Geological characterisation guided by fuzzy k-means clustering of physical properties measured from core samples from the Victoria property, Sudbury Ontario

Abstract

Core logging is a subjective practice done by geologists, which documents the mineralogy, textures, alteration, mineralisation and other features to give core a rock name. Pattern recognition techniques are able to characterise the rocks and link the geophysical and geological data quantitatively. The fuzzy-k means algorithm is an unsupervised pattern recognition technique, which groups data into clusters based on properties measured. This study will use the fuzzy-k means algorithm to characterise core samples from 2 drillholes from the Victoria property in Sudbury with thin section examination to identify how mineralogical changes can affect the measurements. Four different physical properties (density, gamma ray, conductivity and magnetic susceptibility) were measured from a total of 203 core samples of quartz diorite, metagabbro, metabasalt, pyroxenite, olivine diabase and metasedimentary rocks. The samples were classified into 4 different physical units, with additional confusion index values that indicate how well the data was classified. Quartz diorite, metagabbro and metabasalt have the highest confusion index values while the olivine diabase and metasedimentary rocks have the lowest confusion index values. Combining the fuzzy k- means results and thin section examination proved to be successful because heterogeneities in sulphide minerals, ore mineralisation and variation in rock forming minerals cause an overlap in physical properties with other rock samples, increasing the confusion index while homogeneity in mineralogy results in a low confusion index.