A comparison study on the predictive ability of numerical methods for fracturing of rock with different pre-existing flaws

Various numerical methods have been developed for modelling rock fracturing problems. Most of the available studies could reproduce reasonable results for given experimental data, but the ability and accuracy of these methods to predict rock fracturing are still controversial, and a systematic comparison is lacking. In this study, a set of rock mechanics experiments were designed for a numerical modelling contest to test the accuracy of different numerical methods in predicting rock fracturing. Six groups of specially designed experiments on rock fracturing were conducted on three types of rock specimens with pre-existing flaws. The first step was called the white box stage, in which the boundary conditions and experimental results (loading curve and fracture pattern) of two groups of experiments together with classical rock material parameters were provided to the participants who were asked to reproduce the experimental results (strength and fracture pattern) while allowing numerical parameter calibration. In the second stage, called the black box stage, the participants were asked to reproduce the remaining four groups of experiments with their corresponding calibrated numerical parameters. To ensure the solvability of these experimental data, we successfully reproduced the experimental observations and results in both white box and black box stages by using a numerical method with the same set of material parameters. We received complete numerical reports from ten teams using different numerical methods and found that most of the teams were able to accurately reproduce the experimental results in the white box stage but not the black box stage. The discontinuum-based methods with fewer input parameters, clear physical meaning, and easy adjustment provided better performance than the continuum-based methods. The experimental data presented in this study could provide a baseline for further improvement of the predictive ability of numerical methods for more accurate modelling of rock fracturing.