Thermal Digital Twins of Asphalt Pavements using Physics-informed Neural Networks

Abstract

To address the adverse effects of climate change on road surfaces, the creation of a thermal digital twin for asphalt pavements is proposed in this paper. The global increase in temperatures, coupled with heavy traffic loads, has resulted in the premature deterioration of asphalt roads. In response to these early failures, recent efforts have focused on enhancing pavement structural integrity by incorporating asphalt modifiers and cool pavement strategies. Regardless of the chosen approach, continuous monitoring of pavement characteristics using embedded sensors plays a crucial role in enabling timely maintenance and rehabilitation (M&R) decisions. One of the ways to achieve this is through the estimation of the thermal diffusivity of the pavement layers which can be directly related to structural condition. To estimate the diffusivity, an inverse-Physics-informed Neural Network model is proposed, which facilitates the integration of sensor data and the heat transfer mechanisms within the pavement. A finite-difference numerical model is adopted to serve as the groundtruth model, to provide the temperature data within the pavement layer, given the boundary conditions. Using the weather data of Dubai, this study has shown the feasibility of adopting Physics-informed Neural Networks (PINNs) to predict the temperature within the pavement layer, despite the complex mixed boundary conditions. Moreover, the i-PINN can be used to estimate the thermal diffusivity with an error of around 0.0001 m 2 /h, using temperature data taken from five depths of the asphalt layer. The contribution of this study, is therefore, a novel condition monitoring of asphalt pavements that can significantly improve existing road maintenance programs.