Safety Consultation for Prefabricated Construction: A Localized Retrieval-Augmented Generative Question-Answering System
Longhui Liao, Weiwei Mo, Yuhang Wen, Shan Liu, Yang Zou, Mo Li, Chengke Wu, Cheng Fan
Abstract
Despite inherent safety benefits associated with prefabricated construction, risk of on-site hazards persists, particularly during hoisting and splicing phases. Construction workers appear to still lack full mastery of prefabricated construction techniques, thereby augmenting the risk of precipitating on-site safety incidents. This study develops an innovative retrieval-augmented generative (RAG) question-answering system leveraging large language models (LLMs) to bolster safety management in prefabricated construction, offering real-time safety knowledge consultation for construction personnel engaged in prefabricated building projects. The methodology underlying the proposed system synergizes knowledge-base retrieval with LLMs, augmenting answer generation of users’ questions with retrieved safety knowledge to provide more accurate responses. This enables effective addressing of user inquiries in natural language, markedly improving user interaction. Through empirical testing and development of a chatbot interface, this study validates the efficacy of the proposed approach. The design of this system harbors the potential for deployment in construction site environments, facilitating both private deployments for construction firms and offline operations, thereby guaranteeing availability irrespective of network conditions. The methodology offers scalability through the updating of knowledge bases, thus rendering it applicable to myriad construction safety scenarios.