Development of field compaction curves for asphalt mixtures based on laboratory workability tests and machine learning modeling

A clear understanding of asphalt mixtures' workability and compactibility is crucial for optimizing field compaction and mix designs. However, the standard laboratory gyratory compaction test alone often fails to accurately describe in-situ compaction behaviors, leaving a gap between laboratory and field compaction. This paper proposes an innovative yet practical approach to developing field compaction curves and estimating field compaction behavior using laboratory workability test data. A hypothesis of Rotation for Effective Compaction was first introduced, considering particle rotation as an effective parameter linking laboratory and field compaction. It suggests that the trend of particle rotational motion, under given compaction energy, remains consistent across both laboratory and field conditions. Materials from four lanes of the FHWA Turner-Fairbank Highway Research Center (TFHRC) Pavement Testing Facility (PTF) 2023 project were tested using MixWorx™ sensor in accordance with ASTM D8541. An AutoML method was applied for optimizing machine learning models. It identified LightGBM as the optimal model for predicting the field compaction curve, achieving 95.7 % classification accuracy for compaction levels and a 98.4 % R² fit for density prediction. Validation with field sensing and compaction data from Altoona, PA, and Angola, IN projects confirmed model robustness and hypothesis validity. This method offers a promising tool for optimizing asphalt mixture design and identifying workability issues.