Point-Cloud-Based Spatial Shape Inspection of Historic Suspension Bridges: A Case Study of the Longjiang Bridge
Abstract
Spatial shape inspection of modern suspension bridges is relatively well-developed, whereas research on historic suspension bridges remains scarce. Historic bridges usually lack original drawings and maintenance records. Their slender members intersect, vegetation surrounds the structure, and no design geometry is available for comparison. This study develops a point-cloud-based spatial shape inspection method for historic suspension bridges. The method covers point-cloud acquisition and processing, multi-module semantic segmentation, spatial shape anomaly screening, and historical shape reconstruction. It was tested on the Longjiang Bridge in Yunnan, China. On all labeled points, the segmentation achieved an overall accuracy of 98.35%, a macro-F1 score of 96.09%, and an mIoU of 92.78%, outperforming a rule-based configuration by 40.79 percentage points. No continuous profile anomaly was found within the effectively observed span, and a controlled 40 mm local displacement activated the screening criterion. The present main-cable sag was 3.12 m; a 0.3% effective cable-length change gave a historical sag of 2.57 m, and a support range of ±50 mm widened the admissible interval to 2.21–2.92 m. The proposed method can serve as a preliminary step in the health inspection of historic suspension bridges and provides methodological support for building digital archives of historic bridges.