TY - GEN
T1 - Large Language Model-Based BIM-Design for Safety (DfS) Framework for Phase-Specific Safety Mitigation Recommendations during the Design Stage
AU - Gu, Minsang
AU - Jeong, Jaewook
AU - Kumi, Louis
N1 - Publisher Copyright:
© 2026 International Association on Automation and Robotics in Construction. All Rights Reserved.
PY - 2026
Y1 - 2026
N2 - The construction industry exhibits a higher fatality rate compared to other industries, making proactive safety management essential for risk reduction. Existing safety management practices have primarily focused on post-accident responses, presenting challenges in implementing preventive measures. Therefore, this study proposes a framework that integrates the Design for Safety (DfS) concept with Building Information Modeling (BIM) for application during the design phase. While previous research analyzed only completed BIM models and relied on manual matching between DfS information and BIM elements, this study utilizes Large Language Models (LLMs) to automatically infer element information from 488 DfS reports and generates a knowledge graph consisting of 271 nodes and 780 relationships. The proposed system classifies construction progress into four phases based on Uniformat II (Excavation, Substructure, Superstructure, Finishing) and matches them with BIM models. Implemented as an add-in, it enables designers to access empirical data-based mitigation measures for phase-specific elements in real-time within the BIM environment without switching programs. This allows designers with limited DfS-related education or experience to immediately apply safety mitigation measures without interrupting their design workflow, thereby lowering barriers to DfS utilization and enhancing practical applicability in the design phase.
AB - The construction industry exhibits a higher fatality rate compared to other industries, making proactive safety management essential for risk reduction. Existing safety management practices have primarily focused on post-accident responses, presenting challenges in implementing preventive measures. Therefore, this study proposes a framework that integrates the Design for Safety (DfS) concept with Building Information Modeling (BIM) for application during the design phase. While previous research analyzed only completed BIM models and relied on manual matching between DfS information and BIM elements, this study utilizes Large Language Models (LLMs) to automatically infer element information from 488 DfS reports and generates a knowledge graph consisting of 271 nodes and 780 relationships. The proposed system classifies construction progress into four phases based on Uniformat II (Excavation, Substructure, Superstructure, Finishing) and matches them with BIM models. Implemented as an add-in, it enables designers to access empirical data-based mitigation measures for phase-specific elements in real-time within the BIM environment without switching programs. This allows designers with limited DfS-related education or experience to immediately apply safety mitigation measures without interrupting their design workflow, thereby lowering barriers to DfS utilization and enhancing practical applicability in the design phase.
KW - Building Information Modeling (BIM)
KW - Construction Safety
KW - Design for Safety (DfS)
KW - Knowledge Graph
KW - Large Language Model (LLM)
UR - https://www.scopus.com/pages/publications/105046011247
U2 - 10.22260/ISARC2026/0100
DO - 10.22260/ISARC2026/0100
M3 - Conference contribution
AN - SCOPUS:105046011247
T3 - Proceedings of the International Symposium on Automation and Robotics in Construction
SP - 778
EP - 785
BT - Proceedings of the 43rd International Symposium on Automation and Robotics in Construction, ISARC 2026
A2 - Chen, Qian
A2 - Lee, Gaang
A2 - Liang, Ci-Jyun
A2 - Zhang, Jiansong
A2 - Kamat, Vineet R.
PB - International Association for Automation and Robotics in Construction (IAARC)
T2 - 43rd International Symposium on Automation and Robotics in Construction, ISARC 2026
Y2 - 22 June 2026 through 26 June 2026
ER -