Statistical Classification of Joint Roughness Coefficient by Support Vector Machine

Roughness is one of the geometrical properties of rock joints that can be expressed through various methods. In this paper, eight different parameters were used to estimate the joint roughness coefficient (JRC) of 112 joint roughness profiles. The range of variation of these parameters in a given roughness class is relatively large. These roughness values overlap with their adjacent classes. In order to use two parameters simultaneously to estimate the JRC matrix, the interaction of these parameters on the JRC value were evaluated. The resolution of different roughness classes in different scenarios was evaluated using Pearson correlation coefficient and using engineering judgment. So in this paper, a new method based on the classification of joint roughness coefficient (JRC) by support vector machine (SVM) is purposed. So in this paper, a new method based on the classification of joint roughness coefficient (JRC) by support vector machine (SVM) is purposed. Different joint roughness parameters including Z2, RP, Grasselli2D, standard deviation of asperities height (SDH), standard deviation of profiles height variation (SDPHV), standard deviation of asperities angle (SDA), and geostatistical parameters including range (a), sill (C), CA and SRv were evaluated for 112 joint roughness profiles. Using these 8 parameters, an 8 by 8 interaction matrix was created which consequently resulted in 28 individual two-dimensional JRC classification scenarios. A graph with SDH and SDA was selected for the Statistical classification of JRC (SCJRC) because of the relatively obvious boundary between JRC classes and easy calculation. Finally, data classification was performed by SVM. The estimation of SCJRC was checked by 20 experimental direct shear test data. A good agreement is observed between SCJRC and experimental results. The results illustrate that SCJRC is an appropriate method for the estimation of JRC.