Abstract
With the development of the Internet of things, big data, and AI leading the 4th industrial revolution, it has become possible to acquire, manage, and analyze vast and diverse condition signals from various industrial machinery facilities. In addition, it has been revealed that various and large amounts of signals acquired from the facilities can be utilized for fault diagnosis. Currently, while data-driven fault diagnosis techniques applicable to the facilities are being developed, it has been tried to apply the techniques for the development of fully autonomous ships in the shipbuilding and shipping industry. Since the autonomous ships must be able to detect and diagnose the failures on their own in real time, the overall research is required on how to acquire signals from the ship facilities and use them to diagnose their failures. In this study, a fault diagnosis framework was proposed for condition-based maintenance (CBM) of ship oil purifiers, which are an auxiliary facility in the engine system of a ship. First, an oil purifier test-bed for simulating faults was built to obtain data on the state of the equipment. After extracting features using discrete wavelet decomposition from the data, the features were visualized by using t-distributed stochastic neighbor embedding, and were used to train support vector machine-based diagnostic models. Finally, the trained models were evaluated with Accuracy and F1 score, and some models scored 0.99 or higher, confirming high diagnostic performance. This study can be used as a reference for establishing CBM system and fault diagnosis system. Furthermore, this study is expected to improve the safety and reliability of oil purifiers in Degree 4 MASS.