ABSTRACT
The performance maintenance of shield tunnels is of more and more significance with a great number of shield tunnels being put into service. The perception of deformation is an essential part of shield tunnel performance assessment because of the correlation between deformation and all kinds of affecting factors during service. While due to the nature of shield tunnels, the perception of deformation based on the traditional method requires costly effort. For the refined perception of ring-wise deformation, which includes segment rotation and segment dislocation, an automated workflow for processing and managing the point cloud of tunnel lining has been proposed based on deep learning and BIM. An end-to-end deep learning model based on PointNet++ has been trained for the segmentation of ring-wise shield tunnel point clouds. With the proposed data normalization and augmentation method, the accuracy of the deep learning model can achieve promising results. Based on the ring-wise deformation pattern, a robust and efficient optimization algorithm for deformation extraction has been designed. The fusion of accurate data from point clouds and semantic information from BIM models enables the automatic calculation of all types of ring-wise deformation. The proposed methods and workflow provide novel perspectives of digitalization in the life-cycle maintenance for shield tunnels.
