| Tunnel is an important basic infrastructure of highway and railway transportation,and its safe operation is related to national economy and people’s livelihood.Cracks,deformation,leakage and other defects on tunnel linings are the major potential hazards in tunnel operation.Rapid inspection and processing pose a big challenge in tunnel safety operation.The state-ofthe-art approaches to tunnel detection are of low efficiency and high security risks,which cannot meet the demand of massive tunnel safety status census in China.It is of great practical significance to study rapid inspection method of tunnel defect in an automatic and efficient way instead of manual inspection.This thereby motivates me to carry out the researches in this Doctoral thesis.This dissertation presents a rapid measurement technology based on multi-sensor integration,as well as an efficient and automatic detection method of tunnel appearance defects.The devised prototype equipment has been implemented in a large number of highway tunnel and subway tunnel in different provinces and cities in China,and its performance and effectiveness have been confirmed by real-world data.The main technical contributions are given as follows:(1)According to the different measurement characteristics and inspection requirements of the tunnel appearance defects,e.g.cracks and deformation,the approach to rapid inspection of the tunnel appearance defects based on multi-sensor integration is proposed.In order to handle the issues of large tunnel section,high accuracy requirement and none-GNSS signal environment,a rapid tunnel measurement method integrated with array cameras,laser scanner,inertial measurement unit and odometer has been proposed,which solves the problem of sensor attitude maintenance in dynamic measurement,achieves high-precision positioning posture in tunnel,and collaborates to collect high-quality tunnel lining images and laser point cloud data,paving the way for defect automatic detection.(2)Aiming to tackle the complete detection issue of the tiny cracks in tunnel lining images,LinkCrack,an encoder-decoder architecture based deep convolution network is proposed,which is able to utilize the spatial constraint of the neighborhood connection.LinkCrack improves the network performances by upgrading the lightweight encoder network while expanding receptive field of network from a global perspective.On the other hand,LinkCrack enhances the continuity of cracks by constructing the neighborhood connection loss function from the center pixel of the crack,improving the network ability of continuous crack detection.The experimental results show that the crack recognition performance of LinkCrack is superior to those of the state-of-the-art approaches to edge detection and crack detection.(3)In view of the problems that the scale of tunnel deformation is small and the elliptical fitting method is difficult to fit the cross section precisely,a tunnel deformation detection method based on the model of point cloud detail is proposed.In particular,the model of point cloud detail is firstly constructed based on the cylinder geometry features of the tunnel.The shield ring is then split according to the three-dimensional cylindrical expanded feature of the point cloud detail,and the point cloud of cross section is eventually extracted from the central of the ring.By calculating the deviation value of point cloud detail,the horizontal diameter of the cross section is determined by fitting the deviation curve,ending up with a new scheme of tunnel deformation detection.The experimental results show that the proposed deformation detection strategy provides higher fitting accuracy than the elliptical fitting method,and the deformation index is close to reference value from the total station.(4)Two prototype equipment are introduced to carry out the practical inspection of highway tunnel and subway tunnel in many provinces and cities in China,and the validity and practicability of the rapid tunnel measurement and defect detection methods are evaluated.Meanwhile,a real-world application of technical status assessment of lining cracks was introduced to detect the lining cracks from the panoramic images of two tunnels.The results show that the precision of crack detection is satisfied,and the detection efficiency is greatly improved than manual work.Furthermore,another implementation of tunnel horizontal convergence monitoring is addressed to calculate the horizontal diameter change from the threephase subway tunnel laser point cloud,the results show that the convergence deformation of tunnel can be quickly monitored,the efficiency is higher than the traditional total station measuring,and can meet the rapid and comprehensive inspection of the structure deformation of the tunnel. |
