Physics-Informed Neural Network (PINN) modelling for aging prediction in asphalt pavements

Physics-Informed Neural Networks (PINNs) are emerging as an increasingly promising area of research within the scientific community. This growing interest is attributable to their unique ability to integrate fundamental laws of physics into Deep Learning (DL) processes. The present research aims to explore the effectiveness of PINNs in addressing the complexities involved in bitumen aging prediction in order to introduce more efficient and accurate modelling methodologies. In this research project, a PINN was built in the Python programming language following a multi-physics approach, which allowed simultaneous consideration of the three main chemical -physical phenomena governing bitumen aging: heat diffusion, oxygen diffusion, and oxidation kinetic diffusion. Relative to these three phenomena, initial and boundary conditions were considered on a simple geometric model consisting of a bitumen plate. The three partial differential equations (PDEs) describing these processes were integrated into the PINN code to obtain satisfactory results in predicting the aging of bitumen. Some limitations emerged, but several approaches to overcome them were presented, and suggestions for future work were provided. This paves the way for an increasing application of PINNs in pavement engineering, intending to replace traditional Finite Element Models (FEMs), which are computationally expensive and time-consuming.