AI And The Digital Mine

• Read original article here

When you think of the words “data” and “mine”, no doubt the idea of data mining comes first. However, just as much as we find value in mining the rich resources of data, so too can we apply the advanced techniques for dealing with data to real-world mining — that is, extracting natural resources from the earth. The world is just as dependent on natural resources as it is data resources, so it makes sense to see how the evolving areas of artificial intelligence and machine learning have an impact on the world of mining and natural resource extraction.

Mining has always been a dangerous profession, since extracting minerals, natural gas, petroleum, and other resources requires working in conditions that can be dangerous for human life. Increasingly, we are needing to go to harsher climates such as deep under the ocean or deep inside the earth to extract the resources we still need. It should come as little surprise then that mining and resource extraction companies are looking to robotics, autonomous systems, and AI applications of all sorts to minimize risk, maximize return, and also lessen the environmental impact that mining has on our ecosystem.

On a recent AI Today podcast episode, Antoine Desmet of mining technology and equipment company Komatsu shared how they’re using advanced forms of AI, automation, and robotics to make an impact on the organization's operations. Antoine has an interesting background, starting his career as a telecom engineer and receiving a Ph.D in neural network engineering. After getting his Ph.D, he returned to Komatsu and started working in surface analytics. He states that at the time there was a lot of data to work with, but very few analytics in place. He decided to start implementing machine learning and in the last few years his company has seen significant growth through this approach, with his data science team growing from just one person to ten people. 

The role of machine learning in mining

The mining industry uses a lot of big, expensive machinery to perform a wide range of operations at the mine site as well as farther away when the materials need to be processed. Much of this machinery has many sensors that provide large volumes of data that give insights into how the very expensive machines are operating, the conditions in which they operate, and also insights into their performance on specific tasks. Keeping machines up and running is essential to making sure that the mining operation can continue. Any downtime or unnecessary maintenance will result in significant cost and complications for the mining operation. 

Prior to the use of machine learning to help give greater insights into operations, the information coming in from the sensors was just feeding into the control loop and wasn't really being used to provide any sort of pattern identification or predictive analytics value. The use of machine learning has resulted in a shift in how that information is used. By storing and continuously evaluating the huge volumes of sensor and other operational data, the organization can get significantly better insight into what problems are potentially occurring, the evolution of how those problems are occurring, and patterns that can lead to problems down the road. 

To extend the mining metaphor further, Antoine explains the idea of “data dredging” where they stir up all the information they have and start looking for correlations and patterns, and whenever disruptions in the patterns are noticed, they start looking for the reasons why. They were able to analyze their approach and discuss with people around the office on how to come up with a better way to gather this information, analyze the patterns, and uncover even more useful information through a collaborative approach with the machines and the human workforce.

In addition, the use of more advanced analytics prevents the need for humans to have to travel deep into the mine in potentially dangerous situations to try to evaluate problems and determine what is happening. Furthermore, the use of predictive analytics enables more strategic and efficient operations from maintenance to purchasing equipment.

Resource extraction is inextricably linked to the environment around it. It’s near impossible to take something out of the earth without impacting the nature and topology surrounding that extraction. However, it is possible to minimize that impact, and machine learning and AI are helping to do just that. Antoine explains that there is very little benefit to dig in the wrong place or extract resources inefficiently. Machine learning systems are helping to analyze geology and topography with greater precision to identify what exactly you’re digging for and find the most optimal way in which to extract it, maximizing the benefit for the operator and minimizing the overall environmental impact. 

Other uses of machine learning and AI can be applied to using drone footage or satellite imagery to keep a constant watch on waste and output piles to make sure that they are keeping in compliance with environmental regulations as well as minimizing any potential safety hazards. Rather than just spot checking these locations on a regular basis, these AI-enabled computer vision systems can keep a constant watch, spotting potential problems, and alerting management to solve problems before they become hazardous. 

As part of the use of machine learning, Antoine mentions that you have to be careful to avoid human bias, especially when training machine learning models. As a result they’ve also learned to be careful when using humans as models or reference points. In the podcast, Antoine cites a time when the machine operators were analyzed to see how exactly they were running their machines. A machine learning model was then trained around the data collected from this information. However, what they realized is that the model was not performing in an optimal way. They realized that one operator didn’t perform the work like everyone else, and as a result, the model they were using was biased. It helped them to see where error and bias can occur and how best to handle such situations.

They also saw the value of “augmented intelligence”. Rather than replacing the human with an autonomous counterpart, they were able to use the machines to give insights to humans on how to perform their tasks more efficiently. They now see AI as providing a balance where machine learning and human ingenuity cross paths and can be used to help improve one another.

In the podcast, Antoine explains that some of the more tedious tasks that require looking at mountains of information can become very tedious for the human workers, so that type of work is typically left to computers. Once the collection and analysis of the data is done by the machine, it is then handed off to those same workers so that they can spend their time doing higher value tasks with it. It helps them to understand the moving parts as a whole.

To address issues of data access and network connectivity to machines in remote places, Antoine explains that the network is designed for sporadic, remote access. Information is kept on-site and evaluated as close to the edge as possible, with results provided locally to those who need to act on the intelligence and insights. This approach helps them to avoid latency issues so problems aren’t getting away before they can be discovered and it also makes things much safer.

Overall, the use of AI and machine learning is helping to make mines more efficient. While we might want to see one day in the future where we will not have to dig into the earth to extract the resources we need to live and exist on this planet, that day is not here yet. As a result, AI and machine learning are helping to make mining operations more efficient and less impactful on the environment.