DEVELOPMENT AND VALIDATION OF AN AI-BASED PREDICTIVE MODEL FOR THE DYNAMIC MODULUS OF ASPHALT CONCRETE

Abstract

The Asphalt Dynamic Modulus Prediction System (ADMPS) is a state-of-the-art modeling system for predicting the dynamic modulus ( ) of asphalt materials across extended temperature and frequency ranges. Developed to overcome data limitations at high temperatures, ADMPS utilizes experimental data from 10–40°C to accurately extrapolate ( )  values for temperatures up to 150°C. The system, take into account frequency-specific exponential models with machine learning (ML), further improve it by physics-informed (PI) constraints in order to make sure the scientific validity is obtained. Obtaining such a predictive accuracy of 85-90%, the model was validated properly through a custom suite of trend and pattern with accuracy analyses. This paper details this system's development, from initial challenges to the final most accurate solution, thus highlighting its methodologies and significant results at the end.