Fault detection and diagnosis of a class of actuator failures via online approximators

A general framework for the detection and diagnosis of a class of actuator faults is developed. This framework, which addresses both abrupt and incipient faults, utilizes an adaptive detection observer along with a learning scheme for failure diagnosis. The actuator failure is modeled by a multiplicative perturbation of the actuator signal (actuator gain) that is described by a nonlinear function of the measurable output signal. Online approximators (such as neural networks) are used to estimate the unknown fault function and robust adaptive schemes are introduced to account for modeling errors that affect the diagnosis process and may cause false alarms.