This is episode number 228, Data Science in Mining.
Welcome back to the SuperDataScience Podcast, ladies and gentlemen, super excited to have you back here on the show today. We are continuing our series of FiveMinuteFriday episodes about data science in various industries.
Today’s industry is mining, a industry that I am personally extremely fascinated by. Mining is a whole different world to what we’re used to. If you’ve ever been to a mining site or even if you just google some images from mining sites, you’ll notice that it’s radically different to what we see in our daily lives. The vastness of these mines is mind boggling, the size of the operations, the size of the machinery is crazy. Some of the biggest machines in the world are used are those which are used in mining. It’s a massive and very broad industry at the same time and involves extremely different things than what we’re used, and that’s why personally I’ve always been so curious about it.
Here’s some background information on mining before we dive into how data science is used to improve mining operations. The total revenue of the mining industry is actually not as high as you’d expect it to be. It’s only about US $600 billion per year, and that represents just under 1% of the global GDP. That percentage varies depending on the region. For instance, in Africa, South Central Asia, Papua New Guinea, Australia and Peru, that percentage would be higher. In other regions that percentage would be lower. But on average globally it’s about just under 1% of the global GDP.
As you can imagine that is not such a large number given the magnitude and the scale of these mining operations. If you compare it for instance with the global retail sector which is about $25 trillion and represents about 31% of the world’s GDP. Nevertheless, mining is extremely important. That’s where we get a lot of the raw materials that are later used in many different areas ranging from jewelry to microelectronics which go into cars and computers and things like that and all the way to coal which is used in producing energy. As we can see, mining is an extremely important industry.
Now let’s have a look at some of the types of mining, so techniques of mining because you might come across these if you’re researching mining or you are maybe applying for jobs in mining or you’re working in mining. There’s two main types of mining techniques: surface mining and subsurface or underground mining. In terms of surface mining, you’ll hear terms such as open-pit mining, quarrying, strip mining, mountain top removal, landfill mining, and high wall mining. In the sense of underground mining, you will hear terms such as drift mining, slope mining, shaft mining, shrinkage, stope mining, long wall mining, room and pillar mining, retreat mining, and hard rock mining, and block-caving.
Quite a lot of different types of mining, but in general they’re grouped into two main ones, surface mining and underground mining. It’s also useful to know that surface mining today is much more common and produces about 85% of minerals excluding petroleum and natural gas just in the United States alone. When we’re talking mining, usually because of this you’ll probably going to be talking about surface mining but at the same time there’s always also underground mining.
Mining itself is a long process, and that’s also why I really like it because it has lots of steps involved, and those steps range from for instance for surface mining you need to first actually plant … You need to first explore. You need to perform mineral exploration to find where you’re going to be digging where to find those minerals. Then you need to plant explosives and blow up the ground. In fact, like for a couple years one of my roommates was a blaster from Czech Republic, his name is Vojta. If anybody’s listening, if Vojta you are listening, hey, huge shout-out to you. I learned a lot about mining from you Vojta.
You have to blast the minerals, well, the earth in order to shake it up so that it’s easier to excavate. Then you need to excavate it. Then you need to transport it. That involves a lot of logistics. Then you need to also once it’s arrived on site, that’s, all of that first part is the extraction part where you’re getting the ore out of the earth.
Now from here you move on to the recovery phase. The recovery phase is a whole new phase where you need to crush the ore or crush the like whatever other rocks you have and make them small, and then you need to perform different techniques such as flotation or x-rays and others to extract the actual minerals that you’re after from the ore. Then you have to process all of that. And finally then it has to be shipped off to the buyers of the ore. So a massive, massive process. Good to know that it has all these components.
We’re not going to dive into every single stage of the process here in this podcast. We’re only going to look at five main elements where data science can be applied. But just so that you know, if you ever do consider mining or you want to look into it, there’s lots and lots more areas where data science, machine learning, artificial intelligence can be applied to this industry.
The way we’re going to look at how data science is applied is we’re going to go work backwards. We’re not going to start from the mineral exploration and the blasting and then the excavation. We’re going to work backwards from the end of the cycle at the recovery stage. The reason for that and the very actually one way mentors taught me this is that specifically in this industry, in mining, like any improvement that you can make towards increasing efficiency, the further down the line it is, the closer it is to the final product, the higher your overall contribution will be to the whole process.
