by Somdutta Ghosh, Master in Geology at University of Calcutta
Smart work leads to quicker success
I have noticed that in every industry the main challenges are to reduce the cost and to ensure the value of time. The exploration and mining industry is no different and machine learning can be the best tool for this purpose. Nowadays, machine learning algorithms are performing by generating a bunch of code and can carry out thousands of people’s tasks. What are you thinking? The role of humans is going to be diminished?
Not at all, the real scenario is quite different. It is just a smart approach to use Artificial Intelligence (AI) and machine learning to increase the success rate because we are in the age of digital mining. As such, machine learning can be used as a high-tech treasure hunter to find out our desirable minerals. The algorithms can typically give insight into hydrothermal mineral deposits much faster than the core sampling result can. I am not saying this is the replacement of manual fieldwork or laboratory work, but it can be the best way of doing some smart work to ensure the right direction.
So, what is machine learning?
Machine learning is a sub-category of AI. Using this technique, we can solve complex problems even without being an expert in programming (Jooshaki et al., 2021). We can train the computer with the help of large available datasets and then ask it to make predictions when presented with new information.
Machine learning in mining and exploration
Machine learning can make predictions based on geological datasets and also perform data analysis. Examples include:
- It can help us visualize the pattern of datasets by generating figures and diagrams. Datasets can be geochemical data of an igneous complex, a core sampling dataset of an exploration or a geophysical seismic dataset.
- Machine learning can train a model to create prospectivity maps before selecting a mining area, which can give us a better understanding of the presence of desirable minerals for a project.
- AI can analyze aerial photography and satellite imagery. It can assist with identifying the geological features, the alteration of mineralogies, and hydrothermal deposits.
- Risk assessment and environmental sustainability is the biggest concern nowadays. Machine learning can help make decisions on dependable features and physical conditions.
- Missing values are a common headache for every sector. We also face problems with that in mining and exploration. Machine learning fills the gap. We can handle it with deletion and imputation methods. Firstly, if there are a few missing values then we can go with the simple deletion method. Yet, when it comes to a major gap in the dataset, we use some common statistical methods like mean, median, or mode imputation. There are some other imputation methods in machine learning, such as K-Nearest Neighbors (KNN) imputation and other methods like model-based imputation.
The best part of machine learning is that it can study the data and find patterns in it that might be missed by humans.
Different types of machine learning
Supervised learning
Let’s say you already know where the minerals are and few treasures are buried. Now we can teach the computerized algorithms to discover the common spots that could contain our wanted minerals. This supervised model (Figure 1) is initially trained on known data (the training dataset), which includes labeled data — data where each input is paired with a known output. After the model is validated, it can be used to predict or classify the unseen data (the testing dataset) based on what it has learned from the labeled data. There are many algorithms such as Random Forest (RF), KNN, Support Vector Machines (SVM), Artificial Neural Networking (ANN), and so on. Supervised learning techniques are used to classify the rock types based on mineralization patterns and geochemical data. In mineral processing, these models predict potential difficulties in mineral separation and identify the target mineral for extraction. SVM and RF have been used to recognize geological features and make classifications and predictions based on these characteristics. Additionally, KNN and ANN have been successfully applied to predict the ultimate tensile strength (UTS) of materials in exploration (Santos et al., 2009).
by using some lines of code
Unsupervised learning
This one is a little bit mysterious. Because the algorithms have no previous experience, they try to find out on their own, which looks reliable. It can be just like one of your friends who always guides you without even having experience. That means it does not require labeled data like supervised learning. The clustering method and dimensionality reduction are the two major techniques for this type of machine learning. K-means clustering, agglomerative clustering, and density-based clustering are used to group similar mining blocks, materials, or geological features based on some of their inherent properties like grade or mineral content. In material science, the clustering method has been used to classify nanoparticles, such as silver and platinum, according to their structural similarities and other properties. These can help enhance mining planning strategies. Additionally, unsupervised learning can be used to analyze satellite imagery for mineral exploration, which is advantageous for identifying potential deposits and making environmental assessments before digging.
