Ground Penetrating Radar Intelligent Inversion Methods For Tunnel Lining Internal Defects

During the long-term operation periods of the tunnel,some internal defects may easily occur,including cracks,voids and delamination.These defects may even induce safety accidents,such as block falling and collapse,which threaten the healthy service and long-life operation of the tunnel and the safety of people’s lives and property.Performing routine"physical examinations" of the tunnel structure and realizing precise diagnosis of tunnel lining internal defects are critical to ensure safe tunnel operation.Ground penetrating radar(GPR)has become the preferred method in the field of tunnel lining detection owing to its high efficiency,flexible operation and high resolution for structure internal objects.However,the current GPR data interpretation method mainly relies on manual judgment,which can only obtain the rough information of objects,including location and burial depth.GPR data inversion is capable of reconstructing the internal dielectric distribution of the detected structure based on the GPR detection data.However,traditional GPR inversion methods rely on the initial model,which are commonly difficult to accurately reconstruct the geometric and dielectric properties of the objects.Therefore,the study on accurate and efficient automatic GPR inversion method has become an urgent need for tunnel safety detection and health status assessment,which has important academic significance and practical engineering value.The main research work of this paper are as follows:(1)In order to solve the problem of precise inversion of tunnel lining structure complex internal defects,a GPR inversion method based on trace-to-trace encoding was proposed.This method fused the neighborhood information of single GPR trace to enhance the feature of each trace,and then further used the multiple fully connected layers to perform spatial feature mapping.It had the ability to extract the response feature of complex defect and align the feature space with permittivity map,thus enabling accurate permittivity inversion and precise contour imaging of complex tunnel lining internal defects.First,a simulated tunnel lining dataset was established,then a comprehensive comparison with the mainstream inversion methods was conducted to verify the superiority and accuracy of the proposed method for tunnel lining internal complex defects inversion.Finally,concrete model test was performed to demonstrate the applicability of the method to real GPR data.(2)The existing GPR inversion method is prone to cause distortion of the reconstructed object shapes or even the inconsistency of the inverted dialectic values when processing GPR data for consecutive and long survey lines.Thus,a GPR data inversion and identification method for consecutive and long survey lines by integrating convolutional neural network and recurrent neural network was proposed.This method fused the bi-directional correlation information between local GPR B-scans extracted from different locations in consecutive and long survey line to ensure the continuity and accuracy of GPR data inversion and identification method for consecutive and long survey lines.Additionally,it performed tunnel lining structure permittivity inversion and defect categories diagnosis simultaneously using a dual-task structure.Comprehensive simulation experiments and sandbox model test were conducted to verify the effectiveness of this method for GPR data inversion and identification with different consecutive and long survey lines lengths.(3)In order to address the problem of diverse frequencies of GPR equipment and lack of GPR data annotations in tunnel detection scenarios,a semi-supervised cross-frequency GPR inversion method that can mine the internal information of unlabeled GPR data was proposed.This method extracted common response feature of GPP B-Scan data with different frequencies by integrating the generative adversarial network and the Mean-Teacher architecture,and improved the inversion accuracy of target-domain GPR data in terms of global spatial structure coherence and pixel-level permittivity accuracy,which was able to achieve the accurate permittivity inversion of tunnel lining structure under different frequency and small amount of data labeling conditions.The simulation experiment and sandbox test results showed that this method was capable of achieving accurate inversion of GPR data with different frequencies.