Development of Dynamic Modulus from Creep Stiffness of Asphalt Mixtures: A Machine Learning–Based Approach

Dynamic modulus (E*) obtained from the asphalt mixture pavement tester (AMPT) has played a vital role in conventional pavement design. However, the complexities of laboratory fabrication, field sampling of AMPT specimens, and difficulties in measuring E* at extremely low temperatures have led researchers to explore alternative methods. This study attempts to predict E* for asphalt mixtures using a machine learning (ML)–based approach applied to data from the bending beam rheometer (BBR) test. The creep stiffness S(t) data observed from BBR for different mixtures were processed using two different approaches, a two-degree regression-based model and an exponential regression model, to obtain the creep stiffness measurements at different desired frequencies and temperatures, respectively. A further time–temperature superposition-based approach to the processed data results in the creep stiffness over a wide range of frequencies and temperatures. Finally, the E* and S(t) values obtained using the time–temperature superposition approach for a comparable temperature and frequency are used to develop different predictive models. Among the models tested, predictions from the artificial neural network–based model outperformed traditional regression-based approaches. The proposed ML-based model, leveraging subtle variations in S(t) at desired frequencies, shows significant potential to predict E* at intermediate temperatures. This approach offers a better simulation of field conditions while providing a simpler and faster alternative to conventional testing methods like the AMPT, offering a pathway for improved asphalt mixture design in cold regions.