Asphalt mixtures' stiffness modulus prediction using a machine-learning approach based on temperature and frequency conditions

One of the most suitable parameters to summarize the mechanical behaviour of asphalt mixtures is the stiffness modulus. Such performance parameter roughly describes the durability and serviceability provided by road pavements. However, it is strongly influenced by testing conditions, namely the loading frequency and the testing temperature. This study is aimed at investigating this relationship using a machine learning approach based on artificial neural networks. First, the physical and volumetric properties of the asphalt mixture under investigation were determined. Then, a 4-Point Bending Test experimental campaign was carried out and the stiffness modulus was evaluated under several testing conditions. Laboratory results were used to train a neural model that had temperature and frequency as inputs and the stiffness as output. The performance achieved was remarkable. Although the model is limited to only the mixture under investigation, this research is promising in view of an expanded dataset with multiple mixtures considered.