Underground structure internal defect identification method based on cross-hole radar and deep learning

Abstract

The invention discloses an underground structure internal defect identification method based on cross-hole radar and deep learning, and the method comprises the steps of building a numerical simulation data set which comprises a plurality of groups of data pairs of a two-dimensional cross-hole radar time domain oscillogram and a model profile dielectric constant distribution diagram; taking the numerical simulation data set as a learning sample, and training a defect identification model based on deep learning; and performing inversion on cross-hole radar data acquired in real time by using the defect identification model based on deep learning to obtain a corresponding defect dielectric constant distribution prediction image. The method provided by the invention can be applied to actual engineering scenes such as detection of underground diaphragm walls and pile foundations by using the cross-hole radar, engineering geological investigation and the like, and accurate, high-resolution and rapid identification of internal defects of underground structures is realized.

Description

technical field [0001] The invention relates to the technical field of underground structure detection, in particular to a method for identifying internal defects of underground structures based on cross-hole radar and deep learning. Background technique [0002] With the gradual increase in the development of underground space, the number of underground structures put into construction and operation is increasing. However, due to the influence of the natural environment, human activities, building materials, etc., cracks, voids, etc. will inevitably appear in the underground structures. Defects such as mud inclusion and corrosion of steel bars have changed the mechanical properties of the structure to varying degrees, which may lead to safety accidents and damage the safety of people's lives and property. Therefore, efficient and timely detection of internal defects of underground structures during the construction or operation period has become a key task. As a fast and e...

AI Technical Summary

Problems solved by technology

However, the position, shape and filling of the defect cannot be directly observed from the measured two-dimensional time-domain waveform of the trans-hole radar. Therefore, it is necessary to establish an identification algorithm to invert the dielectric rate, impedance, etc.), to reconstruct the defects inside the underground structure from the cross-hole radar data

[0003] At present, the commonly used cross-hole radar data inversion methods include tomography (CT) and full waveform inversion (FWI), but both methods have their limitations: the resolution of tomography is low; Algorithms are expensive to calculate and have high requirements for the selection of parameters

In recent years, deep learning methods rely on their ability to efficiently solve nonlinear mappings between data, and using deep learning methods to interpret GPR data has become a new development trend. However, due to the antenna layout of cross-hole radar and traditional GPR The method is different from the detection system, and there are huge differences in the distribution of the obtained data. Therefore, when the existing deep learning-based ground-penetrating radar recognition method is directly applied to interpret the cross-hole radar data, the ideal inversion effect cannot be obtained.

Images