By using these supervised and unsupervised algorithms, industries have recently made significant improvements in resource discovery, mining planning, sustainable operations, and cost optimization.
Case studies
GoldSpot Discoveries
A January 2021 report from Northstar Gold Corp mentions the advancement of mining strategies in the exploration of the Miller Gold Property in Northeastern Ontario. They have found eight new gold spot discoveries by using the machine learning clustering method and 3D mapping technologies, which were beyond their expectation. The discoveries were made by applying machine learning methodologies to data from previously known gold zones. After 2021, their fund increased by CAD 2.4 million and they started more drilling.
Rio Tinto’s Smart Mining
Rio Tinto is already using AI and data science to get better results. The company aims to modernize the mining industry and introduce digital technology. They employ predictive models that analyze geological data to predict new orebodies. The company also enhanced safety and reduced the risks for their exploration team by applying machine learning techniques to their reported data.
Feature detection in core photography
A September 2024 study from Datarock has mentioned some strategies with neural networks to enhance geological features in drill core photography. For geological core sample images, it is very common to have some artifacts, and sometimes it becomes very difficult to see the important features of the rocks. Using a Convolutional Neural Network (CNN) can help us. This methodology enhances the accuracy of the logging. By using this CNN deep learning method, detecting geological features like veins, fractures, grains, and similar textures has become easier.
So, there is no doubt that machine learning is considered a new technology in some mining industries. AI and machine learning are used globally to improve exploration and help companies discover precious metals in places that might have been overlooked. By mixing historical data with technology, we can reduce the exploration cost and make better estimations where to dig next.
Why is machine learning the future of geologists?
Machine learning is the green card for geologists to make a game-changing contribution to scientific work and industries as well. Now we can imagine the picture of a geologist without being out in the field with a hammer and collecting rock samples one by one. I think machine learning is not only about being dependent on technology, it is just getting smarter. Together, humans and machine learning are redefining exploration. The reasons why machine learning rocks (pardon the pun) include:
- Quick result: It can give us a result faster than manual digging.
- Accuracy: The accuracy depends on the data that the model has been trained on.
- Budget-friendly: It cuts down the cost of exploration by using datasets to predict the target area.
- Smart-learner: The more data we feed it, the smarter result it can give.
According to a report from Market.us, the global Al in mining market size is expected to reach around USD 7.2639 billion by 2033 from USD 939.1 million in 2023.
Challenges: Nothing can be flawless
Now, some might think that machine learning is just like magic and that can fix everything, but it is not without flaws. In some cases, there is insufficient historical data, which makes training the ML model difficult. Another major issue arises when the model performs poorly when applied to real life problems. This could be the result of inaccurate or incomplete data used for training it, or not achieving the optimal balance between model complexity and its ability to generalize on new scenarios. We cannot imagine a tasty recipe without the availability of all the ingredients needed and in appropriate quantities.
The time of advancement
Looking ahead, in the future machine learning will have a promising role in the mining sector. On the other hand, we have already noted some challenges, including data quality and technological adaptations, which are under the supervision of computer scientists. We are proceeding in the right direction to become a part of a graceful future. In mineral exploration, machine learning is going to open a new pathway for discovering the mineral jackpot and helping solve the remaining mysteries of the hidden treasures of our mother Earth.
References
Jooshaki, M., Nad, A., & Michaux, S. (2021). A Systematic Review on the Application of Machine Learning in Exploiting Mineralogical Data in Mining and Mineral Industry. Minerals, 11(8), 816. https://doi.org/10.3390/min11080816
Santos, I., Nieves, J., Penya, Y.K., Bringas, P.G. (2009). Optimising Machine-Learning-Based Fault Prediction in Foundry Production. In: Omatu, S., et al. Distributed Computing, Artificial Intelligence, Bioinformatics, Soft Computing, and Ambient Assisted Living. IWANN 2009. Lecture Notes in Computer Science, vol 5518. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-02481-8_80
For more information: Get in touch with Somdutta on LinkedIn
