Data Driven Business Solutions

We specialise in Data Science and AI solutions for complex engineering and business problems.

  • Varun Rao

Artificial Intelligence in the Mining Industry

For an industry so heavily rooted in the physical sciences, it may surprise readers to learn that the mining industry was among the pioneers of industrial scale artificial intelligence applications. In this article we discuss some pioneering applications of AI/ML in this exciting field.

Although manufacturers and governments are scrambling to introduce autonomous cars on the road, the mining giant Rio Tinto has run autonomous haul trucks to their West Angelas mine in Western Australia for over 10 years. This farsighted decision has reportedly reduced costs by 15%, as well as reduced risks significantly. These behemoths are operated by a supervisory system and a central controller 1500 km away, and use GPS location data to automatically navigate haul roads.

Autonomous Komatsu 930 trucks at Rio Tinto's West Angelas mine in WA [source]

There are other players in the game too. Fortescue's 112 autonomous trucks have reportedly delivered a 30% improvement in productivity. The company's aim is to have the first fully autonomous haulage fleet before 2025. In 2017, BHP Billiton announced its Jimblebar mine would switch completely to driverless trucks. Australia currently dominates the autonomous haul truck, with 80% of the global fleet of 500 trucks rolling wheels at one of our mines. By 2023, is is estimated that this number will triple.

So much for autonomous trucks, an area that Australia is widely considered to be a pioneer in. In the next section, we discuss some further use cases for AI in mining.

Anomaly detection

Machinery failure is a significant issue for mines around the world. Unexpected failure of a machine can lead to lost revenue or worse, injury or death. Given the stakes, advance knowledge of when a machine is likely to fail is a crucial weapon in the miner's armory. Such foreknowledge would also help mines plan maintenance schedules around likely failures.

We discussed our anomaly detection algorithm, Apollo, in detail in another article. Briefly, Apollo helps answer questions such as:

  • Will this truck axle fail in the next 4 days?

  • What is the smallest change I can make to extend the life of this drill?

  • Given the current combination of ambient temperature, soil type and rainfall, are this excavator's readings normal?

Vale used AI to predict rail fracture, resulting in decreased downtime and a saving of $7m per year.

Vale also extended their approach to train wheels, where sensors monitor the wheel bearings to predict imminent failure. Savings in the first year alone were estimated at over $600k. Other companies use specialised machine learning algorithms to detect anomalous behaviour of general mining machinery.

Energy prediction

Deep Blue AI's energy prediction algorithm has been discussed in detail in a previous article, and technical paper. Given recent environmental concerns, energy usage is a significant and growing problem for mining companies. In a bid to reduce their environmental footprint, companies such as Vale have used AI to reduce fuel consumption in off-highway trucks.

  • What is the predicted energy consumption for a particular building?

  • How can I optimise driving routes to minimise fuel consumption?

Sales Volume Prediction

We discussed our retail sales prediction algorithm in a previous article. Armed with accurate predictions, companies can manage dynamic pricing, staff rostering and inventory management to maximum profit and reduce costs. Some questions that could be answered by harnessing the power of AI are:

  • How much iron ore can I expect to sell in 6 months?

  • What are the factors affect my coal sales volumes?

  • What impact will a 10% drop in the exchange rate have on my sales?

Process control and optimisation

Process control is an important aspect of the mining process. These operations are ideally suited to AI applications because of the vast array of ordered data that is available through in-plant sensors. Using these rich datasets, plant behaviour can be modelled to a high degree of accuracy.

Take for example sorting of material. The very nature of digging resources out of the dirt means that the desired product, such as coal or iron ore, is inevitably mixed in with the dirt and socks surrounding it. Smart sorting equipment using AI can be used for automated sorting of material, resulting in a 12% decrease in required mass. In another case, AI algorithms helped the miner Freeport-McMoRan increase copper production by 10% by pushing their Bagdad copper mill beyond the operating envelope set by veteran mining engineers.

Some further questions that could be answered by intelligent applications of AI are:

  • What are the early warning signs of ore impurity?

  • Is the estimated recovery for this process?

  • What input is mostly likely to affect coal quality?

Vale used AI to extend the lifespan of truck tyres by 30%, saving $5m.

Fatigue management

Wearable technologies use innovative materials and miniature sensors to seamlessly integrate into our lives. Although consumer applications are by far more common, the mining industry has adopted some key elements of this fascinating advance. BHP Billiton adopted smart caps for truck drivers in their Escondida copper mine in Chile; the sensors detected brainwaves and identified sleepy or fatigued drivers. Proactive fatigue management leads to significantly improved health, safety and productivity outcomes for the company and its employees. Some typical questions that could be answered are:

  • Are my excavator operators fatigued?

  • What are the telltale signs of tired truck drivers?

AI-driven fatigue management improves health, safety and productivity.

General decision making

Running large businesses is a complex affair, with hundreds or thousands of decisions to be made every day. The sheer complexity of a modern mine renders any optimisation by humans infeasible. In contrast, this is precisely what AI algorithms excel at - processing massive quantities of data, determining relationships between variables, and predicting future outcomes. While the discussion above focused largely on technical issues related to mining, there are more general applications that could leverage the power of AI. Some examples of questions that could be asked are:

  • Is it more advantageous to increase truck speed or load capacity?

  • When is the ideal time to purchase safety helmets, given current and predicted inventory and prices?

  • Is it more efficient to run an extra shift on a weekday, or pay weekend overtime?


Bountiful opportunities notwithstanding, there remain several obstacles to the widespread adoption of sophisticated AI/ML algorithms. In our experience, these issues can largely be attributed to the following three reasons:

In-house talent

Developing in-house talent is a lengthy and expensive process. Data scientists are a unique breed, generally with a narrow but highly specialised set of skills. The degree to which they require domain knowledge of the problem they are working on is frequently underestimated, resulting in theoretically optimum solutions that are in reality unfeasible. Of particular concern to the mining industry, data scientists also require a deep understanding of the engineering processes underpinning their problem to avoid falling into this common trap.

Lack of understanding

Following on from the problem of developing in-house talent, managers are loath to invest time and money in projects they do not fully understand. The proliferation of AI/ML in the public sphere has resulted in a great deal of confusion and misinformation over what the algorithms can and cannot do. In previous articles (see parts I, II, and III) we have laid out a clear, concise vision of the opportunities and limitations of the general machine learning method.

High quality data

The third problem we are familiar with is the issue of collecting high quality data. The upstream data collection task is generally undertaken by operational staff without adequate consideration of the utility of the data for AI/ML algorithms. This is then presented to data scientists as a fait accompli, resulting in a sub-optimal outcome. Our recommendation is for data scientists, operational staff and domain experts to work together right from data conceptualisation to algorithm prediction.

How can Deep Blue AI help?

Our key differentiator is the cross-functional skill set our team brings to the table. Our unique combination of skills in mining, engineering, machine learning and data engineering allows us to generate actionable insights based on real-world experience. Our team can provide advice on every step of your project, right from ideation and data gathering, through to high-fidelity machine learning modelling, culminating in deep insights that help you cut costs, speed up processes, and improve safety.

Read more about our team here.

63 views0 comments

Recent Posts

See All