Date of Award
Summer 7-10-2025
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Majeed Hayat
Second Advisor
James Richie
Third Advisor
Richard Povinelli
Abstract
Dowel bars and tie rods in concrete pavement hold concrete together when they are placed and positioned correctly and according to specifications provided by the federal and state departments of transportation. However, unintentional misalignments and omissions during pavement construction can adversely impact the integrity of pavements, leading to cracks and other types of pavement failure. Analyzing the positions of the dowel bars and tie rods, although crucial, can be tedious and laborious. Ground-penetrating radar (GPR) can be used to detect objects within a material of mostly constant permittivity. When GPR is combined with multiple scans across a road, a three-dimensional model of the road is created, which is referred to as 3DGPR. This project uses 3DGPR data collected over a concrete roadway to find dowel and tie rods in concrete to extract and assess their geometrical features. Once a civil engineer has identified the locations of the dowel bars and tie rods, they can assess whether the bar/rod-laying machinery is functioning properly, if the concrete is settling correctly, and where future roadwork will need to be planned. Manual analysis of GPR data is incredibly time-consuming. The goal of this thesis is to develop and test an algorithm for the automatic detection and assessment of the tie rods and dowel bars in concrete roadways. To locate the dowels and tie rods within the road, a mask is first created from a 2D depth slice of the 3DGPR, which identifies all the rods and joints in two dimensions. Then the depth in time is found using a cross-correlation function. From this, we can extract individual rods from their respective locations using clustering. Then, a curve-fitting function, in conjunction with the ranging equations, is applied to find the true depth in distance units, and the data and respective rods are further filtered to remove spurious signals. Feature extraction is then performed, where a rod can be represented by its center location in three dimensions along with its orientation angles in two directions. These features, in turn, can be used to analyze and assess the alignment and placement attributes of the pavement. The validation results show that velocities and depths are consistent across all rods, suggesting that the multiple datasets are handled in the same way. Additionally, comparing the depths to the overlay images confirms that the 2D location on the road is correct, and because these are used to find the depth with a curve-fit approach, the extracted depth is therefore accurate too. Another validation measure was a third-party analysis of 1.6 miles of data containing 2552 rods by Infrasense, inc. They found the probability of missing a tie rod detection to be 1.3%, and the probability of falsely labeling a missed rod as present is 0.03%. They also determined 0.2% of tie rods were at false depths, and 0.2% of tie rods detected were not in the roadway. From this analysis, 26 missing sections of tie rods were detected successfully along with 2520 rods. Possible limitations discovered are when tie rods are near a road joint, or when a dowel rod is on the edge of the scan equipment. Methods around these limitations are described in the conclusion.