So if you increase efficiency by 5% at the very start where you’ve got mineral exploration or the blasting stages or the excavation, down the line that will erode and will only be about like 1% increase in efficiency in total. Whereas if you manage to increase efficiency by 5% at the very end, in the recovery process, that’s going to translate into a 5% overall increase in efficiency. That’s got to do with bottlenecks, how they’re placed in the mine, where the main bottleneck is and how the mine is designed, and things like that.
That’s something that can be explored further. We’re going to record a whole podcast on that topic, but in general that’s a rule of thumb, like if you want have a maximum impact, then it’s best to look at the mining process from right to left, from the output, from the final stages and then slowly working your way to the start, and improving … using data science in our case improving the mine like that.
Let’s get started. Number one use case of data science in mining is recovery improvement. More and more companies are trying to increase the amount of minerals that they’re recovering from the ore that they have. Depending on the type of mineral, depending on the industry, that percentage can range from like 70% all the way to like 90, in the high 90s, 96%, 98%. The more you can extract, the more you can recover, the better. Because you’re already mined all that ore. You already have all those rocks. It’s already sitting there. Anything that you don’t recover, for instance, if you’re working with gold, any gold that you don’t recover from this ore is just going to go to waste, you’re going to lose it. You want to put in as much effort as you can as long as it’s still efficient to do that in order to recover that ore. This is where data science can come in and help because you can increase the recovery percentage with minimal effort by just analyzing data.
Recovery types vary of course depending on the mineral. For instance, in the case of gold, a recovery process called flotation is used. Highly recommend checking out flotation processes on YouTube. They look pretty cool and very different to what you’d expect. They use bubbles that are passed through a solution of very small particles with dirt and gold and then the gold sticks to those bubbles and goes to the surface, it kind of floats. Then it’s collected there. Very interesting.
The example that we’re going to look at here or the use case, a case study we’re going to look here is there’s a diamond mine called Renard in Quebec which employs a smart system for waste sorting and disposal. This system is primarily used to improve the quality and quantity of the diamond recovery process. The algorithms use data from sensors and x-rays and [inaudible 00:10:38] whatever like use sensors and x-rays to increase the diamond recovery rate which helps recovering at least 96% of the weight of all diamonds larger than one millimeter. That’s the percentage that we talked about. As you can see, using machine learning and data processing algorithms they’ve managed to get very close to 100% which is the ultimate goal of course.
That’s one example of recover improvement. There’s plenty, plenty of ways to use data science to help mining operations improve their recovery.
Number two, moving a step back, predictive maintenance and loss prevention. Another thing that we wanted to look at is we’ve got machines such as like say the trucks that are moving the ore or mills. For instance, there is massive mills that are like three storeys high that are processing this ore to break it down into smaller pieces. We want to make sure that they’re online all the time, not only just because they’re expensive to maintain, but also because of the opportunity cost. Whenever a mill like that is not working, is not breaking down ore, that is lost opportunity cost, that is an extra bottleneck in the process. We want to avoid that because that decreases the amount of gold or whatever other material we are working with that is going to be produced in the end.
What we have here is we can use sensors to monitor temperature, speed, vibration, and other characteristics on machines to tell the teams when it’s time to do maintenance before the breakdowns actually occur. That’s called predictive maintenance. An example here is Newcrest Mining partnering with Petra Data Science to deploy semi-autogenous grinding or SAG mill overload downtime duration algorithms at their Lihir mine. This allowed them to reduce the number of overload events from several per year to zero within the first four months. That’s a massive, massive win because that helps avoid bottlenecks and opportunity costs.
Number three, moving another step back, autonomous transportation. The mining operations as we discussed require a lot of logistics. The ore needs to be moved around. They have massive trucks which you’ve probably seen in one of these photos somewhere, like the wheel of the truck is bigger than an average car you see on the roads. Those trucks can carry about 350 to 400 tons per truck. An opportunity here is to actually make these trucks self-driving.
