ABSTRACT
Segment assembly is a crucial phase in shield tunnel construction, where the initial assembly quality significantly impacts the subsequent segment assembly process and the structural performance. Current practices primarily focus on detecting the assembly quality of completed tunnels, lacking dynamic evaluation and timely guidance for segment alignment adjustments. To enable real-time detection of the assembly quality immediately after segment installation, a segment assembly quality detection method based on hybrid point cloud-image information is proposed. Initially, an orthophoto is generated from segment point cloud. Then, machine learning-based techniques are applied to identify longitudinal joint in the image, facilitating the segmentation of the point cloud. Subsequently, a cylindrical projection of the point cloud is produced, and point cloud slices are extracted from the projection. Finally, linear fitting is employed on these slices to calculate the joint’s dislocation and rotation. Experimental validation demonstrated the method’s excellent practicality and accuracy. Results indicate that the model trained with the Random Forest algorithm accurately identifies line segments of longitudinal joint, and proposed hybrid information segmentation method precisely divides the point cloud along segment joint. Field tests demonstrate that the segment dislocation detection method based on cylindrical projection achieves an average error of 0.60 mm and a maximum error of 1.48 mm, meeting engineering measurement requirements. This study provides a novel automated measurement technique for segment assembly quality assessment during construction, advancing the automation and intelligence of tunnel construction.
