AI-driven Preventive Maintenance Strategies for Asphalt Pavements

Abstract

Taiwan’s urban road infrastructure is increasingly challenged by aging pavements and constrained maintenance budgets. Traditional reactive repair strategies, such as milling and overlay, are not only costly but often lead to accelerated structural deterioration. To address this, we propose an AI-driven preventive maintenance framework that integrates the YOLOv8 deep learning algorithm, high-resolution 3D surface imaging, and a Pavement Condition Index (PCI)-based decision strategy. The system enables real-time identification of cracks, potholes, and rutting, achieving a mean Average Precision (mAP) of 97.2 % and a PCI estimation accuracy with R² = 0.92 and ± 3.5 absolute error compared to manual scoring. Field trials conducted across urban, county, and rural roads in New Taipei over a 12-month period demonstrated significant improvements in pavement condition. PMA and OGFC treatments achieved PCI retention rates above 90 %, while fog and slurry seals exhibited 10 – 12 % declines, particularly under high traffic and wet conditions. Material performance tests confirmed that all sealants and overlays met or exceeded national standards in curing time, abrasion resistance, and strength. Furthermore, integration of Multi-Criteria Decision Analysis (MCDA) and Life-Cycle Cost Analysis (LCCA) showed that condition-based interventions could reduce long-term maintenance costs by up to 20 % compared to reactive strategies. The framework also supports scalable deployment through potential integration with GIS dashboards and cloud-based pavement management systems. This study validates the feasibility of using AI for real-time pavement condition evaluation and strategic maintenance planning. By bridging detection precision, structural analysis, and cost-optimized decision-making, it provides a robust foundation for smart, sustainable road asset management.