Physics-Informed Machine Learning for Predicting Rutting Resistance of Asphalt Mixtures Using Aggregate Shape Parameters

Abstract

Rutting remains one of the most consequential distress modes in asphalt pavements, affecting safety, ride quality, drainage, and life-cycle cost. Conventional empirical and mechanistic models can predict rutting resistance, but they often struggle to represent the coupled influence of aggregate morphology, binder rheology, mixture gradation, compaction, traffic loading, climate, and construction variability. This paper examines physics-informed machine learning as a system-level strategy for predicting rutting resistance of asphalt mixtures using aggregate shape parameters. The central argument is that aggregate shape is not merely a descriptive input but a meso-structural control variable that conditions contact fabric, interlock, stress distribution, and permanent deformation accumulation. Physics-informed machine learning offers a way to embed mechanical constraints, viscoelastic behavior, conservation principles, and domain knowledge into data-driven models, thereby improving extrapolation, uncertainty awareness, and engineering credibility. The paper discusses the architecture of such systems, including imaging and characterization pipelines, laboratory and field data integration, model training and validation, deployment in pavement management, and governance arrangements. It also considers sustainability, equity, robustness, and policy implications, especially where predictive tools influence material specifications, procurement, and maintenance prioritization. The analysis concludes that physics-informed machine learning can support more resilient and sustainable asphalt infrastructure, but only if technical performance is matched by transparent validation, institutional accountability, and careful attention to the socio-technical conditions of deployment.