Enhancing Dynamic Modulus Predictions of Asphalt Mixtures Using Scientific Machine Learning

Abstract: This study presents the development of a scientific machine learning (SciML) model aimed at predicting the dynamic modulus (|E*|) of asphalt mixtures, a crucial parameter for evaluating the stiffness and performance of asphalt pavements. Traditional predictive models, such as the Hirsch model and Witczak equation, rely heavily on mixture volumetric properties, offering reasonable accuracy within certain conditions. However, these models often struggle to maintain high prediction accuracy across a wide variety of asphalt mixtures, temperatures, and loading conditions. This limitation can lead to unreliable performance predictions for pavements in real-world scenarios where conditions vary significantly.

To address these shortcomings, this research integrates physics-informed machine learning techniques with the existing Hirsch and Witczak models to enhance predictive accuracy. The SciML model is trained using a dataset of experimental dynamic modulus values, obtained across multiple temperatures and frequencies, incorporating time-temperature superposition and viscoelastic properties. By leveraging volumetric data as input features, the model not only utilizes the strengths of traditional mechanistic models but also introduces the flexibility and adaptability of data-driven approaches.

The key advantage of this SciML approach lies in its ability to capture the complex interactions between asphalt mixture components that influence the dynamic modulus. This results in more accurate and reliable predictions, particularly when extrapolating to conditions beyond the range of the experimental data. Preliminary results indicate that the SciML model significantly outperforms the Witczak and Hirsch models, especially in scenarios involving temperature and frequency ranges that are challenging for traditional models to predict accurately. This study marks a step forward in the application of machine learning to enhance the understanding and prediction of asphalt pavement behavior under various conditions.