of the of the whole site,” he continues. This information is then sent back to control rooms and processed by AI-based software platforms, which then suggest where to mine immediately, giving a continuous feed of new, updated information. “You could mine in a specific place today, and then receive more data every 10 minutes, a fresh look at the rest of the field, and I can tell in real-time what to do,” Zalevsky explains.

Improvements to operational and energy efficiency, and by extension a mine’s sustainability credentials, are also key drivers of AI adoption by mining companies. Improved mine planning inevitably leads to reductions in environmental harm relating to exploration, and AI-based tools also hold the potential to aid improvements in sustainable decision-making within mining operations by considering factors relating to land use, communities, and governance. 

A recent study into the potential role of AI in the achievement of the UN Sustainable Development Goals (SDGs), found that AI could enable 93% of SDG targets relating to environmental protection and climate action. Specific benefits from AI could come from the technology’s ability to support modelling of climate change, aid the integration of renewable energy into the mining industry, help preserve biological ecosystems, and automate the identification of waste leaks to preserve or restore degraded land and soil. However, the authors of the study warned that the development of AI in a more general sense needs to be supported by the necessary regulatory insight and oversight to prevent gaps in transparency, safety, and ethical standards. 

Zalevsky asks: “Will we see a lot of AI? Yes, if you think of AI as pattern recognition or machine learning. Can you find the pattern in those scenarios? Can you find the best outcome from all these different combinations? Fundamentally, AI is going to be able to do it all much better and faster than any person ever could. 

“But what are the scenarios that you should be looking at? What is the data that you need? I think people will always be involved in that part of the process,” he says, adding that mining remains “a simple industry, so it’ll take time. Most companies don't go from regular maintenance to AI predictive maintenance overnight”.