선박의 기관시스템 보조기기 펌프의 상태기반정비를 위한 HHT 및 CNN을 활용한 고장 분류 = A Study on Fault Classification for Condition-Based Maintenance of Auxiliary Pump in Ship’s Engine Systems Using HHT and CNN

다국어 초록 (Multilingual Abstract)

Purpose: This study focuses on test bed construction for condition-based maintenance of auxiliary pump systems in autonomous ships and fault classification using machine learning.
Methods: Experiments were conducted on a test bed using failure simulation conditions.
Vibration data underwent preprocessing and were converted into images by Hilbert-Huang transform (HHT). The convolutional neural networks (CNN) performed feature extraction and learning to classify faults.
Results: The study confirmed a high percentage classification accuracy of 96% for six operational states (normal and abnormal). Increasing the convolutional layers improved training accuracy but caused overfitting, as indicated by lower validation accuracy. Simplifying the structure and regularization techniques, such as dropout, enhanced the model’s predictive performance.
Conclusion: This study developed a 2D CNN-based algorithm that successfully classified faults with 96% acc