Towards AI-driven real-time deposit modeling: a case study in southern Brazil

Abstract

Mineral exploration and resource estimation play pivotal roles in the mining industry, driving the need for accurate and comprehensive data about mineral deposits. Traditionally, drill core samples have been the primary means of obtaining crucial information regarding the size, shape, and mineral composition of deposits. However, the cost associated with drilling limits the number of samples that can be acquired, posing challenges to achieving a thorough understanding of a mineralized area. In response to these challenges, the mining industry is increasingly turning to cutting-edge technologies to enhance exploration efficiency and reduce costs, such as aerial photogrammetry, hyperspectral imaging, and core scanning. These technologies offer the advantage of acquiring data over larger areas in a relatively short period, providing valuable insights into the geological characteristics of a site. In the context of developing mines, each round of blasting uncovers fresh rock surfaces that harbor new geological information. Leveraging this opportunity to gather real-time data presents an exciting prospect for optimizing mineral exploration. By systematically collecting and processing information from newly exposed surfaces, it becomes possible to enhance the insights obtained from conventional core samples. This research introduces a case study exemplifying the implementation of this paradigm shift in a marble quarry located in southern Brazil. Here, a convolutional neural network is employed to interpret geological features in aerial photographs. Subsequently, the interpreted data is fed into a photogrammetry software, generating a labeled point cloud that complements information derived from traditional core samples. The synergy between aerial data and core sample information allows for the creation of highly detailed lithology models, enabling more accurate short-term forecasting of stripping ratios. A key aspect of this work involves the development and utilization of an in-house Gaussian process implementation for lithological modeling. This technique not only provides insights into the size and shape of the orebody but also offers an invaluable uncertainty estimate. The results from this case study demonstrate the potential of this paradigm shift in mineral exploration and mining practices. Ultimately, this research aims to showcase a pathway towards a more economical, environmentally friendly, and sustainable future for the mining industry.