Project Details

Abstract

An ideal bridge inventory database is a structured, accessible repository of comprehensive information about bridges, such as their condition, inspection history, load capacities, design types, age, and other relevant attributes. Such database is essential to support data-informed and cost-effective bridge asset management and preservation. However, current practices for retrieving information/insights from and updating the databases lack automation, are slow and extremely expert-demanding. The increasing amount and the heterogeneous (multi-modal) nature of the data make it increasingly challenging to manually synthesize and distill useful insights from and/or updating the databases, calling for smart analytics technologies to automate the management, extraction, and interpretation of bridge inventory data. While large language models (LLMs) have shown the capability of comprehending multi-modal data, they remain significantly underutilized in bridge management. This project will investigate the viability of using LLMs to build artificial intelligence (AI) agents that can extract, memorize bridge condition from inspection records/reports, and enable standardized interpretation and organization of insights to support bridge preservation. The AI agents will convert raw and semi-structured bridge inventory data (e.g., inspection narratives, images, sensor signals) into structured database entries, summaries, and actionable recommendations. Users can interact intuitively with the AI agents via natural language queries, enabling efficient retrieval and interpretation of critical insights for bridge management. The agentic AI framework can achieve specified goals with minimal human/expert intervention.

Project Word Files