Joint roughness is one of the most important factors which decide the shear behavior of a rock mass. Conventionally, it is quantified into a parameter called Joint Roughness Coefficient (JRC) by visual comparison between Barton’s standard profiles and line profiles of a joint. Since the line profiles are acquired by hand-mapping using a roughness profiler, it is hard to survey unreachable parts of the rock mass. In this regard, we suggest a method to estimate the roughness of discontinuities using a Terrestrial Laser Scanner (TLS) which enables contactless data acquisition from long range. To resolve problems that TLS point clouds are unordered and containing noise, we used an artificial neural network called PointASNL. We generated synthetic surfaces that resemble joint surfaces scanned by TLS to fill the need of big data for training a model. After training, the model was evaluated with the result of roughness estimation for an outcrop scanned by a TLS device. It is shown that the model can estimate the roughness with average error of JRC 3-4.
Terrestrial Laser Scanner (TLS) is a device which scans surrounding surfaces quickly and translates them into a massive 3D point cloud. Laser emitted from the laser diode goes through the rotating mirror at the center of the device to head towards the surfaces and comes back after reflection. Rangefinder and gimbal inside measure the travel time of the laser, vertical angle between horizon and a point, and the horizontal direction to calculate 3D coordinates of the surfaces. Most TLSs can be operated in a distance of several to hundreds of meters and the rapid rotating mirror enables to get millions of data points in less than tens of minutes. Therefore, one can easily reconstruct the surrounding area or objects into a 3D point cloud for various purposes (e.g. 3D map construction, quality inspection for huge parts).