Surface Roughness Assessment of Natural Rock Joints Based on an Unsupervised Pattern Recognition Technique Using 2D Profiles

Abstract

The stability of a jointed rock mass is generally controlled by its shear strength that significantly depends on surface roughness. So far, different methods have been presented for determining surface roughness using 2D profiles. In this study, a new method based on the unsupervised pattern recognition technique using a combination of statistical, geostatistical, directional, and spectral methods for the quantification of the surface roughness will be proposed. To reach this goal, more than 10,000 profiles gathered from 92 surfaces of natural rock joints were scanned. The samples were collected from limestone cores of the Lar Dam located in the Mazandaran Province, Iran. After introducing a new spectral index, determined from the fast Fourier transform for measuring the unevenness of rough profiles, statistical, geostatistical, directional, and spectral features revealing waviness and unevenness of the 2D profiles were extracted, and a representative vector and profile for each surface were introduced through the weighted mean and median of the profile features. Principal component analysis (PCA) was utilized for finding the direction of the maximum variance of information. Then, clustering of the 92 samples was performed via K-means, and the silhouette measure was used in order to find the optimal number of clusters resulted in the creation of 13 clusters. To verify the procedure, a sample was selected in each cluster, and direct shear tests were performed on the samples. Comparing the experiments and the clustering results shows they are in good agreement. Thus, the method is an efficient tool for the quantitative recognition of surface roughness considering the waviness and unevenness of a surface.