Abstract: This paper offers an integrated model-based and data-driven early fault detection scheme for a ship rudder electro-hydraulic servo system (RESS). First, the state equation of RESS is established, the common faults in the system are analyzed, and the uncertain factors in the system are classified. To reduce the influences of various uncertainties, a hybrid processing method with several steps is proposed next. Using the system input and output data in normal state, the uncertain model parameters can be identified, and then a robust fault detection observer is designed to eliminate the influences of system inherent nonlinearity and unknown external force. To deal with the remaining uncertainties and disturbances, a neural network based compensation model using actual and observed system data is constructed, which can further reduce the influences of uncertainties and to detect the early faults effectively. Both simulation and experimental results show that the combined method is efficient and can be used for on-line fault detection.

Key words: rudder; electro-hydraulic; model-based; data-driven; early fault detection