Prediction of the fundamental viscoelasticity of asphalt mixtures using ML algorithms

The complex modulus (E*) and phase angle (δ) are fundamental viscoelastic (VE) properties of asphalt material, crucial for pavement performance analysis and modeling. This study presents a novel approach to predict the fundamental VE properties of asphalt mixtures using 20 study mixes with disparate mix design information including different Performance Graded (PG grade) binders, recycled asphalt (RAP) contents, with/without polymer modifications, and mix volumetric properties. By employing the XGBoost model for feature importance analysis, integrating the Bailey method to streamline gradation information, and utilizing master curve shape parameters for modeling, this study has effectively reduced data redundancy and dimensionality. Among four machine learning (ML) models compared, the Multi-Hidden-Layer Backpropagation Neural Network (Multi BP) model demonstrates superior applicability to the asphalt mixture dataset, providing accurate predictions for VE properties at any temperature and frequency combinations. Notably, this predictive model relies solely on the mixture design information as inputs, minimizing the need for laborious laboratory experiments. The developed model can be integrated with Mechanistic-Empirical (ME) design, predicting the variations in rutting and cracking performance of asphalt mixtures over the entire service life. This provides essential foundational input support for long-lasting pavement design. •20 study mixes with disparate mix design information were considered in this study.•Four ML models were employed to model |E*| and δ, and the model accuracies were compared.•Feature importance analysis effectively selects feature variables.•The Bailey method efficiently reduces data redundancy and dimensionality.•The developed model can be integrated with Mechanistic-Empirical (ME) design to predict pavement performance.