Data-driven efficient prediction and design optimisation for high-temperature performance of asphalt mixture

Accurately evaluating the high-temperature performance of asphalt mixtures can provide useful performance-oriented support for their design and production. However, conventional performance testing and mixture design methods often involve long cycles and high costs. This study proposed a data-driven and efficient framework for predicting high-temperature performance and enabling automated design assistance of asphalt mixtures. A database was established from laboratory tests covering asphalt binder properties, volumetric characteristics, gradation-related descriptors, and dynamic stability (DS), followed by Grey Relational Analysis (GRA) to screen highly relevant 21-dimensional feature set. A Gaussian Process Regression (GPR) machine learning (ML) model was then developed to predict the dynamic stability (DS) of asphalt mixtures and benchmarked against Support Vector Regression (SVR) and Artificial Neural Network (ANN) models. The results show that the GPR model achieved superior performance in prediction accuracy, generalisation capability, and computational efficiency, with a coefficient of determination (R2) of 0.9889, and a normalised root mean square deviation (NRMSD) of 0.0325. To enhance interpretability, explainable machine learning (XML) techniques were applied to analyze feature contributions within the GPR model, with results compared to those from GRA. Subsequently, the GPR model was integrated with Bayesian Optimization (BO) to develop an automated design optimisation approach for high-temperature performance. Experimental validation yielded a design error of 6.95%, confirming the method’s practical applicability. Furthermore, a multi-module software system was implemented to support intelligent design and optimisation of asphalt mixtures for high-temperature performance. These advancements provide both theoretical foundations and practical tools for promoting digital transformation in pavement engineering.HighlightsData-driven and explainable machine learning models are developed for predicting the dynamic stability of asphalt mixtures.SHAP focuses more on macro-level material characteristics compared to Grey Relational Analysis.A efficient automated design framework for asphalt mixtures targeting dynamic stability is proposed and validated.A multi-module integrated software is developed for asphalt mixtures design.