Algorithms for asphalt concrete density and moisture content prediction using ground penetrating radar data

Cao, Qingqing

Permalink

https://hdl.handle.net/2142/115572 Copy

Description

Title

Algorithms for asphalt concrete density and moisture content prediction using ground penetrating radar data

Author(s)

Cao, Qingqing

Issue Date

2022-04-19

Director of Research (if dissertation) or Advisor (if thesis)

Al-Qadi, Imad L.

Doctoral Committee Chair(s)

Al-Qadi, Imad L.

Committee Member(s)

• Roesler, Jeffery R.
• Popovics, John S.
• Hajj, Ramez M.

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Ground-penetrating radar
• asphalt pavement
• non-destructive testing
• moisture content
• numerical simulation

Abstract

Asphalt concrete (AC) pavement density is directly related to pavement’s structural capacity and its service life. Hence, it has been used for AC quality control and quality assurance during construction. For the past two decades, ground-penetrating radar (GPR) has been used to predict AC pavement density. However, this method does not accurately predict AC density when internal moisture is present. The presence of excessive moisture in flexible pavements causes stripping, raveling, and potholes due to loss of adhesive bond between the aggregate surface and asphalt binder film. In addition, it is critical to monitor moisture content for cold-in-place recycling (CIR) and cold central-plant recycling (CCPR) until it drops to an acceptable threshold before placing pavement overlay or opening the road to traffic. Thus, there is an increasing need for the non-destructive detection and monitoring of moisture content in AC pavements. This study investigates the feasibility of using GPR to estimate the moisture content of AC pavements, employing several models and algorithms to predict both moisture content and density in AC pavements. One challenge of continuous GPR measurements is that the corresponding calculated dielectric constant could be affected by signal stability and antenna height. This would jeopardize the accuracy of the AC density prediction. Using gradient descent and least mean square methods, a least-square adaptive filter was developed to reconstruct the received signal to improve its stability. A height correction was also developed using a power model to correct for the height-change impact. The proposed filter and height correction method were assessed via lab and field tests. The least-square adaptive filter was found to improve signal stability by 50%. The height-correction method removed the effect of shifting antenna height almost entirely. To characterize the electromagnetic wave propagation inside AC pavement, a three-phase numerical method was developed to consider aggregate, binder, and air void components. The goal was to correlate AC specific gravity with its dielectric properties. Using a numerical simulation model, the effects of air-void content, asphalt content, aggregate gradation, and aggregate dielectric constants on the GPR measurements were studied. The effects of vertical or longitudinal heterogeneity on the reflected signals are also discussed. Finally, the model was validated using field data. The developed numerical simulation method was further improved to quantify the effect of internal moisture content on AC pavement dielectric properties. The numerical model was validated using GPR surveys on CIR treated pavements. A moisture-prediction formula was also derived from the simulation to correlate the dielectric constant with the moisture content for non-dry AC pavement. The proposed method, based on numerical simulation, was validated using field test results. This demonstrates the ability of GPR to monitor moisture variation in AC pavements. Using the moisture-prediction formula, a testing approach for predicting internal moisture content of emulsion-stabilized CIR and CCPR layers was developed. GPR signals were preprocessed to improve the signal stability and accuracy. The testing approach was validated and applied to field data collected from IL-100, IL-61, IL-116, and Rt-509 highways in Illinois. The average error of predicting moisture content was 0.33% for the four field projects. To decouple the effect of pavement density and moisture content on the dielectric constant, a deep learning method called “Mois-ResNets” was designed and validated for predicting the internal moisture content of AC pavements. Datasets of GPR signals and moisture contents were collected from field tests and numerical simulations. The architecture of Mois-ResNets contains a short-time Fourier transform, followed by a deep residual network. The testing results showed that Mois-ResNets can achieve a classification accuracy of 91% on testing datasets, outperforming conventional machine learning methods. The prediction from the Mois-ResNets was further used successfully in pavement density prediction when internal moisture content exists in AC pavement.

Graduation Semester

2022-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Qingqing Cao

Owning Collections