Simulation And Test Of Tunnel Internal Defect Based On Ground Penetrating Radar

In recent years,China’s rail transit industry has grown up significantly and the number of tunnel kilometers built is getting longer and longer,and the internal structure of long-term service tunnels may change,and many diseases such as emptying and water leakage will appear.If the interior of the lining is hollowed out,it may cause the tunnel to collapse,seriously endangering the safety of driving,and the safety of the tunnel operation needs to be considered urgently.In the face of the damage caused by traditional testing technology to the line,and the workload is large,it necessitates a significant amount of human and material resources.,and the ground penetrating radar,as one of the most widely used means of nondestructive testing technology for tunnel lining,has the advantages of simple operability,fast detection speed and high detection accuracy.In this thesis,ground penetrating radar combined with machine vision technology is introduced into the field of tunnel lining inspection and experimental research is carried out to assist the staff in inspection.In this thesis,using a combination of software simulation and experiments,GPRMax software is used to simulate the location and size of holes in the reinforced concrete lining structure of the tunnel according to the Time-Domain Finite-Difference(FDTD)method,and use Matlab combined with digital signal processing technology to perform a series of image processing such as DC background removal,noise removal,and gain of the simulated images.The processed images are interpreted and analyzed,the problem of the lack of real data sets of tunnel lining is solved,and the simulated radar images are made into standard data sets according to the data set format required for convolutional neural network learning and training.In terms of algorithms,this thesis conducts intelligent identification research on rebar and holes in the tunnel lining based on deep learning object detection algorithm,and through the study of the latest object detection algorithm Yolov5 network model,combine dataset target characteristics and the algorithm suitable for tunnel lining structure detection system is designed,combined with the method of migration learning and data amplification,and the training is accelerated by Using GPU on the Pytorh framework platform.In this study,the K-means clustering algorithm is used to improve the adaptability of the network detector part,and the SPP module is introduced to improve the network sensing field,and the Recall rate of the rebar and the detection effect of the hole are significantly improved,and the detection accuracy of the hole target is increased to 95.7%,and the average detection accuracy is increased to 97.7%.Based on the above algorithm research,using the YL-GPR ground penetrating radar hardware acquisition system,a reasonable detection scheme is designed to collect data on the road tunnel lining and subway section tunnel lining of a city in the south,and the performance verification of the algorithm model is carried out by using the data collected on site,which realizes the automatic identification of steel bars and hole targets in the tunnel lining radar image.The test results show the effectiveness and reliability of the tunnel lining detection technology proposed in this thesis,and have a good practical application prospect of engineering.