Abstract: To resolve the challenges posed by the complex underwater structures of dams, where traditional contact-based inspection methods carry high risks, low efficiency, and insufficient accuracy, research is being conducted to explore the feasibility of developing a non-contact underwater flaw detection method. This approach integrates a high-resolution multibeam bathymetric system (Ultra High Density mode), high-precision positioning and attitude determination, and sonar imaging. An automated defect detection method based on multi-beam 3D point cloud contour line recognition was developed. This approach establishes high-resolution 3D models and is validated through manual inspection, underwater camera imaging, and automated defect identification, precise localisation, and quantitative analysis. A case study involving safety inspection of a minor seepage pit upstream of a foreign hydropower dam demonstrated: the system's bathymetric accuracy exceeded 0.5% × water depth, with a model resolution of 0.1m. Flaw detection, localisation, and quantitative analysis proved reliable, while the automated recognition algorithm matched manual identification, confirming the method's viability. This provides high-precision data support for subsequent emergency repair works and dam safety diagnostics, offering valuable technical reference for similar engineering flaw detection in steep, deep-water environments.