基于探地雷达时频特征的水泥路面脱空识别
Identification of Cement Pavement Defects Based on GPR Time-Frequency Features
张军 1姜文涛 1余秋琴 2李有鑫 1罗婷倚2
作者信息
- 1. 长安大学道路施工技术与装备教育部重点实验室,西安 710064
- 2. 广西北投公路建设投资集团有限公司,南宁 530028
- 折叠
摘要
脱空病害在水泥路面中不可避免,危害路面结构安全性,迫切需要建立脱空病害区域的准确定位方法.采用探地雷达(GPR)为无损检测工具,通过GPR信号时频特征的提取和正反演实验构建了脱空病害的特征数据集,将特征集进行PCA降维,以PCA主元为输入、正常和脱空为输出分别建立SVM和ANN脱空识别模型;在获得脱空的GPR信号上提出窗口能量法判断脱空区域的深度,在实际路面上对脱空识别进行验证.结果表明:时频特征可表征脱空病害,包含病害的部分数据集更适合用于建模识别;ANN的识别准确度优于SVM,窗口能量法可有效定位脱空区域的深度,并对噪声具有一定的鲁棒性,在室内模型中定位精度平均误差1.51％.研究结果可为路面精准养护提供科学依据.
Abstract
The occurrence of void defects in cement pavement is inevitable and poses a threat to the structural safety of roads.It is urgently necessary to establish an accurate positioning method for void defect areas.Ground-penetrating radar(GPR)was used as a non-destructive testing tool to extract time-frequency features from GPR signals and construct a feature dataset for void defects through forward and inverse experiments.The feature set was then subjected to PCA dimensionality reduction,and SVM and ANN models were respectively established with PCA principal components as input and normal/void output to identify voids.The window energy method was proposed to determine the depth of the void area in the GPR signal of the void.The identification of voids was verified on actual pavement surfaces.The results showed that time-frequency features could characterize void defects,and the subset of data containing parts of the defects was more suitable for modeling and identification.The recognition accuracy of ANN was superior to that of SVM,and the window energy method could effectively locate the depth of the void area and had robustness against noise.In the indoor model,the average positioning error was 1.51％.The research results provide a scientific basis for accurate maintenance of road surfaces.
关键词
道路工程/水泥路面/无损检测/脱空识别/探地雷达Key words
road engineering/cement pavement/NDT/void recognition/GPR引用本文复制引用
基金项目
广西交通运输行业重点科技项目(19-09)
陕西省自然科学基础研究计划(2022JM-249)
陕西省交通厅项目(20-30X)
陕西省重点研发计划(2023-YBSF-104)
出版年
2024
NETL
NSTL
万方数据