Culverts are essential components of roadway and drainage systems, yet many are aging, undersized, and vulnerable to failure during heavy rainfall and flooding. Limited inspection resources and outdated condition data make it difficult for transportation agencies to identify which culverts pose the greatest risk to safety and network performance. This project addresses these challenges by applying artificial intelligence and network analysis to improve culvert condition assessment and maintenance prioritization.
The research focuses on the Deerfield County Watershed, where publicly available culvert condition data are incomplete or outdated. Machine learning models will be developed to predict current culvert condition ratings using historical records and environmental data, with an emphasis on interpretable and physics-informed approaches. These predictions will be integrated with geospatial network simulations to identify culverts whose failure would result in significant connectivity loss, flooding exposure, or service disruption. Targeted field inspections will be conducted to validate model predictions and improve accuracy. The results will support proactive, data-driven culvert management and provide a scalable framework for applying AI to infrastructure risk assessment.