AUTOMATED ORE BODY MODELING USING MACHINE LEARNING-BASED IMPUTATION AND ALPHA-SHAPE CONSTRUCTION

Ore body modeling for resource estimation and mine planning plays a critical role in exploration and mining for mineral resources. Despite significant research and development in this field, the state-of-art approaches continue to rely on manual computer-aided implicit and explicit shape construction techniques. Although such techniques have served the resource industry well, they are still time-consuming and involve a significant amount of subjectivity to account for the diverse nature of geological controls. We have developed a new approach by combining machine learning-based imputation methods and automated shape reconstruction from spatial data using the concept of alpha-shapes. In this approach, we first perform intelligent interpolation that is augmented by the spatial continuity identified from the sparse data such as grades from drill holes and guided by independent geological and geophysical information. We then apply a data adaptive and multi-resolution method using the alpha-shape to form the orebody shapes from the outputs of the machine learning based data imputation methods. The alpha-shape is a generalization of convex hull for shape estimation, and we have enhanced the method by introducing a multi-level cubic-wise shape estimation component. The approach is computationally efficient, requires minimal manual intervention, and removes unnecessary subjectivities. We present the basics of the technique, and illustrate and verify its efficacy using the example of the Eskay Creek volcanogenic massive sulfide deposit rich in precious and base metals. A large drill-hole data set is available from historic production and on-going exploration, providing an ideal study to test the newly developed method for shape construction.