In-Su Han, Myung-Il Roh, Min-Chul Kong, "A Generative AI-based Q&A System for Design Regulations", Proceedings of PRADS(International Symposium on Practical Design of Ships and Other Floating Structures) 2025, Ann Arbor, USA, 2025.10.19-23
International Conference
2025.02.27 11:09
In-Su Han, Myung-Il Roh, Min-Chul Kong, "A Generative AI-based Q&A System for Design Regulations", Proceedings of PRADS 2025, Ann Arbor, USA, 2025.10.19-23
조회 수 3190
| 첨부 '1' |
|---|
| Abstract | Design regulations are constantly being revised to ensure the safety and robust design of ships. Advancements in technology and the growing scale of ships have made design regulations increasingly specialized and detailed. As a result, workers such as designers and inspectors face growing challenges in efficiently finding necessary information within the large volume of design regulation documents. In particular, a significant investment of time and effort is required to thoroughly review each regulation and comprehend the specialized terminology and complex details in order to obtain the desired information. Therefore, this study proposed a method to address these challenges by implementing a generative AI-based Q&A system. The proposed method employed a RAG (Retrieval-Augmented Generation) architecture, which integrates information search modules with generative AI capabilities, to construct a Q&A system that is both accurate and reliable. First, data were extracted from PDF-format design regulations and transformed into a format suitable for the RAG architecture, and used to build the corresponding database. The concept of domain adaptation was then applied to a pre-trained general-purpose language model, resulting in a specialized Q&A system dedicated to design regulations. The proposed Q&A system quickly identifies relevant regulations within the database based on the designer’s queries. It then provides detailed and accurate answers that align with the intent of the query, leveraging generative AI. Applying this system to the regulations of the Korean Register (KR-Rules) demonstrated its ability to deliver fast and precise information to the designer’s queries. By enabling better comprehension of complex ship design regulations and delivering context-relevant answers beyond simple document searches, this study is expected to improve the work efficiency of workers in the field significantly. |
|---|---|
| Publication Date | 2025-10-20 |
