探地雷达目标识别方法及其在隧道衬砌检测中的应用研究

发布时间：2019-06-22 12:23

【摘要】：随着我国高速公路干线网的不断完善和发展,公路隧道的规模与数量也在不断增长。为了避免隧道开挖过程中及建成后的运营阶段发生事故而影响到人们生命财产的安全,需要运用合适的物探工具对隧道衬砌进行质量检测。探地雷达以其快速、高效、高分辨率及无损性等优点而受到隧道工作者的青睐。对于衬砌检测中出现的衬砌脱空、不密实体、空洞充水三种典型病害,可归类于形状、大小各异的空洞中填充了不同物质。本文通过使用探地雷达对物理模型进行探测获取试验数据,研究、建立空洞不同填充物质的分类识别方法,进而将此方法运用于隧道衬砌检测的病害分类识别中,主要研究内容如下:(1)通过物理模型试验分别获取空洞填充干沙、碎石、泥浆、水等物质相应的初始雷达数据,并运用本文总结的探地雷达数据预处理流程对其进行处理;从而进一步总结分析隧道衬砌检测中常见的衬砌脱空、不密实体、空洞积水三种病害典型雷达图像。(2)介绍了探地雷达振幅、频谱、相干性三种属性提取技术,结合雷达属性的物理意义,选取均方根振幅、平均波峰振幅、时域平均能量、相似系数及-3dB带宽平均相位五种雷达属性作为探地雷达目标分类识别的特征参数。(3)通过建立空洞不同填充物的高斯过程机器学习二元分类(GPC)模型,以均方根振幅、平均波峰振幅、时域平均能量、相似系数及-3dB带宽平均相位五种雷达属性作为矢量输入,成功识别出物理模型各空洞中分别填充空气、碎石、泥浆和水时四种物质,并给出相应预测概率。(4)将经物理模型试验验证后的GPC预测模型运用到岑溪大隧道衬砌检测实际工程中,成功对脱空、不密实体、空洞充水三种衬砌病害进行分类识别。说明本文所提出的探地雷达隧道衬砌检测高斯预测模型是可行的,有很好的应用前景。
[Abstract]:With the continuous improvement and development of highway trunk network in China, the scale and number of highway tunnels are also increasing. In order to avoid accidents in the process of tunnel excavation and after completion and affect the safety of people's lives and property, it is necessary to use appropriate geophysical tools to detect the quality of tunnel lining. Ground penetrating radar (GPR) is favored by tunnel workers because of its advantages of fast, high efficiency, high resolution and no damage. For the three typical diseases of lining emptiness, indense solid and empty water filling in lining detection, they can be classified into voids of different sizes and shapes and filled with different substances. In this paper, the experimental data are obtained by using ground penetrating radar (GPR) to detect the physical model, and the classification and identification method of different filling materials of voids is established, and then this method is applied to the disease classification and identification of tunnel lining. The main research contents are as follows: (1) the initial radar data of dry sand, gravel, mud, water and other substances filled with voids are obtained respectively through the physical model test. The data preprocessing process of ground penetrating radar (GPR) summarized in this paper is used to process GPR data. In order to further summarize and analyze the typical radar images of three kinds of diseases such as empty lining, undense entity and empty water in tunnel lining detection. (2) three attribute extraction techniques of GPR amplitude, spectrum and coherence are introduced. combined with the physical meaning of radar attribute, root mean square amplitude, average wave peak amplitude and time domain average energy are selected. Five radar attributes of similarity coefficient and average phase of-3dB bandwidth are used as characteristic parameters of target classification and recognition of ground penetrating radar (GPR). (3) five radar attributes, root mean square amplitude, average peak amplitude, average energy in time domain, similarity coefficient and average phase of-3dB bandwidth, are used as vector inputs by establishing machine learning binary classification (GPC) model for different fillers of voids. The four substances filled with air, gravel, mud and water in each cavity of the physical model are successfully identified, and the corresponding prediction probabilities are given. (4) the GPC prediction model verified by the physical model test is applied to the actual lining detection project of Cengxi tunnel, and the three lining diseases of emptiness, indense entity and empty water filling are successfully classified and identified. It is shown that the Gaussian prediction model for GPR tunnel lining detection proposed in this paper is feasible and has a good application prospect.
【学位授予单位】：广西大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U455.91

【参考文献】