Abstract: Take three kinds common subgrade material (silt, medium sand and gravel sand) as test objects, the GPR was used to test dielectric constant of many samples and the water content and density were tested by laboratory soil tests. The volume mixing models for dielectric constant (root mean square model and linear model) were verified and the MATLAB program was used to process the test value and calculated value of dielectric constant. The result shows there is great difference between the test value and model calculated value of dielectric constant, but there is a good relativity between them. The relativity is used to improve the original models so as to acquire the subgrade compactness by an inversion. Take silt for example, plenty of tests were carried on, the MATLAB program was used to do multiple nonlinear regression and the empirical model was set up. The improved dielectric model and the empirical model for silt were applied to the practical engineering and the test results were in accordance with sand replacement method.

Key words: road engineering, subgrade soil, compactness, dielectric model, ground penetrating radar(GPR), quality detection

摘要： 以3种常见的路基材料(粉土、中砂及砾砂)为试验对象，引入探地雷达测试的各试样介电常数，通过室内土工试验测试各试样的含水量与密度，对介电常数的两种体积混合模型(均方根模型和线性模型)进行验证，并利用MATLAB软件对介电常数的实测值与计算值进行处理。结果表明:介电常数实测值与模型计算值有很大出入，但两者之间有较好的相关性，利用这种相关性可对原模型进行改进，进而可利用改进后的模型反演路基压实度。以粉土为例，通过大量试验，利用MATLAB软件进行多元非线性回归，建立该粉土路基的介电常数经验模型。将改进后的介电常数模型及粉土经验模型应用于实际工程中，并与灌砂法得到的压实度进行对比，结果基本一致。

关键词: 道路工程, 路基土, 压实度, 介电常数模型, 探地雷达(GPR), 质量检测