Abstract: There is a complex nonlinear relationship between rock joint roughness coefficient (JRC) and the statistical parameters. And JRC value calculated using single statistical parameter is not reliable, since the description is one-sided on the morphology of discontinuity surface. Four parameters, average inclination angle , average relative amplitude , standard deviation of inclination angle and standard deviation of amplitude , are selected to describe the morphology of discontinuity surface. The training samples containing 102 profiles with available experimental back-calculated JRC values are chosen to establish the nonlinear relationship between JRC and the selected statistical parameters, then a JRC support vector regression (SVR) prediction model is established. The SVR model is proved to be reliable by comparing the predicted and experimental back-calculated JRC of Barton standard profiles. The discontinuities of Majiagou landslide in Zigui county, Three Gorges Reservoir Region are selected as a case. Three-dimensional laser scanning test is conducted to obtain the morphology data of discontinuity surface and a 3D model of discontinuity surface morphology is developed. Direct shear test is carried out to back calculate its JRC. The results show that JRC predicted by the SVR model and experimental back-calculated value have good consistency, and the relative error is only 4.5%. The JRC estimation results by different regression equations based on statistical parameters for the same profile have larger variation, which indicates that the JRC predicted by SVR model based on the chosen statistical parameters is more reliable. The results may also provide a new approach to quantitatively determine JRC.

Key words: discontinuities, statistical parameter, joint roughness coefficient, support vector machine, three-dimensional laser scanning