Determination of two-dimensional joint roughness coefficient using support vector regression and factor analysis

Joint roughness coefficient (JRC) is an important index in evaluating the mechanical and hydraulic properties of discontinuous rock mass. The correlation between the JRC and statistical parameters of rock joints is one of the commonly used quantitative methods to determine JRC. However, the JRC estimated from a single statistical parameter might be unreliable and inconsistent due to the complexity of the problem. In this study, eight statistical parameters were selected to provide a comprehensive description of the rock joint roughness. To predict the JRC, a nonlinear method based on support vector regression (SVR) and factor analysis was adopted. First, 112 rock joint profiles with available JRC values in the literature are collected; among which, 109 profiles were taken as the training database. The remaining 3 profiles along with another 106 joint profiles extracted from a sandstone joint sample in Majiagou rockslide area were taken as the test database. Second, the selected eight statistical parameters were calculated for those rock joint profiles, from which two independent common factors (i.e., an inclination angle factor and an amplitude height factor) were extracted through factor analysis. Finally, a SVR model was derived based on the extracted common factors and the corresponding JRC values of the rock joint profiles in the training database. The derived SVR model was then validated with the test database. The results show that the JRC predictions with the derived SVR model are more stable and reliable than those obtained with the regression-based correlations, and the derived SVR model could also capture the JRC anisotropy of the rock joint with investigated directions.