Abstract: Due to the poor accessibility in deep and turbid water environments,traditional underwater inspection methods (e.g.,manual diving and underwater photography) struggle to detect corrosion-induced hole defects in underwater steel structures.To advance the development of bridge underwater inspection technology,this study proposes an automated detection method for corrosion voids in underwater anti-collision steel cofferdams based on 3D sonar point cloud modeling.The proposed method integrates second-nearest-neighbor spacing statistical features with an Alpha Shape algorithm to construct an adaptive Alpha Shape-based edge detection model for point clouds.Subsequently,a polygon decomposition technique is applied to segment individual voids from the identified edge point clouds,thereby achieving automated recognition and geometric quantification of corrosion voids in underwater steel cofferdams.Experimental validation through underwater measurements demonstrates that the proposed method achieves an average accuracy of 76.2% in hole defect assessment.Furthermore,the method is successfully applied to inspect thin-walled steel plates on the main pier of a Yangtze River bridge,detecting a total void area of 0.542 m².This research provides a novel technical framework and methodological reference for the digital and intelligent inspection of underwater infrastructure.

Key words: bridge engineering, underwater steel plate corrosion detection, sonar point cloud, point cloud edge detection, hole segmentation