Fault detection and isolation for unknown nonlinear systems using expert methods

In this paper a new comprehensive method for fault detection and isolation (FDI) for unknown nonlinear systems is presented. This method detects and eliminates sensor and actuator faults, as well as plant's component faults. Fault type and location are precisely determined using sensor measurements and controller signals. Fault magnitude of sensor and actuator gain/bias faults is estimated using neuro-fuzzy models and gradient descent method. A fuzzy compensator with an adaptive output gain accommodates the faults and eliminates their effects for a wide range of plant's components. Simulation results on a two-link rigid planar manipulator demonstrate the capability of the proposed technique for detection, diagnosis and accommodation of faults