This study was initiated to address the issue of perfunctory risk assessments in small and medium-sized manufacturing enterprises (SMEs) in South Korea, where a majority of industrial accidents occur. To overcome limitations such as the lack of expert...
This study was initiated to address the issue of perfunctory risk assessments in small and medium-sized manufacturing enterprises (SMEs) in South Korea, where a majority of industrial accidents occur. To overcome limitations such as the lack of expert personnel and information, it proposes an automated risk assessment framework based on Retrieval-Augmented Generation (RAG) technology, designed to help even non-experts conduct swift and accurate assessments.
The core of the proposed model is a Knowledge Base built as a vector database with reliable data, including the Occupational Safety and Health Act, KOSHA Guides, and serious accident investigation reports. The model is also designed to enhance the expertise of its outputs by applying the Hierarchy of Controls principle.
Simulation results for high-risk work scenarios demonstrated that the proposed model is superior to the conventional manual method in terms of accuracy, objectivity, and efficiency. It showcased its potential as an "AI safety expert" by presenting clear evidence based on relevant laws and technical guidelines, systematically analyzing root causes, and generating effective countermeasures.
This research holds academic significance as a pioneering conceptual framework applying RAG technology to manufacturing risk assessment. It also has practical value in its potential to resolve safety information asymmetry for SMEs and contribute to the substantive prevention of industrial accidents.