Development of an ML-Based Georgia Pavement Structural Condition Evaluation System

Details

• Creators:
  Tsai, Yichang (James) ; Yang, Zhongyu ; Lu, Bingjie
• Corporate Creators:
  Georgia Institute of Technology
• Corporate Contributors:
  Georgia. Department of Transportation. Office of Performance-Based Management & Research ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Corridors Deflection Evaluation Ground Penetrating Radar Machine Learning Maintenance Monitoring Pavement Condition Pavement Management Systems Pavement Structural Condition Rehabilitation Threshold-Based Classification Traffic Control Devices Traffic Speed Deflection Device

  More +
• Publication/ Report Number:
  FHWA-GA-26-2304
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; Georgia
• Edition:
  Final Report (January 2024 - July 2026)
• Corporate Publisher:
  Georgia. Department of Transportation. Office of Performance-Based Management & Research
• Abstract:
  Georgia's aging interstate highway network requires more reliable pavement evaluation to support maintenance and rehabilitation (M&R) planning. Current network-level practices rely primarily on surface condition data, which can lead to project-level cost overruns when structural deficiencies are discovered only after budgets have been defined from surface indicators. Traffic Speed Deflection Device (TSDD) technology enables continuous structural response measurement at highway speeds, but Georgia-specific TSDD-based screening rules calibrated against laboratory performance references have not been established. In addition, an integrated framework for combining structural and surface condition in treatment decision-making has not yet been implemented. This study develops Georgia-specific pavement structural condition classification methods and an integrated surface-structural treatment decision framework. The analysis uses TSDD, Ground Penetrating Radar (GPR), Three-Dimensional (3D) pavement surface data, and Hamburg Wheel-Track Testing (HWTT) results from I-59, I-285, and I-575. A quality-controlled dataset of 85 flexible pavement sections was constructed through missing data analysis, pavement-type screening, and a nearest-neighbor spatial joining process. Threshold-based and machine learning (ML) methods were evaluated under the same repeated stratified cross-validation (CV) protocol. The single-variable SCI_8 threshold achieved a mean F1 score of 0.806 and a mean Area Under the Curve (AUC) of 0.888 across 50 validation splits, providing a simple and interpretable candidate baseline for pilot structural screening. The SCI_8 + AC thickness duo-variable threshold achieved the highest mean F1 score among all methods, with F1 = 0.834 and AUC = 0.887, showing that GPR-derived asphalt thickness improves structural detection. Among the ML classifiers, Random Forest (RF) performed best, with F1 = 0.822 and the highest AUC of 0.891. A staged pilot implementation pathway is recommended: SCI_8 threshold screening for near-term pilot use, SCI_8 + AC thickness screening when reliable GPR data are available, and RF as a scalable ML engine as labeled data expand. The treatment framework translates structural and functional condition into four candidate M&R categories: No Action, Milling and Resurfacing, Close Monitoring, and Major Rehabilitation. Corridor-level maps demonstrate that the framework can distinguish surface-driven needs, structure-driven monitoring needs, and combined deterioration patterns, establishing an initial Georgia-specific research-scale foundation for incorporating TSDD-based structural screening into GDOT's pavement management system (PMS), subject to further corridor-level validation.
• Format:
  PDF
• Funding:
  0019906
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:4bff67c95521dac12450ef3d4f8347d0408cb6365ec1af7fb4b756ebe7a76df5780f2443624d75a9513c5052ea470bcb28231db4397e61e280ae0912a6a8691a
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/92748/dot_92748_DS1.pdf
• File Type:
  [PDF - 5.45 MB ]