Using LiDAR-enabled digital mapping and machine learning to identify cracks in Precast Concrete Tunnel Lining at the Ashbridges Bay Treatment Plant Outfall tunnel project

Abstract

Records that are kept during tunnel construction, such as photo records and survey data, are relied upon to generate As-Built drawings in CAD-based design software. Having complete and accurate As-Built data is key to certification upon project completion and as a benchmark for ongoing monitoring during construction and/or operation, however it can be time consuming and resource intensive to collect and organize the data needed. 3D laser scanning can be utilized to capture point clouds for generating detailed 3D models of underground assets and can subsequently be used for engineering decision making. In addition to scanning the built environment, engineers are now able to annotate and characterize features of interest digitally, to further increase the speed of acquisition and usefulness of the data captured during tunnel construction.

This paper presents a workflow for georeferenced digital mapping of Precast Concrete Tunnel Lining (PCTL) in the crown of a 7.0 m internal diameter Ashbridges Bay Treatment Plant Outfall (ABTPO) tunnel in Toronto, Canada using a Light Detection and Ranging (LiDAR) enabled mapping tool. The captured point cloud is used to develop a machine learning (ML) algorithm that is trained to identify deficiencies (e.g., cracks) in the PCTL segments. The hyperparameters of the ML classification algorithm are optimized to prioritize accurate prediction of the location of cracks in the PCTL segments. This study aims to demonstrate the utility of performing digital mapping of tunnels to record conditions, as well as the accessibility of incorporating ML into future tunnel monitoring, particularly when relevant digital data is collected during initial tunnel construction.