Balancing the performance of asphalt mixtures with their environmental and economic impacts is an important goal in sustainable pavement engineering. This study establishes an intelligent optimization framework that integrates physics-constrained neural networks (PCNN) with the non-dominated sorting genetic algorithm III (NSGA-III) to address this multi-objective problem. PCNN models were developed to predict cracking and rutting indices by embedding physical constraints into the training process. NSGA-III was applied to generate Pareto-front solutions that capture trade-offs among cracking resistance, rutting resistance, carbon emissions, and costs. To identify the most balanced mixture, the technique for order of preference by similarity to ideal solution (TOPSIS) was applied to rank the solutions along the Pareto front. The results show that the proposed PCNN model not only achieves higher predictive accuracy than traditional machine learning models, but also improves the generalization and physical consistency of asphalt performance evaluation. The optimized mix designs show better cracking and rutting resistance while also lowering embodied carbon emissions and costs. A clear trade-off between cracking resistance and rutting resistance is observed. The level of improvement varies depending on asphalt binder grade and nominal maximum aggregate size. Overall, this study establishes an integrated and data-driven methodology for low-carbon asphalt mixture design.
Cui et al. (Wed,) studied this question.