Mobile QR Code QR CODE

Journal of the Korea Concrete Institute

J Korea Inst. Struct. Maint. Insp.
  • Indexed by
  • Korea Citation Index (KCI)

Journal Search

XML PDF INFO REF
Title LLM-Based Question-Driven Hazard Identification Methodology for Recognizing Fall Risk in Abnormal and Unplanned Work Situations
Authors 김지우(Jiwoo Kim) ; 오태근(Tae-Keun Oh)
DOI https://doi.org/10.11112/jksmi.2026.30.4.1
Page pp.1-9
ISSN 2234-6937
Keywords 추락사고; 비정상ㆍ돌발작업; 위험성평가; 질문형 유해ㆍ위험요인 식별; 대규모 언어모델 Fall accidents; Abnormal and unplanned work; Risk assessment; Question-driven hazard identification; Large language model
Abstract The construction industry has a high rate of fatal occupational accidents, and fall accidents are one of the most critical accident types.
Non-routine and urgent tasks, such as process changes, emergency inspections, and temporary corrective actions, can create risk-recognition gaps before work begins due to limited preparation, omitted procedures, and time pressure. Conventional risk assessment mainly focuses on planned and repetitive work, making it difficult to reflect rapidly changing site conditions. This study proposes a question-driven hazard identification methodology to systematically identify fall risks before work begins in unexpected field activities. Based on literature and regulatory reviews, a 15-item question pool was developed, including work planning, time pressure, elevated work locations, access routes, platform conditions, guardrails, personal protective equipment, anchorage points, environmental conditions, role allocation, and stop-work judgment. In particular, Q15 was designed to confirm whether identified risk signals lead to work suspension, immediate corrective action, or work resumption after mitigation. The large language model (LLM) was limited to a question generation support tool rather than a risk judgment tool. This study is a methodological development study, and its verification scope is limited to expert validity review and case simulation rather than empirical field data or accident-reduction analysis. The proposed approach can complement the pre-task confirmation gap in abnormal and unplanned work and connect risk signals to on-site corrective actions.