We’ve heard a lot of news about Uber, Google, Tesla creating autonomous vehicles. Well, guess what? There’s a mine in Australia in West Australia called Mine of the Future by Rio Tinto, and it’s already been operating a fleet of 80 autonomous whole trucks, so one of those big trucks, 80 of those trucks completely autonomous since 2008. For over 10 years they’ve been operating a fleet of 80 autonomous trucks. That’s about 20% of the total fleet which is 400 trucks. Those 80 trucks are just going to back and forth all by themselves. These trucks have impacted the company’s bottom line by reducing fuel use by 13% and they’re safer to operate. That is a massive innovation in the space of mining where you can automate transportation. So something that can further be improved. As you can imagine, there’s always room for improvement, there’s always lots more mines that can jump on that autonomous vehicles trend.
Moving even further back, we’re going to skip the blasting parts and the excavation because there’s something else that is really cool at the very start we’ve got mineral exploration. Of course, this is one of the most important parts because if you explore and you don’t find where you have like an estimate that there’s minerals in a certain area and then you start digging and they’re not there, then that’s a lot of wasted time, a lot of wasted resources. Machine learning can come into, come in and help here.
We’ve got an example here, GoldSpot. GoldSpot Discoveries INC using machine learning and they were able to find 86% of the existing gold deposits in Quebec Abitibi. This is the key, that in order to do that they only needed 4% of the total surface area to do so. Basically they used machine learning to very efficiently locate 86% of the total existing gold deposits in that area. That significantly reduced exploration time and costs, and as you can imagine, that increases the profitability of the mine.
That’s four examples of how we can apply and how actually companies already do apply data science, machine learning, and AI in their operations.
Our fifth example is going to concern something also extremely important in the space of mining and that is worker health and safety. Mining is one of the top 10 most dangerous jobs in the world with an average of one death per 3700 workers. Dangers include cave-ins, flooding, elevator problems, lung and respiratory diseases. Although there are no accurate figures, estimates suggest that such accidents kill about 12,000 people per year. That’s a massive number if you think about it, 12,000 people die in mining per year. This is where we can actually use machine learning and artificial intelligence, data science to improve the situation, to help and save people’s lives. I think that’s a very noble cause and something that is worth getting behind if you have the opportunity to do so.
How can we do that? Here’s an example right away. Automated ground control systems installed by many mining companies across the globe are primarily used underground or for pit mining. These systems capture data from the vibrations in the ground and can determine whether the mine is strong enough. Whenever miners face real danger like a ground slide or a tunnel collapse, the monitoring system can send out warning signals for miners to evacuate to safety. Data generated by this ground monitoring system can be used to create safer and cost-effective procedures for drilling and blasting.
That’s just one example of how data science is already used. But if you use, apply your imagination, I’m sure there’s lots of other ways that data science and machine learning and AI can be used to help with this problem ranging from looking at fatigue patterns, for instance, how efficient a worker is and based on that we can see if they’re getting tired or not. Maybe they need a break. Looking at drone footage, maybe sending out drones to fly over mines and exploring for anomalies and seeing, maybe looking at roads and looking at any problems on the roads that may affect the safety of driving on those roads because they’re not paved roads, those are just dirt roads where the trucks are driving.
Using computer vision through surveillance cameras to see if potentially somebody’s in danger or stuck in an elevator or is showing signs again of fatigue. Using artificial intelligence to monitor the rosters and how workers are placed into those rosters and to make sure that they have adequate times to have enough rest. Things like that. There’s lots of ways that we could potentially improve worker health and safety on mines, and that’s something that’s not to be left out from this industry because it is an extremely important component. These are people’s lives. This is probably the most important element in these whole mining operations.
There we go. That’s how data science can and is used in mining to improve operations and increase health and safety. I hope this little excursion into the world of mining was useful and interesting to you. We’ll include all of the links to additional materials that we used in this research at www.superdatascience.com/228. That’s www.superdatascience.com/228. You can go check them out there, and of course if you’re interested, highly encourage you to look a bit more into this industry, very, very fascinating world of mining.
On that note, thank you so much for being here today. I look forward to seeing you back here next time. Until then, happy analyzing.
