무인수상선(USV) 수중 추진기의 고장진단 시스템 연구

Abstract

Unmanned surface vehicle (USV) is one of the marine platforms that conduct missions on the surface of water, and recently, zero-emission research such as the utilization of renewable energy is being promoted because of tight environmental regulations. Simultaneously, the scope of fault diagnosis of the USV is expanding, and the fault of the underwater thruster may cause a fatal danger to the mission performance of the platform. Therefore, an unmanned surface vehicle condition monitoring system and fault diagnosis system are essential. In this study, a condition monitoring software GUI was designed for real-time monitoring of USV. Through the GUI, the status of each element of the electric power system and the electric thruster system was checked, and the user was notified when a fault occurred. For power system operating within the rated operating range, fault diagnostic methods with appropriate threshold settings can be used. However, because the operation range of the electric propulsion system is dynamic, it is difficult to diagnose the fault only based on the threshold. Therefore, in this study, a novel algorithm that applied a data-driven method to diagnose the fault of underwater thruster was developed. The small USV was used to verify the underwater thruster fault diagnosis algorithm, and the fault of the thruster blade and the fault of entanglement such as rope and net were selected as the fault situation. In order to diagnose these faults, the fault diagnosis system based on sensors and a data acquisition system was designed on a small USV. In the data-driven method, it is important to select the fault feature. Therefore, in this study, vibration, current consumption, RPM, and input voltage were selected as the fault features based on the vibration generation owing to the imbalance of the rotating body, the torque of the thruster, and the current formula of the DC motor. In addition, the changes in the fault feature data were predicted in the fault situation, the changes in data were verified through analysis of fault feature data acquired through experiments, and the changes in these data were used as symptoms of fault. The algorithm proposed in this paper measures the data, extracts the data of the fault features, performs the preprocessing process, applies the Principal Component Analysis technique and the entropy technique to detect the fault. Subsequently, this is analyzed through visualization in the three-dimensional principal component space to diagnose the fault. Each verification experiment of fault situations was conducted in an engineering basin, and the fault was classified and diagnosed based on the acquired data to verify the performance of the algorithm.