Application of Machine Learning to the Estimation of Intact Rock Strength from Core Logging Data: A Case Study at the Newcrest Cadia East Mine

Abstract

Understanding the spatial variation in intact and defected rock strength is critical to geomechanical mine design. At the Newcrest Cadia East mine, systematic point load testing (PLT) was used to measure the strength of intact rock and individual defects (e.g., veins) at regular closely spaced intervals along several boreholes. The systematic PLT data collection covers only 1.3% of the 590 km of hole logged at Cadia East. A procedure was developed to homogenize the available geotechnical and geological logging data, infill missing values, and encode raw data into engineered features for use in a machine learning model. A random forest classifier was applied to predict point load index (Is50) from core logging data where tests were not performed. The random forest model predicts the rolling average Is50 value within 1 MPa 48% of the time. The model gives insights into which core logging quantities have the strongest controls on rock strength and provides the basis for developing more detailed geospatial models of intact and defected rock strength.