Physics-Informed and Data-Driven Investigation and Enhancement of Asphalt Pavement Skid Resistance
Date Issued
May 2026
Author(s)
Zhong, Jingtao
The University of Tennessee
Advisor(s)
Baoshan Huang
Additional Advisor(s)
Hairong Qi
Lee Han
Hongyu Zhou
Abstract
Pavement skid resistance is a critical factor for roadway safety, yet current practice still lacks a performance-based framework that links mixture design, texture evolution, and friction guidance. This dissertation develops a systematic framework that advances pavement friction research from empirical deterioration analysis to physics-guided and data-driven generative modeling.
A laboratory-field combined protocol was developed to characterize long-term friction deterioration using an integrated novel testing setup, incorporating the Superpave Gyratory Compactor (SGC)-compacted specimens for the friction performance test. To further elucidate the mechanistic relationship between surface texture evolution and friction response, Close-Range Photogrammetry (CRP) was employed to reconstruct 3D texture models of specimens, which are linked to friction response through Persson-based contact modeling. A texture-image coupled fusion framework was then established by integrating domain-defined texture parameters with computer-encoded image features for machine-learning-based skid resistance prediction. For engineering implementation, explainable machine learning was applied to large-scale field data to assess preventive maintenance effectiveness and quantify key influencing factors. Finally, to address the scarcity and imbalance of high-quality texture datasets, a Power Spectral Density (PSD)-guided generative modeling framework was introduced to synthesize physically consistent multi-scale pavement textures for data augmentation and virtual experimentation.
Results show that ring-shaped specimen setup could be used as an alternative for mixture friction design, providing support for performance-based mixture design. CRP-based 3D reconstruction provides reliable texture descriptors and enables quantitative texture-friction interpretation when combined with contact modeling. Multimodal fusion improves prediction accuracy and robustness compared with unimodal approaches. Explainable analysis identifies treatment- and condition-dependent factors of friction restoration, supporting actionable maintenance planning. The PSD-guided generative model preserves multi-scale spectral characteristics and produces stable, realistic textures, enabling scalable dataset construction and simulation-oriented applications.
Degree
Doctor of Philosophy
Major
Civil Engineering
Embargo Date
May 1, 2029