Pavement Condition Prediction for Communities: A Low-Cost, Ubiquitous, and Network-Wide Approach

Abstract

Effective prediction of pavement deterioration is critical to forecast infrastructure performance and make infrastructure investment decisions under escalating environmental and traffic change. However, most communities often struggle to undertake such predictive tasks due to limited sensing capacity and lack of granular data. With the pavement condition rating (PCR) data generated from artificial intelligence (AI)-powered computer vision technologies and multiple openly available data sets, we propose a low-cost and ubiquitous approach to predict system-level pavement conditions using nine communities across the US as an example. In addition to predicting absolute PCRs as was done in classical models, we develop another set of models to predict the change in PCRs over any time increment (i.e., time lapse between a PCR observation and retrofit decision point) and compare the results. The findings showed that the proposed low-cost prediction approach yields results comparable to existing studies, demonstrating its promising application in supporting pavement management. Furthermore, the PCR change model indicates that, besides current PCR, weather, road classification, socioeconomics, and built environment attributes are important to predicting PCR change. The interactive impacts also show salient interactive effects between variables and current PCR, offering suggestions on better allocating the limited resources in pavement maintenance projects. Finally, the proposed model could enhance climate resiliency and transportation equity during the pavement management process.