Research On Advanced Detection Method Of Metro Shield Machine Based On Generative Adversarial Networks

Tunnel Boring Machine(TBM)has been widely used in subway tunnel construction because of its safe and efficient construction.In the process of TBM construction,it often encounters engineering geological disasters such as collapse,sudden water gushing,rock burst and so on,which brings serious safety problems and great economic losses to subway tunnel construction.Therefore,it is very necessary to detect the geological conditions in front of TBM in advance.Aiming at the difficult problem of advanced detection in TBM construction environment,the focused induced polarization advanced detection technology is adopted and the generative adversarial network algorithm is introduced to carry out two-dimensional resistivity inversion imaging of unfavorable geological body in front of TBM.The specific research work is as follows:(1)Aiming at the problem that bore-tunneling electrical ahead monitoring(BEAM)system cannot be imaged due to its low utilization rate of spatial distribution information,the excitation mode and measurement mode of BEAM system are improved to provide solution conditions for inversion imaging of geological conditions in front of the tunnel face.(2)The finite element forward mathematical model of focused induced polarization for shield advanced detection is established,and the Comsol numerical simulation software is used to forward simulate the response characteristics between the size,shape,location,resistivity of abnormal body and mesured voltage,which provides a prior information for further inversion imaging.(3)In view of the problem that the linear resistivity inversion algorithm depends on the initial model and the imaging real-time performance is poor,a two-dimensional resistivity inversion imaging method based on generative adversarial network is proposed.Taking the voltage data and resistivity distribution obtained from the forward modeling as training sample sets,the appropriate generative adversarial network model is established and trained.The trained generative adversarial network model is used to carry out 2D resistivity inversion imaging test on a single low-resistivity abnormal body model and a multiple low-resistivity abnormal body models,and the imaging results are compared with those of Gauss-Newton linear inversion algorithm.(4)According to the similarity principle,a physical model experiment platform for shield advanced detection is designed to verify the feasibility of the excitation mode and measurement mode of the improved BEAM method and the feasibility of the shield advanced detection method based on generative adversarial network,thus laying a theoretical and experimental foundation for the advanced geological detection research on the project site.