Data-Driven Inspection Planning for Utah Culverts Using Federated Learning

Details

• Creators:
  Mohammadi, Pouria ; Rashidi, Abbas
• Corporate Creators:
  North Dakota State University ; Center for Transformative Infrastructure Preservation and Sustainability (CTIPS) Region 8 University Transportation Center (UTC) ; University of Utah. Department of Civil and Environmental Engineering
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Culverts Data Analysis Data Collection Federated Learning Inspection Machine Learning Planning Predictive Models
• Publication/ Report Number:
  CTIPS-25-001
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; Utah
• Edition:
  Final Report
• Corporate Publisher:
  Center for Transformative Infrastructure Preservation and Sustainability (CTIPS) Region 8 University Transportation Center (UTC)
• Abstract:
  In recent years, transportation agencies have increasingly turned to machine learning (ML) to enhance the effectiveness of infrastructure asset management. However, limited local inventory data often hinders building accurate and reliable ML models. Additionally, data privacy and ownership concerns discourage agencies from sharing raw datasets. Many state departments of transportation (DOTs), including the Utah DOT (UDOT), face challenges in managing culverts due to limited inspection data and privacy concerns. This research proposes using federated learning (FL), an emerging ML paradigm, to enhance culvert condition prediction without requiring centralized data sharing. By leveraging FL, UDOT can collaboratively train predictive models with data from other state DOTs while preserving data confidentiality. The project involves collecting culvert and environmental data from multiple state inventories, preprocessing them to ensure consistency, and developing artificial neural network (ANN)-based models within the FL framework. The resulting FL model achieved an accuracy of 80.4%, performing comparably to the centralized model trained on the fused dataset and significantly outperforming the model trained solely on Utah’s data. The findings demonstrate that FL can effectively support high-performance predictive modeling while preserving data. This research offers a novel approach to infrastructure asset management that balances predictive accuracy with regulatory compliance, setting a precedent for broader adoption of FL in transportation systems.
• Format:
  PDF
• Funding:
  69A3552348308
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:72a7983c17e93e36eb4c3c61f8f6f361f4587b69cea30342fb5619fefcaa280e87d9eec82967a394f1c52e3a7c04ec1d32da6657b1ab4da2d61c856942ff301a
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/86851/dot_86851_DS1.pdf
• File Type:
  [PDF - 1.68 MB ]