Image Instance Segmentation,Evaluation And Grouting Prevention For Apparent Defects Of Shield Tunnel Linings

Affected by soft soil,poor workmanship during the construction process and the nearby construction disturbances,operated shield tunnels often suffer from apparent lining defects（e.g.,leakage,spalling,and cracks）and deformation.These defects promote each other to deteriorate,which will affect the long-term safe operation of shield tunnels.Therefore,it is vital to conduct fast and accurate detection of lining defects,objective and accurate evaluation of structural status,and effective defect control measures to maintain the shield tunnel’s operation.Current defect inspection,evaluation,and prevention techniques have suggested certain limitations.Traditional image-based defect-recognition methods have difficulties performing well when the image backgrounds are complex,restricting such methods from being widely adopted in tunnel lining inspection.Most tunnel health evaluation methods fail to consider the ambiguity,randomness,and correlation of the evaluation indicators in a unified manner.The evaluation results are affected by the weights assigned to the defects.Polyurethane and other organic polymer materials are often grouted into the soil behind tunnel linings to control the water leakage defect.However,polyurethane foam has poor long-term performance and low strength and is often associated with further leakages after grouting treatment.This study is financially supported by NSFC and NKRDPC（No.51778474,52130805,and 2021YFB2600804）.The methods for rapid and accurate tunnel lining defect recognition,structural evaluation,and seepage control using grouting are thoroughly studied via different approaches（e.g.,theoretical analysis,physical model test,and numerical simulation）.The main contributions and findings of this study are as follows:（1）An integrated model for image classification,instance segmentation,and quantification of apparent defects from shield tunnel lining images was proposed.A Res Net-101-based classification model and a Mask R-CNN-based instance segmentation model for leakage and spalling defects were trained and tested using the established datasets.Through optimizing the internal modules of the Mask R-CNN model,the model can use the learned semantic features of defects to quantify the defect area.An integrated model is proposed through optimization and integration of the classification model and the instance segmentation model.Compared with traditional image processing methods and the FCN method,the integrated model can quickly complete the whole classification,segmentation,and quantification of each lining defect image.Moreover,the integrated model can overcome the influence of distractors on the lining surface,and extract and quantify defect information quickly and accurately.（2）The PANet-A*model for fine instance segmentation and quantification of shield tunnel lining cracks was established.The proposed model solves twofold problems:1)cracks identified by deep learning algorithms were discontinuous,and 2)the crack skeleton extracted by skeletonization algorithms at the boundary of the crack tip was not accurate enough.The PANet model was adopted for refined instance segmentation of cracks.An additional semantic branch was constructed and added to the PANet model to extract the semantic information during the instance segmentation process.The morphological closing operation and the A*algorithm are also incorporated into the semantic branch to compute the crack length and width using the semantic information obtained from the semantic branch.Testing results showed that cracks segmented by the proposed model have fewer discontinuities than the same cracks segmented by the Mask R-CNN model,making the segmentation of cracks more refined and improving the crack segmentation performance.The A*algorithm can calculate the length and width of the crack more accurately and avoid the"zigzag"skeleton problem that occurred in the skeletonization algorithm,when extracted the skeleton at the boundary of the crack tip.（3）A multi-source heterogeneous fusion evaluation method based on cloud-Copula-evidence theory was proposed to evaluate the structural health status of shield tunnels in operation.This method adopted the cloud model to consider the fuzziness and randomness of the evaluation indicators,used the Copula function to assess the correlation between the evaluation indicators,and used the D-S evidence theory to fuse the classification results of each defect indicator.The proposed method was applied to evaluate a tunnel interval of Shanghai Metro Line.The performance was compared with the shield tunnel serviceability index（TSI）method and the health evaluation method in the Chinese national standard.The comparison results show that the proposed method can objectively and accurately evaluate the health level of the tunnel linings and has a strong capability to distinguish the severity of different defects.（4）A laboratory apparatus for simulating microbial grouting behind the shield tunnel linings were developed.The indoor physical similarity model experiment based on this apparatus was conducted using different grouting methods,grouting pressure,and bacterial activity to investigate the following grouting responses:the diffusion law of microbial grouting behind the tunnel lining,the distribution law of Ca CO₃ crystals generated by the grouting reaction in the soil pores,the variation law of pore pressure near grouting hole of the tunnel segment,the permeability reduction effect for the lining seam and surrounding soil,and the improvement effect of soil strength.The testing results showed that the microbial grouts diffuse spherical behind the tunnel lining.Higher content of precipitated Ca CO₃ crystals in soil pores results in a better permeability reduction effect and strength improvement effect for surrounding soil.During the grouting process,it was observed that the increasing percentage of pore pressure in the region above and below the grouting hole was different according to different grouting positions on the tunnel segments.（5）The variation law of the tunnel segment’s pore pressure,joint opening,joint dislocation,and internal force was revealed from the numerical simulation results of different grouting positions in the tunnel segments.The method of applying pore pressure and expansion pressure to the soil element nodes in the grouting area was used to investigate the variation law of pore pressure during the grouting process and the influence of grouting compaction on the deformation of the joint and the internal force of the segment.The numerical results show that the grouting compaction and the back pressure generated by the grouting pressure have a modest effect on the opening and dislocation of lining joints.Grouting operation reduces the positive bending moment and axial force of the waist joint and increases the bottom joint’s negative bending moment and axial force.