Project Details

Abstract

As transportation agencies increasingly adopt cutting-edge data analytics to refine infrastructure management strategies, the role of condition prediction models is becoming more critical. These models are pivotal in optimizing maintenance budgets, especially for underrepresented infrastructures like culverts, which have been neglected in the past. Similarly, due to the lack of a comprehensive culvert management system, the Utah Department of Transportation (UDOT) faces significant challenges in inspecting and maintaining culverts. Therefore, this study proposed a data-driven approach using federated learning to enhance Utah's culvert management. Since Utah's culvert dataset was limited, we expanded it by collecting data from several other state DOTs. However, to address data privacy concerns, we employed federated learning approach. This innovative technique avoids direct data sharing. Instead, each DOT trains a local model on its own data, and only the updated model parameters are shared with UDOT. This allows us to leverage the collective knowledge of multiple DOTs while ensuring robust data security. Our findings highlight the efficacy of the proposed federated learning-based models in enhancing prediction accuracy while ensuring data privacy and reducing data transmission overheads.

Project Word Files