작업자 부주의로 인한 건설 현장 떨어짐 사고의 시스템적 분석: HFACS와 LLM의 결합 방법론

Systemic Analysis of Fall Accidents Caused by Worker Inattention at Construction Sites: An Integrated Approach using HFACS and LLM

Accident investigation records have shown that human error is a major cause of accidents in construction. It has been pointed out that instead of blaming the victim who made the error, management approaches and strategies are required to better prevent the occurrence of similar incidents. However, accident records often include only proximal causes rather than root causes that may help determine future preventive actions from a management perspective. This paper thus explores a systemic approach to infer root causes of accidents from the accident descriptions, which include rich information on how they occurred. Specifically, the proposed approach integrates the human factors analysis and classification system with a large language model to analyze 1,816 fall accident cases extracted from publicly available accident reports. Using prompt engineering, we designed the Claude 3.5 Sonnet model for the classification system, achieving a valid rate of 99.34%. Results revealed distinct patterns in unsafe acts at different working heights. For example, as height increased from less than 2 m to 10 m, skill-based errors decreased from 65% to 28%, while routine violations increased from 12% to 42%. However, preconditions of unsafe acts, inadequate supervision, and organizational influences were consistent regardless of height. By leveraging a large language to analyze large-scale data, this research thus uncovered evolving patterns of unsafe acts with increasing working height, suggesting height-specific safety strategies for the prevention of fall accidents in practice.