| While tunnel construction in mountainous areas,construction accidents caused by geological factors occur frequently.The detection of geological information of surrounding rock is the key to ensuring the safety of construction.However,due to the complexity and invisibility of the underground space in mountainous areas,the current methods are still tricky to comprehensively and accurately explore the surrounding rock geological information.There will inevitably be omissions of adverse geological bodies or wrong decisions on design and construction parameters,eventually leading to accidents.Therefore,it is necessary to study the inverse method of the geological information of surrounding rock for the mountainous tunnel to carry out targeted support and construction control and avoid construction accidents.In order to address the needs in mountainous tunnel construction,this study is devoted to developing a set of detection and inversion methods for the geological information of the surrounding rock.By analyzing and summarizing the advantages and disadvantages of the existing geological information detection methods,this paper studied in detail theoretical analysis,data processing,algorithm development,and platform development.Firstly,the interference caused by the complex landform and geological conditions of mountainous to the detection data is serious.The corresponding data processing methods for the surface elevation point,groundpenetrating radar,and tunnel three-dimensional laser point cloud data are proposed.The result provides data support for the subsequent geological information inversion and analysis.Secondly,a multi-physical field inversion method of surrounding rock based on Bayesian theory is proposed.Then,a discontinuities density prediction model based on depth learning is established using the "three instantaneous" characteristics of GPR as input.The applicability of the model is discussed through numerical simulation.The automation extraction algorithm of discontinuities information based on threedimensional laser point cloud data establishes the data for training,and the "random prediction determination" discontinuities network generation method is proposed based on discontinuities density prediction.Finally,based on the research results,the geological information inversion and visualization platform of the surrounding rock of the mountainous tunnel is developed.It has been applied in the Wulaofeng tunnel of "Jian Ge Yuan" expressway and Qinghe tunnel of "Zhao Lu" expressway in Yunnan Province.The main research results of this paper are as follows:(1)The topographic exploration of the mountainous tunnel faces large topographic relief and dense surface vegetation coverage problems.The improved progressive tin encryption(PTD)filtering algorithm realizes the effective filtering of non-ground points in mountainous areas.Compared with the traditional PTD algorithm,the overall filtering error in the mountain area is reduced by more than 16%,and the class I error is reduced by 15%.Aiming at the interference of multiple and diffraction waves of GPR data caused by complex geological conditions of tunnel surrounding rock,a bifrequency coherent backward projection algorithm based on 100 MHz and 200 MHz radar signals is proposed.Numerical simulation results show that the algorithm effectively weakens the energy of multiples and diffracted waves based on ensuring high resolution and has good anti-interference ability.In order to unify the coordinate system of three-dimensional laser multi survey stations of mountain tunnel,the transformation of the point cloud from relative coordinate system to absolute coordinate system is realized in combination with the total station.Through the coherence of multiscale spatial features of the point cloud,the transformation matrix is optimized to realize the registration of the point cloud.The measured results show that the point cloud coherence of different stations is increased by 0.08.(2)The high-density resistivity forward modeling program and Bayesian inversion program are compiled through MATLAB.The probability inversion framework of surrounding rock multi-physical field of mountain tunnel is established,including formation lithology,rock fracture degree,fracture mud content,and fracture water content.The forward simulation,parameter optimization,and post-sampling process are emphasized.The effects of model grid size,parameter compression ratio and borehole density on inversion velocity and inversion results are discussed through numerical simulation,which provides a reference for the selection of inversion parameters.Then,the reliability of the inversion algorithm is verified by numerical simulation and field experiments.The results show that although the inversion error is caused by the existence of geological structure in the local lithology and fracture filling parameters,the effective inversion is still carried out for the lithology interface,rock mass fracture and solution cavity development area,and the fluidity of fracture filling(3)A discontinuity network inversion algorithm based on a residual neural network is proposed in this paper.The "three instantaneous"(instantaneous amplitude,instantaneous phase,and instantaneous frequency)parameters of GPR is the input,and discontinuity density is the output.The feasibility is first discussed through numerical simulation and the effects of detection depth,structural plane parameters,and data acquisition error on the prediction results.According to the error distribution,the best prediction effect can be obtained when the angle between the structural plane and the survey line is 20°～70°,and the linear density of the structural plane is 0.03～0.3 bars/cm.Then,the discontinuity recognition and information extraction algorithm of 3D laser point cloud combined with point cloud classification and point cloud segmentation is proposed,which provides technical support for constructing the data set in practical engineering.Based on the spatial expanded convolution network model,the point cloud classification method breaks through the limitation that the traditional convolution kernel can only extract data features through two-dimensional plane scanning and realizes the automatic extraction of three-dimensional features of point cloud at different scales in space.The overall classification accuracy reaches more than 92%.The point cloud segmentation algorithm based on dynamic DBSCAN avoids the clustering error caused by the surface fluctuation of the discontinuities and the lack of information in meshing and speeds up the clustering speed.Finally,through GPR and3 D laser scanning technology,the sample database of residual network model training is constructed.Based on the prediction data,the "random prediction determination" structural plane network generation algorithm is proposed,which effectively solves the error caused by the lack of deep information in the generation process of discontinuities networks.(4)Combined with the needs of engineering,this paper develops geological information inversion and visualization platforms for the surrounding rock of mountainous tunnel based on MATLAB and Python and introduces the framework of the platform and the functions of each module.Through the application in the geological information detection of Wulaofeng tunnel of "Jian Ge Yuan" expressway and Qinghe tunnel of "Zhao Lu" expressway in Yunnan,the landform along the tunnel,lithology of tunnel surrounding rock,multi-physical field parameters of tunnel surrounding rock,a network of discontinuities and three-dimensional deformation of tunnel surrounding rock is detected and inverted.The complete engineering application example is given... |
PDF Full Text Request