Rudder servo-system fault diagnosis using neural network fault modeling

The model based fault detection and identification scheme for the rudder servo system as the actuator in the closed loop course keeping control system is presented. The nonlinear process model that enables description of a wide class of fault types is given and the fault function as the residual for detecting and identifying the fault is introduced. A fault detection scheme with variable thresholds is suggested and its performance has been tested via simulation in the presence of modeling uncertainty and noise. A fault identification scheme based on comparison between the fault function estimate and the output of a predefined analytical fault model is suggested.