Speaker
Description
The applicability of synergistic Artificial Intelligence (AI) and Digital Twin (DT) technologies for a comprehensive bridge inspection framework depends on the completeness of the available data of the structure. Incomplete or missing data of the structures are often observed, which hinders the processes for effective inspection and maintenance. In this study, a methodology using Large Language Models (LLMs) is presented to overcome the challenge of extracting and predicting incomplete structural information for bridges. The proposed approach involves fine-tuning a closed-source LLM with a bridge-specific dataset that enables accurate information retrieval from an existing database. The generative model then makes predictions and generates high-quality data that is subsequently used to build a comprehensive bridge database. The experimental results highlight the effectiveness of the proposed LLM model in accurately predicting incomplete bridge data within the engineering constraints. This study addresses the existing challenge of incomplete bridge data by proposing an effective framework for data generation. By bridging this gap, the applicability of BIM and digital twin technologies is significantly improved, making them more accessible and feasible for future inspection and maintenance work on infrastructures.
