Deep Learning Based 3-Dimensional Rock Discontinuity Trace Mapping and Orientation Characterization

Abstract

Listen

Translate article

ABSTRACT We propose a 3D rock discontinuity trace mapping technique based on deep learning. Additionally, we present a methodology for characterizing the orientation of rock discontinuity traces using the mapped results. The U-net-VGG 16 deep learning model, which is a transfer learning model based on U-net was used to achieve more precise and accurate discontinuity trace mapping. To verify the proposed methodology, a rock outcrop in the southern Korean peninsula was selected to create a 3D discontinuity trace model. Rock images and automatically mapped trace images obtained with the deep learning model, were combined to generate the 3D discontinuity trace model. The orientation of the rock discontinuity traces were calculated by fitting a plane to the point cloud 3D data of linear segments of the discontinuity traces, and obtaining the orientation of the generated planes. The orientation of discontinuity traces estimated using this method was validated by comparing those results with the orientations of the rock surfaces adjacent to the rock discontinuities. The results indicate that the deep learning model was able to properly map major discontinuity traces. Moreover, the characterization of discontinuity orientations using the 3D discontinuity trace model produced reliable results, with a maximum average difference between the orientations measured through discontinuity traces and adjacent rock surfaces of 2.1°. We believed that rock mass characterization using the proposed method reduces subjectivity and increases reliability. INTRODUCTION In rock engineering, rock mass stability is generally evaluated using classification systems such as the Rock Mass Rating (RMR), the Q-system, or the Geological Strength Index (GSI). The scores or values can serve to determine ground support or blasting patterns. Typically, rock surface mapping is performed manually. However, this process is subjective and can lack reliability. Moreover, areas of difficult access can increase the risk of accidents or diminish the rock survey quality.