Change Your Browser
You´re using a web browser that isn´t supported by this site. To get a better experience, please use another browser of your choice.
New tools for tunnels: how LKAB is putting AI in production
LKAB is using AI to improve the geological mapping of the rocks in their mine. To do so, they have put together a team of IT experts, geologists and mine operators alongside data scientists from Combient Mix. “We do not want to make it a project, rather we want to put AI in production, which is very rare in our type of industry. I think this could be interesting for the other companies to hear about,” says Håkan Tyni Cybersecurity and Enterprise Strategist at LKAB.
The problems with mapping mine walls in real-time
LKAB is mining iron ore in their underground mine in Kiruna, Sweden. They use explosives to blow tunnels into the iron ore body, operating over three shifts and on multiple locations at the same time.
Geologists are also working in the mines, helping the operators to understand the quality of the iron ore and other types of rocks during the tunnel creation. The geologists take their cars and drive down into the mine’s newly blasted tunnels, and perform a task called front mapping. This means that they describe what type of rock types they see, such as iron ore, waste rock and cracks. As the mine runs during three shifts and almost twenty four-seven, it would require either a large number of geologists working inconvenient hours and in shifts, or they would only be able to describe a fraction of the walls available. “When you are building a tunnel, a wall removed is the same as data disappearing. You want to be able to get a picture of every step taken, meaning you have to catch the walls in real-time, before they disappear and the data disappears with them,” says Celine Debras, Mine Geologist at LKAB.
To solve this problem, LKAB brought in Combient Mix to develop an AI model that would automatically detect iron ore, waste rock and cracks based on images of the newly blasted tunnel. “We have implemented a deep learning model, known as U-Net. Given an image, it utilizes convolutional neural network layers to predict where in the image the iron ore, waste rock and cracks are. To improve the model over time, an online annotation tool has been set up for the geologists at LKAB to geologically map the new images. These images will then be used to retrain the model,” says Max Fischer, Data Scientist at Combient Mix.
Start by understanding people’s daily work life
Håkan Tyni is working with LKAB’s future planning and digitalization journey, and is one of the individuals committed to put AI into their production. “It actually began with our participation in Combient. We saw what other companies were pushing for, so we started looking into our own organization and how we could use available technology to improve the way we work. Within the IT department, we had already run some artificial intelligence cases, so we knew about the possibilities. On the other hand, we realized we could not just ask people in the mine what their needs were, because they did not know about these possibilities. Instead we would ask about what they were doing, and how we could make their daily work easier. In the case of describing rocks, the problem is that our people working in the mine have a very limited time for doing their mapping, or they are forced to work in shifts, inconvenient hours, and over holidays,” Håkan says.
He argues that AI efforts typically happen in the back office with analytics running in the background, or as a project that ends with a successful pilot that is never being put into production. This time they want to completely change their operations with the help of new AI tools. Apart from IT and geologists, their efforts therefore also involve business developers, managers and operators working in the mine. “We do not even call it a project, it is something we are integrating to the way we work in the mine, which of course also makes it more complex to implement,” Håkan says. “The IT department is naturally working a bit faster compared to how geologists work, but after some time we have managed to define a common framework,” Celine adds.
As the IT department has already done some commercial AI projects, they have realized they typically need both more domain knowledge and more (high-quality) data than first expected. “We know Combient Mix has many skilled data scientists, and we wanted to see what results we could achieve with really knowledgeable people. Bringing in Mix was kind of a risk minimization, we wanted to make sure we had the right people onboard,” says Håkan. “My role is to assure project fit, and to make sure that the data scientists’ work is meeting our needs from the perspective of what is geologically possible. A part of the job is also to onboard the team at Mix and introduce them to the world of geology. They are really curious to understand how we work,” says Celine.
An app to make it more rewarding
“If you do not have good images, you do not have the proper knowledge to make predictions around the ore body and draw conclusions for planning and decision making on how to dig the tunnels. So with better predictions, we can save costs. The ultimate goal of this first step has been to check the feasibility. Next, we will try to implement it,” says Håkan.
For a geologist’s day-to-day work, both safety conditions and work hours would improve with this new solution as they would not need to adapt to the mine’s running hours, and furthermore, possibly not even depend on being physically present in the mine. “Even if there are reinforcements of the rocks, stone falls or collapse is always a small risk when mining. In the near future, we might be able to replace people with robots to take the pictures, and control it remotely from our phones. The need for geologists will still exist, we will just need to learn how to work with new tools,” says Celine.
Today, operators need to step out of the car and carry a second phone to take a picture, with no clue what the data tells them. “But now we are putting analytics on the phone. We are developing an app so that our operators can see the results from their photos directly, instead of sending data to a black box. It is gamification, enabling people to track their own achievements,” says Håkan. “I have seen so much automation, the need for knowledge is changing but it does not become obsolete. People move from production to maintenance, we will just work in another way.”
Getting everyone on board for phase two
Håkan and his team are solving a real problem for the operators in the mine, using tools that their colleagues did not know was a possibility. “A clear success factor is the combination between understanding people’s problems, and using our technology toolbox to solve them. If you want to change the ways of working, it is important to get everyone on board from the beginning. The team is really everything, that is why we are moving forward with this initiative,” says Håkan.
“The biggest moment of success was when we got results and found it could actually work, even with so little data. We were all surprised. For me personally, I was so excited to work hands-on with artificial intelligence, and see what is possible. For the team, we have learned to cooperate with others. The geologists have never worked with IT before, now we use each other more and more,” Celine concludes.
While writing the article, LKAB and Combient Mix confirmed their continuation of the initiative, now productizing the proof-of-concept. This will include cooperating with Sigma to develop an android app as part of LKAB’s productization plan.