A persistent gap exists between abundant safety documentation and its effective conversion into actionable knowledge in bridge construction safety. While knowledge modeling offers a promising solution, unresolved issues remain regarding knowledge granularity and the boundaries of model advantage. This study develops a Bridge Engineering Construction Safety Knowledge Model (BECS-KM) to narrow this gap and to investigate these challenges.
BECS-KM is built using ontology and knowledge graph techniques grounded in codes and standards. A multilayered representation structure is incorporated to balance knowledge granularity, and a comparative experiment is conducted to evaluate the strengths boundaries of BECS-KM versus ChatGPT-5.
In terms of knowledge modeling, the multilayered structure effectively balances knowledge granularity within BECS-KM. Regarding knowledge application, the comparative experiment demonstrates the superiority of BECS-KM (F1 = 0.97) over ChatGPT-5 (F1 = 0.83) of domain knowledge queries.
This study makes three contributions: (1) transforms fragmented and heterogeneous safety documentation into a structured and operational knowledge resource, (2) provides a multilayered representation structure to balance knowledge granularity and (3) empirically delineates the performance boundary of domain knowledge models against general LLMs. Together, these contributions establish BECS-KM as a robust foundation for future intelligent safety knowledge services.
