Use of spiral cracking patterns in fracture characterization of asphalt materials

Abstract

A new approach for fracture characterization of asphalt materials through use of spiral cracking patterns is presented. Five different asphalt materials each at three different oxidative-aging levels, were utilized in this work. An efficient image processing framework using Convolutional Neural Networks (CNN) was implemented to analyze spiral cracking images. Results showed that spiral tightness parameter was sensitive to fracture energy, performance grade (PG), and oxidative aging level of asphalt binders. Machine learning approaches such as XGBoost and Elastic-Net regularized regression were employed to develop prediction models to estimate fracture in asphalt materials.