On-Going Projects

2nd-Cycle Projects (2024-grant)

• SMART Feedback System for Bridge Asset Management through the Development of Generating Artificial Intelligence Network (GAIN) Members [IBT-ABC-UTC-2024-C2-UGA01] : developing smart bridge components, such as girders, equipped with sensors and/or computer vision technologies to generate data for analysis using artificial intelligence (AI). These components, referred to as Generating Artificial Intelligence Network (GAIN) members, will provide bridge owners with enhanced asset management information. GAIN members will supply data to the bridge network, offering real-world feedback. When deployed at strategic locations, both at the state level and nationwide, GAIN members could deliver valuable insights into state and national networks, supporting bridge construction, mobility, welfare operations, and emergency evacuations.

• Optimizing Strategies in Bridge Asset Management Through Generating Interactive Reinforcement Learning (GI-RL) Methods [IBT-ABC-UTC-2024-C2-UGA02]: A reinforcement learning framework that models the interactions between state transportation agencies and local bridge owners as a multi-agent system. Each agent aims to maximize its immediate and/or long-term payoff by balancing maintenance costs and bridge performance. The RL agents learn optimal strategies through repeated interactions, considering both immediate and future consequences of their actions. Our study uses data from 6,000 bridges in Georgia, incorporating factors such as bridge condition ratings from the National Bridge Inventory and maintenance data, highlighting the costs and benefits of timely repairs and optimal strategies available to specific owners. We propose using advanced reinforcement learning methods with several enhancements to improve learning stability, efficiency, and performance in reinforcement learning environments.

1st-Cycle Projects (2024-grant)

• Quant CR for Transformative Bridge Asset Management [IBT-ABC-UTC-2024-C1-UGA01] : This project aims to empower local and state bridge owners to make informed decisions on MRR, optimizing budget plans across various scenarios. Thus, the research approach centers on developing an AI-powered quantitative condition rating (QUANT CR) system and decision analysis tool for bridges, which leverages historical data, including the extensive records of the National Bridge Inventory (NBI)’s element data, Long-Term Bridge Performance Program, bridge MRR scenarios, inspectors’ narratives, and ultimately risk management including environmental, traffic, vehicle, and other relevant data to predict long-term bridge performance. A web-based Geographic Information System (GIS) platform will store and visually represent bridge network assets, providing uniform access to all databases. A hierarchical approach to quantify bridge condition ratings and assigning them to the network analysis tool will improve long-term bridge performance predictions. Deep learning algorithms will analyze historical data and detailed bridge condition narratives to improve the bridge performance  predictions at element level. Incorporating the element-level inspection results and detailed defect items  is a key task for building a Quant CR model. Text recognition techniques will further interpret inspectors’ descriptive reports, transforming qualitative comments into actionable quantitative data.