Automatic Thickness and Volume Estimation of Sprayed Concrete on Anchored Retaining Walls From Terrestrial Lidar Data

Abstract

ABSTRACT:When ground conditions are weak, particularly in free formed tunnel linings or retaining walls, sprayed concrete can be applied on the exposed surfaces immediately after excavation for shotcreting rock outcrops. In these situations, shotcrete is normally applied conjointly with rock bolts and mesh, thereby supporting the loose material that causes many of the small ground falls. On the other hand, contractors want to determine the thickness and volume of sprayed concrete for both technical and economic reasons: to guarantee their structural strength but also, to not deliver excess material that they will not be paid for. In this paper, we first introduce a terrestrial LiDAR-based method for the automatic detection of rock bolts, as typically used in anchored retaining walls. These ground support elements are segmented based on their geometry and they will serve as control points for the co-registration of two successive scans, before and after shotcreting. Then we compare both point clouds to estimate the sprayed concrete thickness and the expending volume on the wall. This novel methodology is demonstrated on repeated scan data from a retaining wall in the city of Vigo (Spain), resulting in a rock bolts detection rate of 91%, that permits to obtain a detailed information of the thickness and calculate a total volume of 3597 litres of concrete. These results have verified the effectiveness of the developed approach by increasing productivity and improving previous empirical proposals for real time thickness estimation.

A Geometric-Feature-Based Method for Automatic Extraction of Anchor Rod Points from Dense Point Cloud

“…With the rise in artificial intelligence, many feature classifiers, such as Support Vector Machine [ 19 ], Random Forest [ 20 ], and Joint Boost [ 11 ], are also used in point cloud target extraction. Martínez-Sánchez et al [ 21 ] trained a deep network consisting of two autoencoders and an output layer for detecting anchor rods in slopes. In their study, the anchor rods were used as the basis for data alignment, so they were more concerned with finding more anchor rods rather than the identification accuracy.…”