Fault Detection Method for Closed-Loop Control Loop of Water Jet Propulsion Device Based on Convolutional Autoencoder and Support Vector Data Description

The waterjet propulsion device, as an efficient power propulsion method, has been widely used in various types of ships. However, due to its complex structure, harsh working environment, and frequent failures caused by long-term high-load operation, it seriously affects the ship's mission performance. This paper proposes a data-driven fault detection method that introduces convolutional autoencoder (CAE) networks to implement adaptive feature extraction on the raw signal samples of closed-loop control systems, through convolutional encoding and decoding operations, while reducing the data dimension, the nonlinear inherent information in the raw signal samples is obtained, reducing the interference of redundant information. Using Particle swarm optimization and support vector data description (PSO-SVDD) and the Kernel Density Estimation (KDE) to obtain Statistics and Control Limits for Feature Space and Residual Space. Then Using composite statistics combining feature statistics and residual statistics to characterize health status of closed-loop control systems. The effectiveness of the proposed method is verified by physical experiments based on NI Compact RIO. The experimental results indicate that the proposed method has strong engineering value and can improve the reliability of waterjet propulsion devices.