| 수록사항 |
한국대공간건축 논문집 , Vol. 26, No. 3 (통권 105호)(2026-09) |
| 주제어 |
; Large language model; Seismic fragility; Nuclear power plant; Machine learning; Retrieval-augmented generation |
| 요약2 |
This study presents the development of an integrated software program that evaluates the seismic fragility of nuclear power plant (NPP) structures using a large language model (LLM). The program combines machine learning (ML)-based prediction of concrete aging deterioration and seismic floor response with a Monte Carlo simulation-based seismic fragility assessment, and integrates them into a single web-based environment. A distinctive feature of the program is an artificial intelligence assistant built upon an LLM and retrieval-augmented generation (RAG); it interprets natural-language queries, automatically invokes the relevant prediction and fragility-assessment tools, and generates answers together with supporting evidence retrieved from technical documents and automatically produced fragility curves. The ML modules predict chloride diffusion, carbonation, thermal effects, and floor response spectra, and recommend the best-performing algorithm automatically, while the fragility module derives lognormal fragility curves and supports aging-degradation scenarios. By connecting deterioration prediction, response prediction, fragility assessment, and literature-based reasoning into one conversational workflow, the developed program significantly lowers the technical barrier for ML-based seismic fragility evaluation and improves both accessibility and reliability. The overall architecture, development environment, and main functions of the program are described in detail. |
