Estimating the Mechanical Properties of Complex Fractured Rock Masses Using Machine Learning
Wenzhao MENG+, Wei WU#
Nanyang Technological University, Singapore

Natural rock masses contain randomly distributed fractures and exhibit complex mechanical behaviors. These fractures may interact with filling materials (e.g., soil, water, and ice) and behave even more complex. The estimation of fracture parameters in present failure criteria is often subjective and may not be reliable to predict the mechanical properties of fractured rock masses. Here, we present a random forest model to predict the mechanical behaviors of granite specimens with ice-filled fractures. Due to difficulties in collecting fracture parameters from invisible granite in the laboratory and the field, we employ a two-dimensional particle flow code model and validate this model using the uniaxial compression test results of intact and fractured granite specimens. The numerical results show that the persistence factor and the inclination angle significantly influence the uniaxial compression strength and the Young’s modulus of the frozen fractured specimens, while the influences of the fracture number and the ice layer thickness are less important. We collect the uniaxial compressive strength and the Young’s modulus from 186 uniaxial compression tests on the frozen fractured specimens with different ice layer thicknesses, inclination angles, persistence factors, and fracture numbers. We use these parameters to train the random forest model and predict the uniaxial compression strength and the Young’s modulus of frozen fractured specimens with random number and geometry of ice-filled fractures. The prediction results outperform the uniaxial compression strength and the Young’s modulus obtained from the Hoek-Brown failure criterion and the Ramamurthy criterion. Our study indicates that machine learning method can be a reliable option to estimate the mechanical properties of complex fractured rock masses with randomly distributed fractures filled with various materials.