Technical advancements of tools for geoscientific data collection and machine learning techniques are becoming increasingly popular in applications for exploration geology. The aim of this research is to automate aspects of geological drill core logging based on the distinct geochemistry of the rock. A Swedish company called Minalyze developed an X-ray fluorescence (XRF) scanner that is able to detect the multi-elemental composition of the rock for the full length of the core. This machine was implemented into the mining workflow by Glencore plc at the George Fisher mine in Mount Isa, Australia. For this study, Glencore plc provided XRF scans for 31 drill holes. The scanner determined the quantity of each of the 24 elements for every 10 cm of the core. An unsupervised machine learning technique called self-organizing maps (SOM) was applied to the multi-dimensional data set to define anomalies and correlations between the input variables. The results showed a strong correlation between certain elemental compositions from the XRF scans and corresponding rock types, which were previously defined by mine geologists. Based on the SOM analysis of the XRF data, a classification scheme was developed which enables automatic differentiation of the main rock types at the George Fisher mine. This classification scheme can be used to automate the first passes of geological drill core logging at George Fisher. The application of a SOM was proven to be very consistent and accurate in recognizing distinct rock types, since the analysis was based on the input data from XRF scans. Automated geological drill core logging has the potential to enhance the consistency and significantly reduce the processing time and cost of drill core logging.
展开▼
