A streamlined approach for probabilistic pavement life‐cycle performance prediction via physics‐informed neural networks

Abstract

Pavement life cycle management (LCM) is essential for assessing the long-term environmental and economic impacts of roadway infrastructure by means of pavement life cycle assessment (LCA) and life cycle cost analysis (LCCA). However, pavement LCA and LCCA studies frequently overlook the use phase due to the limited availability of performance data. To address this gap, this study proposed a streamlined approach for pavement life-cycle International Roughness Index (IRI) prediction utilizing physics-informed neural networks. Dedicated to pavement LCM, the methodology is built upon input variables that are readily available prior to pavement service, ensuring high prediction accuracy and physical consistency while maintaining computational efficiency. Importantly, the model incorporates the IRI drop following maintenance and rehabilitation (M&R) activities and their subsequent impacts on IRI progression of rehabilitated pavements, capturing critical post-M&R behavior with machine learning (ML). By integrating prior domain knowledge and uncertainty consideration into the IRI progression models, the framework effectively accommodates the variability inherent in pavement deterioration processes, supporting robust probabilistic pavement LCM. Finally, the successful integration of the approach into pavement LCA is demonstrated through multiple case studies, which validate its capability in predicting IRI evolution across various M&R cycles under diverse traffic and climatic conditions. Overall, this research provides a more accessible and robust framework via physics-informed ML, even with minimal yet physically justified priors, for pavement assets management and comprehensive assessment of environmental and economic impacts of roadway infrastructure throughout its service life.

Topics & Concepts