Enhancing understanding of asphalt mixture dynamic modulus prediction through interpretable machine learning method

Abstract

Dynamic modulus is a key parameter in pavement design and pavement mechanics analysis. It is essential to accurately predict dynamic modulus and study the relationships between influencing factors and dynamic modulus. In this study, a hybrid prediction model is developed based on Extreme Gradient Boosting (XGBoost) and Whale Optimization Algorithm (WOA). Based on this model, the effects of asphalt binder properties, test condition, asphalt mixture volume parameters, and asphalt mixture gradation on dynamic modulus are analyzed. The contribution of each variable to the model predictions is quantified through Shapley Additive Explanations (SHAP), and the interaction between dynamic modulus and influencing factors is evaluated by Partial Dependence Plot (PDP). The results indicate that the WOA-XGBoost model has excellent accuracy and robustness in predicting dynamic modulus. The three most important factors affecting dynamic modulus prediction results are the complex shear modulus of binder, the test temperature and the asphalt binder viscosity. The increase in dynamic modulus can be achieved through the utilization of asphalt binders characterized by relatively large complex modulus, high viscosity, small phase angle, and high asphalt PG indexes. Reducing the effective binder volume and air voids of the mixture, optimizing the mixture gradation to a suitable level, and increasing the mineral powder content can also lead to the increase of dynamic modulus. Besides, low test temperature and high frequency generally mean a large value of dynamic modulus. This study clarifies the impact of influencing factors on the performance of asphalt mixtures based on machine learning, which lay a foundation for the intelligent design of asphalt mixtures.