철도 현장 안전 증대를 위한 AI 챗봇 기반 모델링 설계 및 응용 = Application and Design of an AI Chatbot-Based Modeling System for Enhancing Railway Field Safety

다국어 초록 (Multilingual Abstract)

Signaling systems in urban railways are critical components for safe train operations, requiring rapid and accurate identification of causes and countermeasures in the event of a failure. However, the increasing complexity of modern railway systems and the "information silos" phenomenon, where data is disconnected across departments, impose excessive cognitive load on field workers. This has become a primary cause of missed "golden times" or secondary failures resulting from human errors. To address these issues, this study proposes and implements a Hybrid AI Chatbot System combining Retrieval-Augmented Generation (RAG) and Case-Based Reasoning (CBR). A knowledge base was constructed using a Vector DB based on maintenance data and technical regulations of the Busan Urban Railway accumulated from 2010 to 2024. A conversational interface utilizing a Large Language Model (LLM) was developed to respond to field workers' natural language queries in real-time. The proposed system is designed such that the RAG module, which provides legal grounds to prevent 'rule-based mistakes', and the CBR module, which infers past similar failure history to supplement 'knowledge-based mistakes', operate in a complementary manner. Performance evaluation based on actual operational data demonstrated a search recall of 88.9% and an average response time (latency) of 0.41 seconds, proving that the system significantly reduces physical information retrieval time compared to traditional manual methods. Furthermore, in a usability evaluation based on the Technology Acceptance Model (TAM) involving 10 field experts, the system received high scores for Perceived Usefulness (4.8/5.0) and Perceived Ease of Use (4.7/5.0). Analysis from the GEMS model perspective confirmed that the system technically compensates for workers' memory failures and lack of experience, while optimizing human-system interaction from the HOF 5×5 model perspective. This study is significant in that it minimizes human error by applying the latest generative AI technology to the railway maintenance field and establishes a data-driven scientific decision-making support system. It is expected that the system's Safety Intelligence will be further advanced through the introduction of Multi-modal technology and Knowledge Graphs in the future. 주제어: Urban Railway Signaling System, Maintenance, Retrieval-Augmented Generation (RAG), Case-Based Reasoning (CBR), Human Error, Cognitive Overload

목차 (Table of Contents)

• Ⅰ. 서론 1
• Ⅱ. 이론적 배경 및 관련 연구 고찰 5
• 1. 철도 안전의 핵심과제: 인적 오류와 인지 심리학적 접근 5
• 2. 생성형 AI와 검색 증강 생성(RAG) 8
• 3. 사례 기반 추론(CBR) 11
• 4. 하이브리드 아키텍처 모델의 필요성 12
• Ⅲ. 하이브리드 AI 챗봇 시스템 모델 설계 및 구현 13
• 1. 시스템 아키텍처 설계 및 프레임워크 개요 13
• 2. 1단계: 법규와 규정 기반 RAG 시스템 구축 15
• 3. 2단계: 고장 및 장애 해결을 위한 CBR 시스템 구현 18
• 4. 데이터 처리 및 분석을 위한 시스템 최적화 20
• Ⅳ. 실증 실험 및 결과 분석 22
• 1. 실험 환경 및 시나리오 설계 22
• 2. 정량적 성능 평가 25
• 3. 정성적 사용성 평가 28
• Ⅴ. 결론 및 제언 33
• 1. 연구 요약 및 시사점 33
• 2. 연구의 한계 33
• 3. 향후 과제: Multi-Modal RAG and Knowledge Graphical 34
• 참고문헌 37