Intelligent Detection Of Multiple Defect In Tunnel Lining Structure Based On Deep Learning

In recent years,China's tunnel construction has developed rapidly,and a large number of highway,railway and subway tunnels have been put into operation.During its long-term service period,due to complex geological conditions,tunnel lining will inevitably produce cracks,spallings,voids,holes and other structural defects,and even cause safety accidents,which pose a serious threat to the safe operation of the tunnel.Therefore,the effective detection of the apparent and internal defects of the tunnel lining structure is essential to ensure the safe operation of the tunnel.The traditional manual inspection and defect detection methods are time-consuming,laborious,subjective,and have low identification accuracy.Research on efficient intelligent detection methods of tunnel lining structural defects has become the current hotspot of science and technology and the future development trend.This paper takes highway tunnels as the research object during the operation period,designs a method for rapid and intelligent identification of the appearance of tunnel lining structure and multiple internal defect simultaneously.Based on the research of the method,this work compiles a software system for tunnel lining structure defects detection.The main research results of this article are as follows:(1)Combining the SegNet network and Focal Loss function,a method for recognizing the apparent multiple defect of the tunnel lining structure based on the convolutional neural network FL-SegNet is proposed.It mainly solves the problem of pixel conflicts when it is difficult to identify small-scale defects and overlap multiple defect in a real environment.The Focal Loss function is introduced to improve the sample imbalance problem,and the training focus is set on the difficult samples by setting the training weights.In order to verify the applicability of the method in this paper,the network is applied to the real tunnel defect pictures of Nanshibi Tunnel in Jiangxi.And the training results are compared with the basic SegNet network and the two-stream method.The experimental results show that the Focal Loss function can effectively improve the accuracy of defect identification in the case of small-scale crack defects and multiple defect at the same time,and it is more advantageous.(2)Combining ground penetrating radar technology with Faster RCNN target detection network to realize intelligent,fast and non-destructive detection of multiple defect inside the tunnel lining structure.It mainly solves the problem of detecting the position and type of different kinds of defects and their combinations in tunnel lining,as well as the position and quantity of reinforcing bars.Finally,the method of this paper is verified in the simulation and measured radar data,and the area under the PR curve of each defect is calculated.Among them,the recognition accuracy of the steel bar reaches 1.The areas of air-containing cracks,air-containing voids,air-containing holes,water-containing cracks,water-containing voids and water-containing holes defects reaches 0.9073,0.9001,0.8965,0.9059,0.8881 and 0.7993,respectively.(3)This article developes intelligent identification software system for tunnel lining structural defects.Based on Python language,the visual tool PyQt5 framework and Qt Designer software are adopted to write the functions of portability,parameter configuration,embedded apparent multiple defect and internal multiple defect recognition.It integrates data processing and prediction recognition,including data preprocessing,defect recognition,result comparison and display.