Inspection, Survey and Defect Mapping of Dam using Photogrammetry, Thermal IR and Imaging
High resolution inspection and defect mapping over more than 60,000m2 of concrete dam
Project Overview
A dam located in the North West Slopes of NSW has a rock-fill embankment wall with a concrete slab on the upstream face. The dam wall is 954m long and up to 85m high. In lieu of a conventional clay core, the upstream concrete face provides water proofing to the wall and so the integrity of the slabs and construction joints is of high importance to the safety and performance of the dam.
Conventional inspection of the upstream face involves personnel working from a boat to visually inspect the slabs and note the location and extent of any identified defects. Severe drought conditions in early 2019 saw the reservoir levels drop to 6%, exposing most of the upstream face and presenting an ideal opportunity for detailed inspection and crack mapping.
Diodrone was engaged by WaterNSW to develop a reality capture and digital engineering solution for inspection and defect mapping of the upstream face, as well as provide a precision aerial survey over the downstream face for use in ongoing displacement monitoring.
Diodrone Solution
Diodrone developed a drone-based reality capture and digital engineering solution to provide WaterNSW with a detailed crack map for the exposed surfaces of the upstream face, which at the time of capture included more than 60,000m2 of concrete above waterline.
Photogrammetry based 3D modelling, machine-learning, and digital (human) crack mapping developed an ultra-high-resolution 3D model of the site along with a scale-accurate and georeferenced vector layer for cracks and slab boundaries.
PROJECT SCOPE
Precision Aerial Survey
Crack Mapping
3D Modelling
The use of drone-based reality capture provided access and quality that conventional methods could not.
DAM DEFECT MAPPING
A small multirotor drone was used to capture RGB imagery over the extent of the upstream face. Operating at 5-10m above the concrete surface, imagery achieved an average resolution of 2mm per pixel. Images were then processed in photogrammetry software to develop an ultra-high-resolution 3D model and surface aligned orthographic imagery. A digital inspection was then carried out to develop a detailed crack map for each of the 55no. slabs on the upstream face.
Digital crack mapping was carried out by referencing the orthographic imagery of each slab and manually tracing out cracks as they were identified. Quality control was achieved by implementing a closed loop review process whereby cracks were mapped by one technician and then reviewed and improved on by another technician, which where then subsequently reviewed again and so on until no further cracks could be identified. The georeferenced crack layer was then incorporated into a crack map for each slab and delivered in georeferenced CAD and PDF formats.
Photogrammetry derived an ultra-high resolution 3D model and orthographic imagery for digital inspection and crack mapping.
Machine learning was trialled as a means of automating the crack identification and mapping process, though results included a significant number of false positives and failed to identify the finer cracks which were readily identified with human inspection. The project concluded that significant further development of machine learning is required before deployment as a viable solution to crack mapping in concrete structures, particularly where hairline cracks and relatively small defects are of interest.
DELIVERABLES
- Web-based 3D GIS
- Ultra-high-resolution 3D model
- Precision Aerial Survey (point cloud)
- Dam Defect Mapping
- CAD and PDF outputs for each concrete slab
Photogrammetry derived an ultra-high resolution 3D model and orthographic imagery for digital inspection and crack mapping.
Precision Aerial Survey
A small multirotor drone was used to develop a photogrammetry based aerial survey of the downstream face. The drone was operated at an average of 45m above ground level, capturing nadir and oblique imagery at an average of 13mm per pixel.
Utilising a combination of direct georeferencing and a network of 54no. ground control points, photogrammetry processing achieved sub-pixel reprojection errors and an accuracy of 7mm RMSE when compared to 4no. check points distributed throughout the survey area.
The high-resolution point cloud was then gridded into a digital elevation model (DEM) and used to generate elevation contours, which with future surveys will enable change detection and displacement monitoring over the downstream face.
Photogrammetry outputs achieved an accuracy of 7mm RMSE when compared to a network of high-precision check points.
