Aggregate-level 3D analysis of asphalt pavement deterioration using laser scanning and vision transformer

Understanding asphalt pavement microscopic mechanisms deterioration is essential for improving long-term performance, durability, and maintenance strategies. Traditional analyses often rely on 2D texture profiles, overlooking the three-dimensional morphology of aggregates. To address this gap, this paper proposes an advanced microscopic evaluation method based on high-precision 3D laser scanning and deep learning. A Vision Transformer model is developed to segment aggregates using reflection intensity and 3D range data, achieving 85.75 % mean pixel accuracy and 83.78 % mean intersection over union, outperforming the PointGroup. Building on accurate segmentation, 3D morphological indicators such as sphericity and normal vector distribution are extracted to quantify both overall and local aggregate features. The method is validated through indoor accelerated loading and outdoor tracking experiments on raveling and polished pavement. Results demonstrate the framework's effectiveness in assessing deterioration extent and progression, offering deeper insights into failure mechanisms and supporting the development of more resilient asphalt materials.