3D Imaging And Time-shift Change Feature Recognition Of Concrete Internal Defects Based On GPR

As a widely used construction material in large infrastructure,concrete structure is susceptible to internal defects such as voids and cracks due to material deterioration,human damage and environmental erosion,which seriously threaten the load-bearing capacity and safety of large infrastructure and even endanger public life and property safety.Ground penetrating radar(GPR)has become one of the main technologies for concrete internal defect detection because of its convenient operation and strong penetrating ability,and at the same time,the automatic recognition of internal defects based on ground penetrating radar detection data has become a research hotspot.However,at present,the intelligent recognition of concrete internal defects is mainly based on 2D GPR profiles,which is difficult to describe the 3D morphology and spatial location of the defects in an all-round way,and even more difficult to accurately recognize the changing features of the defects in the process of periodic multiple inspections,which restricts the effective assessment of the long-term safety status of concrete structures.In this paper,to address the above-mentioned problems,we study the 3D imaging and time-shift change feature recognition method of concrete internal defects based on GPR,and the main research work and results are as follows:(1)The 3D modeling and GPR data forwarding method of complex concrete internal defects is studied for the demand of high quality GPR data set for the deep learning model used for 3D imaging and time-shift change feature recognition of concrete internal defects.The GPR data response features under the influence of many factors such as complex defect morphology,spatial location,dielectric constant and line placement are analyzed,and 3D models of complex internal concrete defects with complex morphology,random spatial location and wide range of dielectric constants are constructed.The developmental changes of defects are introduced,multiple line locations and directions are considered to generate high-quality simulated GPR data under the 3D models of complex defects,which can provide data support for 3D imaging and time-shift feature recognition of concrete internal defects.(2)In order to solve the problem of 3D imaging of complex concrete internal defects,the method of 3D intelligent imaging of concrete internal defects based on multi-channel 2D GPR B-scan data is studied.The maximum intensity projection(MIP)algorithm and convolutional neural network cascade complex defects 3D imaging method is studied to extract the spatial feature information of GPR data and explore the morphology,location and high level semantic information of the defects,and then the 3D reconstruction module is designed to convert the 2D imaging results output from the convolutional neural network into 3D imaging results.The simulation an test validation results show that the proposed method can achieve the fine portrayal of the 3D morphology and accurate positioning of the spatial location of the complex defects inside the concrete.(3)In order to accurately recognize the time-shift change features of the concrete internal defect in the process of periodic multiple inspections,a method is constructed to recognize the time-shift features of the defect by fusing time sequence information.The projection preprocessing module and feature extraction with time sequence information fusion module are used to fully explore the spatial and temporal feature information of the defect reflection signal in the GPR data.Then,the defect change feature extraction module is built to correct the 3D imaging results at multiple time points voxel by voxel to highlight the change of the defect at different time points precisely,and finally realize the time-shift change feature recognition of the concrete internal defects.On this basis,the effectiveness of the proposed method for timeshift feature recognition is verified based on the simulated GPR data and the actual GPR data.