무인 자동화선박을 위한 친환경 대체연료공급시스템의 AI 기반 고장진단 모드에 관한 연구

다국어 초록 (Multilingual Abstract)

With the global enforcement of carbon neutrality policies and the rapid advance ment of autonomous ship technologies, eco-friendly alternative fuel supply system s have become essential components for future maritime operations. However, the increasing complexity of these systems, combined with limited human intervention in autonomous ship environments, has made reliable and intelligent fault diagnosis a critical requirement. Conventional alarm monitoring systems based on predefined thresholds are insufficient to detect complex fault patterns arising from nonlinear interactions among multiple sensors. This study proposes an AI-based fault diagnosis mode for an eco-friendly altern ative fuel supply system operated in autonomous ships. A pilot-scale experimental fuel supply system using liquefied petroleum gas (LPG) was constructed to emulat e real shipboard operating conditions. Multivariate time-series data, including pres sure, temperature, flow rate, level, and vibration signals, were collected at a samp ling rate of 1 Hz under normal operation and multiple fault scenarios. Representat ive fault modes, such as gas leakage, filter clogging, valve degradation, and pump performance degradation, were experimentally implemented to reflect realistic ope rational anomalies. To effectively capture both local signal variations and long-term temporal depen dencies, a hybrid deep learning architecture combining one-dimensional convolutio nal neural networks (CNN) and bidirectional long short-term memory (BiLSTM) net works was employed. Fault modes were defined from a data-driven perspective as characteristic patterns in multivariate time-series data rather than as isolated phy sical failures. A sliding window approach was adopted to construct training and val idation datasets, enabling continuous fault tracking over time. The experimental results demonstrate that the proposed CNN–BiLSTM-based fau lt diagnosis model can accurately classify multiple fault modes and maintain stable performance during fault transition periods. Probability-based outputs using a Soft max layer provide quantitative confidence measures, allowing the interpretation of uncertainty during transient operating conditions. These results indicate that the p roposed approach extends beyond simple alarm-based diagnostics and enables enh anced state awareness and condition monitoring. The proposed AI-based fault diagnosis mode is expected to contribute to the de velopment of intelligent condition monitoring and predictive maintenance framewo rks for autonomous ships equipped with eco-friendly alternative fuel systems. KEY WORD : AI-based fault diagnosis; Eco-friendly alternative fuel supply system; Multivariate time-series analysis; CNN–BiLSTM; Autonomous ship; Condition monitoring

목차 (Table of Contents)

• List of Tables ⅲ
• List of Figures ⅳ
• Nomenclature and Abbreviation ⅴ
• Abstract ⅶ
• 1. 서론 1
• 1.1 연구배경 1
• 1.2 관련 연구 3
• 1.3 연구 차별성 5
• 2. 시스템 구성 7
• 2.1 시스템 구성 7
• 2.2 주요 구성 8
• 2.3 계측센서 구성 12
• 3. 데이터 수집 16
• 3.1 데이터 정의 16
• 3.2 정상모드 19
• 3.3 고장모드 21
• 3.4 데이터 수집 절차 22
• 3.5 고장모드별 데이터 수집 24
• 4. AI 기반 고장진단 모드 31
• 4.1 고장모드 정의 31
• 4.1.1 AI 기반 고장모드 분류 기준 31
• 4.1.2 AI 기반 고장모드 클래스 정의 32
• 4.2 AI 고장진단 알고리즘 36
• 4.2.1 학습 방법 및 딥러닝 모델 구조 36
• 4.2.2 결과 분석 39
• 4.3 AI 고장진단 알고리즘 평가 43
• 5. 연구 결과 및 고찰 44
• 5.1 AI 고장진단 알고리즘 적용 결과 44
• 5.2 AI 고장진단 결과에 따른 시사점 46
• 5.3 연구의 한계점 및 향후 연구과제 47
• 6. 결론 48
• 참고문헌 50
• 국문초록 52