AI-Based Multiobjective Optimization for Balanced Mix Design of Asphalt Concrete with Generative Adversarial Network-Based Data Augmentation: Integrating Pavement Rutting, Cracking, and Cost

Abstract Balanced mix design (BMD) is a more rigorous asphalt concrete design framework that requires the evaluation of rutting and cracking resistance through lab tests. However, these tests add complexity, making BMD a labor- and time-intensive process. Furthermore, lab tests alone cannot accurately evaluate the performance of asphalt concrete because it depends on the entire material structure for specific loads and environmental conditions. This study innovatively transformed the laboratory BMD procedure into an intelligent multiobjective optimization design framework based on the prediction of pavement rutting and cracking. Laboratory mix design and pavement performance data were collected from literature and the Long-Term Pavement Performance datasets, respectively, and the quality of the dataset was enhanced by integrating data processing methods including autoencoders for dimensionality reduction and generative adversarial networks for data augmentation. Six machine learning models were developed to predict mixture properties and common pavement distress. Through an actual project for designing the asphalt surface course of a highway pavement, the intelligent BMD was then achieved by simultaneously minimizing mixture cost and pavement rut depth, alligator cracking, and longitudinal cracking at the design life of pavement built with the paving mix. The problem was solved using several typical multiobjective metaheuristic algorithms-based algorithms. The design results show that compared to the corresponding laboratory-designed mixtures, the BMD-designed mixtures not only meet the performance criteria but also achieve cost savings of 11.4% for the 12.5-Nominal Maximum Aggregate Size (NMAS) and 5.9% for the 19-NMAS mixtures.