Abstract: According to the features of the fault diagnosis system, a dynamic neural model-output recurrent network, which fits for real-time fault detection is applied to the fault detection of a displacement servo-system. By applying output recurrent network, the neural model becomes the full model of the system. The simulation indicates that fault learning is avoided, and the faults which being very difficult to be detected by other methods can be detected. Comparing with time delayed feedforward network and diagonal recurrent network, output recurrent network shows its advantages in real-time fault detection system.

Key words: Dynamic neural network, Fault detection, Hydraulic servo-system, Output recurrent network