A Multimodal Fault Diagnosis Model Utilizing GAF Transformation for Controllable Pitch Propeller Hydraulic System

• Main Authors: Cai, Huihuang, Chen, Wei, Fu, Huanglong, Wang, Hua, Wang, Jingcheng
• Format: Conference Proceeding
• Language: English
• Subjects: Fault diagnosis; Feature extraction; Gramian angular field; Hydraulic systems; Load modeling; Marine vehicles; multi-head attention; Navigation; Propellers; Sensor phenomena and characterization; Sensor systems; time series classification; Transforms
• ISSN: 2576-3555
• DOI: 10.1109/CoDIT62066.2024.10708444

The hydraulic system of the controllable pitch propeller (CPP) serves as the power source for adjusting the blade pitch. Due to its structural complexity and harsh operating conditions, any malfunction of this system could significantly impair the vessel's navigation. Apart from the pitch command and the pitch feedback of CPP itself, the number and types of sensors added for fault diagnosis are not fixed. Therefore, this study establishes a multimodal fault diagnosis model based on Gramian angular field(GAF). By considering the deviation values of pitch feedback and pitch command as key variables, this model transforms them into a two-dimensional image using GAF, thereby obtaining more effective feature information. Image features are extracted using ResNet, while the temporal features of additional sensors are extracted using GRU. Finally, a multi-head attention mechanism is employed for feature fusion. This method achieves a classification accuracy of 100% in the classification of four types of hydraulic system faults simulated on the experimental platform.