A Method for Preliminary Prediction of Asphalt Mixture Permanent Deformation Test Results

Abstract: The ability of asphalt mixtures to resist permanent deformation is a crucial indicator of asphalt pavement field performance. It is often assessed through laboratory performance tests under cyclic loading, which can take hours or days to complete. With the shift towards Balanced Mix Design (BMD), ensuring the resistance to permanent deformation becomes pivotal during mix design and quality control. This necessitates the characterization of several candidate mixes with varying component combinations to determine their resistance to permanent deformation, which could be labor-intensive and not always feasible. A preliminary forecasting method that can discern mixtures with superior resistance to permanent deformation without undergoing extensive testing procedures is thus highly desirable. The Random Forest (RF) algorithm has been demonstrated to be a successful method in pavement performance modeling and rehabilitation/maintenance strategy selection. This study focused on exploring the RF algorithm's potential to preliminarily predict hot mix asphalt's permanent deformation resistance based on material composition and relevant properties. The Hamburg Wheel Tracking Test was chosen for this study as it is commonly used to indicate the permanent deformation resistance potential of asphalt mixtures. A dataset from a state agency was adopted for analysis. The configuration and hyperparameters of the RF model were defined and optimized, and its robustness was examined using five-fold cross-validation. The model's predictive capability and accuracy were determined and compared with alternative methods such as regression and artificial neural network models. Preliminary results showed that the noise and pre-treatment of the dataset could significantly impact the RF model's predictive accuracy. Recommendations for further improvement of the prediction accuracy are provided.