Physics-informed machine learning for asphalt mixture dynamic modulus: Balancing high accuracy with stability and rationality

The dynamic modulus, a critical parameter for asphalt mixture viscoelasticity and pivotal for pavement engineering, conventionally faces determination challenges due to high costs, extensive data dependency, and limited physical insight. To overcome these, this study introduces a novel dual-driven predictive framework that synergistically integrates physical knowledge with data science principles. Initially, Grey Relational Analysis (GRA) was employed to quantify the influence weights of twelve material composition parameters. Subsequently, leveraging K-Nearest Neighbors (KNN), Gradient Boosting Decision Tree (GBDT), and Random Forest (RF) algorithms, a sensitive feature space was meticulously constructed. Critically, established positive correlation relationships between Softening Point, Asphalt Type, and dynamic modulus were encoded as physics-based constraints within a Backpropagation Neural Network (BP-NN), culminating in a physics-informed, data-driven dynamic modulus prediction model. Rigorous multi-error index cross-validation and stability evaluation confirmed model robustness. Results revealed Void Volume and Asphalt Aggregate Ratio as dominant dynamic modulus influencers, while Apparent Density of Mineral Powder and asphalt Specific Gravity could diminish generalization. Compared to data-driven counterparts, the proposed physical-data-driven model exhibited a marginal increase in Root Mean Square Error (RMSE) (41–46 MPa) and Mean Absolute Error (MAE) (28–29 MPa), yet consistently maintained Coefficient of Determination (R²) > 0.96 and a significantly narrowed prediction confidence interval. These findings underscore that embedding physical constraints substantially enhances model stability and rationality, ensuring adherence to engineering principles while largely preserving high accuracy. This research offers a robust methodology for accurate dynamic modulus prediction and optimized material design in asphalt mixtures, advancing scientific road design and sustainable pavement engineering.