Incipient fault detection of nonlinear dynamical systems with disturbance via deterministic learning

Incipient fault detection (IFD) is becoming an important research issue in practical systems, especially for the systems with disturbance. Small disturbance can cause unexpected dynamics changes, and incipient instability will be indicated in the nonlinear dynamical systems. Based on the gradual change of incipient faults, three kinds of different dynamics patterns of the nonlinear system are defined (i.e., health patterns, sub-health patterns and fault patterns). In particular, sub-health dynamics patterns refer to the incipient instability modes of the system. Firstly, the dynamics of system patterns can be locally accurately approximated and stored in constant radial basis function (RBF) networks through deterministic learning. Secondly, the representative and finite system dynamics patterns are selected to establish a system dynamics pattern database. Finally, a rapid incipient fault detection scheme is suggested, and the rapid detection of system incipient instability and incipient faults is implemented by using dynamical pattern recognition. The fault delectability of nonlinear systems with small disturbance is derived rigorously. Simulation studies are included to demonstrate the effectiveness of the approach.