Construction Quality Monitoring of Hot Mix Asphalt Using UAV Thermal Imagery
📨 Principal Investigator : Hasan Ozer
🤝 Sponsor: National Center for Infrastructure Transformation (NCIT)
đź“… Timeline: 2022 – Ongoing
Highlights
Introduction
Getting the compaction right during asphalt paving is one of the most consequential decisions in pavement construction. Proper compaction improves in-place density, and empirical evidence suggests that even a one percent increase in density can translate to at least a 10 percent longer pavement service life. Yet monitoring the compaction process is challenging because the paving crew operates over hundreds of feet of roadway, coordinating roller passes, speed, and sequencing as the paved mat cools quickly. Areas that cool too quickly may not achieve target density before the mat stiffens, leaving behind hidden zones of reduced performance that will develop as early distress years down the road.
Existing intelligent construction technologies (ICTs) for monitoring thermal and compaction uniformity during paving have meaningful limitations. Paver-mounted thermal profilers (PMTP) capture temperature data only in the narrow strip directly behind the paver screed, missing what happens to the mat after placement. Intelligent Compaction (IC) systems track roller position and stiffness, but exhibit inconsistent correlations and require expensive instrumented rollers. This project, funded by the National Center for Infrastructure Transformation (NCIT) and led by Dr. Hasan Ozer with co-authors Naaga Vedula and Masih Beheshti at Arizona State University, developed an alternative: a UAV equipped with an infrared thermal camera, flying above the paving operation and transmitting real-time thermal maps of the entire freshly placed mat across 13 Arizona construction sites.
Methodology and Framework
The study was structured in two phases. Phase 1 (spring-summer 2023) established the data collection protocol using a DJI Matrice 300 RTK UAV equipped with an H20N thermal camera – a dual-infrared sensor system with 2Ă— and 8Ă— optical zoom and up to 32Ă— digital zoom, capable of capturing 640Ă—512-pixel radiometric thermal images. Eight paving job sites across the Phoenix metropolitan area were visited during active paving operations. The UAV was flown at 30 m (~100 ft) altitude, covering a 200-foot window behind the paver. A protocol was developed to divide each image into 7 horizontal strips and 13 data points per 50-foot sublot, enabling systematic temperature analysis of the mat surface. A new metric, the Mat Differential Matrix (MDM), was developed to locally quantify thermal differentials.
Phase 2 (spring-summer 2024) automated the data collection and analysis pipeline using the DJI Thermal Software Development Kit (SDK) to extract radiometric temperature data at 2-second intervals without operator intervention. Roller-tracking capability, where a YOLOv8 deep learning object detection model was trained to identify both the freshly paved mat and compaction rollers in thermal images, was added in this phase. The trained model enabled continuous, automated tracking of roller position, speed, pass counts, and coverage patterns, which previously required expensive GPS-equipped instrumented roller systems.
Key Findings
Thermal Segregation Patterns Across Arizona Sites
Phase 1 analysis revealed that thermal non-uniformity was present to some degree at all eight sites, though severity varied considerably. Several non-uniform thermal patterns were identified. Longitudinal cold streaks appeared at sites with uneven temperature distribution caused by the gap between the paver auger gearbox and the paved mat. V-shaped temperature anomalies (cooler zones radiating diagonally behind the paver) were observed at one site and traced to a specific issue with the closed hopper wings, which were holding cold material, demonstrating the diagnostic potential of the UAV thermal-mapping approach. Cool-down analysis using the UAV imagery quantified how rapidly different parts of the mat lost temperature, providing insight into where compaction windows were narrowing due to early cooling.
Automated Roller Tracking and Compaction Analysis
The Phase 2 YOLOv8 object detection model achieved high accuracy in identifying and tracking rollers in thermal aerial images, even under challenging conditions like partial occlusion and varying roller orientations. The model was trained using a combination of real images from Phase 1 and synthetic data augmented with real images, and the best-performing model (YOLOv8s-obb) achieved an intersection-over-union (IoU) score sufficient for reliable roller bounding-box detection across diverse site conditions. From the roller detections, the compaction analysis algorithm computed roller pass counts for each section of the mat, roller speed, and identified sections with insufficient passes relative to the job mix formula rolling pattern requirements. A case study at an experimental Phoenix-area site demonstrated the framework’s full capabilities, with two adjacent lanes paved and monitored simultaneously. The UAV data revealed that the west lane had more non-uniform cooling and areas with insufficient roller passes, which were confirmed by core density data collected 1 and 6 months later.Â
Â
Patent and Industry Implications
The integrated thermal-compaction monitoring protocol developed in this study was submitted for intellectual property protection through an ASU patent filing, reflecting its potential for commercial deployment. The protocol offers contractors a non-intrusive, unified monitoring system that does not require modifications to rollers or the paver. It can provide real-time feedback to the paving crew about emerging problem areas before the mat cools below the compaction temperature threshold. It can also be used as a training tool for the operations crew to help mitigate the recurrence of these errors at subsequent job sites.